Model: W-61/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457 Source: Original Platform
478 lines
1.3 MiB
478 lines
1.3 MiB
{"epoch": 0.0, "step": 1, "batch_size": 128, "mean": 0.01704716682434082, "std": 0.7752009630203247, "min": -2.2227783203125, "p10": -0.8393455505371094, "median": 0.0678863525390625, "p90": 0.8599296569824219, "max": 2.196014404296875, "pos_frac": 0.5546875, "sample": [0.6739997863769531, -0.3828582763671875, -1.298828125, 0.071014404296875, 0.8088531494140625, 0.6873779296875, -0.081512451171875, 1.77093505859375, -1.4920806884765625, 1.769927978515625, 0.0503692626953125, -0.1540679931640625, 0.15589141845703125, -0.27809906005859375, -0.009723663330078125, 0.3800621032714844, 0.111541748046875, 0.2325439453125, 1.35693359375, -0.482269287109375, 0.2686614990234375, 1.01312255859375, -0.5781745910644531, -0.1348876953125, 0.15979766845703125, -1.054656982421875, 0.01500701904296875, 0.2495269775390625, 0.5402507781982422, 1.80511474609375, -0.8189926147460938, -0.17108154296875, 0.2899322509765625, -0.8205108642578125, 0.650146484375, -0.6688079833984375, 0.080841064453125, -1.4810791015625, 0.07396697998046875, -1.242919921875, 0.086151123046875, 1.62188720703125, -0.38132476806640625, 0.317474365234375, 2.196014404296875, 0.5562095642089844, 0.1424732208251953, -0.7183074951171875, -0.6070556640625, -0.6271820068359375, -0.5325164794921875, 0.67987060546875, -1.229156494140625, -1.91693115234375, -0.143280029296875, 0.2532958984375, -0.349273681640625, 0.42700958251953125, -0.262420654296875, 1.2039794921875, 0.364410400390625, 0.8435211181640625, 0.00079345703125, -0.19837188720703125, -0.5950164794921875, 0.362579345703125, 0.7775421142578125, -0.015625, -0.0785369873046875, -0.6810150146484375, 0.31806182861328125, -0.27123260498046875, 0.00982666015625, 0.098541259765625, -1.4693603515625, 0.20909881591796875, 0.87872314453125, 0.0, 0.4891510009765625, 0.100677490234375, -0.3092498779296875, 0.839263916015625, -0.5568389892578125, -0.5927276611328125, 1.82940673828125, -1.8787841796875, 0.4121856689453125, 0.1029205322265625, -0.05619049072265625, 0.41302490234375, -0.20123291015625, -1.11163330078125, 0.11328125, -0.6293182373046875, -0.4356689453125, 1.026641845703125, -0.4438629150390625, 0.06475830078125, 0.15139389038085938, 0.4244384765625, -0.3410911560058594, -0.8832931518554688, 0.38405609130859375, -0.6134033203125, 0.14745521545410156, -2.2227783203125, -0.3340606689453125, -0.4396820068359375, 0.543609619140625, 0.310455322265625, -0.6277847290039062, 0.28015899658203125, 0.1476898193359375, 0.3895263671875, 0.0498046875, -0.1731414794921875, 0.24071502685546875, -0.351348876953125, -0.0234375, 0.6951370239257812, 0.0606689453125, 0.8756256103515625, 0.680145263671875, 0.14440345764160156, 0.8532028198242188, -1.6363525390625, 1.4149551391601562, -0.476959228515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/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": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000002.npy"}
|
|
{"epoch": 0.004188481675392671, "step": 3, "batch_size": 128, "mean": 0.03528323769569397, "std": 0.8528732657432556, "min": -2.263885498046875, "p10": -1.0426712036132812, "median": 0.0717010498046875, "p90": 0.918582534790039, "max": 2.94036865234375, "pos_frac": 0.5546875, "sample": [-0.159423828125, 0.06467056274414062, -0.1837158203125, -0.4634208679199219, 0.2940826416015625, -1.276702880859375, 0.31170654296875, 0.40313720703125, 0.0, -0.579925537109375, 0.24652099609375, -0.895843505859375, 0.752593994140625, 1.205413818359375, -0.33465576171875, -0.034564971923828125, -0.99334716796875, 1.020355224609375, 1.9177703857421875, -0.52459716796875, -0.6342315673828125, -0.148895263671875, 0.37066650390625, 0.8455657958984375, 0.083587646484375, -0.12957763671875, 2.94036865234375, -1.36138916015625, 0.79443359375, -0.621734619140625, -0.34604644775390625, 0.08338546752929688, 0.927642822265625, 0.337860107421875, 0.90399169921875, 0.18043136596679688, -0.6065521240234375, -0.260162353515625, -0.54766845703125, 1.1754150390625, -0.1134185791015625, -1.72747802734375, 0.23712158203125, 1.1741790771484375, 0.062744140625, -1.42352294921875, -0.0321044921875, 0.2434234619140625, 1.889404296875, -0.563201904296875, -1.27764892578125, 0.10906219482421875, 0.857330322265625, 0.34612274169921875, 1.5581207275390625, 0.1558837890625, 0.31787872314453125, 2.26800537109375, 0.3165740966796875, 0.3820915222167969, 0.0, -1.406829833984375, 2.76055908203125, -2.263885498046875, -0.6443252563476562, -1.0804290771484375, -1.160858154296875, -1.0264892578125, 0.09748458862304688, 0.13690948486328125, -1.1876220703125, 0.176666259765625, 0.512786865234375, -1.330108642578125, -0.24415206909179688, 0.7548828125, -0.298004150390625, 0.2437744140625, 0.4063720703125, 0.18304443359375, 0.1346282958984375, 0.0597991943359375, 0.9146995544433594, 0.0687255859375, -0.606903076171875, 0.151702880859375, -0.44140625, 0.4043731689453125, -0.413970947265625, 0.60626220703125, 0.262420654296875, 0.139923095703125, -0.722503662109375, -1.88250732421875, -0.42833709716796875, 0.2695770263671875, 0.09971237182617188, 0.37413787841796875, 0.24019241333007812, -0.03401756286621094, -0.5748825073242188, 0.8724365234375, 0.32840728759765625, -0.194976806640625, 0.3956298828125, 0.81109619140625, -0.24166107177734375, 0.323272705078125, -0.68682861328125, 0.01493072509765625, -0.89105224609375, -0.615142822265625, 1.1206207275390625, -0.52032470703125, 0.074676513671875, 0.5386734008789062, -0.3767547607421875, -0.445220947265625, -0.0890350341796875, 0.17203521728515625, 2.40081787109375, -0.23857879638671875, 0.0230560302734375, 0.0044231414794921875, -1.3358993530273438, -0.336029052734375, 0.43817901611328125, 0.180389404296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000003.npy"}
|
|
{"epoch": 0.0062827225130890054, "step": 4, "batch_size": 128, "mean": 0.03976723551750183, "std": 0.8925357460975647, "min": -2.1094970703125, "p10": -1.1655914306640625, "median": 0.123382568359375, "p90": 1.0796112060546874, "max": 2.584716796875, "pos_frac": 0.5546875, "sample": [0.16745758056640625, -0.0113067626953125, -0.12430953979492188, -0.15960693359375, 0.44775390625, 0.00485992431640625, 0.6557121276855469, 1.1419677734375, 1.073028564453125, 1.027862548828125, -0.742279052734375, 0.29583740234375, -0.559814453125, 0.1097869873046875, 0.4580192565917969, 0.983642578125, 0.289520263671875, -1.785736083984375, 0.97637939453125, -0.257476806640625, -1.89599609375, -2.1094970703125, 0.2487640380859375, -1.298858642578125, -0.34423828125, 1.3187713623046875, 1.73736572265625, 0.29291534423828125, -1.158935546875, -0.05326080322265625, -0.98687744140625, -0.040004730224609375, -0.09857177734375, -1.4153900146484375, 0.35247802734375, 0.00537109375, 0.24920654296875, 0.116424560546875, 0.2220611572265625, 0.844146728515625, -0.0001163482666015625, 0.6223983764648438, -0.44336891174316406, 0.657470703125, -0.68780517578125, 0.54986572265625, -0.593780517578125, -0.7223129272460938, -0.08905029296875, -0.220947265625, 0.941558837890625, 0.5704269409179688, 0.24786376953125, -1.069305419921875, 0.5229644775390625, -0.34358978271484375, -1.2397003173828125, 1.094970703125, 2.584716796875, 0.248809814453125, -0.444793701171875, -1.84912109375, 0.5371170043945312, -0.0978240966796875, 0.911224365234375, 0.2464599609375, 0.1204986572265625, -1.181121826171875, -0.15084075927734375, -1.1494140625, 0.5485076904296875, -0.977264404296875, 0.14434814453125, -1.745849609375, 0.220306396484375, 0.148162841796875, -1.12078857421875, -1.562713623046875, -0.07232666015625, -0.39202117919921875, 0.5395584106445312, -0.0562591552734375, -0.25885009765625, -0.98687744140625, 0.8902740478515625, 1.535614013671875, 0.0233917236328125, -0.603973388671875, -0.084930419921875, -0.725860595703125, 0.28833770751953125, 0.015106201171875, -0.7218551635742188, -0.3665771484375, 0.605377197265625, 0.3362274169921875, 2.04205322265625, -0.057849884033203125, -1.4424285888671875, -0.90203857421875, -1.00982666015625, 1.4866943359375, 0.2564811706542969, 0.671173095703125, 1.097564697265625, -1.6971435546875, 0.5212020874023438, 0.175079345703125, -1.4215087890625, 1.34051513671875, 0.1262664794921875, 0.4124755859375, 0.33031463623046875, 0.13885498046875, 1.758331298828125, 0.754547119140625, -0.9674072265625, 0.6370086669921875, 1.242401123046875, -0.2299957275390625, -0.5736846923828125, 0.8507080078125, 0.604705810546875, 0.5851821899414062, -0.02801513671875, 0.2193603515625, 0.549560546875, 2.45013427734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000004.npy"}
|
|
{"epoch": 0.008376963350785341, "step": 5, "batch_size": 128, "mean": -0.020647287368774414, "std": 0.8242490291595459, "min": -3.044219970703125, "p10": -1.0054870605468749, "median": 0.021732330322265625, "p90": 0.9672424316406247, "max": 2.333587646484375, "pos_frac": 0.5234375, "sample": [-0.7140045166015625, 0.23980712890625, -0.0721282958984375, -1.547882080078125, 0.249053955078125, 1.04205322265625, 0.9130859375, -0.0434722900390625, 0.0133209228515625, 0.15875244140625, -0.07745361328125, 0.190948486328125, 0.1552734375, -0.7495803833007812, 0.072418212890625, -1.39208984375, 0.5738868713378906, 0.5364990234375, 0.841064453125, -2.1300048828125, -0.4263916015625, 0.30487060546875, -0.15665435791015625, 0.40997314453125, -0.871978759765625, -0.94281005859375, 0.0800018310546875, 0.3350677490234375, 0.25258636474609375, 0.1431102752685547, -0.47906494140625, 0.2201995849609375, 2.333587646484375, -0.15252685546875, 1.7292633056640625, 0.6808929443359375, 0.38238525390625, 0.083770751953125, -1.062042236328125, 2.1708984375, -0.0804443359375, 0.45904541015625, -0.316925048828125, 0.088531494140625, 0.011962890625, -0.25282859802246094, -3.044219970703125, 1.1857147216796875, -0.11981391906738281, -0.2205810546875, 0.913299560546875, -0.880157470703125, 0.302001953125, -0.28774261474609375, -0.346649169921875, -0.527191162109375, 1.48406982421875, 0.376708984375, 0.5554962158203125, 1.72088623046875, -1.7830810546875, 0.230377197265625, -0.257659912109375, -0.472412109375, -0.059597015380859375, 0.03839111328125, 1.0532684326171875, -1.106201171875, -0.5499267578125, -0.31280517578125, -0.560455322265625, -0.5687942504882812, 0.5737533569335938, -1.0732345581054688, 0.11799049377441406, -0.3743896484375, -0.97314453125, -0.0396728515625, -0.64447021484375, 1.0910568237304688, -0.83123779296875, 0.4505615234375, 0.67431640625, -0.3994140625, 1.33758544921875, -1.10040283203125, -0.67376708984375, 0.106597900390625, 1.36773681640625, -1.113555908203125, -1.806640625, -0.103607177734375, -0.8094940185546875, -0.18096923828125, -0.368072509765625, -0.724151611328125, -0.989044189453125, 0.11026763916015625, 0.15096282958984375, -1.24798583984375, 0.9351806640625, -0.61328125, 0.03014373779296875, 0.476409912109375, -0.44672393798828125, 0.103424072265625, 0.8065185546875, -0.5094795227050781, 0.6144790649414062, -0.600372314453125, 0.1107940673828125, -0.5670013427734375, 0.1966552734375, 0.07733154296875, 0.59722900390625, 0.000278472900390625, 0.7553558349609375, -0.5586700439453125, 1.193572998046875, 0.4959716796875, 0.759490966796875, -1.043853759765625, 1.324798583984375, -0.0883941650390625, 0.77947998046875, 0.218597412109375, -0.2902679443359375, 0.104949951171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000005.npy"}
|
|
{"epoch": 0.010471204188481676, "step": 6, "batch_size": 128, "mean": -0.006272077560424805, "std": 0.8966522216796875, "min": -2.076416015625, "p10": -1.0913604736328124, "median": -0.009037017822265625, "p90": 1.076931762695312, "max": 3.09100341796875, "pos_frac": 0.484375, "sample": [-1.0395355224609375, 2.02386474609375, 0.142242431640625, 0.52734375, 0.0396728515625, -0.4926605224609375, 1.514678955078125, -0.01441192626953125, -0.7167587280273438, -1.6752777099609375, -0.6385879516601562, 2.2102508544921875, -0.9546051025390625, -0.0428466796875, -0.162261962890625, -0.7938690185546875, 0.22711181640625, -0.11107063293457031, -0.29493141174316406, -0.3675537109375, 0.732879638671875, 0.3620185852050781, 0.6196136474609375, 1.4404144287109375, -0.31512451171875, -0.540252685546875, 0.1884765625, -0.3294677734375, -0.469482421875, 0.3910789489746094, -0.1939239501953125, 0.213623046875, -0.802459716796875, -0.003662109375, 0.09285736083984375, 0.129791259765625, -0.079345703125, -0.234375, -0.730316162109375, 2.36688232421875, -0.493011474609375, -0.626953125, -1.81951904296875, -0.50518798828125, -0.20528030395507812, -1.282928466796875, 0.16669654846191406, 1.4063720703125, -0.0715179443359375, -0.7327880859375, 0.37340545654296875, 1.0274658203125, -0.526580810546875, -0.6234130859375, -0.451324462890625, 0.19919586181640625, 1.192352294921875, -0.5495758056640625, -0.77130126953125, -0.5037841796875, 0.2914276123046875, -1.403472900390625, 1.2159271240234375, 0.66448974609375, -2.076416015625, 0.588592529296875, 0.21261978149414062, 1.41937255859375, -1.53753662109375, 0.6983566284179688, 1.6422882080078125, -0.0426483154296875, 0.139739990234375, -1.087432861328125, 0.09302520751953125, 0.3646697998046875, 0.4737091064453125, 0.06751251220703125, 0.0, 0.500579833984375, 0.823028564453125, -1.01824951171875, -0.9470367431640625, 0.032619476318359375, 0.8749847412109375, 0.3979949951171875, 0.113128662109375, -0.6212158203125, -0.13116455078125, 0.3637542724609375, -0.46240234375, -0.502471923828125, 0.9705390930175781, -0.2366943359375, -0.3995361328125, 0.4000091552734375, 0.59869384765625, 0.5936660766601562, 0.047576904296875, -1.1502685546875, 1.99273681640625, -1.19195556640625, 0.225494384765625, 0.58197021484375, -0.439483642578125, 0.0709075927734375, -0.217498779296875, -0.86199951171875, -1.23419189453125, -0.19100189208984375, 0.30548095703125, 0.20961952209472656, 3.09100341796875, 0.17107391357421875, 0.04511260986328125, 2.21728515625, -0.188232421875, 0.7188720703125, -1.10052490234375, 0.4876251220703125, -0.021453857421875, -2.0308837890625, -0.200042724609375, -1.19085693359375, 0.7857666015625, -1.623046875, 0.12697219848632812, -0.7356758117675781], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000006.npy"}
|
|
{"epoch": 0.012565445026178011, "step": 7, "batch_size": 128, "mean": 0.05461117625236511, "std": 0.8927361369132996, "min": -2.671661376953125, "p10": -0.8963287353515624, "median": 0.0045719146728515625, "p90": 1.276936340332031, "max": 3.035888671875, "pos_frac": 0.5078125, "sample": [0.588592529296875, 3.035888671875, -0.07904052734375, 1.297882080078125, -0.133270263671875, -1.3701934814453125, 0.908538818359375, -0.906005859375, -0.163543701171875, -0.08052825927734375, 0.5184860229492188, 0.1918487548828125, 0.092529296875, 0.13770294189453125, 0.4908447265625, -2.671661376953125, 0.0721282958984375, 1.2104873657226562, 0.0, 0.006000518798828125, -0.722137451171875, 0.5817108154296875, -0.434295654296875, -0.3190765380859375, -0.281005859375, 0.57647705078125, -0.813446044921875, -0.9255447387695312, -0.033935546875, -1.0162353515625, 0.56756591796875, 0.043914794921875, -0.080596923828125, -0.2879180908203125, -0.1341552734375, 1.7916259765625, 0.27417755126953125, 2.348175048828125, -2.59307861328125, 0.34214019775390625, -0.95574951171875, -0.2180938720703125, 0.355010986328125, -0.3726806640625, -0.985076904296875, 1.383941650390625, 1.6795806884765625, -0.07318878173828125, -0.0046558380126953125, -0.8751220703125, 0.01397705078125, -0.456146240234375, 0.4373779296875, -0.1076812744140625, -0.69256591796875, 0.003143310546875, -0.278167724609375, 0.0633544921875, -0.8438262939453125, 0.43621826171875, -0.83587646484375, 1.853912353515625, 0.5009078979492188, -0.20501708984375, 0.3394012451171875, 1.392669677734375, -0.412109375, 0.21155929565429688, 0.604736328125, 0.353790283203125, -0.107635498046875, 2.330352783203125, 0.30551719665527344, 0.068328857421875, -0.28435516357421875, -0.31927490234375, -0.2343292236328125, -2.49603271484375, -1.9803466796875, -0.614166259765625, 0.47509765625, 0.2235107421875, 0.1476287841796875, 0.33866119384765625, 0.36474609375, 0.023834228515625, -1.36651611328125, -0.088897705078125, 0.619781494140625, 0.32079315185546875, -0.345062255859375, 0.7179412841796875, -0.4304656982421875, -0.3884010314941406, -0.3757781982421875, -0.68023681640625, -0.44091796875, 0.05871772766113281, 0.29447174072265625, -0.982177734375, -0.118133544921875, -0.892181396484375, -0.33251953125, 0.9561767578125, 0.565338134765625, 0.4449462890625, 0.53314208984375, 0.21429443359375, 1.7025146484375, 0.397430419921875, 0.1908721923828125, -0.184356689453125, 0.036407470703125, 1.2679595947265625, -0.6493911743164062, 0.85491943359375, -1.224853515625, 0.1988525390625, 1.357147216796875, 2.0662841796875, -0.21966552734375, -0.4791259765625, -0.3204803466796875, -0.08928298950195312, -0.16172218322753906, -0.5826416015625, 0.480621337890625, 1.4782180786132812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000007.npy"}
|
|
{"epoch": 0.014659685863874346, "step": 8, "batch_size": 128, "mean": -0.07707980275154114, "std": 1.0771397352218628, "min": -3.396484375, "p10": -1.466558837890625, "median": -0.045963287353515625, "p90": 1.1552062988281249, "max": 3.47198486328125, "pos_frac": 0.46875, "sample": [-1.09466552734375, -0.01654815673828125, -0.5720062255859375, -0.00445556640625, 0.56134033203125, -1.68212890625, -0.7694091796875, 0.18768692016601562, 0.35889625549316406, -2.881591796875, -0.14913177490234375, -1.558074951171875, -0.4002685546875, 0.2327880859375, 1.000396728515625, 0.30464935302734375, 0.1012725830078125, 0.330352783203125, 0.2699604034423828, 0.608306884765625, 0.18878173828125, 1.78399658203125, -1.462860107421875, 0.29888916015625, 0.66619873046875, -0.372283935546875, 0.89605712890625, -0.6837158203125, -1.475189208984375, -3.396484375, -0.65472412109375, 3.47198486328125, -0.34991455078125, -0.3291015625, 0.7005462646484375, -0.722625732421875, -0.726593017578125, 0.085205078125, -2.066131591796875, -2.0821380615234375, 1.22015380859375, -1.22552490234375, 0.74554443359375, 0.046234130859375, 0.2318115234375, -0.812408447265625, -0.05939292907714844, 0.22428131103515625, -0.003082275390625, 1.10736083984375, -1.9400634765625, 0.11395263671875, 0.84735107421875, -0.16400718688964844, -0.183746337890625, -1.03155517578125, 0.648406982421875, -0.4697265625, -0.646820068359375, -2.606658935546875, 0.216339111328125, 1.30322265625, -0.3111572265625, 2.003570556640625, 0.21894073486328125, -0.9071044921875, -0.9698638916015625, 0.0690765380859375, 0.92254638671875, 3.3526611328125, -0.2017822265625, 0.458984375, -0.0400848388671875, -0.8950729370117188, 1.31756591796875, -1.924041748046875, 2.5135498046875, 0.16977691650390625, 0.03171539306640625, -1.1480560302734375, 0.34055328369140625, -0.184814453125, -0.545257568359375, 0.18001747131347656, 0.051025390625, 0.0057125091552734375, -2.0594329833984375, 0.13409423828125, 0.3428497314453125, 1.14141845703125, -1.02001953125, -1.3179931640625, 0.8103485107421875, 1.2180023193359375, -0.2768707275390625, 0.231109619140625, 2.1922607421875, 1.596923828125, 0.5808525085449219, 1.1873779296875, -0.5507888793945312, 1.1126708984375, -0.10211181640625, -0.361236572265625, -0.4934272766113281, 0.198638916015625, -0.3440093994140625, -0.05184173583984375, -0.63525390625, -1.48663330078125, -0.5086936950683594, -0.072235107421875, 0.7342529296875, -0.10137939453125, -0.8580322265625, 0.510894775390625, -0.202056884765625, -1.87103271484375, -0.223724365234375, -0.3371734619140625, -0.2853546142578125, -0.7834930419921875, 1.4012298583984375, -0.6787567138671875, -0.08251190185546875, 0.02774810791015625, 0.85723876953125, -0.10746383666992188], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000008.npy"}
|
|
{"epoch": 0.016753926701570682, "step": 9, "batch_size": 128, "mean": -0.0376712828874588, "std": 0.964229166507721, "min": -3.8287353515625, "p10": -1.4801116943359376, "median": 0.07757568359375, "p90": 1.074800109863281, "max": 2.59002685546875, "pos_frac": 0.5234375, "sample": [-0.5032196044921875, -1.50469970703125, -0.3781890869140625, 0.3136711120605469, 1.048858642578125, -0.181060791015625, 0.5084075927734375, 0.8402900695800781, 0.8338623046875, 0.30223846435546875, -0.113311767578125, -1.75347900390625, -1.0380859375, -0.244598388671875, -1.37322998046875, -0.97210693359375, 1.41790771484375, -1.58001708984375, 0.3712158203125, -1.18731689453125, 0.21797943115234375, 1.3587646484375, 0.4865570068359375, -0.4359130859375, -0.10372161865234375, -0.5914230346679688, 1.494140625, 0.54461669921875, 0.632080078125, -3.8287353515625, -1.962982177734375, -0.2042236328125, -0.333740234375, 0.4456329345703125, 0.0782012939453125, 0.05127716064453125, 1.657806396484375, -0.22930908203125, -0.0832366943359375, 0.252105712890625, -2.093109130859375, -0.537811279296875, 0.62994384765625, -1.5003814697265625, -0.13775634765625, -1.478363037109375, -0.899383544921875, -0.63165283203125, -2.301788330078125, -1.0577011108398438, -0.682281494140625, -1.882843017578125, -1.48419189453125, 0.21588134765625, 0.293792724609375, 0.44475555419921875, 1.69500732421875, 0.080535888671875, -1.16864013671875, -0.396881103515625, 0.740081787109375, 0.23836517333984375, -1.7994384765625, -0.2411327362060547, 0.9628982543945312, 0.51055908203125, -0.5867767333984375, 1.558563232421875, -0.0325775146484375, 1.2707977294921875, -1.69232177734375, 0.12184524536132812, 0.97869873046875, 0.9980659484863281, 1.1353302001953125, 0.20468902587890625, -0.233795166015625, 0.6389236450195312, -0.1423187255859375, 0.139404296875, 0.905487060546875, 1.01544189453125, 0.345916748046875, -0.104583740234375, 0.4382171630859375, -0.7169036865234375, 0.0, 0.2890472412109375, 0.644561767578125, 0.128692626953125, 0.439453125, 0.16273117065429688, 1.227142333984375, 0.56396484375, 0.66522216796875, 0.2685356140136719, -0.8582763671875, -0.31854248046875, -0.590667724609375, 0.9331512451171875, 0.4790611267089844, 0.0, -0.195465087890625, 0.42352294921875, -0.34014892578125, -0.03627777099609375, 0.0, 1.13543701171875, 0.12248992919921875, 0.16196441650390625, 0.36944580078125, 0.12482452392578125, 0.033050537109375, -1.170562744140625, -0.94512939453125, 0.39981842041015625, -0.422637939453125, 0.6812744140625, -0.423492431640625, -0.228759765625, 0.0769500732421875, 2.59002685546875, -0.3131103515625, 0.46954345703125, -1.6699142456054688, -0.8897705078125, 1.6807861328125, 1.496551513671875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000009.npy"}
|
|
{"epoch": 0.018848167539267015, "step": 10, "batch_size": 128, "mean": 0.0028738677501678467, "std": 0.7700600028038025, "min": -2.034423828125, "p10": -0.9447608947753906, "median": -0.012692451477050781, "p90": 1.0258163452148434, "max": 2.086883544921875, "pos_frac": 0.46875, "sample": [0.65179443359375, 0.357818603515625, 0.0933685302734375, 0.2022857666015625, 0.59344482421875, -1.568817138671875, 0.107818603515625, 0.1154022216796875, -0.1080322265625, -0.33935546875, 0.19866943359375, 0.029409408569335938, 0.6733474731445312, 1.899749755859375, -0.819122314453125, -0.283905029296875, -0.19611358642578125, -1.1832275390625, 0.4134521484375, 1.120025634765625, 0.5576171875, 0.38129425048828125, -2.034423828125, -1.11602783203125, 0.6228179931640625, -0.00304412841796875, 1.362335205078125, 0.3593902587890625, -0.08953857421875, 0.2412872314453125, -0.28675079345703125, -0.302093505859375, -0.20236587524414062, -0.4199676513671875, -0.13684463500976562, -0.2559814453125, 1.2593994140625, -0.812652587890625, 1.903717041015625, -0.1009674072265625, 0.009185791015625, 0.24772262573242188, -0.323486328125, -1.327728271484375, -0.33331298828125, -0.4920196533203125, -0.016071319580078125, 0.0090789794921875, 0.5811767578125, -1.362701416015625, 0.10283279418945312, -0.24736785888671875, 0.30517578125, -0.99639892578125, -0.515869140625, -0.4896087646484375, -0.287567138671875, 1.53857421875, -0.1768035888671875, 1.63330078125, -0.80517578125, -0.0055389404296875, 0.5083770751953125, 0.204925537109375, -0.0828704833984375, -1.253448486328125, 1.0041046142578125, -0.41986083984375, -0.1205902099609375, 0.037139892578125, 0.2715606689453125, 1.5370941162109375, 0.695587158203125, -0.017059326171875, -0.838531494140625, 0.40830230712890625, -0.2940635681152344, 1.61773681640625, -0.633026123046875, 1.42413330078125, -0.3335723876953125, 0.83184814453125, 0.310394287109375, 0.2874755859375, 0.26717376708984375, 0.43389892578125, -0.44854736328125, 0.12381744384765625, -1.704345703125, 0.992706298828125, -0.319244384765625, -0.209869384765625, 0.06183624267578125, -0.398681640625, -0.896697998046875, -0.03387451171875, -0.167449951171875, 0.152099609375, -0.3203277587890625, 0.39007568359375, 0.5809783935546875, 0.038166046142578125, -0.391448974609375, 0.9375762939453125, -0.13232421875, -0.9701766967773438, -0.902252197265625, -0.829833984375, 0.33626556396484375, -0.68548583984375, -1.37762451171875, -0.009313583374023438, 0.2884368896484375, 0.0, -0.4120941162109375, -1.56939697265625, -0.033428192138671875, 0.065887451171875, -0.2610893249511719, -0.187408447265625, 1.367584228515625, 0.297088623046875, 1.07647705078125, -1.0522537231445312, 2.086883544921875, -0.933868408203125, 0.60430908203125, -0.56463623046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000010.npy"}
|
|
{"epoch": 0.020942408376963352, "step": 11, "batch_size": 128, "mean": 0.052095331251621246, "std": 0.9345623850822449, "min": -3.006134033203125, "p10": -0.8712211608886719, "median": 0.00902557373046875, "p90": 1.0471961975097657, "max": 3.3447265625, "pos_frac": 0.5078125, "sample": [0.55938720703125, -2.14166259765625, 0.841583251953125, 3.3447265625, 1.0856781005859375, -0.09397125244140625, 0.065887451171875, -0.42291259765625, -0.34912109375, 0.7322540283203125, 0.33748626708984375, 0.42581939697265625, 1.69793701171875, 0.37542724609375, 0.0149993896484375, -0.7743301391601562, 0.10943603515625, 0.14960956573486328, -0.172760009765625, 0.5625762939453125, 0.19402122497558594, -0.2724151611328125, -0.03143310546875, 2.643798828125, -0.4413604736328125, -0.651580810546875, 0.7111358642578125, -0.1461944580078125, 1.514190673828125, -0.45196533203125, -0.758575439453125, 0.6318359375, -0.42520904541015625, -1.59197998046875, -1.4677886962890625, 0.6471710205078125, -0.269989013671875, 0.8788948059082031, -0.6187744140625, 0.015224456787109375, -0.2455291748046875, -2.092193603515625, -0.25146484375, -1.19586181640625, -0.30303955078125, 0.82391357421875, -0.5173873901367188, 0.3480224609375, 1.14959716796875, 1.8580322265625, 0.0030517578125, -0.26727294921875, -0.2937774658203125, -0.790740966796875, -0.764984130859375, -2.640380859375, 0.304962158203125, 0.139923095703125, 0.66973876953125, 0.143310546875, 1.4434661865234375, -1.4360198974609375, -0.10113525390625, -0.009124755859375, -0.39654541015625, -0.1286773681640625, 1.0601043701171875, -0.4106636047363281, 1.0416641235351562, -0.054264068603515625, 1.021514892578125, -0.861358642578125, 0.593841552734375, -0.99884033203125, 0.25367164611816406, -0.660888671875, 0.6228790283203125, 0.40361785888671875, 0.186126708984375, 1.72283935546875, 0.2746467590332031, 0.46710205078125, 0.09093093872070312, -0.114227294921875, -1.831207275390625, 0.6349143981933594, 0.02017974853515625, -0.37762451171875, -0.7508544921875, 0.387908935546875, 0.100341796875, -0.6884231567382812, -3.006134033203125, 0.4475593566894531, 0.293609619140625, -1.10589599609375, -0.082275390625, -1.3319091796875, -0.23133468627929688, -0.15808486938476562, -0.353424072265625, -0.190765380859375, 2.4893798828125, -0.5453033447265625, 0.02313232421875, -0.388885498046875, 0.894744873046875, 1.7975311279296875, 1.2464447021484375, 0.3032989501953125, -0.1071014404296875, -0.4913330078125, -0.152252197265625, 0.30332183837890625, -0.0457305908203125, -0.4534912109375, 0.8921051025390625, 0.857025146484375, 0.2826080322265625, 0.9936676025390625, 0.510009765625, -0.3172149658203125, 0.7961273193359375, 0.6646881103515625, -0.8942337036132812, -0.21395111083984375, 0.3585357666015625, -0.4571380615234375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000011.npy"}
|
|
{"epoch": 0.023036649214659685, "step": 12, "batch_size": 128, "mean": 0.022167682647705078, "std": 0.9175094366073608, "min": -3.950439453125, "p10": -0.9685775756835937, "median": 0.087646484375, "p90": 1.0616851806640624, "max": 2.32421875, "pos_frac": 0.53125, "sample": [-0.687103271484375, 1.2886962890625, 0.54052734375, 0.4500732421875, -0.14925575256347656, 0.41005706787109375, -1.0966033935546875, -0.101104736328125, 0.0972900390625, -1.026031494140625, -1.4488677978515625, 0.79522705078125, 1.108062744140625, 0.12734222412109375, 0.240325927734375, 0.0245819091796875, 0.5424041748046875, -0.2284698486328125, -0.6798095703125, 0.7701873779296875, -1.075103759765625, -0.6197509765625, 2.243133544921875, -0.05036163330078125, 1.0208740234375, -0.5223388671875, -0.8457202911376953, -0.19962310791015625, -0.02899169921875, 0.491119384765625, -1.31182861328125, 0.9641494750976562, -0.5283279418945312, -1.897857666015625, -0.358612060546875, -0.184051513671875, -0.27093505859375, 0.65570068359375, -0.41021728515625, -0.22022247314453125, -0.21811676025390625, 0.24627685546875, -0.52947998046875, 1.159271240234375, -0.21463775634765625, -0.469757080078125, 0.10626220703125, 0.197052001953125, -1.322906494140625, 0.143524169921875, 1.41436767578125, 0.4636688232421875, 0.09515380859375, 0.179046630859375, -0.04632568359375, -0.07266998291015625, 1.15557861328125, -0.5770263671875, -0.9439544677734375, 0.204376220703125, -0.1416778564453125, 0.936187744140625, -0.324951171875, -0.856964111328125, -0.14619064331054688, 1.04180908203125, 0.44159698486328125, 1.64276123046875, 0.504425048828125, 0.984893798828125, 0.48459625244140625, -0.777374267578125, 0.22159576416015625, -0.69940185546875, 0.03368377685546875, 0.873504638671875, 0.8878250122070312, 0.1088409423828125, -0.6391143798828125, -1.85540771484375, 0.4964141845703125, 0.08013916015625, -0.910675048828125, 1.6290664672851562, 0.17171669006347656, 2.32421875, -0.19333457946777344, 0.13104248046875, -3.950439453125, 0.096221923828125, -0.11138916015625, 0.07459259033203125, -0.08036041259765625, 0.55145263671875, -1.6168670654296875, 0.34765625, -0.244140625, -0.0060577392578125, -0.4297332763671875, 0.785308837890625, -0.640838623046875, 0.5950355529785156, 0.461822509765625, 0.616607666015625, 0.38470458984375, 1.1512451171875, 0.7884368896484375, 0.96954345703125, -0.5941925048828125, 1.83599853515625, 1.234161376953125, -0.37054443359375, 0.7444534301757812, -0.7955780029296875, 0.128204345703125, 0.518035888671875, -0.17953109741210938, -3.1861572265625, -1.619873046875, -0.4814910888671875, 0.539154052734375, 1.2822265625, 0.427490234375, 0.200653076171875, -0.934539794921875, -1.2662315368652344, -0.448333740234375, 0.813262939453125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000012.npy"}
|
|
{"epoch": 0.025130890052356022, "step": 13, "batch_size": 128, "mean": 0.027435779571533203, "std": 0.8041146397590637, "min": -1.812774658203125, "p10": -0.87796630859375, "median": -0.021877288818359375, "p90": 0.886785888671875, "max": 3.31988525390625, "pos_frac": 0.484375, "sample": [0.13094329833984375, -0.06751060485839844, -0.3374900817871094, 0.84930419921875, 0.343780517578125, 0.24383544921875, 0.45977783203125, 0.391632080078125, -0.465972900390625, -0.34398651123046875, 0.363677978515625, -1.126739501953125, -0.476043701171875, -0.4227409362792969, -0.06557655334472656, 0.3207244873046875, -0.1640625, -0.7186279296875, 1.4903411865234375, 0.0400848388671875, -0.0181884765625, 0.7283935546875, 3.31988525390625, -0.104827880859375, 0.6157684326171875, -0.2825355529785156, -0.19678688049316406, 0.30926513671875, 0.16149139404296875, -0.437347412109375, -0.427154541015625, -0.44927978515625, 0.03350830078125, 0.4051361083984375, -0.43929290771484375, 0.7231216430664062, -0.04571723937988281, -0.7333526611328125, -0.2938652038574219, -0.044708251953125, 0.24893760681152344, 0.394195556640625, 0.5719146728515625, -0.8753662109375, -0.4191131591796875, 0.052154541015625, -1.812774658203125, -0.1492462158203125, -0.0499267578125, -1.30877685546875, -0.66180419921875, -1.091278076171875, -1.0647621154785156, 0.459869384765625, 1.01531982421875, 0.0, -0.8402099609375, 0.891998291015625, 0.8474807739257812, -0.32840728759765625, 0.18585205078125, -1.2505645751953125, -1.5172958374023438, 0.144775390625, 0.884552001953125, -0.1927490234375, 1.15521240234375, 0.60736083984375, 0.8005828857421875, -0.02556610107421875, 1.75836181640625, -0.4610748291015625, 0.0431671142578125, 0.58013916015625, 0.51519775390625, 1.08184814453125, 0.22371673583984375, 2.13653564453125, -0.47707366943359375, 0.495513916015625, 0.12685775756835938, 0.45072174072265625, 0.5391769409179688, -0.831146240234375, 0.6860198974609375, -0.66094970703125, 0.0898895263671875, -0.21405029296875, 0.48779296875, 0.20529937744140625, 0.875518798828125, -0.14151763916015625, 1.2517776489257812, 0.13896942138671875, -0.6419448852539062, -0.7137374877929688, -0.031398773193359375, -0.5363311767578125, -1.4886627197265625, 0.25048828125, -0.410369873046875, -0.7998046875, 0.0645904541015625, -0.1448822021484375, 0.0290985107421875, -1.14959716796875, 1.698760986328125, -0.884033203125, -0.1703948974609375, -1.147796630859375, 0.062305450439453125, -1.24298095703125, 0.1851043701171875, -0.27883338928222656, 1.487060546875, 0.858978271484375, -0.805389404296875, 2.67840576171875, 1.1926727294921875, -0.42169189453125, 0.1551513671875, -0.59149169921875, -0.614166259765625, -0.3601875305175781, -0.636199951171875, 0.3880577087402344, -1.0282135009765625, -0.28270721435546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000013.npy"}
|
|
{"epoch": 0.027225130890052355, "step": 14, "batch_size": 128, "mean": 0.04740062355995178, "std": 0.867557942867279, "min": -3.354827880859375, "p10": -0.8826339721679688, "median": 0.029745101928710938, "p90": 1.0682556152343747, "max": 3.20458984375, "pos_frac": 0.515625, "sample": [0.909149169921875, 0.4998626708984375, -1.0013580322265625, 1.42523193359375, -0.3220062255859375, -0.5738983154296875, 0.0318756103515625, -0.04638671875, -0.06329345703125, 1.4305267333984375, 0.962310791015625, -0.579437255859375, -1.9603271484375, -1.566070556640625, -0.61492919921875, -0.691802978515625, 0.094085693359375, -0.773468017578125, -0.22010040283203125, 0.1722869873046875, 0.06357955932617188, -0.09976959228515625, 0.80792236328125, -0.0173187255859375, -0.885101318359375, -0.0357666015625, -0.324432373046875, 0.6691741943359375, 0.322479248046875, -0.21502685546875, 1.3069305419921875, -0.3607177734375, -0.3970947265625, 0.947357177734375, 1.5504150390625, 0.553253173828125, 0.92645263671875, 0.045284271240234375, 0.217437744140625, -0.203216552734375, 0.29824256896972656, 1.29705810546875, -0.9122467041015625, 0.472686767578125, 0.0, 0.028087615966796875, -0.68939208984375, 1.0123291015625, -0.07627105712890625, 0.7825698852539062, -0.5871429443359375, -0.200103759765625, 0.009248733520507812, 0.48260498046875, 0.1006622314453125, 0.213470458984375, -0.434661865234375, 2.008697509765625, -0.172210693359375, 0.422454833984375, -1.35919189453125, -0.027252197265625, 1.228240966796875, 0.29754638671875, 0.6888275146484375, 1.189727783203125, -2.248504638671875, -0.04937744140625, 0.031402587890625, 0.46533203125, -0.5313034057617188, 0.2424774169921875, -0.502838134765625, 1.13836669921875, 0.413177490234375, 0.846099853515625, -0.959197998046875, -0.00555419921875, -1.445831298828125, -0.43804931640625, 0.03957366943359375, -0.3958892822265625, -0.41546630859375, 0.14452362060546875, 0.8196258544921875, -0.7081184387207031, -0.32140541076660156, -0.19254302978515625, -0.8815765380859375, 0.2654247283935547, -0.4396820068359375, 0.15760040283203125, 2.44305419921875, 0.647552490234375, -3.354827880859375, -0.8336181640625, -0.40151023864746094, -0.39215087890625, 1.61993408203125, 0.0675201416015625, -0.676025390625, 0.273895263671875, 0.15482330322265625, 0.3023681640625, -0.5094947814941406, -0.0884552001953125, -0.19533538818359375, -1.6580810546875, 0.7168121337890625, 0.2083282470703125, 0.272552490234375, 1.194854736328125, -1.0889892578125, -0.246826171875, 0.09148597717285156, -1.0391845703125, 0.0619049072265625, 0.7256927490234375, 0.525360107421875, 0.14771652221679688, 3.20458984375, -0.4219207763671875, 0.820159912109375, -0.444976806640625, -0.415191650390625, 0.27471923828125, -0.04400634765625, 1.0382080078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000014.npy"}
|
|
{"epoch": 0.02931937172774869, "step": 15, "batch_size": 128, "mean": -0.10837952792644501, "std": 0.94266277551651, "min": -2.67059326171875, "p10": -1.2097740173339844, "median": -0.15191936492919922, "p90": 0.926654052734375, "max": 3.95172119140625, "pos_frac": 0.40625, "sample": [-0.2066497802734375, -0.696380615234375, 0.708282470703125, 0.5179443359375, -0.5431976318359375, 0.48426055908203125, 0.63568115234375, -0.5794677734375, -0.5975799560546875, -0.2479248046875, 0.12514877319335938, -0.363525390625, -0.419891357421875, -0.045978546142578125, -0.10274505615234375, -1.62109375, -0.42819976806640625, -0.2507286071777344, -0.22356414794921875, -0.1083526611328125, 1.173828125, -1.2200546264648438, -0.94769287109375, -2.02459716796875, 0.186370849609375, 0.5340118408203125, 0.507720947265625, -1.1163330078125, 0.731353759765625, -0.52581787109375, 0.31439208984375, -0.03662109375, -0.2178955078125, -0.5814743041992188, -0.1927490234375, -0.3843231201171875, 1.8021240234375, 1.261962890625, 1.891021728515625, -0.01806640625, -0.43593597412109375, -2.67059326171875, 0.947845458984375, 0.42333221435546875, -0.351287841796875, -1.4390869140625, 0.43689727783203125, -1.2053680419921875, -0.1258392333984375, 0.66558837890625, -0.4929828643798828, -0.7395858764648438, 0.557708740234375, 0.34228515625, -0.3690185546875, -1.096160888671875, 0.114410400390625, 0.3186492919921875, 0.154541015625, -0.91497802734375, -0.4891815185546875, 0.2513866424560547, -0.928955078125, -0.029888153076171875, -1.535369873046875, -0.0336151123046875, 0.080047607421875, -0.2576904296875, 1.080047607421875, -0.871978759765625, 0.2767219543457031, -0.15136337280273438, 1.0340423583984375, -1.3099212646484375, -0.559661865234375, 1.641937255859375, 0.0745849609375, 0.5538558959960938, -0.004302978515625, -0.018951416015625, -0.516632080078125, -0.0037670135498046875, -0.1895751953125, -0.41400146484375, 0.100555419921875, -0.54852294921875, -0.5796661376953125, -0.89630126953125, -0.15247535705566406, 2.599334716796875, 0.3968315124511719, -0.7040863037109375, 0.2215118408203125, -1.46673583984375, 1.271453857421875, 3.95172119140625, -0.3775615692138672, -0.61181640625, -1.2869873046875, -0.7855224609375, -0.639923095703125, 2.829376220703125, -0.500396728515625, 1.0121841430664062, 0.304168701171875, 0.247283935546875, -1.99139404296875, -0.3265876770019531, -0.5833587646484375, -0.7425880432128906, 0.7167282104492188, -0.8152618408203125, -0.38149261474609375, 0.917572021484375, -1.773101806640625, 0.024585723876953125, -2.5103759765625, 0.0040740966796875, 0.37853240966796875, 0.2991943359375, 0.279296875, 0.02408599853515625, -1.3199691772460938, -0.696807861328125, 0.628387451171875, 0.572235107421875, -0.2642097473144531, 0.332061767578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000015.npy"}
|
|
{"epoch": 0.031413612565445025, "step": 16, "batch_size": 128, "mean": 0.15118207037448883, "std": 0.9303906559944153, "min": -2.046295166015625, "p10": -0.946246337890625, "median": 0.06344795227050781, "p90": 1.18157958984375, "max": 3.424560546875, "pos_frac": 0.5234375, "sample": [0.69683837890625, 0.20309829711914062, 0.33031463623046875, 0.765655517578125, 1.5390625, 0.1746368408203125, 0.45684814453125, 1.99407958984375, 2.39141845703125, 0.373779296875, 1.1139678955078125, 0.12441635131835938, 0.04779815673828125, -1.21551513671875, 1.827880859375, 0.111663818359375, -0.99456787109375, 2.32086181640625, -0.548431396484375, -0.52337646484375, 0.6135101318359375, 0.04932403564453125, 0.70623779296875, 0.9276123046875, -0.49639892578125, -0.1563720703125, 2.17987060546875, 2.120269775390625, -0.14129638671875, -0.6702728271484375, 0.82757568359375, 0.64056396484375, -1.26348876953125, -1.65814208984375, 0.266082763671875, -0.345184326171875, 3.424560546875, -0.43477630615234375, 0.06674575805664062, -0.017070770263671875, 0.72552490234375, 0.10693359375, 0.987060546875, 1.0184326171875, -1.25830078125, -0.38191986083984375, 0.060150146484375, -0.3140106201171875, 1.18756103515625, 0.2383575439453125, 0.4500999450683594, -0.235107421875, -0.0145416259765625, -1.221221923828125, -1.651458740234375, -0.96124267578125, 0.24308013916015625, -0.04036712646484375, 1.0881729125976562, 0.7298583984375, -0.498779296875, -0.08031463623046875, -0.1494903564453125, -0.559722900390625, -0.43670654296875, 0.12548828125, -0.20056915283203125, 0.1796875, -0.07257080078125, -0.9398193359375, -0.10521697998046875, -0.9932861328125, -0.7415390014648438, 0.462310791015625, -0.360198974609375, -0.5908660888671875, 0.6793212890625, 0.09051513671875, -0.6076774597167969, -0.71295166015625, -0.634002685546875, -0.17897796630859375, -0.8255462646484375, -0.2110595703125, -0.20440673828125, 0.301177978515625, 0.0, 0.536468505859375, -0.285797119140625, -0.836334228515625, 0.408111572265625, 0.6307373046875, -0.180938720703125, -0.90924072265625, 2.19110107421875, 0.7761688232421875, 0.14951705932617188, -0.05511474609375, -0.7153472900390625, 0.6413288116455078, 1.102508544921875, 2.196044921875, -0.756561279296875, 0.307861328125, -0.3155937194824219, -0.222076416015625, -0.007659912109375, 0.47202301025390625, -1.1160888671875, 0.740570068359375, 0.2255096435546875, -1.4324493408203125, -2.046295166015625, -0.37261962890625, -0.318756103515625, 0.630401611328125, 1.1044769287109375, 0.6500396728515625, -0.63763427734375, 0.89569091796875, 0.15146636962890625, -0.1206512451171875, 1.36859130859375, -1.03759765625, 0.18536376953125, 0.401824951171875, 1.17901611328125, 2.45159912109375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000016.npy"}
|
|
{"epoch": 0.033507853403141365, "step": 17, "batch_size": 128, "mean": 0.010174959897994995, "std": 0.7467907071113586, "min": -3.060821533203125, "p10": -0.8280105590820312, "median": -0.0019283294677734375, "p90": 0.8457916259765623, "max": 2.6761474609375, "pos_frac": 0.5, "sample": [-0.04042816162109375, -0.4723052978515625, -1.98834228515625, -1.3504638671875, -0.523101806640625, -0.3263397216796875, 1.213714599609375, 0.377960205078125, 1.036834716796875, 0.11798858642578125, 0.54168701171875, -0.6204833984375, -0.94110107421875, 2.6761474609375, 1.08001708984375, -0.09879302978515625, -0.194793701171875, 0.8140869140625, 0.224090576171875, 0.427490234375, 0.1966552734375, 0.3403472900390625, -1.4185028076171875, -0.13067626953125, 0.4689750671386719, -0.27923583984375, -0.8038330078125, -0.41162109375, -0.07061767578125, -0.637939453125, 0.828094482421875, 0.249267578125, 0.219024658203125, 0.44781494140625, -0.608245849609375, 0.12476730346679688, -0.111907958984375, -3.060821533203125, 0.22941970825195312, 0.3523368835449219, -0.124755859375, -0.60260009765625, 0.16192626953125, -0.7085342407226562, -0.8951950073242188, 0.5750961303710938, 0.29116058349609375, -0.4483222961425781, -0.0147552490234375, 1.8274078369140625, -0.301300048828125, 0.6489410400390625, -0.6962890625, 0.3535614013671875, 0.0537261962890625, -0.0872802734375, 0.12786865234375, -0.398834228515625, -0.10723114013671875, 0.3251953125, 0.457672119140625, 0.0727386474609375, -0.018463134765625, 0.2862548828125, -0.95050048828125, 0.1220855712890625, -0.19384765625, -0.3317298889160156, 0.109954833984375, 0.18518829345703125, 0.817108154296875, -0.2008819580078125, -1.05340576171875, -0.8744049072265625, 0.6864166259765625, 1.57196044921875, -0.7544403076171875, 0.398223876953125, 1.5955810546875, 0.3670806884765625, -0.154388427734375, -0.6873016357421875, -0.0868682861328125, 0.044219970703125, -0.02129364013671875, -0.82684326171875, -0.07281494140625, 0.986083984375, -0.47100830078125, -1.3280029296875, -0.005985260009765625, 0.4025726318359375, -0.24603271484375, 0.699676513671875, 0.708282470703125, 0.148712158203125, -0.2451629638671875, -0.9962158203125, 0.8183746337890625, 0.00897216796875, -0.6699562072753906, 0.20147705078125, 0.78948974609375, -0.56732177734375, 0.00212860107421875, 1.062652587890625, -0.3656730651855469, -0.0248870849609375, 1.090362548828125, 0.013458251953125, -0.3680229187011719, -0.49706268310546875, -0.501312255859375, 0.48564910888671875, 0.18133544921875, 0.415679931640625, -0.038482666015625, 0.14410400390625, -0.109588623046875, -1.38848876953125, -0.2808341979980469, 0.313018798828125, 1.33612060546875, 0.10074615478515625, 0.8870849609375, -0.8307342529296875, -0.26027679443359375, 1.357208251953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000017.npy"}
|
|
{"epoch": 0.0356020942408377, "step": 18, "batch_size": 128, "mean": -0.048301324248313904, "std": 0.889039933681488, "min": -2.951141357421875, "p10": -1.0326263427734375, "median": 0.008962631225585938, "p90": 0.7840255737304684, "max": 2.520111083984375, "pos_frac": 0.5, "sample": [-0.9210052490234375, -1.104705810546875, -0.08987808227539062, 0.579864501953125, 0.4518890380859375, -0.51129150390625, 0.3472328186035156, -0.444732666015625, -0.22626495361328125, 0.2660102844238281, -0.704071044921875, 0.7558135986328125, -0.5024185180664062, -0.07086181640625, 1.554046630859375, 0.37744140625, 0.60406494140625, -1.544097900390625, -0.1344757080078125, -0.5555877685546875, 0.1050872802734375, 0.5087738037109375, -0.35260009765625, -0.1002044677734375, -1.271240234375, -0.7674694061279297, 0.1374969482421875, 0.74835205078125, -0.5805816650390625, -0.01125335693359375, 0.67840576171875, -2.793701171875, 0.14461517333984375, 0.4505615234375, 0.2726898193359375, -0.11162567138671875, 0.689605712890625, 0.15563201904296875, -0.12720489501953125, 1.23681640625, -0.430023193359375, 0.16595458984375, -2.5623779296875, -0.778289794921875, 0.27501678466796875, 0.3394775390625, 0.084075927734375, 0.15003204345703125, 1.0275115966796875, -0.30462646484375, 1.4667205810546875, -0.018585205078125, 1.25604248046875, 0.084808349609375, -1.10858154296875, -1.208831787109375, 0.7483367919921875, -0.5902099609375, 0.38970947265625, 0.26531982421875, 1.194091796875, -0.5691299438476562, -1.01727294921875, -0.25299072265625, -0.174072265625, 0.670257568359375, 0.2489013671875, 2.520111083984375, -0.842010498046875, -0.8485107421875, 0.2791633605957031, 0.6748924255371094, -0.5279693603515625, -0.17437744140625, 0.3418426513671875, 0.05340576171875, 0.27734375, 0.08062744140625, 0.5530853271484375, 1.2478790283203125, -2.951141357421875, 0.32027435302734375, 1.63604736328125, 0.6172943115234375, -0.016876220703125, 0.7001419067382812, -0.5753860473632812, 0.2545166015625, 1.98504638671875, 0.12773513793945312, -0.91192626953125, -0.27204132080078125, 0.3502044677734375, -0.159881591796875, -2.881591796875, 0.7021484375, -0.08405685424804688, -0.46038055419921875, 0.1305999755859375, 0.4845123291015625, -0.07579803466796875, 0.849853515625, 0.05999755859375, -0.61383056640625, -0.006793975830078125, -0.1363525390625, 0.153076171875, -0.550933837890625, -0.277984619140625, -0.442047119140625, -0.915435791015625, 0.21572113037109375, -0.5201416015625, 0.02471923828125, 1.1742095947265625, -0.52496337890625, -0.0536651611328125, 0.304443359375, -1.027252197265625, 2.42156982421875, -0.174468994140625, 0.4770660400390625, -0.4540252685546875, -1.045166015625, -1.29937744140625, -1.910400390625, 0.1219329833984375, -1.07763671875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000018.npy"}
|
|
{"epoch": 0.03769633507853403, "step": 19, "batch_size": 128, "mean": -0.04094433784484863, "std": 0.7840359210968018, "min": -1.60687255859375, "p10": -1.0050506591796875, "median": -0.09058570861816406, "p90": 0.866912841796875, "max": 2.443695068359375, "pos_frac": 0.453125, "sample": [0.13907241821289062, 0.863616943359375, 0.874603271484375, 0.5313720703125, -0.91339111328125, -0.438690185546875, -0.50628662109375, 0.043792724609375, -0.738037109375, 0.18084716796875, 0.19312667846679688, -0.2729530334472656, 0.8216094970703125, -0.11993408203125, 0.3408546447753906, 0.01226806640625, -0.5898895263671875, -1.59442138671875, 1.2966995239257812, 0.26799774169921875, 0.2638092041015625, 0.940032958984375, 0.18227577209472656, 0.336273193359375, -0.071502685546875, -0.1971435546875, 0.4367504119873047, -0.5412216186523438, -1.60687255859375, 0.7025413513183594, 2.15533447265625, 0.9739990234375, -1.32440185546875, 0.531494140625, -0.411346435546875, -0.10966873168945312, -0.246795654296875, 0.4017333984375, -1.2360992431640625, -0.03108978271484375, 1.0705718994140625, -0.035648345947265625, -1.063262939453125, -0.4665679931640625, 0.8511962890625, 2.438232421875, -0.5487251281738281, 1.19586181640625, -0.894287109375, -0.3681640625, -0.41024017333984375, -1.252197265625, -0.2748565673828125, -1.4129180908203125, -0.64361572265625, -0.86712646484375, -1.27789306640625, -1.40753173828125, 0.1618804931640625, -0.46649169921875, -0.30272674560546875, -0.182861328125, -0.6114959716796875, 0.15417861938476562, -0.4428253173828125, -0.6890716552734375, 0.46697998046875, 0.5025177001953125, -1.3640060424804688, 0.3910675048828125, 0.3753662109375, -0.32421875, -1.102081298828125, -0.606964111328125, 0.682403564453125, -0.423583984375, -0.015869140625, 0.25083160400390625, 0.76776123046875, 0.7667083740234375, -0.530731201171875, -0.6370849609375, 0.3538780212402344, -0.8749237060546875, 0.679534912109375, -0.2569427490234375, -0.111114501953125, -0.970947265625, -0.712554931640625, -0.2391357421875, -0.05181884765625, -1.21075439453125, 0.850311279296875, -0.2468414306640625, 0.272003173828125, 0.528564453125, 1.0835418701171875, -0.3977546691894531, 0.21735382080078125, 0.2834510803222656, -0.201507568359375, 0.9323883056640625, -0.865936279296875, 1.2079925537109375, 2.443695068359375, 0.14990234375, 0.3157958984375, -0.4891357421875, -0.27184295654296875, 0.253265380859375, -0.0576629638671875, -0.5362548828125, 1.96185302734375, -0.694061279296875, -0.496368408203125, -1.3154296875, -0.9801025390625, -0.4532012939453125, 0.27008056640625, 0.5266265869140625, 0.5161590576171875, 0.368011474609375, -0.22698974609375, 0.5080718994140625, 0.02028656005859375, -0.838348388671875, 0.7041168212890625, -0.180999755859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000019.npy"}
|
|
{"epoch": 0.039790575916230364, "step": 20, "batch_size": 128, "mean": 0.11721087992191315, "std": 0.9325563311576843, "min": -2.342132568359375, "p10": -0.959014892578125, "median": 0.06838607788085938, "p90": 1.2993927001953125, "max": 3.2412109375, "pos_frac": 0.5390625, "sample": [0.5897064208984375, 1.293304443359375, 0.317901611328125, 2.0059814453125, 0.154937744140625, 0.078338623046875, -0.45294189453125, -0.695343017578125, 0.79351806640625, -1.1364593505859375, -1.256103515625, -0.1457977294921875, 0.0865936279296875, 1.5177001953125, 0.272064208984375, 0.456756591796875, 1.103118896484375, -0.041168212890625, 0.6762847900390625, -0.07472610473632812, -1.1875457763671875, -0.344451904296875, 1.93890380859375, -0.26996612548828125, -0.06349372863769531, 0.7856063842773438, -0.1407012939453125, 0.604827880859375, 0.27951812744140625, -1.517578125, 1.15789794921875, -0.16812896728515625, -0.10577774047851562, -1.7552490234375, -0.4417724609375, -0.640625, 0.10943603515625, 0.526336669921875, -1.7530517578125, -0.8008575439453125, 0.523895263671875, -1.0640869140625, 0.10028076171875, -0.5177459716796875, 0.05712890625, -0.260406494140625, 0.029582977294921875, 0.5090484619140625, -0.561614990234375, -0.39574432373046875, 0.3841094970703125, -0.48754119873046875, 0.0, 0.49969482421875, 3.2412109375, 0.16857147216796875, 0.3011817932128906, 1.3135986328125, 0.7189788818359375, -0.3577117919921875, 0.5291061401367188, -0.34759521484375, 0.485565185546875, 0.719451904296875, 1.364532470703125, 1.834625244140625, 0.26953125, -0.51727294921875, -0.8643341064453125, 2.9222412109375, -2.11474609375, 0.18708038330078125, -0.34234619140625, -0.027126312255859375, 0.506866455078125, 0.31695556640625, -0.052242279052734375, 0.541290283203125, 1.1519012451171875, 1.0146026611328125, 0.187957763671875, 0.0985565185546875, -0.5016899108886719, -0.3407478332519531, -0.9510498046875, -0.5270614624023438, -1.921661376953125, -0.0341796875, -0.0625457763671875, 1.543487548828125, -0.24462127685546875, 0.45697021484375, 1.05419921875, -2.342132568359375, -0.587738037109375, 1.866790771484375, 0.05548095703125, -0.9395828247070312, 0.13068389892578125, -0.12847137451171875, 0.5657958984375, 0.79534912109375, 0.31359100341796875, -0.20092010498046875, 0.143768310546875, -0.7678070068359375, -0.054962158203125, -0.6866455078125, -0.22949981689453125, 1.021575927734375, 1.558624267578125, -0.3458251953125, 0.5068511962890625, 0.44219970703125, 1.0718994140625, 0.3372383117675781, 2.07000732421875, -0.4786376953125, -2.066162109375, 0.025629043579101562, 0.871795654296875, 0.490081787109375, 1.4412841796875, -1.078582763671875, -0.020181655883789062, 0.05843353271484375, -0.1544647216796875, -0.97760009765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000020.npy"}
|
|
{"epoch": 0.041884816753926704, "step": 21, "batch_size": 128, "mean": -0.006047368049621582, "std": 1.0009336471557617, "min": -3.1141357421875, "p10": -1.042205810546875, "median": -0.0186309814453125, "p90": 1.1772575378417969, "max": 2.8883056640625, "pos_frac": 0.484375, "sample": [-0.745452880859375, 1.1714553833007812, -0.6147003173828125, -1.8376922607421875, -1.05609130859375, -0.264801025390625, -0.0555419921875, -0.650970458984375, -0.01483154296875, -0.7274169921875, 0.7073974609375, -0.04296875, 0.346099853515625, -2.3435516357421875, -2.419891357421875, -0.2604522705078125, 0.89154052734375, 1.482025146484375, -2.34906005859375, 0.06909942626953125, 2.8883056640625, 0.3716468811035156, -0.9931640625, 0.01446533203125, 0.34967041015625, 0.3668670654296875, -0.844482421875, 2.449432373046875, -0.8657455444335938, -0.4723663330078125, -0.0875396728515625, 0.85980224609375, -0.92059326171875, -0.04443359375, 0.40869140625, -0.2518768310546875, 0.2838134765625, -0.667694091796875, -2.224517822265625, -0.022430419921875, -0.8977127075195312, -0.331756591796875, 0.4212646484375, -0.4442558288574219, 0.10810089111328125, 1.0396728515625, -0.3209095001220703, -0.067108154296875, 0.34365081787109375, -0.304443359375, -1.428009033203125, 0.8191604614257812, 0.0, -0.2773590087890625, 1.208526611328125, 0.17629241943359375, -1.65863037109375, -0.325714111328125, -0.9694347381591797, 0.471771240234375, -1.51434326171875, 0.624725341796875, 0.009586334228515625, 1.37847900390625, 0.57958984375, 0.8649749755859375, 0.2353515625, -0.768951416015625, 0.233734130859375, 2.62640380859375, -0.06500244140625, 1.0045318603515625, -0.39520263671875, -0.02557373046875, 0.03685760498046875, 1.48779296875, -1.864837646484375, 1.41241455078125, -0.999267578125, -0.722991943359375, -0.5366439819335938, -0.18474578857421875, 1.612548828125, -0.767852783203125, 0.75531005859375, -0.055908203125, 0.07734107971191406, -0.462249755859375, 0.5727081298828125, -0.4926948547363281, -1.0362548828125, 0.61309814453125, 0.7161865234375, 0.11373710632324219, 0.665618896484375, 0.2734565734863281, -0.0359344482421875, -2.22967529296875, 2.166259765625, 0.522369384765625, -0.48443603515625, -0.298492431640625, 1.25244140625, 0.148101806640625, -0.771728515625, 0.641998291015625, -0.1985931396484375, 0.2667083740234375, -0.35648345947265625, 0.7892913818359375, -0.2339019775390625, 0.997589111328125, 0.965118408203125, 1.288604736328125, 0.8227615356445312, 0.025360107421875, 0.5814208984375, 1.002471923828125, 0.549560546875, 1.1907958984375, -3.1141357421875, -1.239349365234375, 0.5484466552734375, -0.27655029296875, -0.5749359130859375, -0.246307373046875, 0.16479110717773438, -0.08470916748046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000021.npy"}
|
|
{"epoch": 0.04397905759162304, "step": 22, "batch_size": 128, "mean": -0.0037176012992858887, "std": 0.8416224718093872, "min": -2.2628173828125, "p10": -1.0149452209472656, "median": -0.006999969482421875, "p90": 0.9498367309570311, "max": 3.05059814453125, "pos_frac": 0.4921875, "sample": [2.878997802734375, 0.3966636657714844, 0.22662353515625, -1.4447021484375, 0.965057373046875, 0.20458984375, 0.1281890869140625, 0.1381072998046875, -1.12469482421875, 0.510833740234375, 0.07794189453125, 0.0460357666015625, -0.64483642578125, -0.30986976623535156, -2.2628173828125, -0.114288330078125, -1.237518310546875, -1.071563720703125, -0.6624755859375, 0.1154937744140625, 0.756317138671875, 0.33946990966796875, -0.67388916015625, -0.2199859619140625, -0.601654052734375, 0.468536376953125, 0.65582275390625, 0.15838623046875, 0.2724609375, 0.26798248291015625, 0.75927734375, 0.175048828125, 0.0, 0.61004638671875, -1.0006332397460938, -0.985565185546875, -0.189208984375, -0.7506561279296875, 0.9433135986328125, -0.51025390625, -0.6044921875, 0.073272705078125, -0.51947021484375, 1.760498046875, 0.295745849609375, 0.1957855224609375, -1.975341796875, -1.13055419921875, 1.4451904296875, 1.2959747314453125, -0.056247711181640625, 0.3930320739746094, -1.37005615234375, 0.329376220703125, 0.10131072998046875, -0.01399993896484375, -0.028942108154296875, 1.04266357421875, 1.4193115234375, 2.792236328125, -0.544586181640625, -1.10455322265625, -0.04095458984375, -0.7905349731445312, -0.11204910278320312, -0.4867706298828125, 1.38519287109375, -0.9794921875, 0.5093536376953125, 0.318939208984375, -0.10532379150390625, -0.77911376953125, 0.03631591796875, -1.37274169921875, -0.407012939453125, 0.221099853515625, -0.1289215087890625, -0.41551971435546875, 0.04022216796875, -0.372802734375, -0.55633544921875, -0.3564453125, 0.278564453125, 0.0828399658203125, -0.2484893798828125, 0.479705810546875, 1.6243438720703125, -0.030059814453125, -1.04833984375, 0.049468994140625, 3.05059814453125, -0.6606063842773438, -0.34400177001953125, 0.13348388671875, -0.2566375732421875, 0.02846527099609375, -0.74261474609375, -0.45196533203125, -0.0430908203125, 0.6412353515625, -1.509246826171875, -0.505279541015625, 1.30340576171875, 1.10565185546875, 0.103515625, 0.1723480224609375, -0.7058258056640625, 0.6700286865234375, -1.169219970703125, 0.2586669921875, 0.5096931457519531, 0.552032470703125, -0.09610748291015625, -0.32971954345703125, -0.19446754455566406, -0.803466796875, 0.37652587890625, -0.026092529296875, 0.20623779296875, 0.941925048828125, -0.21478271484375, -0.083587646484375, -0.3958282470703125, -0.6361846923828125, 0.026660919189453125, -0.092071533203125, 0.09633636474609375, 0.72625732421875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000022.npy"}
|
|
{"epoch": 0.04607329842931937, "step": 23, "batch_size": 128, "mean": 0.08333773910999298, "std": 0.8560036420822144, "min": -2.29888916015625, "p10": -0.9226509094238281, "median": 0.03804779052734375, "p90": 1.0553703308105467, "max": 2.6195068359375, "pos_frac": 0.53125, "sample": [-0.40844154357910156, 0.3544921875, -0.3882598876953125, 0.23822021484375, 2.031402587890625, 1.94384765625, 2.5816650390625, -0.2805023193359375, 0.1945037841796875, -0.365234375, 1.352294921875, 0.463043212890625, -0.00960540771484375, -1.200775146484375, -0.30987548828125, 0.17302322387695312, 0.03961181640625, 0.4174041748046875, -0.15830230712890625, -0.5798492431640625, 0.40845489501953125, -0.9771728515625, -0.44091796875, 0.6192626953125, 0.30893707275390625, -0.08481597900390625, 0.2434539794921875, 0.25042724609375, -0.1278076171875, -0.27440643310546875, 0.0, 1.1122207641601562, -0.63812255859375, 0.30211639404296875, -2.29888916015625, 0.6360321044921875, 0.02788543701171875, -1.055877685546875, -0.5474853515625, -1.66534423828125, -0.271026611328125, -2.14666748046875, 0.37451171875, -1.0092086791992188, 1.521026611328125, -0.534576416015625, 0.4955902099609375, -0.6576080322265625, -0.353271484375, 0.6611404418945312, -0.11547088623046875, -0.0007686614990234375, -0.8398284912109375, -0.5415191650390625, 0.66400146484375, -0.185943603515625, 0.0364837646484375, -0.6384963989257812, 0.2475738525390625, 0.3758697509765625, -1.782440185546875, 0.08812713623046875, 0.4945068359375, -0.099334716796875, 1.44647216796875, -0.1573486328125, 0.65155029296875, 2.6195068359375, 0.601287841796875, -0.00091552734375, 0.8225326538085938, 0.40132904052734375, -0.971405029296875, 1.031005859375, 0.414276123046875, -0.4655609130859375, 0.530609130859375, 0.30803489685058594, 2.174835205078125, 0.7937469482421875, -0.08710479736328125, -0.0044002532958984375, 0.3462333679199219, 0.5750732421875, 1.463531494140625, 0.71197509765625, -0.015869140625, 0.8093109130859375, 0.88702392578125, -1.55572509765625, -0.37871551513671875, -0.4111614227294922, 0.78228759765625, -0.060665130615234375, -0.9017562866210938, -1.137939453125, 0.554412841796875, 0.64605712890625, -0.3957672119140625, -0.59619140625, 1.3258056640625, 0.32631683349609375, 0.1545257568359375, -0.3853607177734375, 0.1768035888671875, -0.12628173828125, 0.47412109375, 0.41153717041015625, 0.0363006591796875, 1.854583740234375, 0.27999114990234375, -0.541015625, 0.16434860229492188, -0.3846435546875, -0.267578125, -0.601806640625, 1.27947998046875, 0.20623779296875, 0.690887451171875, -1.9892578125, 0.531463623046875, -1.1168289184570312, 0.0202789306640625, 0.818634033203125, -0.482147216796875, 0.052581787109375, -0.5294609069824219, -0.8121376037597656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000023.npy"}
|
|
{"epoch": 0.048167539267015703, "step": 24, "batch_size": 128, "mean": 0.05592915415763855, "std": 0.8616992235183716, "min": -2.07025146484375, "p10": -1.1006805419921875, "median": 0.0611419677734375, "p90": 1.0408927917480468, "max": 2.568359375, "pos_frac": 0.53125, "sample": [0.6997909545898438, -0.5924224853515625, -0.16935348510742188, 1.269561767578125, -0.84564208984375, 0.255859375, -0.049774169921875, 2.15936279296875, -0.13442230224609375, 0.347747802734375, 0.6746978759765625, -0.14727783203125, 0.5677490234375, -0.5460700988769531, 0.0344390869140625, -1.89825439453125, 0.167022705078125, 0.78094482421875, 0.161407470703125, 0.5260028839111328, 0.7620849609375, -0.263153076171875, -0.877899169921875, 0.0, -0.1557464599609375, 0.11017608642578125, -1.470306396484375, -0.53704833984375, 0.43221282958984375, -0.666107177734375, 0.7750244140625, 0.09120941162109375, 0.7047119140625, 1.54486083984375, 0.153076171875, -0.51837158203125, -2.07025146484375, -1.09796142578125, -0.4914398193359375, 0.1646575927734375, -0.3239288330078125, -0.18470001220703125, 0.364044189453125, 0.435211181640625, 0.348663330078125, -0.588134765625, 0.0, 0.834930419921875, -0.95343017578125, 0.7896194458007812, 0.5098876953125, 0.24263763427734375, 0.071441650390625, 0.6521453857421875, 0.8363037109375, 0.23532867431640625, -0.647064208984375, 0.185882568359375, 1.100830078125, -0.026763916015625, -0.536224365234375, 0.0595245361328125, 0.8861083984375, 0.10569000244140625, 0.954833984375, 0.43216705322265625, 1.0377273559570312, -0.15087127685546875, 0.1436920166015625, -1.134002685546875, 1.027984619140625, -0.26202392578125, -0.5411605834960938, -1.19366455078125, -0.7385711669921875, -0.259979248046875, 0.738525390625, -1.1729888916015625, 1.656707763671875, 0.15966033935546875, -0.624053955078125, 0.3047027587890625, -0.752685546875, -0.8443145751953125, 0.0119171142578125, -1.24993896484375, 1.2755126953125, 1.1695556640625, -1.107025146484375, 0.6465530395507812, 0.2657470703125, -0.4677734375, -0.464630126953125, -0.887115478515625, -0.1175079345703125, -0.06314277648925781, -1.287628173828125, -0.512298583984375, -0.479278564453125, 0.793060302734375, 2.483154296875, 0.791015625, 0.8719024658203125, 2.46820068359375, 0.02728271484375, -1.8526611328125, 1.04827880859375, 1.308685302734375, 0.570709228515625, -0.42657470703125, 1.6456298828125, -0.6519622802734375, 0.0627593994140625, 0.236663818359375, -0.17736053466796875, -0.093536376953125, -0.8018798828125, -0.18662261962890625, -1.204864501953125, 0.262451171875, 2.568359375, -0.373565673828125, 0.3891563415527344, -1.341461181640625, -1.12445068359375, 0.5971221923828125, -0.6181640625, 0.123870849609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000024.npy"}
|
|
{"epoch": 0.050261780104712044, "step": 25, "batch_size": 128, "mean": 0.07499483227729797, "std": 0.8208662867546082, "min": -2.948822021484375, "p10": -0.9500244140625, "median": 0.12056446075439453, "p90": 1.1824737548828124, "max": 2.109832763671875, "pos_frac": 0.546875, "sample": [0.2930145263671875, -0.023956298828125, -0.593475341796875, 0.751312255859375, -0.022216796875, -0.33984375, 0.702880859375, -0.0159149169921875, 1.683349609375, -1.851776123046875, 0.5107421875, -0.511444091796875, 0.4005584716796875, -0.209442138671875, 0.39007568359375, -0.95404052734375, -1.1014022827148438, -0.194732666015625, -0.765869140625, -0.1663665771484375, -0.7069168090820312, -0.4275035858154297, 0.5679397583007812, 0.114532470703125, -0.316741943359375, -0.6652679443359375, -0.835174560546875, -0.3388519287109375, 0.4081878662109375, 1.34002685546875, 0.3671398162841797, 0.1221466064453125, 0.0, -0.474334716796875, 0.718292236328125, 0.11898231506347656, 0.21355438232421875, -0.1018218994140625, -0.273651123046875, -0.350311279296875, -1.394378662109375, 0.0, 0.6318359375, 1.1739501953125, 0.7036895751953125, -1.6195526123046875, -0.04637908935546875, -0.218017578125, -0.05255126953125, -0.95404052734375, -0.4221687316894531, 1.202362060546875, 1.54949951171875, 0.08089828491210938, 0.868865966796875, -0.0436248779296875, -0.6618576049804688, -0.0404510498046875, 1.0357818603515625, -0.830108642578125, 1.0727310180664062, 0.4938812255859375, -0.69158935546875, -1.68988037109375, -0.5604248046875, 0.3727607727050781, 0.5392913818359375, -1.1107330322265625, -1.75048828125, 2.109832763671875, 1.342529296875, 0.25560760498046875, 0.03275299072265625, 0.2918701171875, 0.3768310546875, 0.126800537109375, -0.754150390625, -0.25011444091796875, 1.3706512451171875, 0.049835205078125, -1.143280029296875, -0.01287841796875, 0.6367645263671875, 1.300537109375, -0.77252197265625, 1.829559326171875, 0.24016761779785156, 0.18577957153320312, 0.652252197265625, 0.132232666015625, 1.698974609375, -0.0757904052734375, 0.492523193359375, 0.6497802734375, 0.619354248046875, 1.29058837890625, 0.5671768188476562, 1.3629608154296875, 1.295013427734375, 0.6287841796875, 0.26141357421875, -0.8751983642578125, 1.011322021484375, -0.09348678588867188, 0.54833984375, 0.81402587890625, -0.842529296875, -0.362884521484375, -1.0029144287109375, -0.94830322265625, -0.9709625244140625, 0.5140762329101562, 0.16375732421875, 0.7966728210449219, 0.145416259765625, 0.7097625732421875, -0.512176513671875, 0.3714027404785156, 0.2722320556640625, 0.74212646484375, 0.18464279174804688, 0.21451377868652344, -0.73748779296875, 0.0387420654296875, -2.948822021484375, 0.45501708984375, 0.154205322265625, -0.13496971130371094], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000025.npy"}
|
|
{"epoch": 0.05235602094240838, "step": 26, "batch_size": 128, "mean": -0.037623971700668335, "std": 0.9261139631271362, "min": -3.37451171875, "p10": -1.0123321533203125, "median": -0.06315231323242188, "p90": 0.929408264160156, "max": 3.39935302734375, "pos_frac": 0.453125, "sample": [-1.71319580078125, 0.012639999389648438, 3.39935302734375, -0.30181884765625, -0.576324462890625, 0.029949188232421875, 0.5736980438232422, 0.8602409362792969, 0.012054443359375, 0.0704193115234375, -0.3481178283691406, -0.27655029296875, 0.0, 0.3958892822265625, -0.571502685546875, 0.412139892578125, 0.169097900390625, 0.01708984375, 2.090667724609375, 0.6211395263671875, 1.1356201171875, 0.4230194091796875, -0.06804656982421875, 0.651763916015625, -0.171051025390625, 0.7121620178222656, -0.61297607421875, -0.8719482421875, 0.0199127197265625, -0.40840911865234375, -0.10823822021484375, -1.132720947265625, 0.28762054443359375, -0.242095947265625, -0.28369140625, -0.24114990234375, 0.18316650390625, -1.2751922607421875, -0.99212646484375, -1.44586181640625, 1.14520263671875, -0.2645416259765625, -0.3909454345703125, -0.487823486328125, 0.776092529296875, -2.70550537109375, -0.21938323974609375, 1.193817138671875, -0.4615631103515625, -0.00833892822265625, -0.129547119140625, -0.269073486328125, -0.3763275146484375, -3.37451171875, -0.8509521484375, -0.058258056640625, 0.1043853759765625, -0.5229339599609375, 0.534454345703125, -1.2301025390625, 0.6025505065917969, -0.3961181640625, -0.389678955078125, 0.6569061279296875, -2.4748382568359375, -0.32143402099609375, -1.20367431640625, 0.77783203125, 3.1895751953125, 0.3734588623046875, 0.185882568359375, -0.24204444885253906, -0.8835906982421875, 1.08056640625, -0.25502777099609375, -0.21446990966796875, 0.04669189453125, 0.07126998901367188, -0.99188232421875, 0.38127899169921875, -0.03912353515625, 1.441009521484375, 0.10445022583007812, 0.2904052734375, 1.781982421875, 0.28587913513183594, 0.462066650390625, 0.010711669921875, -0.30135345458984375, -0.18780517578125, 0.410186767578125, 0.01715087890625, -0.07989120483398438, -0.86395263671875, 0.46026611328125, -1.125274658203125, -0.14117431640625, -0.0028171539306640625, -0.7141513824462891, -0.102691650390625, 0.16222381591796875, -0.5010643005371094, 1.499237060546875, 0.1912689208984375, 0.4690704345703125, -0.745391845703125, -1.07421875, 0.0, -2.28204345703125, -0.4843025207519531, 0.339569091796875, -0.47168731689453125, -0.182373046875, -0.7288970947265625, 0.809295654296875, 0.993804931640625, 0.9018096923828125, 0.732330322265625, 0.831207275390625, 0.461456298828125, 1.611663818359375, 1.20330810546875, -0.35276031494140625, -0.47479248046875, -0.287628173828125, -0.3312530517578125, -1.059478759765625, -0.58612060546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000026.npy"}
|
|
{"epoch": 0.05445026178010471, "step": 27, "batch_size": 128, "mean": 0.11403858661651611, "std": 1.0490643978118896, "min": -3.24798583984375, "p10": -1.1729461669921875, "median": 0.02846527099609375, "p90": 1.586032104492187, "max": 3.244049072265625, "pos_frac": 0.5234375, "sample": [1.788665771484375, 0.32037353515625, -0.3667869567871094, -0.18857955932617188, 0.3411712646484375, -1.3160400390625, -0.243255615234375, -0.8265533447265625, 1.002685546875, 0.0106353759765625, -0.8000335693359375, 0.6641845703125, 0.8669891357421875, -0.553131103515625, 0.60040283203125, 1.76214599609375, 0.10968017578125, 0.01760101318359375, 0.868194580078125, 3.15399169921875, -0.24135589599609375, -0.08042144775390625, 1.29034423828125, -0.148284912109375, 2.253570556640625, 3.244049072265625, 0.4786643981933594, 2.26800537109375, -0.6788177490234375, 0.66632080078125, -1.22308349609375, 0.164031982421875, 0.4270782470703125, -0.20941162109375, 1.103363037109375, -0.654541015625, -0.3011322021484375, 1.2117919921875, -0.2609710693359375, -0.3666229248046875, 1.7266845703125, 1.98419189453125, -3.24798583984375, -0.3124237060546875, 1.544769287109375, 0.3382568359375, -0.24071502685546875, -1.293731689453125, 1.9552001953125, 0.3162384033203125, -0.86944580078125, 1.09844970703125, -0.6089248657226562, -0.32418060302734375, 2.14910888671875, -0.1246337890625, 0.410552978515625, -0.6403350830078125, 0.050079345703125, 1.849853515625, -0.272552490234375, -1.28619384765625, -1.1214447021484375, -1.151458740234375, 0.0279541015625, -0.356964111328125, 0.9019317626953125, 0.3822898864746094, 2.1876220703125, -1.226776123046875, -1.68017578125, -1.255157470703125, 0.315826416015625, -0.9439697265625, 0.889984130859375, 0.9715042114257812, 0.5845947265625, 0.59307861328125, 1.379547119140625, 0.15478515625, -0.30319976806640625, -1.71337890625, 0.338287353515625, -0.6485595703125, -1.7398223876953125, 1.68231201171875, 1.163543701171875, -0.15230560302734375, -0.06739044189453125, 0.37552642822265625, -0.2223663330078125, -1.0294647216796875, -2.28857421875, 0.038818359375, -0.829803466796875, -0.189849853515625, -1.823760986328125, -0.21555328369140625, 0.098358154296875, -0.155517578125, 0.1821441650390625, -0.10682487487792969, -0.8053817749023438, 0.85308837890625, 0.375030517578125, 0.499755859375, 0.439178466796875, -0.37497711181640625, 0.05487060546875, 0.262908935546875, 0.9970703125, 0.4797325134277344, 0.0289764404296875, -0.151397705078125, -0.2371978759765625, 0.7264251708984375, -0.46643829345703125, -0.0463714599609375, -0.68792724609375, 0.669952392578125, 0.97821044921875, 0.364715576171875, 0.0950927734375, -1.631256103515625, -0.909393310546875, 0.568511962890625, -0.7667236328125, -0.12248802185058594], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000027.npy"}
|
|
{"epoch": 0.05654450261780105, "step": 28, "batch_size": 128, "mean": 0.07920265197753906, "std": 0.8954238891601562, "min": -2.9326171875, "p10": -0.9409347534179686, "median": 0.05828571319580078, "p90": 1.2118370056152343, "max": 2.467864990234375, "pos_frac": 0.515625, "sample": [0.468353271484375, -0.2779541015625, -0.843353271484375, -0.75567626953125, -0.3488731384277344, -0.4907264709472656, -0.4075927734375, 0.8937225341796875, -0.027902603149414062, 0.70068359375, 0.3709449768066406, -0.251617431640625, 0.720001220703125, -0.21797943115234375, 0.7309722900390625, 0.3109130859375, 0.2375946044921875, -0.09955596923828125, 1.4201202392578125, 1.2264862060546875, 0.2545642852783203, 0.0339813232421875, -0.1585845947265625, 0.082916259765625, 0.10330963134765625, -0.366607666015625, -1.1273193359375, 0.08730506896972656, 0.118194580078125, -1.1439208984375, 0.837066650390625, 0.322998046875, -0.5042724609375, 1.6754150390625, 0.3130836486816406, -1.35369873046875, 0.2127685546875, -0.0768280029296875, 1.703033447265625, 0.2329864501953125, -0.3884735107421875, -1.7091064453125, -0.745513916015625, -0.91375732421875, 0.7819061279296875, 2.07232666015625, 0.703216552734375, 0.0245361328125, -0.47662353515625, -0.476654052734375, 1.286468505859375, 1.2960052490234375, 0.793792724609375, -1.6541748046875, 0.557464599609375, -0.2071533203125, -0.1127471923828125, 1.586944580078125, 0.891082763671875, 0.83721923828125, -0.497650146484375, 0.42269134521484375, -0.4335784912109375, -0.16184616088867188, 0.9033203125, 0.7774200439453125, -0.5211563110351562, 0.3064117431640625, -0.646514892578125, 0.1230926513671875, -0.4664764404296875, -1.588409423828125, -0.243621826171875, -0.5607070922851562, -0.346527099609375, 0.5867919921875, -2.9326171875, -0.2530937194824219, -0.1079864501953125, 0.9500732421875, 1.2055587768554688, 0.1420440673828125, 0.588104248046875, -0.90618896484375, -1.01080322265625, 0.13019561767578125, 0.2347869873046875, -0.3453826904296875, -0.26476287841796875, 0.22254180908203125, -0.4942474365234375, -0.1053466796875, -0.1288604736328125, -0.831512451171875, -0.3319873809814453, -0.033885955810546875, 2.467864990234375, 0.47998809814453125, 2.168304443359375, -0.1670379638671875, -2.2149658203125, -0.2415008544921875, 0.6630859375, 1.1731719970703125, -0.744293212890625, 0.40087890625, 0.7809219360351562, -0.1942596435546875, 0.993743896484375, 0.09694671630859375, 1.60772705078125, 0.6053924560546875, -1.2686767578125, -0.04947471618652344, -0.541107177734375, -0.06795501708984375, 1.3309478759765625, 0.5311737060546875, 1.13922119140625, 0.084564208984375, -1.0043487548828125, -1.85980224609375, 0.54071044921875, 0.21215057373046875, 2.39971923828125, 0.08259010314941406, 0.0, -1.399322509765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000028.npy"}
|
|
{"epoch": 0.05863874345549738, "step": 29, "batch_size": 128, "mean": 0.07106417417526245, "std": 0.9334477186203003, "min": -2.0072021484375, "p10": -1.1374420166015624, "median": 0.06760597229003906, "p90": 1.1786338806152343, "max": 4.74200439453125, "pos_frac": 0.546875, "sample": [-0.08006477355957031, -2.0072021484375, 0.87652587890625, -0.29693603515625, -1.67205810546875, -1.08746337890625, 0.33612060546875, 4.74200439453125, -1.708770751953125, 0.640289306640625, 0.828521728515625, 1.087738037109375, 1.062957763671875, 0.3226776123046875, 0.42504119873046875, 0.0, -1.840362548828125, 0.12158203125, 0.06732177734375, 0.11724090576171875, 0.317108154296875, 2.541656494140625, 0.5103607177734375, -1.62310791015625, -0.159271240234375, 0.700927734375, 1.100128173828125, 0.45343780517578125, -0.174407958984375, 0.6720123291015625, 0.083648681640625, -1.641876220703125, 0.6162872314453125, 0.9840087890625, -1.312744140625, -1.119384765625, 0.29254150390625, -1.209869384765625, 1.23712158203125, 0.0015716552734375, -0.884613037109375, 0.048740386962890625, 0.1185302734375, 0.30010986328125, 1.549774169921875, -0.11357879638671875, -0.42528533935546875, -0.60247802734375, 1.171051025390625, -0.4604644775390625, 1.35107421875, -0.8360595703125, 0.6177978515625, -0.783935546875, 0.55487060546875, 1.548980712890625, 0.8779296875, -0.0835723876953125, -0.273468017578125, 0.39350128173828125, -1.126556396484375, -1.2281494140625, -0.57598876953125, -0.027984619140625, 0.2812347412109375, 0.2894287109375, -1.117919921875, -0.1561737060546875, 0.0, -0.37947845458984375, 1.2506103515625, -0.0885009765625, 0.4295501708984375, -1.02301025390625, -0.01964569091796875, -0.865753173828125, 0.342437744140625, 0.030733108520507812, -1.050048828125, 0.39288330078125, -0.156585693359375, -0.32094573974609375, -0.19793701171875, 1.1963272094726562, 0.7620697021484375, 0.162322998046875, 1.2085723876953125, -0.4165802001953125, -0.273712158203125, 0.520294189453125, 0.5131988525390625, 1.40087890625, 0.44722747802734375, -0.410491943359375, 0.586151123046875, -0.3171234130859375, -0.002960205078125, 0.8376312255859375, 0.3076171875, 0.0908660888671875, -0.9349517822265625, 0.18686676025390625, 1.4289398193359375, -1.1731491088867188, 0.463043212890625, 1.2456130981445312, -0.5837554931640625, -0.18560791015625, -1.42340087890625, -1.195220947265625, -0.124053955078125, -0.2359771728515625, 1.056488037109375, -0.355194091796875, 1.69287109375, -1.162841796875, 0.8057098388671875, -0.185394287109375, 0.02386474609375, 0.18780517578125, 0.06789016723632812, 0.00555419921875, 0.09466552734375, -0.7911224365234375, 0.071441650390625, 1.061492919921875, -0.647369384765625, 0.13330078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000029.npy"}
|
|
{"epoch": 0.060732984293193716, "step": 30, "batch_size": 128, "mean": 0.007129386067390442, "std": 0.9013113975524902, "min": -3.07562255859375, "p10": -1.0052413940429688, "median": 0.020862579345703125, "p90": 1.0491874694824213, "max": 2.97430419921875, "pos_frac": 0.5078125, "sample": [0.357025146484375, 1.7570343017578125, -1.32568359375, -0.94635009765625, 0.3860206604003906, 0.818145751953125, 0.594512939453125, -0.2792816162109375, 0.4239501953125, 0.15473175048828125, -0.002227783203125, -1.0302886962890625, 0.833099365234375, -0.846343994140625, 0.6671295166015625, -1.495208740234375, -0.277130126953125, 0.916290283203125, -0.6030044555664062, -1.1580963134765625, -2.765594482421875, 1.67266845703125, -0.9848480224609375, 0.25948333740234375, 0.1041412353515625, -1.1431884765625, 0.38995361328125, 0.20752334594726562, -0.2989311218261719, 1.241851806640625, 1.7389373779296875, -0.19007110595703125, 0.353271484375, 0.574188232421875, -0.14045333862304688, -0.12274169921875, 0.8856964111328125, -0.4873542785644531, 0.2244415283203125, -0.454345703125, -0.736419677734375, 0.9083328247070312, 0.366455078125, 0.2901611328125, -0.51092529296875, 1.186431884765625, -0.63421630859375, -0.09521484375, 0.6866607666015625, 0.09478759765625, 1.179962158203125, 0.6312789916992188, 0.626495361328125, 0.0618133544921875, 0.796356201171875, -0.18846893310546875, -1.6875, 0.9023284912109375, -0.67376708984375, -0.4664764404296875, 0.7098388671875, -1.160430908203125, 1.8295440673828125, -0.9945068359375, -0.1294097900390625, 0.214569091796875, -1.184295654296875, -0.44134521484375, -0.16989517211914062, 0.414215087890625, 0.10814666748046875, -0.188323974609375, -0.4031562805175781, 0.14879608154296875, -0.96990966796875, 2.450439453125, -0.20188140869140625, -0.309417724609375, 0.271148681640625, -1.3722686767578125, 0.9931411743164062, 2.97430419921875, -3.07562255859375, 1.248016357421875, -0.04651451110839844, 0.305328369140625, 2.03704833984375, -0.5302581787109375, 0.8116302490234375, -0.5615997314453125, 0.03022003173828125, 0.011505126953125, 0.4414825439453125, -2.09539794921875, 0.033966064453125, 0.19842529296875, 0.3371734619140625, 0.1436786651611328, 0.0677490234375, 0.56488037109375, -0.47772216796875, 0.0336761474609375, 0.1142120361328125, -0.13559722900390625, 1.20013427734375, -0.40814208984375, -0.33892822265625, -0.299957275390625, -0.1213836669921875, -1.48309326171875, -0.510009765625, -0.365509033203125, 1.3587646484375, -0.1729736328125, -0.1798572540283203, 0.03119659423828125, 0.331878662109375, 0.3587493896484375, -0.8266658782958984, -0.842803955078125, -0.38507080078125, -0.274749755859375, 0.2199554443359375, -0.761932373046875, -0.19490623474121094, -0.16515350341796875, -0.346221923828125, 0.296630859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000030.npy"}
|
|
{"epoch": 0.06282722513089005, "step": 31, "batch_size": 128, "mean": 0.13683582842350006, "std": 0.8606264591217041, "min": -1.7993621826171875, "p10": -1.0370597839355469, "median": 0.14349365234375, "p90": 1.2359100341796874, "max": 2.9486083984375, "pos_frac": 0.5703125, "sample": [-0.7267608642578125, -0.57293701171875, -1.2298583984375, 0.932830810546875, 0.41583251953125, 0.82867431640625, 0.37469482421875, 1.20501708984375, 0.28852081298828125, 0.71124267578125, 2.15106201171875, -0.48471832275390625, 0.26995849609375, -1.41912841796875, -0.191192626953125, -1.672760009765625, 1.3398628234863281, -0.06179046630859375, 0.429718017578125, -0.6877288818359375, 0.97491455078125, 0.4053955078125, 0.300384521484375, 1.36212158203125, 0.3861083984375, 1.40118408203125, 0.0414581298828125, 1.768951416015625, -0.516448974609375, 0.46167755126953125, 0.14666748046875, 1.317352294921875, -0.364990234375, -0.06259536743164062, -0.873077392578125, 1.213470458984375, 0.3136749267578125, 0.342926025390625, -1.3921051025390625, 0.0, 0.651092529296875, 0.3159637451171875, 0.0, 1.15496826171875, 0.7893142700195312, -0.3706207275390625, 0.834228515625, 0.6520500183105469, -0.427459716796875, -0.2725830078125, -0.908843994140625, 0.0566864013671875, -0.757232666015625, 0.567352294921875, 0.08688735961914062, -0.072021484375, -0.25634765625, 2.823272705078125, -0.167327880859375, -1.0899734497070312, 0.0, -0.03515625, -1.23876953125, -1.171783447265625, 0.76007080078125, -0.1245574951171875, 0.418365478515625, -1.5955657958984375, -1.19091796875, 0.16130447387695312, -1.7993621826171875, -1.47052001953125, 0.3041877746582031, 0.5761032104492188, 0.24676513671875, 0.2359161376953125, -0.8114013671875, -0.2437896728515625, 1.19451904296875, 0.5374603271484375, 0.08826828002929688, -0.4474334716796875, -1.0399169921875, -0.14519691467285156, 0.44838714599609375, 0.0532989501953125, 1.1534423828125, -0.29693603515625, -0.4434356689453125, 1.94866943359375, -1.0358352661132812, -0.1805877685546875, -0.494110107421875, 1.366790771484375, 0.076751708984375, -0.87213134765625, 0.480621337890625, 1.4425201416015625, -0.2817535400390625, 0.0, 0.214080810546875, -0.38372802734375, 0.05419921875, 0.2839546203613281, -0.21916961669921875, 0.4976806640625, -0.6524505615234375, -0.6806240081787109, 0.459320068359375, -0.0126495361328125, 0.2501220703125, -0.43790435791015625, -0.088134765625, 2.9486083984375, 1.28826904296875, 0.544189453125, 0.14031982421875, 0.49884033203125, 1.70355224609375, 0.73504638671875, 0.35174560546875, 0.82440185546875, 0.262542724609375, 0.2052001953125, 0.5089969635009766, 0.09344482421875, -1.336334228515625, 0.15016937255859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000031.npy"}
|
|
{"epoch": 0.06492146596858639, "step": 32, "batch_size": 128, "mean": -0.034656181931495667, "std": 0.9526011347770691, "min": -3.0780029296875, "p10": -0.9811767578125, "median": -0.0135498046875, "p90": 1.1725479125976561, "max": 2.9312744140625, "pos_frac": 0.4921875, "sample": [-0.8465576171875, -0.05457878112792969, -2.526397705078125, -1.06884765625, 0.506317138671875, -0.8384246826171875, -0.4511451721191406, 1.2132110595703125, 0.4176177978515625, 1.14544677734375, -0.743438720703125, -0.26718902587890625, 1.478546142578125, 0.522247314453125, 0.1363983154296875, -0.8243045806884766, -0.972076416015625, 0.185546875, 1.54718017578125, -0.924041748046875, 0.4767608642578125, -0.888427734375, -0.6610260009765625, -0.1495819091796875, 1.3850250244140625, -0.42645263671875, 0.2656402587890625, -0.49038124084472656, 0.658782958984375, 0.5369873046875, 0.09947967529296875, -0.04444122314453125, -0.6976776123046875, -0.4556884765625, -0.375213623046875, 0.944305419921875, -1.132049560546875, 0.633270263671875, 0.461151123046875, -0.3022918701171875, 0.7650146484375, -0.93212890625, -1.006103515625, -0.0708465576171875, 0.6957244873046875, 0.55340576171875, 1.155120849609375, 0.0213470458984375, -0.04486083984375, 0.171661376953125, 1.46685791015625, 1.3262939453125, -2.1209716796875, -0.776763916015625, -1.23046875, -0.784332275390625, 2.9312744140625, 2.1622314453125, -0.06390380859375, -3.0780029296875, -0.0208587646484375, 0.083984375, 0.2253875732421875, -0.2725830078125, 0.038543701171875, -1.5814056396484375, -1.002410888671875, -0.8715667724609375, 0.2926025390625, 1.015594482421875, 2.518310546875, 0.19207763671875, 0.5577507019042969, -2.52490234375, 1.462890625, -0.491363525390625, -0.5839691162109375, -0.94720458984375, 0.5191879272460938, 1.267547607421875, 0.63232421875, -0.81011962890625, 0.037494659423828125, -0.36118125915527344, 1.071380615234375, 0.5512542724609375, 0.2604827880859375, 0.088165283203125, -0.491455078125, 0.007080078125, -0.0400543212890625, 0.0528106689453125, 0.2671356201171875, -0.2203369140625, 0.05548095703125, 0.78643798828125, -0.673736572265625, -0.965789794921875, 0.1202545166015625, -0.3819427490234375, -0.5210342407226562, -0.50640869140625, -0.569244384765625, -0.14934539794921875, 0.37841796875, -0.01279449462890625, -0.03801155090332031, 0.39322662353515625, -0.7529754638671875, -0.024204254150390625, 0.805419921875, -0.11458587646484375, 0.0662689208984375, -0.423919677734375, 0.71234130859375, -1.4697265625, -1.72491455078125, 0.075286865234375, 0.513214111328125, -2.2462310791015625, -0.4962615966796875, -0.01430511474609375, 0.1505126953125, 0.22674560546875, 0.49188232421875, 1.437530517578125, -0.810638427734375, 1.710235595703125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000032.npy"}
|
|
{"epoch": 0.06701570680628273, "step": 33, "batch_size": 128, "mean": 0.2310909926891327, "std": 0.9724738001823425, "min": -2.7220458984375, "p10": -0.9778404235839844, "median": 0.21956634521484375, "p90": 1.42735595703125, "max": 3.587158203125, "pos_frac": 0.625, "sample": [0.7323570251464844, 0.618408203125, 0.1598358154296875, 2.3688507080078125, -0.8363113403320312, 0.8074493408203125, -0.1506195068359375, -0.415924072265625, 1.821075439453125, 0.347412109375, 1.55621337890625, -0.7191696166992188, 1.89501953125, -2.7220458984375, 0.0, -0.625640869140625, 1.27197265625, -0.148956298828125, -1.4262619018554688, -0.96173095703125, 0.13519287109375, 0.29877471923828125, 0.10321044921875, 0.48175048828125, 1.8506011962890625, -0.77783203125, -0.3219757080078125, -0.4828338623046875, 1.400970458984375, 1.2139892578125, 0.49774169921875, -1.413818359375, 0.32562255859375, -2.195159912109375, -0.1651325225830078, 2.78167724609375, 0.8186187744140625, -0.18792724609375, -1.425628662109375, -0.51715087890625, 0.08258819580078125, -0.9704360961914062, -0.02263641357421875, 1.51239013671875, -0.9951171875, 0.3785247802734375, 1.04156494140625, -0.46954345703125, 0.3577117919921875, -0.8740386962890625, 0.9104766845703125, 0.5712432861328125, -0.24835205078125, 0.0, -0.526214599609375, 0.5272979736328125, 0.2506389617919922, 0.704437255859375, 0.9524383544921875, 0.01995849609375, -0.446197509765625, 0.232666015625, 0.5482749938964844, -0.27680206298828125, 0.3687458038330078, 0.2969970703125, -0.10116767883300781, 0.33039093017578125, -1.583251953125, 1.0663604736328125, 0.64630126953125, 0.47206878662109375, -0.1737060546875, 0.68255615234375, 0.18316650390625, -0.22589111328125, -1.2922515869140625, 1.97784423828125, 0.143463134765625, -0.339599609375, -1.028564453125, 0.5255126953125, 2.2176513671875, -1.0285110473632812, -0.028167724609375, 0.100677490234375, 0.555023193359375, -1.0938720703125, 0.594146728515625, 1.4430999755859375, -0.2796630859375, 0.16393280029296875, 0.41757965087890625, 0.09743499755859375, 0.3281898498535156, 0.12322998046875, 0.4841156005859375, 0.8133544921875, -1.962890625, 0.72119140625, 0.1407470703125, 0.27672576904296875, 1.4206085205078125, 2.1246337890625, 0.6049880981445312, -0.29265594482421875, -0.260101318359375, 0.60809326171875, 1.1128997802734375, -1.558197021484375, 1.76092529296875, 0.2064666748046875, -0.5264892578125, 1.39349365234375, 0.7480087280273438, -0.484832763671875, 0.2025909423828125, 0.4744110107421875, 3.587158203125, 0.9494171142578125, 0.390625, 0.7490730285644531, 0.74298095703125, -0.0541229248046875, -0.29278564453125, 0.13488388061523438, 0.035175323486328125, 0.51593017578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000033.npy"}
|
|
{"epoch": 0.06910994764397906, "step": 34, "batch_size": 128, "mean": 0.19447600841522217, "std": 0.992289662361145, "min": -2.410552978515625, "p10": -0.8728164672851562, "median": 0.1326141357421875, "p90": 1.5277954101562499, "max": 4.0291748046875, "pos_frac": 0.5703125, "sample": [-0.0225067138671875, 0.6353607177734375, 1.751953125, -0.03786468505859375, 0.35965728759765625, -2.374176025390625, -0.1636962890625, -0.23669052124023438, -0.1325225830078125, 1.871337890625, -1.43939208984375, 0.379547119140625, 0.41265869140625, -0.5679931640625, 0.148712158203125, 4.0291748046875, 0.5208740234375, 0.5556182861328125, 0.4775428771972656, -0.32387542724609375, 0.16366958618164062, -1.128173828125, -0.987548828125, 1.376373291015625, 0.11419677734375, -0.5465545654296875, 2.493896484375, -0.11269378662109375, 0.05377674102783203, -0.279541015625, 0.4459075927734375, 1.6854782104492188, 0.825927734375, -0.0670166015625, 0.1413726806640625, -0.7412872314453125, 0.300872802734375, -1.0272369384765625, -2.016448974609375, -0.3067626953125, 0.7749176025390625, -1.102783203125, 1.861236572265625, 2.974365234375, -0.552490234375, 0.0540618896484375, -0.546630859375, 0.4451904296875, 0.127410888671875, -0.671142578125, -0.333587646484375, -0.1694183349609375, -0.0709228515625, 1.3084716796875, -0.1077880859375, 0.850738525390625, 0.06500244140625, 0.897705078125, 2.01812744140625, -0.553466796875, 0.441253662109375, 0.19205856323242188, -0.058197021484375, 1.0446319580078125, -0.20318603515625, -0.94091796875, -0.1730499267578125, -0.15970802307128906, 0.7871932983398438, -0.659271240234375, 0.4410858154296875, 0.316558837890625, -1.293365478515625, 0.138458251953125, 1.141815185546875, 0.7664260864257812, -0.320709228515625, 0.09773445129394531, 1.1555099487304688, -0.2542133331298828, 1.50860595703125, 1.705535888671875, 0.22467041015625, -0.7562408447265625, 0.3238983154296875, -0.441619873046875, 1.57257080078125, 0.5079345703125, 0.655914306640625, 1.625885009765625, -0.07160568237304688, -1.94842529296875, 2.5972900390625, -0.805633544921875, -0.059173583984375, -0.19598388671875, -0.86328125, -1.619537353515625, 0.775634765625, 0.1333770751953125, -0.296142578125, -0.537750244140625, 0.464935302734375, 0.39215087890625, 0.45367431640625, -0.2879486083984375, -0.8950653076171875, 0.00925445556640625, -0.3625030517578125, 0.955108642578125, 0.520843505859375, 0.87274169921875, 0.562103271484375, -0.67303466796875, -0.574981689453125, 0.04138946533203125, -2.410552978515625, 0.4164581298828125, 0.381134033203125, 0.684967041015625, 1.114471435546875, 0.905426025390625, 0.13387298583984375, -0.738372802734375, 1.99273681640625, 0.1429595947265625, 0.662353515625, 0.1318511962890625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000034.npy"}
|
|
{"epoch": 0.0712041884816754, "step": 35, "batch_size": 128, "mean": -0.03497309982776642, "std": 1.0839654207229614, "min": -2.703948974609375, "p10": -1.3714927673339843, "median": 0.019134521484375, "p90": 1.0348052978515623, "max": 4.00616455078125, "pos_frac": 0.53125, "sample": [-0.8846511840820312, -2.63714599609375, -0.440704345703125, 0.353515625, 0.226318359375, 0.915863037109375, 0.986328125, 0.3576812744140625, -2.0743408203125, 0.018402099609375, 0.98931884765625, 0.02898406982421875, -2.44580078125, 0.6021881103515625, 0.9264144897460938, 0.043212890625, -2.703948974609375, -1.39898681640625, -0.273895263671875, 0.2496051788330078, 0.6095962524414062, -1.290008544921875, -0.7939453125, -0.22027587890625, 0.48691558837890625, -0.1535491943359375, 1.31158447265625, -0.47493743896484375, 1.7619171142578125, -0.63616943359375, 0.1102752685546875, -0.32273101806640625, 0.0662841796875, 0.0106048583984375, 1.743011474609375, 1.655670166015625, -0.3931732177734375, 0.8017578125, 0.22574806213378906, 0.0, -1.96258544921875, 4.00616455078125, 0.3105010986328125, 0.468841552734375, 0.5163726806640625, -1.282379150390625, 2.013946533203125, -1.368194580078125, 0.1466827392578125, -2.4407958984375, -1.4654388427734375, -0.6662826538085938, 0.164459228515625, -1.065948486328125, -0.24395751953125, -0.926361083984375, 1.021575927734375, 1.743194580078125, 0.8340377807617188, 0.008241653442382812, -0.878570556640625, 2.527099609375, 1.012908935546875, 0.1204071044921875, -1.4091796875, -0.19232177734375, 0.0828704833984375, 0.6082057952880859, 0.49298095703125, -0.19200515747070312, -2.5464935302734375, 0.307861328125, -0.58953857421875, 0.524932861328125, -0.36956214904785156, 1.065673828125, -0.6130485534667969, 0.4934844970703125, -0.175994873046875, -0.714599609375, 0.589691162109375, 0.013275146484375, -1.219757080078125, 0.3647003173828125, 1.9733810424804688, -1.01470947265625, -0.3393230438232422, -1.3791885375976562, -1.89337158203125, 0.3599510192871094, -0.873779296875, -1.3121719360351562, 0.0380096435546875, 1.6542205810546875, 0.18134117126464844, 0.3927001953125, -0.39990234375, -0.703125, 0.2395782470703125, 2.928985595703125, -1.396820068359375, 0.631103515625, 0.0, -0.2466278076171875, 0.6512603759765625, 0.546051025390625, 0.097900390625, 0.4530029296875, 0.23431396484375, -0.47347259521484375, -0.701019287109375, -0.2848663330078125, -0.671783447265625, -0.3514862060546875, 0.5674858093261719, 1.80908203125, 0.2755584716796875, 0.019866943359375, -0.33795166015625, -0.00441741943359375, 0.8042068481445312, -0.1732177734375, -1.0052642822265625, -0.557464599609375, 0.3787422180175781, -0.85162353515625, -0.6736602783203125, 0.47991943359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000035.npy"}
|
|
{"epoch": 0.07329842931937172, "step": 36, "batch_size": 128, "mean": 0.20210357010364532, "std": 1.0580976009368896, "min": -4.28778076171875, "p10": -1.0827667236328125, "median": 0.2021636962890625, "p90": 1.5838211059570309, "max": 2.81744384765625, "pos_frac": 0.578125, "sample": [-0.1456298828125, 0.532196044921875, 0.983367919921875, -0.3377838134765625, -0.25787353515625, 0.5052490234375, 1.2838134765625, -0.19339752197265625, -0.8314208984375, 0.1440277099609375, 0.25909423828125, 0.2146759033203125, 0.25531005859375, 0.082000732421875, 0.05926513671875, 0.333282470703125, 0.31993865966796875, 1.160736083984375, -1.2176513671875, -0.06833648681640625, 0.8982009887695312, -0.85791015625, -0.6615066528320312, 0.0, 0.5137405395507812, 0.336029052734375, -1.1104736328125, -1.57489013671875, -0.7332916259765625, -0.11395263671875, 1.68890380859375, 0.099151611328125, -0.1944580078125, -0.811553955078125, -0.841583251953125, -0.5533447265625, 0.36449432373046875, 0.3223114013671875, 0.15410614013671875, 2.5238037109375, 1.939666748046875, -0.1305084228515625, 0.67950439453125, 0.938995361328125, -1.154449462890625, -0.2891845703125, 1.1462860107421875, 2.2494659423828125, -1.1109619140625, -0.0811767578125, 0.49365234375, 0.6823577880859375, 0.880859375, -0.04938507080078125, -0.54193115234375, 0.9123687744140625, 0.60528564453125, 0.7357025146484375, 2.81744384765625, 2.070037841796875, 0.0, -0.316253662109375, 1.800201416015625, 0.623992919921875, 1.54998779296875, 1.2239990234375, 0.26397705078125, 1.6627655029296875, 0.2611236572265625, -0.2701568603515625, 1.79705810546875, 0.484893798828125, -1.48748779296875, 0.3282318115234375, -1.070892333984375, 0.3474922180175781, 0.1896514892578125, 1.06658935546875, -2.652008056640625, -1.129364013671875, 0.32281494140625, -0.520904541015625, 0.7505340576171875, 1.0102005004882812, -4.28778076171875, -0.14919090270996094, 1.5379638671875, -0.6239776611328125, -0.13134765625, 1.225311279296875, 1.7392578125, 1.97503662109375, 0.4688568115234375, -0.5469589233398438, -2.28057861328125, -0.5599365234375, -0.34639930725097656, 1.423095703125, -0.6194305419921875, 0.436370849609375, 1.01922607421875, 2.4835205078125, 0.4237823486328125, 0.9642791748046875, -0.73486328125, 0.38553619384765625, 2.50555419921875, 0.090789794921875, 0.933807373046875, -1.285125732421875, -0.39832115173339844, 0.42873382568359375, -0.6588363647460938, -0.0859375, -0.3978767395019531, -0.5833969116210938, -0.39910888671875, 1.409423828125, -0.2027587890625, -0.7841949462890625, 0.74755859375, 0.79534912109375, -1.1389923095703125, 0.31412506103515625, 0.051849365234375, 0.187408447265625, 0.183074951171875, -1.2047576904296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000036.npy"}
|
|
{"epoch": 0.07539267015706806, "step": 37, "batch_size": 128, "mean": 0.13524064421653748, "std": 0.9948042631149292, "min": -2.805328369140625, "p10": -1.16112060546875, "median": 0.13520050048828125, "p90": 1.5039001464843749, "max": 3.3990478515625, "pos_frac": 0.59375, "sample": [0.0547943115234375, 0.9178009033203125, 0.213836669921875, 0.576416015625, 1.66998291015625, 0.1738109588623047, -1.29644775390625, 0.4426155090332031, 1.1973724365234375, 0.171661376953125, 1.92169189453125, -0.15074539184570312, -0.25909423828125, 0.26467132568359375, 0.0028076171875, 0.5614013671875, 1.09271240234375, 0.28627777099609375, -0.044830322265625, -1.56097412109375, 0.20172882080078125, 0.179718017578125, -0.08814430236816406, -0.572967529296875, -1.7425537109375, -0.4488372802734375, 0.13580322265625, 1.54364013671875, 0.22495269775390625, -1.170135498046875, 0.9095306396484375, -0.3552284240722656, -0.38410377502441406, 0.226287841796875, 0.0673980712890625, 1.4913330078125, 0.0771484375, 2.4453125, 1.941162109375, 0.29998779296875, 0.011735916137695312, -0.9320068359375, -0.4156608581542969, -0.64752197265625, -0.812652587890625, 3.3990478515625, 0.12220096588134766, -0.06960296630859375, 0.1345977783203125, -2.020416259765625, -0.2267303466796875, 0.0716094970703125, -1.708892822265625, 1.49981689453125, -2.805328369140625, -0.38671875, 0.5367584228515625, 1.3675689697265625, -1.157257080078125, 0.628204345703125, 0.912017822265625, 0.3441162109375, 0.0829925537109375, 0.3870849609375, -0.443634033203125, -0.40020751953125, -0.55029296875, -1.56500244140625, 1.533905029296875, 1.69573974609375, 0.144866943359375, 0.1134490966796875, -0.041290283203125, -0.6180572509765625, 0.2974395751953125, 0.604278564453125, 0.33560943603515625, 0.274810791015625, 0.318756103515625, -0.54412841796875, -0.10198974609375, 3.101348876953125, 0.7371139526367188, -0.681365966796875, -0.6132354736328125, -1.412567138671875, -1.117767333984375, 0.17950439453125, 1.76812744140625, 0.23992919921875, 0.728912353515625, -0.829254150390625, -0.630950927734375, 0.060028076171875, 0.26563262939453125, 0.6193504333496094, 0.34185791015625, 0.08394241333007812, -0.08011627197265625, 0.31329345703125, -0.7266845703125, -0.169769287109375, -0.478118896484375, -0.6453361511230469, 0.56280517578125, 0.589202880859375, 1.513427734375, 1.8302001953125, 1.0698699951171875, 0.3817100524902344, -0.132049560546875, -1.25982666015625, -0.3892364501953125, -0.022186279296875, 0.764556884765625, 0.51580810546875, 0.531524658203125, -1.40435791015625, -0.625885009765625, -0.2445220947265625, 0.2774658203125, -1.2293853759765625, -0.196502685546875, 0.21673583984375, 0.50634765625, 1.1497802734375, -1.522430419921875, 2.786865234375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000037.npy"}
|
|
{"epoch": 0.0774869109947644, "step": 38, "batch_size": 128, "mean": 0.242706760764122, "std": 1.1111340522766113, "min": -2.65289306640625, "p10": -1.011700439453125, "median": 0.20087432861328125, "p90": 1.476116943359375, "max": 4.494537353515625, "pos_frac": 0.6171875, "sample": [1.1678466796875, 0.11289215087890625, -0.3640098571777344, 0.558624267578125, 0.25594329833984375, -0.12874603271484375, 0.602508544921875, 1.748199462890625, 1.4794921875, 1.051513671875, -0.73516845703125, -0.0533905029296875, -0.1275787353515625, 4.494537353515625, -0.8116455078125, 2.211181640625, 0.3226165771484375, -0.8825607299804688, -0.3881683349609375, -0.5135955810546875, 0.48590087890625, -1.41455078125, 0.3053627014160156, 0.085601806640625, 0.146209716796875, -0.72540283203125, -2.3177490234375, -0.17010498046875, -1.21063232421875, -0.3558158874511719, 0.0953369140625, 1.3436279296875, -0.5802001953125, -1.49786376953125, 0.34998321533203125, 0.3330535888671875, -0.19573974609375, -0.1917743682861328, 1.111968994140625, -1.38519287109375, -0.43695068359375, -0.27191925048828125, -0.06787109375, 0.0175018310546875, -0.864715576171875, 1.2024154663085938, 0.1270751953125, 0.406707763671875, 1.1466217041015625, 0.76483154296875, 0.3916015625, 0.30077362060546875, -1.048828125, -0.5123062133789062, 0.5076255798339844, -0.5962982177734375, -0.637420654296875, 0.1080322265625, -0.99578857421875, 0.278533935546875, 0.25521087646484375, 0.16363525390625, 1.974212646484375, -0.2147674560546875, 0.2521514892578125, -0.04094123840332031, 0.6729965209960938, -2.5959625244140625, 0.9056243896484375, -0.46453857421875, 1.0203857421875, 0.0435791015625, -0.23858642578125, 1.47467041015625, 1.5462188720703125, -0.58905029296875, 2.8165130615234375, -0.38623809814453125, 0.5113449096679688, -1.4093017578125, 0.6760406494140625, -1.1671676635742188, 0.0120849609375, 2.17999267578125, 0.0920867919921875, 3.246856689453125, -0.03692626953125, -0.600921630859375, 0.6158447265625, 0.3923492431640625, 1.1690216064453125, -1.338043212890625, 2.663909912109375, 0.2381134033203125, 0.2869873046875, 0.24918365478515625, 0.0321502685546875, 1.1304779052734375, 3.0909423828125, 0.037811279296875, 0.9255599975585938, 0.0, 0.397308349609375, 0.85870361328125, 0.571563720703125, -0.5363922119140625, 0.33868408203125, 0.722930908203125, 1.3426513671875, -0.47662353515625, 0.450286865234375, -2.65289306640625, -1.2911529541015625, 0.9454345703125, 1.745849609375, 0.12451171875, 2.633544921875, 0.7749786376953125, 0.682342529296875, 1.18988037109375, -0.8330078125, 0.2450408935546875, -2.09124755859375, 0.7277679443359375, -0.362457275390625, 1.4349517822265625, 0.031175613403320312, 0.46698760986328125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000038.npy"}
|
|
{"epoch": 0.07958115183246073, "step": 39, "batch_size": 128, "mean": 0.42272233963012695, "std": 0.9566536545753479, "min": -2.32769775390625, "p10": -0.6391891479492188, "median": 0.3371009826660156, "p90": 1.6508056640624997, "max": 3.8160400390625, "pos_frac": 0.6640625, "sample": [0.1107940673828125, -0.4925537109375, -1.045013427734375, 0.6086502075195312, -0.38531494140625, -0.808074951171875, -0.0535430908203125, -0.07492828369140625, -1.680389404296875, 0.6044921875, 0.079925537109375, 1.3861618041992188, -0.124267578125, 0.909393310546875, 0.2977294921875, 1.8353271484375, -0.459259033203125, 1.693572998046875, 0.3444366455078125, -0.6502227783203125, 0.912567138671875, 1.0484619140625, 0.7430534362792969, 0.50408935546875, 2.929840087890625, 1.2831573486328125, 1.1067428588867188, -0.656982421875, -0.1175994873046875, -0.21636962890625, 1.632476806640625, 1.262451171875, 1.425567626953125, 0.2191314697265625, 1.126678466796875, 0.336944580078125, -0.015329360961914062, 0.014617919921875, 1.80999755859375, 0.42215728759765625, 0.33538818359375, -0.06871604919433594, 0.9560394287109375, 0.4279327392578125, 0.24161529541015625, 2.26629638671875, 1.61907958984375, 0.0008716583251953125, 0.8779754638671875, 2.01104736328125, -1.303985595703125, -0.5831069946289062, -0.5642852783203125, 1.12835693359375, 0.50982666015625, -0.531158447265625, 0.414703369140625, 0.916748046875, -2.32769775390625, -0.29669189453125, -0.3719673156738281, -0.0890960693359375, -0.619659423828125, 0.39984130859375, 0.896759033203125, 1.900238037109375, -0.0621185302734375, 0.07967567443847656, 1.7150726318359375, -0.3210906982421875, -1.60015869140625, 0.239654541015625, -0.1397228240966797, -1.688934326171875, 0.519134521484375, 0.92242431640625, 0.7691650390625, -0.546661376953125, 1.1348190307617188, -0.6715087890625, 1.404266357421875, 0.1526947021484375, 0.9736328125, 0.6934356689453125, 0.162933349609375, -0.770538330078125, 3.8160400390625, 1.41668701171875, 1.8219757080078125, -0.09215545654296875, 0.15319442749023438, -0.793914794921875, 0.17218780517578125, -0.8576126098632812, 0.0, 0.025665283203125, 0.32196807861328125, -0.36834716796875, 1.8758544921875, 0.53851318359375, 2.024169921875, -0.02722930908203125, 0.702117919921875, 0.2723541259765625, 0.533905029296875, 0.6277618408203125, 0.730804443359375, 0.5561981201171875, 1.165069580078125, -0.05352783203125, 1.051513671875, 0.1481170654296875, 0.33725738525390625, 0.4735832214355469, 1.08148193359375, -0.2940025329589844, 0.8900146484375, 1.617431640625, 0.385040283203125, 0.13214874267578125, -0.06229400634765625, 0.19667816162109375, -0.63446044921875, 3.34344482421875, 0.903350830078125, 0.6260528564453125, -0.31430816650390625, 0.68463134765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000039.npy"}
|
|
{"epoch": 0.08167539267015707, "step": 40, "batch_size": 128, "mean": 0.46768683195114136, "std": 1.0331898927688599, "min": -1.573760986328125, "p10": -0.5890495300292968, "median": 0.3453845977783203, "p90": 1.7363388061523433, "max": 4.6051025390625, "pos_frac": 0.6328125, "sample": [-0.23302268981933594, -1.10064697265625, -0.27227783203125, -0.2977752685546875, -0.14510154724121094, 0.36138916015625, -0.9139404296875, 3.7381591796875, -0.1002349853515625, 2.29364013671875, 0.6113433837890625, 0.39905548095703125, -0.252838134765625, 1.157958984375, -0.4598541259765625, 0.6917991638183594, 0.3685264587402344, 0.45501708984375, 0.26360511779785156, 0.35614776611328125, 0.01692962646484375, 1.8603515625, -0.04662513732910156, -0.13488388061523438, 0.48009490966796875, 1.585174560546875, -0.075286865234375, 1.3002471923828125, 0.264007568359375, 1.27935791015625, -0.8697013854980469, 0.464324951171875, 0.36423492431640625, 0.13482666015625, -0.89947509765625, 2.40716552734375, 0.169342041015625, -0.5712432861328125, 3.31524658203125, -0.4078369140625, 4.6051025390625, 0.928741455078125, -0.45684814453125, -0.13763427734375, -0.44828033447265625, -0.6123771667480469, -0.3798675537109375, 1.8438873291015625, 2.00262451171875, 1.524871826171875, -0.5860671997070312, -0.25634765625, 1.0805511474609375, 1.45745849609375, 1.546234130859375, 0.1937255859375, -0.0104522705078125, 0.6820602416992188, 1.69024658203125, 0.03594970703125, 0.40386962890625, 0.10920906066894531, 0.76116943359375, -0.353424072265625, -0.876953125, 0.25921630859375, 1.52020263671875, 1.2216796875, 2.283843994140625, 0.722320556640625, -0.2769622802734375, -0.803009033203125, 0.6285552978515625, 1.627197265625, 0.13568115234375, -1.0297126770019531, 1.4029541015625, 0.3434410095214844, -0.2958984375, 0.149658203125, -0.0930328369140625, 1.08355712890625, -0.0803680419921875, -0.0184478759765625, -0.459564208984375, 0.5721015930175781, 0.07866668701171875, 0.9964599609375, 0.3833656311035156, -0.0812225341796875, -0.59600830078125, 2.0860595703125, 0.265869140625, 2.089385986328125, -1.4775390625, 0.78253173828125, 0.39892578125, 0.09125518798828125, -0.36663818359375, 1.329925537109375, 1.264312744140625, 0.55133056640625, 0.40570068359375, 0.34732818603515625, -0.5189666748046875, 2.5997314453125, 0.21209716796875, -0.1616363525390625, -0.19696807861328125, 0.505157470703125, 0.6439208984375, 0.766265869140625, 1.533538818359375, 0.5180587768554688, 1.548248291015625, -1.258392333984375, -1.573760986328125, -0.2789459228515625, 1.3338623046875, 0.74993896484375, 0.6478271484375, -0.21912384033203125, -0.2703399658203125, 1.24609375, 0.0429534912109375, 2.562042236328125, -1.332763671875, 1.017303466796875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000040.npy"}
|
|
{"epoch": 0.08376963350785341, "step": 41, "batch_size": 128, "mean": 0.23396101593971252, "std": 1.0367461442947388, "min": -3.170654296875, "p10": -0.9260665893554686, "median": 0.27954864501953125, "p90": 1.3573722839355469, "max": 3.3392333984375, "pos_frac": 0.609375, "sample": [1.60498046875, 0.798309326171875, 0.26043701171875, -1.2199249267578125, -0.322509765625, -2.576812744140625, 0.058319091796875, 1.201934814453125, 1.11505126953125, 0.7780303955078125, 1.2444381713867188, 1.244171142578125, -0.1454315185546875, 0.1636505126953125, -0.4529266357421875, -0.001708984375, 0.2073516845703125, -0.4495849609375, 0.8963623046875, 0.538116455078125, -0.05743408203125, 1.5198974609375, -0.1147308349609375, 1.1497650146484375, -0.17120361328125, 0.55859375, 0.96484375, 1.701019287109375, -1.720367431640625, 0.29231834411621094, 1.3778457641601562, -0.652496337890625, -0.18939208984375, 0.28814697265625, -0.0906219482421875, -0.5029945373535156, 1.99981689453125, 1.351226806640625, 0.448577880859375, 0.791961669921875, 2.8636474609375, 0.8788986206054688, 0.55950927734375, 1.80413818359375, 0.56524658203125, -0.18341064453125, 0.6622314453125, -0.28226470947265625, 0.982452392578125, 0.520660400390625, 0.0046539306640625, -0.763275146484375, -1.05987548828125, -1.0316162109375, 3.3392333984375, -0.888671875, 0.5922317504882812, -0.51605224609375, -0.171844482421875, 0.3253631591796875, -0.03509521484375, 0.682891845703125, -0.300994873046875, 1.0111083984375, -0.449554443359375, 0.2539520263671875, -1.689971923828125, 0.198974609375, 0.0, 0.464019775390625, -0.686767578125, -0.3128662109375, -1.0133209228515625, 0.040985107421875, -2.06402587890625, -1.33526611328125, 0.5791397094726562, 0.376220703125, -0.83734130859375, -1.250885009765625, 0.8387451171875, 1.0361099243164062, 0.2709503173828125, 1.3717117309570312, 0.5751953125, 0.3049774169921875, 0.6063079833984375, 0.8250732421875, 1.673065185546875, -0.496917724609375, -1.191925048828125, 0.5195465087890625, 0.9910926818847656, -0.21697998046875, -0.44525146484375, 1.098602294921875, 0.354278564453125, -0.603118896484375, 0.044104576110839844, -0.5672760009765625, 2.739013671875, 0.6019287109375, 0.867828369140625, 0.11627960205078125, -0.2279052734375, 0.9260482788085938, 0.16649627685546875, -0.6912460327148438, 0.68841552734375, -0.8269195556640625, 1.0695934295654297, 0.97906494140625, 0.04400634765625, -2.747406005859375, 2.270355224609375, -0.756866455078125, 0.89520263671875, 0.00079345703125, 0.6598281860351562, -0.813568115234375, -3.170654296875, 1.193511962890625, 0.853546142578125, -0.03045654296875, 1.8177490234375, 0.351898193359375, 1.15045166015625, -0.88775634765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000041.npy"}
|
|
{"epoch": 0.08586387434554973, "step": 42, "batch_size": 128, "mean": 0.3963726758956909, "std": 1.1262952089309692, "min": -3.71826171875, "p10": -0.655322265625, "median": 0.2911977767944336, "p90": 1.6536834716796873, "max": 3.195068359375, "pos_frac": 0.6875, "sample": [-0.08811569213867188, -0.03980255126953125, 0.1259002685546875, 1.457275390625, 1.76959228515625, 0.99407958984375, -0.1807708740234375, 2.65777587890625, 0.782379150390625, -0.07212066650390625, 1.868011474609375, 3.1468505859375, 2.037109375, 0.29965972900390625, -2.298004150390625, -0.1796436309814453, 0.482177734375, 1.228515625, 0.4338531494140625, 2.79473876953125, 0.23081207275390625, 0.0530853271484375, 0.0061492919921875, 0.503753662109375, 1.26861572265625, 0.16802978515625, 0.437408447265625, 0.0325775146484375, 1.6815185546875, -0.6585693359375, -0.088531494140625, 1.0093631744384766, 1.1456298828125, 0.63433837890625, 1.13909912109375, 1.2267303466796875, 1.3718719482421875, 0.35843658447265625, -0.622528076171875, -0.0500335693359375, 0.6321029663085938, -0.015438079833984375, 0.16229248046875, -0.2273101806640625, 0.7882080078125, -1.1365814208984375, 1.3910064697265625, 0.41650390625, 2.219696044921875, 0.19139480590820312, 0.5502853393554688, 0.27472686767578125, 0.8792724609375, -0.04070281982421875, 0.8369827270507812, -0.0113372802734375, -0.055023193359375, -0.987060546875, -0.16400146484375, 0.5070419311523438, 1.51416015625, -0.6539306640625, -0.8709259033203125, -1.8674774169921875, 0.4755859375, 1.916412353515625, -0.6339950561523438, 0.10475921630859375, -0.441802978515625, 0.28273582458496094, 0.118499755859375, 0.09085845947265625, 0.634979248046875, 1.5936279296875, -3.71826171875, 0.012022018432617188, -3.02642822265625, 0.10083770751953125, 1.031982421875, -0.037006378173828125, 0.03166961669921875, 2.28857421875, -0.662811279296875, 1.12628173828125, 1.280517578125, 0.980194091796875, 1.308013916015625, 1.0919189453125, 0.49358558654785156, 1.281829833984375, 1.0904998779296875, -1.680572509765625, 0.15863037109375, 0.5356826782226562, 1.2977294921875, -0.083465576171875, 0.08453369140625, 3.195068359375, 0.48258209228515625, -0.4095306396484375, 1.641754150390625, -0.840423583984375, 0.0, 0.04730224609375, -0.059326171875, -0.4284782409667969, 0.811431884765625, 0.137298583984375, -0.35553741455078125, 0.836822509765625, 0.229736328125, -0.41998291015625, 0.6557159423828125, 0.16810989379882812, 0.7708740234375, 2.479949951171875, 1.44305419921875, 0.1651611328125, 1.4297409057617188, 0.1215057373046875, -0.192291259765625, 2.087615966796875, -3.5572509765625, 1.24493408203125, -2.16546630859375, 0.3025970458984375, -0.119354248046875, 0.5033645629882812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000042.npy"}
|
|
{"epoch": 0.08795811518324607, "step": 43, "batch_size": 128, "mean": 0.4291083514690399, "std": 1.2131116390228271, "min": -4.632720947265625, "p10": -0.865086555480957, "median": 0.49472808837890625, "p90": 1.8734222412109371, "max": 3.48651123046875, "pos_frac": 0.65625, "sample": [1.067779541015625, 2.119720458984375, 1.36920166015625, -3.2406005859375, 1.104705810546875, 2.0093231201171875, 1.511962890625, -0.0652313232421875, -0.1273651123046875, 0.16843414306640625, -0.17985153198242188, -0.3843231201171875, 0.4926910400390625, 1.303497314453125, -0.808807373046875, -1.121612548828125, -0.4096832275390625, -0.33160400390625, 1.239501953125, 0.3468017578125, 1.98992919921875, 0.599609375, 2.38616943359375, 0.2460784912109375, -0.9644317626953125, -0.916412353515625, -1.361297607421875, 1.1800689697265625, 0.18906021118164062, 1.966156005859375, 1.4937744140625, 2.03948974609375, 0.9456787109375, -4.632720947265625, 1.83367919921875, 0.21043014526367188, -0.32714080810546875, 0.43703460693359375, 0.6832942962646484, 0.5953826904296875, -0.1697845458984375, -0.0810546875, -0.8827571868896484, -0.857513427734375, 0.20156097412109375, 0.2086353302001953, -0.2216949462890625, -0.7605819702148438, -3.566375732421875, -0.1846771240234375, -0.68890380859375, 1.1668243408203125, 0.09053611755371094, 1.7569656372070312, 1.5711669921875, 1.074462890625, 1.07061767578125, 1.803314208984375, -1.4202957153320312, 0.7939453125, 0.6741943359375, -0.113525390625, 0.3743896484375, -0.30340576171875, 2.08013916015625, -1.21563720703125, 1.13604736328125, -1.9569091796875, 1.36553955078125, 2.5800628662109375, 0.566375732421875, 0.958251953125, -0.4876861572265625, -0.20977783203125, 0.9217529296875, -0.51593017578125, -0.4007720947265625, -0.56793212890625, -0.27146148681640625, 2.009796142578125, 0.6184844970703125, 2.3673095703125, 0.123779296875, 0.5283050537109375, 1.4344940185546875, 1.2186737060546875, 1.1782684326171875, 0.1398162841796875, 0.310089111328125, 0.8513412475585938, 1.5091552734375, 0.597900390625, -2.2646484375, 0.8092575073242188, 2.662994384765625, -0.3401298522949219, 1.0107498168945312, 3.48651123046875, -0.85284423828125, 0.53009033203125, 0.49676513671875, 1.2473220825195312, 1.451690673828125, -0.4624061584472656, -0.852996826171875, 1.402587890625, 1.448089599609375, 0.28106689453125, 1.1054229736328125, 1.323211669921875, 0.241729736328125, 1.027587890625, 0.36197662353515625, -1.1246109008789062, 1.8048095703125, 0.9700164794921875, 1.2016448974609375, 0.7146224975585938, 1.13165283203125, 0.385986328125, -0.024658203125, -0.4267578125, -0.1923503875732422, 2.1993408203125, 0.968048095703125, 0.0043487548828125, 0.3702545166015625, -0.23439979553222656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000043.npy"}
|
|
{"epoch": 0.09005235602094241, "step": 44, "batch_size": 128, "mean": 0.6096642017364502, "std": 1.360891342163086, "min": -3.3184814453125, "p10": -0.939532470703125, "median": 0.5214958190917969, "p90": 2.4410552978515625, "max": 4.196685791015625, "pos_frac": 0.703125, "sample": [0.1278076171875, 0.2373809814453125, 0.867431640625, -0.48836517333984375, -0.526123046875, 2.576751708984375, 0.014362335205078125, -0.16405868530273438, -0.0055084228515625, -0.193572998046875, -0.0728607177734375, -0.485443115234375, 0.8860015869140625, -0.054462432861328125, 2.434326171875, 0.16717529296875, -0.948760986328125, -0.0022220611572265625, -2.36016845703125, 0.6937713623046875, 0.30657958984375, 0.89324951171875, -0.338287353515625, 0.512725830078125, 0.36078643798828125, 0.5213699340820312, 2.456756591796875, 0.14478492736816406, -3.3184814453125, 0.1010894775390625, 1.8267364501953125, -0.47802734375, -0.018951416015625, -0.2560882568359375, 0.17454147338867188, -0.473876953125, 0.857818603515625, -0.6717033386230469, 3.680419921875, 2.148651123046875, 2.76715087890625, -0.5744380950927734, 0.3029937744140625, 2.790435791015625, 0.362030029296875, 2.46710205078125, 1.245269775390625, 1.7742919921875, 1.391204833984375, -0.5206756591796875, 0.21147727966308594, 2.26165771484375, -1.1747894287109375, -0.39324951171875, 0.116851806640625, 1.699981689453125, 0.96673583984375, 2.2178955078125, 0.11692428588867188, -0.0699462890625, 0.67694091796875, 0.68359375, 0.7048530578613281, 3.6378173828125, 1.555908203125, 1.008514404296875, 1.23858642578125, -1.312347412109375, 0.983734130859375, 3.93792724609375, -0.1502685546875, 0.6338348388671875, -0.935577392578125, 2.04345703125, 2.56109619140625, 1.4149017333984375, 1.85516357421875, -2.152679443359375, 0.6729736328125, -3.1956787109375, 2.386383056640625, 0.7391815185546875, 2.01873779296875, 0.5726757049560547, 1.78271484375, 0.482177734375, 0.1593017578125, 1.0140380859375, 4.196685791015625, 0.302398681640625, -0.19537353515625, 1.76751708984375, 0.6407470703125, 0.818634033203125, 0.5567474365234375, 1.747711181640625, -1.13214111328125, 0.397430419921875, 0.12908935546875, 0.142608642578125, 1.7742919921875, 0.07500267028808594, -1.161895751953125, 0.421478271484375, 0.0953521728515625, 1.03607177734375, -0.439971923828125, 2.194671630859375, 0.39467620849609375, -1.295562744140625, 3.18603515625, -0.239013671875, 0.5370254516601562, 3.171051025390625, 0.6286544799804688, 0.94500732421875, 0.6959686279296875, 0.5216217041015625, -3.176544189453125, -0.5412139892578125, -1.612091064453125, 1.3360824584960938, 0.9761505126953125, 0.929901123046875, 2.536102294921875, -1.073974609375, 0.6667327880859375, 1.972930908203125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000044.npy"}
|
|
{"epoch": 0.09214659685863874, "step": 45, "batch_size": 128, "mean": 0.5594825744628906, "std": 1.3970685005187988, "min": -2.990142822265625, "p10": -0.9256103515624999, "median": 0.391632080078125, "p90": 2.2191192626953122, "max": 6.416015625, "pos_frac": 0.65625, "sample": [-0.072265625, 0.0449676513671875, 2.609161376953125, 0.7561492919921875, -0.1784229278564453, -0.103302001953125, 0.35485076904296875, 0.971343994140625, 1.0481719970703125, 2.695068359375, 0.798614501953125, 2.206512451171875, 3.0787353515625, -0.5063705444335938, -0.084686279296875, 0.724273681640625, -0.10611152648925781, 0.2783203125, 0.0, -0.9083251953125, 0.16192626953125, 0.3828125, -0.40045166015625, 0.55096435546875, -2.055145263671875, 3.527435302734375, -0.38103485107421875, 0.828155517578125, 0.5247802734375, 0.20552825927734375, 0.40045166015625, -0.2583122253417969, 0.1580810546875, 0.3769378662109375, -0.38506317138671875, 1.135040283203125, 0.20416259765625, 0.13193893432617188, 2.761260986328125, 1.66656494140625, 1.969024658203125, 0.99658203125, 1.1398773193359375, -0.026458740234375, 1.998779296875, 0.045909881591796875, 1.215057373046875, 0.3382415771484375, 0.69598388671875, 1.125213623046875, -0.027435302734375, 1.335540771484375, -0.3265380859375, -1.0502166748046875, -0.0858306884765625, -1.1346893310546875, 0.6374664306640625, 0.83758544921875, -0.17277145385742188, -2.62005615234375, -2.22381591796875, 2.423614501953125, -0.7239151000976562, 0.2547149658203125, -1.6253662109375, -0.1529388427734375, 0.550384521484375, 0.7160911560058594, 1.4266510009765625, -0.022378921508789062, -0.570831298828125, 2.172393798828125, -0.3249359130859375, 1.4569511413574219, 5.6708984375, 0.3023681640625, 0.6559295654296875, -0.038299560546875, 0.812744140625, -0.48883056640625, 1.1579208374023438, 2.5838623046875, 0.4564552307128906, 1.8069915771484375, 0.662078857421875, -1.615234375, 0.3143310546875, 0.5027542114257812, -0.29486083984375, -2.990142822265625, 2.009857177734375, 2.24853515625, 0.04925537109375, -1.0095977783203125, 3.549560546875, 0.5824050903320312, 1.958984375, 1.721435546875, 1.5614013671875, -1.363800048828125, -2.706787109375, 0.7076873779296875, -0.17934417724609375, -0.6647186279296875, 0.3116912841796875, 3.46630859375, 0.8181915283203125, -1.2569122314453125, 0.40853118896484375, 0.2331695556640625, 0.5312652587890625, -0.095855712890625, 1.01727294921875, 0.84588623046875, -0.1692962646484375, -0.9659423828125, 2.113433837890625, 0.6102294921875, 6.416015625, 0.27410888671875, 1.12286376953125, 2.316558837890625, 0.3504219055175781, 1.0253143310546875, 1.33349609375, -0.6337738037109375, -0.9036865234375, 1.0900344848632812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000045.npy"}
|
|
{"epoch": 0.09424083769633508, "step": 46, "batch_size": 128, "mean": 0.6571169495582581, "std": 1.3843001127243042, "min": -3.346588134765625, "p10": -1.04361572265625, "median": 0.41839599609375, "p90": 2.4087265014648436, "max": 4.96478271484375, "pos_frac": 0.6953125, "sample": [2.3912353515625, 4.96478271484375, 1.8516693115234375, 0.47845458984375, 0.473052978515625, 1.22296142578125, 2.2962646484375, 0.83050537109375, -1.083648681640625, -0.5492401123046875, -1.026458740234375, 1.13470458984375, 4.29534912109375, -0.482635498046875, 0.8966064453125, -1.929046630859375, -1.11505126953125, -0.059024810791015625, 1.677947998046875, 0.32196044921875, 0.38739013671875, 0.1135406494140625, 1.04827880859375, 0.946441650390625, -1.2255859375, -0.18756103515625, -1.141326904296875, -0.34497833251953125, 4.27447509765625, -0.24554061889648438, -0.225341796875, 0.6065597534179688, 1.099456787109375, 0.300689697265625, 1.824737548828125, 4.72625732421875, 0.295440673828125, 0.551422119140625, -0.4043426513671875, 0.534454345703125, -0.935638427734375, 0.772552490234375, 0.8387985229492188, 0.3980712890625, 1.76287841796875, -2.43011474609375, 1.718414306640625, 0.2502632141113281, 0.9279098510742188, 0.19158935546875, 1.97943115234375, 2.4003448486328125, 0.6872329711914062, 1.1829833984375, -0.069000244140625, 2.4839019775390625, 0.34796142578125, 3.018524169921875, 0.225311279296875, 2.36236572265625, 0.07440948486328125, 0.3846473693847656, -0.5473175048828125, 0.063934326171875, 2.42828369140625, -0.05123138427734375, 0.3034229278564453, -1.2142333984375, -1.5643157958984375, 1.937591552734375, 1.005035400390625, 0.15192413330078125, 1.79791259765625, 0.2476043701171875, -0.6251754760742188, -1.371612548828125, -0.5695838928222656, -0.38730621337890625, 0.5142307281494141, -0.055450439453125, 0.23661041259765625, 1.013580322265625, 1.5800933837890625, -1.09246826171875, 1.551544189453125, 0.5726547241210938, -3.346588134765625, -0.274322509765625, 0.332916259765625, 0.136749267578125, -0.75811767578125, 0.9214935302734375, 2.48321533203125, 0.05731201171875, -0.4545326232910156, 3.69744873046875, -0.12659454345703125, -1.2287139892578125, -0.98016357421875, 1.2142486572265625, -0.19081878662109375, 1.1090240478515625, 2.1254119873046875, 0.0053253173828125, 1.26678466796875, 0.7935028076171875, 1.192962646484375, 1.0187835693359375, 1.4603271484375, 0.438720703125, 1.719268798828125, -1.3580322265625, 0.20684432983398438, 0.02466583251953125, -0.1074676513671875, 2.850860595703125, 0.84625244140625, 2.264556884765625, 3.20098876953125, 1.27374267578125, 0.0, 0.909515380859375, 0.0380706787109375, -0.07097625732421875, 0.1251678466796875, 1.9390869140625, 2.61260986328125, 2.718017578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000046.npy"}
|
|
{"epoch": 0.09633507853403141, "step": 47, "batch_size": 128, "mean": 0.3610554039478302, "std": 1.358116865158081, "min": -3.928741455078125, "p10": -1.1648849487304687, "median": 0.2357177734375, "p90": 1.8476562499999998, "max": 4.433868408203125, "pos_frac": 0.625, "sample": [-0.153564453125, 0.0569915771484375, 0.9281845092773438, -0.39935302734375, 4.433868408203125, -2.077789306640625, -3.568634033203125, -0.1422576904296875, 2.470306396484375, 0.8875732421875, 1.105926513671875, 0.10056304931640625, 0.20599365234375, -0.31545257568359375, -1.4580078125, -0.776763916015625, 1.1316375732421875, -0.530487060546875, 3.0318603515625, -0.29315185546875, 2.078643798828125, -2.612030029296875, 0.8486366271972656, -1.0626220703125, 0.2999267578125, 1.12017822265625, 0.081298828125, 0.934661865234375, -2.80609130859375, -0.7474212646484375, 0.509918212890625, 0.0, 0.8744049072265625, -1.1369247436523438, 0.12896728515625, 1.496978759765625, 3.11053466796875, -0.124755859375, -0.8203887939453125, 0.7389755249023438, 0.1640625, -0.016490936279296875, 1.25152587890625, 3.38299560546875, 0.8607635498046875, 0.1868896484375, 0.044708251953125, 1.1717529296875, -0.2748870849609375, 0.1680316925048828, 0.621429443359375, 0.11104202270507812, 1.912384033203125, 0.239166259765625, -1.48577880859375, 4.11456298828125, -3.928741455078125, 1.646240234375, 0.232269287109375, 1.441558837890625, -1.1645050048828125, 1.131317138671875, -0.4228057861328125, 1.4239501953125, 0.658721923828125, -0.1860198974609375, -0.28948974609375, 0.1679229736328125, 0.7252349853515625, 2.0786895751953125, 0.34173583984375, 0.22653770446777344, -0.26416015625, -0.092193603515625, 0.4735565185546875, 1.7746200561523438, 0.0, 1.5335235595703125, 1.959014892578125, 1.819915771484375, 1.0855636596679688, 0.32172393798828125, -0.575531005859375, -0.248748779296875, 1.288665771484375, -0.18414306640625, -0.014743804931640625, -0.083099365234375, 0.46767425537109375, 1.2954864501953125, 1.2688140869140625, -1.165771484375, 0.118743896484375, 2.883941650390625, 0.6578598022460938, 1.164794921875, 0.500244140625, -0.6506500244140625, 1.59765625, 1.2186279296875, -0.182281494140625, 1.2644805908203125, 0.5457916259765625, 1.602996826171875, -0.80743408203125, -1.510406494140625, 0.10329055786132812, -1.90020751953125, -0.3237152099609375, -0.03575897216796875, -0.053009033203125, 0.0198974609375, 2.296142578125, 0.33274269104003906, 1.26416015625, 1.6728515625, -1.2021102905273438, 1.099843978881836, -0.1037750244140625, 0.3199462890625, -0.6411895751953125, 2.53466796875, 0.86700439453125, 0.35528564453125, 1.02972412109375, -3.70648193359375, -1.46832275390625, 0.604461669921875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000047.npy"}
|
|
{"epoch": 0.09842931937172775, "step": 48, "batch_size": 128, "mean": 0.6560736894607544, "std": 1.5277864933013916, "min": -2.2156829833984375, "p10": -0.9509155273437498, "median": 0.4579925537109375, "p90": 2.2228546142578125, "max": 7.03594970703125, "pos_frac": 0.6875, "sample": [0.543792724609375, 1.955596923828125, 0.8665008544921875, 1.6714859008789062, 0.969329833984375, 1.45452880859375, 0.38507080078125, 3.437652587890625, 1.537811279296875, 0.1284332275390625, -0.14560699462890625, -0.7908782958984375, -0.20322418212890625, -2.2156829833984375, -0.4278411865234375, 3.18914794921875, -1.79595947265625, -1.423980712890625, -2.0861358642578125, 0.2947235107421875, 2.70599365234375, 0.46710968017578125, 0.337371826171875, -1.0391845703125, 1.2386016845703125, -0.9130859375, -0.7841796875, 1.536712646484375, -0.61151123046875, 0.5338134765625, 0.6002635955810547, 1.04498291015625, 1.4459228515625, 1.00164794921875, 4.1904296875, 0.8933372497558594, -0.05803871154785156, 1.68927001953125, 1.8703460693359375, 2.378753662109375, 0.06203460693359375, 2.209625244140625, 0.0840301513671875, 0.12903594970703125, 0.9097137451171875, 0.7219696044921875, -2.005523681640625, 0.201385498046875, -0.53570556640625, 0.43329620361328125, 5.366363525390625, 0.8377647399902344, 0.799774169921875, 0.20203399658203125, -2.068450927734375, 0.856201171875, 0.32390594482421875, 0.250244140625, 0.2830619812011719, 1.463958740234375, 1.23675537109375, 0.9653778076171875, -0.6462249755859375, 1.97027587890625, 0.02197265625, -0.52337646484375, -1.247833251953125, -0.142120361328125, 0.8735275268554688, 1.1712188720703125, -2.1688232421875, 1.805419921875, 0.0, -1.2319488525390625, 1.89483642578125, 0.20020675659179688, 1.67193603515625, 0.45855712890625, 0.48961639404296875, 2.25372314453125, 0.533294677734375, 1.939788818359375, -0.509002685546875, 6.32098388671875, 0.149658203125, 3.310546875, -1.214324951171875, 0.6348190307617188, -0.27602386474609375, 0.0112152099609375, -0.566192626953125, 0.31439208984375, -1.1111297607421875, -0.7333145141601562, 2.171295166015625, 4.0062255859375, 0.977996826171875, 0.9412994384765625, 0.109039306640625, 1.21441650390625, -0.27825927734375, 1.9817047119140625, 0.0, 0.360382080078125, 2.444183349609375, 0.457427978515625, 7.03594970703125, -0.12224578857421875, -0.881072998046875, 1.2714309692382812, -0.5197296142578125, -0.0266876220703125, 0.636505126953125, 0.2640380859375, 0.14767074584960938, 1.258636474609375, -0.6213226318359375, -0.7989349365234375, 0.5409317016601562, 0.5609798431396484, 0.731903076171875, 1.70318603515625, -0.3058929443359375, 0.063385009765625, -1.1749725341796875, -0.128662109375, 3.865234375, 1.835540771484375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000048.npy"}
|
|
{"epoch": 0.10052356020942409, "step": 49, "batch_size": 128, "mean": 0.518486738204956, "std": 1.5895745754241943, "min": -4.39190673828125, "p10": -0.96104736328125, "median": 0.42653656005859375, "p90": 2.482843017578125, "max": 5.68212890625, "pos_frac": 0.671875, "sample": [1.1367111206054688, 0.74981689453125, 0.8779335021972656, 0.5811386108398438, -0.712615966796875, 0.1477203369140625, -0.139251708984375, 4.92974853515625, 3.62542724609375, 0.8345661163330078, 0.5355072021484375, 1.868682861328125, -0.416778564453125, 0.3495635986328125, -0.3725128173828125, 0.151092529296875, 1.7885894775390625, 1.296987533569336, 0.31616973876953125, 2.58709716796875, -0.639617919921875, 0.4532585144042969, 2.82257080078125, 0.78594970703125, 1.089263916015625, -0.73992919921875, -1.2962799072265625, 5.68212890625, 0.04119110107421875, 1.25909423828125, 0.4285125732421875, 0.601898193359375, 1.11175537109375, 0.17260360717773438, 2.481781005859375, 0.8751869201660156, 1.136138916015625, 1.018951416015625, 0.2334747314453125, -0.3856201171875, 0.23240089416503906, 2.017578125, -2.20880126953125, 2.3187332153320312, 1.07818603515625, 0.2983245849609375, 2.3734130859375, 2.606170654296875, 2.2093505859375, -1.49468994140625, -0.64215087890625, -1.6718215942382812, -0.3800048828125, 1.580596923828125, 0.43024444580078125, 3.34326171875, 0.6099472045898438, 0.12644577026367188, -0.313934326171875, 0.681884765625, 0.04949188232421875, 2.02789306640625, -1.02410888671875, -0.26171875, -0.11483001708984375, 0.630157470703125, -0.084075927734375, -1.5441741943359375, 0.7275390625, 0.4306640625, 0.9906005859375, 2.020965576171875, 0.1904296875, -0.4637565612792969, 0.24735641479492188, 2.485321044921875, -4.342864990234375, -0.93402099609375, 1.4864501953125, -1.97906494140625, 0.3841114044189453, 3.49462890625, 0.424560546875, -0.06909942626953125, 0.49956512451171875, 1.40826416015625, -3.96832275390625, 1.1602325439453125, 2.2633056640625, 3.1732177734375, 0.04486083984375, -0.19316864013671875, -0.07305145263671875, 1.876953125, 0.13535308837890625, -0.2802734375, -0.004150390625, -0.116241455078125, 0.74566650390625, -1.697265625, -0.39276123046875, 0.026824951171875, -4.39190673828125, 0.8022918701171875, 3.510498046875, -0.8983001708984375, 1.156280517578125, 1.383544921875, 0.8456268310546875, 0.27207374572753906, 3.193695068359375, 0.7559661865234375, 0.9182929992675781, -0.4503669738769531, -0.063446044921875, 0.47928619384765625, -0.4468498229980469, 3.9619140625, 0.562713623046875, 0.1408843994140625, -4.36810302734375, 0.00897216796875, 0.265655517578125, -0.329345703125, 1.325927734375, -1.387420654296875, -0.154296875, -0.5717926025390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000049.npy"}
|
|
{"epoch": 0.10261780104712041, "step": 50, "batch_size": 128, "mean": 0.36489278078079224, "std": 1.618477702140808, "min": -4.44482421875, "p10": -1.4258972167968749, "median": 0.2673187255859375, "p90": 2.240673828125, "max": 7.711395263671875, "pos_frac": 0.578125, "sample": [0.94091796875, -0.9041290283203125, -0.3914794921875, -0.010686874389648438, -0.28509521484375, -0.6542434692382812, 0.24725341796875, -0.13448715209960938, 0.575347900390625, 7.711395263671875, -0.7183837890625, -0.1239156723022461, -0.30859375, 1.209075927734375, 2.972412109375, 1.2309646606445312, 1.8546142578125, -2.967254638671875, 1.4713592529296875, -0.8605422973632812, -0.9036865234375, -4.44482421875, -1.382080078125, 0.7919464111328125, 2.136077880859375, 2.735748291015625, 1.4042129516601562, -0.00815582275390625, 2.2314453125, -0.136077880859375, -1.917755126953125, 2.26220703125, 0.5331878662109375, -1.52813720703125, 0.4793701171875, -0.69482421875, -0.2353973388671875, -0.9638671875, -1.318603515625, 0.38128662109375, -0.60833740234375, -2.235443115234375, 1.671295166015625, 1.87982177734375, 0.07568359375, -0.017547607421875, 0.64764404296875, 1.27728271484375, -0.38741493225097656, 0.11141204833984375, 1.163330078125, 0.0928955078125, 0.055267333984375, 0.3938560485839844, 0.9524765014648438, 1.81610107421875, 2.49658203125, -0.927520751953125, 1.5440826416015625, 1.917236328125, 1.0242767333984375, 1.1601409912109375, -3.13909912109375, -0.028179168701171875, -0.097076416015625, -0.989471435546875, -2.20635986328125, 0.3565254211425781, 1.57952880859375, 0.08582687377929688, 1.060546875, -0.5202121734619141, -0.10027503967285156, 0.5246429443359375, -0.0594329833984375, -2.76934814453125, 2.27288818359375, 1.0892333984375, 1.72296142578125, -0.06588554382324219, -0.446533203125, -0.4151611328125, -2.285797119140625, -0.9409027099609375, 1.0030899047851562, 3.4193572998046875, -2.0732421875, -2.30352783203125, 1.3115005493164062, -2.990814208984375, 0.287384033203125, 0.195068359375, 2.703369140625, 1.079071044921875, 1.02276611328125, 0.092437744140625, -0.39300537109375, 0.6988372802734375, 1.6332855224609375, -0.3111114501953125, 0.7200927734375, 1.408233642578125, 1.8509674072265625, 1.323883056640625, 0.325958251953125, 0.5071029663085938, 0.4473457336425781, 1.619476318359375, -0.7109909057617188, -1.326568603515625, 1.3277587890625, 2.3860931396484375, 5.2493896484375, -0.429656982421875, -0.946441650390625, -2.822998046875, 0.883880615234375, 2.788726806640625, 0.2042694091796875, 0.4160957336425781, 1.61077880859375, 0.2034912109375, 0.97113037109375, -0.057220458984375, 2.306060791015625, -0.117828369140625, -0.21194076538085938, 2.396575927734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000050.npy"}
|
|
{"epoch": 0.10471204188481675, "step": 51, "batch_size": 128, "mean": 0.628907322883606, "std": 1.8532851934432983, "min": -5.44134521484375, "p10": -1.3327667236328122, "median": 0.43032264709472656, "p90": 3.050820922851562, "max": 5.5799560546875, "pos_frac": 0.6328125, "sample": [1.28802490234375, 0.8994522094726562, 1.5036029815673828, 0.9042472839355469, 0.9013080596923828, -2.1586227416992188, 2.908660888671875, 3.30535888671875, -1.54949951171875, -0.5023193359375, 1.1987762451171875, 2.129791259765625, 1.48828125, 3.927337646484375, -0.9631538391113281, -1.2587127685546875, 0.43487548828125, -2.43768310546875, 1.045196533203125, 2.33795166015625, -5.44134521484375, 0.736724853515625, 1.2328338623046875, 0.2279815673828125, 2.456939697265625, 5.5799560546875, -0.6001434326171875, -1.22027587890625, 3.917572021484375, -0.2001190185546875, -1.2294921875, 2.81268310546875, 3.884490966796875, 0.927764892578125, 1.3656463623046875, 4.289306640625, 2.90997314453125, 1.1593856811523438, -0.3477783203125, 0.019008636474609375, 0.7711105346679688, 1.3508758544921875, -1.2209320068359375, 1.971282958984375, -0.37332725524902344, 2.905426025390625, -0.4072113037109375, 0.4257698059082031, -0.63275146484375, -3.71185302734375, 2.595458984375, 0.653961181640625, -3.777435302734375, 2.02264404296875, 0.401824951171875, 1.8451690673828125, 0.2018890380859375, -0.0036163330078125, -0.0574951171875, 2.7046051025390625, -0.3548583984375, 0.97265625, -1.282012939453125, 2.0790634155273438, -2.363677978515625, -2.42388916015625, 0.12145233154296875, 0.7232818603515625, 0.26422119140625, 3.01409912109375, 0.1201019287109375, 0.6668701171875, -0.753448486328125, 3.3914794921875, -1.438446044921875, 0.0, 3.136505126953125, -1.2166900634765625, 2.2158126831054688, 3.263671875, -0.2728271484375, -0.4970664978027344, 0.006099700927734375, 0.949432373046875, -1.51947021484375, -0.299774169921875, -0.5149078369140625, -0.8974380493164062, -1.1683349609375, 2.8214111328125, 0.2295989990234375, -0.081146240234375, -0.2114715576171875, 0.3790130615234375, -0.3026924133300781, 1.716156005859375, 0.0004425048828125, 1.0315093994140625, 1.58795166015625, -1.0386810302734375, -2.653167724609375, -1.2874755859375, 0.35211181640625, 5.27374267578125, 4.419921875, 3.6138916015625, 1.1924896240234375, 1.77520751953125, 1.235107421875, 2.8172607421875, 1.82427978515625, -0.11104202270507812, -1.26922607421875, 1.6025848388671875, 1.4051361083984375, 3.146728515625, -1.46917724609375, -2.9822998046875, 1.22265625, 0.97723388671875, 0.182342529296875, 0.32373046875, 0.2711334228515625, -0.4875640869140625, 0.23516845703125, 0.8385772705078125, -0.116943359375, 0.5643463134765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000051.npy"}
|
|
{"epoch": 0.1068062827225131, "step": 52, "batch_size": 128, "mean": 0.9603263139724731, "std": 1.9496344327926636, "min": -4.09466552734375, "p10": -1.21646728515625, "median": 0.8086280822753906, "p90": 3.5725433349609372, "max": 7.336822509765625, "pos_frac": 0.71875, "sample": [1.25115966796875, 2.56646728515625, -2.29864501953125, -3.136962890625, 0.404022216796875, -0.4649505615234375, 3.44354248046875, 0.5785903930664062, -0.3419170379638672, 5.09765625, 0.56622314453125, 1.105255126953125, 0.0048370361328125, -4.09466552734375, 1.331787109375, 6.17156982421875, -0.9333343505859375, 1.353790283203125, -0.28786468505859375, 0.20993804931640625, 1.697418212890625, 0.3659172058105469, 1.8098602294921875, 0.9649467468261719, 0.4426422119140625, -0.92669677734375, -1.38885498046875, 4.19403076171875, 2.241302490234375, -2.168212890625, -1.27252197265625, -0.8382568359375, -0.16445541381835938, 1.5591812133789062, 0.848388671875, 1.28680419921875, 0.0637969970703125, -1.70977783203125, 0.008762359619140625, 0.35455322265625, 4.77471923828125, 0.4298439025878906, -1.3941650390625, 0.893951416015625, 1.92340087890625, 0.0307769775390625, 0.9948577880859375, 6.10443115234375, 1.271240234375, 1.9366455078125, 1.0049819946289062, 2.620452880859375, 0.91558837890625, 1.0992889404296875, -0.1669750213623047, 2.651519775390625, 0.25327301025390625, 0.859100341796875, -0.3702239990234375, -0.0556640625, 0.8065261840820312, 0.11846923828125, 1.98577880859375, -0.47137451171875, 3.55902099609375, 0.88916015625, -1.8564453125, -0.23453903198242188, 1.5518646240234375, -0.14443016052246094, 0.3861083984375, 0.9548282623291016, 2.32080078125, 1.8809661865234375, -1.19244384765625, -1.166107177734375, 1.1873531341552734, -1.10693359375, 0.8861236572265625, 1.371337890625, 2.638671875, 4.6695556640625, 1.12554931640625, 1.4267578125, 3.560302734375, -1.844390869140625, -0.655303955078125, 3.252960205078125, 5.19891357421875, 1.90374755859375, 2.0789642333984375, 0.19777679443359375, -0.288116455078125, 3.601104736328125, -1.824005126953125, 0.4855613708496094, 0.3644561767578125, 4.190643310546875, 1.0492095947265625, -1.1018295288085938, 1.9501953125, 3.754486083984375, 1.4224395751953125, 0.18145751953125, 4.9993896484375, 2.1434669494628906, -2.8319091796875, 0.64337158203125, 7.336822509765625, 2.05438232421875, 0.81072998046875, 4.841278076171875, -0.143341064453125, -0.6211166381835938, 2.597320556640625, 0.01605224609375, 0.4627838134765625, 0.5185794830322266, 0.413665771484375, 1.328460693359375, -0.034637451171875, 1.5032806396484375, -2.31103515625, -0.209869384765625, 0.56475830078125, 0.31163787841796875, 3.46197509765625, 0.3281707763671875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000052.npy"}
|
|
{"epoch": 0.10890052356020942, "step": 53, "batch_size": 128, "mean": 0.8132294416427612, "std": 2.0989110469818115, "min": -6.304656982421875, "p10": -1.3123649597167968, "median": 0.5991363525390625, "p90": 3.3358474731445313, "max": 6.9273223876953125, "pos_frac": 0.65625, "sample": [0.8277587890625, -0.6078338623046875, 1.8692169189453125, -1.8396682739257812, 0.1305999755859375, 1.142120361328125, 6.9273223876953125, -1.949462890625, -0.298553466796875, 0.1374359130859375, 3.243682861328125, -0.778839111328125, -0.369415283203125, -4.227294921875, 0.559844970703125, 2.711029052734375, 2.20684814453125, 0.0145416259765625, 3.08154296875, 1.195831298828125, 0.020721435546875, -0.082305908203125, 0.806488037109375, 1.0645751953125, 4.80938720703125, 4.63690185546875, 2.8744354248046875, 4.3074951171875, 0.431884765625, 3.3121337890625, 0.965728759765625, 5.46966552734375, 0.5020713806152344, 0.1234283447265625, 1.104207992553711, -0.4449958801269531, -0.584686279296875, -3.5516357421875, 0.053619384765625, 1.3595962524414062, 0.92279052734375, -0.06621170043945312, 0.19845199584960938, -2.3228759765625, -0.14069175720214844, 0.27081298828125, -1.46551513671875, -0.590972900390625, 3.259490966796875, -2.41796875, 0.81060791015625, 6.6510009765625, 1.2307510375976562, -2.43865966796875, 0.6679229736328125, -1.2819900512695312, 1.92413330078125, 0.8387184143066406, -0.5567550659179688, 3.3339385986328125, 2.661865234375, 0.5951881408691406, -0.12994384765625, 0.619171142578125, 1.510223388671875, -0.26297760009765625, 1.235107421875, -0.23099517822265625, 0.6030845642089844, 0.5251998901367188, 0.5678253173828125, 2.38006591796875, 3.434661865234375, 0.757110595703125, 3.172210693359375, 2.22857666015625, -0.9852752685546875, 0.82177734375, 3.55560302734375, -0.31219482421875, 3.340301513671875, 0.513336181640625, 0.633758544921875, -0.402557373046875, -0.6735305786132812, 4.901611328125, -5.03759765625, 2.289642333984375, 0.3523540496826172, -1.38323974609375, 1.4204788208007812, -0.4397125244140625, 2.612518310546875, -1.458709716796875, 1.78839111328125, 6.84979248046875, 0.1411285400390625, -1.20025634765625, -0.16841888427734375, -0.1309661865234375, -0.504730224609375, 2.06610107421875, -1.0902786254882812, -1.8387451171875, -0.2114105224609375, -0.49562835693359375, -0.39190673828125, 2.7729034423828125, 0.9486770629882812, -6.304656982421875, -0.13037872314453125, 0.8465499877929688, 3.894073486328125, 1.59747314453125, 0.6295166015625, 1.07757568359375, 2.88275146484375, 0.0, 2.66790771484375, 0.375213623046875, 1.842681884765625, 0.5080413818359375, 1.3823776245117188, -0.885223388671875, 0.894134521484375, 1.0015945434570312, 0.178619384765625, 3.703125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000053.npy"}
|
|
{"epoch": 0.11099476439790576, "step": 54, "batch_size": 128, "mean": 0.7389856576919556, "std": 2.1060919761657715, "min": -6.185699462890625, "p10": -1.7823699951171872, "median": 0.5434341430664062, "p90": 3.4836669921874996, "max": 7.2138671875, "pos_frac": 0.6953125, "sample": [3.113525390625, 3.712493896484375, 1.84393310546875, 1.791595458984375, 1.878753662109375, 0.971160888671875, 0.106903076171875, 4.4132080078125, -0.3068046569824219, -1.920135498046875, 0.970458984375, -1.2242431640625, -0.3603401184082031, -1.969757080078125, 4.4036865234375, 0.01247406005859375, 5.04473876953125, 0.15821075439453125, 0.214324951171875, 0.8048171997070312, 1.0488739013671875, 3.553863525390625, 0.9724578857421875, 1.083404541015625, -1.938201904296875, -0.04660797119140625, 2.91998291015625, -0.337066650390625, 1.1173095703125, -0.037899017333984375, -0.0478515625, 0.189971923828125, -1.023101806640625, 0.5131301879882812, 0.85137939453125, 2.6353759765625, 4.82421875, -4.2626953125, 1.7650146484375, 0.46392059326171875, 1.1596755981445312, 0.5737380981445312, 4.46038818359375, -0.807708740234375, 5.50433349609375, 0.0449371337890625, -3.4993743896484375, 0.374359130859375, -0.8088150024414062, -1.72332763671875, 2.003143310546875, 5.867218017578125, 0.7584304809570312, 0.2493438720703125, -0.25463104248046875, -0.12009048461914062, -0.358642578125, 1.3642349243164062, 0.842620849609375, 0.32575225830078125, 1.5937423706054688, 1.4982452392578125, -1.357940673828125, -0.73382568359375, 2.81805419921875, 1.5347824096679688, 0.924713134765625, 0.6500091552734375, 2.3438873291015625, 2.195953369140625, -0.6693382263183594, 1.1752395629882812, 0.502166748046875, 1.5029754638671875, -0.018585205078125, 0.5009384155273438, 0.99212646484375, -0.517852783203125, 0.8599472045898438, 0.3680572509765625, -0.3614082336425781, -1.0666961669921875, 1.1178436279296875, 0.8177413940429688, 0.77008056640625, 0.631256103515625, 0.225616455078125, -0.8814849853515625, 2.3987388610839844, 0.5129852294921875, 0.509124755859375, 2.86053466796875, -3.2635498046875, 3.194671630859375, 0.7637481689453125, 0.09583282470703125, -2.9635009765625, 7.2138671875, 5.5628662109375, 2.94482421875, 0.42462158203125, 3.923492431640625, 2.6739501953125, 0.374481201171875, 3.453582763671875, 0.3693351745605469, -3.472564697265625, 0.400360107421875, -6.185699462890625, 3.09228515625, -0.038726806640625, -2.3773345947265625, -1.64849853515625, 0.240692138671875, 4.41754150390625, -0.170379638671875, -2.0144882202148438, 0.05010986328125, 1.58526611328125, -0.1302032470703125, 0.727203369140625, 0.486572265625, 1.2604103088378906, -3.7498779296875, -2.9080810546875, 1.11517333984375, 0.7574729919433594, 0.8270111083984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000054.npy"}
|
|
{"epoch": 0.1130890052356021, "step": 55, "batch_size": 128, "mean": 0.5742762088775635, "std": 1.9454877376556396, "min": -6.48779296875, "p10": -1.28092041015625, "median": 0.3409881591796875, "p90": 3.052760314941406, "max": 8.529205322265625, "pos_frac": 0.6015625, "sample": [-1.067474365234375, 2.535858154296875, -0.251983642578125, 0.83709716796875, -0.857574462890625, -0.70147705078125, 0.5217990875244141, 1.217437744140625, 1.64849853515625, -2.7166595458984375, 3.9283905029296875, 2.032958984375, 2.6704559326171875, -0.2275390625, 0.5089569091796875, 0.04478263854980469, 1.1259765625, 0.55303955078125, -0.06445693969726562, 0.704193115234375, 0.13539886474609375, 1.9472198486328125, 1.19317626953125, 2.3901824951171875, 1.5287017822265625, -0.1060791015625, 0.466552734375, 1.26702880859375, 0.4547882080078125, -0.7420654296875, -0.13726806640625, 0.121490478515625, -0.716522216796875, 3.411712646484375, -0.122039794921875, 1.493927001953125, -0.692169189453125, 2.151519775390625, -0.5538482666015625, 3.635528564453125, -3.7225341796875, -0.4039459228515625, -0.5985870361328125, 0.98199462890625, -0.19146728515625, 1.3759002685546875, -0.455352783203125, 1.0710906982421875, 0.3662109375, 4.56585693359375, 0.0471038818359375, -2.0941009521484375, -3.0933837890625, 0.06397247314453125, 0.568084716796875, -2.351043701171875, -2.4550018310546875, 1.809844970703125, -0.627044677734375, 1.36553955078125, -0.9102935791015625, 0.5966033935546875, 4.934783935546875, -0.24216842651367188, 1.638153076171875, -0.3375244140625, -1.9624557495117188, 3.393951416015625, 1.069854736328125, -1.6202278137207031, -0.1752166748046875, 0.540618896484375, -0.040618896484375, 0.1992645263671875, -0.9601364135742188, -0.10837936401367188, -0.747955322265625, 2.313568115234375, -0.15155410766601562, 1.3244171142578125, -0.42353057861328125, 4.51995849609375, -3.09136962890625, 0.4017982482910156, -0.0777740478515625, 0.28302764892578125, -0.5153312683105469, 3.1544342041015625, 2.989715576171875, 0.99383544921875, -0.1789703369140625, 2.4369354248046875, -0.83355712890625, 0.37042236328125, 8.529205322265625, 0.062114715576171875, 3.31463623046875, -1.1738204956054688, 0.48409271240234375, 2.74664306640625, -1.684051513671875, 0.99749755859375, 0.5604591369628906, 3.44000244140625, 0.23252105712890625, -2.8753814697265625, 2.51434326171875, 1.6085052490234375, 2.0907745361328125, -0.8636245727539062, -6.48779296875, 0.19379806518554688, 1.6113128662109375, -0.096954345703125, -1.2401885986328125, 0.017486572265625, 3.009185791015625, -1.3759613037109375, 1.0473518371582031, -1.171661376953125, 0.9903488159179688, 0.02976226806640625, -0.126373291015625, 1.90142822265625, 0.315765380859375, 5.50830078125, 3.9597930908203125, 0.86090087890625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000055.npy"}
|
|
{"epoch": 0.11518324607329843, "step": 56, "batch_size": 128, "mean": 0.9824748635292053, "std": 2.523247718811035, "min": -6.92828369140625, "p10": -1.5934814453125, "median": 0.7009620666503906, "p90": 3.9203948974609375, "max": 8.5765380859375, "pos_frac": 0.7109375, "sample": [0.94488525390625, -2.7470703125, 3.41943359375, 5.46368408203125, 1.2844390869140625, 1.9815826416015625, 0.92645263671875, 0.619720458984375, 0.316497802734375, 6.37353515625, 2.1411895751953125, 2.960540771484375, 0.16461181640625, -1.307220458984375, 0.8864593505859375, 0.19609832763671875, -0.6904640197753906, 0.05884552001953125, 3.28680419921875, -2.617218017578125, 0.349609375, -0.773773193359375, 2.80047607421875, -1.4327392578125, 0.20192718505859375, 5.692169189453125, 1.25762939453125, 4.50274658203125, 2.9485015869140625, 0.414947509765625, 3.573760986328125, 3.916229248046875, 0.16342926025390625, -1.0, 7.1158447265625, 2.3018646240234375, -3.74359130859375, 0.5027389526367188, 1.0316085815429688, -0.3298683166503906, 5.55328369140625, -0.53326416015625, 0.9217948913574219, 0.08689117431640625, 0.14531517028808594, 8.5765380859375, 2.5976409912109375, 0.18588638305664062, -1.5280303955078125, -0.410369873046875, -1.232818603515625, -1.8079452514648438, 1.745391845703125, -6.00164794921875, 0.3131256103515625, -4.69549560546875, 0.06630897521972656, -0.09521102905273438, 1.11346435546875, 0.2494049072265625, 1.98583984375, 6.55535888671875, 3.61181640625, 2.710845947265625, -0.10308837890625, -1.574493408203125, 3.108978271484375, 2.5336761474609375, 4.1661376953125, 0.7999954223632812, -1.5053787231445312, -0.6099300384521484, -4.325714111328125, 0.4128265380859375, -0.16029739379882812, -0.30572509765625, -2.164825439453125, 3.79595947265625, 0.6138763427734375, 1.532745361328125, -6.92828369140625, 0.459381103515625, 2.3049163818359375, -1.637786865234375, 0.6134033203125, 7.6558837890625, 1.203826904296875, 3.93011474609375, -2.4024505615234375, 1.772491455078125, -0.25026702880859375, 6.06591796875, 0.80419921875, 2.36639404296875, -2.135009765625, 3.30572509765625, 1.2993736267089844, -0.6224365234375, 0.03912353515625, 0.39532470703125, 2.051300048828125, -0.6033935546875, 1.1829376220703125, -0.928863525390625, -3.5831298828125, 4.989501953125, 0.39337158203125, 2.6590576171875, 0.1086273193359375, 1.9677734375, 2.170867919921875, 1.325103759765625, 2.20166015625, -0.9842529296875, 0.05602455139160156, 1.8122711181640625, -0.640106201171875, 0.14602279663085938, 1.434356689453125, 1.4769287109375, 1.618988037109375, -0.42023468017578125, 1.73248291015625, 0.91436767578125, 0.3066253662109375, 3.591766357421875, 2.269500732421875, 0.7822036743164062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000056.npy"}
|
|
{"epoch": 0.11727748691099477, "step": 57, "batch_size": 128, "mean": 1.3333923816680908, "std": 2.1127402782440186, "min": -1.93316650390625, "p10": -1.0641960144042968, "median": 0.9749755859375, "p90": 3.549621582031249, "max": 13.77410888671875, "pos_frac": 0.7578125, "sample": [-1.4534225463867188, 2.787353515625, -1.4094696044921875, 2.35565185546875, 0.11269378662109375, -1.0595703125, 3.9788818359375, -0.09396743774414062, -1.93316650390625, 1.748291015625, -0.255279541015625, 0.5096435546875, 2.16534423828125, 0.4109039306640625, 3.2528305053710938, 0.07599449157714844, 0.923309326171875, 2.301177978515625, 1.23040771484375, 0.5279541015625, 0.9515895843505859, 3.0920867919921875, 0.4893798828125, -0.2045745849609375, -0.3663177490234375, 2.587310791015625, -1.594635009765625, -1.294189453125, -0.874359130859375, -1.0749893188476562, 2.5416259765625, 3.271209716796875, 0.9757080078125, 1.119354248046875, 5.6573486328125, 13.77410888671875, 1.17559814453125, -0.2589111328125, -1.6688995361328125, 2.034912109375, 5.0943603515625, 0.1865081787109375, -0.9508056640625, 4.6502227783203125, 0.8655319213867188, 0.9253921508789062, -0.5592269897460938, -1.743743896484375, -0.599365234375, 2.6856689453125, 1.085906982421875, 0.5211563110351562, 1.36138916015625, 4.940887451171875, 0.9293670654296875, 1.94525146484375, 1.30572509765625, 0.44945526123046875, 2.442596435546875, 0.23760986328125, 1.933502197265625, -0.54248046875, 0.6559638977050781, 2.602142333984375, 1.3961181640625, 3.212890625, 5.0262451171875, 3.057373046875, -1.6865386962890625, 0.077728271484375, 2.5467529296875, 0.532928466796875, -0.7315216064453125, -0.45381927490234375, 2.930908203125, 2.33868408203125, 1.204345703125, 3.280242919921875, 2.0311431884765625, 1.6630859375, 0.272735595703125, -0.07342529296875, 5.350341796875, 4.5865478515625, 0.036529541015625, 0.17218017578125, -1.2132339477539062, 2.5361328125, -1.78656005859375, -0.87139892578125, 2.046051025390625, 4.09234619140625, 0.357025146484375, 0.4395904541015625, 1.847900390625, -0.3906822204589844, 3.47894287109375, 0.30275726318359375, 0.9742431640625, -1.526763916015625, -1.23779296875, 0.280120849609375, 6.1551513671875, 2.5068359375, 0.8283004760742188, 4.6795654296875, 0.9848251342773438, 3.095184326171875, 0.5311737060546875, 0.2895660400390625, 1.649810791015625, 1.2322998046875, 3.016326904296875, -0.1904296875, 1.9926795959472656, 2.177886962890625, 1.21136474609375, 0.3465862274169922, 1.0547866821289062, 0.13325119018554688, 0.9158935546875, -0.9492034912109375, 3.71453857421875, 2.4569549560546875, 3.27734375, 0.29743194580078125, 3.033111572265625, 3.200897216796875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000057.npy"}
|
|
{"epoch": 0.1193717277486911, "step": 58, "batch_size": 128, "mean": 1.0254621505737305, "std": 2.1047592163085938, "min": -11.23828125, "p10": -1.0366851806640625, "median": 1.0725517272949219, "p90": 3.444408416748046, "max": 6.61407470703125, "pos_frac": 0.734375, "sample": [2.31817626953125, 4.56573486328125, 1.501312255859375, 1.767669677734375, 2.0612030029296875, 0.2910804748535156, 1.7159881591796875, -0.37322998046875, 0.0770263671875, 4.81097412109375, 1.930877685546875, 2.267791748046875, 1.2121734619140625, 5.25390625, 1.472503662109375, -0.9843902587890625, 4.3239593505859375, 4.51446533203125, 3.3512649536132812, -1.14410400390625, 6.340911865234375, 3.7396240234375, 0.058135986328125, -1.95428466796875, -0.2965087890625, 1.364410400390625, 2.1660003662109375, 1.755126953125, 0.83526611328125, -2.8907470703125, 0.450286865234375, 2.2097320556640625, 0.1083984375, -0.2466278076171875, 3.01519775390625, 1.0573806762695312, 2.64447021484375, 1.9635696411132812, -0.6449470520019531, -1.03497314453125, -3.5037841796875, 0.73883056640625, 0.6041221618652344, 0.9385223388671875, 2.3604736328125, -1.4508819580078125, 0.4962005615234375, 6.61407470703125, -0.1405353546142578, -11.23828125, -0.217254638671875, 5.10064697265625, -0.607940673828125, -1.7059822082519531, 2.2730712890625, 0.58074951171875, -0.2957611083984375, 1.5470867156982422, -0.87908935546875, 3.25164794921875, 1.15960693359375, -0.26844024658203125, 2.31610107421875, 0.798828125, 2.266387939453125, 1.78857421875, 0.99139404296875, 2.075592041015625, 0.171051025390625, 1.957550048828125, -2.034515380859375, -1.5314102172851562, -1.0326766967773438, 1.95440673828125, 1.1509170532226562, 0.931793212890625, 4.65777587890625, 0.4069175720214844, 1.2454833984375, 0.1331939697265625, 2.2312698364257812, -0.79241943359375, 5.24188232421875, -1.040679931640625, 0.753082275390625, 0.9618072509765625, -0.0813751220703125, 0.486053466796875, 0.28424835205078125, 2.5166015625, 0.4830780029296875, 1.0877227783203125, 0.3597412109375, 1.42413330078125, 1.4246826171875, 0.21929931640625, 2.56817626953125, -0.899383544921875, 2.35888671875, 1.6414794921875, 0.6460418701171875, -0.3885765075683594, -0.01273345947265625, -1.674407958984375, 2.5256805419921875, 0.6709442138671875, 0.8263282775878906, 1.6370697021484375, 4.117973327636719, 1.214385986328125, -1.49554443359375, 1.8018798828125, 1.107574462890625, 3.6617431640625, 1.2353744506835938, 1.1117782592773438, -0.58306884765625, -0.5257568359375, 2.0864028930664062, 1.48284912109375, 1.7491607666015625, 1.2923450469970703, 0.1107177734375, 1.815704345703125, -0.7585983276367188, -1.126800537109375, 1.996337890625, 0.326812744140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000058.npy"}
|
|
{"epoch": 0.12146596858638743, "step": 59, "batch_size": 128, "mean": 0.907480776309967, "std": 2.571038246154785, "min": -6.7550048828125, "p10": -1.4458351135253906, "median": 0.5153961181640625, "p90": 4.319656372070312, "max": 9.80401611328125, "pos_frac": 0.6484375, "sample": [-0.935150146484375, 1.7282562255859375, 0.95843505859375, -6.7550048828125, 1.4453277587890625, 1.8808746337890625, 0.2445068359375, 1.91961669921875, 0.8627471923828125, 0.420074462890625, 0.43038177490234375, 1.792083740234375, 2.0579071044921875, 2.5039215087890625, -4.261138916015625, 0.4556427001953125, 0.12139892578125, 0.01613616943359375, -0.86505126953125, 1.495147705078125, -0.3315887451171875, 0.7222900390625, 0.20377540588378906, -0.04524993896484375, 2.1769866943359375, 1.2061538696289062, 1.8441925048828125, 3.38275146484375, -4.1308135986328125, 0.5064697265625, 0.524322509765625, 1.0202484130859375, 0.1508636474609375, 2.1905975341796875, 0.0, 6.635498046875, -0.21367454528808594, 0.9265899658203125, -0.07495307922363281, -2.822662353515625, 0.22484588623046875, 6.556884765625, 4.8548583984375, 6.302734375, 0.3614349365234375, -3.809844970703125, 2.4422836303710938, 9.80401611328125, 2.703826904296875, 8.65869140625, 2.2784500122070312, 0.848236083984375, 5.2496337890625, 5.07073974609375, 0.0980072021484375, 1.76483154296875, -0.06156158447265625, 2.6039886474609375, -5.896942138671875, -1.4673843383789062, 0.027923583984375, 2.1503067016601562, -0.4334716796875, -0.18719482421875, 0.7690277099609375, 4.170440673828125, 6.019073486328125, -1.24285888671875, 4.212677001953125, 2.3539581298828125, -0.5709304809570312, 1.0951690673828125, -1.7496337890625, 5.761871337890625, -1.9071578979492188, -2.98663330078125, 1.0373096466064453, -3.6492919921875, 0.069091796875, 0.06120109558105469, 2.34088134765625, 1.087890625, 1.53326416015625, 2.8939361572265625, 1.0436325073242188, -0.4257659912109375, 2.503814697265625, -0.6942596435546875, -0.6669921875, -0.4598236083984375, 1.84490966796875, 2.886688232421875, -0.17284584045410156, -1.1737327575683594, -0.888458251953125, 4.56927490234375, 2.350311279296875, 0.0426025390625, -1.0958404541015625, 0.234466552734375, -1.1099853515625, -0.652099609375, 0.8766403198242188, 1.179443359375, -0.15820693969726562, 0.002471923828125, -1.4365997314453125, -3.260467529296875, -0.19441986083984375, 0.45654296875, 0.6025390625, 1.149383544921875, 5.15704345703125, -0.453826904296875, -0.18861770629882812, -0.47949981689453125, 7.07208251953125, 1.4737396240234375, 1.867828369140625, -0.213104248046875, -2.924591064453125, 0.5878677368164062, 2.732452392578125, 1.51556396484375, -0.09572410583496094, -0.35488128662109375, 3.0311279296875, -0.7556381225585938], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000059.npy"}
|
|
{"epoch": 0.12356020942408377, "step": 60, "batch_size": 128, "mean": 1.261034607887268, "std": 2.2859678268432617, "min": -7.39471435546875, "p10": -1.276998710632324, "median": 1.0446853637695312, "p90": 3.9529815673828126, "max": 7.679473876953125, "pos_frac": 0.7109375, "sample": [-3.85211181640625, 3.800689697265625, 4.39129638671875, -1.6215667724609375, 3.542510986328125, 2.1109619140625, -0.12417030334472656, 2.1949615478515625, 3.129730224609375, 0.8079833984375, 0.49878692626953125, -0.3687858581542969, -0.2957878112792969, 2.0879898071289062, 0.3823394775390625, -0.87060546875, 0.584381103515625, -2.2882080078125, 3.4783172607421875, 2.578369140625, 0.9962692260742188, -0.92474365234375, 3.34967041015625, -0.734039306640625, -1.6596565246582031, 1.0497283935546875, -0.05767822265625, 0.0, -0.6246871948242188, 3.74493408203125, -0.43902587890625, 4.650146484375, 1.423187255859375, 2.0301055908203125, 1.1014404296875, 0.2447052001953125, 3.35467529296875, -1.7203369140625, 3.161895751953125, -0.20420074462890625, -0.6407394409179688, 2.3296432495117188, 1.271942138671875, 2.917022705078125, 1.3443222045898438, -0.437225341796875, -1.392404556274414, 0.344390869140625, -0.0923309326171875, 0.31310272216796875, 0.38587188720703125, -1.62762451171875, 2.075958251953125, 5.197845458984375, -0.1334209442138672, -0.944793701171875, 4.061431884765625, -1.2275390625, 1.690277099609375, 3.951141357421875, 0.4411582946777344, 1.7208251953125, 2.8558807373046875, 0.8116436004638672, -7.39471435546875, 2.176239013671875, 0.20635986328125, 1.039642333984375, 0.5964889526367188, -1.519256591796875, -2.1958770751953125, 3.6448516845703125, 3.6432571411132812, 6.27960205078125, 3.00738525390625, 0.8374176025390625, -0.1210174560546875, 2.23626708984375, 1.2795944213867188, 7.502738952636719, 4.5593719482421875, 0.9141387939453125, -0.3341255187988281, 2.020355224609375, 4.3370819091796875, 1.4007110595703125, 1.324432373046875, -4.8900146484375, 1.319406509399414, 3.2909393310546875, 2.3174896240234375, 0.4157295227050781, -0.83306884765625, 0.18414306640625, 0.2618408203125, 3.9902801513671875, -0.36383819580078125, 3.957275390625, 0.46600341796875, 0.02197265625, 1.7315673828125, 3.6019287109375, 0.048492431640625, -1.8042144775390625, 5.031890869140625, 2.6486663818359375, -0.5238418579101562, 2.8970947265625, 7.679473876953125, 0.0, 6.4151611328125, 0.8486328125, 2.7189178466796875, 0.819580078125, -0.6540718078613281, 0.84112548828125, 1.88604736328125, 0.5683670043945312, 2.331329345703125, -3.5587158203125, 1.8202438354492188, 3.3148345947265625, 1.7940635681152344, 2.975311279296875, 2.121246337890625, 0.4226837158203125, 3.293670654296875, 2.43798828125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000060.npy"}
|
|
{"epoch": 0.1256544502617801, "step": 61, "batch_size": 128, "mean": 1.0385408401489258, "std": 2.530125141143799, "min": -7.0372314453125, "p10": -1.7814453125, "median": 0.8018531799316406, "p90": 4.336917114257813, "max": 8.998138427734375, "pos_frac": 0.65625, "sample": [-1.88604736328125, -2.096588134765625, 0.9283657073974609, -0.3038787841796875, 4.172607421875, 4.3560791015625, -0.0123138427734375, 0.45489501953125, 0.001705169677734375, 0.5735855102539062, 0.75738525390625, -5.29962158203125, -1.7130126953125, 1.0701446533203125, 0.5121307373046875, 0.9117698669433594, -3.098907470703125, -0.824493408203125, -2.582305908203125, -0.36962127685546875, 1.308258056640625, 1.1648483276367188, 5.8455963134765625, 3.3734130859375, 4.65325927734375, 0.170684814453125, 4.167633056640625, -0.1534881591796875, 3.752349853515625, 0.864044189453125, 0.045566558837890625, 0.425689697265625, 0.3553810119628906, -0.3687477111816406, 4.743255615234375, 0.39306640625, -0.3341522216796875, 4.328704833984375, -0.300140380859375, 0.78631591796875, -0.7779083251953125, 0.799102783203125, 2.4260711669921875, -0.6252975463867188, 2.693572998046875, 0.2627105712890625, 4.26519775390625, 2.036895751953125, 1.665557861328125, 1.048858642578125, 0.099853515625, 2.6804962158203125, 6.9384765625, 4.388519287109375, 4.1819610595703125, 2.769805908203125, 2.696624755859375, 1.3397502899169922, 6.6767425537109375, -0.78717041015625, -0.4290771484375, 3.129108428955078, 1.963134765625, -1.4097137451171875, 0.446075439453125, 4.15838623046875, 2.605377197265625, 4.815948486328125, 3.259033203125, 0.9969635009765625, 2.47247314453125, 0.4502105712890625, 1.7233467102050781, -0.850006103515625, -1.5072174072265625, 0.377655029296875, -0.25260353088378906, 0.5455322265625, 2.2420501708984375, -1.252838134765625, 6.65252685546875, 1.1631622314453125, 3.0701904296875, 2.506317138671875, 1.6925201416015625, 2.8297348022460938, 0.9053115844726562, -1.77166748046875, -0.726104736328125, -0.614166259765625, -0.3233203887939453, -4.30511474609375, -1.80426025390625, 1.575897216796875, 0.1883258819580078, 1.4732666015625, 1.077392578125, 4.88616943359375, -0.9837493896484375, 5.7374267578125, -0.27215576171875, 8.998138427734375, 2.391815185546875, 0.0, -0.471221923828125, 1.3620147705078125, 2.62750244140625, 0.53509521484375, 1.2820777893066406, -7.0372314453125, 3.375640869140625, -2.839447021484375, -1.059295654296875, -1.2490692138671875, -3.92156982421875, 2.27001953125, -3.158447265625, -0.3804931640625, -0.954925537109375, 2.984375, -3.01171875, 0.8046035766601562, -0.6259918212890625, 1.5642166137695312, 1.4615631103515625, 5.2091064453125, 2.8058319091796875, -2.024139404296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000061.npy"}
|
|
{"epoch": 0.12774869109947645, "step": 62, "batch_size": 128, "mean": 1.1891518831253052, "std": 2.6068332195281982, "min": -6.1116943359375, "p10": -1.8687744140625, "median": 0.7253646850585938, "p90": 5.04232177734375, "max": 9.09832763671875, "pos_frac": 0.671875, "sample": [3.777862548828125, 0.6498641967773438, 8.4132080078125, 2.7044448852539062, 4.12371826171875, 0.16780853271484375, 1.82757568359375, 1.039520263671875, -1.47442626953125, 0.3534393310546875, -0.7705535888671875, -1.2795791625976562, -1.1974334716796875, 3.408355712890625, 3.289794921875, 1.1160507202148438, 1.082244873046875, 2.1481857299804688, -0.3020477294921875, -1.898681640625, 2.0420913696289062, 0.551300048828125, 1.9545211791992188, 0.7644119262695312, -1.85595703125, 5.02313232421875, -3.3837890625, 0.650360107421875, 1.02069091796875, 0.2220458984375, -0.018218994140625, 1.176910400390625, 2.0309371948242188, 1.5856475830078125, 0.15079689025878906, 0.22882461547851562, 1.7078857421875, -0.4800262451171875, 1.3433494567871094, -0.504486083984375, 0.077880859375, 4.501495361328125, 3.0971908569335938, -0.05120086669921875, -0.8841094970703125, 0.3046875, 0.5596656799316406, 3.78900146484375, 3.738494873046875, 9.09832763671875, -6.1116943359375, 5.11376953125, -2.0838623046875, 2.95660400390625, 1.52593994140625, -2.78131103515625, 6.13507080078125, 0.0, 5.47930908203125, -0.2603607177734375, -2.957672119140625, -1.724365234375, 5.08709716796875, -0.27191925048828125, 1.1019287109375, -0.3963775634765625, 0.6863174438476562, 0.5322265625, 0.229156494140625, 0.7843017578125, 4.008544921875, 1.05950927734375, 0.38881874084472656, -2.6320838928222656, 1.5860748291015625, 5.0168304443359375, 0.9134674072265625, -3.348876953125, 0.1968994140625, 1.9146461486816406, 0.491180419921875, -0.7877655029296875, 2.4331207275390625, 2.325439453125, -3.1407470703125, 4.437408447265625, 7.87420654296875, 2.075439453125, 0.229248046875, 5.52459716796875, 0.8646793365478516, -2.572418212890625, 6.00640869140625, 2.607696533203125, 6.17486572265625, 2.861328125, -2.1173095703125, 6.3492431640625, -1.50726318359375, -2.3737335205078125, 5.2647705078125, -0.957611083984375, 0.5326995849609375, 1.671051025390625, 5.3926544189453125, 2.3818511962890625, 1.7869415283203125, 0.8605117797851562, -0.36421966552734375, 4.251556396484375, 2.234344482421875, 0.2086639404296875, 0.499267578125, -2.0037841796875, -0.391815185546875, -0.36965179443359375, -1.1253662109375, -0.6685333251953125, 2.7157058715820312, 0.5851860046386719, -0.071746826171875, -1.174072265625, 4.4140777587890625, 2.10113525390625, -0.53692626953125, 2.069244384765625, -1.54840087890625, -1.0709228515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000062.npy"}
|
|
{"epoch": 0.12984293193717278, "step": 63, "batch_size": 128, "mean": 1.4187986850738525, "std": 2.914341926574707, "min": -8.17926025390625, "p10": -1.8557876586914062, "median": 1.2001819610595703, "p90": 5.1058197021484375, "max": 9.56109619140625, "pos_frac": 0.6796875, "sample": [1.355865478515625, 3.662109375, 0.2139263153076172, -4.8450927734375, -3.5537109375, -0.5750465393066406, 3.088104248046875, -0.4281463623046875, -1.268402099609375, 3.606964111328125, -3.54241943359375, -2.9289703369140625, -2.1879119873046875, -2.3478927612304688, 3.490386962890625, -0.4136199951171875, 2.502471923828125, 5.0909423828125, 7.58270263671875, 7.16778564453125, -2.442413330078125, -1.934295654296875, 9.56109619140625, -0.49810791015625, 2.70318603515625, 2.30987548828125, 1.977752685546875, -3.2846603393554688, -0.3138427734375, 4.0673980712890625, 0.057453155517578125, 1.820556640625, 0.6391143798828125, 0.4950752258300781, 1.2081336975097656, 9.35809326171875, -1.4755859375, 0.8256683349609375, -1.86468505859375, 3.0257568359375, -0.952972412109375, 1.0821666717529297, 3.34210205078125, -3.9790496826171875, 0.767181396484375, 3.0018310546875, 8.50054931640625, -8.17926025390625, -1.8145751953125, 4.45880126953125, 0.12537193298339844, 4.114837646484375, 2.4068603515625, 0.0, 0.08856582641601562, 0.720977783203125, 3.822906494140625, -0.6399154663085938, -0.10799217224121094, 0.867889404296875, 1.1809768676757812, 0.64874267578125, 1.290924072265625, 5.21929931640625, -1.07916259765625, 0.46944427490234375, 2.5257720947265625, 4.279296875, -1.7470855712890625, 4.47216796875, 5.6893310546875, 2.4747772216796875, 4.485626220703125, 1.24627685546875, 5.45513916015625, -1.793212890625, 4.2281036376953125, 1.7098159790039062, 1.9764404296875, -0.954986572265625, -0.7833251953125, 7.7513885498046875, 1.5857086181640625, -1.8519744873046875, 2.7564315795898438, 0.667144775390625, 2.289257049560547, -0.088592529296875, 1.3401718139648438, 2.4799957275390625, 2.6089019775390625, -1.0850830078125, -1.175048828125, 5.140533447265625, 0.6448287963867188, 3.468902587890625, 0.37176513671875, 3.322021484375, 3.416229248046875, 5.146728515625, 3.0902099609375, 2.0904998779296875, 3.695709228515625, -0.8351287841796875, 2.534942626953125, -0.00122833251953125, -0.8426513671875, -1.0524444580078125, 4.304473876953125, 1.0753021240234375, 1.17425537109375, -1.124176025390625, 1.9253387451171875, 3.35760498046875, 1.8947219848632812, 5.97052001953125, 0.46966552734375, 2.0066986083984375, 0.66259765625, 1.192230224609375, -4.379119873046875, -0.2024383544921875, 2.248321533203125, 2.167724609375, 0.418609619140625, -0.2620868682861328, 5.05511474609375, 5.6553955078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000063.npy"}
|
|
{"epoch": 0.1319371727748691, "step": 64, "batch_size": 128, "mean": 1.5104295015335083, "std": 3.095769166946411, "min": -6.27288818359375, "p10": -1.8743186950683592, "median": 1.3211135864257812, "p90": 5.665896606445313, "max": 10.5062255859375, "pos_frac": 0.71875, "sample": [3.7847900390625, -0.8560791015625, -6.27288818359375, 6.19696044921875, -3.169769287109375, 4.00244140625, 2.6957855224609375, 1.8258056640625, 7.39971923828125, -0.02154541015625, 2.65899658203125, -1.275390625, 7.61309814453125, 2.913360595703125, 0.0890350341796875, -4.220062255859375, 7.757568359375, 1.5882186889648438, 1.0576934814453125, -4.697784423828125, 2.004119873046875, 1.3385772705078125, -1.7627410888671875, -1.8658294677734375, 0.487213134765625, -5.9117279052734375, 2.106433868408203, 1.646026611328125, 2.31884765625, 5.075897216796875, 1.30364990234375, 1.59686279296875, 3.87347412109375, 0.7500381469726562, 1.81805419921875, 0.27816009521484375, 0.700775146484375, 3.36199951171875, 2.6119003295898438, 0.6785888671875, 3.962554931640625, -0.4954338073730469, -1.0076675415039062, 2.0803451538085938, 1.0482330322265625, 8.274017333984375, 5.6756591796875, -0.4143829345703125, 6.111114501953125, 1.244049072265625, 4.064208984375, -1.3894500732421875, 0.31109046936035156, 0.81207275390625, 0.7654571533203125, -1.8941268920898438, -3.2535552978515625, 2.2667083740234375, 3.0921783447265625, 2.8441314697265625, 3.3817138671875, 2.579151153564453, 0.430450439453125, 6.0433349609375, 1.465850830078125, 0.18890380859375, -0.29483795166015625, -5.138916015625, 0.2147216796875, 2.74169921875, 0.054443359375, 2.7711944580078125, 9.44287109375, 2.2333221435546875, 0.06612777709960938, 1.679412841796875, 3.06646728515625, 0.68768310546875, -0.4662437438964844, 0.3889923095703125, -1.7668190002441406, 4.228759765625, -0.6101493835449219, -0.2491626739501953, 3.58636474609375, 5.40869140625, 1.4562606811523438, 1.9639968872070312, 1.0610694885253906, -2.59564208984375, 1.8017425537109375, 10.5062255859375, 6.26605224609375, 5.537841796875, 1.0215606689453125, 0.393341064453125, -3.30877685546875, 1.288848876953125, -1.759490966796875, 0.169036865234375, 3.597991943359375, 5.661712646484375, 4.0073394775390625, 0.0, -0.07091331481933594, 0.4725475311279297, -0.7646484375, 1.244415283203125, 3.4214248657226562, -2.777984619140625, 5.463775634765625, 1.7227096557617188, -0.07756805419921875, 6.1861724853515625, 1.5335693359375, -0.4269142150878906, -0.9845046997070312, 6.6260986328125, -4.461517333984375, 0.8833389282226562, 0.0, 2.06378173828125, 5.60394287109375, 1.92138671875, 4.0070343017578125, 3.3816375732421875, -5.9776611328125, -0.435760498046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000064.npy"}
|
|
{"epoch": 0.13403141361256546, "step": 65, "batch_size": 128, "mean": 1.187312126159668, "std": 2.716702461242676, "min": -8.107208251953125, "p10": -2.3033599853515625, "median": 1.0794868469238281, "p90": 4.876705932617186, "max": 7.415771484375, "pos_frac": 0.71875, "sample": [0.3245277404785156, -3.0382080078125, 1.74212646484375, 0.2579612731933594, 6.389556884765625, 1.6347160339355469, 1.76934814453125, 2.6888427734375, 2.9190826416015625, -5.419219970703125, -2.556884765625, -6.121490478515625, -0.4620361328125, 4.22906494140625, -1.162872314453125, 5.880615234375, 4.038116455078125, 0.0126953125, 0.9502143859863281, -0.483551025390625, -3.078155517578125, 0.48598480224609375, 3.5131683349609375, 5.236602783203125, 2.5136966705322266, 3.5358047485351562, 5.05194091796875, 2.2938461303710938, -2.36639404296875, 0.747650146484375, 1.7419281005859375, -0.9350967407226562, 0.0495452880859375, 2.83123779296875, 5.263916015625, -0.4084625244140625, 0.06459426879882812, 4.7718505859375, 0.4703216552734375, -0.15016555786132812, 2.0369873046875, -3.97930908203125, 4.01507568359375, 1.9154891967773438, 4.302001953125, -1.594024658203125, 3.9372406005859375, 1.9500274658203125, 5.410888671875, 0.8032379150390625, 1.779937744140625, 3.254241943359375, -1.7933349609375, 1.777008056640625, 2.2417221069335938, -0.0003662109375, 5.8387451171875, 0.361114501953125, 0.10432052612304688, 2.548614501953125, -2.5826416015625, 1.0369338989257812, 0.6087493896484375, -0.9509124755859375, 1.269622802734375, 0.595550537109375, 1.5466461181640625, -0.4449310302734375, -1.427001953125, 0.580718994140625, 2.7894134521484375, 4.801605224609375, -0.74530029296875, 0.660858154296875, 2.9540863037109375, 0.8544540405273438, -1.2049407958984375, 0.6144142150878906, 0.3473358154296875, -0.6167831420898438, -4.1640625, -1.02734375, 0.0, 1.150421142578125, 2.9734039306640625, 2.6799850463867188, -0.5279083251953125, 1.1573944091796875, 5.7639923095703125, 3.3900146484375, 1.025634765625, 3.94024658203125, 5.917236328125, 2.602996826171875, 4.3504638671875, -2.281524658203125, 4.3823394775390625, 1.5326995849609375, -0.5372695922851562, 7.415771484375, 1.118804931640625, 2.055389404296875, -8.107208251953125, 2.00250244140625, 1.5475845336914062, 5.4029693603515625, 1.5902099609375, 5.77825927734375, 0.74224853515625, 6.53338623046875, 1.5775833129882812, 0.9907073974609375, 2.3654022216796875, 0.4608001708984375, -4.36883544921875, -1.4292793273925781, 3.050689697265625, -3.302093505859375, 0.30859375, 1.6210479736328125, -2.35430908203125, -1.3987884521484375, 1.654266357421875, 0.054107666015625, 3.082244873046875, 0.011156082153320312, 1.0401687622070312, -0.6200714111328125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000065.npy"}
|
|
{"epoch": 0.13612565445026178, "step": 66, "batch_size": 128, "mean": 1.3755590915679932, "std": 3.6460189819335938, "min": -10.924774169921875, "p10": -2.2432174682617188, "median": 1.2580642700195312, "p90": 5.43733673095703, "max": 11.405349731445312, "pos_frac": 0.6796875, "sample": [0.7505569458007812, 1.32470703125, 5.320098876953125, 0.03765869140625, -0.5071487426757812, 6.21978759765625, -0.6650390625, 1.7404098510742188, 0.81756591796875, 1.035247802734375, -0.21213912963867188, -0.057468414306640625, 2.2006568908691406, 1.1398544311523438, 1.17803955078125, -0.3394012451171875, -0.3582878112792969, 2.1283721923828125, 9.346466064453125, 1.9903411865234375, 2.344696044921875, -5.961151123046875, 4.861854553222656, 3.625, 3.066375732421875, 1.840911865234375, -0.299774169921875, 0.23895263671875, -0.5055427551269531, 5.7108917236328125, 1.631988525390625, 0.43648529052734375, 11.405349731445312, -4.65386962890625, 0.584320068359375, -0.5045318603515625, 1.47698974609375, 10.516143798828125, 3.7403106689453125, -1.08087158203125, -5.7355194091796875, 2.3375072479248047, 5.176666259765625, 2.585113525390625, 0.7282485961914062, 4.074737548828125, -2.09429931640625, 7.84014892578125, 6.614471435546875, -2.22021484375, -7.6900634765625, 1.69488525390625, 1.308990478515625, 11.0460205078125, 9.6676025390625, 1.2071380615234375, 0.7596282958984375, 3.441619873046875, -10.924774169921875, -0.2498016357421875, -1.290771484375, 0.3077545166015625, 9.424774169921875, 0.601837158203125, 3.18572998046875, -6.5231170654296875, 0.816741943359375, -1.4345703125, 2.876617431640625, -1.3491287231445312, 0.315948486328125, -1.2427978515625, 6.442718505859375, -3.477691650390625, 4.06439208984375, 1.7115325927734375, 3.16448974609375, -0.9347820281982422, 4.940338134765625, -4.1289825439453125, 3.684478759765625, 0.7889747619628906, -1.951202392578125, 4.262908935546875, -0.537109375, 2.0333709716796875, 0.21142578125, 0.85955810546875, 2.0494384765625, 2.39813232421875, 0.68768310546875, 2.8419189453125, 5.22607421875, -8.88970947265625, -0.3569793701171875, -2.007476806640625, -2.7618408203125, -2.0561065673828125, 2.9437255859375, 2.678314208984375, 5.86871337890625, 2.9565277099609375, 2.63916015625, 0.0, 2.824005126953125, -1.5495986938476562, -0.960845947265625, 6.73382568359375, 1.8561019897460938, 3.56683349609375, 2.003173828125, 2.407623291015625, 0.17834854125976562, -3.918243408203125, 0.606903076171875, 1.4503173828125, -0.7803802490234375, 2.86260986328125, 3.722259521484375, 4.2437591552734375, -3.313873291015625, -0.39764404296875, 2.2529144287109375, -2.2968902587890625, 2.8277587890625, 0.7285232543945312, 4.931884765625, 3.91827392578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000066.npy"}
|
|
{"epoch": 0.1382198952879581, "step": 67, "batch_size": 128, "mean": 1.1753947734832764, "std": 3.4083991050720215, "min": -9.09320068359375, "p10": -2.6726577758789056, "median": 1.1468048095703125, "p90": 5.3160552978515625, "max": 10.0836181640625, "pos_frac": 0.6796875, "sample": [-0.751007080078125, 4.61431884765625, -1.11810302734375, 9.2738037109375, 0.4109764099121094, -5.28680419921875, 0.4035682678222656, 4.987152099609375, 5.4381103515625, -0.4619140625, 1.9170188903808594, 7.7396240234375, 0.5235137939453125, 1.81121826171875, 2.54931640625, 2.328369140625, 2.7842254638671875, 5.31463623046875, -0.9091796875, -0.83319091796875, 1.0541229248046875, -1.2915306091308594, 2.9876251220703125, -0.0214691162109375, 0.2752532958984375, 5.96929931640625, -2.13201904296875, -0.018585205078125, 2.158111572265625, 0.749481201171875, -6.850494384765625, 0.15076828002929688, -1.7548828125, 2.089813232421875, 2.543975830078125, -4.460968017578125, 0.0, 1.723846435546875, 4.007110595703125, 1.9539966583251953, -1.535369873046875, 3.5013427734375, 3.2465896606445312, 6.286773681640625, 0.8603057861328125, -1.5431890487670898, 3.6459197998046875, -2.50396728515625, -5.89056396484375, 0.03267478942871094, -3.603485107421875, 4.9188079833984375, -0.5722427368164062, 6.941162109375, -1.25079345703125, 2.3576507568359375, 7.4896240234375, -4.033233642578125, 1.4854888916015625, 1.1766357421875, 2.76788330078125, 3.0198326110839844, 6.3739013671875, -3.0662689208984375, 2.6003341674804688, 0.06394004821777344, 0.6307029724121094, -9.09320068359375, -1.335693359375, 0.17557716369628906, 5.8150482177734375, 1.035125732421875, 0.5831413269042969, 3.71307373046875, -7.04058837890625, 10.0836181640625, 8.20831298828125, 4.496002197265625, -0.5139846801757812, 1.5015106201171875, 2.3946380615234375, -1.336141586303711, 1.937164306640625, -0.7376708984375, -5.32891845703125, -0.1167144775390625, 5.319366455078125, 0.4969635009765625, 1.116973876953125, 3.5023193359375, 7.362548828125, -8.7298583984375, 4.226898193359375, 2.5284271240234375, -4.21759033203125, 4.549530029296875, 0.262542724609375, 2.47515869140625, 2.3016815185546875, 5.021636962890625, 3.084808349609375, 1.7029647827148438, -0.5058746337890625, 0.8311309814453125, 1.0284423828125, 0.78985595703125, -1.4855537414550781, -4.836334228515625, -1.8302764892578125, 1.2822036743164062, 2.343414306640625, 3.041900634765625, 1.2065277099609375, -1.6560630798339844, -0.8671302795410156, 4.6778564453125, 3.56756591796875, 0.34726715087890625, 3.490631103515625, 1.494964599609375, 3.540740966796875, 1.4907455444335938, 1.900390625, 0.77105712890625, 0.8053054809570312, -1.0220489501953125, 2.47723388671875, -1.1476669311523438], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000067.npy"}
|
|
{"epoch": 0.14031413612565444, "step": 68, "batch_size": 128, "mean": 1.722196340560913, "std": 4.068597316741943, "min": -8.913116455078125, "p10": -3.040636444091797, "median": 1.299142837524414, "p90": 6.515597534179688, "max": 15.33001708984375, "pos_frac": 0.6484375, "sample": [7.531440734863281, -0.06665420532226562, 0.7352294921875, -1.84417724609375, -0.5902061462402344, -0.13446807861328125, 1.70611572265625, 5.97869873046875, -0.2824440002441406, 1.8483963012695312, 1.5724334716796875, -0.3466949462890625, -0.2471923828125, 5.493499755859375, 6.41876220703125, -3.522674560546875, 0.1655731201171875, 0.1205596923828125, 3.21722412109375, 2.579334259033203, 5.763427734375, 0.80126953125, -2.4945144653320312, 1.3569183349609375, 15.33001708984375, 1.2875633239746094, 1.542938232421875, 1.6839599609375, -0.4691658020019531, -5.4547119140625, 4.617828369140625, 4.2447509765625, 10.760101318359375, 8.590301513671875, 0.5890235900878906, 6.549072265625, -0.2718658447265625, 1.9298248291015625, -3.1151123046875, -6.8214111328125, -0.497772216796875, -0.11637115478515625, 4.491970062255859, 3.6083984375, 0.437255859375, -4.774200439453125, -0.1494293212890625, 2.7168045043945312, 5.4549560546875, 6.501251220703125, 2.3622894287109375, -3.2006912231445312, 2.540771484375, 4.92376708984375, 10.79620361328125, -0.6897125244140625, 2.96356201171875, 7.64404296875, -0.8388252258300781, 0.9253616333007812, 4.409759521484375, 2.1378173828125, -0.4733619689941406, 2.243865966796875, 0.31650543212890625, -0.3590888977050781, -8.913116455078125, 4.438507080078125, -0.55859375, 12.7923583984375, 0.06069755554199219, -4.58447265625, 1.6066131591796875, 1.3107223510742188, -6.80462646484375, 2.20269775390625, -0.154296875, -4.009735107421875, 8.945587158203125, -0.330810546875, -2.5126800537109375, 2.1336898803710938, -0.3733978271484375, 3.3178558349609375, 2.78399658203125, -0.304473876953125, -0.992828369140625, 6.188751220703125, -8.7427978515625, 9.081275939941406, 4.667205810546875, 1.1257400512695312, 0.5513916015625, -1.33087158203125, 0.8199462890625, 5.688507080078125, -5.60430908203125, -3.010467529296875, 5.9932861328125, 1.6499557495117188, 4.4972381591796875, 0.8902473449707031, 1.5258636474609375, 8.001129150390625, 8.01824951171875, 1.71759033203125, 5.146759033203125, 5.314239501953125, 2.212860107421875, 4.0193939208984375, 2.823760986328125, -3.1110305786132812, 1.283905029296875, 1.0512237548828125, -0.591583251953125, 1.447509765625, 4.550445556640625, 0.39266204833984375, -0.7243804931640625, 2.0618133544921875, -0.25218963623046875, -2.0557403564453125, 0.6238250732421875, 11.500564575195312, 5.0467529296875, -1.9618606567382812, -1.0841140747070312, 0.8365936279296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000068.npy"}
|
|
{"epoch": 0.1424083769633508, "step": 69, "batch_size": 128, "mean": 2.1263132095336914, "std": 3.8656177520751953, "min": -8.243438720703125, "p10": -1.8362308502197264, "median": 1.5124311447143555, "p90": 6.933737182617188, "max": 14.5982666015625, "pos_frac": 0.7578125, "sample": [5.20855712890625, 0.896240234375, 1.9450721740722656, 5.735076904296875, 0.83172607421875, -8.243438720703125, 4.4145355224609375, 1.6923370361328125, -1.91375732421875, 0.24964141845703125, 4.397087097167969, 2.009166717529297, 9.982025146484375, 0.4154548645019531, 8.17401123046875, -0.6804962158203125, 5.879852294921875, 0.2827320098876953, 2.8204345703125, 0.4521942138671875, -1.295928955078125, 0.3754234313964844, -3.1211700439453125, -5.0380859375, 7.240447998046875, 10.079803466796875, 4.573516845703125, 3.72625732421875, 0.034564971923828125, 2.6446533203125, -7.903289794921875, 4.335906982421875, 0.24393081665039062, 1.38671875, 2.2432937622070312, -0.8388633728027344, 1.4046058654785156, 0.824798583984375, 6.613800048828125, -0.258941650390625, 10.466064453125, -5.9232177734375, -0.0763397216796875, 2.85662841796875, 2.0243682861328125, 1.5078468322753906, -3.64404296875, 6.6922607421875, 3.1952362060546875, 0.4797821044921875, 6.363525390625, 0.70635986328125, 6.936920166015625, 6.23626708984375, 4.27154541015625, -4.8792724609375, -5.468658447265625, 0.7120819091796875, 2.80645751953125, -0.5524559020996094, 0.8872833251953125, 5.778076171875, 0.9091835021972656, 3.2232208251953125, -0.812896728515625, 4.60272216796875, 1.4401092529296875, 5.229248046875, 4.062469482421875, 5.63861083984375, 9.4886474609375, 2.804595947265625, 1.5950698852539062, 0.0416717529296875, 2.991668701171875, -0.4455108642578125, 6.248291015625, 1.5273551940917969, 1.5035858154296875, 2.755279541015625, -1.8030052185058594, 0.3765869140625, 1.5170154571533203, 1.0712776184082031, -1.687957763671875, 0.5917510986328125, 3.4298095703125, 0.9614105224609375, 3.5643768310546875, 10.94268798828125, 6.932373046875, 8.043807983398438, 1.2154769897460938, 4.77557373046875, -3.56793212890625, -2.026702880859375, 0.3227996826171875, 0.498138427734375, 8.88330078125, -0.24421310424804688, 4.2283782958984375, 2.94500732421875, 14.5982666015625, 1.1993408203125, 5.05908203125, 3.2153167724609375, -2.7918701171875, 1.9300689697265625, -1.185028076171875, 1.7316741943359375, -1.235504150390625, -1.1278076171875, 0.5434341430664062, 5.958404541015625, 5.36602783203125, -6.30780029296875, -1.0729141235351562, -1.118682861328125, 3.5775794982910156, 9.595932006835938, 0.84088134765625, 3.339813232421875, 0.7640419006347656, -0.37747764587402344, -0.3102149963378906, 3.993072509765625, 7.931854248046875, 1.0806922912597656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000069.npy"}
|
|
{"epoch": 0.14450261780104712, "step": 70, "batch_size": 128, "mean": 2.0472443103790283, "std": 3.868495225906372, "min": -12.75946044921875, "p10": -1.6016372680664062, "median": 1.5019302368164062, "p90": 6.840914916992188, "max": 15.6068115234375, "pos_frac": 0.734375, "sample": [-1.0697021484375, 3.102752685546875, 2.2873458862304688, -0.931610107421875, -1.710296630859375, 0.8760833740234375, 3.779388427734375, 2.5909042358398438, -4.652099609375, 6.872039794921875, -1.4995269775390625, -0.131561279296875, 0.383758544921875, 5.46044921875, 6.26263427734375, 4.545372009277344, 2.348358154296875, 4.02685546875, 0.26448822021484375, 2.6221923828125, 1.0088653564453125, 6.9068603515625, -0.065704345703125, 2.2754383087158203, -5.38397216796875, 2.76287841796875, 0.814117431640625, 2.23193359375, 2.69091796875, -1.2954635620117188, 7.2576904296875, 0.3686676025390625, 0.2904815673828125, -1.9888916015625, 1.27978515625, 1.1032257080078125, 6.393218994140625, -3.7559051513671875, -4.2711029052734375, 0.1333770751953125, 1.5691680908203125, 0.188690185546875, 10.601318359375, 1.6546630859375, 1.6035537719726562, -0.959686279296875, -0.61688232421875, 1.4973907470703125, 0.0549774169921875, -3.0059814453125, 0.5675048828125, 1.2841415405273438, -1.57635498046875, 0.0327606201171875, -0.19071578979492188, 5.654052734375, 9.406158447265625, 4.781494140625, 6.429840087890625, 1.4684858322143555, 2.0920257568359375, 3.3973388671875, -1.0620574951171875, 1.2435684204101562, 1.222747802734375, 4.53717041015625, 1.5064697265625, 5.22998046875, 1.2980003356933594, 1.690643310546875, -3.19775390625, 1.7734375, 0.22530746459960938, 1.3167724609375, -0.1531829833984375, 7.89300537109375, 0.888916015625, 1.9823646545410156, 9.00958251953125, 3.1522216796875, 14.395660400390625, 6.2576904296875, 2.3238906860351562, 3.357177734375, 4.4471893310546875, 1.7943115234375, -1.4305572509765625, 3.6610107421875, 0.31026458740234375, 10.2181396484375, 6.18255615234375, 5.75299072265625, 0.0, 0.19915008544921875, 2.19879150390625, 7.90765380859375, 5.4066162109375, 3.41943359375, 0.675537109375, 10.5345458984375, 4.12164306640625, 6.82757568359375, 0.46722412109375, -2.174297332763672, 2.6672515869140625, -3.43585205078125, 8.556884765625, 0.7980804443359375, 4.1595306396484375, -1.6606292724609375, -0.04393768310546875, -0.8353042602539062, 1.2063064575195312, 2.469085693359375, -6.48345947265625, -12.75946044921875, 1.368255615234375, -0.9797286987304688, 3.96673583984375, 6.2401123046875, 2.614501953125, 2.0407562255859375, 2.4608154296875, -0.344329833984375, 15.6068115234375, 0.0, -0.29633331298828125, -0.79840087890625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000070.npy"}
|
|
{"epoch": 0.14659685863874344, "step": 71, "batch_size": 128, "mean": 1.9616045951843262, "std": 4.641745090484619, "min": -14.49188232421875, "p10": -3.573047637939453, "median": 2.211758613586426, "p90": 8.142111206054688, "max": 12.7884521484375, "pos_frac": 0.6953125, "sample": [-0.06844329833984375, 10.896209716796875, -1.1632347106933594, -6.08380126953125, -2.2177734375, -9.146026611328125, -0.36429595947265625, 10.51953125, -7.0580902099609375, 3.9581298828125, 1.9591560363769531, 3.6589088439941406, -3.8470458984375, 4.3670654296875, 2.2605972290039062, 5.1263427734375, 5.163604736328125, -0.83319091796875, 3.03131103515625, 9.8916015625, 3.3636398315429688, -0.602020263671875, 2.4302825927734375, 10.388458251953125, 2.32049560546875, -1.230224609375, 1.5497589111328125, 6.676300048828125, 2.4695205688476562, 5.105316162109375, -3.4365692138671875, 5.488792419433594, 2.4019775390625, 10.6046142578125, -0.839111328125, 3.18511962890625, 7.26239013671875, 0.4693756103515625, 6.896209716796875, -5.58233642578125, -1.8155517578125, 0.905426025390625, -5.3995361328125, 2.9516448974609375, 2.192474365234375, -0.1356353759765625, 2.6923828125, -12.977294921875, -3.61083984375, -0.9129219055175781, 1.926483154296875, 4.375640869140625, -0.7057762145996094, 4.732616424560547, 1.0244598388671875, 6.1680145263671875, 9.42474365234375, 2.247589111328125, -0.35271644592285156, -1.046142578125, 4.5050201416015625, -0.961212158203125, 7.33843994140625, 8.39825439453125, 0.3143444061279297, -3.3720703125, -0.2841758728027344, -1.962738037109375, 3.4290771484375, 0.004047393798828125, 1.378509521484375, 10.713043212890625, 4.04046630859375, -4.885498046875, -3.6033554077148438, 3.327728271484375, 8.5294189453125, -8.035675048828125, -2.7038726806640625, 0.34075927734375, 0.55859375, -1.644012451171875, -0.283905029296875, 3.0601577758789062, 6.613861083984375, 7.630950927734375, 1.0029296875, 5.452880859375, 6.4251708984375, 3.21234130859375, 1.5208091735839844, 3.829620361328125, 0.445098876953125, 0.56158447265625, 1.2066497802734375, 11.0712890625, -1.3148880004882812, 8.171783447265625, 2.9898223876953125, 4.77301025390625, 8.12939453125, -3.9590682983398438, 0.6715297698974609, 5.368255615234375, 9.502532958984375, -0.12969970703125, 2.2781219482421875, 0.9615116119384766, -3.56005859375, 1.1096649169921875, 3.457427978515625, 1.1476631164550781, 6.805511474609375, 2.24237060546875, 2.6262359619140625, 2.2310428619384766, 2.9198150634765625, 4.8731536865234375, 1.8402328491210938, 0.842529296875, 2.7247467041015625, 1.1980667114257812, 3.3179931640625, 0.00241851806640625, -0.3078460693359375, 12.7884521484375, -14.49188232421875, 4.04339599609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000071.npy"}
|
|
{"epoch": 0.1486910994764398, "step": 72, "batch_size": 128, "mean": 1.9562186002731323, "std": 4.2105913162231445, "min": -11.04571533203125, "p10": -3.1011589050292967, "median": 1.8680648803710938, "p90": 8.049079895019531, "max": 13.894805908203125, "pos_frac": 0.6796875, "sample": [3.154083251953125, 2.8008270263671875, -3.4649581909179688, 0.7978515625, 10.158935546875, -2.371978759765625, 1.851043701171875, 4.4709014892578125, 2.504486083984375, -0.6043777465820312, -1.3238525390625, 0.353759765625, 1.9013671875, 3.4291839599609375, 3.0467529296875, 9.10882568359375, -2.6102676391601562, 1.251220703125, -2.373321533203125, 2.7872314453125, 3.1870269775390625, 6.0450439453125, 2.106414794921875, 11.28802490234375, 3.6140899658203125, 0.8673171997070312, -0.25946807861328125, 3.6048812866210938, -5.2034912109375, 0.4053192138671875, 3.324371337890625, -11.04571533203125, -2.139617919921875, -4.637977600097656, 2.2941741943359375, 2.721405029296875, 1.82183837890625, -0.634033203125, 1.143585205078125, 1.6283416748046875, 5.3603515625, -2.92816162109375, 8.181610107421875, 5.86334228515625, 8.738189697265625, 2.6806983947753906, 3.3476791381835938, -0.50140380859375, -4.72760009765625, -4.016998291015625, -3.1056289672851562, 5.07928466796875, -3.0992431640625, 0.0, 7.982177734375, 7.58209228515625, 8.3751220703125, 0.14715957641601562, 4.0472412109375, 8.9185791015625, -0.13569259643554688, -3.033447265625, 3.573810577392578, 6.061622619628906, 2.9582748413085938, 5.504364013671875, 1.877532958984375, 0.5358200073242188, -0.3703880310058594, 4.076904296875, -4.71136474609375, 2.3065242767333984, 0.7150497436523438, 3.9052581787109375, 8.160430908203125, 1.3124198913574219, 0.020107269287109375, 6.93524169921875, 0.274383544921875, 9.687774658203125, 2.4956512451171875, 1.9015026092529297, -1.6289825439453125, -2.581939697265625, -2.5545806884765625, -2.781585693359375, 0.888824462890625, -0.102996826171875, -1.9720077514648438, 8.248870849609375, -5.61431884765625, 8.89874267578125, 0.127899169921875, 4.7939453125, 1.8585968017578125, 5.7483978271484375, 6.2218475341796875, -0.46514892578125, 1.7208404541015625, 2.688446044921875, 1.68829345703125, 6.9918365478515625, 0.0, 4.5603179931640625, -3.3254051208496094, 0.7204017639160156, 7.0914306640625, 13.894805908203125, 4.6148834228515625, 6.463233947753906, 4.0773468017578125, -1.898590087890625, -1.34027099609375, -7.707489013671875, 10.44927978515625, -0.7610931396484375, -2.17584228515625, 0.0, 1.3524436950683594, 3.476287841796875, 2.5035934448242188, 2.7456893920898438, -5.343048095703125, 4.343536376953125, 2.206634521484375, -3.6234588623046875, 0.91943359375, 8.001358032226562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000072.npy"}
|
|
{"epoch": 0.15078534031413612, "step": 73, "batch_size": 128, "mean": 1.898528814315796, "std": 4.659677982330322, "min": -18.11981201171875, "p10": -3.2364723205566404, "median": 1.622711181640625, "p90": 7.592080688476562, "max": 13.673492431640625, "pos_frac": 0.703125, "sample": [1.3510932922363281, 1.61114501953125, 7.5833740234375, 5.5149993896484375, -1.196624755859375, 2.2056121826171875, 0.31048583984375, -1.094818115234375, -0.0874481201171875, 6.1009063720703125, 3.45928955078125, 2.545318603515625, 0.918212890625, 9.93780517578125, 0.1994915008544922, 0.27962493896484375, -9.80206298828125, -9.91192626953125, 0.20749664306640625, 1.1215362548828125, 1.89617919921875, 0.004119873046875, 0.5434074401855469, 4.841339111328125, 0.8376541137695312, 0.05584716796875, 2.0163040161132812, 1.3637275695800781, 2.2128448486328125, 5.0334930419921875, -5.71331787109375, -2.5734024047851562, 4.4423828125, 10.25341796875, -1.20806884765625, 1.7943801879882812, -3.3718185424804688, 11.4515380859375, 5.3693389892578125, 2.614501953125, 3.7481536865234375, 0.7118682861328125, 5.3699798583984375, -1.8399124145507812, 12.8414306640625, -1.3337860107421875, -1.4518280029296875, -0.26739501953125, 3.5800247192382812, 11.083465576171875, -6.606048583984375, 1.63427734375, 9.543853759765625, -18.11981201171875, 3.99658203125, 1.4510345458984375, 3.228179931640625, -5.2166900634765625, 6.1146240234375, -2.4294052124023438, 3.56402587890625, -0.6904296875, 3.9002227783203125, -1.37152099609375, -2.5702972412109375, 6.03826904296875, 4.469673156738281, -4.1351318359375, 8.237640380859375, 0.096832275390625, -0.9853134155273438, -1.71026611328125, 4.927764892578125, 4.135833740234375, 0.7764892578125, 2.0676441192626953, -5.5224151611328125, -5.105987548828125, -1.6778411865234375, 1.5058364868164062, 6.72021484375, 5.058929443359375, 4.376007080078125, 1.24664306640625, -2.792144775390625, 2.9945144653320312, -0.048488616943359375, 2.600177764892578, -0.8720703125, 5.564239501953125, 0.6429786682128906, 1.93792724609375, 8.039581298828125, 3.3030319213867188, 8.191192626953125, 6.17681884765625, 2.6544837951660156, 3.433135986328125, -6.9650726318359375, 4.6959991455078125, -0.3315582275390625, 4.8830108642578125, 4.6798248291015625, 4.91748046875, 7.21588134765625, 6.1452484130859375, 0.07257080078125, -2.4071044921875, 13.673492431640625, 6.897796630859375, 2.975341796875, -3.4622802734375, 1.601531982421875, -0.06354713439941406, 9.182632446289062, -3.178466796875, -0.11251068115234375, 7.612396240234375, -3.4827880859375, 0.36496734619140625, 7.565826416015625, 1.4813194274902344, 7.647308349609375, 3.2352294921875, 1.4076461791992188, 0.21260833740234375, 4.98760986328125, -2.7748870849609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000073.npy"}
|
|
{"epoch": 0.15287958115183245, "step": 74, "batch_size": 128, "mean": 2.003473997116089, "std": 4.903952598571777, "min": -12.530853271484375, "p10": -4.318698120117187, "median": 1.7440757751464844, "p90": 8.672845458984375, "max": 15.12982177734375, "pos_frac": 0.703125, "sample": [3.992919921875, 8.9534912109375, 4.19390869140625, 0.19390869140625, 9.206390380859375, -1.08563232421875, -5.1466827392578125, 1.098876953125, 7.2813873291015625, 6.9869384765625, -0.6796875, 9.660079956054688, 0.0, 1.13641357421875, 1.4964752197265625, 9.1962890625, 4.67266845703125, 10.7110595703125, -3.032257080078125, 0.614990234375, 3.842193603515625, 4.633819580078125, 3.265869140625, 1.34161376953125, 8.29217529296875, -5.246917724609375, -4.23797607421875, 0.3592529296875, -2.43218994140625, -3.8529815673828125, -1.14617919921875, -1.2099609375, -4.08709716796875, -0.5846023559570312, 5.82415771484375, 1.382904052734375, 0.5373001098632812, 6.891834259033203, 0.2373638153076172, 8.937652587890625, -0.4787406921386719, -1.314727783203125, -8.436370849609375, -8.2032470703125, 0.5755462646484375, 4.7620849609375, 0.214691162109375, -1.56048583984375, -0.711517333984375, 4.498046875, -10.13372802734375, -7.8489990234375, -1.221435546875, 3.316314697265625, 2.3843841552734375, -1.069986343383789, 1.5253639221191406, 1.4385986328125, -9.41241455078125, 6.4073486328125, 5.4349365234375, 3.6973876953125, 1.085784912109375, 7.9793701171875, 8.1007080078125, 2.396636962890625, 9.67974853515625, 3.07330322265625, 1.3616561889648438, 15.12982177734375, 1.730560302734375, 0.9066009521484375, -1.4850311279296875, 5.933685302734375, -0.6361923217773438, 2.4964599609375, 2.437286376953125, 2.976409912109375, 0.0841064453125, 5.2196502685546875, 2.3871688842773438, -0.0275421142578125, 1.8846435546875, -12.530853271484375, 11.8643798828125, 1.91729736328125, 0.38262939453125, 2.005645751953125, 1.9386482238769531, -0.8101119995117188, -0.28083038330078125, -4.507049560546875, 0.00115966796875, 5.82647705078125, 3.3021011352539062, 4.4193115234375, 2.995513916015625, -8.8209228515625, -5.400848388671875, 4.201324462890625, 4.9014129638671875, -5.6873016357421875, 2.584625244140625, 8.559356689453125, 0.98675537109375, 6.138587951660156, 3.427459716796875, -4.565032958984375, 5.98040771484375, -1.25115966796875, -0.5353126525878906, 1.4265975952148438, 3.306610107421875, 3.1546859741210938, 13.0587158203125, -0.1124267578125, 1.7180404663085938, 4.8198089599609375, 4.76873779296875, 1.307281494140625, 4.57012939453125, 2.3160171508789062, 1.7575912475585938, 9.013504028320312, 13.90625, 4.67327880859375, 0.598876953125, 10.339630126953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000074.npy"}
|
|
{"epoch": 0.1549738219895288, "step": 75, "batch_size": 128, "mean": 2.7650890350341797, "std": 4.443413734436035, "min": -9.81219482421875, "p10": -2.116897583007812, "median": 2.394491195678711, "p90": 8.45962219238281, "max": 14.789306640625, "pos_frac": 0.765625, "sample": [2.439464569091797, 3.0307693481445312, 1.5439529418945312, 1.596405029296875, -2.004425048828125, 6.972869873046875, 6.022705078125, -3.044219970703125, 4.42156982421875, 3.6618576049804688, 0.6698989868164062, -4.37017822265625, 1.3884696960449219, -0.8079185485839844, 9.36175537109375, 0.193695068359375, 14.789306640625, 3.2314453125, 5.826171875, 4.067436218261719, 8.161026000976562, 6.709869384765625, 6.46466064453125, -4.592376708984375, 0.8888931274414062, 11.95001220703125, -1.8980712890625, 7.1240234375, -0.13775634765625, 1.965545654296875, 3.1966552734375, 7.9934234619140625, 6.0723876953125, 3.7187042236328125, 0.08434295654296875, -1.5542926788330078, 3.7427978515625, 4.49212646484375, 3.7686538696289062, 0.880859375, 10.104949951171875, 6.797435760498047, 0.987701416015625, -9.81219482421875, 0.6977691650390625, 1.1007080078125, 0.21558570861816406, -0.0474090576171875, 9.029541015625, 0.839111328125, 4.688262939453125, 4.417816162109375, 2.349517822265625, 0.33984375, 4.49176025390625, 5.083831787109375, -2.64202880859375, 9.0706787109375, -0.5338897705078125, 5.84173583984375, 2.5234375, 2.175647735595703, 0.546112060546875, 7.3016357421875, 1.2917022705078125, 1.8343048095703125, 8.236297607421875, -1.4706497192382812, 1.8515129089355469, 0.970458984375, 2.9940338134765625, -1.682769775390625, 1.7255268096923828, 11.6068115234375, 9.9466552734375, 0.39772987365722656, -8.444244384765625, -6.58233642578125, -0.96844482421875, 4.3054962158203125, 10.984405517578125, 5.4413299560546875, 7.676910400390625, -4.375785827636719, -1.133209228515625, 8.980712890625, 0.612945556640625, 0.330322265625, 2.9366912841796875, 5.011444091796875, 6.2296142578125, -0.9859466552734375, 5.54949951171875, 3.53851318359375, 5.857208251953125, 12.6446533203125, -0.30792236328125, 3.90667724609375, 0.331085205078125, 7.007728576660156, -7.6588134765625, 0.788726806640625, 4.810150146484375, 10.63360595703125, 7.4124755859375, 3.264129638671875, -0.2703704833984375, 3.998443603515625, 1.8259010314941406, -4.728515625, 6.82879638671875, -0.2526721954345703, 0.697021484375, 5.960182189941406, 4.717462539672852, 7.22021484375, -0.64990234375, 1.05523681640625, 4.6773681640625, -2.475677490234375, 2.2131500244140625, 0.62945556640625, 4.681804656982422, 9.098876953125, -2.0410919189453125, -2.2937774658203125, -6.60919189453125, 0.5593719482421875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000075.npy"}
|
|
{"epoch": 0.15706806282722513, "step": 76, "batch_size": 128, "mean": 3.029268264770508, "std": 5.109252452850342, "min": -10.294677734375, "p10": -2.1500579833984372, "median": 2.2644271850585938, "p90": 10.804782104492187, "max": 21.210906982421875, "pos_frac": 0.7578125, "sample": [5.93548583984375, -2.072113037109375, 1.9847946166992188, 1.851837158203125, 3.2199935913085938, 5.4243927001953125, 1.416412353515625, 1.3273239135742188, 2.462484359741211, 3.9766769409179688, 9.731201171875, 0.076324462890625, 0.550689697265625, -2.261077880859375, 14.7164306640625, 5.981353759765625, -0.2906646728515625, -2.6594696044921875, 1.4287872314453125, -1.1266326904296875, 12.663299560546875, 1.3069686889648438, 4.141754150390625, 2.5835800170898438, 3.1603546142578125, 11.99542236328125, 1.458984375, 3.1265182495117188, 1.4333000183105469, 1.073394775390625, 2.611236572265625, -0.7426605224609375, 5.3772735595703125, 8.260269165039062, -10.294677734375, 16.09307861328125, 3.929229736328125, 3.782623291015625, 2.929290771484375, 3.877685546875, 11.12664794921875, 5.37591552734375, 7.04656982421875, 6.5574493408203125, -5.68463134765625, 1.417428970336914, 2.3235626220703125, 3.0717620849609375, 0.2577056884765625, 8.15142822265625, 0.0, 3.8051300048828125, 0.1129150390625, 0.8799972534179688, 7.297798156738281, -2.8602294921875, 11.211700439453125, -6.886474609375, -2.624725341796875, -4.143280029296875, 5.3812408447265625, 11.23388671875, -2.10247802734375, 1.71661376953125, 4.0293426513671875, 1.279397964477539, 4.079925537109375, 4.005455017089844, 5.572662353515625, -1.9767837524414062, 0.22894287109375, 11.08050537109375, -3.4417724609375, 1.2139205932617188, -3.70635986328125, 0.0, 10.686614990234375, 4.295867919921875, 4.8429412841796875, 14.391754150390625, 2.405426025390625, 0.07472610473632812, 0.93524169921875, 2.0516891479492188, 1.3215866088867188, 3.0749053955078125, 1.829864501953125, 6.9215087890625, 2.0740966796875, 3.252410888671875, 4.259368896484375, 1.3269882202148438, -0.23389434814453125, 2.8330230712890625, -0.24227333068847656, -1.1120986938476562, -4.4384765625, -0.983856201171875, 1.7002716064453125, -0.6092987060546875, 5.9776611328125, 0.6644210815429688, 8.047332763671875, 13.076751708984375, 9.99639892578125, 18.309173583984375, 12.668563842773438, 0.9080810546875, 2.602996826171875, -3.5552825927734375, 0.8643646240234375, -1.7417221069335938, 21.210906982421875, -0.738128662109375, -10.090087890625, -0.0032806396484375, 3.27947998046875, 3.2382049560546875, 9.04595947265625, 1.3751220703125, 2.4365196228027344, 7.1220703125, 2.4657440185546875, 2.205291748046875, -0.8192825317382812, 2.8281936645507812, -0.2846832275390625, 0.5248565673828125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000076.npy"}
|
|
{"epoch": 0.15916230366492146, "step": 77, "batch_size": 128, "mean": 2.3984451293945312, "std": 5.40907096862793, "min": -14.6854248046875, "p10": -3.6241775512695313, "median": 1.8541679382324219, "p90": 9.029614257812499, "max": 20.0672607421875, "pos_frac": 0.6640625, "sample": [-0.8019790649414062, 1.9376220703125, 0.846771240234375, 3.9680328369140625, 1.7867431640625, 0.6084136962890625, 8.4415283203125, 5.5395660400390625, -0.2386608123779297, 4.3085479736328125, -3.524993896484375, -0.2572135925292969, 8.95794677734375, 5.50006103515625, 10.7086181640625, 2.0443496704101562, 0.4383544921875, 3.261138916015625, -14.6854248046875, 6.6220703125, -0.3194732666015625, -0.143890380859375, -2.82952880859375, -2.65631103515625, 7.61431884765625, 8.607421875, 5.4452667236328125, 2.3066787719726562, 4.1156158447265625, 4.74993896484375, -4.56585693359375, -0.18206787109375, 1.0756072998046875, 8.4141845703125, 0.02655029296875, 1.388214111328125, 1.3137969970703125, 4.67254638671875, 9.10198974609375, -1.5919876098632812, 1.159088134765625, 0.8785324096679688, 5.056549072265625, 1.34478759765625, 4.275703430175781, 1.82122802734375, 0.5290145874023438, 18.500885009765625, -1.6000823974609375, 2.4212646484375, 5.67828369140625, -0.6180419921875, 3.48907470703125, 8.389190673828125, 8.6820068359375, 5.669647216796875, 4.13262939453125, -1.448760986328125, 0.0340118408203125, 0.09220314025878906, -0.515655517578125, -4.155097961425781, 8.99859619140625, 20.0672607421875, -3.6204833984375, 6.59161376953125, 3.360424041748047, 0.362945556640625, 11.751556396484375, 2.5089645385742188, 10.37115478515625, 3.1239089965820312, -8.886138916015625, -3.6834716796875, -3.6327972412109375, -2.3884429931640625, 5.271575927734375, 1.8427581787109375, 0.87506103515625, -0.209442138671875, -2.618927001953125, 10.733139038085938, -3.1070175170898438, -3.569549560546875, 1.1752853393554688, 4.82415771484375, 9.572860717773438, -7.9595947265625, -4.51373291015625, 4.072013854980469, 2.5810699462890625, 3.2831802368164062, -2.296154022216797, 4.02252197265625, 5.4937591552734375, -3.4779052734375, -1.08551025390625, -7.646575927734375, 9.969757080078125, 3.57958984375, -6.02667236328125, 7.339935302734375, -0.8203639984130859, 2.883270263671875, 6.290901184082031, -0.532958984375, -0.2223663330078125, 11.9775390625, 11.88470458984375, -0.93621826171875, 1.8655776977539062, -5.886566162109375, -4.7975616455078125, 15.8577880859375, -0.25778961181640625, 3.62738037109375, -5.201904296875, 2.779998779296875, -2.827434539794922, 0.53546142578125, 7.747802734375, 2.1839752197265625, 4.5708770751953125, 6.98822021484375, 4.352020263671875, 14.440216064453125, -3.1861572265625, 0.8129043579101562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000077.npy"}
|
|
{"epoch": 0.1612565445026178, "step": 78, "batch_size": 128, "mean": 2.953214406967163, "std": 5.3980841636657715, "min": -12.607101440429688, "p10": -3.1187667846679688, "median": 2.3061161041259766, "p90": 10.896171569824219, "max": 22.6007080078125, "pos_frac": 0.734375, "sample": [-3.052947998046875, 6.375457763671875, -6.12933349609375, 4.834075927734375, 9.230072021484375, 1.4936294555664062, 3.3554306030273438, 4.674835205078125, -3.663299560546875, 5.1755828857421875, 1.1421737670898438, 1.8158302307128906, 4.36260986328125, 4.367218017578125, -0.9702529907226562, 3.081451416015625, 3.7743377685546875, 6.3725433349609375, 10.847305297851562, 0.4495849609375, 0.0, 11.70703125, -3.66986083984375, 3.5725860595703125, 0.9289169311523438, 1.55950927734375, 4.7117919921875, 2.7480697631835938, -2.361907958984375, 5.697410583496094, -2.2028656005859375, 4.7991943359375, 8.9599609375, -0.372650146484375, 7.001861572265625, -4.5100555419921875, 0.3522796630859375, 13.460052490234375, 1.196197509765625, -4.9158935546875, 4.800361633300781, 5.232666015625, -1.2202606201171875, 22.6007080078125, 11.46453857421875, -3.082489013671875, 2.8548583984375, 3.589569091796875, 1.47235107421875, 3.815460205078125, 5.4715576171875, -12.607101440429688, 2.248626708984375, 2.2171173095703125, -0.0211639404296875, -0.5904865264892578, 6.019287109375, 4.4548492431640625, 2.5969619750976562, -1.4097557067871094, 11.9295654296875, -0.178253173828125, 5.122138977050781, 5.6876678466796875, 2.503826141357422, -3.2034149169921875, 0.917755126953125, 3.534820556640625, 3.7197265625, -10.274261474609375, 14.731231689453125, 2.0026397705078125, 1.72589111328125, 0.279052734375, 11.3106689453125, 8.6102294921875, 0.18379783630371094, 2.333904266357422, -1.4423046112060547, 0.7472686767578125, -3.5558319091796875, -1.2388114929199219, 9.285003662109375, 15.058013916015625, 1.67230224609375, 2.5010223388671875, -0.7874717712402344, 4.251678466796875, -5.7620391845703125, 5.71234130859375, 2.9426918029785156, -2.7241058349609375, 9.07403564453125, -1.1709060668945312, -8.134429931640625, 1.308074951171875, 3.154388427734375, -3.7839813232421875, 12.2249755859375, 0.83990478515625, 0.5559654235839844, 1.2902297973632812, -1.4686279296875, 11.01019287109375, 7.244415283203125, 1.9155120849609375, 6.234100341796875, 0.3769378662109375, 1.792327880859375, 2.4484024047851562, 13.43463134765625, 0.236663818359375, 3.31817626953125, 5.0111083984375, 1.498626708984375, 2.6278305053710938, 7.498077392578125, 12.46856689453125, 9.390716552734375, -3.5208892822265625, -2.8370361328125, -0.3662109375, 17.865753173828125, 1.7322425842285156, 1.293731689453125, 2.2783279418945312, -0.9638214111328125, 8.43109130859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000078.npy"}
|
|
{"epoch": 0.16335078534031414, "step": 79, "batch_size": 128, "mean": 3.0336763858795166, "std": 6.734210968017578, "min": -39.93603515625, "p10": -3.154351806640625, "median": 2.4750137329101562, "p90": 9.961065673828125, "max": 23.1495361328125, "pos_frac": 0.6796875, "sample": [-0.3841552734375, 6.365104675292969, 8.056900024414062, -3.128936767578125, 6.399505615234375, -3.096649169921875, 1.6738128662109375, -1.637298583984375, 3.4141998291015625, -2.5566558837890625, 11.43756103515625, -1.142425537109375, 0.0, 8.374267578125, -0.43756103515625, -0.8067092895507812, -0.168609619140625, -1.89312744140625, -6.004737854003906, -0.79718017578125, 2.548919677734375, -2.467620849609375, -3.2707290649414062, -2.216215133666992, -3.704620361328125, 4.4417877197265625, 3.009033203125, 0.0, 9.805908203125, 0.4425830841064453, -7.0703125, 11.117034912109375, 5.5079345703125, 4.8787689208984375, 3.8601913452148438, 1.5227203369140625, 4.196533203125, 2.002838134765625, 3.8207626342773438, -0.487945556640625, 4.970916748046875, -7.6005401611328125, 9.22650146484375, 0.41313743591308594, 0.685272216796875, 5.6124267578125, 6.615501403808594, 15.837127685546875, 2.0552291870117188, 2.4717254638671875, 2.761077880859375, -3.284088134765625, -39.93603515625, -7.060089111328125, 3.5475311279296875, 8.248870849609375, 0.45220947265625, 13.46234130859375, 9.9210205078125, 9.70611572265625, -0.2044677734375, 3.700103759765625, 2.906280517578125, 5.4251251220703125, 11.144119262695312, 4.2845458984375, 3.3165969848632812, 2.375345230102539, -2.0232696533203125, 2.478302001953125, -0.1134033203125, 4.608978271484375, 0.6913890838623047, 1.55438232421875, -3.7457275390625, -4.235137939453125, 8.40093994140625, -0.9466552734375, 9.2989501953125, 1.7441558837890625, -4.0281982421875, 17.869598388671875, 10.05450439453125, 7.75872802734375, 3.097686767578125, 6.61761474609375, 4.55023193359375, 0.5998382568359375, 0.5619659423828125, 0.9549102783203125, 0.28955841064453125, 5.570640563964844, 0.8936614990234375, 0.9547119140625, 2.0217933654785156, 23.1495361328125, -0.27932167053222656, 5.1258392333984375, -4.4550628662109375, 8.228668212890625, 13.8577880859375, 9.28582763671875, 5.612762451171875, 4.2136993408203125, -2.42047119140625, 9.010364532470703, 8.528839111328125, -0.9442253112792969, 8.983139038085938, -1.6123199462890625, -3.213653564453125, 17.464080810546875, 0.8577880859375, -0.6334762573242188, 4.838769912719727, 5.9902496337890625, 6.141815185546875, 6.29461669921875, 0.6883773803710938, 4.9846038818359375, -0.08889007568359375, 9.790008544921875, 12.107421875, 18.981109619140625, -0.8677749633789062, -1.4492645263671875, 0.8075942993164062, 11.1932373046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000079.npy"}
|
|
{"epoch": 0.16544502617801046, "step": 80, "batch_size": 128, "mean": 2.538395881652832, "std": 6.234681606292725, "min": -21.693115234375, "p10": -3.2822784423828124, "median": 1.76446533203125, "p90": 10.880935668945312, "max": 24.35888671875, "pos_frac": 0.671875, "sample": [1.9375724792480469, 4.261848449707031, -1.175567626953125, 1.63250732421875, 1.03466796875, -2.39483642578125, 1.375579833984375, -5.6446533203125, 5.36407470703125, -3.0619354248046875, 0.3240394592285156, -2.290374755859375, 5.322509765625, 3.997314453125, 4.4860076904296875, -2.653076171875, -0.28338623046875, 8.77392578125, 3.3482666015625, 5.10845947265625, 15.973602294921875, -0.78472900390625, 2.40997314453125, 5.27288818359375, 0.7917633056640625, 9.933990478515625, 6.10614013671875, -3.115814208984375, 3.0422134399414062, 10.131988525390625, 2.73468017578125, 0.60748291015625, 0.35833740234375, -1.6494140625, -13.43157958984375, 1.5623779296875, -21.693115234375, 5.74200439453125, 0.0, 14.293701171875, -1.816009521484375, 9.04278564453125, -2.844512939453125, -4.70135498046875, 1.89642333984375, -9.83740234375, -0.194061279296875, 5.6364288330078125, -0.714263916015625, -1.705108642578125, 1.294961929321289, 2.72064208984375, 2.1126785278320312, 6.092010498046875, 7.509613037109375, -0.04392051696777344, -0.028654098510742188, -0.4325408935546875, -1.6323089599609375, 12.4432373046875, 3.3918609619140625, 1.487884521484375, 6.124103546142578, 4.297821044921875, -3.196868896484375, 1.184844970703125, 4.770660400390625, 0.496612548828125, 7.094856262207031, 3.132608413696289, -2.7168045043945312, 5.9603729248046875, 3.18353271484375, -4.062080383300781, 17.9296875, 0.408843994140625, -1.0123672485351562, 0.22846221923828125, 6.085479736328125, -8.11798095703125, 12.012542724609375, 4.48529052734375, -0.607757568359375, 10.84490966796875, 6.04248046875, 2.7575912475585938, -6.2410430908203125, 0.58367919921875, -1.2385101318359375, 15.13348388671875, 6.048667907714844, 2.1871490478515625, 12.481643676757812, 5.58319091796875, 4.237209320068359, 12.4361572265625, 7.3568572998046875, 1.416717529296875, -12.22113037109375, 0.08249282836914062, -7.21441650390625, 10.964996337890625, -3.4815673828125, 5.18865966796875, 3.79595947265625, 11.24993896484375, -0.2530670166015625, -1.2495574951171875, 5.926513671875, 7.6591796875, -1.094635009765625, -0.2947425842285156, 0.37395477294921875, 0.18523406982421875, 0.585662841796875, -5.660003662109375, 3.541584014892578, -1.6677970886230469, 0.8946781158447266, -3.823028564453125, 5.5711212158203125, 12.65972900390625, 9.98663330078125, 4.6703643798828125, 24.35888671875, 0.15074729919433594, 3.610992431640625, 15.67742919921875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000080.npy"}
|
|
{"epoch": 0.16753926701570682, "step": 81, "batch_size": 128, "mean": 3.0399296283721924, "std": 6.605613708496094, "min": -17.96832275390625, "p10": -3.890701293945312, "median": 2.450397491455078, "p90": 11.915283203125, "max": 16.376861572265625, "pos_frac": 0.6953125, "sample": [-0.20523834228515625, -0.12397384643554688, 10.507896423339844, -10.575057983398438, 4.77099609375, -0.5857200622558594, -9.475250244140625, 2.4377212524414062, 15.921539306640625, 8.284576416015625, 16.376861572265625, 12.87152099609375, -2.2773208618164062, -3.7607421875, 8.63433837890625, 0.47264671325683594, 4.529144287109375, 8.33984375, 6.105072021484375, 4.6089935302734375, 0.6447525024414062, 10.517425537109375, 0.7535247802734375, -2.5230865478515625, 4.8936767578125, 3.316009521484375, 2.971954345703125, -0.310791015625, 9.502532958984375, 7.06591796875, 11.86181640625, 4.0941009521484375, 8.534912109375, 5.695831298828125, -1.5081787109375, 3.0486679077148438, -4.193939208984375, 14.283935546875, 13.970504760742188, 2.5967864990234375, -4.40570068359375, 1.30963134765625, 11.112548828125, -0.0471038818359375, -0.978240966796875, 2.1806182861328125, -14.40911865234375, -2.007537841796875, 8.0816650390625, 10.81719970703125, 1.5663909912109375, 4.31884765625, -6.841499328613281, 9.37188720703125, -5.511249542236328, -1.2224922180175781, 5.256195068359375, 4.9089813232421875, 3.697998046875, -15.469070434570312, -5.037353515625, 2.9423980712890625, 4.992919921875, 1.860076904296875, 0.3397064208984375, 3.8238372802734375, 0.349853515625, 0.485260009765625, 2.998504638671875, 3.63140869140625, 0.4552459716796875, -2.1147918701171875, -0.4927215576171875, 12.5191650390625, 13.623291015625, 9.170867919921875, -0.8500213623046875, 4.7458038330078125, 1.9321060180664062, 5.863250732421875, -13.0452880859375, 5.946144104003906, 11.849334716796875, -2.752716064453125, 2.33819580078125, 2.2700119018554688, -0.9565925598144531, -4.508514404296875, -0.076202392578125, 2.2644729614257812, -17.96832275390625, -2.0581512451171875, -13.870697021484375, 0.9120635986328125, 2.46307373046875, 2.0755615234375, 0.0051422119140625, -0.0010814666748046875, 12.0400390625, 14.8077392578125, 12.267822265625, 14.607421875, 4.7183685302734375, -1.9217529296875, 10.36492919921875, 2.409780502319336, -0.71826171875, 1.3659286499023438, 0.38353919982910156, 7.2198486328125, -2.6153182983398438, 5.274223327636719, 0.41950225830078125, -1.1700439453125, 2.5879364013671875, 10.793739318847656, 7.7192230224609375, -1.5118560791015625, 9.403594970703125, 6.5499267578125, 7.18450927734375, 13.29644775390625, 13.9881591796875, 11.104621887207031, 2.07183837890625, 0.9366683959960938, 6.998878479003906, -1.3958587646484375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000081.npy"}
|
|
{"epoch": 0.16963350785340314, "step": 82, "batch_size": 128, "mean": 3.179635763168335, "std": 6.063204765319824, "min": -15.33685302734375, "p10": -3.464209747314453, "median": 3.095348358154297, "p90": 11.48803253173828, "max": 23.52325439453125, "pos_frac": 0.75, "sample": [2.45587158203125, 6.93878173828125, 0.45703125, -7.0743865966796875, 0.225860595703125, 5.1336669921875, 0.1517333984375, 11.566421508789062, 8.00543212890625, 0.1371307373046875, 0.189117431640625, 7.326568603515625, -4.9796905517578125, -1.953460693359375, -2.321502685546875, 6.571258544921875, 5.35247802734375, 2.8235015869140625, 11.454437255859375, 5.547332763671875, 1.2875595092773438, -3.306671142578125, 7.493919372558594, -1.8834381103515625, 0.4306488037109375, 10.359771728515625, -9.985809326171875, 4.359809875488281, 3.49774169921875, 5.40753173828125, 18.21685791015625, 3.111083984375, 11.700698852539062, 5.487884521484375, 11.090621948242188, 14.33038330078125, -1.28448486328125, 5.0475006103515625, 4.08831787109375, -15.33685302734375, -5.6610565185546875, 0.0430908203125, 2.4467926025390625, 3.7448501586914062, -2.4776611328125, 11.585540771484375, 0.05950927734375, 6.56488037109375, 19.0194091796875, -1.920501708984375, -1.0612945556640625, 18.660507202148438, 4.937503814697266, 4.9002532958984375, 2.877044677734375, 0.5382041931152344, 13.0250244140625, -2.865264892578125, 0.7420501708984375, 2.2060546875, 3.6822662353515625, -5.22064208984375, -11.776519775390625, 3.4410934448242188, -4.1702880859375, 1.944244384765625, -3.8554229736328125, 1.7229232788085938, 11.735626220703125, 3.30267333984375, -0.5422611236572266, 6.9319000244140625, 1.309906005859375, 3.1545639038085938, -0.965301513671875, 3.3932647705078125, 0.00640869140625, 1.5590572357177734, 5.197761535644531, 1.8789596557617188, 1.2970848083496094, 23.52325439453125, -2.7796859741210938, 7.30474853515625, -0.332733154296875, 3.1298599243164062, 6.880767822265625, 10.795658111572266, 2.0056686401367188, 2.8584213256835938, 3.0796127319335938, -1.2944793701171875, 2.776927947998047, 4.4942626953125, 6.361186981201172, 7.5225830078125, -6.006591796875, 3.4377288818359375, 5.310791015625, 11.2109375, 4.972442626953125, -0.28617095947265625, 0.7909317016601562, 4.66485595703125, 7.61224365234375, 3.602693557739258, -0.2736835479736328, -1.82611083984375, -4.74615478515625, 1.766876220703125, -2.7560577392578125, 4.797119140625, 6.7484588623046875, 8.36075210571289, -3.6550445556640625, 11.808746337890625, 14.21832275390625, 3.1225433349609375, 3.3038482666015625, 6.5782470703125, 2.8490142822265625, 5.0360107421875, 12.791107177734375, 6.239585876464844, 0.9976806640625, 0.3434600830078125, -10.47772216796875, -3.3824234008789062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000082.npy"}
|
|
{"epoch": 0.17172774869109947, "step": 83, "batch_size": 128, "mean": 3.5498406887054443, "std": 6.986107349395752, "min": -14.82073974609375, "p10": -3.4361236572265623, "median": 2.7264652252197266, "p90": 12.116179656982421, "max": 30.452392578125, "pos_frac": 0.6875, "sample": [6.41796875, 0.0, 11.16619873046875, 4.0343017578125, 3.11761474609375, 0.11346435546875, -0.9016342163085938, 9.6658935546875, 2.19976806640625, -0.04801177978515625, -2.1436920166015625, -13.855926513671875, -0.344970703125, -11.66339111328125, 0.1767730712890625, 0.07196044921875, 4.137542724609375, 11.234161376953125, 8.20703125, 2.5476036071777344, 9.51861572265625, 6.422554016113281, -2.5403900146484375, 5.496673583984375, 10.3526611328125, -3.48211669921875, -1.95013427734375, -1.0103607177734375, -1.18658447265625, 12.356842041015625, 6.1219024658203125, -0.390716552734375, 4.60235595703125, 1.1477584838867188, -6.356536865234375, 15.50537109375, 4.560173034667969, 3.536407470703125, 20.321029663085938, 2.162445068359375, 4.778308868408203, -1.8407211303710938, 2.9053268432617188, 2.9725341796875, -3.497894287109375, 10.546630859375, 4.6849517822265625, 5.1783599853515625, -1.3768310546875, -3.7915191650390625, 6.6026763916015625, 5.40325927734375, 19.06988525390625, 9.1390380859375, -11.860107421875, 0.997528076171875, 11.00274658203125, 3.47119140625, 30.452392578125, 0.308349609375, -2.6847991943359375, 12.013038635253906, 4.61187744140625, 7.15069580078125, 2.285491943359375, -0.072662353515625, 1.41925048828125, 0.30615234375, -4.3788299560546875, 9.051345825195312, -8.112335205078125, 5.9576416015625, 0.1806488037109375, -0.8708038330078125, -3.13970947265625, 3.5110015869140625, 16.134033203125, -3.259979248046875, 3.44879150390625, 9.20880126953125, -3.2022018432617188, 14.519378662109375, 5.16680908203125, 4.99041748046875, -14.82073974609375, -4.99249267578125, 6.487945556640625, 13.30072021484375, -8.91583251953125, 11.5489501953125, -0.4149436950683594, -2.1068649291992188, 0.96319580078125, -4.37432861328125, 5.288265228271484, 15.972625732421875, 15.9666748046875, 0.921142578125, 2.309112548828125, 1.24456787109375, 14.448394775390625, -0.722808837890625, 4.73077392578125, 2.3931884765625, -1.9417266845703125, -0.031612396240234375, 6.2432098388671875, 5.33746337890625, 1.870361328125, -1.27825927734375, 9.614776611328125, -3.416412353515625, 9.448577880859375, 9.3037109375, 2.4405517578125, 5.8648529052734375, 2.35821533203125, 17.58135986328125, 19.2164306640625, -2.6894378662109375, 1.7787017822265625, -0.918304443359375, 9.06195068359375, 0.5747756958007812, 4.179534912109375, 5.984588623046875, 3.3746185302734375, 2.4933624267578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000083.npy"}
|
|
{"epoch": 0.17382198952879582, "step": 84, "batch_size": 128, "mean": 2.802483081817627, "std": 6.952788352966309, "min": -21.7791748046875, "p10": -3.840992736816406, "median": 2.754974365234375, "p90": 12.05281982421875, "max": 19.616943359375, "pos_frac": 0.7109375, "sample": [-1.2171134948730469, 7.0440673828125, 16.6878662109375, -0.2588653564453125, -3.63018798828125, -3.658721923828125, 5.15789794921875, 4.270591735839844, 4.930656433105469, -4.821929931640625, 0.9093399047851562, 8.26171875, 1.497894287109375, -2.6940383911132812, 2.1654052734375, 3.9546966552734375, 5.8910064697265625, 1.0625381469726562, -0.457855224609375, 10.610183715820312, -2.9633712768554688, 7.588470458984375, 2.475006103515625, 3.228515625, 9.73480224609375, 0.242919921875, -7.882568359375, 7.5517578125, 5.36077880859375, 2.6067047119140625, -1.3529052734375, -16.798583984375, -21.7791748046875, 12.2939453125, 2.83074951171875, -1.5563201904296875, 7.10223388671875, 2.298431396484375, 1.09002685546875, -1.2898368835449219, 5.238128662109375, 2.916168212890625, 10.605926513671875, -3.98284912109375, 12.50775146484375, -1.922088623046875, 1.646240234375, -3.7801971435546875, 13.001922607421875, 4.32403564453125, 18.440216064453125, -17.8603515625, 10.2730712890625, 1.940521240234375, 3.2790298461914062, -4.725860595703125, 2.6497802734375, -4.406707763671875, 6.26934814453125, 7.730690002441406, 6.320770263671875, 2.904571533203125, 0.113677978515625, 4.864898681640625, 12.111053466796875, 2.6432037353515625, 5.18939208984375, 0.20221710205078125, 1.03497314453125, 3.6673583984375, -3.77899169921875, 3.435272216796875, 2.907196044921875, -0.03200531005859375, 2.0893516540527344, 12.803924560546875, 8.02386474609375, 8.573799133300781, -12.251800537109375, 4.347900390625, 2.8955230712890625, 1.9409637451171875, -3.5541763305664062, 2.289703369140625, -3.079620361328125, -0.066650390625, -0.70135498046875, 9.634063720703125, 17.427886962890625, 0.8138427734375, 12.027862548828125, 4.0064697265625, -7.72674560546875, 5.533416748046875, 0.2069244384765625, -0.5073356628417969, 3.1457061767578125, 2.478290557861328, 3.8902740478515625, 4.2694091796875, 4.2288818359375, -1.0484123229980469, 2.67919921875, 2.0430068969726562, 9.184696197509766, 5.7471923828125, -3.4266357421875, 19.616943359375, -11.61541748046875, 10.297027587890625, 16.048431396484375, 2.3687057495117188, 6.565195083618164, 5.22802734375, 16.596893310546875, 9.780517578125, -6.84405517578125, -15.204986572265625, 4.666473388671875, -0.707763671875, 0.5807609558105469, 12.70965576171875, 14.012542724609375, -3.2504196166992188, 3.8538742065429688, 5.4701385498046875, 1.415252685546875, -3.0045013427734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000084.npy"}
|
|
{"epoch": 0.17591623036649215, "step": 85, "batch_size": 128, "mean": 2.0604825019836426, "std": 5.759313106536865, "min": -15.99517822265625, "p10": -5.743463897705078, "median": 2.4455461502075195, "p90": 8.896917724609374, "max": 16.1527099609375, "pos_frac": 0.6484375, "sample": [4.4948577880859375, -3.5047607421875, -1.8088035583496094, 16.062774658203125, 6.797359466552734, 4.9801177978515625, -1.9017333984375, -2.234344482421875, 6.3170166015625, 2.22021484375, 1.6787395477294922, -3.02398681640625, 2.2436485290527344, 9.22650146484375, 6.5269775390625, 4.252349853515625, 6.735565185546875, 5.990814208984375, 5.7676239013671875, 3.3918838500976562, -0.1783447265625, 0.152984619140625, 9.741897583007812, 14.473663330078125, 1.9856834411621094, -1.468505859375, -1.530029296875, -4.462638854980469, -7.411346435546875, 5.17596435546875, 6.0582275390625, 3.932830810546875, -1.885528564453125, 3.83294677734375, 7.993988037109375, 10.058731079101562, -4.2870635986328125, 5.221954345703125, -0.8268661499023438, -0.843994140625, 10.877426147460938, 7.077117919921875, -1.848968505859375, 4.7910614013671875, 10.307586669921875, -15.99517822265625, 2.5483627319335938, 3.4922561645507812, 3.352579116821289, 16.1527099609375, 0.47454833984375, -6.110382080078125, 4.896144866943359, 1.58001708984375, 2.480203628540039, -5.850547790527344, 1.71282958984375, 1.8157691955566406, 4.894935607910156, 1.6958770751953125, -4.635467529296875, 5.5392303466796875, 3.105926513671875, -3.631256103515625, 8.782623291015625, 6.965301513671875, 6.100494384765625, 4.55743408203125, -5.6649627685546875, -9.5430908203125, -2.70172119140625, 6.170581817626953, -8.83563232421875, -3.0672607421875, -0.42828369140625, -6.33697509765625, 10.178955078125, 6.99456787109375, -0.0447235107421875, 5.034523010253906, 3.09771728515625, 8.544952392578125, 1.1296234130859375, 6.09735107421875, 16.04541015625, -12.0985107421875, -0.601226806640625, 9.163604736328125, 3.8605518341064453, -4.8511505126953125, -9.818809509277344, 1.9167633056640625, 6.3272705078125, 3.1203136444091797, 9.9102783203125, 2.353494644165039, -0.7416610717773438, 3.4921112060546875, -5.69757080078125, 4.22589111328125, -0.17351150512695312, 4.992431640625, 0.5797462463378906, 0.758148193359375, -0.0334625244140625, 0.12921142578125, -6.5072021484375, 2.906524658203125, 6.866973876953125, 4.117347717285156, -6.5961456298828125, -6.0048828125, -0.901153564453125, 6.71514892578125, -3.608184814453125, 2.37750244140625, 6.3104705810546875, 1.7621917724609375, 14.59210205078125, -7.396888732910156, -0.26788330078125, 4.4231109619140625, -5.214790344238281, 2.410888671875, 3.795013427734375, -2.5630435943603516, 3.0301971435546875, 4.933506011962891], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000085.npy"}
|
|
{"epoch": 0.17801047120418848, "step": 86, "batch_size": 128, "mean": 3.6548938751220703, "std": 7.477386474609375, "min": -16.11871337890625, "p10": -4.007577514648437, "median": 3.2682456970214844, "p90": 12.198757934570313, "max": 25.256439208984375, "pos_frac": 0.6796875, "sample": [4.1798248291015625, -9.86712646484375, 2.316558837890625, -6.84564208984375, 4.4246978759765625, 2.642303466796875, 4.416107177734375, 2.904926300048828, 6.3632049560546875, 5.3964691162109375, 25.256439208984375, -3.611846923828125, 1.52301025390625, 3.80792236328125, 6.726509094238281, -10.882347106933594, -11.33062744140625, -3.84356689453125, 1.0463638305664062, -0.9569435119628906, 2.8195724487304688, 5.944957733154297, -2.6309051513671875, -0.4217529296875, 13.96319580078125, -2.0772552490234375, -14.06768798828125, 1.5936279296875, 19.802978515625, 12.622390747070312, -7.9927978515625, 10.5086669921875, -1.176025390625, 10.9058837890625, 20.9906005859375, 11.62322998046875, 11.619537353515625, -4.609100341796875, 2.584228515625, 12.70526123046875, 3.5101470947265625, -1.9943389892578125, 9.151596069335938, -1.7241363525390625, 3.3065109252929688, 1.4229736328125, 8.203971862792969, 11.7257080078125, 10.7720947265625, 5.818519592285156, -15.227813720703125, 11.70745849609375, 1.87579345703125, 7.4865570068359375, 5.3942718505859375, -2.0339813232421875, 11.042510986328125, -16.11871337890625, 1.9407615661621094, 10.33245849609375, 19.98309326171875, 3.745208740234375, 2.85003662109375, 11.115692138671875, 7.06353759765625, -1.341064453125, 0.0, -2.0145263671875, 3.22998046875, 6.69281005859375, 3.227752685546875, 12.18609619140625, -11.7869873046875, 4.3032073974609375, 2.2647533416748047, 1.05633544921875, -0.1426239013671875, -1.032867431640625, 11.657012939453125, 9.292724609375, -2.6048355102539062, 7.96331787109375, 11.3255615234375, -0.41362762451171875, 5.105766296386719, -8.288925170898438, -9.74139404296875, 8.79150390625, -3.847625732421875, 0.736846923828125, -0.6477317810058594, 5.144073486328125, 1.9785614013671875, 1.6228599548339844, 3.17840576171875, -4.38079833984375, 4.0218505859375, 5.6597900390625, 4.665802001953125, 3.4901981353759766, -1.36602783203125, -2.315032958984375, -2.874786376953125, 12.228302001953125, 12.6925048828125, 8.605880737304688, -0.7605876922607422, 25.129119873046875, 2.2085418701171875, 5.53961181640625, -2.505584716796875, 5.7675018310546875, 13.261444091796875, 5.811614990234375, 12.14501953125, -1.378173828125, 6.9293670654296875, 14.928085327148438, -0.4371452331542969, 6.40411376953125, 5.475860595703125, 10.101055145263672, 8.193862915039062, 12.53228759765625, -1.900543212890625, -1.669586181640625, 1.337738037109375, 2.6649932861328125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000086.npy"}
|
|
{"epoch": 0.18010471204188483, "step": 87, "batch_size": 128, "mean": 4.016026496887207, "std": 6.657456398010254, "min": -18.958892822265625, "p10": -2.602792739868163, "median": 3.36334228515625, "p90": 13.5628173828125, "max": 25.818572998046875, "pos_frac": 0.6953125, "sample": [4.2204437255859375, 8.919769287109375, 10.249603271484375, 8.65740966796875, 16.40380859375, 0.0, -0.11907005310058594, -2.3148765563964844, 7.8425750732421875, 14.25067138671875, -5.077056884765625, 7.992408752441406, -3.27459716796875, 9.703340530395508, 25.818572998046875, -0.23055076599121094, 0.4268951416015625, 6.812347412109375, 15.60211181640625, 3.8415679931640625, -1.24493408203125, -7.6555938720703125, 1.611083984375, 5.1455230712890625, 3.3594970703125, -1.46905517578125, -1.6494140625, 6.44073486328125, 15.84527587890625, 0.37207794189453125, 10.8634033203125, -8.74285888671875, 6.2032470703125, 16.90692138671875, 5.24517822265625, 4.579475402832031, 4.4383697509765625, 4.37744140625, 5.68572998046875, 3.3671875, -0.9550151824951172, -0.081695556640625, 6.624755859375, -3.907318115234375, -1.8388290405273438, -0.7151107788085938, 9.07159423828125, -4.4969482421875, 10.45587158203125, 2.76092529296875, 7.916267395019531, -0.23133087158203125, 12.65673828125, 15.343505859375, 10.943756103515625, 7.817230224609375, 5.2462310791015625, 8.2418212890625, 11.75262451171875, 16.607330322265625, -1.06512451171875, 2.7397613525390625, 1.847503662109375, 9.933250427246094, 3.505950927734375, 21.97454833984375, -1.5731048583984375, 0.0, 3.1139373779296875, 5.38897705078125, 4.2689361572265625, 6.083740234375, 13.551849365234375, 3.9964447021484375, 13.588409423828125, 3.5406341552734375, 3.8507080078125, -0.9779129028320312, 8.63189697265625, -0.23143768310546875, -7.597015380859375, -4.108306884765625, -1.5548629760742188, 0.984130859375, -0.73016357421875, 9.18701171875, -3.508026123046875, 6.5980987548828125, 2.4791412353515625, 5.88427734375, 4.473052978515625, 1.4166412353515625, 3.2417678833007812, -1.022003173828125, 0.809783935546875, -9.08868408203125, 1.1910476684570312, 0.6615982055664062, -6.093109130859375, 12.842742919921875, -0.07924652099609375, 3.96234130859375, -0.4656486511230469, -0.20526123046875, 14.071258544921875, -18.958892822265625, 3.5950775146484375, -5.8766632080078125, 0.1762409210205078, 2.471832275390625, 12.7498779296875, 15.033599853515625, -0.4558563232421875, 8.424285888671875, 0.456787109375, -0.3506927490234375, 9.935394287109375, 2.6458663940429688, 2.203338623046875, 0.5361709594726562, 3.963226318359375, 0.3561859130859375, -1.9566192626953125, 2.4463043212890625, 6.445762634277344, 1.4593048095703125, 15.38873291015625, 1.2254810333251953], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000087.npy"}
|
|
{"epoch": 0.18219895287958116, "step": 88, "batch_size": 128, "mean": 3.3358511924743652, "std": 7.365321159362793, "min": -28.171661376953125, "p10": -4.091219329833985, "median": 2.4151611328125, "p90": 13.269705200195313, "max": 25.29010009765625, "pos_frac": 0.671875, "sample": [8.754352569580078, 14.32763671875, 13.240234375, -0.00732421875, 17.7431640625, 2.9652099609375, -3.5155487060546875, 1.1388397216796875, 8.02783203125, 2.7014083862304688, 1.8588371276855469, -4.731353759765625, 3.194244384765625, 22.20294189453125, -0.09666824340820312, -0.55224609375, 0.24068069458007812, 7.454254150390625, 0.7759323120117188, -2.9661865234375, 13.562606811523438, -4.5135040283203125, 9.0723876953125, -4.120277404785156, -2.75286865234375, 6.86810302734375, -5.201515197753906, 0.9049530029296875, 15.0015869140625, -6.95709228515625, -5.2598876953125, 9.55224609375, 3.725799560546875, 4.55401611328125, 3.6588668823242188, -1.2949333190917969, 20.20318603515625, 0.0, -0.18250083923339844, 2.61602783203125, 6.24407958984375, 5.5484619140625, 10.111312866210938, -0.776702880859375, 14.835128784179688, 13.338470458984375, 3.58251953125, -5.5059814453125, 2.471405029296875, 3.8593292236328125, 0.0, 3.041473388671875, 2.275360107421875, 6.65985107421875, 12.467742919921875, 0.879974365234375, -0.8946571350097656, 1.2978744506835938, -1.075347900390625, 3.92791748046875, 10.474372863769531, -0.27449607849121094, 0.341949462890625, -5.8889312744140625, 5.5159912109375, 1.7728271484375, -4.078765869140625, 10.503936767578125, 0.7625389099121094, 10.66363525390625, 3.340259552001953, -3.66827392578125, 9.66064453125, -0.09870338439941406, -2.855224609375, -19.841796875, 4.514068603515625, 1.464578628540039, -0.452178955078125, -2.4219741821289062, 8.048294067382812, 7.5979156494140625, 8.5413818359375, -1.6904621124267578, 0.387237548828125, -4.8956298828125, -1.7347488403320312, -6.962890625, 6.288047790527344, 7.022247314453125, 0.2860126495361328, 4.56158447265625, 2.358917236328125, 2.3519287109375, 5.03289794921875, 16.489776611328125, 1.103515625, -1.656829833984375, -0.9629974365234375, 17.7266845703125, 4.284423828125, 3.6290969848632812, -1.2193889617919922, 9.808685302734375, -1.343109130859375, 25.29010009765625, 15.18017578125, -0.2828073501586914, 8.74981689453125, -7.854583740234375, 9.048355102539062, 0.9341278076171875, 1.971588134765625, 3.740203857421875, 23.5732421875, 4.057218551635742, -1.1758594512939453, 0.382568359375, 3.5521240234375, 3.541015625, -28.171661376953125, 5.59161376953125, -2.0762939453125, 7.880653381347656, 1.4783248901367188, 7.837066650390625, 4.7589111328125, 0.014373779296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000088.npy"}
|
|
{"epoch": 0.18429319371727748, "step": 89, "batch_size": 128, "mean": 3.73172664642334, "std": 6.784300327301025, "min": -19.07794189453125, "p10": -2.8471908569335933, "median": 2.9953556060791016, "p90": 12.607259368896484, "max": 24.42724609375, "pos_frac": 0.78125, "sample": [0.1974029541015625, -11.4891357421875, 15.731719970703125, 1.5622406005859375, 5.3384552001953125, 16.552734375, -1.2328205108642578, 1.5662994384765625, 1.4569244384765625, 7.87139892578125, 2.40625, 3.42218017578125, 3.4384918212890625, 11.314727783203125, 1.952239990234375, 11.4012451171875, 0.9380664825439453, -3.0621490478515625, 5.7122344970703125, 2.34930419921875, 3.99700927734375, 12.9022216796875, -2.75506591796875, 0.5084743499755859, -0.0733795166015625, 1.7067794799804688, 5.262969970703125, 0.310760498046875, -2.4317626953125, 2.93365478515625, 2.8565826416015625, 2.3418426513671875, -0.598968505859375, 12.823974609375, 13.64324951171875, 24.42724609375, 13.2061767578125, -7.11083984375, 1.4790802001953125, 2.1812744140625, -19.07794189453125, -1.710479736328125, 3.2528076171875, -9.17694091796875, -0.38018798828125, 4.24334716796875, 8.501068115234375, -12.94244384765625, 0.826202392578125, -2.6636199951171875, -0.8393878936767578, 7.091400146484375, 3.825042724609375, -0.027191162109375, 9.762893676757812, 9.69110107421875, 6.243865966796875, -3.254100799560547, 7.068962097167969, 9.590179443359375, 1.9737091064453125, 15.8719482421875, 7.475311279296875, 7.209686279296875, 8.59326171875, 1.0602264404296875, -6.2517852783203125, 0.645233154296875, 0.8866195678710938, -4.1175384521484375, 4.361503601074219, 6.482879638671875, 3.8317489624023438, 6.0089569091796875, -7.1017303466796875, -1.3865432739257812, 2.9670372009277344, 6.748291015625, 23.14886474609375, -6.430267333984375, -0.531341552734375, 6.730987548828125, 3.1510391235351562, 8.523178100585938, 1.2100372314453125, 12.26885986328125, 2.971118927001953, 12.839752197265625, 9.4244384765625, 19.576904296875, 4.799522399902344, -3.62664794921875, 12.514381408691406, 2.7610092163085938, 3.5679702758789062, 3.5401687622070312, 24.019287109375, -1.36480712890625, 1.2312088012695312, 3.9689407348632812, 6.69183349609375, 1.8873786926269531, 3.0155792236328125, 0.04620361328125, 3.6251068115234375, 4.8200531005859375, -1.6707839965820312, 0.0973052978515625, 0.37786865234375, 4.409423828125, 3.1097259521484375, 2.9751319885253906, 9.110191345214844, 13.39141845703125, 6.54693603515625, 4.40997314453125, 3.9732513427734375, 2.8988037109375, 8.623321533203125, -13.297576904296875, 1.7763690948486328, 12.1531982421875, 5.00921630859375, 1.9552383422851562, 2.130891799926758, 3.835357666015625, 2.2533416748046875, -1.11083984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000089.npy"}
|
|
{"epoch": 0.18638743455497384, "step": 90, "batch_size": 128, "mean": 3.412227153778076, "std": 7.372861862182617, "min": -22.80120849609375, "p10": -5.097956848144531, "median": 2.6579360961914062, "p90": 12.607379150390624, "max": 24.020294189453125, "pos_frac": 0.7109375, "sample": [8.301933288574219, 4.6024169921875, 1.1819000244140625, 17.6673583984375, -10.722808837890625, 0.0072021484375, -4.676910400390625, 1.1319580078125, -2.775787353515625, 11.26947021484375, 1.9377307891845703, 1.9371566772460938, -7.5993499755859375, 0.00286102294921875, -0.4355316162109375, 6.267730712890625, 23.299728393554688, 18.33502197265625, 13.31390380859375, 1.6584548950195312, 6.31353759765625, -0.347412109375, 1.1411895751953125, 2.8613739013671875, -12.505859375, -5.0969085693359375, 12.129859924316406, 6.88714599609375, 7.243438720703125, 2.7205810546875, -0.4192962646484375, 7.337898254394531, 9.243438720703125, 3.48223876953125, 7.823394775390625, 2.5192489624023438, -5.23974609375, -5.073028564453125, 3.43115234375, 5.932655334472656, -5.293952941894531, 2.9107513427734375, -0.9215240478515625, -10.0748291015625, 8.111358642578125, 10.1156005859375, 4.613365173339844, 8.822158813476562, 5.5359039306640625, 1.635589599609375, 7.746490478515625, -3.1095199584960938, 0.823974609375, 1.9062728881835938, 0.55816650390625, 0.95849609375, -10.35882568359375, -0.59100341796875, 0.7324066162109375, 5.971290588378906, 0.5152435302734375, 0.0, 12.8697509765625, 6.8806304931640625, 9.981330871582031, -0.379974365234375, 8.4754638671875, 3.1480026245117188, -1.14373779296875, 12.667236328125, 1.3021392822265625, 1.147125244140625, 7.196136474609375, 16.72381591796875, 12.178985595703125, 1.196685791015625, 0.3440990447998047, 3.9005508422851562, 3.1851577758789062, 10.23333740234375, 10.601318359375, -22.80120849609375, 9.17718505859375, -11.058502197265625, 0.41750335693359375, 3.372528076171875, 5.0952911376953125, 2.5952911376953125, -0.3956413269042969, -3.8524627685546875, -1.125762939453125, -5.5943603515625, 11.88702392578125, 10.953704833984375, -3.6540679931640625, 2.434600830078125, -4.9632568359375, 5.141510009765625, 9.412322998046875, 10.14007568359375, -2.10400390625, 11.752197265625, 6.830078125, -0.20302581787109375, -2.61920166015625, 24.020294189453125, -0.8390960693359375, 4.1646728515625, -5.7758331298828125, -12.789703369140625, 5.336669921875, 1.36004638671875, -5.10040283203125, 13.6268310546875, 3.127288818359375, 0.1823596954345703, 12.58172607421875, 3.11590576171875, 14.306655883789062, 2.2264404296875, -1.3240432739257812, -2.289093017578125, 19.46197509765625, 14.2557373046875, 3.2190723419189453, 11.4288330078125, 2.5712890625, 12.8548583984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000090.npy"}
|
|
{"epoch": 0.18848167539267016, "step": 91, "batch_size": 128, "mean": 3.396705150604248, "std": 7.739069938659668, "min": -12.0577392578125, "p10": -6.253641510009765, "median": 3.2162094116210938, "p90": 13.717295837402343, "max": 26.77703857421875, "pos_frac": 0.6328125, "sample": [5.138824462890625, 1.4592514038085938, -1.4334716796875, 16.373916625976562, 8.904449462890625, 14.38568115234375, -0.208892822265625, 4.327842712402344, 5.2024993896484375, -3.4014663696289062, 1.296630859375, 5.77386474609375, 6.8299713134765625, 7.736045837402344, 6.485044479370117, 7.6960601806640625, 0.8988800048828125, -12.0577392578125, -4.077430725097656, -0.2948932647705078, 6.615745544433594, 4.1532440185546875, 5.560020446777344, 7.13079833984375, -1.4374828338623047, 5.704010009765625, 0.052310943603515625, -0.955230712890625, -0.5933609008789062, -8.68621826171875, 13.581253051757812, -0.3068733215332031, -2.0790863037109375, -6.57574462890625, -3.1819610595703125, 8.861053466796875, -5.569358825683594, 5.71343994140625, 6.076995849609375, 6.8383636474609375, -9.757843017578125, -0.817291259765625, 3.2773590087890625, -1.75372314453125, 3.155059814453125, 19.859771728515625, -6.4658203125, -5.5484466552734375, 0.0, -10.823745727539062, -5.796722412109375, -4.923133850097656, -9.476043701171875, 2.62744140625, 6.5696868896484375, 6.653594970703125, -0.39385986328125, -6.166526794433594, 14.655029296875, 5.285125732421875, 10.550811767578125, -8.660598754882812, 3.8315277099609375, -0.20963287353515625, -2.584188461303711, -2.005340576171875, 24.440521240234375, -0.824798583984375, 0.2270355224609375, 10.999343872070312, -8.635101318359375, -2.58282470703125, 26.77703857421875, 3.6909332275390625, 14.03472900390625, 0.5086574554443359, 4.5771636962890625, 1.1430778503417969, 3.46923828125, 12.46881103515625, -9.338050842285156, -4.290069580078125, 22.822967529296875, 10.113014221191406, 15.0467529296875, 0.0, 8.676689147949219, 2.7678985595703125, 8.892333984375, 11.240966796875, 3.2967681884765625, 8.17730712890625, 3.7859268188476562, -6.4569091796875, 0.428680419921875, 12.836700439453125, -7.691764831542969, 2.3697471618652344, 4.045928955078125, 21.04913330078125, 8.731414794921875, 8.400421142578125, 7.51422119140625, 4.80328369140625, 3.54205322265625, -9.564369201660156, -2.95111083984375, 11.889266967773438, -1.4898681640625, 16.87060546875, -3.70550537109375, 18.193359375, 21.130126953125, -0.5398445129394531, -2.474090576171875, 11.115371704101562, 2.320648193359375, -3.33624267578125, 7.834564208984375, 0.5673675537109375, 9.234161376953125, -3.5927734375, 2.00262451171875, 1.5417823791503906, 4.86279296875, 3.369873046875, 6.565460205078125, 0.8553390502929688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000091.npy"}
|
|
{"epoch": 0.1905759162303665, "step": 92, "batch_size": 128, "mean": 3.8129653930664062, "std": 7.977285861968994, "min": -29.488037109375, "p10": -5.840748596191406, "median": 3.5032806396484375, "p90": 14.089506530761717, "max": 22.878692626953125, "pos_frac": 0.7109375, "sample": [-6.914459228515625, 0.27123260498046875, -5.800506591796875, 3.5373001098632812, 3.489471435546875, 6.91925048828125, 4.376350402832031, 4.035125732421875, 12.86767578125, 5.934814453125, 1.0847015380859375, 2.0915756225585938, 4.04437255859375, -0.901397705078125, 19.21844482421875, 19.1094970703125, 18.17578125, -6.70050048828125, 7.3825531005859375, 1.2747726440429688, -6.583076477050781, 7.0924072265625, 14.378524780273438, 12.505096435546875, 3.155517578125, -1.2167816162109375, 3.698577880859375, 12.471435546875, 1.176025390625, -1.365875244140625, 4.9521484375, 0.6492156982421875, 12.377410888671875, 7.838165283203125, 1.7327880859375, 0.045928955078125, -0.136260986328125, 4.5047149658203125, 14.71124267578125, 10.518798828125, 1.626220703125, 1.968292236328125, -0.03512001037597656, -29.488037109375, 12.673934936523438, 2.6822662353515625, 3.369537353515625, 3.149627685546875, -0.123016357421875, 1.5925636291503906, -4.73345947265625, -8.655181884765625, -7.651092529296875, 1.110687255859375, 17.032257080078125, -5.9346466064453125, -9.94415283203125, 0.8690528869628906, -6.28662109375, 6.34326171875, 6.285697937011719, 5.35211181640625, 8.068115234375, 5.6050872802734375, 6.01470947265625, 4.356620788574219, 9.940299987792969, 0.0, -3.154754638671875, 13.621826171875, -2.6964569091796875, 3.326171875, 3.51708984375, -0.7218208312988281, -2.0624923706054688, 8.023468017578125, -1.3294258117675781, 2.2017745971679688, 12.500167846679688, -0.8025398254394531, -7.3248443603515625, 14.02874755859375, 5.9825592041015625, -22.740142822265625, -1.6706161499023438, 1.96868896484375, 6.2174530029296875, 20.809478759765625, 20.73553466796875, 4.9493255615234375, 7.468902587890625, 7.1006011962890625, -0.1878643035888672, -0.518402099609375, -0.20737075805664062, 22.878692626953125, -1.521148681640625, -4.3940887451171875, 19.2587890625, 4.940116882324219, 10.90838623046875, 11.046890258789062, -11.2939453125, 14.231277465820312, 8.140594482421875, 4.121002197265625, 2.437408447265625, 3.0319366455078125, 2.8633575439453125, 4.31964111328125, 9.753814697265625, 3.7254409790039062, 9.91274642944336, -10.35919189453125, 5.6283416748046875, 7.93621826171875, 0.6090488433837891, 0.704925537109375, -1.472381591796875, 8.226226806640625, -2.76611328125, 8.275663375854492, 3.6481246948242188, 2.1491432189941406, 15.7105712890625, 17.048736572265625, 9.805404663085938, -1.6721878051757812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000092.npy"}
|
|
{"epoch": 0.19267015706806281, "step": 93, "batch_size": 128, "mean": 3.4336280822753906, "std": 7.5146002769470215, "min": -19.403900146484375, "p10": -4.740724182128906, "median": 2.4893646240234375, "p90": 14.66195068359375, "max": 24.49407958984375, "pos_frac": 0.6875, "sample": [1.0968246459960938, 12.83746337890625, 1.2181549072265625, 21.92364501953125, -6.10113525390625, 8.219253540039062, -2.3546905517578125, 11.576141357421875, 3.2769775390625, 4.795131683349609, 4.700958251953125, -8.4180908203125, 0.878814697265625, 3.15252685546875, 5.00286865234375, 11.87689208984375, 2.5646820068359375, 16.70623779296875, -2.8323593139648438, 4.847404479980469, 6.1884918212890625, 4.7901763916015625, 3.029062271118164, 4.33380126953125, 15.341133117675781, 2.4140472412109375, -0.3932647705078125, 3.615875244140625, 11.77764892578125, 10.690963745117188, 3.4260025024414062, -8.63336181640625, 0.8799667358398438, -3.7452850341796875, 2.2908172607421875, 6.1672210693359375, 4.8584747314453125, 5.109130859375, -4.080535888671875, 12.959716796875, 1.9750518798828125, 6.2060394287109375, 3.49560546875, -5.04339599609375, 1.089752197265625, 4.158416748046875, 1.9788818359375, 15.6038818359375, 1.503936767578125, -1.0682296752929688, 1.3095932006835938, 6.185943603515625, 17.445037841796875, -6.723968505859375, 4.9920654296875, 1.4581756591796875, -1.4080886840820312, -0.39282989501953125, 9.587188720703125, 0.94635009765625, 6.926788330078125, 4.186126708984375, 14.85711669921875, -2.1343994140625, -1.9268035888671875, -10.20794677734375, 10.141448974609375, 17.00445556640625, -6.8080596923828125, -0.1691741943359375, 3.539661407470703, 3.645101547241211, -3.7538604736328125, -2.0850372314453125, 6.8580780029296875, -2.9326019287109375, 14.57830810546875, 0.07720947265625, 21.430938720703125, -19.403900146484375, 20.0689697265625, 7.9986572265625, 15.0146484375, -7.98834228515625, 5.7722320556640625, 1.7934188842773438, 0.0, -8.7276611328125, 2.123188018798828, 4.3773040771484375, 17.200408935546875, 12.2723388671875, -2.301483154296875, 2.045989990234375, 5.699920654296875, -13.26531982421875, 6.425567626953125, -0.9641551971435547, -2.856292724609375, -6.607915878295898, 9.865020751953125, 8.080230712890625, 2.155059814453125, 0.296295166015625, 5.3253326416015625, 0.0, 1.5430374145507812, -1.3336715698242188, -4.6110076904296875, -0.8624343872070312, -1.0945510864257812, 0.5638198852539062, 5.819114685058594, 1.8077774047851562, -11.60980224609375, 22.47052001953125, 24.49407958984375, -1.9583930969238281, -1.3595504760742188, -0.9691619873046875, 4.38873291015625, 2.63360595703125, -0.790283203125, 13.232635498046875, 0.9499111175537109, 2.9081192016601562, 0.5993576049804688, 5.768463134765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000093.npy"}
|
|
{"epoch": 0.19476439790575917, "step": 94, "batch_size": 128, "mean": 4.078984260559082, "std": 9.016536712646484, "min": -17.950653076171875, "p10": -6.599555206298828, "median": 2.9920778274536133, "p90": 14.920635986328124, "max": 31.6959228515625, "pos_frac": 0.6875, "sample": [-1.643035888671875, 10.929824829101562, -6.0865020751953125, 16.8680419921875, 5.35760498046875, 10.781005859375, 19.853302001953125, 2.9895267486572266, 2.737823486328125, 8.761871337890625, 3.56048583984375, -5.1253204345703125, -14.361373901367188, 0.8648719787597656, 12.021514892578125, -0.241119384765625, 14.822784423828125, 1.220367431640625, -0.47454833984375, 0.5742950439453125, 12.986083984375, 14.3060302734375, 4.0821685791015625, -2.7699737548828125, 19.5380859375, 14.56463623046875, 2.029582977294922, 4.4599151611328125, -10.286163330078125, -3.070159912109375, 11.2847900390625, 21.50823974609375, -11.692230224609375, 23.33441162109375, 7.403186798095703, 6.01806640625, 7.791015625, 1.8903236389160156, -6.5528564453125, 20.50677490234375, 4.738121032714844, 31.6959228515625, 13.3717041015625, -1.007354736328125, 11.34832763671875, 3.7966766357421875, 14.355293273925781, -4.960418701171875, 7.863365173339844, -2.673816680908203, 1.191925048828125, 1.1131134033203125, -1.1160717010498047, 7.075347900390625, 3.1846466064453125, -5.6846466064453125, 7.845855712890625, -1.4481201171875, 9.61090087890625, -7.98626708984375, 0.13726806640625, -4.765472412109375, 3.3315658569335938, -4.30523681640625, 1.151702880859375, 14.544708251953125, 15.148956298828125, 0.9003219604492188, -6.708518981933594, 12.670639038085938, -0.90576171875, 0.7685890197753906, 7.9939422607421875, 12.271820068359375, 19.12005615234375, -6.721588134765625, 3.318572998046875, -2.1408233642578125, 2.56512451171875, 5.59375, 3.6175384521484375, -0.12066650390625, -5.3305816650390625, -4.231903076171875, -12.039474487304688, 22.061279296875, 12.723892211914062, -8.12347412109375, 16.021514892578125, -11.474853515625, -17.950653076171875, 4.81524658203125, 2.3447799682617188, 8.61605453491211, 5.68988037109375, 12.4161376953125, 9.375732421875, 8.626235961914062, 18.075607299804688, -6.4271087646484375, 6.4468841552734375, -7.631378173828125, 13.999603271484375, 1.4381828308105469, 3.2643585205078125, 6.8524322509765625, 2.1505355834960938, 9.5869140625, -0.894683837890625, 1.6881637573242188, 2.99462890625, 5.378997802734375, 0.77703857421875, -17.0675048828125, 0.159912109375, -0.09185791015625, 9.016693115234375, 2.185791015625, 1.1106719970703125, -6.5496826171875, 10.320327758789062, 0.7935638427734375, 27.5369873046875, 12.178932189941406, -9.918014526367188, 1.449493408203125, -0.2271108627319336, -0.552520751953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000094.npy"}
|
|
{"epoch": 0.1968586387434555, "step": 95, "batch_size": 128, "mean": 4.478293418884277, "std": 9.111438751220703, "min": -23.45977783203125, "p10": -7.482466125488281, "median": 4.09808349609375, "p90": 17.366510009765623, "max": 27.39300537109375, "pos_frac": 0.7109375, "sample": [-3.1818599700927734, 0.161285400390625, 3.8805618286132812, 24.21514892578125, 7.6055908203125, 1.489013671875, 1.37969970703125, 6.88043212890625, 25.4532470703125, 9.66912841796875, -0.26666259765625, -0.8363494873046875, 2.2631683349609375, 2.3043251037597656, -8.595413208007812, 5.878814697265625, 3.433197021484375, 6.509735107421875, 12.933425903320312, 1.9830169677734375, 12.106231689453125, 7.172477722167969, -10.8179931640625, 8.075042724609375, -0.645782470703125, -1.6952667236328125, 26.697540283203125, -4.44879150390625, 16.177947998046875, -23.45977783203125, 20.97882080078125, -2.956329345703125, 9.391143798828125, -7.23626708984375, 23.5009765625, 9.8206787109375, 3.21099853515625, 7.204315185546875, 2.59442138671875, 9.860107421875, 3.2205963134765625, 4.78875732421875, -3.30328369140625, -0.2073650360107422, 0.6350860595703125, 9.277145385742188, 5.048133850097656, 9.874221801757812, 23.37811279296875, 0.9622039794921875, -0.121246337890625, 9.50201416015625, 5.2806854248046875, -0.14287567138671875, 6.59332275390625, 7.0892333984375, 5.532798767089844, 3.5820846557617188, 5.3355255126953125, 11.66595458984375, 6.64764404296875, -1.5192413330078125, 1.6072235107421875, 0.734161376953125, 3.1276626586914062, 8.107147216796875, -3.605072021484375, 11.18927001953125, 6.0174560546875, 3.486743927001953, -6.042724609375, 18.0191650390625, 9.5872802734375, 17.910842895507812, 1.15020751953125, -0.8918838500976562, -0.1442089080810547, 17.133224487304688, 2.4837646484375, 6.2783966064453125, -9.165283203125, 9.952701568603516, 3.8918304443359375, 1.1631813049316406, 5.45562744140625, 24.4146728515625, 7.2794189453125, 9.403564453125, -0.87939453125, -14.699981689453125, 4.6258544921875, 7.260894775390625, -2.35272216796875, 1.9033126831054688, -13.25506591796875, 1.83758544921875, 8.9599609375, 5.578887939453125, -9.34808349609375, 4.750457763671875, -3.8846435546875, -2.9968185424804688, -8.056930541992188, 1.98590087890625, 11.564403533935547, -8.353874206542969, 5.405391693115234, -1.6260986328125, 24.76788330078125, -10.049560546875, 4.3043365478515625, 6.147613525390625, -0.3910064697265625, 8.9102783203125, 1.68646240234375, 15.371246337890625, 1.5321044921875, 9.785430908203125, 5.348724365234375, -13.6365966796875, -9.200103759765625, -0.6212482452392578, 5.7296295166015625, 19.57916259765625, 4.822784423828125, 24.49713134765625, -11.528564453125, 27.39300537109375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000095.npy"}
|
|
{"epoch": 0.19895287958115182, "step": 96, "batch_size": 128, "mean": 4.3049421310424805, "std": 8.279412269592285, "min": -19.4976806640625, "p10": -4.085034942626951, "median": 3.6801280975341797, "p90": 16.46982345581055, "max": 27.66534423828125, "pos_frac": 0.6953125, "sample": [-3.51202392578125, -7.9224853515625, -6.6700439453125, 1.8031234741210938, 16.4471435546875, 8.097938537597656, -13.08751106262207, -6.38818359375, 0.951385498046875, 0.434539794921875, 3.7010765075683594, -5.2799835205078125, -5.265777587890625, -1.502685546875, 8.449443817138672, 1.5992050170898438, 5.39306640625, -6.345245361328125, 1.801788330078125, 5.11663818359375, 8.9595947265625, 2.1519012451171875, 9.116119384765625, 7.0271759033203125, -3.062042236328125, 3.91790771484375, 17.66546630859375, 12.2542724609375, 2.1250686645507812, 4.4710693359375, 26.094192504882812, 5.653900146484375, 10.503753662109375, 11.082054138183594, 6.80615234375, 3.8677749633789062, 3.769397735595703, 3.5435028076171875, 1.61773681640625, 1.1556320190429688, 5.86737060546875, 4.4184722900390625, -2.8619384765625, 20.4326171875, -3.573638916015625, 5.789405822753906, -8.400665283203125, -0.5465908050537109, 4.6334686279296875, -1.885406494140625, 4.762115478515625, -1.4744873046875, 9.82421875, -3.110015869140625, 3.6591796875, -2.4771575927734375, 5.451164245605469, 14.174957275390625, 4.230316162109375, 13.4461669921875, -0.9450111389160156, 4.085052490234375, 10.68115234375, 9.104248046875, 7.0091552734375, 22.2469482421875, 27.66534423828125, 3.371337890625, -0.19744110107421875, 16.592559814453125, 4.35748291015625, 16.522743225097656, 1.9932022094726562, -9.115890502929688, 24.476531982421875, 2.991790771484375, 9.433441162109375, -2.24041748046875, 18.8814697265625, 0.33807373046875, 8.205810546875, 0.374420166015625, 5.044746398925781, -6.9010009765625, -0.051971435546875, -1.9872245788574219, -0.61883544921875, 11.97686767578125, 17.012481689453125, 0.2145538330078125, -1.0019893646240234, -0.6001205444335938, 4.782045364379883, 16.255401611328125, 19.9842529296875, -6.379112243652344, 4.09246826171875, 7.46636962890625, 10.574468612670898, 12.99456787109375, -15.492950439453125, 20.0731201171875, 2.4858551025390625, -2.0784950256347656, -0.2724761962890625, 10.405288696289062, 0.269195556640625, 25.7493896484375, 0.7496414184570312, 2.87542724609375, 9.603363037109375, 0.0, 0.935028076171875, 2.343158721923828, 0.8379135131835938, 4.657585144042969, -3.339508056640625, 9.44696044921875, 5.030059814453125, 12.08770751953125, -3.5790023803710938, 8.03708267211914, -2.754505157470703, -3.011627197265625, 2.5214157104492188, 6.6781005859375, -1.3190460205078125, -19.4976806640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000096.npy"}
|
|
{"epoch": 0.20104712041884817, "step": 97, "batch_size": 128, "mean": 4.653496742248535, "std": 10.008232116699219, "min": -16.9100341796875, "p10": -7.3302368164062495, "median": 3.951244354248047, "p90": 17.86153869628906, "max": 31.21478271484375, "pos_frac": 0.6875, "sample": [5.5379791259765625, 7.3624267578125, 6.1962127685546875, -6.7001190185546875, 6.873390197753906, 2.9495468139648438, -1.8612098693847656, 5.880428314208984, 8.4569091796875, 7.2928924560546875, -1.3971519470214844, 31.21478271484375, 11.093215942382812, 16.20587158203125, 0.3586883544921875, 5.02777099609375, 17.693878173828125, -0.38848876953125, -4.9334716796875, 13.482376098632812, 3.200531005859375, 2.6164932250976562, -2.8232421875, -1.371551513671875, 10.086090087890625, -7.287078857421875, 4.65460205078125, -6.388236999511719, 8.230663299560547, 8.785057067871094, -7.149681091308594, 0.47795867919921875, -4.0134429931640625, -8.880401611328125, 7.5254974365234375, 27.901596069335938, 10.832511901855469, -3.6023712158203125, -6.027435302734375, -10.40740966796875, 1.883697509765625, 1.21087646484375, 3.965057373046875, 23.3675537109375, -8.097976684570312, 6.1767578125, 5.403228759765625, 9.131332397460938, 0.328033447265625, 9.184577941894531, 11.6495361328125, -5.135040283203125, 25.3065185546875, 1.165008544921875, 8.0166015625, 11.471893310546875, -7.430938720703125, 8.2510986328125, 1.92236328125, -1.882568359375, 9.653572082519531, 6.8311309814453125, -1.86163330078125, 7.861454010009766, 5.2086181640625, 0.0, 27.154052734375, 8.032852172851562, 9.824798583984375, 30.1163330078125, 0.9100780487060547, 14.121749877929688, 8.535812377929688, 4.9018402099609375, -0.6349754333496094, 27.8560791015625, 3.6822509765625, -15.11846923828125, -7.831901550292969, 19.140380859375, 6.8325958251953125, 1.1051788330078125, -5.69354248046875, 3.9374313354492188, 12.494110107421875, 5.448917388916016, -0.06498336791992188, 28.34912109375, 5.9514923095703125, -3.86871337890625, 7.603557586669922, -10.50885009765625, 7.100372314453125, 2.1173248291015625, 6.908721923828125, -16.9100341796875, 0.0, 6.766998291015625, 13.383537292480469, 1.603424072265625, 1.5016555786132812, 29.468719482421875, -10.113922119140625, 2.3815155029296875, 3.1286392211914062, 2.2541046142578125, -1.78814697265625, -7.11578369140625, 27.003692626953125, 9.439697265625, -5.076171875, 23.238250732421875, -8.547256469726562, -12.28155517578125, 0.3924560546875, 12.7481689453125, 2.9072113037109375, 6.404937744140625, -0.2750205993652344, -13.675460815429688, 9.755561828613281, -4.529815673828125, 1.9140472412109375, 13.631103515625, 18.25274658203125, -12.421478271484375, 3.8876953125, 9.62762451171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000097.npy"}
|
|
{"epoch": 0.2031413612565445, "step": 98, "batch_size": 128, "mean": 4.7015838623046875, "std": 10.211080551147461, "min": -27.59552001953125, "p10": -5.5452577590942385, "median": 3.4617576599121094, "p90": 17.708757019042967, "max": 33.56884765625, "pos_frac": 0.6875, "sample": [12.036701202392578, 12.960418701171875, 22.619384765625, 3.0596752166748047, 29.63958740234375, 18.1378173828125, 13.247390747070312, -5.0373992919921875, 3.71978759765625, -1.1899452209472656, 16.118316650390625, -12.820907592773438, -1.2384033203125, 5.536224365234375, 7.2818450927734375, 7.696685791015625, 2.2817840576171875, -13.501953125, -6.407539367675781, 2.1563758850097656, 11.5989990234375, -3.90509033203125, -6.28594970703125, 3.959259033203125, 2.2927169799804688, 3.8623123168945312, 0.0662078857421875, -4.63397216796875, -5.91876220703125, -16.991363525390625, 2.010164260864258, -27.59552001953125, -5.69989013671875, -6.605224609375, 3.4855880737304688, -1.4416961669921875, -5.478986740112305, 14.70172119140625, 2.981719970703125, 3.3637313842773438, 2.4637908935546875, 8.790802001953125, 0.9818572998046875, 7.3807220458984375, -0.701446533203125, 18.67559814453125, 0.72186279296875, 17.662078857421875, -4.6911468505859375, 14.078977584838867, 32.8360595703125, 14.20928955078125, -1.2842483520507812, 4.606170654296875, 0.01761627197265625, 0.12640380859375, 28.29864501953125, 12.975677490234375, -1.5340576171875, 22.885009765625, -15.695648193359375, -3.0666275024414062, -2.584930419921875, -4.772735595703125, 24.498870849609375, -0.9105682373046875, 11.630401611328125, 4.131988525390625, 6.332370758056641, -2.1207427978515625, 7.582183837890625, 17.2099609375, -15.81243896484375, 2.0908050537109375, 20.414703369140625, 14.811325073242188, 4.33648681640625, 12.462158203125, 0.5284423828125, 15.72039794921875, 5.5383148193359375, 17.817672729492188, -8.86456298828125, -0.4792213439941406, 16.16162109375, -4.6461639404296875, -0.0741424560546875, 1.6003551483154297, 3.43792724609375, -1.21441650390625, 23.07989501953125, 8.057464599609375, 5.1719207763671875, -1.4063873291015625, 7.510826110839844, 8.238311767578125, 0.5886764526367188, 6.9947357177734375, 2.734771728515625, 5.756988525390625, 1.500335693359375, 15.190109252929688, 33.56884765625, 3.249114990234375, 7.132843017578125, -4.948753356933594, -4.1979827880859375, 7.074554443359375, 5.8044586181640625, 3.837432861328125, -0.946685791015625, 2.064697265625, 3.8633270263671875, 7.167047500610352, -0.6052207946777344, 8.648681640625, 2.591522216796875, -2.6532745361328125, 7.119071960449219, 10.925857543945312, 12.76706314086914, -4.244476318359375, 15.75030517578125, 7.213470458984375, 20.472320556640625, 2.0929412841796875, -19.872406005859375, 9.88507080078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000098.npy"}
|
|
{"epoch": 0.20523560209424083, "step": 99, "batch_size": 128, "mean": 5.051894664764404, "std": 10.555581092834473, "min": -24.116943359375, "p10": -8.54915313720703, "median": 4.76650333404541, "p90": 18.79938659667969, "max": 25.9600830078125, "pos_frac": 0.71875, "sample": [-18.1590576171875, 0.0812225341796875, -10.855224609375, 25.823959350585938, 5.082366943359375, 12.913909912109375, -1.422515869140625, 19.227874755859375, 0.18115234375, -4.745880126953125, 5.82086181640625, -11.062637329101562, 0.056884765625, 6.819801330566406, -8.471282958984375, 15.528594970703125, 17.60272216796875, 8.601119995117188, 8.315521240234375, -1.9249420166015625, 5.248819351196289, 3.1108551025390625, 10.33123779296875, -18.0391845703125, 24.120635986328125, 8.175735473632812, 4.785865783691406, -10.29473876953125, 7.694580078125, 16.87066650390625, 13.52044677734375, -3.7884521484375, 20.60345458984375, 2.120513916015625, 5.827545166015625, -0.702880859375, 2.70635986328125, 8.571090698242188, 25.9600830078125, 4.621490478515625, 10.770767211914062, 17.5457763671875, 7.017921447753906, 17.1842041015625, 5.5781097412109375, -4.703125, 12.688438415527344, 1.6245880126953125, -1.3894805908203125, -0.85028076171875, -14.181892395019531, -21.17205810546875, -18.441314697265625, 11.132789611816406, 12.031875610351562, -3.7118301391601562, 14.512847900390625, -18.637710571289062, 15.48028564453125, 8.836578369140625, 5.6266021728515625, 1.547454833984375, 20.361846923828125, 1.3882217407226562, -4.1047515869140625, 16.7122802734375, -23.859039306640625, 8.1705322265625, 3.4383468627929688, 18.8521728515625, 1.812896728515625, 15.054412841796875, -1.252105712890625, 0.5892620086669922, -0.585693359375, 8.6495361328125, 4.884653091430664, 18.776763916015625, 5.938507080078125, 3.3945159912109375, 9.105743408203125, 1.2052841186523438, -1.866973876953125, 3.8028106689453125, 17.254196166992188, 4.225238800048828, 12.945892333984375, -2.985260009765625, 6.689979553222656, 4.6643524169921875, -8.730850219726562, 2.5679054260253906, 1.0191802978515625, 6.930503845214844, 19.157211303710938, 0.8719635009765625, 16.90869140625, -10.42132568359375, 1.6118392944335938, -3.160552978515625, 20.61993408203125, -0.374481201171875, 0.0, 4.747140884399414, -1.674346923828125, 16.71197509765625, 2.3857574462890625, -3.002227783203125, 16.243606567382812, 22.51800537109375, 21.60882568359375, 12.760116577148438, -24.116943359375, 15.42822265625, 2.700611114501953, 6.4165802001953125, 11.68267822265625, 22.47418212890625, -2.426727294921875, 2.05816650390625, 2.9785308837890625, -2.84857177734375, 14.078399658203125, -0.673583984375, 13.76104736328125, 7.104408264160156, 22.60699462890625, 3.50689697265625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000099.npy"}
|
|
{"epoch": 0.20732984293193718, "step": 100, "batch_size": 128, "mean": 4.343875885009766, "std": 9.610705375671387, "min": -19.14654541015625, "p10": -6.505128479003906, "median": 3.122274398803711, "p90": 18.21948547363281, "max": 27.424713134765625, "pos_frac": 0.6953125, "sample": [7.80804443359375, 4.0030517578125, 12.20294189453125, 3.69171142578125, 15.666748046875, 3.8453369140625, -11.4459228515625, 17.952545166015625, 0.0694580078125, 9.915985107421875, -4.9559326171875, 1.46392822265625, 0.2510833740234375, 5.159759521484375, 13.952362060546875, 25.34014892578125, 0.20487213134765625, 2.017425537109375, -8.890792846679688, 8.133407592773438, -1.229400634765625, 4.45196533203125, -8.9554443359375, -1.8718070983886719, 0.9724884033203125, 9.248733520507812, 2.18731689453125, -8.298820495605469, 0.16082763671875, 16.04718017578125, -10.8238525390625, 2.08056640625, -1.311279296875, 4.814353942871094, 4.832275390625, 4.476295471191406, -0.096466064453125, 18.84234619140625, 2.7248458862304688, 14.403579711914062, -2.0479583740234375, 12.542022705078125, 5.47991943359375, 2.69757080078125, -1.7364425659179688, 0.850341796875, 8.220481872558594, 20.8441162109375, 24.380126953125, 13.431427001953125, -1.1979293823242188, -6.1496429443359375, -5.4497833251953125, 5.7341766357421875, 3.347869873046875, 11.702301025390625, -12.8309326171875, -2.8150634765625, -19.14654541015625, 2.07000732421875, 0.8288803100585938, 24.38421630859375, -5.114990234375, 5.535400390625, 0.24267578125, 6.0997314453125, -6.3761444091796875, -13.565994262695312, 5.02740478515625, 6.5767822265625, 27.424713134765625, 7.97015380859375, 3.8966217041015625, 5.692583084106445, -1.1819381713867188, -6.80609130859375, -5.9286956787109375, 6.7509307861328125, 0.3493194580078125, 11.637287139892578, -11.861274719238281, 4.5465240478515625, 5.549102783203125, 2.15216064453125, 6.360076904296875, 3.134502410888672, -4.183937072753906, -0.7756156921386719, 15.427276611328125, 27.11846923828125, 0.27947998046875, 2.9295501708984375, -5.246673583984375, 3.11004638671875, 6.282012939453125, 1.2893753051757812, 3.9739418029785156, 2.801511764526367, -0.8655796051025391, 12.374496459960938, 22.98687744140625, -0.33087158203125, 24.15869140625, 16.793212890625, 5.0301055908203125, -4.451202392578125, 16.06597900390625, 24.385498046875, 0.9686832427978516, 0.0388946533203125, 25.140411376953125, 5.234710693359375, -0.719207763671875, -3.838836669921875, -12.640411376953125, 15.40484619140625, -0.11662483215332031, 4.47271728515625, 21.331878662109375, -9.10601806640625, 0.5339450836181641, 10.202911376953125, 15.510284423828125, -4.371311187744141, 8.927814483642578, -1.1748504638671875, 21.307861328125, -12.5361328125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000100.npy"}
|
|
{"epoch": 0.2094240837696335, "step": 101, "batch_size": 128, "mean": 4.461676120758057, "std": 10.333577156066895, "min": -23.9049072265625, "p10": -6.308200073242187, "median": 4.593280792236328, "p90": 15.6172119140625, "max": 35.50390625, "pos_frac": 0.71875, "sample": [6.890098571777344, -23.9049072265625, 4.786773681640625, 15.819671630859375, 3.5091629028320312, 2.380462646484375, 3.3706092834472656, 4.978851318359375, -4.8712158203125, 1.127227783203125, 15.819427490234375, 11.633880615234375, -4.52655029296875, 21.456573486328125, 32.5347900390625, 25.802520751953125, 2.39678955078125, 4.698890686035156, -1.4982719421386719, 10.88165283203125, 7.211090087890625, 5.7962799072265625, 10.974151611328125, 9.193084716796875, 1.2722148895263672, 20.571273803710938, 5.777008056640625, 1.979766845703125, -0.6819305419921875, 26.45111083984375, -1.8837966918945312, 5.055267333984375, 4.250881195068359, 5.3248291015625, 15.129150390625, 9.017166137695312, 23.261642456054688, 14.763320922851562, 12.176143646240234, 1.2354736328125, -8.31683349609375, 4.599525451660156, 10.0032958984375, 11.43914794921875, 18.76697540283203, 15.343597412109375, 6.74041748046875, -4.629058837890625, -0.23202896118164062, -11.94244384765625, -1.8797607421875, 6.3951873779296875, -15.773773193359375, 11.653358459472656, 19.41827392578125, 8.474838256835938, 6.9312591552734375, -18.698974609375, -1.4411678314208984, 3.3213729858398438, -5.462806701660156, 8.2537841796875, 4.975982666015625, 11.109474182128906, -0.296630859375, -4.190706253051758, 5.788200378417969, 10.199050903320312, -1.4029693603515625, 15.530548095703125, -0.111358642578125, 3.614276885986328, -0.578857421875, 5.53466796875, 0.8963241577148438, 2.421489715576172, 19.744232177734375, 1.2974720001220703, 3.5400848388671875, -22.59222412109375, -21.349517822265625, 7.655059814453125, -9.363214492797852, 14.5457763671875, 3.7433319091796875, 2.426788330078125, 12.091766357421875, 14.464183807373047, 1.8474922180175781, -2.6179656982421875, 10.029865264892578, 5.176116943359375, 8.0648193359375, 4.516387939453125, 8.5186767578125, 3.4353485107421875, -3.624237060546875, 31.120849609375, 13.8651123046875, 5.38116455078125, -17.619140625, -20.590301513671875, 9.348094940185547, 1.383697509765625, 4.5870361328125, -2.637115478515625, 4.64459228515625, 4.4811553955078125, 35.50390625, -6.2230377197265625, 8.73809814453125, 0.24092864990234375, 11.532257080078125, -1.62738037109375, 0.8971176147460938, 7.028465270996094, -1.7454833984375, -6.5069122314453125, -5.167236328125, 7.11767578125, -9.581710815429688, -0.37908935546875, 2.2112579345703125, 0.5770835876464844, -13.667228698730469, 1.2470979690551758, 8.141021728515625, 10.628093719482422], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000101.npy"}
|
|
{"epoch": 0.21151832460732983, "step": 102, "batch_size": 128, "mean": 6.39826774597168, "std": 9.566093444824219, "min": -19.129852294921875, "p10": -5.922801208496093, "median": 6.019712448120117, "p90": 17.757369232177734, "max": 32.3282470703125, "pos_frac": 0.734375, "sample": [-5.668792724609375, -0.8195953369140625, 13.170547485351562, 5.842964172363281, 1.74652099609375, 17.899391174316406, 8.981613159179688, -0.2836456298828125, 3.250579833984375, 14.042572021484375, 5.6214599609375, 6.304317474365234, 11.674102783203125, 7.86602783203125, -0.04290771484375, 15.413375854492188, 10.557830810546875, -8.853485107421875, -3.3676910400390625, -0.916473388671875, -3.0672760009765625, 13.0458984375, 13.624481201171875, 24.03997802734375, 16.84027099609375, -3.30224609375, 5.884944915771484, -9.90032958984375, 4.653053283691406, -9.05242919921875, -6.5729827880859375, 6.847007751464844, 3.4520034790039062, 8.47271728515625, -6.5154876708984375, 20.57073974609375, 14.94256591796875, -1.6016616821289062, 1.5506973266601562, 1.3001670837402344, -2.4151611328125, 1.9321060180664062, -2.1089611053466797, 16.834381103515625, 9.32720947265625, 5.1792144775390625, 2.7466812133789062, -7.291961669921875, -8.56396484375, -6.91278076171875, 17.5802001953125, -4.562103271484375, 5.29541015625, 7.309356689453125, 5.789581298828125, 17.302825927734375, 7.264068603515625, 32.3282470703125, 10.343658447265625, 15.954345703125, 1.4536361694335938, 0.76416015625, 16.858489990234375, 12.371124267578125, 9.115631103515625, -4.801239013671875, -9.01043701171875, 2.71856689453125, -3.9553565979003906, 5.324760437011719, 10.383037567138672, 16.49364471435547, 21.906875610351562, 15.090530395507812, 5.732791900634766, 21.431991577148438, 18.474475860595703, 5.1149139404296875, 7.64739990234375, -14.229949951171875, 0.8520450592041016, 8.44183349609375, 0.0, 6.30023193359375, -2.1341400146484375, 18.37646484375, 0.0, 6.388511657714844, 13.813720703125, -0.1337604522705078, 12.05450439453125, 12.271575927734375, -1.9760894775390625, 2.833099365234375, 16.268951416015625, 6.15447998046875, 13.540985107421875, 0.37317657470703125, 11.866119384765625, 15.863609313964844, 17.696502685546875, 6.627464294433594, 2.602569580078125, 25.199462890625, 7.22882080078125, 10.560882568359375, 2.640756607055664, 18.17657470703125, 0.40849876403808594, 18.12957763671875, -19.129852294921875, 10.857574462890625, -1.995025634765625, 3.07342529296875, 12.94793701171875, -14.701690673828125, 15.282569885253906, 2.0059814453125, 30.82763671875, 3.3424072265625, 15.90423583984375, 5.43194580078125, 11.31329345703125, 9.813812255859375, 9.61138916015625, 28.389251708984375, -0.92138671875, -13.349868774414062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000102.npy"}
|
|
{"epoch": 0.2136125654450262, "step": 103, "batch_size": 128, "mean": 4.841897487640381, "std": 10.708672523498535, "min": -22.043212890625, "p10": -6.673909950256345, "median": 2.2939300537109375, "p90": 19.01605224609375, "max": 46.43212890625, "pos_frac": 0.71875, "sample": [-17.759857177734375, -9.81585693359375, 6.6875, 11.36285400390625, 19.3355712890625, 15.462970733642578, -0.6380538940429688, -3.3033409118652344, 23.84930419921875, 0.0562744140625, 0.394561767578125, 15.9747314453125, 19.14251708984375, 2.549591064453125, 0.0, -1.49249267578125, -17.30322265625, -11.460113525390625, -3.789337158203125, 9.784698486328125, -0.6207046508789062, 1.2776336669921875, 8.842987060546875, -22.043212890625, 2.438446044921875, 5.82879638671875, 1.3834152221679688, 2.1844940185546875, 6.2777099609375, 1.2971725463867188, 11.047393798828125, 10.189865112304688, 0.7638282775878906, 4.755157470703125, -3.52886962890625, 7.4850616455078125, -1.1260528564453125, 16.08203125, 7.6031494140625, 7.250431060791016, -1.7284908294677734, 14.013458251953125, 2.25225830078125, 31.450408935546875, -1.6427001953125, 13.7921142578125, 1.943267822265625, 9.14276123046875, 14.2105712890625, 46.43212890625, 19.231185913085938, 13.19427490234375, 7.697113037109375, -11.5960693359375, 1.3834991455078125, 0.42456817626953125, -9.475318908691406, 10.2943115234375, 10.05126953125, 11.100860595703125, 17.70904541015625, 7.5921630859375, 1.2197341918945312, 11.80194091796875, 4.604095458984375, 21.364990234375, 10.597412109375, 12.301513671875, 3.8613967895507812, 7.3763275146484375, 1.3367233276367188, 0.575836181640625, 1.52978515625, 4.64678955078125, -0.0131378173828125, 0.9154300689697266, -8.612676620483398, -19.967041015625, 1.13067626953125, 4.4731903076171875, 20.956451416015625, 13.057456970214844, 0.9671649932861328, 12.1004638671875, -0.6775436401367188, 9.690765380859375, 5.928657531738281, 2.0745849609375, 27.017791748046875, -2.796142578125, 2.3182296752929688, 2.15472412109375, 16.186233520507812, 8.168914794921875, -1.9595260620117188, 1.9571762084960938, 24.4793701171875, -2.1749267578125, -5.803253173828125, -12.619415283203125, 0.0, 2.2696304321289062, 3.0468902587890625, 16.80059814453125, -2.399627685546875, 18.80670166015625, -11.57232666015625, 1.009246826171875, 1.5028457641601562, 12.118579864501953, 12.4881591796875, -4.512451171875, -3.641399383544922, 21.344085693359375, 0.3497467041015625, 0.05403709411621094, 21.3355712890625, 24.36376953125, 9.729644775390625, 18.96185302734375, -4.449005126953125, 6.520519256591797, -0.897216796875, -14.03759765625, 2.0325775146484375, -19.839019775390625, -5.843009948730469, 0.1501617431640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000103.npy"}
|
|
{"epoch": 0.2157068062827225, "step": 104, "batch_size": 128, "mean": 4.400773048400879, "std": 11.499898910522461, "min": -27.36865234375, "p10": -8.66700096130371, "median": 3.3540725708007812, "p90": 20.013775634765626, "max": 47.1217041015625, "pos_frac": 0.6875, "sample": [3.0191650390625, -4.62408447265625, 7.814323425292969, 16.33544921875, -10.253082275390625, 5.277435302734375, 4.3661041259765625, -0.3971271514892578, 19.35064697265625, -5.230560302734375, 18.456939697265625, -8.526981353759766, 3.59735107421875, -16.00201416015625, 7.9410400390625, 7.603366851806641, 28.287811279296875, 22.652435302734375, -2.03131103515625, -3.0906295776367188, 2.3147125244140625, -2.6645755767822266, 7.2144317626953125, -22.116607666015625, 25.558837890625, -3.279449462890625, 3.4441986083984375, -1.4248409271240234, 3.1469593048095703, 23.823150634765625, 7.430122375488281, 15.214813232421875, -13.39361572265625, 0.29936981201171875, 3.1156005859375, -8.99371337890625, 1.1760711669921875, -7.786201477050781, 22.143768310546875, -2.13787841796875, 11.684295654296875, 0.53607177734375, 0.9785423278808594, -11.887054443359375, -10.873809814453125, 10.38043212890625, 0.1888275146484375, 8.259857177734375, -2.322052001953125, -1.964935302734375, -3.313629150390625, 9.764968872070312, 8.713455200195312, 4.322265625, 26.60418701171875, 5.743106842041016, 19.947509765625, -5.9181060791015625, -3.2177200317382812, 2.46209716796875, 4.537567138671875, 9.8492431640625, 7.050750732421875, 6.55504035949707, 6.5579833984375, 3.2231597900390625, -1.838165283203125, 15.6142578125, 4.08367919921875, -14.637252807617188, 10.45440673828125, 20.16839599609375, 0.45316314697265625, -27.36865234375, 47.1217041015625, 5.1185455322265625, 23.539581298828125, 9.763168334960938, 11.004783630371094, 20.83428955078125, 5.5237274169921875, -26.545318603515625, -12.895751953125, 11.704132080078125, 2.8819427490234375, -0.0845184326171875, 0.9415740966796875, 7.626861572265625, -7.59521484375, 17.769241333007812, 23.50408935546875, 0.382904052734375, 5.26513671875, 10.349212646484375, 0.7764739990234375, 0.14044189453125, -6.9608154296875, 0.7170314788818359, 24.0484619140625, 12.390335083007812, 15.265869140625, 27.757568359375, -5.302635192871094, 16.062408447265625, -18.62896728515625, 0.42841339111328125, -5.202789306640625, 3.7527694702148438, 5.9275970458984375, -15.043701171875, 0.0, 9.900497436523438, -0.85797119140625, 12.789848327636719, 3.263946533203125, 6.9172821044921875, 11.491546630859375, -6.521919250488281, 8.370674133300781, 2.2361907958984375, -0.2338409423828125, 15.672500610351562, 3.003324508666992, 6.545135498046875, 15.549560546875, -2.38055419921875, 0.312255859375, 2.4785995483398438], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000104.npy"}
|
|
{"epoch": 0.21780104712041884, "step": 105, "batch_size": 128, "mean": 8.086766242980957, "std": 11.203606605529785, "min": -23.019744873046875, "p10": -5.014460754394531, "median": 7.243831634521484, "p90": 22.583802795410154, "max": 33.745361328125, "pos_frac": 0.8046875, "sample": [6.209075927734375, 4.4957275390625, 14.383544921875, 1.428232192993164, -6.4693603515625, 17.335540771484375, 9.074073791503906, 15.830482482910156, 11.717399597167969, 2.5701751708984375, 15.90869140625, 7.0353851318359375, 19.743560791015625, 19.372283935546875, -10.9752197265625, 22.44342041015625, 6.533599853515625, -0.9604682922363281, 27.894943237304688, 10.267440795898438, -1.2737770080566406, -0.32830810546875, 7.9781494140625, 7.563995361328125, 22.911361694335938, 28.260711669921875, -22.295928955078125, 1.0471725463867188, -5.0865478515625, 27.41552734375, 3.716094970703125, 13.80963134765625, 0.01059722900390625, 18.070083618164062, -1.887420654296875, 20.896697998046875, 24.608169555664062, 4.5691070556640625, 19.44976806640625, -10.608779907226562, 3.044219970703125, 23.714599609375, 16.029815673828125, 16.13232421875, 2.712432861328125, 18.650848388671875, 9.036407470703125, -1.843170166015625, 21.511962890625, 20.248382568359375, 2.7512741088867188, 13.2640380859375, 0.2976837158203125, 14.07135009765625, -2.044830322265625, 13.729995727539062, 9.457122802734375, 21.680633544921875, 7.968292236328125, -10.43988037109375, 3.98797607421875, -13.931732177734375, 0.298828125, 4.705474853515625, 6.269775390625, 7.808502197265625, 14.341217041015625, 0.5103759765625, 3.052255630493164, 2.701446533203125, 19.880218505859375, -19.880340576171875, 16.98275375366211, 21.23046875, 7.086151123046875, -4.9835662841796875, 12.274169921875, 17.38616943359375, 27.04461669921875, 1.2651214599609375, 8.23553466796875, 3.1982269287109375, 7.736114501953125, 1.4499664306640625, 6.92950439453125, 19.50079345703125, -13.05145263671875, 23.13299560546875, 7.679176330566406, -23.019744873046875, 3.406818389892578, 11.31298828125, 25.025619506835938, -12.680389404296875, 7.401512145996094, 3.18304443359375, 9.231460571289062, -8.0997314453125, 14.3089599609375, 0.0, 4.709678649902344, 5.834625244140625, 0.3271369934082031, -1.1990890502929688, 15.25946044921875, -3.1825103759765625, 33.282684326171875, 26.76800537109375, 6.2116241455078125, 4.820365905761719, 6.843109130859375, 23.342971801757812, 0.8296432495117188, 4.903205871582031, 18.0347900390625, 10.649551391601562, -4.7191925048828125, 11.7479248046875, 2.823436737060547, -8.465911865234375, 8.166267395019531, 21.669471740722656, 4.4461517333984375, 3.6607627868652344, -0.00848388671875, 33.745361328125, 19.28253936767578, 9.77288818359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000105.npy"}
|
|
{"epoch": 0.2198952879581152, "step": 106, "batch_size": 128, "mean": 5.417588233947754, "std": 11.562724113464355, "min": -20.598876953125, "p10": -9.121853637695311, "median": 4.6478271484375, "p90": 21.347671508789062, "max": 36.364593505859375, "pos_frac": 0.6953125, "sample": [-6.6215362548828125, -15.1458740234375, 14.04766845703125, -3.444915771484375, -2.258270263671875, 18.34716796875, -5.804168701171875, 8.100677490234375, -13.140289306640625, 15.107902526855469, -12.703765869140625, 0.0631103515625, 4.8396148681640625, 19.247406005859375, 15.18798828125, 29.971588134765625, 26.472320556640625, -9.3387451171875, -3.554779052734375, -3.1509017944335938, 13.8846435546875, 3.0529937744140625, 4.093292236328125, 4.08203125, 11.772064208984375, 0.705352783203125, 3.5629377365112305, 6.397361755371094, 11.399795532226562, 24.7899169921875, 1.483123779296875, 8.829978942871094, 16.393905639648438, 0.4942474365234375, 0.7052345275878906, -5.9593658447265625, 10.5040283203125, 2.6343231201171875, 5.599456787109375, 7.9072418212890625, 21.6929931640625, 1.63555908203125, 7.403648376464844, -16.83978271484375, -3.11669921875, 13.6412353515625, -6.6102294921875, -0.056488037109375, 12.672531127929688, 5.342338562011719, 8.25921630859375, 8.26571273803711, -5.090616226196289, 17.298492431640625, -19.493423461914062, -20.598876953125, 8.77874755859375, 31.202667236328125, 28.137542724609375, 21.199676513671875, 5.3021240234375, 7.3711090087890625, -12.39996337890625, -6.831146240234375, 12.700958251953125, -0.00313568115234375, -5.837066650390625, -3.9669036865234375, 22.5404052734375, -12.057952880859375, 12.387008666992188, 6.9054412841796875, -0.35655975341796875, 5.1866455078125, -13.251495361328125, 1.5256824493408203, -3.5406875610351562, 1.2426300048828125, 17.5614013671875, 2.4335784912109375, 1.251556396484375, 36.364593505859375, 8.8553466796875, 1.7039642333984375, -5.352386474609375, 2.975311279296875, 21.865936279296875, -0.34044647216796875, 15.187591552734375, -9.028900146484375, 9.920433044433594, 4.044334411621094, 15.95751953125, -0.68560791015625, 0.6882781982421875, 4.356315612792969, 12.852264404296875, 19.438201904296875, 5.22021484375, 7.94281005859375, 29.698516845703125, -15.9010009765625, 1.3671951293945312, -2.8163604736328125, 6.40386962890625, 12.316448211669922, 13.749961853027344, -11.3958740234375, 4.4560394287109375, -5.144256591796875, -10.9127197265625, -3.17230224609375, 32.26873779296875, 7.9088134765625, -7.921875, 5.31683349609375, 11.419113159179688, -5.552406311035156, 20.548431396484375, 3.66778564453125, 23.02447509765625, 6.797210693359375, 8.558685302734375, 22.920318603515625, 20.89483642578125, 0.36283111572265625, 10.361114501953125, 1.818450927734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000106.npy"}
|
|
{"epoch": 0.22198952879581152, "step": 107, "batch_size": 128, "mean": 7.253628730773926, "std": 12.32365894317627, "min": -54.4224853515625, "p10": -5.493386077880859, "median": 5.75311279296875, "p90": 24.515606689453122, "max": 33.508544921875, "pos_frac": 0.7265625, "sample": [-0.1635284423828125, -3.92364501953125, 4.058174133300781, 5.5595245361328125, 10.14208984375, -5.8590087890625, 12.80560302734375, 27.2601318359375, 5.847686767578125, 4.1896514892578125, -5.437042236328125, 4.1351165771484375, 1.0149078369140625, 13.084503173828125, 13.224990844726562, -1.82147216796875, 13.29541015625, -6.067901611328125, 10.278961181640625, 19.20880126953125, -2.7971038818359375, 21.556427001953125, 7.720939636230469, 13.871673583984375, -11.72039794921875, 27.501495361328125, 8.317230224609375, -13.642059326171875, 18.04400634765625, -6.67431640625, 17.7315673828125, 0.9074935913085938, 1.152435302734375, 17.6632080078125, 21.2342529296875, 24.247283935546875, 10.389984130859375, 5.7802276611328125, 33.508544921875, -0.8499221801757812, 27.135482788085938, 6.2135009765625, 7.7253570556640625, 12.651679992675781, 3.7801513671875, 1.1031417846679688, -3.581634521484375, 5.9108734130859375, 13.487272262573242, 6.8388671875, 7.804267883300781, -0.6199722290039062, 3.3491287231445312, 3.65618896484375, 22.86761474609375, -4.65478515625, 3.727508544921875, 1.8444709777832031, 20.5833740234375, 10.924713134765625, 6.057464599609375, 2.1092376708984375, -5.006591796875, 32.21112060546875, 0.1583099365234375, -5.624855041503906, 5.150947570800781, 32.32769775390625, 23.303695678710938, 19.489501953125, -13.204864501953125, 25.141693115234375, -4.5376739501953125, 15.731689453125, 19.94122314453125, 1.5448760986328125, 2.6834716796875, 19.45086669921875, 14.520248413085938, 12.620468139648438, 28.06964111328125, 14.017166137695312, -2.867919921875, -11.04388427734375, -1.83380126953125, 5.10882568359375, 6.671119689941406, -0.8597335815429688, 33.485992431640625, -8.153335571289062, 27.870147705078125, 12.421035766601562, -1.006622314453125, 7.6759033203125, 5.7259979248046875, 1.3381824493408203, 15.59295654296875, 2.9969482421875, 0.2976837158203125, 0.17479705810546875, -54.4224853515625, -10.88739013671875, 10.8248291015625, 27.008056640625, 17.349884033203125, -2.5950393676757812, 27.031829833984375, -0.97454833984375, 6.5977783203125, 1.4906845092773438, 20.516082763671875, -1.169219970703125, 9.406234741210938, 2.111358642578125, 4.2939453125, 17.97589111328125, -3.747650146484375, -2.26123046875, -6.2716522216796875, -1.600830078125, 1.23846435546875, 8.7257080078125, -5.852020263671875, 6.9664306640625, -0.3125782012939453, 29.143508911132812, 5.537078857421875, 17.068603515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000107.npy"}
|
|
{"epoch": 0.22408376963350785, "step": 108, "batch_size": 128, "mean": 8.056560516357422, "std": 10.96955394744873, "min": -18.944290161132812, "p10": -4.900703430175781, "median": 7.2608642578125, "p90": 25.12657775878906, "max": 33.620574951171875, "pos_frac": 0.78125, "sample": [25.191070556640625, 15.111984252929688, -1.2980804443359375, 6.056671142578125, 13.402496337890625, -6.8725128173828125, 19.38400650024414, 9.490009307861328, 3.5742435455322266, 12.3594970703125, 2.9853591918945312, -7.158355712890625, -3.8390960693359375, 12.011322021484375, 4.5188446044921875, -7.283897399902344, 27.99700927734375, 5.957550048828125, -2.6157989501953125, 9.366119384765625, 11.64697265625, 6.1103363037109375, 10.5712890625, -7.688751220703125, 0.215545654296875, 9.15032958984375, 27.7716064453125, -2.2812957763671875, 0.0, -18.875091552734375, 13.885395050048828, 5.149383544921875, 9.911605834960938, 5.53985595703125, 16.6871337890625, 2.177734375, 26.4161376953125, 6.2991943359375, 18.469467163085938, -0.2803306579589844, 3.4200611114501953, 10.2322998046875, 2.37353515625, -3.82012939453125, 4.36346435546875, 26.699020385742188, 10.4993896484375, -15.674560546875, 4.860931396484375, 26.25146484375, 14.515846252441406, 2.8006744384765625, 7.677764892578125, 33.54205322265625, 6.766803741455078, 1.79827880859375, 0.0, -4.934295654296875, 1.6438236236572266, 0.6368331909179688, 1.8742485046386719, 23.6519775390625, 13.008316040039062, 9.109451293945312, -18.944290161132812, 8.380203247070312, 20.515625, 11.327136993408203, -1.2422866821289062, 10.993133544921875, 9.187164306640625, 4.632476806640625, -2.558746337890625, 0.71722412109375, -4.8863067626953125, 24.42254638671875, 19.762847900390625, 1.4338874816894531, 12.770263671875, 19.47821044921875, 9.88134765625, -6.3745574951171875, 25.8006591796875, 10.476608276367188, 3.8316650390625, -0.332183837890625, 22.522247314453125, 10.730796813964844, 9.371978759765625, 28.8248291015625, 7.669586181640625, 19.17633056640625, 3.036346435546875, 10.932296752929688, 8.112371444702148, -2.5993194580078125, 32.119140625, 14.712387084960938, 30.271240234375, 20.75482177734375, 16.023014068603516, 4.1043853759765625, 7.216705322265625, 0.0, -1.1988677978515625, 9.878334045410156, 6.2989044189453125, 33.620574951171875, 5.522037506103516, 25.09893798828125, 7.305023193359375, -8.80657958984375, 4.2389068603515625, 0.9657211303710938, 16.255661010742188, 25.450347900390625, 21.94317626953125, 2.5415802001953125, 2.9282684326171875, 10.48297119140625, -11.11053466796875, -11.054313659667969, -11.398548126220703, 7.4461822509765625, 0.1563873291015625, 10.7015380859375, 4.006317138671875, 17.203765869140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000108.npy"}
|
|
{"epoch": 0.2261780104712042, "step": 109, "batch_size": 128, "mean": 8.413018226623535, "std": 13.538355827331543, "min": -26.675819396972656, "p10": -6.147621536254882, "median": 6.751380920410156, "p90": 23.538375854492188, "max": 56.803955078125, "pos_frac": 0.7421875, "sample": [3.927825927734375, 11.748001098632812, -10.590629577636719, 22.006866455078125, 4.528778076171875, 7.620460510253906, 16.34088134765625, 17.245391845703125, -11.491180419921875, 6.539093017578125, 5.1031494140625, 19.127685546875, 0.5622005462646484, -3.1722412109375, -8.135162353515625, -4.40057373046875, -1.771820068359375, 33.1065673828125, 40.682159423828125, 1.551544189453125, 1.226715087890625, 0.5972442626953125, -2.529083251953125, -13.088958740234375, -6.670589447021484, 11.830764770507812, 0.88385009765625, -1.3411064147949219, -0.4644432067871094, 9.701675415039062, 6.5692291259765625, -2.330322265625, 15.48779296875, 35.352447509765625, 25.4256591796875, 10.31939697265625, 12.605392456054688, 20.8023681640625, 1.7818222045898438, 21.187347412109375, 13.952178955078125, 18.04034423828125, 6.021965026855469, -0.33381080627441406, 13.958831787109375, 33.283599853515625, -0.1829376220703125, 28.989013671875, 7.435630798339844, 7.717323303222656, 13.101104736328125, 5.33349609375, 7.346923828125, 6.068359375, -21.12554931640625, 6.333404541015625, 7.1778564453125, 7.028511047363281, 12.959342956542969, 8.42864990234375, 15.75628662109375, 3.77655029296875, -0.851287841796875, 22.117523193359375, 6.365234375, 11.861221313476562, -5.923492431640625, 9.68060302734375, 9.360847473144531, 12.496551513671875, 16.907684326171875, 4.5694580078125, -7.2495574951171875, 17.040206909179688, 10.833549499511719, 23.504547119140625, 48.025054931640625, -2.98736572265625, 0.8275070190429688, 30.77593994140625, 5.3388671875, 5.1022491455078125, 18.908050537109375, 13.888618469238281, 0.150146484375, 1.0969314575195312, 14.873504638671875, 1.156585693359375, -7.297149658203125, 43.219970703125, 23.0885009765625, 18.1641845703125, 6.93353271484375, 56.803955078125, 9.503620147705078, -2.478424072265625, 1.229400634765625, 38.4547119140625, -12.633056640625, 3.212127685546875, -8.131103515625, 18.712158203125, 0.0, 1.2759552001953125, 29.204010009765625, 9.390548706054688, 14.036941528320312, 9.540237426757812, -0.40753173828125, 10.81353759765625, -0.05633544921875, 5.224853515625, 2.143798828125, -2.487558364868164, 20.444320678710938, -0.852447509765625, 13.438453674316406, 19.28662109375, -12.232025146484375, -25.475311279296875, -26.675819396972656, 23.6173095703125, 4.8225250244140625, 17.068374633789062, 5.41851806640625, -5.696533203125, -5.165618896484375, 14.594635009765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000109.npy"}
|
|
{"epoch": 0.22827225130890053, "step": 110, "batch_size": 128, "mean": 8.153325080871582, "std": 13.665023803710938, "min": -38.99700927734375, "p10": -5.807595062255859, "median": 5.967170715332031, "p90": 27.143865966796874, "max": 44.03765869140625, "pos_frac": 0.7265625, "sample": [-0.5382232666015625, -38.99700927734375, -11.274505615234375, 28.723236083984375, -1.7970161437988281, 1.1425418853759766, -3.492431640625, 0.717315673828125, -0.43353271484375, 27.68798828125, 5.33489990234375, 4.4592132568359375, 3.479045867919922, 23.564697265625, -5.604541778564453, -0.49646759033203125, 11.371826171875, 6.322784423828125, 16.186279296875, 4.7203369140625, -9.683807373046875, 20.4178466796875, 2.018665313720703, 25.707916259765625, 8.949935913085938, 4.373085021972656, -6.106544494628906, 13.996871948242188, 0.06231689453125, -4.380035400390625, 28.7906494140625, 13.880020141601562, 28.5721435546875, 2.537893295288086, 34.314910888671875, -3.4625244140625, 16.684463500976562, 17.731033325195312, 10.145294189453125, -3.49005126953125, 5.398651123046875, 7.05340576171875, 12.408218383789062, 26.600341796875, -10.62481689453125, 0.5741157531738281, 18.27880859375, -7.90673828125, 11.880126953125, 0.5478134155273438, 22.08904266357422, -20.724029541015625, 11.73583984375, 1.8572540283203125, 31.452239990234375, 11.222686767578125, 4.5929718017578125, -31.533447265625, 16.487091064453125, 19.767822265625, 4.2024688720703125, 6.335845947265625, 3.1693572998046875, -0.43280029296875, 2.986236572265625, 27.451080322265625, -6.527687072753906, 27.0751953125, 18.827728271484375, -15.711090087890625, -0.6094856262207031, 19.6553955078125, 4.8412933349609375, 3.120147705078125, 12.019134521484375, 6.1963958740234375, 0.5346527099609375, -2.9314346313476562, -1.5625, 13.126190185546875, 0.0, 4.4834747314453125, 22.04327392578125, 14.659805297851562, -9.969898223876953, 16.151031494140625, 0.6453056335449219, -5.2284393310546875, 6.8056182861328125, 27.267318725585938, 15.981063842773438, 36.29010009765625, 15.632614135742188, 25.625457763671875, 5.737945556640625, 19.36798095703125, 27.5390625, -3.0509185791015625, -2.3008460998535156, 23.07861328125, 27.119384765625, 6.67919921875, 1.376704216003418, 23.844696044921875, -17.794944763183594, 8.95068359375, 4.305778503417969, -5.679473876953125, 24.842193603515625, 20.771652221679688, -1.7059707641601562, 30.319717407226562, -3.1383895874023438, 9.89984130859375, 10.563766479492188, 27.20098876953125, 44.03765869140625, -3.47393798828125, 17.717987060546875, 23.812469482421875, 0.1887664794921875, 19.6947021484375, 6.7486724853515625, 7.482845306396484, -13.157562255859375, 2.50537109375, -0.8081741333007812, 5.5023193359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000110.npy"}
|
|
{"epoch": 0.23036649214659685, "step": 111, "batch_size": 128, "mean": 8.678546905517578, "std": 13.727838516235352, "min": -21.539657592773438, "p10": -8.243379592895506, "median": 7.20538330078125, "p90": 26.66218566894531, "max": 47.91253662109375, "pos_frac": 0.7421875, "sample": [4.396453857421875, -10.094635009765625, 14.63299560546875, -21.539657592773438, 47.341064453125, -7.720470428466797, 11.0166015625, 27.139312744140625, 0.1561737060546875, 0.662017822265625, 28.439178466796875, 6.08160400390625, 6.7243499755859375, 4.92755126953125, 7.01885986328125, 15.002082824707031, 8.258697509765625, 35.22869873046875, -17.947418212890625, -13.62109375, 11.504684448242188, 5.378643035888672, 15.757858276367188, 1.056396484375, 12.9892578125, 3.40155029296875, 4.454193115234375, 25.940689086914062, 1.2936782836914062, 11.327484130859375, 15.304397583007812, 19.0362548828125, -2.135875701904297, 32.7799072265625, -3.350372314453125, 5.493980407714844, -6.138580322265625, 16.13116455078125, -0.6405181884765625, 11.062210083007812, 1.9676513671875, 10.732582092285156, -7.692420959472656, 22.265869140625, 11.578582763671875, 2.80072021484375, 16.81011962890625, -2.3147430419921875, 32.292816162109375, 7.509864807128906, 1.3486862182617188, 47.91253662109375, -9.88360595703125, -2.35162353515625, -0.11199951171875, 18.920654296875, 12.1024169921875, 21.702972412109375, 5.042762756347656, 31.909393310546875, 4.3177490234375, 10.571456909179688, 5.368633270263672, 11.452606201171875, 3.7868881225585938, -5.6005096435546875, 2.6218719482421875, 15.105720520019531, 40.0655517578125, 10.6915283203125, 5.432830810546875, -1.2522125244140625, 7.1868896484375, -0.7727813720703125, 3.1175689697265625, -0.27297019958496094, -0.51708984375, 26.45770263671875, 10.669052124023438, -16.58026123046875, 0.8111076354980469, 9.51837158203125, -9.4635009765625, 3.18084716796875, -3.407684326171875, -6.2017364501953125, 0.17877197265625, 1.8516845703125, 17.3809814453125, 29.7430419921875, 16.900009155273438, 32.81353759765625, 18.400657653808594, -2.3733863830566406, 16.479598999023438, 10.585311889648438, 16.343616485595703, -2.5295639038085938, 44.467926025390625, 3.1652755737304688, 21.36280059814453, 9.082229614257812, -15.628814697265625, -9.76751708984375, -2.512359619140625, 25.921905517578125, -12.728668212890625, 19.55926513671875, -2.8654022216796875, 20.3431396484375, 15.086868286132812, 7.223876953125, 17.50927734375, 36.1119384765625, 21.534454345703125, 18.869354248046875, 24.989715576171875, -16.7626953125, 11.467041015625, 14.545366287231445, 8.153396606445312, -10.2120361328125, 11.837005615234375, 7.514892578125, 19.2139892578125, 1.5979156494140625, -11.896148681640625, 2.31561279296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000111.npy"}
|
|
{"epoch": 0.2324607329842932, "step": 112, "batch_size": 128, "mean": 4.338167667388916, "std": 15.254352569580078, "min": -48.01995849609375, "p10": -13.29689483642578, "median": 3.7663707733154297, "p90": 25.380017089843747, "max": 45.552703857421875, "pos_frac": 0.5859375, "sample": [15.079055786132812, 3.5650787353515625, -5.50274658203125, 11.549652099609375, -0.44467926025390625, 8.840385437011719, -0.656768798828125, -32.478851318359375, -3.2526512145996094, 1.585662841796875, -2.1956844329833984, 6.109748840332031, -10.48284912109375, 20.246978759765625, -0.333587646484375, 2.6982574462890625, 12.6611328125, 16.157989501953125, 6.180229187011719, 25.25836181640625, -9.080169677734375, 19.5596923828125, 13.604217529296875, 13.120758056640625, -48.01995849609375, -13.83929443359375, -8.945724487304688, 11.698440551757812, 11.013595581054688, 8.913124084472656, -9.510612487792969, 4.4694061279296875, -26.012115478515625, 5.4707794189453125, 4.3906402587890625, -18.626068115234375, 4.86785888671875, 5.737089157104492, -0.512054443359375, 3.8318748474121094, 17.75653076171875, -0.82855224609375, 5.62261962890625, 14.709945678710938, 13.198188781738281, -1.6435623168945312, 12.963775634765625, 0.083740234375, 18.021331787109375, -6.898406982421875, 8.490310668945312, 5.4226226806640625, 0.5342864990234375, 15.86663818359375, -18.099853515625, -3.1209945678710938, -15.861068725585938, -8.359771728515625, -34.9439697265625, -8.457759857177734, -1.7527446746826172, -0.7890625, 18.0120849609375, 35.646759033203125, -3.538482666015625, 3.833465576171875, 31.631072998046875, 16.214019775390625, 1.47650146484375, 6.611289978027344, 7.279108047485352, -5.733213424682617, 18.215667724609375, 13.71588134765625, 7.040435791015625, 38.325408935546875, -3.343372344970703, -0.072235107421875, 29.26971435546875, -17.72698974609375, -0.091552734375, 10.824493408203125, -0.28894805908203125, -1.3751487731933594, 21.1495361328125, 21.7606201171875, 12.052932739257812, -0.217803955078125, -4.859626770019531, 2.865875244140625, 26.48114013671875, -13.064437866210938, 30.55902099609375, 4.8648223876953125, -6.3712158203125, 16.703781127929688, -2.214366912841797, 0.2872943878173828, -36.35498046875, -5.9689483642578125, -7.9198760986328125, 25.036376953125, -2.324005126953125, -2.6180419921875, 26.322494506835938, 16.393783569335938, -14.050323486328125, 27.8194580078125, 5.00482177734375, 17.754425048828125, 34.4156494140625, -11.34710693359375, 3.70086669921875, 25.66387939453125, 6.25238037109375, 1.0706939697265625, -17.025360107421875, -2.4471282958984375, 1.8084716796875, -0.30584716796875, 7.2073974609375, -5.14788818359375, 28.1571044921875, 26.785079956054688, 45.552703857421875, -18.457794189453125, -4.436607360839844, 10.189834594726562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000112.npy"}
|
|
{"epoch": 0.23455497382198953, "step": 113, "batch_size": 128, "mean": 10.301370620727539, "std": 14.967475891113281, "min": -29.92822265625, "p10": -8.580038452148438, "median": 8.873031616210938, "p90": 30.98782501220703, "max": 57.6683349609375, "pos_frac": 0.75, "sample": [5.279869079589844, 17.162811279296875, 9.047607421875, 0.6012992858886719, 16.94921875, 0.8885040283203125, -6.365875244140625, 14.66473388671875, 27.91082763671875, 4.7496337890625, 5.817718505859375, 8.698455810546875, 19.310546875, -11.049100875854492, -1.7640838623046875, 3.8267459869384766, 4.097146987915039, 2.8476104736328125, 4.604339599609375, 13.516881942749023, 57.6683349609375, -2.62994384765625, -1.94830322265625, 41.70916748046875, 12.0262451171875, -29.92822265625, 4.93341064453125, 23.84588623046875, 13.710968017578125, -1.7448654174804688, -2.2571258544921875, -10.98095703125, 23.02947998046875, 32.6292724609375, 39.87841796875, 0.909210205078125, 14.976669311523438, -6.3253173828125, 28.486297607421875, -9.368026733398438, 12.006103515625, 30.422256469726562, -0.52978515625, 15.018966674804688, 9.72039794921875, 4.807399749755859, -10.2747802734375, 1.2400131225585938, 0.47361183166503906, 30.8515625, 5.0764312744140625, 17.24115753173828, -25.0335693359375, 27.94256591796875, 34.38568115234375, 24.887344360351562, 12.145454406738281, 33.767974853515625, 5.7623138427734375, -11.543746948242188, 10.146934509277344, -1.1968994140625, -5.295661926269531, 25.181060791015625, -0.7989425659179688, 8.19970703125, 6.053005218505859, 7.14874267578125, -13.377349853515625, 16.710113525390625, 5.81536865234375, 33.238983154296875, -0.9326038360595703, 11.60113525390625, -8.576934814453125, 5.155097961425781, 13.97021484375, 31.305770874023438, 14.816741943359375, 15.926437377929688, 14.065361022949219, -11.81268310546875, 17.520156860351562, 28.937850952148438, 34.63128662109375, 14.873390197753906, 12.985519409179688, -4.076393127441406, -3.15765380859375, 5.938323974609375, 33.572418212890625, 26.458175659179688, 20.880157470703125, 13.486572265625, -8.682647705078125, 24.75982666015625, 0.5475006103515625, 14.578765869140625, 16.4144287109375, 8.0404052734375, 5.746406555175781, -2.56103515625, 13.3829345703125, 3.0580673217773438, 24.23040771484375, 39.745849609375, 24.293869018554688, 38.933563232421875, -9.953125, 2.98822021484375, 22.58001708984375, 17.9932861328125, 11.1805419921875, 43.01031494140625, -5.3184051513671875, -8.5872802734375, 1.414031982421875, 3.042022705078125, 20.825881958007812, 17.344619750976562, 0.430511474609375, 16.7779541015625, -18.85369873046875, 15.14654541015625, 5.2532196044921875, -1.69476318359375, 20.45489501953125, -5.1238861083984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000113.npy"}
|
|
{"epoch": 0.23664921465968586, "step": 114, "batch_size": 128, "mean": 5.183666229248047, "std": 13.754703521728516, "min": -26.27880859375, "p10": -12.807805252075195, "median": 4.613616943359375, "p90": 23.15022735595703, "max": 41.54351806640625, "pos_frac": 0.703125, "sample": [3.9224891662597656, 21.220703125, 2.962371826171875, 26.0635986328125, 0.44309234619140625, 6.982086181640625, 1.3056259155273438, 4.867397308349609, -1.330291748046875, -4.5355377197265625, -2.8413848876953125, 23.6328125, 1.1011962890625, 29.352874755859375, 16.278045654296875, 39.800994873046875, -10.921939849853516, 2.374664306640625, -21.906494140625, 1.4135284423828125, -0.337371826171875, -16.144241333007812, 4.626609802246094, 15.708808898925781, 1.2376556396484375, 15.758026123046875, 9.675148010253906, 8.9222412109375, 11.92047119140625, 15.12481689453125, 3.671661376953125, 12.670791625976562, 4.644523620605469, 7.8279571533203125, 16.37091064453125, -12.795730590820312, 10.570915222167969, -25.65643310546875, 26.218170166015625, 5.55889892578125, -5.150733947753906, -4.72320556640625, 2.443300247192383, -26.27880859375, 5.60076904296875, 1.8019943237304688, 0.5260009765625, 0.73358154296875, 7.671485900878906, -9.474395751953125, 0.1751708984375, -12.835979461669922, 12.0576171875, -8.611129760742188, -16.373046875, -18.075973510742188, 11.5166015625, 41.54351806640625, 24.394927978515625, -8.291122436523438, 25.25469970703125, -23.70257568359375, 20.347412109375, -7.8021087646484375, -3.009002685546875, 19.055206298828125, 37.57923126220703, -9.4998779296875, 1.6263465881347656, 13.98748779296875, -24.18048095703125, 0.09394454956054688, -16.46856689453125, 9.741500854492188, 3.567962646484375, 33.838623046875, 0.58160400390625, 0.0018157958984375, -2.9890289306640625, -16.861602783203125, 3.1641082763671875, 9.231025695800781, 9.772186279296875, -17.38555908203125, 0.5302524566650391, -0.6973819732666016, -18.65704345703125, -4.246070861816406, 0.0, 22.943405151367188, 10.726272583007812, -2.102123260498047, 13.552978515625, 5.8013458251953125, 0.0704345703125, 22.606475830078125, -3.0911865234375, 18.628890991210938, 2.659027099609375, 6.31329345703125, 9.714805603027344, 31.4154052734375, 12.591190338134766, 11.77996826171875, 37.6533203125, 15.759048461914062, 5.454254150390625, 4.600624084472656, -7.5404052734375, 0.0, 34.395050048828125, -2.43597412109375, -7.24066162109375, 13.414154052734375, 8.8651123046875, 20.82685089111328, 7.301910400390625, 5.34161376953125, 6.205036163330078, 6.1238861083984375, 3.1837005615234375, 7.95245361328125, 6.723358154296875, -1.76397705078125, 12.818389892578125, 7.558128356933594, 3.078155517578125, 14.34075927734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000114.npy"}
|
|
{"epoch": 0.2387434554973822, "step": 115, "batch_size": 128, "mean": 5.693040370941162, "std": 15.076605796813965, "min": -29.720916748046875, "p10": -11.296039962768555, "median": 4.716560363769531, "p90": 26.896705627441403, "max": 56.72265625, "pos_frac": 0.6640625, "sample": [29.7569580078125, -15.182907104492188, 12.838134765625, -10.678855895996094, -0.69525146484375, -13.018447875976562, -8.954010009765625, 3.28948974609375, 2.3116302490234375, 19.915029525756836, -5.248922348022461, 8.638153076171875, 8.986190795898438, 8.29629135131836, 3.1807708740234375, -10.024200439453125, 9.060649871826172, -4.617431640625, 15.082550048828125, 2.4550628662109375, -11.19088363647461, 3.153564453125, 6.596656799316406, 6.707763671875, 4.75421142578125, -4.01446533203125, 2.47320556640625, 12.853462219238281, 12.6761474609375, 14.88690185546875, -4.924468994140625, 11.69744873046875, 12.874603271484375, 4.127098083496094, 3.7584457397460938, 9.39190673828125, -7.097869873046875, -21.219680786132812, -1.6043243408203125, 31.3363037109375, 6.12298583984375, 28.3505859375, 38.49267578125, 7.958793640136719, -20.070770263671875, -19.260330200195312, 23.635475158691406, -10.968673706054688, 26.396896362304688, 42.756195068359375, 5.3858642578125, 56.72265625, -10.956146240234375, -1.216156005859375, -13.32830810546875, 15.967193603515625, 25.964202880859375, 24.887786865234375, 13.18597412109375, -22.20977783203125, -16.214111328125, -7.2095947265625, 28.06292724609375, 3.8319091796875, -4.785552978515625, 12.43646240234375, 4.52679443359375, 1.082275390625, 4.6789093017578125, 33.7403564453125, 1.0773983001708984, 9.31201171875, 15.006214141845703, 8.8817138671875, -3.324869155883789, 6.131259918212891, 43.95782470703125, 0.315460205078125, -19.74853515625, -6.688453674316406, -0.8707275390625, 2.251434326171875, -11.87103271484375, 35.12493896484375, 20.98755645751953, 4.05712890625, -27.2288818359375, 9.087738037109375, 2.392192840576172, 31.369171142578125, -3.808145523071289, 4.328269958496094, -6.711151123046875, 32.775146484375, 14.886444091796875, -4.862701416015625, 2.2110061645507812, 6.606170654296875, -9.048690795898438, 7.451141357421875, 10.674163818359375, 5.63568115234375, 9.057441711425781, 7.8464508056640625, 7.67919921875, 3.2918243408203125, 41.4337158203125, 5.0175018310546875, -9.074188232421875, -11.541404724121094, 12.447219848632812, 4.876076698303223, 16.782882690429688, 7.218727111816406, 11.967811584472656, -10.55145263671875, -6.10101318359375, 13.66585922241211, -0.5369033813476562, 22.13275146484375, -29.720916748046875, -2.9928512573242188, 8.809463500976562, -0.25539398193359375, 13.767364501953125, 3.514556884765625, -1.981353759765625, 11.004486083984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000115.npy"}
|
|
{"epoch": 0.24083769633507854, "step": 116, "batch_size": 128, "mean": 8.296485900878906, "std": 14.547768592834473, "min": -40.978302001953125, "p10": -7.088724517822265, "median": 7.34747314453125, "p90": 24.728570556640626, "max": 55.83953857421875, "pos_frac": 0.71875, "sample": [4.648468017578125, -4.299594879150391, 4.9052886962890625, 3.4005813598632812, 6.3804168701171875, 12.12054443359375, 10.745849609375, 24.72723388671875, 43.776519775390625, 7.858001708984375, 18.99139404296875, 10.10400390625, 20.83667755126953, 3.1476516723632812, 16.96923828125, 12.1995849609375, -5.9633941650390625, 12.764633178710938, 4.164201736450195, -10.7305908203125, -20.713623046875, -5.969970703125, 23.167068481445312, 23.822052001953125, 0.3915901184082031, -8.77642822265625, 10.29083251953125, -0.8777732849121094, 6.504524230957031, -0.04425048828125, 1.6944942474365234, 23.5731201171875, -9.697357177734375, 18.077178955078125, -6.1720428466796875, 1.926177978515625, 23.233238220214844, 14.1614990234375, -6.9807586669921875, -4.9775238037109375, -2.5819740295410156, 20.825103759765625, 18.39398193359375, -9.056114196777344, 8.308975219726562, 4.950225830078125, -1.97528076171875, 5.7957763671875, 22.634185791015625, 32.54351806640625, -40.978302001953125, 3.299713134765625, 14.698318481445312, -15.76251220703125, -0.4927978515625, 28.0093994140625, 12.650543212890625, 24.731689453125, -4.1092529296875, 0.45899200439453125, 9.431716918945312, 11.282318115234375, 6.727264404296875, 4.402740478515625, 9.693851470947266, 0.0, 8.525596618652344, 35.80780029296875, -2.2807159423828125, -8.163665771484375, 18.26068115234375, 32.72418212890625, 0.367645263671875, -4.6672515869140625, -7.340644836425781, 2.0576019287109375, 1.7505264282226562, 1.7260856628417969, 6.386417388916016, 23.06121826171875, 3.4350051879882812, 36.61029052734375, 19.441497802734375, 0.0, 12.614856719970703, 6.3383941650390625, 20.195358276367188, 4.4418792724609375, 34.066680908203125, -1.3173675537109375, -1.8527946472167969, 55.83953857421875, 13.237411499023438, -34.64324951171875, 7.95098876953125, 31.7939453125, 18.34363555908203, 4.531562805175781, 9.834386825561523, 10.28369140625, 11.283050537109375, -1.26513671875, 8.548065185546875, 26.859344482421875, 21.78021240234375, -21.894065856933594, 9.602218627929688, 13.739227294921875, 3.6009368896484375, -13.0484619140625, 11.731826782226562, 7.7295684814453125, 45.075592041015625, -1.9941787719726562, 23.975509643554688, 16.111495971679688, 6.9653778076171875, 15.54315185546875, 21.996658325195312, 13.258522033691406, -6.893951416015625, 7.9354248046875, -0.2390899658203125, 29.2208251953125, -11.511871337890625, 5.839447021484375, -2.64208984375, 18.0206298828125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000116.npy"}
|
|
{"epoch": 0.24293193717277486, "step": 117, "batch_size": 128, "mean": 8.465743064880371, "std": 13.864516258239746, "min": -23.517974853515625, "p10": -6.76133575439453, "median": 7.124088287353516, "p90": 28.91600646972656, "max": 44.178558349609375, "pos_frac": 0.7265625, "sample": [2.614360809326172, 8.05038070678711, 12.184917449951172, 16.610458374023438, 5.593971252441406, 13.977783203125, 1.8928680419921875, -1.7114143371582031, 10.03631591796875, -4.130226135253906, 14.784469604492188, -6.490142822265625, 0.66070556640625, 14.476699829101562, 18.501880645751953, 2.84112548828125, 9.260177612304688, 9.651107788085938, 25.769195556640625, 3.5762939453125, 18.15924072265625, -9.308395385742188, 7.800968170166016, 2.235363006591797, 14.249649047851562, 28.981353759765625, 1.1400909423828125, -19.21929168701172, 13.05877685546875, 44.178558349609375, 16.143518447875977, 30.256317138671875, 34.43896484375, 14.315797805786133, 10.227052688598633, -11.475555419921875, 24.758682250976562, 0.316162109375, 1.6072921752929688, -9.39764404296875, 34.161712646484375, 6.8710479736328125, 4.51177978515625, 30.58135986328125, 0.0, 23.2843017578125, -4.8477630615234375, 6.89166259765625, 38.165069580078125, 24.60595703125, 8.290115356445312, 36.12164306640625, -22.2081298828125, 19.19537353515625, 22.42108154296875, 21.35748291015625, -6.18719482421875, 2.631551742553711, 2.7641525268554688, 8.0224609375, 17.88275146484375, 14.488525390625, 13.78350830078125, 11.901443481445312, 5.42962646484375, 11.870758056640625, 13.410537719726562, -1.3527679443359375, -7.5174713134765625, -7.3941192626953125, 8.874885559082031, -21.56170654296875, -0.06704521179199219, 10.567352294921875, 26.774627685546875, -13.790924072265625, 10.8121337890625, 12.4976806640625, 21.45751953125, 1.6810111999511719, -2.2921600341796875, -2.2891845703125, 5.9542236328125, 32.788970947265625, 4.8468017578125, -5.296600341796875, -6.38189697265625, -15.7054443359375, -0.798553466796875, 3.653064727783203, -23.517974853515625, 15.4442138671875, 30.692703247070312, 28.88800048828125, 8.294567108154297, 4.62969970703125, -9.739578247070312, -2.052886962890625, -0.8922233581542969, 31.2396240234375, 3.743743896484375, -3.71630859375, 2.3796749114990234, -3.603790283203125, -5.5451507568359375, 7.3543548583984375, 27.358734130859375, 6.893821716308594, 25.7899169921875, -5.008876800537109, 16.06710433959961, 17.226394653320312, 20.9466552734375, 0.6694488525390625, 0.37717437744140625, -4.058563232421875, 4.90632438659668, -0.117919921875, 4.5870819091796875, -3.033679962158203, -19.170257568359375, 38.62190246582031, 12.33538818359375, 13.727352142333984, 1.7962799072265625, 17.435009002685547, 36.5859375, 14.6021728515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000117.npy"}
|
|
{"epoch": 0.2450261780104712, "step": 118, "batch_size": 128, "mean": 9.107654571533203, "std": 15.242619514465332, "min": -25.833267211914062, "p10": -10.247073364257812, "median": 7.6334228515625, "p90": 26.867254638671874, "max": 50.08697509765625, "pos_frac": 0.7421875, "sample": [5.955085754394531, 10.62939453125, 19.52642822265625, 3.8746337890625, 30.294754028320312, 24.936370849609375, -23.093505859375, -13.969310760498047, 20.67169189453125, 26.771728515625, 13.40655517578125, 6.3970794677734375, 6.933963775634766, 6.24798583984375, -12.65435791015625, 33.647613525390625, 17.711212158203125, 19.971702575683594, 22.873291015625, 6.851051330566406, 11.8677978515625, 19.40594482421875, 33.129112243652344, 2.238189697265625, -25.1661376953125, -25.833267211914062, 16.9027099609375, 3.25140380859375, -19.557052612304688, 23.0892333984375, 18.176483154296875, 25.547203063964844, 44.1256103515625, 12.534423828125, 3.5876235961914062, 5.550567626953125, 2.7224884033203125, 6.96881103515625, 18.981964111328125, 20.698638916015625, -4.45172119140625, 23.09307861328125, 32.350128173828125, -2.3411197662353516, 2.5165252685546875, -4.4119720458984375, 19.798141479492188, 18.201904296875, 12.32330322265625, 0.5995597839355469, 0.799163818359375, 1.65545654296875, 27.09014892578125, 13.84197998046875, 28.302947998046875, 39.5396728515625, -7.775932312011719, 9.620391845703125, -0.6791572570800781, 23.30987548828125, -5.79736328125, 0.0, -8.12799072265625, 10.90203857421875, -13.013458251953125, -7.828042984008789, -18.922698974609375, 6.484001159667969, 16.2969970703125, -10.585174560546875, -0.255645751953125, 32.3265380859375, 0.8338165283203125, 16.984832763671875, 7.1254730224609375, 24.0316162109375, 23.420379638671875, 16.190597534179688, -10.1021728515625, 26.6551513671875, 4.7697296142578125, 23.499099731445312, 8.72747802734375, -5.1353302001953125, 13.273704528808594, 25.721771240234375, -20.594940185546875, 2.1441497802734375, -11.56549072265625, -8.902931213378906, 49.607666015625, -4.45404052734375, 25.243743896484375, 8.424652099609375, -1.8540267944335938, 9.288337707519531, 1.3977012634277344, -13.657142639160156, 12.18817138671875, -8.834548950195312, 5.279544830322266, 35.70037841796875, 8.1400146484375, 13.288177490234375, -0.5112152099609375, 11.768173217773438, 5.933052062988281, 5.6779327392578125, 10.608720779418945, 32.990631103515625, -6.479705810546875, -7.913665771484375, 18.82427215576172, 13.536605834960938, 7.1268310546875, 2.76373291015625, -20.6734619140625, 10.561553955078125, 6.8819122314453125, 21.637451171875, 2.2102584838867188, 15.529693603515625, 4.334417343139648, 22.269515991210938, 0.0, 6.5995330810546875, 15.1143798828125, 50.08697509765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000118.npy"}
|
|
{"epoch": 0.24712041884816754, "step": 119, "batch_size": 128, "mean": 7.730894088745117, "std": 18.3520450592041, "min": -38.04112243652344, "p10": -14.321496582031248, "median": 5.499183654785156, "p90": 31.009225463867185, "max": 69.39602661132812, "pos_frac": 0.6796875, "sample": [-38.04112243652344, -18.15673065185547, 25.188507080078125, 3.198455810546875, 2.1831283569335938, 3.6088485717773438, -8.661460876464844, -6.4454193115234375, 7.3612060546875, -4.0175628662109375, 23.52435302734375, -11.326011657714844, -11.99505615234375, 32.1285400390625, -17.1890869140625, 9.58331298828125, -13.90240478515625, 24.49188232421875, 27.870155334472656, -2.79876708984375, 7.617431640625, 17.506301879882812, -1.3648052215576172, -8.166511535644531, 7.29022216796875, 15.272842407226562, 21.07489013671875, 6.37054443359375, 33.94207763671875, 3.9287490844726562, 4.105588912963867, 2.1735382080078125, -3.8766326904296875, 0.7462615966796875, 12.164962768554688, 10.8126220703125, 11.746795654296875, 4.1165313720703125, 7.4818267822265625, 3.692596435546875, 22.22540283203125, -8.796737670898438, -1.61346435546875, -0.3074493408203125, 30.802947998046875, 1.2593917846679688, 3.81353759765625, -34.74456787109375, -13.89202880859375, -19.632293701171875, 50.568756103515625, 11.146881103515625, 3.8465118408203125, 29.884857177734375, -5.214073181152344, -1.0636444091796875, 14.33404541015625, 2.2029476165771484, 16.242645263671875, 22.552635192871094, 4.595703125, 1.18988037109375, 24.3721923828125, 16.0037841796875, 20.429534912109375, -9.44073486328125, 39.88104248046875, 12.312479019165039, 32.468505859375, 19.369110107421875, -0.078094482421875, 8.392501831054688, 13.995651245117188, 0.0, 37.65983581542969, -2.596649169921875, -2.2807235717773438, -25.50860595703125, 19.3463134765625, -23.061431884765625, 20.494064331054688, 12.477340698242188, 5.75726318359375, -15.29937744140625, 12.337905883789062, 0.176513671875, 31.49053955078125, -7.441312789916992, 15.693588256835938, 6.996856689453125, 24.535919189453125, -16.145355224609375, -2.990234375, 61.235107421875, -4.5739593505859375, 69.39602661132812, 8.2628173828125, 0.58673095703125, 7.48736572265625, -26.32098388671875, 25.257415771484375, 32.399169921875, 4.6656494140625, 18.614212036132812, -8.94482421875, 23.419677734375, -7.593025207519531, 6.028472900390625, 5.768516540527344, -15.3671875, 3.2047576904296875, -22.271514892578125, 49.007568359375, 14.837165832519531, 58.3983154296875, 5.3162994384765625, -1.0669403076171875, 15.023895263671875, 0.161376953125, 20.60009765625, 6.8635406494140625, 3.7641754150390625, 26.850341796875, -7.831615447998047, 40.261322021484375, 5.68206787109375, 5.209442138671875, -16.7679443359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000119.npy"}
|
|
{"epoch": 0.24921465968586387, "step": 120, "batch_size": 128, "mean": 9.094857215881348, "std": 17.263002395629883, "min": -32.76080322265625, "p10": -12.871966552734373, "median": 9.61667251586914, "p90": 31.8508056640625, "max": 56.444976806640625, "pos_frac": 0.6796875, "sample": [-11.78924560546875, 23.729644775390625, -6.321998596191406, 11.192962646484375, -3.03851318359375, 11.097976684570312, 23.756866455078125, -12.317596435546875, 7.7615966796875, 18.147796630859375, -2.774707794189453, -3.319580078125, 20.863685607910156, 4.54913330078125, -18.29443359375, -1.5308837890625, 32.582763671875, -3.7298316955566406, 9.890182495117188, 19.50754165649414, -6.64093017578125, 32.2999267578125, 30.417999267578125, 27.0035400390625, 26.4189453125, 11.963531494140625, 6.353607177734375, -0.858856201171875, -5.581413269042969, 31.6583251953125, 27.87969970703125, 0.225616455078125, 4.562835693359375, 4.57122802734375, -23.4483642578125, 7.36700439453125, 0.3557586669921875, 15.842483520507812, 20.64306640625, 20.63690185546875, 0.2526893615722656, -2.2495880126953125, 13.05206298828125, -24.89312744140625, 0.0, -1.2589950561523438, -20.433563232421875, -1.587432861328125, -7.378059387207031, 6.8172607421875, 11.08013916015625, 12.251396179199219, 51.4720458984375, 18.77081298828125, 16.289825439453125, 9.179718017578125, -14.165496826171875, 13.773468017578125, 7.8206634521484375, 4.027824401855469, -3.4752273559570312, 0.0, 10.885009765625, 20.775558471679688, 56.444976806640625, 14.237823486328125, 18.037750244140625, 39.13165283203125, -16.17352294921875, 19.18724822998047, 23.04669189453125, 10.644989013671875, 3.9444427490234375, 17.031005859375, 13.655242919921875, -8.7735595703125, 9.127410888671875, 33.7628173828125, 23.824798583984375, 20.88848876953125, 11.67608642578125, -16.482513427734375, -18.132598876953125, -9.099105834960938, -11.863052368164062, 37.7557373046875, 9.343162536621094, 37.71307373046875, 44.62457275390625, 15.685813903808594, -0.26639556884765625, 26.066162109375, -30.0423583984375, 32.751617431640625, 37.59852600097656, 5.14825439453125, 0.037200927734375, 18.038665771484375, 11.906120300292969, 19.133941650390625, -23.07520294189453, 2.556306838989258, 0.32685089111328125, 18.259361267089844, 26.4888916015625, -1.213052749633789, 11.689727783203125, -1.8464736938476562, 42.0758056640625, -32.76080322265625, 21.353759765625, 21.3184814453125, 19.35826873779297, -17.399017333984375, -23.683990478515625, 15.50067138671875, 13.5526123046875, 20.0728759765625, 5.59808349609375, -0.180572509765625, -5.091468811035156, 15.752235412597656, -3.42352294921875, 51.160369873046875, 2.1123008728027344, 3.0025634765625, -4.62554931640625, 15.009765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000120.npy"}
|
|
{"epoch": 0.2513089005235602, "step": 121, "batch_size": 128, "mean": 5.361635208129883, "std": 15.34766674041748, "min": -33.56915283203125, "p10": -13.547648620605468, "median": 3.9297409057617188, "p90": 23.129589843749986, "max": 46.685516357421875, "pos_frac": 0.6484375, "sample": [-1.8304901123046875, 0.67047119140625, 3.3915557861328125, 6.904930114746094, 0.4837188720703125, 3.7981033325195312, -3.24908447265625, 7.272575378417969, 9.132377624511719, 28.061126708984375, 3.1479644775390625, -0.6881561279296875, 33.810028076171875, 5.117944717407227, 16.471954345703125, 0.289581298828125, -2.4573211669921875, 4.061378479003906, -0.11777877807617188, 21.66925048828125, 19.077987670898438, -15.738784790039062, 18.253021240234375, 15.951904296875, -25.620574951171875, 0.0, 8.791572570800781, 2.11669921875, 0.769073486328125, 19.642791748046875, 10.443273544311523, 8.957313537597656, -33.56915283203125, 26.53704833984375, 5.721168518066406, -33.370513916015625, -3.1791915893554688, 5.4683380126953125, 21.262542724609375, 14.153076171875, -4.079864501953125, -10.406402587890625, 7.9961395263671875, 1.4285249710083008, 7.140968322753906, 1.574951171875, 16.681838989257812, -13.843505859375, -9.713897705078125, 0.0, 8.550735473632812, -4.192551612854004, -1.8520660400390625, -2.2584228515625, 15.265243530273438, -7.950935363769531, -0.414703369140625, 6.17034912109375, 18.3406982421875, -13.420852661132812, -3.285369873046875, -18.552017211914062, -2.3633575439453125, 46.685516357421875, 13.153694152832031, 8.792037963867188, 43.421875, -7.08538818359375, -4.626678466796875, -1.5480194091796875, -5.2107391357421875, 15.637939453125, -2.774169921875, 14.727462768554688, -9.70947265625, 1.013427734375, 4.068107604980469, 7.800926208496094, 1.230499267578125, 15.961933135986328, 15.631591796875, -1.2415695190429688, -6.71990966796875, 28.53600311279297, 45.850341796875, 5.1623687744140625, -25.22515106201172, -1.90631103515625, -20.57373046875, 46.21990966796875, 16.883392333984375, 17.871963500976562, 17.28045654296875, 2.82257080078125, 33.395294189453125, 2.6448516845703125, -24.321929931640625, 3.0020599365234375, 13.302230834960938, 0.8727264404296875, 2.665813446044922, 8.308837890625, 14.311767578125, 8.522552490234375, 4.7292022705078125, 14.7850341796875, -2.73345947265625, 5.353424072265625, -3.6689682006835938, -3.535930633544922, 5.745361328125, 34.192169189453125, 10.952362060546875, 3.34393310546875, 9.1126708984375, 8.322654724121094, 29.379776000976562, -26.09918212890625, 28.16094970703125, 14.263275146484375, -0.9760532379150391, -14.13720703125, 41.4091796875, 14.518714904785156, 1.3224906921386719, -18.32086181640625, -24.04656982421875, 16.957992553710938], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000121.npy"}
|
|
{"epoch": 0.2534031413612565, "step": 122, "batch_size": 128, "mean": 7.4063310623168945, "std": 18.257713317871094, "min": -37.6368408203125, "p10": -16.5485595703125, "median": 6.896396636962891, "p90": 29.58761596679687, "max": 59.73516845703125, "pos_frac": 0.6484375, "sample": [21.73748779296875, -16.8302001953125, 26.357467651367188, -14.41900634765625, 13.869964599609375, 22.397445678710938, 8.846405029296875, 2.9507293701171875, -34.81256103515625, 29.27423095703125, 27.028564453125, 0.0, 9.167938232421875, -25.586151123046875, 1.6695404052734375, -37.6368408203125, 10.841537475585938, 16.334716796875, -13.181381225585938, 14.270095825195312, -6.25787353515625, 22.93408203125, -22.8135986328125, -29.7340087890625, 7.736328125, 35.19586181640625, -12.689666748046875, 9.77570915222168, 47.097198486328125, 2.5882186889648438, 7.273895263671875, -2.7383270263671875, 7.7257843017578125, 6.953895568847656, 18.83563232421875, -5.893959045410156, 4.624324798583984, -20.58990478515625, 1.88720703125, 9.745933532714844, -28.90240478515625, 11.558624267578125, 6.598541259765625, 45.34112548828125, -2.0191497802734375, -22.295196533203125, -10.408828735351562, 9.224321365356445, 39.867919921875, -8.24822998046875, -6.8131561279296875, 9.567512512207031, 12.02337646484375, -21.237167358398438, 12.230224609375, 14.8731689453125, -0.2114543914794922, 23.484817504882812, -29.93072509765625, 10.96484375, 15.282135009765625, 20.364166259765625, 18.010482788085938, -0.604156494140625, -9.055046081542969, 33.22119140625, 14.066070556640625, 26.8826904296875, 21.94921875, 28.387298583984375, 44.978912353515625, 13.253816604614258, 0.918060302734375, 38.06390380859375, 59.73516845703125, 7.220916748046875, 5.023956298828125, -0.791046142578125, 31.755615234375, 26.71044921875, -3.5915298461914062, -16.4278564453125, 4.954315185546875, -10.028961181640625, -6.5228271484375, -3.06011962890625, 0.25597381591796875, 13.650520324707031, 6.838897705078125, -5.477630615234375, 17.703277587890625, -0.8958282470703125, 2.0633506774902344, -18.22303009033203, -3.6450424194335938, -18.83099365234375, -2.521148681640625, 26.855682373046875, 23.891082763671875, -0.753692626953125, 5.6269683837890625, -5.72576904296875, 21.992843627929688, 20.361114501953125, 3.108715057373047, 25.226654052734375, 20.5267333984375, 28.42241668701172, 4.433280944824219, 2.1623764038085938, -0.7650794982910156, -0.216705322265625, -8.43890380859375, 14.935379028320312, 49.42169189453125, 0.5852813720703125, 32.3330078125, -3.3307342529296875, 1.54119873046875, 24.792465209960938, 11.928512573242188, 21.498779296875, -4.051723480224609, 35.46051025390625, -5.232460021972656, 2.143829345703125, 30.31884765625, 7.737945556640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000122.npy"}
|
|
{"epoch": 0.2554973821989529, "step": 123, "batch_size": 128, "mean": 10.575752258300781, "std": 18.829790115356445, "min": -49.607666015625, "p10": -10.549185180664061, "median": 9.928062438964844, "p90": 33.957942962646484, "max": 66.31298828125, "pos_frac": 0.734375, "sample": [27.033615112304688, -9.260894775390625, 29.785720825195312, -12.03277587890625, 5.77008056640625, 5.609107971191406, 11.591522216796875, 3.570068359375, 14.19525146484375, 10.605911254882812, -0.5475730895996094, 1.768280029296875, -2.9039154052734375, 2.49920654296875, 6.960969924926758, -9.913360595703125, -9.844894409179688, 16.89111328125, 44.03369140625, 9.474563598632812, -23.81817626953125, -22.4281005859375, 8.4212646484375, 16.723119735717773, 48.28802490234375, 3.6785736083984375, 19.7275390625, 29.793670654296875, 29.02728271484375, 0.9650039672851562, 28.575714111328125, 24.168853759765625, 19.68927001953125, 25.040817260742188, 7.657745361328125, 24.596649169921875, 23.865142822265625, -24.651611328125, 2.127593994140625, 29.892059326171875, 18.370376586914062, 4.6812286376953125, 1.4331817626953125, 33.81451416015625, 10.790237426757812, 58.2174072265625, 0.6383724212646484, 3.143251419067383, 18.967864990234375, 31.4241943359375, 3.0581321716308594, 19.524429321289062, 21.89508056640625, 66.31298828125, 47.1864013671875, 12.093421936035156, -6.500917434692383, -2.2302093505859375, 0.07757568359375, 47.871368408203125, -5.3232269287109375, 5.954160690307617, -49.607666015625, 34.276161193847656, 10.512924194335938, -20.01861572265625, 22.186904907226562, 5.871429443359375, 36.341796875, 7.365869522094727, 11.0289306640625, -0.3629302978515625, 8.74299430847168, -18.53759765625, 10.270248413085938, 26.98529052734375, 21.824951171875, -4.61553955078125, -19.299407958984375, 41.901458740234375, -5.263031005859375, 10.975296020507812, 20.542316436767578, -7.0111083984375, 7.606864929199219, 19.2056884765625, -18.402740478515625, 17.707427978515625, 18.857147216796875, -6.49725341796875, 0.0, -2.910186767578125, 36.7308349609375, 13.05023193359375, 9.945831298828125, 0.0, 14.652587890625, 9.910293579101562, 0.08165168762207031, 19.32061767578125, -6.025291442871094, -13.41217041015625, 5.311897277832031, 26.728179931640625, 0.5429229736328125, 11.970733642578125, 10.263336181640625, 35.7451171875, 16.021255493164062, -16.4246826171875, -2.28466796875, -3.44439697265625, 32.26458740234375, 8.90020751953125, -31.575576782226562, 49.52734375, 14.99346923828125, -7.81378173828125, 14.2674560546875, -2.029632568359375, 0.049762725830078125, 11.8326416015625, 5.8963775634765625, 33.821563720703125, 47.37831115722656, -19.64117431640625, 16.606704711914062, 22.830230712890625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000123.npy"}
|
|
{"epoch": 0.25759162303664923, "step": 124, "batch_size": 128, "mean": 9.563409805297852, "std": 17.383522033691406, "min": -49.80718994140625, "p10": -7.0763935089111305, "median": 7.271461486816406, "p90": 33.43193664550781, "max": 57.300689697265625, "pos_frac": 0.765625, "sample": [1.24798583984375, 29.36785888671875, -49.80718994140625, 17.94427490234375, -4.000728607177734, 2.4609909057617188, 26.755508422851562, 8.080886840820312, 35.238311767578125, 45.272705078125, 4.4027557373046875, -28.51318359375, 7.3853607177734375, -6.521759033203125, 21.9232177734375, 4.657844543457031, 6.217742919921875, 19.55804443359375, -6.442527770996094, 14.1856689453125, 5.2677001953125, 5.2625274658203125, -0.62066650390625, 9.925796508789062, 20.8387451171875, 2.625396728515625, -0.42929840087890625, -5.986846923828125, 4.751861572265625, 7.47406005859375, 28.670272827148438, -14.877532958984375, 4.016597747802734, 23.7344970703125, 45.573150634765625, 37.35798645019531, 18.95941162109375, 12.0206298828125, 5.634552001953125, 23.21954345703125, 20.21893310546875, 18.191131591796875, 38.782073974609375, 10.002471923828125, 3.2016353607177734, -2.7672576904296875, -0.438812255859375, 14.0570068359375, 46.2452392578125, -8.370540618896484, 9.08135986328125, -18.114444732666016, 6.642204284667969, 0.0, 2.324838638305664, -2.713623046875, 25.645263671875, 6.255767822265625, 3.4130706787109375, -11.022300720214844, 33.0999755859375, -39.667022705078125, -25.384002685546875, 42.45904541015625, 12.544342041015625, 22.988525390625, 8.536651611328125, 29.599395751953125, 25.394760131835938, -0.8769683837890625, 9.06463623046875, 5.64111328125, -18.983642578125, -28.763641357421875, 22.677642822265625, 1.9081001281738281, 3.3211746215820312, 12.870391845703125, 35.653045654296875, 20.867034912109375, 9.935760498046875, 30.64117431640625, 6.65399169921875, -15.446929931640625, -5.7265777587890625, 5.847450256347656, 2.90228271484375, 14.603759765625, 35.204315185546875, 16.40426254272461, 3.2803192138671875, 13.524864196777344, -4.0665435791015625, 12.39605712890625, 11.151260375976562, 8.366947174072266, -2.287384033203125, 57.300689697265625, 13.604486465454102, 21.51690673828125, -3.2094879150390625, 1.82574462890625, 15.409454345703125, 2.1135177612304688, 2.673919677734375, 54.36761474609375, -10.00628662109375, 0.287078857421875, 7.050434112548828, 15.030029296875, -3.413543701171875, 1.894989013671875, 7.157562255859375, 11.890106201171875, 7.599700927734375, 21.988109588623047, 8.596359252929688, 34.206512451171875, 0.1353130340576172, -1.319671630859375, 4.772430419921875, 16.348907470703125, 23.451416015625, 45.3697509765625, 5.999359130859375, 11.46868896484375, 4.283477783203125, -14.078968048095703], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000124.npy"}
|
|
{"epoch": 0.25968586387434556, "step": 125, "batch_size": 128, "mean": 9.365135192871094, "std": 19.051164627075195, "min": -33.9981689453125, "p10": -11.141900634765625, "median": 6.7734832763671875, "p90": 35.98381042480469, "max": 63.16302490234375, "pos_frac": 0.65625, "sample": [6.6074371337890625, -10.028383255004883, -33.189048767089844, 38.41424560546875, 1.5918960571289062, 14.930496215820312, 15.70526123046875, -11.032623291015625, 3.4964332580566406, -14.76300048828125, 4.4456787109375, -3.0143890380859375, 33.099945068359375, -1.4801025390625, 12.313323974609375, 24.738494873046875, -28.90863037109375, 14.16119384765625, 13.351669311523438, 15.125984191894531, -8.293609619140625, -4.196727752685547, 19.5771484375, -4.511177062988281, 16.79034423828125, 13.537651062011719, 11.315032958984375, 14.29449462890625, -3.2368621826171875, 46.5364990234375, 4.678108215332031, -7.110237121582031, 9.145187377929688, -1.3932247161865234, 6.9395294189453125, 31.77691650390625, 21.048095703125, 36.331695556640625, 15.474685668945312, -4.5970306396484375, 1.7935333251953125, -23.193328857421875, -11.396881103515625, -6.801666259765625, -5.931854248046875, 43.0167236328125, 1.826873779296875, -6.674713134765625, -33.9981689453125, 15.863090515136719, 17.0201416015625, -15.918060302734375, 24.5225830078125, 63.16302490234375, 6.090583801269531, 15.862129211425781, 7.846168518066406, 20.12646484375, 17.969940185546875, 27.37530517578125, -3.28887939453125, 36.3939208984375, 14.507186889648438, -0.7858524322509766, 0.03558349609375, 14.55572509765625, 2.430572509765625, 1.1324882507324219, 7.399688720703125, 5.641143798828125, 57.001953125, 12.20974349975586, 28.81829833984375, 44.20947265625, -12.741096496582031, -19.437301635742188, 7.686370849609375, 21.590667724609375, 35.834716796875, 23.81344985961914, 31.458709716796875, 4.3623809814453125, -2.4780426025390625, 44.341400146484375, 0.485931396484375, 13.07464599609375, -14.772018432617188, -7.955322265625, 21.235702514648438, 30.865936279296875, 28.946876525878906, 2.197052001953125, -5.041748046875, 14.47991943359375, -8.581573486328125, -1.2520904541015625, -8.05572509765625, 28.56427001953125, 27.446990966796875, -0.7877273559570312, -6.600372314453125, 0.0, 3.1191349029541016, -5.8386688232421875, 42.121429443359375, -8.615753173828125, 1.671661376953125, 4.394416809082031, 6.3177337646484375, 47.771148681640625, -6.420806884765625, -10.868148803710938, 17.6658935546875, 31.77496337890625, 0.60394287109375, 26.58447265625, 50.199859619140625, 10.85638427734375, -15.66058349609375, 17.70343017578125, 27.226104736328125, -30.730194091796875, -7.552825927734375, 47.529052734375, 13.217384338378906, 18.988494873046875, -3.44427490234375, -17.05438232421875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000125.npy"}
|
|
{"epoch": 0.2617801047120419, "step": 126, "batch_size": 128, "mean": 10.706304550170898, "std": 19.07395362854004, "min": -43.0147705078125, "p10": -10.908261108398436, "median": 8.983123779296875, "p90": 34.19221801757812, "max": 52.172943115234375, "pos_frac": 0.7109375, "sample": [21.764984130859375, 52.172943115234375, 14.009033203125, 18.41644287109375, 27.383056640625, 40.9658203125, 12.740097045898438, 39.50511932373047, 3.603700637817383, -1.6792373657226562, -43.0147705078125, 24.744888305664062, 2.5573997497558594, 18.729949951171875, 27.812366485595703, 5.612060546875, 28.444091796875, -5.1763458251953125, -21.858184814453125, -6.3218994140625, -14.391334533691406, 28.721023559570312, -11.30523681640625, -7.303592681884766, 3.802978515625, -9.353118896484375, -3.9017333984375, -3.5586624145507812, 5.722412109375, -7.253631591796875, 7.2066650390625, -2.2929229736328125, 16.92670440673828, 14.023513793945312, 30.326019287109375, -10.738128662109375, -9.000869750976562, 1.650238037109375, -1.3997802734375, 4.325689315795898, 7.693798065185547, 23.198287963867188, 33.546234130859375, -0.9059410095214844, 16.704376220703125, 25.04412841796875, 7.6397857666015625, 15.42529296875, -14.537628173828125, -3.1998443603515625, 33.764007568359375, 0.49444580078125, 11.373443603515625, 50.05133056640625, 3.5783042907714844, 38.081787109375, 16.2669677734375, 6.183349609375, 3.2845458984375, 21.55108642578125, 11.891815185546875, 29.9090576171875, 21.3204345703125, 23.38470458984375, -14.508026123046875, 18.022613525390625, 5.1249237060546875, 25.516326904296875, 20.743667602539062, 46.956298828125, 40.8065185546875, -5.522430419921875, 27.773086547851562, 27.730697631835938, -3.2298011779785156, 7.5865325927734375, 10.26885986328125, -8.345001220703125, 0.0049610137939453125, -25.015869140625, 27.66368865966797, 12.34521484375, -2.9550724029541016, 46.61126708984375, 22.172027587890625, 2.6109085083007812, -29.381378173828125, 6.46722412109375, 23.1634521484375, -32.15191650390625, 12.561553955078125, -9.793693542480469, 25.489944458007812, 32.26519775390625, -42.769378662109375, -10.029983520507812, 4.230436325073242, 8.94451904296875, 5.786590576171875, -16.516021728515625, 23.563751220703125, 27.65288543701172, 33.19189453125, 43.89056396484375, 9.021728515625, 19.403228759765625, -6.92059326171875, 44.82867431640625, 22.629669189453125, 39.50933837890625, -11.68359375, 20.528564453125, 16.103050231933594, 35.191375732421875, 19.784637451171875, 7.013092041015625, 3.9798507690429688, -1.70654296875, 9.72442626953125, -1.9277801513671875, 13.962310791015625, -8.9786376953125, 4.9630126953125, 4.518402099609375, 29.789093017578125, 51.521575927734375, 3.8562240600585938, -17.992767333984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000126.npy"}
|
|
{"epoch": 0.2638743455497382, "step": 127, "batch_size": 128, "mean": 12.328869819641113, "std": 19.239173889160156, "min": -42.164794921875, "p10": -10.019789123535157, "median": 8.749064445495605, "p90": 37.951647949218746, "max": 55.891143798828125, "pos_frac": 0.734375, "sample": [-0.6392745971679688, -0.82806396484375, -9.636474609375, 23.490478515625, 10.821533203125, 3.3966522216796875, 20.689178466796875, 12.132362365722656, -9.904029846191406, -42.164794921875, 29.965576171875, 3.2133865356445312, 20.18084716796875, 15.894144058227539, 9.8310546875, -33.885345458984375, -7.62646484375, 11.90081787109375, -17.438194274902344, 5.73193359375, 8.63018798828125, 7.0084381103515625, 19.896820068359375, 4.256195068359375, 20.473464965820312, 47.14642333984375, 22.06854248046875, 34.054901123046875, 13.536529541015625, 25.257713317871094, 38.63336181640625, 14.944238662719727, 30.2630615234375, 3.6063385009765625, 10.7392578125, 45.2818603515625, -10.289894104003906, -20.823638916015625, -0.364715576171875, 9.21929931640625, 47.03741455078125, 6.897087097167969, 8.434616088867188, 7.521392822265625, -15.228607177734375, -24.428421020507812, 35.691436767578125, 36.649444580078125, 9.678131103515625, -20.65289306640625, 17.590499877929688, 35.34033203125, 22.122650146484375, -7.948333740234375, 27.981170654296875, 40.68927001953125, 24.390625, 16.08184814453125, -2.9722900390625, 2.3521270751953125, 2.713714599609375, 4.4501953125, 37.343292236328125, 23.430267333984375, 8.565223693847656, 33.2891845703125, -5.7935028076171875, 5.4651336669921875, 30.137725830078125, 37.65948486328125, 27.113693237304688, -2.646087646484375, -11.828201293945312, -2.5166664123535156, 9.240966796875, -10.862823486328125, -1.42578125, -15.080177307128906, 7.8796539306640625, -3.084850311279297, 5.859619140625, 22.389511108398438, 55.891143798828125, -1.1400032043457031, 40.461456298828125, 19.634353637695312, 5.0199432373046875, 55.1611328125, 31.17401123046875, 20.496826171875, 47.214874267578125, -8.133468627929688, 6.24560546875, 5.855224609375, 3.9764251708984375, -2.460845947265625, -20.496049880981445, 6.914031982421875, 8.867940902709961, 30.550277709960938, -6.8911895751953125, 8.137100219726562, -9.671432495117188, 29.589157104492188, 23.452682495117188, 4.140510559082031, 38.69580078125, 48.928924560546875, 2.117706298828125, -3.967193603515625, 6.132598876953125, 3.2166748046875, 24.473175048828125, 35.7640380859375, -18.31902313232422, 4.262325286865234, 37.4376220703125, 39.22998046875, -2.037006378173828, 12.799224853515625, 30.947296142578125, 24.2120361328125, 4.11248779296875, 31.424835205078125, 1.65478515625, -7.2669677734375, 23.711441040039062, 48.3841552734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000127.npy"}
|
|
{"epoch": 0.26596858638743454, "step": 128, "batch_size": 128, "mean": 13.580965042114258, "std": 20.842586517333984, "min": -44.5576171875, "p10": -11.083181762695311, "median": 12.640172958374023, "p90": 37.731224822998044, "max": 74.542724609375, "pos_frac": 0.7265625, "sample": [27.18719482421875, 2.15966796875, 20.076339721679688, 25.874969482421875, -31.989044189453125, -33.534149169921875, -20.03668212890625, 7.6236114501953125, 26.1492919921875, 27.488510131835938, 23.790504455566406, 2.77606201171875, 35.43629455566406, 0.5790252685546875, 39.13508605957031, 0.8304004669189453, 3.1667404174804688, 23.6552734375, 28.0806884765625, 25.590606689453125, -3.2885589599609375, -0.44058990478515625, 35.87592315673828, 65.56884765625, 4.256134033203125, 7.7361602783203125, -1.746734619140625, 4.845703125, 29.479400634765625, 20.385833740234375, 9.208770751953125, 11.405672073364258, -19.68011474609375, 33.52159118652344, -7.614034652709961, 31.488845825195312, -16.288162231445312, 6.1119842529296875, 1.1589813232421875, -20.610336303710938, 24.55669403076172, 3.264678955078125, -3.0188827514648438, 17.2432861328125, 33.02191162109375, -3.292572021484375, 0.5226936340332031, 23.904701232910156, 20.00592041015625, 37.12957000732422, 35.298431396484375, 25.1763916015625, -3.9789352416992188, 34.59307861328125, -3.99932861328125, -6.1511993408203125, 3.688232421875, 25.904682159423828, 20.7940673828125, 41.332244873046875, 27.1590576171875, 4.8520660400390625, 13.756847381591797, -0.45810699462890625, 0.5241012573242188, 8.681198120117188, 40.765533447265625, 29.39324951171875, 28.81884765625, 14.017807006835938, 15.456619262695312, 28.63751220703125, 51.3861083984375, 17.47491455078125, 47.053314208984375, 35.17620849609375, 11.1436767578125, -5.5732421875, -1.31427001953125, -0.8925819396972656, 40.35740661621094, 32.24993896484375, -14.510498046875, 19.078201293945312, 29.020416259765625, -44.5576171875, 21.939208984375, -20.76885986328125, 18.61432647705078, 74.542724609375, 9.069427490234375, -11.812255859375, 35.91270446777344, 15.182647705078125, -7.895660400390625, 1.0024242401123047, -15.959053039550781, -7.156494140625, 4.765869140625, 11.52349853515625, 37.006134033203125, -5.6743927001953125, 5.2901611328125, 44.86402130126953, 22.969161987304688, 35.18707275390625, -6.559299468994141, -6.366607666015625, 56.111297607421875, 1.0456428527832031, 8.358856201171875, 0.0, -21.759658813476562, -7.61883544921875, 26.49981689453125, -12.412528991699219, -10.770721435546875, 14.946533203125, 23.93902587890625, 43.779754638671875, 5.048431396484375, 18.386474609375, 10.8758544921875, 41.402252197265625, 61.574371337890625, 26.036392211914062, 19.94647216796875, -1.8107223510742188], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000128.npy"}
|
|
{"epoch": 0.2680628272251309, "step": 129, "batch_size": 128, "mean": 12.522164344787598, "std": 21.541791915893555, "min": -49.316558837890625, "p10": -12.588343811035154, "median": 11.058734893798828, "p90": 42.888616943359374, "max": 78.99417114257812, "pos_frac": 0.765625, "sample": [-49.316558837890625, 29.72833251953125, -24.285324096679688, -14.44940185546875, 13.161140441894531, 45.09189987182617, 26.481597900390625, 37.108375549316406, 30.731582641601562, 2.9690093994140625, 30.40460205078125, 2.7166309356689453, 13.23760986328125, -0.3734130859375, -6.571739196777344, -10.79595947265625, 27.53314208984375, 35.478790283203125, 12.696914672851562, -11.156532287597656, 12.311370849609375, 11.616188049316406, 1.005462646484375, 18.77325439453125, 43.22991943359375, -5.987701416015625, 16.471206665039062, -1.9906005859375, 78.99417114257812, 61.322998046875, 5.7747802734375, 22.686874389648438, 52.73042297363281, 16.88702392578125, 1.3606796264648438, -44.49725341796875, 34.14569091796875, 6.33013916015625, 62.96258544921875, 10.59677505493164, -4.826915740966797, 13.266876220703125, 39.875457763671875, 43.418914794921875, 10.250198364257812, 45.27825927734375, -25.963897705078125, 9.645851135253906, 1.1597442626953125, 4.0399017333984375, 27.499374389648438, 18.02935791015625, 9.886123657226562, 26.549346923828125, -12.05230712890625, 8.583009719848633, 5.3883819580078125, 42.85626220703125, 2.8306884765625, 29.454055786132812, 1.6942329406738281, 3.1314163208007812, 42.964111328125, 22.349563598632812, 60.361175537109375, -7.972900390625, 15.970230102539062, 16.617965698242188, 27.42742919921875, 4.26220703125, 4.1083984375, 6.631538391113281, 3.5495452880859375, -5.07843017578125, 2.91595458984375, -15.118499755859375, 27.355758666992188, 20.400848388671875, -2.621021270751953, 59.278106689453125, 26.906600952148438, 9.450965881347656, 19.80792236328125, 3.6135025024414062, 2.6966285705566406, 3.37469482421875, 29.396804809570312, 4.400054931640625, 17.94891357421875, 39.1871337890625, -10.04840087890625, -26.341796875, -2.6494827270507812, 0.684600830078125, -7.224632263183594, 14.76641845703125, -9.695404052734375, -31.67193603515625, 15.606399536132812, 10.93804931640625, 46.547515869140625, -13.839096069335938, 11.978729248046875, -3.5805435180664062, -18.758468627929688, -21.4481201171875, 13.263320922851562, 3.250732421875, 14.477264404296875, 22.77166748046875, 27.427841186523438, 6.445556640625, 15.43682861328125, 9.26904296875, 9.767295837402344, 46.20538330078125, 11.644195556640625, 30.66168212890625, -17.41949462890625, 18.932159423828125, 11.179420471191406, -0.6576938629150391, 1.90185546875, 31.166290283203125, 13.324920654296875, -14.361518859863281, 25.5982666015625, 31.994029998779297], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000129.npy"}
|
|
{"epoch": 0.27015706806282724, "step": 130, "batch_size": 128, "mean": 11.675660133361816, "std": 19.71907615661621, "min": -40.0186767578125, "p10": -11.55416564941406, "median": 9.612533569335938, "p90": 36.51520996093749, "max": 56.232086181640625, "pos_frac": 0.7265625, "sample": [7.5616455078125, -3.587127685546875, 29.93878173828125, 4.6812896728515625, 5.285453796386719, 40.7869873046875, 9.587646484375, 29.307098388671875, 39.10633087158203, -8.48553466796875, -2.701465606689453, 9.07342529296875, 24.190765380859375, 41.253662109375, -0.10357666015625, 7.124420166015625, -3.8944091796875, -17.314849853515625, 7.6759033203125, 21.13177490234375, 17.19359588623047, 38.937225341796875, -1.0370712280273438, 10.386138916015625, -2.5492477416992188, 56.232086181640625, 2.173492431640625, 22.708984375, 48.104217529296875, 50.105194091796875, -31.270477294921875, -1.99224853515625, 0.0, -10.722152709960938, 26.602935791015625, 11.4561767578125, 24.41058349609375, -39.9521484375, 8.881332397460938, 22.4000244140625, 25.709014892578125, 13.474502563476562, 48.644317626953125, -10.440956115722656, -40.0186767578125, -35.904876708984375, -4.2657470703125, 34.38917541503906, 2.07659912109375, 0.0847625732421875, 35.3116455078125, 3.5392074584960938, 10.994552612304688, 33.555503845214844, -16.04449462890625, -1.5760498046875, 12.747377395629883, 20.464324951171875, 35.477203369140625, -4.17236328125, 10.10589599609375, 0.7718925476074219, 22.65826416015625, 17.395584106445312, 4.3179168701171875, 43.861724853515625, 29.4012451171875, -17.38714599609375, -13.495529174804688, -1.1914520263671875, 40.28802490234375, 9.16497802734375, 30.00146484375, -24.758392333984375, 6.0627288818359375, 18.26751708984375, -15.165985107421875, 27.594635009765625, 9.637420654296875, 15.955682754516602, 28.850692749023438, 16.2786865234375, 29.6856689453125, 18.03143310546875, 32.57208251953125, 1.7350101470947266, 32.172698974609375, 9.009353637695312, 25.98394775390625, -3.7778854370117188, 12.34759521484375, 6.40618896484375, 5.125823974609375, 3.75390625, 23.755111694335938, -28.615570068359375, -4.70135498046875, 6.732879638671875, 3.062042236328125, 49.64739990234375, 44.3843994140625, 34.82501220703125, 20.486862182617188, 1.5068359375, -24.57659912109375, 26.446685791015625, 4.4557342529296875, 6.05560302734375, 20.2249755859375, -4.527313232421875, 34.67566680908203, 7.605274200439453, 27.969314575195312, -6.3246612548828125, 19.026214599609375, 19.959457397460938, 39.096038818359375, 20.3629150390625, -0.6587371826171875, 17.035842895507812, 16.362472534179688, -5.314727783203125, -2.2232666015625, 0.1672210693359375, 31.5361328125, -24.08026123046875, 4.949993133544922, 26.783294677734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000130.npy"}
|
|
{"epoch": 0.27225130890052357, "step": 131, "batch_size": 128, "mean": 9.607320785522461, "std": 22.001131057739258, "min": -46.87713623046875, "p10": -18.581686401367186, "median": 9.088603973388672, "p90": 39.41763610839844, "max": 71.43307495117188, "pos_frac": 0.6328125, "sample": [30.416412353515625, 9.098686218261719, 37.174041748046875, -22.54205322265625, 11.794601440429688, 1.0643463134765625, 1.7223167419433594, 24.794784545898438, -17.392990112304688, 39.415283203125, 31.173614501953125, 31.25848388671875, 43.150726318359375, 19.781341552734375, -17.884033203125, 2.9276123046875, 4.12030029296875, -17.61553955078125, 10.808624267578125, 48.354522705078125, 32.29046630859375, 10.616134643554688, -6.20831298828125, -5.459625244140625, 30.32878875732422, -5.1488800048828125, -10.123321533203125, 42.585418701171875, 9.908866882324219, 12.786102294921875, -1.5149002075195312, 57.459564208984375, -6.896697998046875, 0.70758056640625, 20.63092041015625, -2.3185253143310547, -23.258895874023438, 44.40966796875, -2.79718017578125, 30.5244140625, 9.0015869140625, 3.823974609375, -2.4846839904785156, -16.862060546875, 25.619873046875, 6.791893005371094, 56.19873046875, -17.026092529296875, 19.922592163085938, 24.324737548828125, -1.9920806884765625, -4.0986328125, -18.480804443359375, 16.04693603515625, 13.5679931640625, 39.423126220703125, -23.145339965820312, -5.27691650390625, 25.588558197021484, 34.99803161621094, -8.534500122070312, -18.81707763671875, -20.65283203125, 30.849769592285156, 7.62255859375, 6.442752838134766, 16.811859130859375, 27.651123046875, 71.43307495117188, 9.078521728515625, -46.87713623046875, 4.237152099609375, 9.72503662109375, -10.73417854309082, -3.774158477783203, 19.006683349609375, -27.924591064453125, 60.4632568359375, -6.844158172607422, 35.79463195800781, 11.031219482421875, -22.947784423828125, 1.74560546875, -13.281524658203125, 11.97802734375, 18.577301025390625, -15.8897705078125, 37.87774658203125, 41.652313232421875, -23.1334228515625, -0.9357833862304688, -3.087158203125, 43.7733154296875, -20.227506637573242, 26.8258056640625, 5.0697784423828125, 2.1780948638916016, 8.570457458496094, -7.0008697509765625, 16.88776397705078, 12.570137023925781, -20.831085205078125, 28.846954345703125, -3.6567840576171875, 9.786346435546875, 34.128990173339844, -2.70361328125, -32.65191650390625, 52.342041015625, 9.285369873046875, 31.7818603515625, 40.005523681640625, 13.40472412109375, 19.804412841796875, -11.8564453125, -3.32568359375, 20.623428344726562, -10.912078857421875, 27.44571304321289, 9.241455078125, 21.740753173828125, 15.157180786132812, 25.7542724609375, 17.485946655273438, -6.0694580078125, -21.577003479003906, -8.770843505859375, 1.9814186096191406], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000131.npy"}
|
|
{"epoch": 0.2743455497382199, "step": 132, "batch_size": 128, "mean": 12.99841022491455, "std": 24.779481887817383, "min": -46.78973388671875, "p10": -17.691838073730466, "median": 10.632087707519531, "p90": 43.89382629394531, "max": 96.48382568359375, "pos_frac": 0.6953125, "sample": [-9.4508056640625, 26.08837890625, 38.54615783691406, 13.378654479980469, 0.01064300537109375, -9.1588134765625, 27.3243408203125, 11.635650634765625, -2.4317626953125, 37.6751708984375, 4.188899993896484, -10.566329956054688, -19.3089599609375, 35.94696044921875, 20.342498779296875, 8.723762512207031, 1.19287109375, 44.956024169921875, 42.91888427734375, 59.175994873046875, 35.070098876953125, 32.13494873046875, 70.492431640625, 49.50640869140625, 14.972488403320312, 1.5803756713867188, -0.6066093444824219, -34.71513366699219, 14.273895263671875, 21.584014892578125, -1.1546630859375, 47.412139892578125, 2.0080108642578125, 20.72064208984375, -20.68341064453125, -15.415138244628906, -17.327407836914062, 10.417098999023438, 2.7650604248046875, 0.930084228515625, 41.78546142578125, 15.599868774414062, -9.46142578125, -14.659072875976562, -23.171661376953125, -0.0243682861328125, 5.90594482421875, -18.54217529296875, 10.847076416015625, -9.628284454345703, 15.952163696289062, -10.985288619995117, 16.772802352905273, 30.28441619873047, 3.7550048828125, 1.300689697265625, 6.103851318359375, 17.297515869140625, 96.48382568359375, -0.3865814208984375, 24.93359375, 29.49853515625, 23.160430908203125, -23.38653564453125, 36.47944641113281, -1.62432861328125, -1.8796157836914062, 12.270462036132812, -19.8743896484375, 1.5081787109375, 17.802947998046875, 4.555194854736328, 13.321380615234375, 36.91242980957031, -5.6976318359375, -1.742889404296875, 0.0, 38.400482177734375, 11.748443603515625, -18.78436279296875, 8.678482055664062, 35.1217041015625, 51.1771240234375, -33.1190185546875, -46.78973388671875, 55.057830810546875, 13.340560913085938, 18.766326904296875, 4.346527099609375, 9.424560546875, 21.0482177734375, 13.269241333007812, 32.38397216796875, 24.943992614746094, 52.61326599121094, 92.52346801757812, -15.603729248046875, 26.468826293945312, -0.256378173828125, 15.58807373046875, 23.3388671875, 38.64398193359375, 3.620361328125, 0.9570636749267578, 10.3104248046875, 51.707427978515625, 7.009674072265625, 5.422821044921875, 0.0, -18.796295166015625, 32.408172607421875, 39.84393310546875, -9.676300048828125, 1.4380054473876953, -9.513916015625, 11.27587890625, -2.787994384765625, 57.053955078125, -28.32196044921875, 43.4385986328125, 34.06874084472656, 29.487884521484375, -24.389358520507812, 14.399703979492188, -14.5128173828125, 16.27935791015625, 0.581878662109375, 63.53997802734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000132.npy"}
|
|
{"epoch": 0.2764397905759162, "step": 133, "batch_size": 128, "mean": 13.394699096679688, "std": 22.222286224365234, "min": -64.46368408203125, "p10": -12.621294403076169, "median": 11.499774932861328, "p90": 42.0314697265625, "max": 71.81964111328125, "pos_frac": 0.765625, "sample": [20.1802978515625, 1.6170063018798828, -33.9549560546875, 0.0, 1.3399658203125, 32.9443359375, -0.9100265502929688, 13.965255737304688, 3.6245803833007812, 30.626602172851562, 16.762802124023438, 17.97430419921875, 6.5013427734375, 25.326950073242188, 38.504150390625, 49.9534912109375, -9.928024291992188, -1.3172607421875, -1.466583251953125, -15.093505859375, 34.01470947265625, 49.04852294921875, 71.81964111328125, 4.609222412109375, 24.813690185546875, 43.036376953125, 22.213119506835938, 53.540496826171875, 0.0, 0.0655364990234375, 19.427398681640625, 20.773300170898438, -5.554140090942383, 1.8552093505859375, 25.953628540039062, 33.68000793457031, 20.784072875976562, 27.055389404296875, -28.51519775390625, 32.10455322265625, 39.437408447265625, 6.2957611083984375, -5.2175140380859375, -8.64288330078125, 9.88885498046875, -0.054534912109375, -14.915504455566406, 8.474479675292969, 4.98065185546875, 5.849609375, 16.39813232421875, 41.97705078125, 2.05206298828125, 29.083736419677734, 49.08770751953125, 9.757980346679688, 6.410552978515625, 42.158447265625, -4.3675689697265625, 30.623733520507812, -22.68084716796875, 10.309226989746094, 34.471923828125, 1.498382568359375, 16.715341567993164, 8.192012786865234, 25.292617797851562, 41.338134765625, 29.7498779296875, 28.707542419433594, -37.22943115234375, 0.3701019287109375, 12.690322875976562, 9.826416015625, 49.80023193359375, 8.599336624145508, -5.158073425292969, -21.72528076171875, 3.874971389770508, 42.99957275390625, -5.966865539550781, 2.1270751953125, 9.819839477539062, -11.6380615234375, 20.79327392578125, -9.517562866210938, 7.623809814453125, 26.90161895751953, 26.04058837890625, 43.16520690917969, -4.272705078125, 45.939239501953125, 28.01605224609375, 13.492095947265625, 14.835182189941406, 4.80810546875, 9.3486328125, 40.213623046875, 25.977203369140625, 42.93353271484375, 10.147598266601562, 31.60430908203125, 21.381607055664062, 26.622478485107422, -27.955108642578125, 18.83989715576172, 23.36932373046875, 28.091644287109375, -25.383193969726562, 47.152099609375, 21.432952880859375, 7.945610046386719, 5.003173828125, -18.33990478515625, -49.920928955078125, 19.114303588867188, -64.46368408203125, 5.056850433349609, 3.7824325561523438, 36.09765625, 3.8514404296875, 24.717620849609375, 35.62937545776367, -5.7484283447265625, -21.4425048828125, 2.106813430786133, 31.132598876953125, 39.75868225097656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000133.npy"}
|
|
{"epoch": 0.27853403141361255, "step": 134, "batch_size": 128, "mean": 9.493376731872559, "std": 21.198904037475586, "min": -43.2845458984375, "p10": -15.086764526367187, "median": 7.374603271484375, "p90": 34.19742736816406, "max": 90.15252685546875, "pos_frac": 0.671875, "sample": [38.24369812011719, 27.540664672851562, -17.530517578125, -6.81561279296875, -26.352279663085938, -14.016464233398438, 7.9705810546875, -18.529373168945312, -4.48486328125, 26.85748291015625, -39.7861328125, -5.82647705078125, 12.08209228515625, 18.1729736328125, 27.662246704101562, -1.0896015167236328, -4.703496932983398, 9.687808990478516, 22.486846923828125, 7.770214080810547, 13.5120849609375, 59.78106689453125, 0.0, 17.8375244140625, 23.25250244140625, 21.82855224609375, 31.0360107421875, -2.7224960327148438, 4.58294677734375, 47.55108642578125, -29.395734786987305, -6.3784637451171875, 20.378692626953125, 2.4720535278320312, 28.662059783935547, 4.183174133300781, 27.58624267578125, 16.274368286132812, 15.276947021484375, 5.95318603515625, -10.014923095703125, 24.582550048828125, 14.382522583007812, 24.40789794921875, 90.15252685546875, 34.8233642578125, 5.46240234375, -13.58477783203125, 69.75094604492188, 28.44781494140625, -22.833251953125, 35.37824249267578, 10.87225341796875, -12.781147003173828, -0.5565032958984375, 2.070892333984375, 32.1590576171875, 22.656951904296875, 38.39198303222656, 8.306076049804688, -12.057636260986328, -2.73541259765625, 7.187042236328125, 7.562164306640625, -26.52166748046875, 40.65345764160156, -3.293212890625, 15.373016357421875, 26.04443359375, 10.040847778320312, 2.5725574493408203, 18.38751220703125, -17.017112731933594, -4.422527313232422, 2.51971435546875, 40.71092224121094, 11.928863525390625, 25.568878173828125, 6.428401947021484, -33.73750305175781, 17.319091796875, 33.929168701171875, 28.4674072265625, -10.308837890625, -1.49462890625, -3.600860595703125, 9.832962036132812, 15.779928207397461, 42.76531982421875, 1.5456180572509766, -0.7094879150390625, -4.818115234375, 9.234912872314453, 1.12591552734375, -43.2845458984375, 27.123641967773438, 20.951995849609375, 13.41363525390625, 6.8170928955078125, -9.186233520507812, 5.643714904785156, 15.18035888671875, -15.36676025390625, -42.07208251953125, -14.966766357421875, 18.7581787109375, -16.51104736328125, 1.7454833984375, 35.476715087890625, 0.923919677734375, 5.96466064453125, 32.11844253540039, 0.6406478881835938, 23.7762451171875, 40.58856201171875, 28.70065689086914, 33.789119720458984, 4.9017181396484375, 4.28497314453125, -6.536613464355469, -1.436767578125, 0.916046142578125, -3.3736495971679688, -7.44915771484375, 32.80535888671875, 19.293760299682617, 4.17333984375, 0.0], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000134.npy"}
|
|
{"epoch": 0.2806282722513089, "step": 135, "batch_size": 128, "mean": 13.207748413085938, "std": 20.73897361755371, "min": -49.46867370605469, "p10": -9.911980819702148, "median": 10.10927963256836, "p90": 44.43466491699218, "max": 58.794677734375, "pos_frac": 0.78125, "sample": [0.3339500427246094, -27.3861083984375, 12.215347290039062, 22.81842041015625, -9.501605987548828, 41.646820068359375, 0.8490619659423828, 13.29730224609375, 31.64825439453125, 3.2741870880126953, 52.050018310546875, 18.240020751953125, -6.694488525390625, -2.3779449462890625, 11.972412109375, 8.45721435546875, 8.82232666015625, 24.31805419921875, 24.982940673828125, 2.23236083984375, 2.2279891967773438, 7.0485382080078125, -0.708465576171875, 2.3410873413085938, -30.522735595703125, 0.279052734375, 34.703948974609375, 26.242149353027344, 18.000770568847656, 46.958404541015625, 33.39256286621094, 0.20241165161132812, 30.489044189453125, 5.203094482421875, -16.581085205078125, 37.991851806640625, 4.89373779296875, 18.1280517578125, 26.3040771484375, 8.513153076171875, 5.762542724609375, 11.812286376953125, 13.238510131835938, 37.00732421875, 58.794677734375, 20.01769256591797, 44.865264892578125, 10.229904174804688, 3.8011474609375, 0.41162109375, 19.724533081054688, 20.288253784179688, 1.9183502197265625, -10.869522094726562, -18.531951904296875, -8.6234130859375, 44.2501220703125, 5.5776519775390625, 37.61468505859375, 18.4698486328125, 23.892654418945312, 16.181488037109375, 31.881927490234375, -5.28375244140625, -4.238914489746094, 11.687469482421875, -3.734619140625, 21.06005859375, 20.106170654296875, 4.3349151611328125, 16.479347229003906, 5.274616241455078, 51.125885009765625, 16.659423828125, 34.45121765136719, 30.97039031982422, 16.1663818359375, 7.53857421875, -8.2646484375, 49.494140625, 19.58721923828125, -17.137130737304688, -1.7950210571289062, 3.30322265625, 7.28179931640625, 50.32281494140625, 1.92327880859375, 50.99920654296875, 36.653900146484375, 49.207275390625, 20.110504150390625, 38.593994140625, 4.09619140625, -40.35894775390625, 47.6378173828125, 5.5042572021484375, 6.878944396972656, 5.81379508972168, 1.8258056640625, 23.2567138671875, 7.249237060546875, 11.2899169921875, 30.822433471679688, 58.29229736328125, 2.5574684143066406, -0.1800537109375, 4.3751678466796875, 13.805023193359375, 46.48199462890625, 57.452545166015625, -15.0572509765625, -15.969497680664062, 4.146728515625, 11.2508544921875, 0.0, 21.37109375, -20.171142578125, -49.46867370605469, -0.15370941162109375, 9.988655090332031, -5.33148193359375, 20.332717895507812, 6.251983642578125, 43.964569091796875, 28.5570068359375, -12.373062133789062, -20.19085693359375, -0.25628662109375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000135.npy"}
|
|
{"epoch": 0.28272251308900526, "step": 136, "batch_size": 128, "mean": 12.43630599975586, "std": 22.894947052001953, "min": -54.27105712890625, "p10": -9.678085327148438, "median": 9.987274169921875, "p90": 41.40351257324218, "max": 82.04867553710938, "pos_frac": 0.765625, "sample": [42.927764892578125, 16.923202514648438, 1.3349456787109375, 16.177135467529297, 34.867652893066406, 40.0477294921875, 45.88726806640625, 11.222976684570312, 33.19224548339844, 60.535125732421875, 6.028724670410156, 2.9806766510009766, 43.5946044921875, 6.607673645019531, 26.543956756591797, 39.23374938964844, 13.821456909179688, 38.703887939453125, 9.83154296875, 0.470367431640625, -43.075164794921875, 35.966182708740234, 7.71820068359375, 25.034488677978516, -19.091766357421875, 4.642181396484375, -6.84881591796875, -41.227325439453125, 7.189666748046875, 13.90118408203125, 4.6130828857421875, -0.480438232421875, -6.5809326171875, 8.179290771484375, 24.451889038085938, -29.160964965820312, -9.765777587890625, 29.484535217285156, 13.005378723144531, -5.71051025390625, 26.654815673828125, -20.091720581054688, 57.6156005859375, -54.27105712890625, 11.944671630859375, -7.4174957275390625, 3.0550537109375, 4.31121826171875, 17.706817626953125, -4.2154998779296875, 15.54376220703125, -32.934844970703125, 43.85292053222656, 16.209915161132812, 30.93072509765625, 71.36761474609375, -9.6405029296875, 7.3504791259765625, 8.81219482421875, 31.731246948242188, 42.28857421875, 19.75494384765625, 36.724212646484375, -11.770332336425781, 10.755157470703125, 0.1428680419921875, -0.735626220703125, 10.70135498046875, 33.54853820800781, -2.938720703125, 4.3958587646484375, 16.135009765625, 5.679252624511719, 10.223175048828125, 4.363983154296875, 51.473541259765625, -0.6787643432617188, 13.622344970703125, 8.537792205810547, -8.475555419921875, 0.140960693359375, -25.9267578125, 23.2164306640625, 56.836517333984375, 35.6015625, 23.8406982421875, 1.7205543518066406, 16.683349609375, -7.1148681640625, 42.508758544921875, 16.730751037597656, 10.14300537109375, -9.515945434570312, 13.791717529296875, 4.778594970703125, 0.8123703002929688, 57.407623291015625, 14.643165588378906, 29.852096557617188, 22.531890869140625, -0.4574909210205078, 6.76141357421875, 36.8748779296875, 0.1367645263671875, 29.56353759765625, -13.4078369140625, -6.974029541015625, 2.375701904296875, 34.560089111328125, 41.024200439453125, 35.6590576171875, 8.801471710205078, -22.108367919921875, -6.792877197265625, 5.404571533203125, 9.1248779296875, 82.04867553710938, 4.822517395019531, -0.27178955078125, 8.901565551757812, 11.5140380859375, 14.316604614257812, 11.439537048339844, 3.3175201416015625, 16.44451904296875, -50.865142822265625, 30.499908447265625, 5.010246276855469], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000136.npy"}
|
|
{"epoch": 0.2848167539267016, "step": 137, "batch_size": 128, "mean": 11.281706809997559, "std": 21.904430389404297, "min": -47.19598388671875, "p10": -15.777109909057616, "median": 9.023681640625, "p90": 36.968832397460936, "max": 73.71539306640625, "pos_frac": 0.75, "sample": [-30.34130859375, 3.3546142578125, 37.03289794921875, -6.3979644775390625, 11.8350830078125, 18.63323974609375, 38.77687072753906, 12.952301025390625, 2.4023590087890625, 9.500808715820312, 19.733428955078125, 15.944000244140625, 24.942138671875, 55.7432861328125, 17.26903533935547, 3.55218505859375, -1.0083160400390625, 32.56201171875, 1.47467041015625, -13.357666015625, 29.69769287109375, 20.3701171875, 6.437957763671875, 9.92437744140625, -44.430694580078125, 21.541976928710938, 6.03076171875, -16.417388916015625, 29.299484252929688, 7.070465087890625, 48.998931884765625, 15.447509765625, -2.3137969970703125, 38.88458251953125, -18.49273681640625, -0.50372314453125, 3.67498779296875, 7.769981384277344, 0.74627685546875, -8.698974609375, 8.743011474609375, -3.885814666748047, -19.134376525878906, 3.612457275390625, -47.19598388671875, 5.302947998046875, 7.221305847167969, 14.077484130859375, -43.12579345703125, 15.932758331298828, 6.806732177734375, 43.99713134765625, 6.842193603515625, 10.634521484375, 14.254470825195312, 35.306488037109375, -16.701751708984375, 40.73297119140625, -17.91009521484375, 1.113922119140625, 7.300537109375, -0.8274383544921875, 27.6014404296875, -0.8753509521484375, 9.304351806640625, 58.026885986328125, -27.078704833984375, 25.313806533813477, 8.68408203125, -2.684783935546875, 2.1744232177734375, 25.502197265625, 30.4476318359375, 11.754310607910156, 34.48066329956055, 2.220722198486328, 1.2588672637939453, 73.71539306640625, 29.277862548828125, 14.472793579101562, 9.7626953125, 59.42633056640625, -1.1189651489257812, 36.00511169433594, 17.281524658203125, 18.00970458984375, 56.9920654296875, 19.065872192382812, -5.426490783691406, 36.09686279296875, 1.4188461303710938, -14.444198608398438, -1.2417068481445312, 7.189666748046875, 1.8320465087890625, 64.67547607421875, 27.751220703125, 2.9133071899414062, 19.604408264160156, 14.657150268554688, -36.25970458984375, -32.063018798828125, 6.248291015625, -31.339614868164062, -8.755867004394531, 52.901275634765625, -15.502704620361328, 29.94805908203125, 36.941375732421875, 33.53868865966797, 23.042266845703125, 33.912384033203125, -1.0152587890625, 19.790740966796875, 14.78326416015625, 23.045719146728516, -3.4161376953125, 6.964271545410156, 1.060577392578125, 6.16656494140625, 8.50368881225586, 21.055347442626953, 6.3116455078125, 12.348724365234375, 16.120819091796875, 19.974273681640625, 29.076507568359375, -6.11041259765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000137.npy"}
|
|
{"epoch": 0.2869109947643979, "step": 138, "batch_size": 128, "mean": 11.058513641357422, "std": 24.624513626098633, "min": -47.02789306640625, "p10": -17.952958679199217, "median": 8.465286254882812, "p90": 41.51779022216797, "max": 82.01834106445312, "pos_frac": 0.6796875, "sample": [68.036376953125, 41.21844482421875, 46.243316650390625, 18.308448791503906, -13.433570861816406, 6.42791748046875, -6.370361328125, 65.95046997070312, 8.1883544921875, 22.596954345703125, 11.780311584472656, 8.29083251953125, 4.888397216796875, 13.625320434570312, 11.375335693359375, -31.125389099121094, -17.820587158203125, 29.032135009765625, -12.03448486328125, -29.075942993164062, 20.513641357421875, 5.499481201171875, -13.390594482421875, -8.56329345703125, -47.02789306640625, 25.552688598632812, -5.3077392578125, 35.24268341064453, 2.6206741333007812, 82.01834106445312, -6.38287353515625, 6.755584716796875, 3.5228271484375, 42.21626281738281, 1.3480339050292969, 24.1016845703125, -26.471176147460938, 9.605636596679688, 40.59783935546875, -1.2738418579101562, 11.478057861328125, -2.2532005310058594, 31.07904052734375, 36.002349853515625, 16.2801513671875, 32.81907653808594, 4.495147705078125, 2.9358177185058594, 13.4942626953125, -15.801742553710938, 11.841476440429688, 0.043792724609375, 37.694671630859375, -18.261825561523438, 44.674652099609375, 11.96563720703125, 34.900848388671875, -0.6960906982421875, 19.354637145996094, -19.70025634765625, -18.948776245117188, 0.504608154296875, -2.12646484375, -1.190399169921875, 20.904541015625, -30.008071899414062, 65.32135009765625, 32.059532165527344, 27.430389404296875, -0.7734375, -27.417572021484375, 15.5321044921875, -6.711822509765625, 54.273162841796875, 15.351242065429688, -9.464813232421875, -3.8301544189453125, 21.0478515625, 13.224861145019531, 1.0196075439453125, 14.76025390625, 36.9932861328125, 29.424713134765625, 33.624053955078125, -39.9471435546875, -11.01824951171875, 14.49798583984375, 0.20278167724609375, -13.907920837402344, -17.421417236328125, 27.861618041992188, 56.698150634765625, 45.72193908691406, 3.07952880859375, -2.8656749725341797, 29.36663818359375, 26.188262939453125, 27.13165283203125, 3.470123291015625, -5.2064208984375, 29.822296142578125, -16.30413818359375, 72.4874267578125, -5.253074645996094, 8.639739990234375, 14.307861328125, -7.6738739013671875, -17.259796142578125, 6.1246185302734375, 39.628440856933594, -20.043121337890625, 3.505462646484375, 0.6371688842773438, 18.728118896484375, 27.145477294921875, 2.2954330444335938, 56.88330078125, -34.184906005859375, 34.660888671875, -39.25634765625, 4.336658477783203, 22.249053955078125, 6.5411224365234375, 20.542098999023438, 49.3419189453125, 27.902664184570312, 8.884719848632812, -13.678085327148438], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000138.npy"}
|
|
{"epoch": 0.28900523560209423, "step": 139, "batch_size": 128, "mean": 8.110418319702148, "std": 24.183950424194336, "min": -47.3792724609375, "p10": -22.45168762207031, "median": 6.0638427734375, "p90": 40.990425109863274, "max": 68.03799438476562, "pos_frac": 0.6328125, "sample": [-37.5550537109375, -3.7688446044921875, -30.984222412109375, 6.44580078125, -30.93218994140625, 14.596603393554688, 31.49700164794922, 17.8372802734375, -1.372344970703125, -11.24154281616211, 50.448577880859375, 4.5228424072265625, 21.323455810546875, 11.6444091796875, -27.217239379882812, 45.65678405761719, 54.0513916015625, 10.99285888671875, -14.818611145019531, 10.222869873046875, 5.148448944091797, 23.699005126953125, 31.850784301757812, -15.628005981445312, 14.7501220703125, -13.276107788085938, 4.911357879638672, 36.38201904296875, -6.1146240234375, -47.3792724609375, 0.6115837097167969, -7.661346435546875, 36.15458679199219, 40.49761962890625, 25.059112548828125, 0.3275642395019531, 12.356452941894531, -9.637664794921875, 2.716583251953125, 29.416961669921875, -3.0812759399414062, -18.798629760742188, 2.0950927734375, 21.89935302734375, 14.066120147705078, -7.474029541015625, -40.541229248046875, 1.03033447265625, -11.981658935546875, 37.830841064453125, -8.543914794921875, 0.739044189453125, -22.020492553710938, -23.457809448242188, 7.6380615234375, 43.5064697265625, -0.9523773193359375, 1.564453125, -24.16558837890625, -19.345703125, 39.3720703125, -13.363445281982422, 42.14030456542969, 24.243484497070312, 27.886566162109375, 13.6376953125, 0.5795993804931641, -0.4140777587890625, 38.1136474609375, 22.017417907714844, -16.434738159179688, 19.449737548828125, 14.96502685546875, 42.4766845703125, -4.322517395019531, -10.826568603515625, 4.58294677734375, 68.03799438476562, -11.01806640625, 42.84869384765625, 24.736358642578125, 20.80133056640625, 16.156234741210938, 51.96809387207031, 5.681884765625, -14.525520324707031, 54.918212890625, -37.695068359375, 17.210845947265625, -4.701276779174805, 28.11700439453125, 45.931854248046875, 17.50604248046875, 15.324127197265625, -19.681549072265625, 13.325515747070312, -2.1020050048828125, -38.595672607421875, -7.051422119140625, 2.3432884216308594, 23.962326049804688, 6.886474609375, -9.5743408203125, 3.1463851928710938, 13.435165405273438, -9.73699951171875, 27.539047241210938, 66.886474609375, 9.869735717773438, 59.27996826171875, 39.300445556640625, 19.1083984375, 3.137451171875, -34.000823974609375, 6.824592590332031, -32.100189208984375, 0.0, 7.3448333740234375, 13.703201293945312, 33.07828140258789, -7.694549560546875, -34.530670166015625, 20.071609497070312, -4.7637481689453125, 20.92333984375, -14.3319091796875, 35.787017822265625, 5.399192810058594], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000139.npy"}
|
|
{"epoch": 0.29109947643979056, "step": 140, "batch_size": 128, "mean": 14.92657470703125, "std": 22.572914123535156, "min": -42.50605773925781, "p10": -9.982501220703124, "median": 10.344749450683594, "p90": 47.67088012695312, "max": 77.502685546875, "pos_frac": 0.7109375, "sample": [-8.507293701171875, 23.514007568359375, 32.873291015625, -9.801002502441406, -10.405998229980469, 49.59368896484375, -5.459377288818359, 40.818084716796875, -3.4279632568359375, 0.5986328125, -16.439117431640625, 10.353561401367188, 32.1981201171875, 20.759788513183594, 5.9513397216796875, 7.541633605957031, 2.02020263671875, 20.713211059570312, 47.522491455078125, 10.3359375, 8.087814331054688, -17.625778198242188, 28.3946533203125, -26.82110595703125, 20.475189208984375, 39.723114013671875, -4.52166748046875, 6.5549774169921875, 44.04762268066406, 41.318572998046875, 29.7835693359375, 32.433868408203125, 63.67230224609375, 37.822608947753906, -8.850677490234375, 56.712799072265625, -3.484619140625, 4.035003662109375, 48.745391845703125, 49.51625061035156, -2.084228515625, -25.54541015625, 11.360977172851562, -0.3292350769042969, -2.12579345703125, 36.52783203125, 51.29547119140625, 8.45770263671875, -9.191162109375, 77.502685546875, 9.519004821777344, 24.279815673828125, 31.3133544921875, 11.565834045410156, 2.4845733642578125, 21.25250244140625, 21.86248779296875, 5.987775802612305, -8.193115234375, 23.97039031982422, 27.387832641601562, 0.544158935546875, 3.6059188842773438, 21.912811279296875, 5.8039093017578125, 0.789459228515625, -4.424015045166016, -14.37109375, 0.0, -42.50605773925781, -1.173797607421875, -26.83746337890625, 65.77328491210938, 2.2351760864257812, 5.9802398681640625, 14.2998046875, 23.82830810546875, 32.125518798828125, 2.7136077880859375, 48.017120361328125, 10.721153259277344, 3.9187088012695312, 0.22454452514648438, 44.074798583984375, 51.488861083984375, 35.34075164794922, 28.57177734375, -3.145008087158203, 35.018585205078125, 37.98700714111328, 6.169830322265625, 6.450157165527344, 18.321533203125, -4.0498199462890625, 29.787620544433594, 32.42071533203125, 0.0, -5.6059722900390625, 18.27410888671875, -0.7179107666015625, -8.133026123046875, 48.857147216796875, -17.309417724609375, 45.818145751953125, 9.672470092773438, -18.312469482421875, 44.146484375, 19.478805541992188, 22.742523193359375, 22.73760986328125, 43.54742431640625, 22.51873779296875, 50.45916748046875, 20.81664276123047, 26.464569091796875, -8.889175415039062, -1.04754638671875, -15.260528564453125, 8.760360717773438, -25.18012237548828, 15.073028564453125, 54.9288330078125, 10.13665771484375, -2.8503799438476562, 7.390106201171875, -17.20928955078125, 26.618194580078125, 22.985870361328125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000140.npy"}
|
|
{"epoch": 0.2931937172774869, "step": 141, "batch_size": 128, "mean": 14.254927635192871, "std": 25.203052520751953, "min": -76.00277709960938, "p10": -10.683622741699217, "median": 12.65909194946289, "p90": 47.476626586914065, "max": 87.748291015625, "pos_frac": 0.734375, "sample": [41.17921447753906, 18.77154541015625, -25.401844024658203, -9.338104248046875, 6.50982666015625, 13.985382080078125, -8.234222412109375, 21.707550048828125, -47.92901611328125, 87.748291015625, 5.796516418457031, 10.9825439453125, 30.974594116210938, 32.798583984375, 11.8681640625, 44.254119873046875, 23.769775390625, 18.18609619140625, 16.98305892944336, 22.487884521484375, 57.669189453125, -4.7489013671875, 1.7665023803710938, 43.452667236328125, -0.18494606018066406, -14.402996063232422, 38.24213790893555, 12.865837097167969, 2.4139862060546875, 49.50953674316406, 34.7596435546875, -16.359222412109375, -10.44122314453125, 69.10430908203125, 3.7327880859375, 17.239837646484375, -25.605796813964844, -8.259679794311523, 10.892562866210938, -18.55474853515625, 41.2432861328125, 36.01904296875, 12.972990036010742, 19.980911254882812, -6.463214874267578, 8.6998291015625, 1.57745361328125, -7.8839263916015625, 59.85223388671875, 7.826934814453125, 11.51922607421875, 24.901214599609375, 2.9148998260498047, -10.409439086914062, 28.6712646484375, 23.096466064453125, 1.5720367431640625, 1.16400146484375, 26.067047119140625, 0.515045166015625, 34.326019287109375, 33.818382263183594, -15.3861083984375, -44.20973205566406, -5.5083770751953125, 68.67559814453125, -1.8659381866455078, 25.95769500732422, -5.4915924072265625, 36.66813659667969, 3.621917724609375, 12.636360168457031, 6.404483795166016, 24.09166717529297, 28.91680908203125, 35.59198760986328, 32.94403076171875, -7.639495849609375, 0.0, -13.977432250976562, 41.84831237792969, -2.19775390625, -0.249664306640625, 59.11175537109375, 6.975730895996094, 53.54779052734375, -76.00277709960938, 17.8050537109375, 18.13330078125, -1.182830810546875, 23.849929809570312, 18.53160858154297, 32.264625549316406, -44.89433288574219, -6.0724945068359375, 17.203933715820312, 0.7236480712890625, 55.36824035644531, -0.502716064453125, 32.257835388183594, -7.3623809814453125, 12.53375244140625, 19.253387451171875, 39.0018310546875, 14.233245849609375, 4.2125396728515625, 25.66375732421875, 47.625244140625, 42.02259826660156, -11.249221801757812, 7.581027984619141, 12.68182373046875, 58.9176025390625, -36.11378479003906, 1.981414794921875, 9.6361083984375, 10.98388671875, 12.985931396484375, 20.2159423828125, 20.066070556640625, -8.500389099121094, 47.412933349609375, 48.40234375, 7.889495849609375, 11.91851806640625, 55.76153564453125, 15.217819213867188, 1.53533935546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000141.npy"}
|
|
{"epoch": 0.29528795811518327, "step": 142, "batch_size": 128, "mean": 13.999675750732422, "std": 25.368560791015625, "min": -66.05197143554688, "p10": -13.96692504882812, "median": 9.304386138916016, "p90": 48.38865356445312, "max": 81.23388671875, "pos_frac": 0.671875, "sample": [-2.005462646484375, -1.8288345336914062, 9.456485748291016, 27.90167236328125, 2.5727405548095703, -18.160400390625, 8.963531494140625, -1.61065673828125, 72.99832153320312, 40.64350891113281, 44.0634765625, -3.79638671875, -7.185028076171875, 36.329071044921875, 16.93017578125, 28.965240478515625, 0.6335334777832031, -2.8672332763671875, 52.28961181640625, 25.649551391601562, -1.97869873046875, 25.354888916015625, 0.0, 20.9007568359375, 21.19329833984375, 48.301055908203125, 11.083984375, 3.4012908935546875, -3.0951156616210938, 0.381011962890625, -3.128692626953125, 27.231689453125, -2.6297569274902344, -24.5062255859375, 2.76458740234375, 2.3817710876464844, 23.9317626953125, 13.713638305664062, -22.077041625976562, 27.85906982421875, 20.354736328125, 50.699462890625, 42.50189208984375, 16.583755493164062, 17.327880859375, -10.176025390625, -1.1789283752441406, 45.30010986328125, 21.045860290527344, 69.70770263671875, 15.2315673828125, -6.330793380737305, 4.05096435546875, 13.559249877929688, -12.43914794921875, 2.607097625732422, 0.313934326171875, 35.27227783203125, -17.53173828125, -19.782913208007812, 21.21337890625, 33.88818359375, -12.086196899414062, -8.2568359375, 48.593048095703125, 37.61115264892578, 18.84893798828125, 1.1449394226074219, 45.75639343261719, -3.1069602966308594, 7.6986846923828125, 20.9339599609375, 0.152740478515625, 6.31239128112793, -24.645416259765625, 29.045936584472656, -39.423248291015625, 27.57823944091797, 44.148193359375, -25.210540771484375, -4.383453369140625, 68.83151245117188, 81.23388671875, -1.294464111328125, -1.8475875854492188, 10.675849914550781, 9.409011840820312, 51.447021484375, 9.199760437011719, 6.242382049560547, -2.6227455139160156, -3.655750274658203, 33.813568115234375, 41.77178955078125, 56.63775634765625, 6.070320129394531, 36.042144775390625, 26.869308471679688, 20.16265869140625, 53.577545166015625, -66.05197143554688, 53.56085205078125, -30.498809814453125, 1.1251392364501953, -0.3957691192626953, -27.549591064453125, 35.168426513671875, 17.758132934570312, 31.52374267578125, -2.01348876953125, 36.90374755859375, -35.779815673828125, 3.6828231811523438, 5.9765625, 19.368995666503906, -6.966239929199219, 29.65521240234375, -1.4797325134277344, 3.646697998046875, 1.899749755859375, -23.16864013671875, 62.514404296875, -0.13477134704589844, -0.62908935546875, 31.95654296875, 28.108917236328125, 66.06706237792969, 41.18888854980469], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000142.npy"}
|
|
{"epoch": 0.2973821989528796, "step": 143, "batch_size": 128, "mean": 16.023521423339844, "std": 27.279359817504883, "min": -78.172607421875, "p10": -10.750225830078124, "median": 12.724166870117188, "p90": 48.02594299316406, "max": 98.9505615234375, "pos_frac": 0.7421875, "sample": [23.353071212768555, 48.638763427734375, 54.94537353515625, 28.4539794921875, 39.90740203857422, 16.8714599609375, 30.575286865234375, 9.467475891113281, -16.274856567382812, -7.554826736450195, 28.5660400390625, 0.04532623291015625, 22.271194458007812, 13.041183471679688, 2.3290939331054688, -11.211181640625, 47.7633056640625, 3.88726806640625, -17.324722290039062, 98.9505615234375, -19.7711181640625, -6.880950927734375, 44.2041015625, 46.534515380859375, 21.759376525878906, 44.661712646484375, 34.831817626953125, 62.25848388671875, 87.28836059570312, 0.39093017578125, -3.34564208984375, 42.36048126220703, 27.884719848632812, 33.00595474243164, 25.32927703857422, -5.36956787109375, -78.172607421875, -2.4771499633789062, 28.724403381347656, 7.78912353515625, 8.952461242675781, -31.83136749267578, 8.059249877929688, 40.06072998046875, 5.008880615234375, 6.24700927734375, 10.714681625366211, 21.391647338867188, 15.397586822509766, -63.97216796875, 5.77630615234375, 26.35186767578125, 2.5304946899414062, 38.56982421875, 19.787841796875, 23.5859375, -0.75634765625, 21.548736572265625, -13.531524658203125, -0.7366180419921875, -10.06787109375, 0.9715995788574219, 68.64178466796875, 20.640472412109375, 5.446889877319336, -2.3496322631835938, 58.05426025390625, 32.84135437011719, 40.777099609375, 2.101470947265625, 31.328094482421875, 35.26069641113281, -6.434947967529297, 3.852458953857422, 32.673912048339844, -9.3345947265625, -37.91680908203125, -0.0808258056640625, 12.093719482421875, 12.407150268554688, 4.802337646484375, -1.35491943359375, 7.863609313964844, 17.461456298828125, 10.79339599609375, -2.1448974609375, -18.089279174804688, 2.6239051818847656, -0.8683319091796875, 13.310928344726562, 35.25437927246094, 15.297470092773438, -3.4370803833007812, 21.983779907226562, 20.6767578125, 27.85089111328125, 0.3548851013183594, 15.9505615234375, -4.694684982299805, 1.5389251708984375, 81.76626586914062, 29.683319091796875, -6.290618896484375, 11.598182678222656, -30.76214599609375, 74.55862426757812, 33.24359893798828, 3.3802433013916016, 3.838184356689453, -1.7444610595703125, 27.9224853515625, 57.363128662109375, 28.172943115234375, 65.67044067382812, 25.251373291015625, 62.467529296875, 42.195343017578125, 43.6385498046875, -38.773651123046875, 26.205570220947266, -16.594207763671875, 50.40380859375, -10.55267333984375, 10.9666748046875, 46.55322265625, 7.164581298828125, 15.55426025390625, 1.1873245239257812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000143.npy"}
|
|
{"epoch": 0.2994764397905759, "step": 144, "batch_size": 128, "mean": 13.223828315734863, "std": 24.80944061279297, "min": -66.84848022460938, "p10": -12.91422348022461, "median": 6.61932373046875, "p90": 45.23160858154297, "max": 90.33837890625, "pos_frac": 0.734375, "sample": [-0.6902618408203125, 18.525070190429688, 42.850494384765625, 4.938682556152344, 4.670196533203125, 12.63616943359375, -22.395111083984375, -10.145462036132812, 1.2817916870117188, 90.33837890625, 2.544952392578125, -11.455291748046875, 6.797252655029297, 58.260162353515625, 3.952301025390625, 34.302978515625, 1.566925048828125, 45.17100524902344, 9.26953125, 0.78216552734375, 3.1751556396484375, 20.4803466796875, -12.905830383300781, 3.0229263305664062, 2.7762451171875, -6.17755126953125, -21.61443328857422, -2.521442413330078, 26.952682495117188, 5.008516311645508, -35.7908935546875, 5.790639877319336, -1.502685546875, 24.60601806640625, 50.98185729980469, 39.40116882324219, 29.760955810546875, 72.71295166015625, 6.0810699462890625, 57.311187744140625, -1.0968456268310547, -9.012359619140625, 5.69140625, 18.85064697265625, -0.6081829071044922, -16.56787109375, 8.3682861328125, 34.978729248046875, -16.417999267578125, 22.681182861328125, 12.788009643554688, 55.148468017578125, 36.825286865234375, -16.89422607421875, -66.84848022460938, -26.848663330078125, 15.00775146484375, -12.357292175292969, -11.01898193359375, 1.1424713134765625, 33.1116943359375, 0.66436767578125, -8.6683349609375, 6.4515380859375, 27.667823791503906, 11.37713623046875, 66.360595703125, 62.21343994140625, 38.37132263183594, 40.046630859375, 1.7352142333984375, 20.601829528808594, 46.83477783203125, -28.638198852539062, 21.894699096679688, 24.83599853515625, 3.955108642578125, 10.058319091796875, 46.032958984375, 39.2960205078125, 43.82460021972656, 37.501033782958984, 4.897796630859375, -6.675666809082031, 14.062408447265625, 24.699790954589844, -0.699737548828125, 4.053985595703125, 2.7579078674316406, 32.41455078125, 69.052001953125, 32.33154296875, 12.231826782226562, 0.741485595703125, 11.584512710571289, 45.373016357421875, 2.59027099609375, -2.43194580078125, 6.83392333984375, 35.02392578125, 9.583953857421875, 0.21507835388183594, 0.0, 1.2406463623046875, -12.933807373046875, -11.492919921875, 3.56890869140625, -33.33314514160156, 26.251220703125, 24.841415405273438, 18.6260986328125, -2.176361083984375, 41.88446044921875, -7.733268737792969, 26.808624267578125, 3.3729248046875, 19.927764892578125, 5.402198791503906, 18.1451416015625, -23.962249755859375, 11.36279296875, 37.69380187988281, 6.787109375, 43.63482666015625, -18.36328125, -0.70294189453125, 70.02215576171875, 5.046539306640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000144.npy"}
|
|
{"epoch": 0.30157068062827225, "step": 145, "batch_size": 128, "mean": 16.933469772338867, "std": 25.59073829650879, "min": -53.67051696777344, "p10": -13.466324615478516, "median": 15.468765258789062, "p90": 50.14303283691406, "max": 81.91082763671875, "pos_frac": 0.703125, "sample": [-13.697418212890625, 38.57122802734375, -40.776214599609375, 35.796142578125, 13.634815216064453, -27.949310302734375, 40.12152099609375, 18.422088623046875, 75.79269409179688, 26.0885009765625, 22.58648681640625, 11.1129150390625, 23.482093811035156, 33.35209655761719, 30.005409240722656, 62.32720947265625, 3.994293212890625, 29.078643798828125, 9.648887634277344, 6.828483581542969, 2.0685577392578125, -6.77069091796875, 50.2979736328125, -7.107151031494141, -7.738670349121094, 68.35809326171875, -27.712318420410156, 48.39532470703125, 1.0521240234375, 19.00006103515625, 3.690185546875, 30.398521423339844, 8.743461608886719, 35.52215576171875, 54.889617919921875, 22.4591064453125, 72.01144409179688, -53.67051696777344, 8.67364501953125, -1.131561279296875, 16.814361572265625, 18.1324462890625, -13.519767761230469, 61.454864501953125, 67.56182861328125, 14.3514404296875, 37.041412353515625, 34.348236083984375, 12.077049255371094, 19.8492431640625, 18.38623809814453, -7.2049407958984375, -6.451042175292969, 33.20794677734375, 46.49293518066406, 41.95616149902344, -0.52166748046875, -4.367347717285156, -14.586219787597656, 3.6527748107910156, 16.586090087890625, 6.672454833984375, -4.63531494140625, 43.561126708984375, 61.732177734375, -5.045013427734375, 56.796905517578125, -9.20379638671875, 55.15840148925781, 29.912424087524414, 48.592193603515625, -10.607101440429688, 6.274700164794922, 32.49873352050781, 44.8968505859375, 11.814178466796875, 0.5463371276855469, 11.26708984375, 26.695205688476562, -23.189132690429688, -18.1363525390625, 5.54620361328125, 8.921642303466797, 27.941986083984375, 81.91082763671875, 20.49896240234375, 5.26611328125, 25.7391357421875, -23.8013916015625, -11.076675415039062, -2.8481178283691406, 0.0, 8.169891357421875, 19.004913330078125, 9.135379791259766, 3.31689453125, 40.448486328125, 70.85589599609375, 38.973907470703125, 24.079238891601562, -1.8702926635742188, -10.223419189453125, 24.5855712890625, 24.114524841308594, 12.79058837890625, -13.44342041015625, 25.014976501464844, 27.91851806640625, -2.8541831970214844, -1.1809778213500977, -15.44525146484375, 48.141876220703125, 26.96643829345703, 44.9229736328125, -2.0140457153320312, -2.864715576171875, 3.6309890747070312, 28.933536529541016, 39.713531494140625, 50.076629638671875, 32.275848388671875, -13.962997436523438, -1.8513851165771484, -4.0532379150390625, 26.5045166015625, -21.295257568359375, -8.555145263671875, 18.710464477539062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000145.npy"}
|
|
{"epoch": 0.3036649214659686, "step": 146, "batch_size": 128, "mean": 13.377705574035645, "std": 24.73737144470215, "min": -60.68902587890625, "p10": -14.760281372070311, "median": 7.5244140625, "p90": 44.53443298339843, "max": 77.521728515625, "pos_frac": 0.7265625, "sample": [5.223319053649902, 2.298004150390625, 18.95572280883789, -30.313140869140625, 10.64459228515625, 39.737579345703125, -25.728378295898438, 10.5194091796875, 33.46319580078125, -23.405548095703125, 77.521728515625, 72.130126953125, 11.37750244140625, -8.15191650390625, 15.381072998046875, 2.4924240112304688, 39.848907470703125, 20.682708740234375, 2.9651641845703125, 2.594207763671875, 28.60223388671875, 2.1419448852539062, 11.101558685302734, 9.563934326171875, -18.52935791015625, -10.248550415039062, 3.7022933959960938, 19.751800537109375, 17.22675323486328, 19.44354248046875, 11.770118713378906, 0.7885189056396484, 38.8765869140625, 0.386016845703125, 6.22509765625, 18.650787353515625, 15.525222778320312, -21.9451904296875, 10.543663024902344, 3.2427749633789062, 7.433349609375, 48.033935546875, 1.1639862060546875, 34.64208984375, 0.0, 46.358154296875, -5.912353515625, -1.282339096069336, 28.483383178710938, 26.74395751953125, 7.615478515625, -14.969268798828125, 7.2137603759765625, -1.6025390625, 50.7115478515625, 59.12127685546875, -59.92303466796875, 30.822509765625, 5.3307037353515625, 33.65956115722656, 35.131622314453125, 5.040740966796875, 2.7707443237304688, 47.19158935546875, 34.661865234375, -12.89715576171875, -7.9791259765625, -12.789627075195312, 6.820709228515625, -19.508880615234375, 50.02522277832031, -17.998443603515625, 6.13238525390625, 37.735626220703125, -21.667510986328125, 60.11541748046875, 8.1585693359375, 6.974250793457031, -10.406158447265625, 69.603759765625, 7.366546630859375, 32.5946044921875, 2.6493358612060547, -11.2164306640625, 34.131134033203125, -0.3895072937011719, -6.856201171875, 4.3209686279296875, 0.0, 38.011756896972656, -3.3089447021484375, -4.1209564208984375, 21.319473266601562, -23.210205078125, 3.2144775390625, 58.785858154296875, 22.147705078125, 38.672088623046875, 7.036674499511719, 50.55218505859375, -22.1397705078125, -14.67071533203125, 1.5993270874023438, 12.47955322265625, 38.427978515625, -8.275375366210938, 36.281494140625, 5.850006103515625, 50.195953369140625, 43.752838134765625, 43.468780517578125, -60.68902587890625, 28.51266860961914, 42.083473205566406, 15.1697998046875, 36.575439453125, -9.596481323242188, 42.67754364013672, 1.2798194885253906, 35.16233825683594, -0.918792724609375, 5.894695281982422, 19.78778076171875, 42.248626708984375, 29.394622802734375, -11.27492904663086, 30.987442016601562, -9.43548583984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000146.npy"}
|
|
{"epoch": 0.3057591623036649, "step": 147, "batch_size": 128, "mean": 11.431533813476562, "std": 27.80303955078125, "min": -75.2735595703125, "p10": -18.713284301757813, "median": 8.40777587890625, "p90": 47.82854690551758, "max": 86.2650146484375, "pos_frac": 0.671875, "sample": [0.517364501953125, -2.69921875, 6.840137481689453, 33.67344665527344, -8.927513122558594, -2.3753204345703125, 83.10623168945312, -17.919204711914062, 0.0, 68.46575927734375, -18.53070068359375, 22.728363037109375, 13.532501220703125, 8.119140625, 25.664331436157227, 4.456390380859375, -19.139312744140625, 42.693756103515625, 53.4498291015625, 3.4997100830078125, -4.492401123046875, 20.08837890625, 0.4295501708984375, 36.42346954345703, -1.31561279296875, 49.23481750488281, 0.8994102478027344, 42.83323669433594, 11.889007568359375, 2.7481689453125, 67.63656616210938, 34.56976318359375, 17.550090789794922, 21.789657592773438, 14.93719482421875, -30.10614013671875, -41.199012756347656, -6.233428955078125, -17.43463134765625, 16.973243713378906, 12.3983154296875, 31.136600494384766, 31.800464630126953, 5.4772796630859375, -26.90279769897461, 26.98114013671875, -28.70477294921875, -1.9112091064453125, -0.6124362945556641, -0.37332916259765625, -3.6058197021484375, 49.958953857421875, 10.883132934570312, 12.086685180664062, 27.24341583251953, 29.777366638183594, 78.20428466796875, 11.8865966796875, 4.5775604248046875, -4.6074676513671875, 21.260986328125, 19.1602783203125, 58.30101013183594, 6.711322784423828, 50.0592041015625, 1.557708740234375, 1.783203125, 57.669952392578125, 38.467952728271484, 3.6654510498046875, -41.587646484375, 10.5435791015625, 2.05926513671875, 0.406036376953125, 30.06005859375, 6.0205841064453125, 10.913778305053711, 5.435691833496094, -0.046504974365234375, -0.2396240234375, 26.999221801757812, -8.547225952148438, 0.4231414794921875, 35.55438232421875, 16.041595458984375, 36.26507568359375, 8.6964111328125, 32.08013916015625, -12.564453125, 29.848983764648438, 12.979827880859375, 43.70751953125, -27.21630859375, -75.2735595703125, -37.39546203613281, -1.0994186401367188, 11.180862426757812, 24.490570068359375, -16.87310791015625, -41.28675842285156, -9.370437622070312, 4.4217071533203125, 47.564666748046875, 13.6483154296875, -18.4508056640625, -8.22475814819336, 44.46852111816406, -3.304962158203125, -37.79127502441406, -14.291854858398438, -64.2608413696289, 9.56378173828125, 7.9449310302734375, 37.46863555908203, 48.44426727294922, 37.0318603515625, 8.014236450195312, 54.042327880859375, -3.45660400390625, -7.439605712890625, 9.419502258300781, 41.19673156738281, 43.895233154296875, 18.4742431640625, 86.2650146484375, 9.466758728027344, -4.328348159790039, -27.45953369140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000147.npy"}
|
|
{"epoch": 0.3078534031413613, "step": 148, "batch_size": 128, "mean": 12.034950256347656, "std": 25.816137313842773, "min": -55.734100341796875, "p10": -22.275946044921874, "median": 10.424476623535156, "p90": 46.74147415161133, "max": 91.00704956054688, "pos_frac": 0.6484375, "sample": [8.6270751953125, 12.935592651367188, 18.617233276367188, -14.243377685546875, -1.8265609741210938, 67.95857238769531, -0.2741050720214844, 18.5771484375, 72.62786865234375, 24.179725646972656, 5.381610870361328, 21.340988159179688, -4.735870361328125, 53.624114990234375, 15.62890625, 36.38124084472656, 10.3778076171875, -20.744140625, 17.784698486328125, 22.476306915283203, -21.8526611328125, 41.44439697265625, -46.53111267089844, 0.3168449401855469, 7.9652099609375, -23.26361083984375, 13.770248413085938, -5.662269592285156, 13.744049072265625, -8.576021194458008, 17.00194549560547, 16.94732666015625, -8.328681945800781, -55.734100341796875, -3.26849365234375, -13.04644775390625, 60.488677978515625, 7.10443115234375, -21.71563720703125, 20.443267822265625, 56.75799560546875, 1.2925033569335938, -20.478424072265625, -24.09552001953125, 1.5416717529296875, 5.880216598510742, 1.922149658203125, -14.764678955078125, 18.2047119140625, 25.495941162109375, 7.782550811767578, 8.8524169921875, 6.8045654296875, 18.275848388671875, -25.258956909179688, 62.912506103515625, 9.426017761230469, 36.958160400390625, 10.471145629882812, 45.23065185546875, 22.068878173828125, -3.5819854736328125, -0.03453826904296875, -1.5290374755859375, 0.0, 7.979667663574219, 54.53070068359375, 19.694625854492188, 35.34624481201172, 39.54486083984375, -3.53900146484375, -26.48424530029297, -5.7083587646484375, -33.3441162109375, 34.97418212890625, 11.058296203613281, -25.692710876464844, 32.364105224609375, -2.447967529296875, 29.256515502929688, -1.2034664154052734, 18.01709747314453, -24.47991943359375, 22.99615478515625, 54.20942687988281, 14.65066146850586, 1.8731632232666016, 15.9346923828125, -10.150901794433594, -13.97100830078125, 25.732154846191406, 24.42330551147461, -0.8279876708984375, 27.166824340820312, 21.295562744140625, 46.61161804199219, -23.268524169921875, 8.315528869628906, -10.750907897949219, 61.62123107910156, 2.7675628662109375, 24.09027099609375, 91.00704956054688, 36.238861083984375, 2.15472412109375, -6.514995574951172, 26.038646697998047, -26.172271728515625, 31.70751953125, 25.860069274902344, 47.044471740722656, 44.12548828125, 16.467742919921875, 27.923492431640625, -31.712852478027344, -2.9458541870117188, -25.171142578125, 15.671539306640625, 62.096466064453125, 0.0, -8.44940185546875, 12.0709228515625, 33.83832550048828, -0.7811622619628906, 14.91632080078125, 43.52479553222656, 56.61592102050781, -3.7432937622070312], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000148.npy"}
|
|
{"epoch": 0.3099476439790576, "step": 149, "batch_size": 128, "mean": 14.814956665039062, "std": 24.98700523376465, "min": -40.90972900390625, "p10": -14.360104370117185, "median": 11.840805053710938, "p90": 44.81778411865234, "max": 95.3284912109375, "pos_frac": 0.7578125, "sample": [46.148590087890625, -15.659210205078125, 23.66973876953125, 33.81050109863281, 43.253997802734375, 46.27978515625, 17.513900756835938, 4.657398223876953, 8.4256591796875, 5.8945770263671875, 24.46551513671875, 3.40679931640625, 44.54681396484375, 20.008220672607422, 23.130828857421875, -2.900421142578125, 45.38507080078125, 10.370468139648438, -1.542510986328125, -8.847854614257812, 13.899620056152344, 4.586845397949219, 19.30816650390625, 3.4563751220703125, 17.768291473388672, 12.777130126953125, -38.59465789794922, 28.7525634765625, 48.088836669921875, 14.202667236328125, 8.727249145507812, 19.74676513671875, 26.47479248046875, 13.302993774414062, 1.754817008972168, -0.4425506591796875, 44.57466125488281, 2.2617015838623047, -13.8033447265625, 1.08905029296875, 33.61933898925781, 71.5135498046875, -6.728034973144531, 2.3896331787109375, 74.92042541503906, 40.202880859375, 10.135025024414062, -28.422164916992188, 37.132537841796875, 50.6937255859375, 19.811325073242188, -3.2185211181640625, 9.270965576171875, 19.49018096923828, 5.7790069580078125, 75.9158935546875, -31.04278564453125, 14.37591552734375, 15.1766357421875, 10.90447998046875, -9.11328125, 15.532997131347656, -23.413726806640625, 32.83717346191406, -25.816375732421875, 3.5970458984375, 8.70947265625, -4.1107177734375, 21.589599609375, 21.21629524230957, 39.07179641723633, 24.00921630859375, 27.093158721923828, 12.907501220703125, 36.743682861328125, 13.535514831542969, 43.0052490234375, 7.343952178955078, 55.57421875, -0.08275222778320312, 40.95172119140625, 0.3948516845703125, 31.81732177734375, 25.7459716796875, 10.48760986328125, -9.12939453125, 51.42266845703125, 34.475830078125, -38.400238037109375, 0.6147232055664062, 9.960296630859375, 27.26873779296875, -3.95098876953125, -23.019775390625, -3.427703857421875, -4.999176025390625, 62.965972900390625, 7.22735595703125, 39.598876953125, 8.03082275390625, -1.9383544921875, 4.9708404541015625, 31.518325805664062, -0.07586669921875, 27.709091186523438, 5.89324951171875, -27.52740478515625, -23.933868408203125, 0.911834716796875, 95.3284912109375, -1.6256103515625, 2.312591552734375, 9.59063720703125, 25.881332397460938, 5.212554931640625, -11.989212036132812, 40.943519592285156, 30.089035034179688, -40.90972900390625, -39.57232666015625, 21.210189819335938, 37.03192138671875, -25.290145874023438, 70.40158081054688, 41.85536193847656, 17.660751342773438, 10.401336669921875, 0.1188812255859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000149.npy"}
|
|
{"epoch": 0.31204188481675393, "step": 150, "batch_size": 128, "mean": 14.235536575317383, "std": 24.06322479248047, "min": -45.4034423828125, "p10": -10.479516601562498, "median": 11.356636047363281, "p90": 43.119817352294916, "max": 79.87570190429688, "pos_frac": 0.75, "sample": [22.663604736328125, 27.294395446777344, -3.723346710205078, 40.109344482421875, 12.96746826171875, -20.23388671875, -6.38916015625, 23.033523559570312, -29.39544677734375, 36.7811279296875, 24.72808074951172, 27.973388671875, 39.70721435546875, 15.38836669921875, -13.52203369140625, 4.65203857421875, 7.59393310546875, 33.465606689453125, 7.542236328125, 36.33868408203125, 24.043502807617188, -10.069259643554688, -2.073638916015625, 3.7350006103515625, -5.849164962768555, 10.98095703125, 8.92694091796875, 1.670806884765625, 63.92201232910156, 28.63188934326172, 65.2493896484375, 1.57135009765625, 14.57330322265625, 8.371271133422852, 17.303054809570312, 25.315811157226562, 20.650054931640625, -44.11578369140625, 38.77396774291992, 11.330886840820312, 18.595184326171875, 41.74943542480469, 15.262908935546875, 38.48152160644531, -6.6067352294921875, -3.3621597290039062, 1.8313064575195312, 3.4439697265625, 12.59686279296875, 60.2900390625, 10.177248001098633, 73.77622985839844, 9.334197998046875, 31.822525024414062, 43.9906005859375, 54.628631591796875, 42.827880859375, 6.250001907348633, 0.5875396728515625, 1.25946044921875, 2.806396484375, -1.0770339965820312, -24.102325439453125, 43.801002502441406, -0.736053466796875, 0.1080780029296875, -0.590240478515625, 0.6641998291015625, 41.60487365722656, 46.189056396484375, 19.170196533203125, 38.0909423828125, -0.7104949951171875, 41.1859130859375, 17.963043212890625, -9.900909423828125, -40.0081787109375, 7.8275146484375, 2.8465805053710938, 23.323211669921875, 9.971168518066406, -39.315704345703125, -4.8550262451171875, -26.733688354492188, -4.86090087890625, 15.92974853515625, -18.53472900390625, 79.87570190429688, 34.26385498046875, 9.85107421875, 12.594085693359375, -9.694000244140625, 6.049774169921875, 29.19952392578125, -45.4034423828125, 8.101959228515625, 38.74400329589844, -1.6326904296875, -18.780242919921875, 1.0798721313476562, 23.520843505859375, 37.679481506347656, 14.713531494140625, 50.574920654296875, 7.926898956298828, 4.37091064453125, 39.2646484375, 13.843093872070312, 40.557159423828125, 29.177703857421875, -0.778594970703125, 11.38238525390625, 3.0476417541503906, 51.896240234375, 11.676895141601562, 0.35074615478515625, 22.22032928466797, -11.436782836914062, 54.36663818359375, -8.898635864257812, 0.0, 22.742095947265625, 72.1251220703125, 0.994720458984375, 14.146244049072266, 13.929229736328125, 20.93376922607422, -29.3387451171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000150.npy"}
|
|
{"epoch": 0.31413612565445026, "step": 151, "batch_size": 128, "mean": 20.67343521118164, "std": 27.296838760375977, "min": -51.286895751953125, "p10": -13.421916198730468, "median": 23.102447509765625, "p90": 52.22318115234374, "max": 82.56488037109375, "pos_frac": 0.75, "sample": [31.064590454101562, 21.175445556640625, 0.4060096740722656, -23.269691467285156, 0.08539199829101562, 75.96954345703125, 51.2669677734375, -21.38927459716797, 5.727294921875, 39.615020751953125, -1.795074462890625, -9.891914367675781, 29.25387191772461, 35.680511474609375, -7.05596923828125, -13.079536437988281, -10.264389038085938, 33.400238037109375, 36.021697998046875, 27.427757263183594, -51.286895751953125, 4.405792236328125, -6.93035888671875, 11.998458862304688, 2.01611328125, 31.8568115234375, -2.4296417236328125, -4.143402099609375, 31.298370361328125, 45.903472900390625, 50.58251953125, -0.1908111572265625, -1.564117431640625, 17.272186279296875, 33.93562316894531, 21.250457763671875, 12.15386962890625, 15.663185119628906, -4.809154510498047, 77.23797607421875, 24.737930297851562, 31.205596923828125, 12.961982727050781, 18.205596923828125, 44.46449279785156, -26.44635009765625, 30.700714111328125, 9.1697998046875, 43.259674072265625, 43.97200012207031, 42.02545928955078, -9.469482421875, 2.1421260833740234, 0.8798446655273438, 30.386444091796875, 40.00959777832031, 15.13189697265625, 35.26885986328125, -27.264617919921875, 18.916915893554688, -4.952880859375, 72.0704345703125, 41.6658935546875, -3.4233016967773438, 17.240554809570312, 32.98103332519531, 37.707275390625, -21.3306884765625, 27.06549072265625, 58.87908935546875, -35.32929992675781, -17.89813232421875, 41.300537109375, 37.76091766357422, 40.20105743408203, 19.546417236328125, 49.84832763671875, 37.83877182006836, 63.7222900390625, -23.4364013671875, -26.2430419921875, 38.040557861328125, 68.1883544921875, 39.35736083984375, 73.28402709960938, 7.741020202636719, 43.249176025390625, -1.0254020690917969, 0.0, 23.34844970703125, 32.744476318359375, 34.008056640625, 31.073287963867188, -12.25592041015625, 24.21502685546875, 32.279693603515625, 49.6024169921875, 67.49221801757812, -14.220802307128906, 82.56488037109375, -36.155914306640625, 24.87774658203125, 3.2674560546875, 30.270980834960938, 54.454345703125, 44.8359375, 7.141685485839844, -9.771713256835938, 30.372257232666016, 21.406402587890625, 29.309677124023438, 58.2247314453125, 11.257415771484375, -8.178810119628906, 76.0302734375, 19.627685546875, 48.93701171875, 50.026123046875, 63.368316650390625, 22.8564453125, 28.67352294921875, 4.9623565673828125, 9.275909423828125, 21.650390625, 28.797550201416016, 2.455810546875, 10.0806884765625, -31.583271026611328], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000151.npy"}
|
|
{"epoch": 0.3162303664921466, "step": 152, "batch_size": 128, "mean": 15.709592819213867, "std": 27.156776428222656, "min": -54.31108093261719, "p10": -15.061877441406242, "median": 13.781688690185547, "p90": 52.866809082031246, "max": 97.18634033203125, "pos_frac": 0.703125, "sample": [32.70758056640625, 46.54217529296875, -0.1204681396484375, 13.613410949707031, 5.78070068359375, 34.6263427734375, 11.493621826171875, 32.464019775390625, 53.7506103515625, 18.429046630859375, 7.830257415771484, 34.626678466796875, 66.31968688964844, 7.785430908203125, 29.1527099609375, -20.112045288085938, 1.725860595703125, 1.691131591796875, 1.8580322265625, -21.768829345703125, -54.31108093261719, 29.6595458984375, -47.615081787109375, 24.2689208984375, -8.856670379638672, 23.212188720703125, 23.250770568847656, -0.8056640625, 49.446990966796875, -1.010284423828125, -0.7035675048828125, 44.57994079589844, 52.488037109375, 29.09686279296875, 33.626220703125, 33.370513916015625, 46.745574951171875, 47.33184814453125, 60.429443359375, 29.03510284423828, -36.270381927490234, 97.18634033203125, -28.4210205078125, 1.1396484375, 29.203018188476562, -1.9771194458007812, -10.268379211425781, 20.4737548828125, -8.662443161010742, 1.2401351928710938, 12.355789184570312, -7.30889892578125, 13.949966430664062, -6.3279876708984375, -6.618478775024414, 23.644973754882812, 5.92559814453125, -20.1630859375, -8.87701416015625, 0.5649032592773438, 59.761474609375, 14.838821411132812, -20.839263916015625, 18.045013427734375, 66.88323974609375, -19.9541015625, 21.6962890625, -7.3486328125, -1.584564208984375, 37.662841796875, 6.159605026245117, -9.865432739257812, -3.903881072998047, 9.476654052734375, 1.1443252563476562, 63.333038330078125, 36.658836364746094, 18.752960205078125, 61.27253723144531, 30.500370025634766, 54.281768798828125, -33.106231689453125, 1.1397342681884766, -5.565826416015625, -1.761322021484375, 5.6584625244140625, 28.711761474609375, 30.70599365234375, 65.4029541015625, 8.852367401123047, 4.0199737548828125, -39.390106201171875, -9.909027099609375, 63.980224609375, 36.14287567138672, 21.01850128173828, -21.96209716796875, 6.01287841796875, 24.035011291503906, -12.9652099609375, 18.375946044921875, 46.3958740234375, 25.323394775390625, 12.90557861328125, -10.06658935546875, 61.8626708984375, 4.309165954589844, 14.226612091064453, -6.068634033203125, -5.267845153808594, 40.76442337036133, 14.45916748046875, -40.228759765625, 23.156005859375, -2.5657119750976562, 10.498367309570312, 24.002059936523438, 3.00537109375, 38.593963623046875, 56.944000244140625, 43.785186767578125, 51.37774658203125, 45.51329803466797, 36.3948974609375, 5.788974761962891, 21.145523071289062, -10.3807373046875, 36.16040802001953], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000152.npy"}
|
|
{"epoch": 0.3183246073298429, "step": 153, "batch_size": 128, "mean": 23.34060287475586, "std": 25.866191864013672, "min": -50.655120849609375, "p10": -2.6461380004882797, "median": 19.632484436035156, "p90": 59.896664428710935, "max": 88.61679077148438, "pos_frac": 0.84375, "sample": [14.395660400390625, 0.17101287841796875, 71.61419677734375, 12.3978271484375, 58.335693359375, 45.003807067871094, 41.956817626953125, 1.3251495361328125, 88.61679077148438, 8.665130615234375, 45.738677978515625, 3.4272499084472656, -7.429229736328125, 24.217132568359375, 33.797607421875, 4.758552551269531, -16.7154541015625, 3.5730438232421875, 24.873382568359375, 65.32283020019531, 58.68048095703125, 68.62069702148438, 33.4293327331543, 2.21453857421875, 11.000030517578125, -3.6321258544921875, 36.63018798828125, -0.6951065063476562, 1.2709274291992188, 2.376678466796875, -14.065460205078125, 3.3371047973632812, 46.5513916015625, 3.9991455078125, 11.866058349609375, 14.535476684570312, 50.42242431640625, 41.374359130859375, 26.9129695892334, 0.0, 59.2481689453125, 35.41169738769531, 53.54541015625, 24.627273559570312, 47.00469970703125, 36.965057373046875, 42.82818603515625, -2.22357177734375, 17.60040283203125, 22.258872985839844, 18.061248779296875, 49.215087890625, -50.655120849609375, 4.9628143310546875, -4.0308990478515625, 32.08673095703125, 47.119384765625, 35.281585693359375, -0.7332000732421875, 64.60580444335938, 17.94061279296875, 4.3900604248046875, 19.816986083984375, 67.6939697265625, -10.548614501953125, 48.0181884765625, 10.3905029296875, 17.309234619140625, 19.447982788085938, 21.50860595703125, -11.075469970703125, 25.217681884765625, 7.2217254638671875, 36.27520751953125, 1.385833740234375, -1.152862548828125, 9.173004150390625, 28.00408935546875, 17.231704711914062, 31.60455322265625, 34.44415283203125, -10.035499572753906, 44.4044189453125, 67.984619140625, 23.772247314453125, 22.722885131835938, 4.720947265625, 14.707275390625, 38.149635314941406, -1.26202392578125, 2.6178741455078125, 19.336029052734375, 63.98200988769531, 43.937744140625, 52.76823425292969, 10.32864761352539, 6.376520156860352, 15.740325927734375, 10.426513671875, 14.089569091796875, 47.4697265625, 45.11076354980469, 38.49638366699219, 7.431327819824219, 4.3212432861328125, -1.7998733520507812, 18.475341796875, 75.9261474609375, 28.42839813232422, 69.24114990234375, 61.811279296875, 21.839004516601562, 5.435546875, 61.409820556640625, -18.873178482055664, 26.828033447265625, 10.718551635742188, 58.91448974609375, 33.77168273925781, 7.17144775390625, -19.7322998046875, 68.24411010742188, 52.90777587890625, 11.697708129882812, -46.642234802246094, -35.134307861328125, 32.21923828125, 30.79022216796875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000153.npy"}
|
|
{"epoch": 0.3204188481675393, "step": 154, "batch_size": 128, "mean": 16.78602409362793, "std": 29.917850494384766, "min": -73.49127197265625, "p10": -19.50521545410156, "median": 17.054046630859375, "p90": 57.10330505371093, "max": 82.31365966796875, "pos_frac": 0.734375, "sample": [24.326187133789062, 47.4136962890625, -25.094482421875, 59.4970703125, 31.595550537109375, 7.529106140136719, -5.143718719482422, 56.22306823730469, 5.808929443359375, 17.051727294921875, -46.84455871582031, 37.08006286621094, 13.683403015136719, 12.487091064453125, 29.339599609375, -73.49127197265625, 34.819732666015625, -24.102630615234375, 28.1708984375, 2.678436279296875, 11.635025024414062, 1.2833480834960938, 6.750213623046875, -18.374420166015625, 1.1395416259765625, 1.8336410522460938, 33.764404296875, 7.3048858642578125, 4.36358642578125, 70.3992919921875, 13.21441650390625, -6.75782585144043, 56.920684814453125, 60.511688232421875, 3.79705810546875, 48.550907135009766, 31.5546875, 64.0499267578125, 35.39996337890625, 61.2333984375, 4.207122802734375, 57.5294189453125, 61.871337890625, 28.73383331298828, -51.9307861328125, 18.295928955078125, 75.75970458984375, -1.0209197998046875, -6.162139892578125, -24.21246337890625, -3.529022216796875, -2.2770462036132812, 16.913345336914062, -23.32080078125, 26.67645263671875, 39.70013427734375, 28.644561767578125, -4.339813232421875, 3.304931640625, -22.14373779296875, -2.3104286193847656, 18.199676513671875, 44.27928161621094, 30.63678741455078, 26.90777587890625, -50.813629150390625, -17.739728927612305, 0.185150146484375, -3.4775924682617188, -11.76470947265625, -3.16961669921875, -43.274993896484375, 50.15000915527344, 65.42153930664062, -7.504234313964844, -9.44195556640625, 10.621681213378906, -6.6102294921875, 29.933837890625, 8.201644897460938, -30.489913940429688, 10.58486557006836, -1.9668350219726562, 27.148696899414062, 32.397796630859375, 10.86859130859375, 48.5557861328125, 28.057525634765625, -5.41009521484375, 32.769287109375, 16.88739013671875, 21.750091552734375, 56.460540771484375, 47.533485412597656, 35.86883544921875, -11.278778076171875, 19.447540283203125, 9.784210205078125, 82.31365966796875, 13.442398071289062, 21.57598876953125, 47.524566650390625, -18.277008056640625, -60.762542724609375, 39.724998474121094, 24.441299438476562, 33.72819519042969, 30.556793212890625, 17.056365966796875, 20.247146606445312, 63.05744934082031, 33.68730163574219, 31.129013061523438, 56.03338623046875, 0.22796630859375, 4.350017547607422, 75.51690673828125, 18.47356414794922, 16.440093994140625, 21.906280517578125, -39.8929443359375, -12.416999816894531, 55.959503173828125, 27.063507080078125, 3.8590774536132812, 22.10650634765625, 36.21354675292969, 61.623313903808594], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000154.npy"}
|
|
{"epoch": 0.3225130890052356, "step": 155, "batch_size": 128, "mean": 12.270386695861816, "std": 27.45819854736328, "min": -56.470703125, "p10": -21.230739593505856, "median": 9.21479606628418, "p90": 47.29134368896484, "max": 80.895751953125, "pos_frac": 0.671875, "sample": [64.8524169921875, 23.585983276367188, 12.493377685546875, -6.495147705078125, 32.500030517578125, -0.3219127655029297, 10.779769897460938, -4.26324462890625, 6.10455322265625, 46.608428955078125, 42.0418586730957, 4.140596389770508, 29.57098388671875, -22.9490966796875, 23.75408935546875, 28.2647705078125, -17.521820068359375, 69.522216796875, 80.895751953125, -6.4219207763671875, 5.584259033203125, 26.448684692382812, 45.172027587890625, 79.1146240234375, 3.8784942626953125, 40.4295654296875, -40.63104248046875, 0.0, -29.344955444335938, -7.891265869140625, -19.102500915527344, 55.41021728515625, 26.367324829101562, 43.692466735839844, 4.46087646484375, -14.584228515625, -20.494300842285156, 37.651329040527344, -7.14874267578125, -0.498992919921875, 8.8250732421875, 30.098159790039062, 15.9254150390625, 0.8844223022460938, 36.67327880859375, 53.213104248046875, 30.34588623046875, 11.79498291015625, 8.057952880859375, 33.670745849609375, -8.134284973144531, -50.732177734375, 7.8568115234375, 7.413116455078125, 2.685396194458008, -35.470245361328125, -0.65997314453125, 37.1658935546875, 68.3392333984375, 48.88481140136719, 3.4542694091796875, 4.160942077636719, 18.232818603515625, -8.708755493164062, -39.49492645263672, -6.244211196899414, 43.229095458984375, 55.911529541015625, 15.739185333251953, 16.463943481445312, -2.3589630126953125, -28.33515167236328, 38.364288330078125, 42.041748046875, -12.903472900390625, 31.05609130859375, 8.10107421875, -6.182014465332031, 34.033843994140625, -26.552825927734375, 32.57330322265625, 50.77458190917969, -23.063034057617188, -14.523574829101562, 8.324798583984375, 7.993011474609375, 34.815704345703125, -40.3858642578125, 2.079742431640625, 37.972320556640625, 17.009521484375, 25.857467651367188, 21.68780517578125, -1.3326416015625, -37.737548828125, 13.453826904296875, 7.92138671875, 60.97303771972656, 11.367828369140625, -14.044563293457031, -16.0704345703125, -12.970252990722656, -4.356941223144531, 9.60451889038086, 19.198333740234375, 38.551910400390625, 4.658515930175781, -5.404563903808594, -56.470703125, 2.3373260498046875, -39.13775634765625, 11.954780578613281, 36.09581756591797, -9.797149658203125, 51.001220703125, 27.1026611328125, 35.42932891845703, 24.543869018554688, 14.690399169921875, 57.6739501953125, -19.3115234375, 0.5575408935546875, 1.147430419921875, 13.43438720703125, 18.7210693359375, 20.512054443359375, 13.200485229492188, -0.50567626953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000155.npy"}
|
|
{"epoch": 0.32460732984293195, "step": 156, "batch_size": 128, "mean": 20.725866317749023, "std": 31.423086166381836, "min": -73.52313232421875, "p10": -22.608444213867188, "median": 23.61650848388672, "p90": 57.270941162109374, "max": 90.72735595703125, "pos_frac": 0.78125, "sample": [36.21038818359375, 40.329872131347656, 27.518722534179688, 21.63604736328125, -1.953857421875, 3.4144973754882812, 43.59431457519531, 44.840423583984375, 28.04852294921875, 26.441587448120117, 41.55799865722656, 26.988189697265625, 48.8057861328125, 33.95391845703125, 2.263671875, -22.453125, -5.776527404785156, 33.48974609375, 7.158355712890625, 40.201812744140625, 2.358551025390625, 45.0185546875, -13.816650390625, -26.330108642578125, 27.8662109375, 10.676559448242188, -3.1570892333984375, 19.216697692871094, 0.9293479919433594, 41.58685302734375, 68.828125, -5.3678436279296875, 69.593505859375, -6.06170654296875, 9.554544448852539, -22.970855712890625, -36.44987487792969, 5.200901031494141, 46.978546142578125, 2.7506561279296875, 28.660476684570312, 56.063323974609375, 27.48681640625, 9.686614990234375, 6.5123748779296875, 3.319122314453125, 41.65495300292969, -33.350555419921875, 44.20009231567383, 31.946014404296875, 2.354999542236328, 33.72985076904297, 24.3519287109375, 24.368377685546875, 81.0826416015625, -24.6268310546875, -35.53086853027344, 82.6798095703125, 0.28106689453125, 8.43707275390625, -5.739349365234375, 82.7415771484375, 33.06193542480469, 48.06364440917969, 29.20907211303711, 55.0977783203125, 48.95867919921875, 10.300888061523438, 57.9232177734375, 6.016639709472656, 17.37091064453125, 0.1510162353515625, -4.816616058349609, 64.46893310546875, 34.1724853515625, 32.136962890625, 90.72735595703125, 20.754119873046875, 0.3027534484863281, -3.7037487030029297, 78.65690612792969, 35.89067077636719, 66.42437744140625, 1.4401092529296875, 8.68157958984375, 22.364517211914062, 2.420156478881836, 31.8662109375, 29.277854919433594, -26.48114013671875, 42.270538330078125, 49.34474182128906, 36.55201721191406, 43.449462890625, 26.390869140625, 12.31561279296875, -19.466278076171875, 59.51689147949219, 22.881088256835938, -25.357818603515625, 56.99139404296875, 12.12091064453125, 16.315887451171875, -23.52288818359375, -73.52313232421875, 50.174407958984375, -1.5012664794921875, 77.62451171875, -2.6470794677734375, -36.62886047363281, 6.819305419921875, 31.849761962890625, 84.28668212890625, 10.99322509765625, -72.42715454101562, -34.6176872253418, 2.25750732421875, 44.09132385253906, 51.6885986328125, 52.83050537109375, 0.3913383483886719, 27.419952392578125, 39.57909393310547, 55.1624755859375, -16.602951049804688, 56.06268310546875, -7.441902160644531, 3.5450592041015625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000156.npy"}
|
|
{"epoch": 0.3267015706806283, "step": 157, "batch_size": 128, "mean": 14.312121391296387, "std": 28.17302703857422, "min": -58.40260314941406, "p10": -19.65106201171875, "median": 10.527023315429688, "p90": 49.53560180664062, "max": 94.35104370117188, "pos_frac": 0.703125, "sample": [34.91876220703125, -0.5409660339355469, 22.9698486328125, 61.58607482910156, -1.49560546875, -19.206069946289062, 11.59661865234375, 17.181549072265625, 50.53839111328125, -6.84552001953125, 25.29022216796875, -0.49069976806640625, -41.956878662109375, 27.164321899414062, 1.2515220642089844, 33.2899169921875, 9.827239990234375, 1.1532211303710938, 2.8465614318847656, -13.588088989257812, -17.284408569335938, 61.44122314453125, 27.66253662109375, -9.684555053710938, 46.42959976196289, -3.4883594512939453, -32.38037109375, -0.8458099365234375, 4.8095855712890625, 15.8585205078125, 7.1313934326171875, -6.369659423828125, 9.5916748046875, -48.0321044921875, 33.91126251220703, 46.53746032714844, 20.0806884765625, 30.683670043945312, -42.464752197265625, -58.40260314941406, 77.75225830078125, 16.038528442382812, 32.47673034667969, 20.50042724609375, 78.1190185546875, 35.79725646972656, 11.125335693359375, -7.293708801269531, 20.965240478515625, 0.0, 15.022048950195312, 47.90296936035156, 7.25885009765625, -5.344818115234375, 62.52593994140625, 12.5799560546875, 37.40449523925781, -6.5313720703125, -5.511077880859375, 1.409616470336914, 49.1058349609375, 25.983123779296875, 51.132362365722656, 10.0673828125, 30.113937377929688, -3.2975730895996094, 48.212310791015625, -6.04815673828125, -43.8179931640625, 35.237518310546875, 1.73187255859375, -16.355911254882812, -42.811004638671875, 10.986663818359375, -11.47637939453125, 18.578369140625, -33.14422607421875, 39.3748779296875, 6.47796630859375, 53.688629150390625, 18.67572021484375, 38.583770751953125, 0.9291000366210938, -7.1737518310546875, 3.005340576171875, 9.847671508789062, 6.5435791015625, 14.135578155517578, 24.001636505126953, 0.0, -25.062347412109375, 57.678009033203125, 23.9573974609375, 14.331924438476562, 5.330596923828125, -1.3381500244140625, 27.145143508911133, 52.07274627685547, 94.35104370117188, 47.96772766113281, 9.330169677734375, -20.689376831054688, 30.3974609375, 8.580167770385742, 5.064666748046875, 2.1440505981445312, 32.90129089355469, 33.19765853881836, 22.226318359375, -6.651519775390625, -24.714401245117188, 43.48052978515625, 2.9358062744140625, 37.90887451171875, 31.038787841796875, 31.757781982421875, -4.99407958984375, 0.9823942184448242, -22.9093017578125, 43.831336975097656, -27.827545166015625, 6.9470062255859375, 6.070333480834961, 33.253929138183594, 7.953582763671875, 18.355369567871094, 92.72689819335938, 57.057891845703125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000157.npy"}
|
|
{"epoch": 0.3287958115183246, "step": 158, "batch_size": 128, "mean": 15.748287200927734, "std": 27.516281127929688, "min": -45.544921875, "p10": -22.099305725097654, "median": 15.877670288085938, "p90": 49.98393859863281, "max": 79.80340576171875, "pos_frac": 0.71875, "sample": [4.925054550170898, -2.5822372436523438, 23.3280029296875, 49.516845703125, -32.82366943359375, 3.87139892578125, -1.6698760986328125, 19.8211669921875, -15.758712768554688, 52.246490478515625, -15.2974853515625, 1.814544677734375, -39.152496337890625, -35.049285888671875, 23.821624755859375, 19.02337646484375, 32.12782669067383, 22.633102416992188, 9.240768432617188, 26.794082641601562, 15.6595458984375, 57.485870361328125, 38.717926025390625, 4.5489959716796875, 16.095794677734375, 34.67408752441406, 13.747856140136719, -39.2041015625, 37.4462890625, 24.130462646484375, -10.270217895507812, 32.64390563964844, -14.493072509765625, 50.63482666015625, 16.64739990234375, 7.6526336669921875, 37.420013427734375, -33.858150482177734, 18.366912841796875, -39.092315673828125, 37.37858581542969, 7.7237548828125, 14.142744064331055, -24.176692962646484, 14.61993408203125, -1.9447479248046875, -6.9423675537109375, 16.93524169921875, 36.29937744140625, 21.568843841552734, 60.49774169921875, -4.725282669067383, 1.8089885711669922, 11.967231750488281, 32.71257019042969, 44.693206787109375, -34.73406219482422, -19.13421630859375, 61.35865783691406, 15.365707397460938, 59.177642822265625, 7.75396728515625, -6.29443359375, 9.94140625, 34.31854248046875, 12.715728759765625, -21.5565185546875, 0.0, 29.065841674804688, 14.0404052734375, 65.49661254882812, -6.0711822509765625, -3.82318115234375, 45.319091796875, 23.59869384765625, 49.704986572265625, 4.6729583740234375, -23.140472412109375, -10.566085815429688, -2.219573974609375, 47.70213317871094, 9.4052734375, -44.16514587402344, 14.397628784179688, 30.84461212158203, 35.21856689453125, 34.254638671875, 0.3568134307861328, 41.15699768066406, 36.90008544921875, 41.720794677734375, 29.111419677734375, 11.166435241699219, 17.95526123046875, 64.914306640625, 41.75048828125, 75.72442626953125, -5.707561492919922, -25.27862548828125, -21.653091430664062, 12.454544067382812, 79.80340576171875, -14.87261962890625, 36.025970458984375, 46.378944396972656, 43.04541015625, 50.87901306152344, 29.122528076171875, 34.791229248046875, -1.1299552917480469, 16.7401123046875, 32.004241943359375, 33.965606689453125, 9.904541015625, 28.206649780273438, -14.108352661132812, 50.794097900390625, 3.7536697387695312, -29.793899536132812, -7.4437255859375, 30.2374267578125, 41.468292236328125, -45.544921875, 11.169921875, 77.90620422363281, 18.302230834960938, 20.817981719970703, 1.7918853759765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000158.npy"}
|
|
{"epoch": 0.3308900523560209, "step": 159, "batch_size": 128, "mean": 12.699447631835938, "std": 32.537994384765625, "min": -87.10955810546875, "p10": -27.834120178222655, "median": 10.235313415527344, "p90": 56.067370605468746, "max": 93.14845275878906, "pos_frac": 0.6171875, "sample": [-18.898040771484375, 54.10676574707031, 23.29364013671875, 32.005859375, -3.4689483642578125, -14.42041015625, -37.332275390625, 50.05023193359375, -2.17535400390625, 58.667724609375, -50.44331359863281, 14.237068176269531, -87.10955810546875, -11.622810363769531, 10.538215637207031, 34.445098876953125, 12.387947082519531, -5.4781494140625, -33.56475830078125, 42.59234619140625, 27.58251953125, 1.109283447265625, -11.991806030273438, -24.96709442138672, 22.68756103515625, 54.763671875, 24.635879516601562, 56.830474853515625, 30.88018798828125, -6.3049163818359375, 22.193359375, 2.70111083984375, 38.1231689453125, 15.829032897949219, 14.981468200683594, -17.97943115234375, -2.468505859375, -26.72564697265625, -3.6588096618652344, -31.773834228515625, 44.45649719238281, 16.470809936523438, 26.509933471679688, 39.629249572753906, -9.326995849609375, 20.21014404296875, -27.347366333007812, 3.6660690307617188, -28.969879150390625, -19.57434844970703, 0.9505538940429688, -8.074432373046875, 47.499847412109375, 8.862060546875, 17.32281494140625, 47.21339416503906, 2.614727020263672, 46.51263427734375, -7.256927490234375, -5.296173095703125, 93.14845275878906, 9.305099487304688, -27.01842498779297, 12.614913940429688, 71.76287841796875, 13.385284423828125, 50.104217529296875, 2.45068359375, -9.3836669921875, 12.691024780273438, 5.2152099609375, 34.027099609375, -5.712650299072266, 43.79998779296875, 35.758880615234375, -33.574310302734375, 1.87774658203125, 60.194793701171875, -16.72021484375, 18.3084716796875, -18.99273681640625, -37.119049072265625, 28.232215881347656, 71.21951293945312, 75.89208984375, -10.71450424194336, -13.392082214355469, -31.402999877929688, -8.630630493164062, 55.740325927734375, 26.893356323242188, 16.349594116210938, -19.642913818359375, 21.491851806640625, 10.438766479492188, 32.19482421875, -31.4683837890625, 58.571533203125, -9.458335876464844, 40.02616500854492, 29.565391540527344, -25.165231704711914, -31.812057495117188, -1.3327484130859375, 84.45327758789062, 74.85174560546875, 22.884124755859375, -48.18312072753906, 43.40153503417969, 4.589443206787109, -5.074737548828125, -11.4747314453125, 60.44775390625, -43.087493896484375, 74.0947265625, -7.681915283203125, 47.95855712890625, 33.636077880859375, 49.78987121582031, 44.445953369140625, 10.0318603515625, 7.271854400634766, -2.893218994140625, 67.5494384765625, 9.442276000976562, 7.831298828125, -10.6190185546875, 41.81264877319336], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000159.npy"}
|
|
{"epoch": 0.33298429319371725, "step": 160, "batch_size": 128, "mean": 18.86957550048828, "std": 28.079017639160156, "min": -45.77734375, "p10": -14.066427612304686, "median": 16.7085018157959, "p90": 54.748063659667956, "max": 99.930908203125, "pos_frac": 0.75, "sample": [-7.464729309082031, 43.518280029296875, 0.0, 78.2679443359375, 22.902618408203125, 24.869384765625, 28.224884033203125, 17.7103271484375, 73.41326904296875, -0.5020751953125, 48.43792724609375, 66.31390380859375, 26.42546272277832, 65.62750244140625, 18.109451293945312, -3.29681396484375, 69.447509765625, 24.565521240234375, 8.136245727539062, 17.960281372070312, 7.7752227783203125, 8.402923583984375, 25.70825958251953, 31.917205810546875, 1.53387451171875, -7.195709228515625, -5.9591064453125, 76.67208862304688, 2.400463104248047, -43.86236572265625, 13.369598388671875, 98.5416259765625, -32.660133361816406, 44.924034118652344, 18.27057647705078, 3.624032974243164, 7.684120178222656, 4.260677337646484, 23.744659423828125, 15.706676483154297, 32.78052520751953, 47.29345703125, 11.465576171875, 0.0, 36.720458984375, -11.6617431640625, 32.23511505126953, -23.389694213867188, -14.616485595703125, 0.98297119140625, 19.91363525390625, 34.19178771972656, 99.930908203125, 67.07025146484375, -16.211532592773438, 14.7945556640625, 24.96038818359375, -14.74652099609375, 9.407554626464844, -11.320755004882812, -10.08282470703125, 50.43434143066406, 7.069244384765625, -28.232666015625, -10.6680908203125, 53.227081298828125, 52.98236083984375, 53.59483337402344, 4.384805679321289, 12.71234130859375, 28.548179626464844, 2.1584014892578125, -1.0182819366455078, 12.3824462890625, 25.596969604492188, 31.588165283203125, 64.958984375, 46.429718017578125, 33.735595703125, 18.69525146484375, -24.429473876953125, -6.294593811035156, 39.94800567626953, -30.59478759765625, -13.8306884765625, 59.87431335449219, 28.57928466796875, 5.113525390625, 40.28343200683594, -12.973419189453125, 7.442596435546875, 4.43487548828125, 1.1517410278320312, -25.796478271484375, 28.684906005859375, -45.77734375, -1.5286483764648438, 27.250946044921875, 57.438934326171875, 1.17254638671875, 12.54660415649414, 15.136093139648438, -15.436208724975586, 24.707183837890625, 18.214569091796875, 10.69573974609375, 9.680624008178711, -6.9921112060546875, 40.58575439453125, 27.603134155273438, 51.4676513671875, -0.931121826171875, 15.469879150390625, 40.67961120605469, -5.70733642578125, 25.133407592773438, 49.550567626953125, 11.687957763671875, 32.074127197265625, 50.1746826171875, 57.45159912109375, 28.784622192382812, -31.51068115234375, 32.2528076171875, 19.41619873046875, 49.171295166015625, 3.025970458984375, 4.400478363037109], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000160.npy"}
|
|
{"epoch": 0.33507853403141363, "step": 161, "batch_size": 128, "mean": 19.651905059814453, "std": 30.320959091186523, "min": -64.33071899414062, "p10": -14.235157775878905, "median": 17.622940063476562, "p90": 64.5651641845703, "max": 76.92080688476562, "pos_frac": 0.71875, "sample": [-3.6389694213867188, 1.6587104797363281, -0.997772216796875, -61.7220458984375, 0.3663463592529297, 21.785568237304688, 3.254730224609375, 6.17913818359375, 6.23291015625, 3.90911865234375, 4.178791046142578, 73.99932861328125, 28.344940185546875, -29.797042846679688, 3.3408203125, 2.5560836791992188, 71.28036499023438, 20.42034912109375, 36.80925369262695, -3.7171630859375, 41.668121337890625, 15.5147705078125, 53.03192138671875, 75.7974853515625, -20.652557373046875, 4.00213623046875, 5.937885284423828, 71.4132080078125, 7.606910705566406, 29.101734161376953, 5.33013916015625, 25.929906845092773, -4.3691864013671875, -13.992141723632812, 31.801300048828125, 24.214370727539062, 33.95945739746094, 48.25341796875, -0.9693260192871094, 10.130607604980469, -4.36663818359375, -38.48828125, -27.1727294921875, -8.455528259277344, 12.136894226074219, 43.768280029296875, -14.122222900390625, -5.071563720703125, 0.0, 37.563316345214844, 17.2615966796875, 33.32989501953125, 6.935760498046875, 0.10020828247070312, 12.114013671875, -1.29327392578125, 48.802825927734375, 51.68354034423828, 38.889686584472656, 26.12262725830078, 51.99554443359375, 60.1937255859375, -18.98614501953125, 18.362396240234375, -11.820465087890625, 63.732818603515625, 54.25531005859375, 67.48419189453125, -16.779922485351562, 13.707687377929688, 71.5445556640625, 25.00246810913086, 35.619537353515625, 64.62200927734375, -31.14739990234375, 67.39321899414062, 31.1270751953125, -1.71868896484375, 1.513519287109375, 36.14862060546875, -13.535087585449219, 0.466461181640625, 68.30560302734375, 64.03652954101562, 26.155776977539062, 67.75149536132812, 1.65582275390625, 48.9364013671875, 76.92080688476562, -1.143341064453125, -2.39007568359375, 25.176727294921875, -15.4715576171875, 37.638336181640625, -9.454132080078125, 25.81658935546875, 65.748046875, -5.042137145996094, 35.2523193359375, -7.8094329833984375, 23.862045288085938, 56.72630310058594, 20.2972412109375, -14.8585205078125, 13.744346618652344, 39.68328857421875, -14.498672485351562, 26.304931640625, 37.14405059814453, 46.93218994140625, 53.8978271484375, 63.3707275390625, -10.873237609863281, 30.246429443359375, 25.29902458190918, -49.354736328125, 64.54080200195312, 9.712997436523438, 12.274162292480469, 0.0, -3.2012939453125, 65.60723876953125, 55.48020935058594, -64.33071899414062, 28.153076171875, 17.984283447265625, 2.3296661376953125, 49.81304931640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000161.npy"}
|
|
{"epoch": 0.33717277486910996, "step": 162, "batch_size": 128, "mean": 17.607093811035156, "std": 29.177364349365234, "min": -69.807373046875, "p10": -19.571892547607423, "median": 16.930435180664062, "p90": 57.203179931640626, "max": 85.76788330078125, "pos_frac": 0.765625, "sample": [-6.348480224609375, 7.8593292236328125, 60.890869140625, 38.60883331298828, 2.0500335693359375, 16.748321533203125, -20.312347412109375, 27.629714965820312, 7.3350372314453125, 25.575088500976562, 47.1986083984375, 72.55914306640625, 11.005538940429688, 21.013633728027344, -7.025146484375, -10.604598999023438, 22.91400146484375, 28.2728271484375, 1.7189445495605469, 4.387996673583984, -29.592254638671875, 34.26129150390625, -23.483009338378906, 28.615036010742188, 17.2984619140625, 27.49658966064453, 57.3173828125, 10.8658447265625, -37.90936279296875, 6.4472503662109375, -0.14745330810546875, 20.4635009765625, 34.2210693359375, -26.06841278076172, 64.2633056640625, 5.98895263671875, 17.112548828125, -58.928253173828125, -10.581947326660156, 34.40525817871094, 2.2355422973632812, 51.72259521484375, 2.997295379638672, 60.30908203125, 58.7373046875, -69.807373046875, 55.83148193359375, 27.40771484375, 27.01885986328125, 57.0615234375, 1.348550796508789, 35.044708251953125, 41.104278564453125, 4.43145751953125, 29.442733764648438, -3.0708541870117188, -16.479522705078125, 39.13154602050781, 67.9454345703125, 25.655181884765625, 65.7191162109375, 82.48440551757812, -1.64630126953125, 0.50439453125, 36.297119140625, 50.73854064941406, 53.6976318359375, 62.34602355957031, -17.80865478515625, 44.68702697753906, 0.423675537109375, 8.28415298461914, -23.066436767578125, 17.698699951171875, 85.76788330078125, -20.07367706298828, -25.998870849609375, 3.38482666015625, 14.504638671875, 48.85662841796875, -6.279386520385742, 10.47265625, 39.3437614440918, 26.096343994140625, 57.0352783203125, 51.535003662109375, 6.487600326538086, -1.18011474609375, 25.969375610351562, 21.797271728515625, 8.491622924804688, 11.141021728515625, 8.356405258178711, -54.34339904785156, 1.600494384765625, 27.023818969726562, 49.889892578125, 30.89739990234375, 17.335205078125, 57.15423583984375, 39.038330078125, -37.0498046875, 44.685791015625, 7.293304443359375, 0.43756103515625, 60.3712158203125, -13.681182861328125, 5.0735321044921875, -3.0103073120117188, 45.13254165649414, -29.734756469726562, 4.0042266845703125, -10.457260131835938, 36.4095458984375, 26.91522216796875, 34.22959899902344, 12.021163940429688, -3.6409530639648438, 25.207061767578125, 19.08184814453125, 7.876092910766602, -19.356842041015625, -0.20319366455078125, 4.0523681640625, 6.625587463378906, 57.3541259765625, 26.76824951171875, 11.04888916015625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000162.npy"}
|
|
{"epoch": 0.3392670157068063, "step": 163, "batch_size": 128, "mean": 20.232545852661133, "std": 25.238651275634766, "min": -40.331695556640625, "p10": -11.860713958740233, "median": 20.222915649414062, "p90": 54.199903869628905, "max": 78.55551147460938, "pos_frac": 0.75, "sample": [20.148468017578125, -0.1794281005859375, 38.190635681152344, -12.80401611328125, 23.012847900390625, -12.793098449707031, 18.719619750976562, 18.577926635742188, 31.342987060546875, 2.7109832763671875, -0.0749359130859375, 58.50029754638672, 39.480926513671875, 14.10205078125, 65.978759765625, 13.12628173828125, 11.443344116210938, -11.46112060546875, 8.075653076171875, -7.325340270996094, -13.312225341796875, 34.80172348022461, 23.9576416015625, -2.0143508911132812, 14.0980224609375, 45.65379333496094, -5.35285758972168, 12.60205078125, 44.79156494140625, -2.737518310546875, -5.277557373046875, 53.78236389160156, -0.073333740234375, 41.1614990234375, 28.169281005859375, 21.812744140625, 10.38494873046875, 42.739715576171875, -8.6207275390625, -5.9935302734375, 36.178497314453125, 29.998626708984375, -27.42578125, 8.797988891601562, 11.79608154296875, 20.873687744140625, -39.094635009765625, 56.095306396484375, 42.38786315917969, -28.157379150390625, 32.531585693359375, 6.93121337890625, 36.235198974609375, 33.04142761230469, -1.445037841796875, 51.02647399902344, 18.6484375, 55.49560546875, 52.36433410644531, 35.47149658203125, 2.35479736328125, -3.57452392578125, -8.683349609375, 29.913009643554688, -8.442138671875, -10.907928466796875, 9.856956481933594, 44.350563049316406, 6.89984130859375, 35.081756591796875, -40.331695556640625, 56.89129638671875, 23.606895446777344, 20.103240966796875, 12.8123779296875, -16.693878173828125, 68.70138549804688, -14.131927490234375, 57.421783447265625, 58.4564208984375, 30.3974609375, 41.4197998046875, 20.29736328125, 17.86208724975586, 36.08342742919922, 36.203208923339844, 31.677474975585938, 0.206390380859375, 55.174163818359375, 44.86293029785156, 4.175748825073242, 6.0648193359375, 20.45636749267578, 78.55551147460938, 32.865142822265625, -23.655181884765625, -25.001426696777344, 24.459367752075195, 71.464599609375, 28.589218139648438, 48.861602783203125, 44.606689453125, -16.161346435546875, 45.6580810546875, -2.7583770751953125, 15.34185791015625, 55.476165771484375, 16.717605590820312, 23.047332763671875, 30.28777313232422, 52.171722412109375, 31.551376342773438, 18.8702392578125, 1.0057735443115234, 16.019851684570312, 14.365142822265625, 28.639739990234375, 51.57295227050781, 29.16180419921875, 60.698158264160156, -3.5340309143066406, 2.90618896484375, -36.870391845703125, 50.278106689453125, 42.39152526855469, 25.020492553710938, 9.030553817749023, -1.5311965942382812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000163.npy"}
|
|
{"epoch": 0.3413612565445026, "step": 164, "batch_size": 128, "mean": 17.71393585205078, "std": 27.672412872314453, "min": -45.992340087890625, "p10": -10.526345825195312, "median": 15.641372680664062, "p90": 55.640890502929686, "max": 93.3096923828125, "pos_frac": 0.7265625, "sample": [-9.3446044921875, 17.201683044433594, -2.395570755004883, 49.81524658203125, 1.0921897888183594, -0.1930999755859375, -4.7451629638671875, -9.757431030273438, 23.740768432617188, 31.8604736328125, 29.688018798828125, -12.092453002929688, 36.49101257324219, 39.934539794921875, 20.0855712890625, -45.76605224609375, -1.53997802734375, 49.002166748046875, 11.706182479858398, 61.7554931640625, -0.66864013671875, -38.53759765625, 44.14183044433594, 14.522735595703125, 10.853485107421875, -22.22100830078125, 24.404327392578125, 2.906951904296875, -6.7870330810546875, 5.950897216796875, 31.264205932617188, 0.20703125, 3.504058837890625, 53.06941223144531, -2.4900970458984375, 45.46881103515625, -40.60777282714844, 3.165771484375, -0.33514404296875, -36.644744873046875, -7.4275665283203125, 64.04354858398438, 39.12152099609375, 29.04248046875, 56.684539794921875, 5.5022430419921875, 49.5137939453125, 0.67303466796875, 0.59716796875, 4.232757568359375, -25.543785095214844, 12.803695678710938, 18.481597900390625, 26.5688533782959, -12.067230224609375, 57.692176818847656, -1.625457763671875, 38.26261901855469, 28.948135375976562, 27.978057861328125, 1.734771728515625, 9.1162109375, 56.005950927734375, 57.40545654296875, -9.667800903320312, 59.3536376953125, 21.83031463623047, -2.5681610107421875, 42.50286865234375, 13.29925537109375, 75.105712890625, 28.54730987548828, 62.634521484375, -1.032562255859375, -7.896701812744141, 1.0460548400878906, 18.142929077148438, 6.015476226806641, -42.949737548828125, 18.69048309326172, 27.74176025390625, 48.168609619140625, -10.1878662109375, 2.5110321044921875, -1.1121978759765625, 33.200531005859375, 44.784210205078125, 57.607666015625, 37.66455078125, 25.01568603515625, -3.2276039123535156, 42.23682403564453, 2.0032806396484375, 43.041778564453125, 10.02276611328125, 47.241668701171875, 44.44207763671875, 55.48443603515625, -11.316131591796875, 0.6772994995117188, -31.85430908203125, 24.46405029296875, 1.2608795166015625, 26.009124755859375, 7.377143859863281, -4.308135986328125, 50.79730224609375, -9.10986328125, 38.85152816772461, 10.116363525390625, 28.389678955078125, 42.6583251953125, 6.823310852050781, 16.760009765625, -45.992340087890625, -19.389678955078125, 8.20379638671875, -6.209014892578125, 93.3096923828125, 17.638702392578125, 26.914169311523438, 41.129547119140625, 13.704437255859375, 31.0958251953125, 39.09783935546875, 56.420379638671875, 89.51983642578125, 19.2000789642334], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000164.npy"}
|
|
{"epoch": 0.34345549738219894, "step": 165, "batch_size": 128, "mean": 14.55510139465332, "std": 26.446836471557617, "min": -55.762603759765625, "p10": -14.628693389892577, "median": 8.674407958984375, "p90": 49.981991577148435, "max": 80.52175903320312, "pos_frac": 0.65625, "sample": [4.39959716796875, 48.440093994140625, -4.611427307128906, 38.0975341796875, -17.8843994140625, 3.0648880004882812, 28.571609497070312, 5.704559326171875, 40.88372802734375, 42.66973876953125, 63.1275634765625, 3.578510284423828, -52.642913818359375, 37.33843994140625, 53.850341796875, 14.18743896484375, -5.77789306640625, 61.70843505859375, -7.11236572265625, 43.156585693359375, 28.768325805664062, -45.4915771484375, 49.880706787109375, 35.75675964355469, 6.3953704833984375, -22.753002166748047, -22.1964111328125, -8.625595092773438, 18.687591552734375, 32.313018798828125, 5.960968017578125, -8.7281494140625, -1.39788818359375, 0.597442626953125, 1.1068572998046875, 11.741539001464844, 26.635147094726562, -4.4838104248046875, -8.17875862121582, -8.020034790039062, 17.58514404296875, 14.749336242675781, 49.485931396484375, -16.71551513671875, 15.564697265625, 50.21832275390625, -0.8524227142333984, -0.8106842041015625, -15.64434814453125, 8.182083129882812, 78.4407958984375, 2.009765625, 3.6758346557617188, -0.3066864013671875, 55.2655029296875, 21.391677856445312, -9.522003173828125, 4.875171661376953, -5.510929107666016, 14.521087646484375, -0.536590576171875, 57.6661376953125, -6.499717712402344, 43.44439697265625, 23.610137939453125, -13.743194580078125, 22.37469482421875, -1.4066925048828125, -3.6131591796875, -6.3436737060546875, -0.515380859375, -2.4533233642578125, 10.1824951171875, 18.808761596679688, 23.264572143554688, 4.686553955078125, 34.88139343261719, -23.68511962890625, 28.8333740234375, 14.85546875, -22.857925415039062, 24.36809539794922, 64.40725708007812, -15.857177734375, 7.972160339355469, 35.08514404296875, 0.5828933715820312, 58.5135498046875, 47.51885986328125, 18.8978271484375, 11.256256103515625, -0.3590240478515625, 35.208221435546875, 39.52645492553711, 47.52398681640625, -14.193412780761719, -8.703865051269531, 21.8978271484375, 38.33735656738281, 5.534019470214844, 80.52175903320312, -2.6924667358398438, 32.191131591796875, 37.13841247558594, 58.04499816894531, -28.84282684326172, -9.82540512084961, 15.052764892578125, 48.50849914550781, 3.29010009765625, 0.8960914611816406, 5.0433197021484375, 53.571044921875, 30.7091064453125, 47.18341064453125, 41.346435546875, 4.000244140625, 11.73468017578125, 9.166732788085938, 42.84992980957031, -1.8825531005859375, 37.109954833984375, -55.762603759765625, -26.527130126953125, -2.8154983520507812, 52.90765380859375, -8.177413940429688, -5.4782867431640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000165.npy"}
|
|
{"epoch": 0.34554973821989526, "step": 166, "batch_size": 128, "mean": 13.634095191955566, "std": 32.04103469848633, "min": -69.2960205078125, "p10": -22.405337524414062, "median": 10.125251770019531, "p90": 56.89071350097655, "max": 96.87625122070312, "pos_frac": 0.65625, "sample": [14.075347900390625, -1.049560546875, -23.2794189453125, 12.69671630859375, -1.507080078125, 11.323036193847656, 60.015869140625, 52.17547607421875, -2.6662120819091797, 2.940927505493164, 8.988761901855469, -7.258811950683594, 96.87625122070312, -14.177711486816406, 2.58074951171875, -7.68865966796875, 21.159591674804688, 26.715301513671875, 3.8268508911132812, 1.8750762939453125, -1.4040756225585938, -30.9224853515625, 17.86785888671875, 58.63362121582031, -54.39959716796875, -22.030731201171875, 1.6650104522705078, 28.450439453125, -49.69255065917969, -14.751571655273438, -7.550018310546875, 15.26898193359375, 26.519256591796875, 64.90771484375, 55.72679138183594, 72.411376953125, 2.4723777770996094, -7.7204742431640625, -15.421623229980469, -66.5804443359375, 6.212900161743164, -26.366683959960938, 5.127586364746094, -2.2357177734375, -9.45538330078125, 69.00064086914062, 20.144386291503906, 18.622238159179688, -16.925323486328125, 33.74554443359375, -0.985931396484375, 80.27633666992188, -36.8719482421875, 60.95281982421875, 85.66534423828125, -46.71144104003906, -28.58935546875, 85.42340087890625, 27.210968017578125, 36.928245544433594, -20.022171020507812, -9.122711181640625, -21.51763916015625, 32.0943603515625, 44.946441650390625, -24.447509765625, 48.38250732421875, 33.50927734375, 5.54364013671875, -11.393096923828125, 20.49285888671875, 21.8719482421875, 67.00253295898438, -8.419305801391602, -10.124160766601562, 18.58208465576172, 13.522941589355469, -0.1549358367919922, 13.418380737304688, 8.740997314453125, -4.250011444091797, 39.6668701171875, 17.238265991210938, 9.057052612304688, 11.623199462890625, 68.6654052734375, 1.68658447265625, 34.39568328857422, 16.725494384765625, -32.92240905761719, 6.77459716796875, -6.9359283447265625, 14.393791198730469, 45.432891845703125, 46.13380432128906, 25.688552856445312, 16.373138427734375, 45.98168182373047, 25.762229919433594, 6.066650390625, 20.30548095703125, 0.5150966644287109, -3.4578170776367188, 41.66725158691406, 1.9427356719970703, 30.248687744140625, 47.774383544921875, 42.16925048828125, -10.466514587402344, 56.14375305175781, -37.184478759765625, -69.2960205078125, 9.8946533203125, 42.83103942871094, 52.92376708984375, -19.728660583496094, 11.187339782714844, -1.54534912109375, 19.07177734375, -7.113983154296875, 54.6363525390625, -0.4829139709472656, 96.2255859375, 12.910625457763672, 2.562591552734375, 28.59185791015625, 10.355850219726562, 9.776809692382812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000166.npy"}
|
|
{"epoch": 0.34764397905759165, "step": 167, "batch_size": 128, "mean": 15.789811134338379, "std": 28.645395278930664, "min": -53.852569580078125, "p10": -15.618978881835936, "median": 11.731101989746094, "p90": 55.63718719482421, "max": 88.31878662109375, "pos_frac": 0.7109375, "sample": [42.27227020263672, 3.8495635986328125, 13.28814697265625, 39.798614501953125, 9.632415771484375, 44.18548583984375, 38.312255859375, 65.31155395507812, 31.03424072265625, 14.6339111328125, 51.53173828125, -40.931732177734375, 59.266632080078125, 1.4260902404785156, 10.734832763671875, 68.35931396484375, 14.96539306640625, 62.398712158203125, -0.24420166015625, 13.59686279296875, 2.2327880859375, 36.33099365234375, -47.624359130859375, -53.852569580078125, -2.6802139282226562, 0.7156219482421875, -21.774612426757812, 48.38996887207031, 27.420867919921875, 48.0042724609375, 25.818572998046875, 54.76075744628906, 12.115798950195312, 57.77044677734375, 31.987838745117188, 6.811492919921875, 14.639122009277344, -30.156494140625, 17.499664306640625, -7.731330871582031, 61.05133056640625, -3.0052490234375, 12.77838134765625, 2.515594482421875, 0.0104827880859375, -53.660400390625, 0.0, 2.20556640625, 2.14105224609375, 28.672607421875, 11.234909057617188, 66.6734619140625, -25.07562255859375, 57.68218994140625, -17.110366821289062, 14.2298583984375, 40.113990783691406, 9.123046875, 25.394287109375, 0.5941162109375, -31.12957763671875, 2.341552734375, 73.29022216796875, 0.45693397521972656, 23.568450927734375, 11.346405029296875, 16.714126586914062, -13.218795776367188, 33.74726867675781, 4.032356262207031, -0.8815460205078125, 1.1203556060791016, 34.275550842285156, -32.63313293457031, 42.146766662597656, 30.386550903320312, 50.03028869628906, -1.29931640625, 23.28948974609375, 2.1899566650390625, -1.0197906494140625, 44.54032897949219, 31.822113037109375, -23.837539672851562, 50.4090576171875, -5.993415832519531, 16.59356689453125, -7.429891586303711, 35.47174072265625, 33.72483825683594, 32.309967041015625, 5.138053894042969, 43.557220458984375, 18.78155517578125, 32.833709716796875, 72.66680908203125, -5.121120452880859, -14.404953002929688, -14.979812622070312, -3.6282196044921875, -5.649467468261719, 7.521745681762695, -7.0455474853515625, 62.64837646484375, -9.808349609375, 6.73980712890625, 88.31878662109375, 18.28118896484375, 39.554840087890625, -11.19818115234375, 6.96295166015625, 34.592529296875, -28.882492065429688, -4.312034606933594, -1.9725494384765625, 4.728752136230469, 33.06500244140625, -2.69403076171875, -4.95477294921875, 45.68773651123047, 37.19383239746094, 19.57379913330078, 10.14764404296875, -2.955841064453125, -49.37664794921875, 47.17926025390625, 7.481109619140625, 69.39030456542969], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000167.npy"}
|
|
{"epoch": 0.34973821989528797, "step": 168, "batch_size": 128, "mean": 20.818775177001953, "std": 29.020523071289062, "min": -63.350311279296875, "p10": -10.931014251708984, "median": 17.721786499023438, "p90": 56.95883636474609, "max": 88.55508422851562, "pos_frac": 0.75, "sample": [26.730758666992188, 16.28863525390625, 65.89065551757812, 55.63018798828125, 38.9710693359375, 21.37842559814453, 41.12091064453125, 18.57543182373047, 46.500244140625, -6.577301025390625, 48.6712646484375, -11.09124755859375, 0.808197021484375, 9.661285400390625, -0.2437591552734375, 3.939605712890625, 23.562042236328125, 29.985076904296875, -6.424594879150391, 45.97431182861328, 15.560447692871094, 54.84814453125, 17.677520751953125, 23.078033447265625, 53.06024169921875, -1.1982345581054688, 20.432373046875, 45.30094528198242, -3.2783126831054688, 2.7076797485351562, 0.0, 14.046798706054688, 73.13897705078125, 13.6080322265625, 56.44883728027344, 28.926681518554688, 14.906951904296875, 7.100311279296875, 28.0936279296875, 2.9940185546875, 60.89385986328125, 15.04449462890625, -0.67742919921875, 35.70452880859375, -6.858673095703125, 86.1236572265625, 17.048797607421875, 32.913848876953125, 74.29144287109375, -13.6220703125, 48.990234375, 0.1686859130859375, 25.53908920288086, 13.74176025390625, -40.665374755859375, 32.326171875, 45.30119705200195, 48.60284423828125, 41.231224060058594, 4.7562408447265625, -20.466766357421875, -21.776992797851562, 1.79595947265625, 1.0255889892578125, 41.03302001953125, -0.899749755859375, 50.023040771484375, 43.76025390625, 88.55508422851562, 25.211395263671875, 65.84100341796875, 10.428081512451172, -63.350311279296875, 38.785125732421875, 58.2052001953125, 0.40988731384277344, 5.97039794921875, 42.68061828613281, 5.2398834228515625, -3.265655517578125, -0.513275146484375, 6.411895751953125, 35.37615966796875, 20.47442626953125, 60.357391357421875, 44.25653076171875, 54.582366943359375, 35.995941162109375, 69.91943359375, 46.661903381347656, -23.844024658203125, -4.814453125, 53.62030029296875, -11.98760986328125, 6.1552581787109375, 43.202423095703125, -10.862342834472656, 17.42755126953125, 27.402725219726562, -47.74601745605469, 17.76605224609375, 8.192863464355469, 48.22004318237305, 30.625091552734375, 26.420547485351562, 65.30453491210938, -1.7658920288085938, -3.735870361328125, 4.16461181640625, -1.5617523193359375, 7.05035400390625, -0.38360595703125, -25.23870849609375, -35.37158203125, 46.18299865722656, 17.54254150390625, 39.63311767578125, 58.148834228515625, 51.36895751953125, 41.900421142578125, -1.754180908203125, 29.93505859375, 15.569061279296875, -8.81365966796875, -33.06866455078125, 10.903060913085938, 64.296630859375, -55.66827392578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000168.npy"}
|
|
{"epoch": 0.3518324607329843, "step": 169, "batch_size": 128, "mean": 23.864307403564453, "std": 29.788509368896484, "min": -48.8389892578125, "p10": -7.254790496826172, "median": 17.763256072998047, "p90": 64.57955932617188, "max": 92.9136962890625, "pos_frac": 0.7734375, "sample": [-2.31671142578125, 37.41485595703125, 78.70751953125, 32.60223388671875, -14.317413330078125, 13.839515686035156, 5.036083221435547, 6.5212554931640625, 3.1390380859375, 18.804534912109375, 18.064102172851562, 49.55126953125, 50.99699783325195, 52.49786376953125, -28.772491455078125, -1.251434326171875, 47.72332763671875, 8.32598876953125, 4.63422966003418, -0.307952880859375, -2.087799072265625, 17.76897430419922, 44.11000061035156, 14.601531982421875, 45.417938232421875, 58.94683837890625, 24.657623291015625, -16.434783935546875, 25.98974609375, 41.75823974609375, 55.893157958984375, 57.224517822265625, 44.7354736328125, -7.1710662841796875, 3.9774703979492188, 8.101097106933594, -12.598226547241211, -48.8389892578125, 25.94024658203125, -2.76922607421875, -7.251487731933594, 60.36187744140625, -6.597198486328125, 13.193862915039062, 0.0, -34.768821716308594, 9.203266143798828, 28.457077026367188, 14.95977783203125, 25.56036376953125, -1.71124267578125, 9.822113037109375, 92.9136962890625, 56.104095458984375, 33.809669494628906, 43.58216857910156, -42.337059020996094, 45.345001220703125, 0.4484081268310547, -1.3804092407226562, 68.20257568359375, 41.682159423828125, 17.757537841796875, -9.145751953125, 36.365478515625, 63.889801025390625, 55.8179931640625, 76.41217041015625, 16.12493896484375, 60.597137451171875, 3.4523544311523438, 77.80126953125, 6.727806091308594, 1.3525390625, 37.696685791015625, -29.734054565429688, 8.799606323242188, 69.65863037109375, 36.271697998046875, -8.2763671875, 45.83537292480469, 8.906303405761719, 11.03428840637207, 43.9246826171875, 21.697738647460938, 46.1998291015625, 54.624298095703125, -7.2624969482421875, 35.722930908203125, 28.899215698242188, 0.36061859130859375, 35.29168701171875, 5.517677307128906, 90.8009033203125, 30.764862060546875, 9.77083969116211, 11.606536865234375, 3.33209228515625, 75.8673095703125, 34.258636474609375, 6.113882064819336, 44.59918975830078, 82.93234252929688, 15.491058349609375, -25.865386962890625, -25.085113525390625, 7.318939208984375, -1.7617874145507812, 84.105712890625, 20.158721923828125, 41.56671142578125, -0.15449905395507812, 62.06158447265625, 49.766448974609375, 44.555145263671875, 0.0, 14.088714599609375, 4.1299285888671875, 21.4844970703125, 40.199745178222656, 12.627655029296875, -1.3708858489990234, 8.447357177734375, 12.335887908935547, 77.68472290039062, 79.6234130859375, 66.18899536132812, -5.050018310546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000169.npy"}
|
|
{"epoch": 0.3539267015706806, "step": 170, "batch_size": 128, "mean": 20.330013275146484, "std": 27.899906158447266, "min": -48.905792236328125, "p10": -9.37613067626953, "median": 18.539710998535156, "p90": 56.74753265380859, "max": 95.01568603515625, "pos_frac": 0.7578125, "sample": [5.733978271484375, 11.665435791015625, 43.66822814941406, -8.761001586914062, 42.144866943359375, 10.42889404296875, 0.5284042358398438, 93.8824462890625, 12.547386169433594, 9.949188232421875, 28.18585205078125, 78.0740966796875, 36.214599609375, 24.115997314453125, -36.6910400390625, 89.59548950195312, -19.39892578125, 29.631866455078125, -1.3906326293945312, 12.48959732055664, -26.52740478515625, 5.351806640625, 23.502914428710938, 29.99017333984375, 8.296768188476562, -25.459800720214844, -18.41119384765625, 3.076268196105957, 54.20477294921875, -6.2540283203125, 24.050277709960938, 22.702531814575195, 35.46199035644531, 4.01243782043457, 15.737632751464844, -0.031036376953125, 0.228790283203125, 4.6938629150390625, 38.582427978515625, 52.469512939453125, 1.4084014892578125, 30.279037475585938, 3.519439697265625, 60.04388427734375, 39.542633056640625, 31.922622680664062, 34.09674072265625, -1.4150390625, -18.64959716796875, 59.54632568359375, 56.37879943847656, 46.926902770996094, 40.678955078125, -39.95836639404297, 23.195159912109375, -3.602996826171875, 60.08795166015625, -10.811431884765625, 54.59669494628906, 43.03551483154297, 13.334716796875, 65.12054443359375, 50.72503662109375, -20.050674438476562, 38.341705322265625, 40.09229278564453, -2.945556640625, 12.21685791015625, 52.58702087402344, 3.1209716796875, 24.991249084472656, 47.078643798828125, 8.408163070678711, 17.84539794921875, 42.917640686035156, 37.0443115234375, 0.0, 57.60791015625, 17.683242797851562, 41.96195983886719, -25.483779907226562, 39.35858154296875, 6.083251953125, 45.09246826171875, 50.224334716796875, 27.908798217773438, -26.390716552734375, 12.90234375, 35.71563720703125, -48.905792236328125, 23.2022705078125, 22.23626708984375, -3.05157470703125, 38.30975341796875, 0.20256805419921875, -4.648313522338867, 58.318267822265625, -4.286407470703125, 26.0198974609375, 95.01568603515625, -0.072021484375, -8.320083618164062, 24.790740966796875, -3.79730224609375, -6.47589111328125, 1.064544677734375, 22.051437377929688, 10.735260009765625, 2.118877410888672, 20.014801025390625, -2.1798477172851562, 3.215484619140625, 74.201904296875, 15.701812744140625, -25.05278778076172, 19.234024047851562, 7.627349853515625, 49.77459716796875, 25.825599670410156, 2.721860885620117, -1.2359542846679688, 33.27069091796875, 60.82066345214844, 52.639305114746094, 5.950721740722656, 31.244346618652344, -7.5435791015625, 58.89520263671875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000170.npy"}
|
|
{"epoch": 0.35602094240837695, "step": 171, "batch_size": 128, "mean": 20.834836959838867, "std": 28.405961990356445, "min": -34.9837646484375, "p10": -14.97576560974121, "median": 18.078948974609375, "p90": 59.81724548339842, "max": 86.53237915039062, "pos_frac": 0.75, "sample": [80.25277709960938, 14.36810302734375, 20.290573120117188, 12.49761962890625, 53.30194091796875, 57.45361328125, 54.46619415283203, -22.1533203125, -15.942615509033203, 14.668960571289062, -17.309982299804688, 1.7322540283203125, 67.51646423339844, -14.5614013671875, 0.7122802734375, 0.8470535278320312, -13.512077331542969, 29.91240692138672, 29.63250732421875, 66.84454345703125, 47.471923828125, -2.4225845336914062, 18.11590576171875, 72.2823486328125, 54.276947021484375, 23.366165161132812, -22.885101318359375, 11.839561462402344, -21.29425048828125, 16.794189453125, 7.153228759765625, 10.131500244140625, -11.162353515625, 31.934158325195312, 45.326019287109375, 67.294189453125, 48.08750915527344, 2.8653564453125, 20.957603454589844, -3.0076065063476562, 13.6640625, -11.266021728515625, -6.088504791259766, -7.14898681640625, -11.320152282714844, 73.69720458984375, 55.08941650390625, 11.230329513549805, -34.9837646484375, 3.930206298828125, 57.39720153808594, 26.33544921875, 8.1297607421875, 55.178375244140625, 29.40789794921875, 20.56048583984375, 34.990631103515625, 64.95498657226562, 70.095947265625, -18.590789794921875, 47.44281005859375, 27.40984344482422, -5.45989990234375, 49.235694885253906, 27.067184448242188, 52.02537536621094, 24.3193359375, 13.207361221313477, 46.88188171386719, 35.449920654296875, 23.170211791992188, 68.369140625, -2.0427093505859375, 4.2429656982421875, -25.419525146484375, -9.445159912109375, 64.29020690917969, 6.220855712890625, -24.783721923828125, 2.6590728759765625, -0.6812210083007812, 3.3214569091796875, -12.177322387695312, -24.757919311523438, 14.33721923828125, -24.415115356445312, 37.99209976196289, 37.1719970703125, -2.19061279296875, 25.318878173828125, 42.23735809326172, 83.29000854492188, -30.473098754882812, 13.02667236328125, 24.9180908203125, 49.07318115234375, 45.377105712890625, 2.5660533905029297, 58.190399169921875, 25.4443359375, 18.220855712890625, 37.077117919921875, -23.686767578125, 12.272705078125, 18.0419921875, 15.6751708984375, 27.02020263671875, 53.3359375, 17.568939208984375, 4.358207702636719, 19.82708740234375, -13.25225830078125, 50.659088134765625, 45.708465576171875, 43.80694580078125, 14.56591796875, 23.66119384765625, 4.484275817871094, 20.549514770507812, 10.156753540039062, 86.53237915039062, 43.21807861328125, -7.070650100708008, 1.0129833221435547, -2.3630828857421875, 63.61322021484375, 23.0108642578125, -6.962921142578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000171.npy"}
|
|
{"epoch": 0.3581151832460733, "step": 172, "batch_size": 128, "mean": 13.93291187286377, "std": 29.282901763916016, "min": -72.40234375, "p10": -14.77611083984375, "median": 13.260818481445312, "p90": 54.448840332031246, "max": 75.872314453125, "pos_frac": 0.6953125, "sample": [13.41107177734375, -7.010772705078125, 6.639923095703125, 15.912841796875, 27.623260498046875, 14.708549499511719, -2.556304931640625, -8.4881591796875, 6.673530578613281, -0.66754150390625, 0.22536468505859375, 0.309814453125, -56.053985595703125, 5.229588508605957, 34.093597412109375, 11.123846054077148, 31.44109344482422, 62.74755859375, 31.74249267578125, 12.723472595214844, -7.947418212890625, 7.599334716796875, 22.988250732421875, 64.55780029296875, 4.513641357421875, 38.652862548828125, 31.349136352539062, -6.406829833984375, -7.3614654541015625, 47.856414794921875, 4.402587890625, 25.89068603515625, -57.7852783203125, 4.5643463134765625, 16.389404296875, 40.21117401123047, -14.518646240234375, 64.9976806640625, -15.376861572265625, -1.0638961791992188, -12.275106430053711, 15.078842163085938, -2.440185546875, -1.7727184295654297, 67.36589050292969, -72.40234375, -6.3252105712890625, 29.080810546875, 60.57073974609375, 41.08819580078125, -7.3341064453125, -25.791015625, 18.404571533203125, 1.3616714477539062, 24.9515380859375, 13.546623229980469, 53.9305419921875, 59.02037048339844, -12.27685546875, -4.448944091796875, -2.720428466796875, 16.8646240234375, 21.300201416015625, -2.3199310302734375, 48.70361328125, 37.841766357421875, -3.847412109375, 6.4757080078125, 16.93090057373047, 6.63079833984375, 72.4559326171875, -47.05908203125, 3.3293838500976562, 38.134857177734375, 62.3076171875, 49.614959716796875, 2.0711669921875, 4.883934020996094, 20.44732666015625, -55.278076171875, 28.65093994140625, 8.647537231445312, 24.436248779296875, -48.525848388671875, 16.33135986328125, 39.701873779296875, 12.067344665527344, 57.479248046875, 13.313323974609375, -38.00758361816406, 61.4049072265625, -8.869421005249023, 55.658203125, -29.32861328125, -34.40226745605469, 17.118026733398438, 20.408966064453125, 18.652587890625, -20.107372283935547, 12.136795043945312, -5.818145751953125, 75.872314453125, 33.726898193359375, 36.064205169677734, 25.343643188476562, 24.977951049804688, -8.485282897949219, 19.758865356445312, 32.238407135009766, 7.299896240234375, 13.20831298828125, 44.46971130371094, 26.321941375732422, 51.87727355957031, 49.2308349609375, 34.04058837890625, -6.651031494140625, 7.890380859375, 7.651739120483398, 46.78607177734375, -63.8238525390625, 56.45386505126953, -0.9058837890625, 9.056594848632812, 0.0, 37.46685791015625, 29.785888671875, -6.6348876953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000172.npy"}
|
|
{"epoch": 0.36020942408376966, "step": 173, "batch_size": 128, "mean": 17.42889404296875, "std": 28.225862503051758, "min": -58.990447998046875, "p10": -16.39081420898437, "median": 16.593698501586914, "p90": 52.90798645019531, "max": 102.40670776367188, "pos_frac": 0.71875, "sample": [4.68585205078125, -58.990447998046875, -38.751930236816406, 28.083206176757812, 26.552978515625, -2.477935791015625, 46.21441650390625, -36.999786376953125, 44.44244384765625, 86.37188720703125, -3.2430267333984375, 24.560943603515625, -14.925933837890625, 34.20684814453125, 38.961669921875, 51.796630859375, 5.8924407958984375, 4.071083068847656, 17.99810791015625, 67.48641967773438, 56.697479248046875, 11.050369262695312, 6.6038360595703125, 33.61346435546875, 18.460128784179688, 15.8018798828125, -1.462677001953125, 6.97265625, 25.93634033203125, 44.18980407714844, 3.77911376953125, 35.84092712402344, -3.2285003662109375, 5.512119293212891, 7.381744384765625, 31.367843627929688, -31.928634643554688, 24.02301025390625, 16.101001739501953, 40.273712158203125, -9.775787353515625, 25.3271484375, 32.920654296875, 34.71051788330078, 43.72975158691406, 32.427276611328125, -4.5854034423828125, -4.4646453857421875, 3.5653228759765625, 9.61981201171875, -3.6368484497070312, -33.550445556640625, -20.871337890625, 102.40670776367188, 14.202545166015625, -25.528900146484375, 9.387451171875, -19.808868408203125, 24.06256866455078, 84.87493896484375, 51.507850646972656, 55.9027099609375, 2.161396026611328, 46.09248352050781, 47.27947998046875, 54.33099365234375, -41.391845703125, 24.6011962890625, 17.086395263671875, 17.175735473632812, 23.186058044433594, 2.9054336547851562, 57.9788818359375, 39.316741943359375, 1.91021728515625, 39.82647705078125, 7.996246337890625, 60.19903564453125, -3.720874786376953, 15.624496459960938, 20.28338623046875, -26.66082763671875, 2.884319305419922, 5.410858154296875, 45.79289245605469, 6.269950866699219, 32.61730194091797, 50.09040832519531, -3.08056640625, 8.0313720703125, 18.56005859375, -1.38299560546875, 44.104644775390625, 24.659881591796875, 57.917388916015625, 24.697662353515625, 43.241424560546875, 32.664093017578125, 38.994903564453125, 49.286529541015625, 58.05506896972656, 9.031951904296875, -4.911151885986328, -10.16973876953125, -8.867391586303711, 52.298126220703125, -25.46771240234375, 47.69136047363281, 30.70244598388672, 17.36578369140625, -4.36907958984375, 0.4798583984375, -11.150390625, 8.51617431640625, 1.361917495727539, -6.54217529296875, -1.833953857421875, -26.252029418945312, 61.757293701171875, -6.928077697753906, 57.49444580078125, -10.207733154296875, 0.0, 26.90473175048828, 24.853759765625, -30.411880493164062, -4.7637481689453125, 23.974838256835938], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000173.npy"}
|
|
{"epoch": 0.362303664921466, "step": 174, "batch_size": 128, "mean": 21.637727737426758, "std": 28.191997528076172, "min": -47.955596923828125, "p10": -9.912945556640624, "median": 17.015281677246094, "p90": 61.784933471679686, "max": 104.08563232421875, "pos_frac": 0.78125, "sample": [45.87151336669922, 1.92572021484375, 4.76161003112793, 1.2552108764648438, 74.357666015625, 51.0992431640625, 62.5797119140625, -30.639129638671875, 9.560562133789062, 1.1811447143554688, 54.53460693359375, 42.76270294189453, 1.708587646484375, 43.14275360107422, 10.92193603515625, 26.751678466796875, 55.74078369140625, 87.76318359375, 10.49920654296875, 11.601760864257812, -3.3272705078125, 62.28948974609375, 2.77203369140625, 69.9935302734375, 67.3177490234375, -5.4158935546875, 36.82505798339844, 8.170242309570312, 69.10679626464844, 41.20527648925781, -5.529899597167969, 42.563751220703125, 104.08563232421875, -21.11944580078125, -16.832626342773438, -6.8097381591796875, 51.11688232421875, 61.568695068359375, 75.46157836914062, 2.5093994140625, 15.787004470825195, 27.129013061523438, 29.448257446289062, 25.562728881835938, -2.505117416381836, 67.27682495117188, 43.209991455078125, 31.396240234375, -4.05487060546875, 87.29608154296875, -10.120758056640625, -22.145050048828125, 45.936119079589844, 31.774169921875, -6.0795745849609375, 1.69921875, 7.133350372314453, -0.260284423828125, 23.80096435546875, 20.14520263671875, 15.043182373046875, 4.520896911621094, 62.54833984375, 46.657684326171875, 17.700958251953125, -12.586395263671875, -4.4353485107421875, 37.535736083984375, 13.086181640625, 31.189651489257812, 19.45630645751953, 42.25762939453125, 51.53075408935547, 47.60711669921875, 38.54895782470703, 21.346961975097656, 7.475677490234375, 11.755767822265625, 54.60821533203125, -9.823883056640625, -29.117965698242188, 35.75586700439453, 1.540557861328125, 34.9051513671875, 18.379043579101562, 2.121397018432617, 54.04606628417969, 8.505678176879883, -17.029083251953125, 13.384321212768555, 7.743736267089844, 16.19842529296875, 46.333251953125, 31.90740966796875, 39.411865234375, 37.6844482421875, 1.0321044921875, 73.42221069335938, -0.7330684661865234, 15.753631591796875, 45.304962158203125, -15.837127685546875, 37.064453125, 20.392547607421875, -7.76708984375, 11.152946472167969, -8.80413818359375, 0.35367584228515625, 0.671875, -28.850555419921875, 35.94464111328125, -1.7064208984375, 2.5269699096679688, 29.797035217285156, 24.24017333984375, -15.1385498046875, 57.496307373046875, -13.00787353515625, 6.52618408203125, -47.955596923828125, 11.703353881835938, 16.329605102539062, 31.490890502929688, 20.57550048828125, 39.02825927734375, 0.8852310180664062, 20.89935302734375, -6.7179718017578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000174.npy"}
|
|
{"epoch": 0.3643979057591623, "step": 175, "batch_size": 128, "mean": 17.543521881103516, "std": 30.197956085205078, "min": -82.47657775878906, "p10": -16.143058776855465, "median": 15.152191162109375, "p90": 53.43164978027343, "max": 104.8623046875, "pos_frac": 0.765625, "sample": [44.6607666015625, 23.31549072265625, 33.6385498046875, -2.2809276580810547, 1.1829147338867188, -36.42857360839844, 21.411773681640625, 3.9722900390625, 77.46102905273438, -38.74530029296875, 11.655410766601562, 10.95806884765625, 27.076950073242188, 22.695541381835938, 41.26608657836914, 104.8623046875, 4.146942138671875, 40.755096435546875, 30.346237182617188, 61.20379638671875, -28.85443115234375, 47.19953918457031, 49.70198059082031, 40.08697509765625, 36.86351013183594, 1.09808349609375, -1.9825897216796875, -27.51141357421875, 25.884017944335938, -29.293701171875, -4.845268249511719, -4.6651611328125, 20.009521484375, 48.610321044921875, -3.08074951171875, 47.967987060546875, 9.7034912109375, 0.792083740234375, 60.958160400390625, 38.3067626953125, 2.9983367919921875, 9.930328369140625, 12.197784423828125, 18.359619140625, 8.046905517578125, 60.286712646484375, 47.000030517578125, 45.486602783203125, -12.40045166015625, 71.3858642578125, 26.628448486328125, 11.206798553466797, 0.83172607421875, 11.106475830078125, -11.636581420898438, -18.090606689453125, 14.793731689453125, -0.869964599609375, -1.3222808837890625, 61.291595458984375, 19.507020950317383, 44.550933837890625, 7.357086181640625, 32.622642517089844, 48.141143798828125, 6.2020263671875, 1.5016365051269531, 4.383205413818359, 3.54119873046875, 5.8809814453125, 50.514129638671875, -2.5212860107421875, 9.85321044921875, 55.0101318359375, 41.185585021972656, -33.95770263671875, -0.411590576171875, 52.755157470703125, 0.0, -82.47657775878906, 17.572540283203125, 33.12945556640625, 66.20285034179688, 7.11676025390625, 20.858154296875, -1.510467529296875, 20.81915283203125, -15.308395385742188, -31.506591796875, 72.028564453125, 37.9451904296875, 3.6791458129882812, 42.654266357421875, 40.8721923828125, 42.93965148925781, 30.556427001953125, -8.51544189453125, 10.376842498779297, -22.20806884765625, 5.89501953125, 39.911865234375, 44.27099609375, 8.759239196777344, 17.9412841796875, 3.497802734375, -63.93994140625, 18.83612060546875, 15.510650634765625, -9.005645751953125, 22.964691162109375, 29.783649444580078, 19.57891845703125, 25.284027099609375, 28.026344299316406, 67.5970458984375, 5.31134033203125, 21.843231201171875, 72.8814697265625, 44.02496337890625, 13.48065185546875, 11.02423095703125, 8.597396850585938, 19.9166259765625, -62.21319580078125, -43.214874267578125, 0.0, 4.135009765625, 76.19622802734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000175.npy"}
|
|
{"epoch": 0.36649214659685864, "step": 176, "batch_size": 128, "mean": 12.85472297668457, "std": 31.457786560058594, "min": -93.03717041015625, "p10": -20.078851318359373, "median": 9.457326889038086, "p90": 51.19303970336914, "max": 89.06619262695312, "pos_frac": 0.71875, "sample": [-17.041961669921875, 11.656723022460938, -0.1080322265625, -19.1187744140625, 18.94097900390625, 37.81165313720703, -41.19712829589844, -44.050567626953125, 12.790336608886719, 13.477935791015625, -69.8115234375, 3.3066558837890625, 19.96286392211914, 47.079437255859375, 10.190933227539062, 31.959365844726562, 38.59828186035156, 3.4138412475585938, 2.21343994140625, -7.3498077392578125, -36.321136474609375, 4.12579345703125, 7.9677886962890625, -59.0672607421875, 11.199310302734375, 65.57974243164062, 25.361854553222656, 83.148193359375, 29.018585205078125, -1.919586181640625, 77.3753662109375, -28.049652099609375, -1.4371757507324219, 16.467498779296875, 9.283184051513672, 3.022186279296875, -10.71142578125, 22.625518798828125, 13.5670166015625, 8.508758544921875, -39.406280517578125, 10.959091186523438, 41.220001220703125, -2.4991455078125, 6.651359558105469, 50.82134246826172, 5.3778839111328125, 1.8419265747070312, 30.723602294921875, 66.5498046875, -28.538795471191406, 44.01420593261719, 39.85328674316406, -14.185394287109375, 14.47967529296875, 46.11553955078125, -3.31201171875, 37.79838562011719, 32.61695861816406, 4.222709655761719, 3.723480224609375, 1.2217330932617188, -3.519073486328125, -14.119331359863281, 39.70684814453125, 52.060333251953125, 33.2027587890625, -9.664306640625, 58.16253662109375, 1.804168701171875, -14.334537506103516, 62.59783935546875, 12.711746215820312, 2.436370849609375, 89.06619262695312, 46.316986083984375, 0.5965290069580078, -14.007080078125, 14.698959350585938, 9.6314697265625, 28.412399291992188, 34.573272705078125, 43.605323791503906, 36.27002716064453, 34.37640380859375, 48.150123596191406, -6.47882080078125, 6.291957855224609, 3.084320068359375, 70.48101806640625, 26.761581420898438, 4.6074981689453125, -27.061431884765625, 26.140884399414062, -11.878631591796875, 3.0558910369873047, 46.76617431640625, -63.2064208984375, -22.31903076171875, 29.989349365234375, -12.211334228515625, 45.49633026123047, 3.72161865234375, -13.479034423828125, 6.665534973144531, -54.95184326171875, 23.251358032226562, 11.563323974609375, -13.253250122070312, 4.712589263916016, 56.46674346923828, -6.7887115478515625, 69.17507934570312, 45.054473876953125, 70.40388488769531, 30.404212951660156, 4.096435546875, 19.21167755126953, -18.688446044921875, -93.03717041015625, 52.0994873046875, 1.3746337890625, 0.9091968536376953, 26.44367218017578, 31.1983642578125, 8.51580810546875, 27.33464813232422, -1.9736404418945312], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000176.npy"}
|
|
{"epoch": 0.36858638743455496, "step": 177, "batch_size": 128, "mean": 12.943015098571777, "std": 33.11043930053711, "min": -93.11062622070312, "p10": -25.33723526000976, "median": 10.013603210449219, "p90": 55.27855529785156, "max": 88.9703369140625, "pos_frac": 0.6796875, "sample": [41.58512878417969, -2.2680130004882812, 2.1378936767578125, 69.26904296875, -3.0586013793945312, 57.388092041015625, 11.006145477294922, -23.7159423828125, 26.77557373046875, 5.283958435058594, 51.291534423828125, -36.23488998413086, -70.0205078125, -11.388525009155273, 3.0115509033203125, -14.315826416015625, 35.277435302734375, 4.053375244140625, 41.43324279785156, 3.9144287109375, 32.4315185546875, 29.126800537109375, 37.291839599609375, 55.0848388671875, -20.88140869140625, 31.44744873046875, -32.22613525390625, 70.95852661132812, 46.668182373046875, 67.13360595703125, 6.3384246826171875, 25.555984497070312, -45.03826904296875, 76.97857666015625, -4.01495361328125, 25.860687255859375, 54.336181640625, 3.2652244567871094, -9.484502792358398, 5.20794677734375, 32.829681396484375, -22.781723022460938, -3.588411331176758, 2.5293045043945312, -4.58929443359375, -23.744956970214844, 4.141990661621094, 12.394508361816406, 41.32469177246094, 35.21583557128906, -10.3175048828125, 2.2992401123046875, 76.16915893554688, 22.182388305664062, -2.487152099609375, 54.36408996582031, 20.192962646484375, -4.434600830078125, 60.98335266113281, 4.483642578125, -36.1468505859375, 64.83912658691406, 0.6641387939453125, -14.770751953125, 52.17535400390625, -0.4148406982421875, -35.22917175292969, -7.2672119140625, 5.375118255615234, 55.730560302734375, -13.68511962890625, 83.64764404296875, -6.34063720703125, 7.484615325927734, -4.537269592285156, 40.07411193847656, 40.65936279296875, -71.09722900390625, 20.99371337890625, 20.728729248046875, 21.92596435546875, -35.244293212890625, -42.931785583496094, 18.341552734375, -55.257843017578125, 10.864501953125, 29.56171417236328, 1.7379150390625, -9.7261962890625, 9.162704467773438, 18.19158935546875, 50.0645751953125, 33.84881591796875, 32.17616271972656, -9.3426513671875, -15.894134521484375, 14.626113891601562, 11.77880859375, 48.61175537109375, 8.303565979003906, -20.615310668945312, -4.7557220458984375, 11.079315185546875, 36.829071044921875, 8.804569244384766, -64.96783447265625, 7.1214752197265625, 1.3798675537109375, -29.05255126953125, -93.11062622070312, 22.936912536621094, 32.78504943847656, 59.934600830078125, 22.35498046875, 13.312313079833984, 57.8251953125, 33.95068359375, 28.438720703125, 88.9703369140625, 11.025405883789062, 48.21527099609375, 2.0746536254882812, -1.026031494140625, -2.5864486694335938, 17.1690673828125, 2.4609375, 28.09930419921875, 49.73772430419922], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000177.npy"}
|
|
{"epoch": 0.3706806282722513, "step": 178, "batch_size": 128, "mean": 20.42443084716797, "std": 31.769960403442383, "min": -61.15008544921875, "p10": -13.702149200439452, "median": 17.3427734375, "p90": 64.7998291015625, "max": 119.52972412109375, "pos_frac": 0.734375, "sample": [49.71568298339844, 13.985527038574219, 18.848243713378906, -3.897247314453125, -6.443534851074219, 6.68867301940918, 38.308837890625, 38.8182373046875, 26.144437789916992, 31.7412109375, 119.52972412109375, 45.08647155761719, 31.08409881591797, 13.169677734375, -11.46148681640625, 70.72601318359375, 0.0, 50.63075256347656, 47.55976867675781, 21.443359375, 12.392852783203125, 76.89492797851562, 6.776288986206055, 64.8138427734375, 44.74755859375, -14.246810913085938, -7.15155029296875, 85.40673828125, -15.741790771484375, 30.723052978515625, 3.8828659057617188, -3.694549560546875, 0.28887939453125, 41.591651916503906, -56.68052673339844, 10.09893798828125, 23.978607177734375, 58.79963684082031, -27.90667724609375, 45.452484130859375, 64.7938232421875, 29.219268798828125, 26.723297119140625, 34.113922119140625, 59.904205322265625, 54.534332275390625, 5.557823181152344, -38.838287353515625, -3.155426025390625, 80.12016296386719, 47.95195007324219, 76.93130493164062, -36.01727294921875, -61.15008544921875, -4.774871826171875, 10.546043395996094, 13.9420166015625, 39.593162536621094, -26.610244750976562, 21.6422119140625, 0.15520477294921875, 86.84341430664062, 7.427276611328125, 34.763427734375, 53.3056640625, 19.6627197265625, 22.76837158203125, 27.64276123046875, 30.39520263671875, 10.696548461914062, 76.80392456054688, 16.224411010742188, -11.004383087158203, -1.6180419921875, 11.595972061157227, 70.198486328125, -37.402252197265625, 10.986236572265625, -5.846916198730469, 3.581634521484375, 37.103057861328125, 23.00811767578125, 29.5982666015625, -4.5096435546875, -52.184600830078125, -3.9715805053710938, 71.69915771484375, -3.1381988525390625, 15.188087463378906, 23.34527587890625, 11.348289489746094, 3.173643112182617, -0.18804931640625, 8.920257568359375, 11.088714599609375, 61.97918701171875, 4.351715087890625, 6.660743713378906, 0.718475341796875, 56.24444580078125, 8.124107360839844, -8.988893508911133, -13.868919372558594, 81.7239990234375, 23.513893127441406, -13.63067626953125, 18.461135864257812, 51.295623779296875, 48.22528076171875, -11.978042602539062, -3.9969253540039062, 84.5098876953125, -23.281646728515625, 9.604736328125, 35.0406494140625, 20.46246337890625, -19.227203369140625, 14.657218933105469, 35.707366943359375, 4.74658203125, 21.846405029296875, 29.00396728515625, 24.045562744140625, -0.61102294921875, -6.213371276855469, 24.649383544921875, 42.88325881958008, 32.87501525878906], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000178.npy"}
|
|
{"epoch": 0.37277486910994767, "step": 179, "batch_size": 128, "mean": 14.26394271850586, "std": 30.53413963317871, "min": -55.9342041015625, "p10": -24.748720550537104, "median": 10.792922973632812, "p90": 57.79271469116211, "max": 85.58078002929688, "pos_frac": 0.6875, "sample": [24.959808349609375, -0.6537208557128906, 57.2894287109375, -7.21002197265625, 10.796356201171875, 56.09477233886719, 6.738067626953125, 10.35207748413086, -43.08209228515625, -0.12127685546875, 10.481048583984375, -0.83251953125, 21.559234619140625, 5.8282318115234375, 58.335052490234375, -33.60236358642578, 39.656036376953125, 24.03192138671875, 8.337982177734375, -4.541597366333008, 21.59820556640625, 36.93480682373047, 5.162773132324219, 24.23004150390625, 17.794036865234375, -23.69556427001953, -34.15650939941406, -4.830196380615234, -3.5466537475585938, -43.32954406738281, 40.67228698730469, 5.538017272949219, -37.641632080078125, 5.6250152587890625, -9.116546630859375, 25.271881103515625, 39.523773193359375, 50.00718688964844, 18.58953857421875, -5.7619781494140625, -27.206085205078125, -40.892791748046875, -3.855865478515625, 18.143157958984375, -17.348052978515625, 26.578781127929688, 6.1726837158203125, -8.276477813720703, 52.147705078125, -51.4708251953125, -20.273651123046875, -22.862945556640625, 14.85727310180664, 19.5382080078125, 5.577850341796875, 68.82745361328125, 58.211143493652344, 51.647300720214844, 20.3704833984375, -2.5562744140625, -55.9342041015625, -9.53350830078125, 78.36691284179688, 13.188140869140625, -6.910488128662109, 31.217132568359375, 57.61338806152344, -14.477859497070312, -1.9894866943359375, 10.78948974609375, 6.170986175537109, 37.80023193359375, 17.4603271484375, 29.63739013671875, 1.8383941650390625, 5.678102493286133, 62.81257629394531, 9.415008544921875, 2.2622642517089844, 30.539756774902344, 4.09589958190918, 7.9798736572265625, 80.486083984375, -7.46478271484375, 11.887649536132812, -35.3734130859375, 69.27648162841797, 61.255615234375, 1.1052780151367188, -30.18841552734375, 40.784942626953125, 3.0264053344726562, 68.7484130859375, 29.70135498046875, 26.890167236328125, 21.5731201171875, 23.934951782226562, 30.619186401367188, 14.662849426269531, 27.313247680664062, 69.865234375, 36.67829895019531, 56.50709533691406, 11.783195495605469, -36.8782958984375, 23.987823486328125, -43.97015380859375, 10.495559692382812, 15.047691345214844, 53.687477111816406, 7.670249938964844, 32.25067138671875, -23.401199340820312, 85.58078002929688, -14.650726318359375, -12.646705627441406, 39.51464080810547, 42.4306640625, 2.8543701171875, -9.18060302734375, 67.09228515625, 43.25373840332031, 0.14104461669921875, 11.18642807006836, -7.2875823974609375, -16.73992919921875, 59.80607604980469, 43.83251953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000179.npy"}
|
|
{"epoch": 0.374869109947644, "step": 180, "batch_size": 128, "mean": 19.137468338012695, "std": 29.240571975708008, "min": -54.73017883300781, "p10": -14.31068534851074, "median": 12.38021469116211, "p90": 61.12967834472656, "max": 84.6689453125, "pos_frac": 0.7578125, "sample": [0.17339324951171875, -4.154598236083984, -29.001419067382812, 3.5874099731445312, 6.7263336181640625, 29.30596923828125, 57.17108154296875, 6.95220947265625, -2.229583740234375, 62.127960205078125, -2.81805419921875, 44.329986572265625, 69.9410400390625, -48.40876007080078, 5.7491455078125, 55.77207946777344, -8.997802734375, 6.1103515625, 11.209064483642578, -54.444366455078125, 26.313262939453125, 64.87753295898438, 31.833984375, 30.9853515625, 20.778228759765625, 42.311676025390625, 18.746124267578125, 51.706268310546875, 84.6689453125, -5.430131912231445, -54.73017883300781, 10.822898864746094, 59.18658447265625, 82.14920043945312, 65.40341186523438, 7.3362884521484375, 10.463211059570312, 7.3590545654296875, 9.169784545898438, 12.894790649414062, 56.898681640625, -5.521873474121094, 41.36627960205078, 27.188953399658203, 75.11224365234375, 15.53826904296875, -16.420730590820312, -2.055145263671875, 5.6656494140625, 1.397613525390625, -11.994476318359375, -9.393386840820312, 4.32757568359375, 25.2882080078125, 62.63616943359375, 1.9813232421875, 16.83068084716797, 38.052574157714844, 10.210968017578125, 23.842540740966797, 75.0517578125, 37.05797576904297, -6.575469970703125, 82.64849853515625, 44.790138244628906, -29.329681396484375, 19.988441467285156, 30.4918212890625, -17.113067626953125, 13.369537353515625, 36.2884521484375, 6.169929504394531, 11.865638732910156, 48.307861328125, 57.764556884765625, 60.70184326171875, -0.5968399047851562, 0.21526145935058594, 41.779510498046875, 39.301849365234375, 80.88369750976562, 17.046009063720703, -7.451202392578125, 43.8326416015625, 41.70695495605469, 36.919090270996094, 17.158281326293945, -33.91942596435547, 3.571014404296875, 44.15089416503906, -0.17694091796875, 16.763580322265625, 1.9151382446289062, 27.16949462890625, 46.955841064453125, 9.9296875, 23.77606201171875, 27.41351318359375, 5.1000213623046875, -18.539138793945312, 4.714347839355469, 19.388916015625, 55.73394775390625, 22.081268310546875, 5.162580490112305, -16.21780776977539, 4.267429351806641, -18.01837158203125, 64.95526123046875, 3.7290115356445312, -13.49334716796875, 36.412349700927734, -4.000873565673828, -17.24298095703125, 41.320106506347656, 31.060699462890625, 29.001541137695312, -1.821807861328125, -10.9947509765625, 42.44329833984375, 2.487638473510742, 8.81817626953125, 3.1172256469726562, 4.837139129638672, -1.15216064453125, 74.64370727539062, -16.638473510742188, 11.71875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000180.npy"}
|
|
{"epoch": 0.3769633507853403, "step": 181, "batch_size": 128, "mean": 17.90673065185547, "std": 30.793968200683594, "min": -61.3377685546875, "p10": -20.452730560302733, "median": 15.464406967163086, "p90": 54.981964874267575, "max": 89.70761108398438, "pos_frac": 0.7421875, "sample": [0.8019294738769531, 5.308570861816406, 1.06964111328125, 8.943328857421875, 0.0, 32.69183349609375, 35.578399658203125, 28.99822998046875, 8.709098815917969, 25.791751861572266, 60.706298828125, -9.244621276855469, 42.116302490234375, -5.7976531982421875, 77.42584228515625, 12.161590576171875, 0.0, 45.98191833496094, 41.297149658203125, -1.0584716796875, 52.795204162597656, 8.9298095703125, 40.9439697265625, 7.9404296875, -60.67054748535156, 77.32275390625, -9.491302490234375, 13.586933135986328, 5.810882568359375, 23.609344482421875, 53.257328033447266, 2.4139480590820312, 15.504955291748047, 51.795806884765625, 25.7462158203125, 30.577957153320312, 54.80870056152344, 15.423858642578125, 15.593841552734375, 28.4935302734375, 16.840545654296875, 23.94769287109375, -33.457611083984375, -5.414323806762695, 9.448822021484375, 13.677413940429688, 8.883377075195312, 11.695068359375, 80.7173843383789, -38.044700622558594, 26.78118896484375, 8.486747741699219, 42.24552917480469, 52.94140625, -13.258392333984375, -55.96148681640625, -1.8661689758300781, 14.214996337890625, -2.60430908203125, 18.324493408203125, 35.633544921875, 4.3897705078125, -20.406494140625, 62.130096435546875, 23.15380859375, 21.11429214477539, 29.3355712890625, 54.103851318359375, 1.401092529296875, 25.52899169921875, 18.517059326171875, 38.87959289550781, 10.181434631347656, 39.219390869140625, 86.23291015625, 89.70761108398438, -40.55755615234375, 11.097488403320312, -23.1171875, 52.047332763671875, -22.640838623046875, 12.934814453125, -4.843076705932617, -51.68695068359375, -8.014419555664062, 55.199310302734375, 13.837310791015625, -7.628517150878906, 10.07952880859375, 2.59783935546875, -15.007125854492188, 28.45294189453125, 65.27944946289062, -15.563690185546875, 12.412742614746094, 18.05640411376953, 9.610061645507812, 18.452957153320312, 16.2513427734375, 69.96829223632812, 66.9111328125, -23.04443359375, 13.869140625, 11.621185302734375, -28.1051025390625, -10.342857360839844, 17.639572143554688, 75.66543579101562, -16.98284912109375, -61.3377685546875, 51.91815185546875, 22.53582763671875, 54.757259368896484, -0.1278076171875, 23.53289794921875, 70.95973205566406, -38.12091064453125, 16.42236328125, 19.6533203125, -0.903350830078125, 14.617080688476562, 46.76502990722656, 53.277557373046875, -20.56061553955078, 54.888816833496094, 23.809661865234375, 37.44720458984375, 45.482330322265625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000181.npy"}
|
|
{"epoch": 0.37905759162303665, "step": 182, "batch_size": 128, "mean": 22.094989776611328, "std": 26.746337890625, "min": -55.807952880859375, "p10": -6.775614166259765, "median": 22.945770263671875, "p90": 58.81895446777343, "max": 79.92141723632812, "pos_frac": 0.8125, "sample": [36.148284912109375, 48.54255676269531, -2.102783203125, -0.8106784820556641, 19.930694580078125, 8.471221923828125, 58.052520751953125, 27.416961669921875, 19.613510131835938, 5.3525238037109375, 57.48101806640625, 25.8201904296875, 23.72271728515625, 45.50445556640625, 29.022674560546875, -11.233062744140625, 33.02642822265625, 46.808502197265625, 61.91059875488281, 38.298927307128906, 40.419700622558594, 21.159420013427734, 39.34748840332031, 67.36474609375, -3.67852783203125, 63.82879638671875, -6.632835388183594, 29.960205078125, 77.37210083007812, 27.086212158203125, 46.71813201904297, 13.531463623046875, 28.48065185546875, 28.77117919921875, 9.898818969726562, -2.5443801879882812, -10.894424438476562, 9.2869873046875, 15.946823120117188, 7.424468994140625, -40.007476806640625, 17.389007568359375, -55.807952880859375, 21.46990203857422, 55.722442626953125, -6.2565460205078125, 71.8555908203125, -55.77972412109375, 51.50770568847656, 66.22555541992188, 25.655426025390625, 52.362579345703125, 40.57684326171875, 60.947052001953125, 36.9857177734375, -5.787353515625, 9.07766342163086, 6.406806945800781, 5.973388671875, 33.818939208984375, 25.6219482421875, 55.96372985839844, 40.944488525390625, 27.237213134765625, 26.217056274414062, 9.252243041992188, 38.494361877441406, 27.498199462890625, 79.92141723632812, 10.295005798339844, -5.411750793457031, 8.293676376342773, 22.1688232421875, 15.39996337890625, 2.3590087890625, 9.804168701171875, 42.130889892578125, 37.378753662109375, 6.234184265136719, 8.575469970703125, 5.4857635498046875, -7.1087646484375, 0.8428497314453125, 26.120529174804688, 1.99603271484375, 0.3168182373046875, 33.731781005859375, 3.7874679565429688, 15.21209716796875, 2.5547866821289062, 60.6072998046875, 50.00537109375, 69.03756713867188, 61.120269775390625, -35.25291442871094, -9.25423812866211, 67.82049560546875, 40.92418670654297, 21.2164306640625, 32.597747802734375, -15.196578979492188, -4.1485595703125, 3.19097900390625, 10.110700607299805, -3.925262451171875, 31.212162017822266, 28.710601806640625, 24.881271362304688, 42.1461181640625, 7.53607177734375, -22.8702392578125, 44.162872314453125, 56.9464111328125, 2.279306411743164, 2.2027511596679688, 5.118743896484375, 4.893669128417969, 54.0478515625, 33.92101287841797, -14.193962097167969, -15.918487548828125, 54.81146240234375, 28.950927734375, 3.205810546875, 60.836700439453125, -0.7737274169921875, 25.8616943359375, -16.14197540283203], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000182.npy"}
|
|
{"epoch": 0.381151832460733, "step": 183, "batch_size": 128, "mean": 18.37460708618164, "std": 29.4113712310791, "min": -51.50940704345703, "p10": -19.65650177001953, "median": 16.842016220092773, "p90": 56.42259674072265, "max": 88.17529296875, "pos_frac": 0.734375, "sample": [-20.823150634765625, 35.235107421875, -19.156509399414062, 47.28325653076172, -22.53277587890625, 61.386474609375, -15.758026123046875, 13.468582153320312, 53.66957092285156, 30.39966583251953, -12.22174072265625, 17.069183349609375, 28.390228271484375, 0.459228515625, 26.7235107421875, -34.2713623046875, 2.254558563232422, 29.147079467773438, 14.892822265625, 2.2163333892822266, 20.552764892578125, 6.0780029296875, 43.22117614746094, -20.879257202148438, 15.520484924316406, -43.974029541015625, 53.910888671875, 13.555328369140625, 9.829742431640625, 14.670974731445312, -4.5841064453125, -26.90985107421875, 11.712387084960938, 11.1365966796875, 7.3296051025390625, 52.15678405761719, 9.010452270507812, 59.02757263183594, 25.289093017578125, -5.274505615234375, 56.31195068359375, -4.240867614746094, 27.59696388244629, -16.0997314453125, 50.78533935546875, 32.36125183105469, 47.32124328613281, -18.068771362304688, 68.9840087890625, 38.19662094116211, -1.57421875, 32.752655029296875, 47.96293640136719, -36.42462158203125, 0.0, 0.5084953308105469, -6.10479736328125, 18.174362182617188, -16.4569091796875, 40.569000244140625, 12.933448791503906, 69.85052490234375, 0.25633811950683594, 75.83673095703125, 41.656280517578125, -26.859130859375, 10.169273376464844, 72.1815185546875, 39.25982666015625, -15.519088745117188, 34.48497772216797, 77.47564697265625, 88.17529296875, 45.233131408691406, 0.587188720703125, 11.4046630859375, -35.083892822265625, 23.934722900390625, -3.443603515625, 47.84135437011719, 30.663299560546875, 20.911895751953125, 8.303840637207031, 16.614849090576172, -7.122020721435547, 10.531036376953125, 20.876556396484375, 31.767562866210938, 10.163482666015625, 12.080963134765625, -0.04863739013671875, 21.458389282226562, 3.4530029296875, 2.9646682739257812, -3.882720947265625, 57.1861572265625, 41.76374816894531, 42.93414306640625, 33.31512451171875, 0.0559234619140625, -51.50940704345703, 6.9450531005859375, -23.188858032226562, 37.948883056640625, -3.805561065673828, 43.60032653808594, 0.7455062866210938, 56.68077087402344, -3.3100051879882812, 71.56851196289062, 17.99078369140625, 75.54150390625, 31.519866943359375, 48.884071350097656, 20.503692626953125, -42.011810302734375, 18.427001953125, 35.63427734375, -34.700042724609375, 82.58062744140625, -6.37017822265625, 35.204872131347656, 27.21368408203125, 29.702850341796875, -5.861480712890625, 20.23308563232422, 51.42308044433594, 34.22515869140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000183.npy"}
|
|
{"epoch": 0.3832460732984293, "step": 184, "batch_size": 128, "mean": 26.141279220581055, "std": 32.443511962890625, "min": -49.14512634277344, "p10": -14.841909790039061, "median": 24.250228881835938, "p90": 70.4194839477539, "max": 97.53604125976562, "pos_frac": 0.7890625, "sample": [26.641021728515625, 97.53604125976562, 64.4940185546875, 24.112442016601562, 24.388015747070312, -10.082489013671875, 14.35748291015625, -6.6524810791015625, 70.3187255859375, 53.615509033203125, 23.785507202148438, 17.14947509765625, -24.9537353515625, 15.19000244140625, -3.4271087646484375, 3.8395919799804688, 82.32229614257812, 42.11799621582031, 19.66680908203125, 43.236488342285156, 49.939239501953125, 65.8424072265625, -45.83673095703125, 54.86199951171875, -6.2711181640625, 72.25936889648438, 32.0445556640625, -9.774795532226562, -18.565635681152344, 77.769775390625, 9.221420288085938, 73.42794799804688, 23.066421508789062, -12.884368896484375, 41.6978759765625, 66.3555908203125, 5.86834716796875, 41.50408935546875, 39.85272216796875, 77.66268920898438, 21.380828857421875, 83.60397338867188, -49.14512634277344, 38.96146011352539, 11.965972900390625, 89.29022216796875, -39.03033447265625, 59.178466796875, -14.510040283203125, 44.165771484375, 16.35483169555664, -9.892822265625, 19.906463623046875, 35.431793212890625, -18.578033447265625, 30.843475341796875, 31.20989990234375, 15.630325317382812, -34.60188293457031, 89.25631713867188, 7.627685546875, 6.808389663696289, 30.574432373046875, -41.889991760253906, 32.749755859375, 39.5120849609375, 52.01983642578125, 52.64299011230469, 20.888702392578125, 37.259185791015625, 13.17913818359375, 12.606704711914062, 47.66553497314453, 2.0330734252929688, 49.6922607421875, 4.235370635986328, 23.22821044921875, 56.19708251953125, 39.48661804199219, 40.6373291015625, 63.87652587890625, 35.055206298828125, -10.691146850585938, 60.36354064941406, 5.7611083984375, -22.287338256835938, 6.307098388671875, 70.65458679199219, 4.4320831298828125, 25.137237548828125, 5.1769561767578125, 42.221343994140625, 12.406387329101562, 13.507316589355469, 41.325469970703125, -1.2142486572265625, -16.357789993286133, 19.921615600585938, 0.9933128356933594, 63.10992431640625, 42.88294982910156, 42.00860595703125, 12.57586669921875, 63.25537109375, 1.0049285888671875, 2.352468490600586, 52.71942138671875, 8.1600341796875, -8.086326599121094, 78.86639404296875, 24.53887939453125, -20.76812744140625, 56.56488037109375, 30.875587463378906, 47.1386833190918, -4.50665283203125, 73.21490478515625, 5.6986236572265625, -11.794700622558594, 31.27855682373047, -15.61627197265625, -23.54412841796875, 68.6612548828125, 15.5076904296875, 87.4609375, 69.49066162109375, -6.609565734863281, 38.780208587646484], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000184.npy"}
|
|
{"epoch": 0.38534031413612563, "step": 185, "batch_size": 128, "mean": 19.300125122070312, "std": 32.28630065917969, "min": -58.33258056640625, "p10": -19.6685791015625, "median": 15.384063720703125, "p90": 61.577441406249996, "max": 93.14743041992188, "pos_frac": 0.6796875, "sample": [47.521331787109375, -23.784698486328125, 47.543212890625, -15.541595458984375, -2.0040130615234375, 68.69293212890625, 53.42854309082031, 61.371246337890625, 43.716796875, -6.557518005371094, 45.669281005859375, 7.5453338623046875, 48.6597900390625, 44.24543380737305, 11.463096618652344, 81.41497802734375, 38.07342529296875, -11.5196533203125, 8.898109436035156, 59.46136474609375, 29.854263305664062, 9.0213623046875, 46.89565658569336, 44.98701477050781, 24.02734375, 13.4769287109375, 35.405426025390625, 6.926898956298828, 1.9247417449951172, 69.92877197265625, -58.33258056640625, 47.750511169433594, 0.990081787109375, 0.359771728515625, 20.120872497558594, 38.959564208984375, 22.72198486328125, 18.059402465820312, 25.2850341796875, 62.058563232421875, -4.831878662109375, 19.464630126953125, -26.413848876953125, 44.218170166015625, 2.82330322265625, 68.01693725585938, -11.723983764648438, 37.57978057861328, 22.38385009765625, 54.10791015625, -24.520187377929688, -24.273788452148438, -41.13580322265625, -13.87261962890625, 6.5892486572265625, 0.4750938415527344, 11.085800170898438, 0.0, 7.3151702880859375, 36.175323486328125, -9.858734130859375, -19.611129760742188, 10.304901123046875, -1.8828201293945312, 32.04731750488281, 41.84013366699219, 30.11893653869629, 16.3499755859375, 40.7093505859375, -2.689870834350586, -6.392120361328125, 55.545166015625, 45.529632568359375, 51.013710021972656, -7.783733367919922, -1.22833251953125, -7.105865478515625, 14.41815185546875, 18.893535614013672, 54.857513427734375, 93.14743041992188, -13.389846801757812, 7.9922027587890625, 0.0, -7.9134674072265625, 46.16609573364258, 65.75772094726562, -18.70770263671875, 79.87994384765625, -27.844375610351562, 43.66059112548828, 39.8389892578125, -10.141921997070312, -2.3188915252685547, 38.94061279296875, 77.42266845703125, -31.635101318359375, 33.693260192871094, 36.602256774902344, 10.713714599609375, -4.011909484863281, 19.649002075195312, 36.34613037109375, -4.92425537109375, 80.98159790039062, -45.66778564453125, -14.114227294921875, 7.619842529296875, 91.50350952148438, 89.28436279296875, 50.46026611328125, 5.477287292480469, 10.230476379394531, -9.365234375, 48.927886962890625, -43.32696533203125, 35.807891845703125, 75.58758544921875, 16.696731567382812, -21.725616455078125, 43.122474670410156, 1.9224395751953125, -11.2982177734375, -11.5352783203125, -19.802627563476562, -38.912986755371094, 12.915657043457031, 19.44580078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000185.npy"}
|
|
{"epoch": 0.387434554973822, "step": 186, "batch_size": 128, "mean": 18.90727996826172, "std": 28.8080997467041, "min": -59.56195068359375, "p10": -9.580294799804687, "median": 16.937576293945312, "p90": 60.6880630493164, "max": 83.96218872070312, "pos_frac": 0.7421875, "sample": [21.686355590820312, 2.171642303466797, 25.6571044921875, 17.5181884765625, 19.793792724609375, 72.45367431640625, 26.2919921875, 7.06695556640625, 12.842514038085938, 10.900577545166016, 20.57575225830078, 26.5054931640625, 6.5512847900390625, 51.792694091796875, -16.761871337890625, 27.08978271484375, 28.75146484375, 2.3008193969726562, 62.402313232421875, 32.173240661621094, 12.4886474609375, 6.283790588378906, -6.0085296630859375, 3.3389739990234375, 5.9285736083984375, 46.137420654296875, 13.224777221679688, 23.83673095703125, -2.813873291015625, -0.669891357421875, 8.03592300415039, 30.582763671875, 80.50984191894531, -0.5376434326171875, -5.835693359375, -44.403968811035156, 25.381988525390625, 70.71664428710938, 22.85205078125, 24.56964111328125, 6.550422668457031, 53.70831298828125, 47.95619201660156, 42.507205963134766, 37.8741569519043, 47.88824462890625, -28.3609619140625, 3.6653270721435547, 43.062416076660156, 37.963623046875, 13.659088134765625, 13.9906005859375, 36.43214416503906, 18.258270263671875, -12.7039794921875, -23.03839111328125, 33.446044921875, 17.575485229492188, 1.6669921875, -9.232513427734375, -6.238683700561523, -10.39178466796875, 65.75091552734375, -41.3304443359375, 59.95338439941406, -4.33251953125, 27.85443115234375, 17.02984619140625, -2.5211181640625, 63.08497619628906, -2.2344741821289062, 16.744140625, 81.88217163085938, 38.48118591308594, 6.481414794921875, -0.096954345703125, 47.3931884765625, -59.56195068359375, 0.8225193023681641, 67.43023681640625, 21.172622680664062, 3.41357421875, 2.2264404296875, 73.89727783203125, 42.36444091796875, -29.679779052734375, 30.689300537109375, -8.090248107910156, 10.402610778808594, -32.610595703125, 16.396575927734375, 17.453948974609375, 34.70349884033203, -2.3835296630859375, 80.26104736328125, 72.11392211914062, 32.92524719238281, 16.845306396484375, -47.433746337890625, 1.0260753631591797, 25.231929779052734, 53.27976989746094, 5.0689849853515625, -5.2774658203125, 54.783721923828125, 45.492919921875, 21.879043579101562, -0.6435699462890625, 1.3533706665039062, 36.928035736083984, 39.79694366455078, -8.948471069335938, 31.337738037109375, -5.5865020751953125, 77.34100341796875, 0.0, 83.96218872070312, 39.5582160949707, -22.94036865234375, 40.33197021484375, -17.475908279418945, -5.085666656494141, 1.573638916015625, 2.4081192016601562, 25.8826904296875, -7.625419616699219, 47.246063232421875, 4.113525390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000186.npy"}
|
|
{"epoch": 0.38952879581151834, "step": 187, "batch_size": 128, "mean": 19.67531967163086, "std": 33.95091247558594, "min": -59.44740295410156, "p10": -21.713504028320305, "median": 17.706409454345703, "p90": 62.292776489257804, "max": 92.99745178222656, "pos_frac": 0.6875, "sample": [57.25457763671875, 60.587799072265625, 36.227294921875, 55.4666748046875, -48.873046875, 33.58953857421875, 72.883056640625, 5.884252548217773, -29.588302612304688, 88.43743896484375, 63.962646484375, 1.8656387329101562, -5.52142333984375, 6.938873291015625, 57.35890197753906, 61.771026611328125, 43.96763610839844, 13.538436889648438, 46.71305847167969, 41.612640380859375, 25.241607666015625, -8.089126586914062, 46.503509521484375, 0.4782257080078125, -8.558349609375, -35.378082275390625, 35.96098327636719, -28.21624755859375, -14.1214599609375, -25.90386962890625, -1.62164306640625, 63.51019287109375, 21.1968994140625, 17.540985107421875, -19.917633056640625, 40.02839660644531, -9.401336669921875, -8.111297607421875, 40.17051696777344, 44.77685546875, 33.81654357910156, -6.303466796875, 79.38494873046875, -3.2353668212890625, -11.0933837890625, 40.0560302734375, 70.505615234375, 5.2620849609375, -16.470062255859375, 35.891998291015625, 3.680419921875, 30.34001922607422, 17.87183380126953, 40.86531066894531, -43.6646728515625, -4.67864990234375, 40.601318359375, -6.8317718505859375, 46.096736907958984, 52.987579345703125, 24.711868286132812, -6.416343688964844, 56.69390869140625, -59.44740295410156, 92.99745178222656, 59.577606201171875, -44.97892761230469, -54.0751953125, 6.54510498046875, 36.6678466796875, 73.266845703125, -45.05390167236328, -4.861610412597656, -17.5477294921875, -9.9898681640625, -34.60296630859375, 56.00712585449219, 9.996650695800781, 59.594482421875, 8.6546630859375, 0.353515625, 12.068099975585938, -38.11674499511719, 4.556671142578125, 12.717117309570312, -0.49515533447265625, 16.726959228515625, 4.2665557861328125, -43.593994140625, 36.59136962890625, 25.894378662109375, 61.60223388671875, 57.54869079589844, 41.72883605957031, 34.8323974609375, 73.52735900878906, 0.0274658203125, 39.78631591796875, 48.96276092529297, 40.576507568359375, -4.260528564453125, 48.31150817871094, 88.979736328125, 13.3287353515625, 1.7591438293457031, 45.89691162109375, -0.09490966796875, 20.321365356445312, -18.763702392578125, 72.82357788085938, 45.345703125, 23.9447021484375, 12.842056274414062, 23.182334899902344, 7.754552841186523, 43.044403076171875, 69.40803527832031, -0.14585113525390625, 1.8303794860839844, 6.2792816162109375, -11.74188232421875, 42.405662536621094, 33.522605895996094, -6.765892028808594, 0.0, 67.42669677734375, -12.367889404296875, 21.624404907226562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000187.npy"}
|
|
{"epoch": 0.39162303664921466, "step": 188, "batch_size": 128, "mean": 19.767297744750977, "std": 33.90901565551758, "min": -74.71095275878906, "p10": -20.69364013671875, "median": 16.23664093017578, "p90": 63.81544647216797, "max": 89.65695190429688, "pos_frac": 0.6953125, "sample": [3.190685272216797, 23.94244384765625, 40.26220703125, 0.0, 13.77972412109375, 3.594451904296875, 6.908355712890625, 70.82778930664062, -3.4420204162597656, 44.64429473876953, 51.481353759765625, 4.242271423339844, 82.41485595703125, -7.868125915527344, 24.8697509765625, -24.02532196044922, 52.65936279296875, 19.286651611328125, 14.624923706054688, 88.73321533203125, 73.29998779296875, 49.295379638671875, -14.356536865234375, 0.5839958190917969, -8.560264587402344, -18.6875, -23.869674682617188, 8.654937744140625, -20.601959228515625, 47.30528259277344, 61.63482666015625, -16.356948852539062, 47.65223693847656, 25.898056030273438, -2.11834716796875, 39.388275146484375, 39.097412109375, 62.356842041015625, 54.76791000366211, 8.803726196289062, 12.8978271484375, 55.103668212890625, 8.244453430175781, 57.91119384765625, 78.7041015625, -63.0947265625, 13.69805908203125, 49.40399169921875, 21.852783203125, 40.19599914550781, 6.66046142578125, 46.729888916015625, -19.50684356689453, -3.85015869140625, 28.861053466796875, 76.822265625, 83.10958862304688, 8.018444061279297, 41.53192138671875, 77.673095703125, 37.25640869140625, 5.865531921386719, 38.0994987487793, 57.11149597167969, 6.782596588134766, -0.78704833984375, 15.017593383789062, 63.61390686035156, 14.076622009277344, 44.64442443847656, -1.021759033203125, 40.19830322265625, 89.65695190429688, -1.2488040924072266, -5.96099853515625, 2.486053466796875, 54.908538818359375, 64.76498413085938, -10.456878662109375, 22.276580810546875, -31.120269775390625, 48.63580322265625, 44.893280029296875, 58.981048583984375, -0.97955322265625, -5.893165588378906, 31.704681396484375, 52.06818389892578, 23.57293701171875, 1.04949951171875, -74.71095275878906, 25.0333251953125, -34.786407470703125, -27.33575439453125, 25.144378662109375, -20.520050048828125, -32.461265563964844, -27.08758544921875, 39.049617767333984, 21.38849639892578, 23.1593017578125, 0.6390609741210938, -35.405555725097656, -68.15643310546875, 47.51434326171875, 45.643798828125, 43.53449249267578, 56.71617889404297, 74.79751586914062, -6.549365997314453, -37.279632568359375, 1.1932048797607422, 19.15892791748047, 8.75927734375, 64.28570556640625, -11.80426025390625, 71.61572265625, -4.774501800537109, 0.073486328125, -6.8037109375, -1.19927978515625, 17.4556884765625, 0.0, -6.4058837890625, 32.71031188964844, 19.395030975341797, 3.656494140625, -20.907562255859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000188.npy"}
|
|
{"epoch": 0.393717277486911, "step": 189, "batch_size": 128, "mean": 23.14593505859375, "std": 27.969860076904297, "min": -47.629638671875, "p10": -10.026705932617187, "median": 18.280288696289062, "p90": 63.62325134277342, "max": 88.54104614257812, "pos_frac": 0.7890625, "sample": [35.28581237792969, 15.743133544921875, -47.629638671875, 10.425636291503906, 10.3201904296875, 10.580078125, 33.300193786621094, 84.92840576171875, 39.860206604003906, 22.216766357421875, 17.063507080078125, 39.5279541015625, 73.31707763671875, 40.631622314453125, 51.42308807373047, 47.641845703125, 58.74774169921875, -1.4562244415283203, -11.501571655273438, 31.44140625, 66.64962768554688, 69.14691162109375, -13.637794494628906, 77.4268798828125, 32.911468505859375, 10.958724975585938, 31.46734619140625, 29.817230224609375, 3.603546142578125, -5.44915771484375, -23.914276123046875, 2.1369400024414062, 22.77196502685547, 35.94956970214844, 1.77178955078125, 31.079940795898438, 17.137611389160156, 2.793741226196289, 11.656097412109375, 28.42962646484375, 1.2081050872802734, 71.211669921875, 54.83892822265625, 49.280853271484375, 53.4666748046875, 37.52606201171875, 16.52618408203125, -15.35845947265625, 58.939361572265625, -22.509552001953125, -26.0799560546875, -7.9210205078125, -27.09735107421875, 37.79534912109375, 4.441497802734375, 18.12799072265625, 32.30218505859375, 5.1595001220703125, 8.577911376953125, 12.190338134765625, 43.745391845703125, 13.239059448242188, 29.178466796875, -3.0337677001953125, 15.56085205078125, 22.46426773071289, 15.612947463989258, 11.277244567871094, 4.614269256591797, 17.752296447753906, 16.174156188964844, 9.970573425292969, -10.107231140136719, 52.15008544921875, -2.4151229858398438, 1.4857139587402344, 12.55438232421875, 3.7766036987304688, -14.118408203125, 14.316192626953125, 31.252731323242188, 24.19366455078125, 21.22564697265625, -0.48638916015625, 36.44548416137695, 44.606201171875, 76.09037780761719, -10.534820556640625, -0.8735561370849609, -0.1565074920654297, 11.455657958984375, 66.8287353515625, 22.06085205078125, -42.696807861328125, 0.77056884765625, 0.0, 88.54104614257812, 22.98699951171875, 38.785064697265625, 73.51164245605469, 39.69366455078125, 46.410888671875, 61.88433074951172, 17.69104766845703, 19.71648406982422, 0.0, 19.721282958984375, 83.1917724609375, -3.3162841796875, 67.136962890625, 28.651885986328125, 11.577239990234375, 60.84454345703125, 8.289663314819336, 18.432586669921875, -13.023040771484375, 46.658294677734375, 57.85371398925781, 49.9652214050293, 29.808746337890625, 24.78240966796875, 79.44818115234375, -7.845619201660156, 62.32623291015625, 45.80181884765625, 3.5076751708984375, -1.94366455078125, -9.992195129394531], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000189.npy"}
|
|
{"epoch": 0.3958115183246073, "step": 190, "batch_size": 128, "mean": 22.32888412475586, "std": 31.94244384765625, "min": -63.8111572265625, "p10": -16.73185577392578, "median": 21.865737915039062, "p90": 59.280023193359376, "max": 103.235107421875, "pos_frac": 0.765625, "sample": [-16.540748596191406, 13.570648193359375, 84.9744873046875, 36.73518371582031, -5.3065185546875, 14.993219375610352, -47.3773193359375, -3.334512710571289, -23.1439208984375, 0.55340576171875, 42.10734558105469, -48.95189666748047, -17.177772521972656, 58.4493408203125, 52.817718505859375, -2.1329345703125, 3.294097900390625, -13.31463623046875, 7.69903564453125, -45.909088134765625, -6.088470458984375, -12.13775634765625, -30.9383544921875, 55.14610290527344, 50.397830963134766, 4.7516632080078125, 42.1397705078125, -1.828338623046875, 35.893646240234375, 69.91891479492188, 93.19439697265625, -12.759185791015625, 29.7298583984375, 16.363754272460938, 52.262969970703125, -4.71380615234375, 57.58197021484375, 19.324535369873047, 46.356529235839844, 35.588226318359375, 53.38946533203125, -63.8111572265625, 8.628067016601562, 83.05085754394531, 50.78071594238281, 27.291351318359375, 52.3193359375, 17.39495086669922, 69.27346801757812, 85.44000244140625, 17.699722290039062, 7.276607513427734, 44.24304962158203, -1.1708831787109375, 30.758285522460938, 26.701194763183594, 5.031513214111328, -4.819122314453125, 5.761810302734375, -7.0727691650390625, -3.8165283203125, 51.38554382324219, 28.54547119140625, 23.232742309570312, 42.2127685546875, 25.717361450195312, 22.447998046875, -23.55462646484375, 17.738922119140625, 48.60791015625, 45.077728271484375, 35.07999801635742, 27.396484375, 0.0, 57.29327392578125, 63.65394592285156, 26.190204620361328, 30.908668518066406, 58.101654052734375, 15.873355865478516, 2.498046875, -0.103485107421875, 32.604705810546875, 38.90544128417969, 57.52801513671875, 22.6529541015625, 18.748077392578125, 44.109771728515625, 48.681671142578125, 9.410400390625, 16.616607666015625, 103.235107421875, 42.528953552246094, 59.25762939453125, -2.2133865356445312, -49.81121826171875, 11.270278930664062, -18.875091552734375, 62.92705535888672, -22.055404663085938, 43.537109375, 18.260772705078125, 25.1336669921875, 59.332275390625, 43.028472900390625, 35.977813720703125, 12.791519165039062, 6.147979736328125, 1.1607818603515625, 49.133758544921875, 4.265623092651367, 72.50985717773438, 27.703439712524414, 31.266571044921875, 5.904779434204102, 11.84002685546875, 3.616485595703125, 41.898597717285156, -25.589080810546875, 21.283477783203125, 21.268035888671875, 0.6753616333007812, 4.294523239135742, 42.471065521240234, -53.0167236328125, 69.177734375, 6.42620849609375, 67.23240661621094], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000190.npy"}
|
|
{"epoch": 0.39790575916230364, "step": 191, "batch_size": 128, "mean": 17.799724578857422, "std": 33.644500732421875, "min": -88.56781005859375, "p10": -20.68417510986328, "median": 12.162803649902344, "p90": 63.994451904296874, "max": 89.67630004882812, "pos_frac": 0.7421875, "sample": [51.559539794921875, 29.7825927734375, -27.793853759765625, -0.3459014892578125, 38.463584899902344, 73.45135498046875, 8.998626708984375, 13.622554779052734, 3.5211734771728516, -20.2247314453125, 1.079803466796875, 39.60688018798828, 14.628129959106445, 5.724082946777344, 0.037811279296875, -7.669189453125, 26.00999641418457, 0.445556640625, 1.6093978881835938, 18.412399291992188, 28.285293579101562, 40.29766845703125, 10.428955078125, 0.8422775268554688, -0.3526611328125, 13.276710510253906, -4.47088623046875, -46.040740966796875, 58.19757080078125, 39.00987243652344, 9.29791259765625, 3.1811180114746094, -12.977035522460938, 46.060546875, 5.743896484375, -49.842529296875, 43.780906677246094, -9.676315307617188, -88.56781005859375, -17.942901611328125, 16.693031311035156, 62.43017578125, 87.99162292480469, 12.451492309570312, -0.3120155334472656, 83.17857360839844, 12.67303466796875, -15.3472900390625, 6.497858047485352, 8.26422119140625, 8.335433959960938, 52.17889404296875, 87.85617065429688, 68.98404693603516, 9.640899658203125, 7.4528656005859375, 16.53741455078125, 67.95974731445312, 14.54241943359375, 9.755790710449219, 6.182744979858398, 3.797882080078125, 30.100683212280273, -16.28881072998047, 58.86279296875, 5.42938232421875, 80.23736572265625, 26.981704711914062, 16.9423828125, 3.3378257751464844, 4.249786376953125, 27.150962829589844, 10.765823364257812, -27.2464599609375, -1.5089340209960938, -48.52374267578125, 18.3409423828125, 20.19049072265625, 51.739959716796875, -28.2633056640625, 2.806976318359375, 64.3485107421875, 1.3036746978759766, 46.37725830078125, -35.227142333984375, 72.623779296875, 66.18841552734375, -25.44561767578125, 35.498443603515625, 71.88821411132812, 53.6409912109375, 58.571563720703125, 14.83258056640625, 1.3775577545166016, -1.3707809448242188, -7.26373291015625, 46.5445556640625, 27.142333984375, 83.881103515625, 31.216949462890625, 29.348663330078125, 6.335357666015625, 58.14412307739258, 37.28547668457031, 89.67630004882812, 62.69242858886719, 4.484764099121094, -18.899765014648438, -44.84877014160156, 15.040771484375, -9.876953125, -53.11572265625, 26.32242202758789, 22.059539794921875, 49.274810791015625, 58.44610595703125, -4.66552734375, -3.67083740234375, -21.756210327148438, 63.84271240234375, 32.615196228027344, 8.722564697265625, 11.874114990234375, -35.8101806640625, 0.0, 44.460845947265625, -20.195030212402344, 61.946563720703125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000191.npy"}
|
|
{"epoch": 0.4, "step": 192, "batch_size": 128, "mean": 17.461965560913086, "std": 34.065799713134766, "min": -88.21696472167969, "p10": -27.707627868652338, "median": 14.393331527709961, "p90": 63.36333465576172, "max": 87.21859741210938, "pos_frac": 0.6953125, "sample": [21.630294799804688, 48.102813720703125, 61.156005859375, -35.4503173828125, 65.26458740234375, 16.295928955078125, -18.198089599609375, 13.98431396484375, -9.152801513671875, 73.44677734375, 44.85125732421875, 15.03369140625, 47.895240783691406, 34.774261474609375, 55.956207275390625, -6.6924285888671875, -33.32343292236328, 43.6707763671875, -88.21696472167969, 77.720703125, -63.4464111328125, 72.03616333007812, -9.198226928710938, -49.337039947509766, 35.626068115234375, 48.1107177734375, -1.442474365234375, 8.107002258300781, 7.709663391113281, 50.45698547363281, -72.8013916015625, 25.319419860839844, 39.192718505859375, 52.215576171875, -0.311676025390625, 2.3679561614990234, -2.8645782470703125, 51.22601318359375, 9.500152587890625, 13.030181884765625, 13.575210571289062, -36.05096435546875, -7.860443115234375, 52.299285888671875, 56.569061279296875, -42.17301940917969, 33.38201904296875, -30.70013427734375, 0.0, 0.0, 2.8128662109375, -1.148834228515625, -34.891212463378906, 26.984619140625, -18.432785034179688, 49.35289001464844, 3.80389404296875, 42.188140869140625, -15.100372314453125, 0.739288330078125, 32.931678771972656, 25.950082778930664, -18.162139892578125, -15.418121337890625, 59.997314453125, 38.524566650390625, 34.86029052734375, 63.17668151855469, -1.0806999206542969, 57.301300048828125, -22.95989990234375, 4.1016082763671875, 25.548095703125, -9.694091796875, 18.84343719482422, 40.06542205810547, 36.53082275390625, 72.9482421875, 72.20462036132812, 8.093208312988281, 4.834434509277344, 63.798858642578125, -1.1453018188476562, -30.58734130859375, -26.473464965820312, 14.960113525390625, 10.908863067626953, -4.002838134765625, 46.50016784667969, 26.248573303222656, 0.55242919921875, 16.87591552734375, -32.37980651855469, 86.31365966796875, 9.403993606567383, 36.26322937011719, 1.7769775390625, -6.965587615966797, 68.30877685546875, 59.878265380859375, 10.161369323730469, 13.89080810546875, 4.162200927734375, -12.493171691894531, -1.1736793518066406, 17.497802734375, 37.87547302246094, 7.743888854980469, 87.21859741210938, 35.27952575683594, 49.30186462402344, -4.854034423828125, 37.19007110595703, 16.823150634765625, 17.05133056640625, 10.04421615600586, 50.52632141113281, 14.770946502685547, -34.281951904296875, 0.04779052734375, 3.7855987548828125, 73.839599609375, 26.418731689453125, 14.015716552734375, 15.625991821289062, 68.47396850585938, -2.54888916015625, 70.31085205078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000192.npy"}
|
|
{"epoch": 0.40209424083769635, "step": 193, "batch_size": 128, "mean": 18.330469131469727, "std": 32.62166213989258, "min": -58.010894775390625, "p10": -23.223577880859374, "median": 18.531082153320312, "p90": 59.299112701416014, "max": 82.87149047851562, "pos_frac": 0.6953125, "sample": [59.34093475341797, -4.542686462402344, -3.099761962890625, -4.3090362548828125, -8.322341918945312, 48.72259521484375, 34.063255310058594, 57.17024230957031, 20.606990814208984, 29.2291259765625, 31.472625732421875, 2.633974075317383, 13.427032470703125, 56.83177185058594, -9.31939697265625, 54.30792236328125, 0.703369140625, 11.548637390136719, -0.7600250244140625, 3.4131832122802734, 56.968505859375, -18.109130859375, 48.28944396972656, 5.545379638671875, 24.6292724609375, 30.230377197265625, 13.820648193359375, -45.344085693359375, 49.4588623046875, -3.193267822265625, -31.77569580078125, 3.472198486328125, 39.66839599609375, 17.28271484375, 46.42467498779297, -10.23260498046875, 53.177978515625, -13.847625732421875, 0.5481052398681641, 25.17125701904297, -22.35003662109375, -6.9735107421875, 26.378692626953125, 38.027069091796875, 6.1607818603515625, 15.4154052734375, 66.4473876953125, 32.019775390625, 54.66278076171875, 48.96697998046875, -9.180450439453125, 10.043609619140625, 43.921607971191406, 7.076622009277344, 55.61273193359375, 3.0859527587890625, 75.99396514892578, 25.359954833984375, -10.30487060546875, 32.86138916015625, -6.6197357177734375, 32.05555725097656, 50.732200622558594, -19.193206787109375, 82.1209716796875, -8.0169677734375, 26.013282775878906, 43.94940185546875, 18.727081298828125, -7.65447998046875, -58.010894775390625, -55.59999084472656, 21.470184326171875, 82.87149047851562, -35.87640380859375, 48.0892333984375, 62.561309814453125, 70.304931640625, 18.3350830078125, 75.21893310546875, 54.902374267578125, 23.6402587890625, 55.438507080078125, -44.54779052734375, 76.375732421875, 59.28118896484375, 9.083877563476562, 6.512641906738281, -2.65679931640625, 21.12860107421875, 23.3470458984375, -0.04944610595703125, 75.76351928710938, 58.300445556640625, 21.95207977294922, -0.69732666015625, 24.072265625, 6.118843078613281, 22.68988037109375, 53.590606689453125, 49.15484619140625, 66.978759765625, 9.178817749023438, -27.932174682617188, 25.676605224609375, -20.124038696289062, 77.14337158203125, -25.2618408203125, -48.502655029296875, 0.5159912109375, -3.0605850219726562, 24.740509033203125, -35.45965576171875, 25.033355712890625, 5.3900146484375, 15.682628631591797, 60.077301025390625, -32.340087890625, 10.6768798828125, -7.929840087890625, -18.4425048828125, -3.302276611328125, -48.13629150390625, 23.951416015625, 34.729949951171875, -31.039505004882812, 2.937835693359375, 51.709075927734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000193.npy"}
|
|
{"epoch": 0.4041884816753927, "step": 194, "batch_size": 128, "mean": 13.88785171508789, "std": 31.616161346435547, "min": -58.44902038574219, "p10": -23.143519592285156, "median": 7.902568817138672, "p90": 59.514045715332024, "max": 95.79611206054688, "pos_frac": 0.6328125, "sample": [95.79611206054688, -10.695404052734375, 41.13298034667969, -39.04084777832031, 18.41082763671875, 63.25933837890625, 31.130828857421875, -30.224288940429688, 2.628631591796875, -0.9472770690917969, 53.28840637207031, -19.7137451171875, -3.7590694427490234, 5.5121612548828125, 11.1156005859375, 1.2941093444824219, -5.088249206542969, 23.400894165039062, 56.945892333984375, 2.890838623046875, 17.15228271484375, -5.961055755615234, 21.744823455810547, 2.9280242919921875, 12.91326904296875, -6.078224182128906, 2.5714111328125, 24.805755615234375, -23.299667358398438, -31.42852783203125, -23.727432250976562, 21.56743621826172, -9.70071792602539, 15.97918701171875, -0.5183258056640625, 23.697235107421875, 79.83273315429688, 43.365135192871094, 8.652374267578125, 51.8178596496582, -14.163650512695312, 90.85784912109375, -6.3664703369140625, -16.591712951660156, -8.787521362304688, 44.91328430175781, -12.118606567382812, 62.70795440673828, 58.829437255859375, 9.18035888671875, -9.71990966796875, 21.04296875, 29.762481689453125, 29.338623046875, 87.30859375, 21.637908935546875, 9.043586730957031, 23.3104248046875, 16.744056701660156, -6.29010009765625, 10.12225341796875, 9.999542236328125, -3.14361572265625, 43.586700439453125, -40.312255859375, -13.47418212890625, 3.5712127685546875, -49.256378173828125, 24.23486328125, -13.1387939453125, 80.7540283203125, -3.449188232421875, -24.428253173828125, 70.9853515625, -0.47137451171875, 66.23159790039062, -2.1130752563476562, -17.083106994628906, 32.057952880859375, 51.2186279296875, -23.07659912109375, 10.82177734375, 50.840301513671875, 61.11146545410156, -5.195466995239258, -44.98382568359375, 42.81367492675781, -0.18715667724609375, 50.9232177734375, 62.8956298828125, 31.81396484375, -33.997711181640625, 21.209136962890625, -6.624946594238281, 5.003150939941406, 5.854705810546875, 6.1387939453125, -58.44902038574219, 84.7579345703125, 13.247329711914062, -8.876251220703125, 20.44219970703125, 1.9561920166015625, 7.152763366699219, -40.32670593261719, 8.885726928710938, -11.697052001953125, 43.0076904296875, 2.6155548095703125, -12.830841064453125, 13.343002319335938, 75.10299682617188, 44.506011962890625, 6.8846435546875, 27.163284301757812, 22.8056640625, 52.0882568359375, 2.1467113494873047, -24.457420349121094, 3.578582763671875, -0.9052734375, -5.91619873046875, 52.19097900390625, -11.605506896972656, 2.6069297790527344, -13.9986572265625, 36.236572265625, 30.445953369140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000194.npy"}
|
|
{"epoch": 0.406282722513089, "step": 195, "batch_size": 128, "mean": 13.843399047851562, "std": 31.355363845825195, "min": -60.338592529296875, "p10": -22.442285156249998, "median": 7.364484786987305, "p90": 55.771702575683584, "max": 98.228515625, "pos_frac": 0.6328125, "sample": [-8.41363525390625, -21.827728271484375, -1.8628997802734375, 53.9986572265625, 57.42835998535156, 49.929351806640625, -27.59832763671875, 0.9794921875, 44.1923828125, 7.298496246337891, 29.067138671875, 8.352912902832031, 2.1549644470214844, -7.639316558837891, -1.2738494873046875, 21.094085693359375, -9.558563232421875, -6.67431640625, 15.535858154296875, 41.3470458984375, -9.604507446289062, 12.311065673828125, 49.785797119140625, -2.4302597045898438, -15.539794921875, -1.6590347290039062, 1.1620712280273438, 0.43682861328125, -27.23553466796875, 7.5498809814453125, -10.01300048828125, 45.8509521484375, -4.276031494140625, -0.58642578125, 2.857940673828125, 3.391387939453125, 66.36927795410156, 48.341064453125, -3.040679931640625, -4.6053009033203125, -30.808303833007812, -11.542388916015625, 11.033882141113281, 7.4670867919921875, 55.06170654296875, 5.723297119140625, 81.66366577148438, -26.3031005859375, -15.9605712890625, 10.8492431640625, -35.52915954589844, 25.245330810546875, 30.639694213867188, 45.01167297363281, -52.727142333984375, 18.886688232421875, -7.129486083984375, -54.01879119873047, 61.882850646972656, -4.158050537109375, 22.3873291015625, -60.338592529296875, 0.7603302001953125, 43.5179443359375, -13.816429138183594, -23.876251220703125, -47.644020080566406, -11.373260498046875, -2.5803375244140625, 36.74774169921875, 63.353546142578125, 28.79974365234375, 45.085540771484375, 1.1157302856445312, 29.183349609375, 76.54498291015625, -8.91522216796875, -1.3604736328125, 23.43396759033203, 13.43115234375, 48.17357635498047, 53.18145751953125, 25.54449462890625, 36.7379150390625, -3.6752166748046875, -2.2505264282226562, 3.177276611328125, -28.07012939453125, -1.0562210083007812, 5.193511962890625, 2.580810546875, 98.228515625, 45.16792297363281, -0.4701995849609375, 49.024009704589844, 19.4814453125, 74.14682006835938, 2.58843994140625, 22.218002319335938, 73.55429077148438, -19.980323791503906, 38.522308349609375, 10.410980224609375, 27.452301025390625, 4.780364990234375, 7.430473327636719, 8.376861572265625, 54.1451416015625, 40.998046875, 59.48497009277344, 83.22296142578125, 29.82635498046875, -40.59356689453125, 12.220611572265625, -19.242515563964844, 6.5001983642578125, 38.993804931640625, -49.056884765625, 5.72283935546875, 37.80940246582031, 10.6376953125, -11.293434143066406, -2.17572021484375, 29.495849609375, 60.15235900878906, 39.76092529296875, 58.973297119140625, -9.437286376953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000195.npy"}
|
|
{"epoch": 0.4083769633507853, "step": 196, "batch_size": 128, "mean": 20.750043869018555, "std": 30.06943702697754, "min": -52.7637939453125, "p10": -12.68731002807617, "median": 16.687698364257812, "p90": 61.441018676757814, "max": 96.00430297851562, "pos_frac": 0.7734375, "sample": [1.5490856170654297, 5.23858642578125, 33.149452209472656, 1.136474609375, 15.584817886352539, 7.821624755859375, -18.996490478515625, 8.73097038269043, 45.91050720214844, 1.4575653076171875, 2.9880752563476562, 66.45628356933594, 27.621368408203125, 37.693389892578125, 74.85736083984375, -22.124114990234375, 4.788055419921875, 3.77117919921875, 12.361831665039062, 17.05816650390625, 33.2392578125, 4.73699951171875, 39.4864501953125, 58.12055969238281, 0.4311790466308594, 18.73614501953125, 1.56036376953125, 0.2528667449951172, -40.61595153808594, 54.468658447265625, 88.06869506835938, 72.23320770263672, 44.55206298828125, 32.474609375, -1.8626708984375, 41.733306884765625, 6.5569000244140625, -32.848175048828125, 24.915069580078125, 35.897003173828125, -24.839675903320312, 72.41302490234375, 30.231781005859375, -8.606201171875, 8.846118927001953, 9.52410888671875, 18.582763671875, 80.19085693359375, 14.764617919921875, 24.293548583984375, -23.926422119140625, -0.18300437927246094, -0.1546783447265625, 2.143932342529297, -1.7820816040039062, 62.6361083984375, 55.48095703125, 3.4956207275390625, 5.79107666015625, 20.0162353515625, 35.203704833984375, 8.379913330078125, 42.944496154785156, 4.83465576171875, 18.919097900390625, 60.8271484375, -24.857418060302734, 17.320159912109375, 0.876068115234375, -15.27203369140625, 10.136955261230469, 17.213043212890625, 0.37100982666015625, 47.77384948730469, 34.628448486328125, 52.9342041015625, 0.750335693359375, 16.056427001953125, 41.648101806640625, 52.209381103515625, 24.06507110595703, -37.76756286621094, 25.76854705810547, -3.2878646850585938, -9.62054443359375, 49.307373046875, 31.92333984375, 56.73138427734375, 25.205291748046875, 1.948089599609375, 42.4222412109375, -52.7637939453125, 16.317230224609375, 86.22848510742188, 21.4937744140625, -5.649360656738281, -12.373908996582031, -9.169448852539062, 62.16949462890625, 60.433013916015625, 28.38128662109375, -13.4185791015625, 6.005767822265625, 96.00430297851562, 70.34487915039062, 9.834196090698242, -22.872055053710938, 53.73236846923828, 61.9520263671875, 57.201416015625, 45.42701721191406, 83.65118408203125, 55.814422607421875, 61.222015380859375, -5.57623291015625, -11.339683532714844, -3.653064727783203, 48.01116943359375, 30.4244384765625, 29.50360107421875, -1.8897705078125, -6.802986145019531, -33.409393310546875, 33.3173828125, 8.429534912109375, 21.363616943359375, -3.0894775390625, 3.0483322143554688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000196.npy"}
|
|
{"epoch": 0.41047120418848165, "step": 197, "batch_size": 128, "mean": 26.823715209960938, "std": 32.09828567504883, "min": -65.20166015625, "p10": -8.896005249023435, "median": 24.858084678649902, "p90": 64.73815002441405, "max": 97.975830078125, "pos_frac": 0.8046875, "sample": [-34.14057922363281, 10.92279052734375, -6.16033935546875, 42.110748291015625, 57.884613037109375, -8.267715454101562, 49.896026611328125, 19.141311645507812, 7.003566741943359, 62.42254638671875, 50.59368896484375, 49.000823974609375, -10.362014770507812, 26.929779052734375, 47.35456848144531, -24.7873477935791, 97.975830078125, 79.42156219482422, 19.3173828125, 79.3741455078125, -0.25608253479003906, 57.44389343261719, 72.81554412841797, 18.192840576171875, 35.53186798095703, 20.614517211914062, 54.212249755859375, 67.528564453125, 65.663330078125, 64.11458587646484, 55.95273208618164, 40.68394470214844, -37.112884521484375, 42.922203063964844, 22.02838134765625, 49.448524475097656, 51.32549285888672, -1.8514938354492188, 15.069122314453125, 71.44757080078125, 20.48700714111328, 41.2984619140625, -65.20166015625, 62.89373779296875, -37.6527099609375, 24.125213623046875, 78.20315551757812, -1.0733566284179688, 2.0440673828125, -0.777252197265625, 39.93455505371094, 22.405792236328125, 11.21527099609375, -15.08831787109375, 11.415565490722656, 0.5369071960449219, 55.99171447753906, 11.833251953125, 64.34164428710938, 51.90325927734375, 47.722869873046875, 91.61505126953125, 3.8448829650878906, 52.6771240234375, -31.1614990234375, -2.855487823486328, 47.60255432128906, 58.53582000732422, 25.21278190612793, 11.32635498046875, 12.496070861816406, 34.074951171875, -2.2077102661132812, 56.87223815917969, 1.2053070068359375, 32.81135559082031, 16.419403076171875, 14.489471435546875, -30.113304138183594, 3.64605712890625, 12.298095703125, 15.046676635742188, 30.82513427734375, 45.925933837890625, -24.37176513671875, 44.2640380859375, 4.73199462890625, 33.81676483154297, 39.7862548828125, 4.496551513671875, 49.380096435546875, -0.6251220703125, 64.240234375, 38.2977294921875, 48.608001708984375, 8.21917724609375, 24.503387451171875, -37.006683349609375, 12.175811767578125, -43.53485107421875, 10.572998046875, 31.714492797851562, -4.891448974609375, 48.939208984375, 62.639892578125, 31.807464599609375, 29.19195556640625, 22.018417358398438, 61.20085144042969, 34.258544921875, 89.94931030273438, 82.72529602050781, 6.86834716796875, 4.2006072998046875, 22.143524169921875, 15.66845703125, 68.56890869140625, 0.6755542755126953, 53.00307846069336, 13.139747619628906, 0.0, 93.01708984375, 21.690582275390625, 29.55474853515625, 25.95270538330078, -33.895660400390625, 41.115936279296875, -3.9293899536132812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000197.npy"}
|
|
{"epoch": 0.41256544502617803, "step": 198, "batch_size": 128, "mean": 20.420949935913086, "std": 37.42236328125, "min": -89.18133544921875, "p10": -19.47696762084961, "median": 17.20923614501953, "p90": 72.91470489501953, "max": 99.835693359375, "pos_frac": 0.734375, "sample": [55.297119140625, 18.79779052734375, 7.18994140625, -3.9216766357421875, 47.86187744140625, 1.585418701171875, 49.01995849609375, 65.49432373046875, 28.873504638671875, 32.96893310546875, -12.713836669921875, -6.9859466552734375, 34.1983642578125, 30.227386474609375, 9.912227630615234, -9.763671875, 78.57283020019531, 39.20458984375, 34.511474609375, 7.181854248046875, 72.61419677734375, -70.83428955078125, -38.7379150390625, 73.61589050292969, 74.6910400390625, 96.70721435546875, -63.15625, 12.860321044921875, 8.829437255859375, 17.0244140625, 74.2652587890625, 55.05345916748047, 55.4273681640625, 45.88104248046875, 6.59503173828125, 26.979568481445312, -29.377349853515625, 99.835693359375, -41.01788330078125, 91.6668701171875, -2.13800048828125, 0.351348876953125, 45.5693359375, 53.38108444213867, -23.24810791015625, 0.5344429016113281, 83.70587158203125, -19.563148498535156, -16.709964752197266, 5.532417297363281, 21.367523193359375, 28.435638427734375, -2.112285614013672, 30.333648681640625, 15.087310791015625, 58.970123291015625, 71.55276489257812, 2.4282302856445312, 20.73291015625, 52.17822265625, 12.986618041992188, 19.9075927734375, -6.262306213378906, 53.67059326171875, 43.280548095703125, 24.959930419921875, 1.9084243774414062, 81.26681518554688, 25.021011352539062, 13.767936706542969, 14.42474365234375, -56.524261474609375, -16.077301025390625, 22.631744384765625, -89.18133544921875, -17.08270263671875, 17.394058227539062, 59.6761474609375, -14.028450012207031, 53.8970947265625, 8.667953491210938, 54.04011535644531, 46.19697570800781, -17.30279541015625, 53.676292419433594, -2.6468276977539062, 20.89008331298828, 67.53176879882812, -52.11463928222656, 12.80084228515625, 0.0, 48.1539306640625, 10.914840698242188, -13.3328857421875, -19.440032958984375, 7.13128662109375, 5.813211441040039, 34.787330627441406, 79.02203369140625, 2.901763916015625, 83.37103271484375, 1.032684326171875, 58.29603576660156, 40.61248779296875, 25.676544189453125, 40.90233612060547, 29.508941650390625, 3.272174835205078, 0.321807861328125, -84.0111083984375, 81.69869995117188, -19.408721923828125, 2.8228721618652344, -8.568016052246094, -4.680259704589844, -5.843017578125, 77.97015380859375, 24.45123291015625, 43.79840087890625, 6.96533203125, 19.784698486328125, -22.113037109375, 4.86358642578125, -24.704076766967773, 15.428445816040039, 66.12109375, 60.3189697265625, -0.16082763671875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000198.npy"}
|
|
{"epoch": 0.41465968586387436, "step": 199, "batch_size": 128, "mean": 21.234352111816406, "std": 32.56523132324219, "min": -54.532501220703125, "p10": -16.925830841064453, "median": 15.115875244140625, "p90": 67.19842834472657, "max": 100.87152099609375, "pos_frac": 0.7421875, "sample": [12.530731201171875, 31.81005859375, 49.17681884765625, 53.01719665527344, 75.16200256347656, 50.674461364746094, 50.441162109375, 1.4459609985351562, 66.99026489257812, 15.07403564453125, 97.14309692382812, 53.42713165283203, 40.52764892578125, 46.705413818359375, 87.8062744140625, -44.9266357421875, 33.512908935546875, 46.381561279296875, 7.528837203979492, -1.8090095520019531, 1.8359146118164062, 2.3586807250976562, -1.8082828521728516, 3.5834693908691406, 3.374298095703125, 25.871414184570312, 13.287330627441406, -0.23731231689453125, 18.547561645507812, -3.92779541015625, 30.027664184570312, 33.802330017089844, 47.05628204345703, 7.0650634765625, 26.621726989746094, 32.59730529785156, -35.73640441894531, -7.010261535644531, 54.346595764160156, 12.017425537109375, 9.421157836914062, 50.9510498046875, 11.869140625, 14.999406814575195, 4.3363494873046875, 68.86114501953125, -2.70751953125, -3.7594146728515625, 1.1638946533203125, -40.675872802734375, 51.11759948730469, 57.291778564453125, 9.652374267578125, 76.94430541992188, -6.4484405517578125, 21.845108032226562, 12.815032958984375, 20.049285888671875, 68.06074523925781, -27.089860916137695, -34.553741455078125, 57.717193603515625, 60.857513427734375, 24.639434814453125, -17.63909149169922, -33.7108039855957, 41.323158264160156, 11.17596435546875, -16.620147705078125, 26.329208374023438, 83.90811157226562, 0.6142730712890625, 3.5310726165771484, 71.1939697265625, 12.221115112304688, -0.8616180419921875, 35.201019287109375, 21.371414184570312, -1.4548263549804688, 19.738006591796875, 64.20516967773438, 9.686309814453125, 67.68414306640625, -9.227935791015625, 64.65078735351562, 69.1490478515625, -48.030433654785156, 52.53411865234375, 59.1488037109375, 26.085052490234375, -24.201507568359375, -14.555511474609375, 40.864227294921875, 69.88540649414062, -8.473220825195312, 19.371856689453125, -0.1283111572265625, 12.22100830078125, 22.021530151367188, 32.336517333984375, 20.02911376953125, -16.427734375, 15.15771484375, 10.256683349609375, 42.7396240234375, -54.532501220703125, -38.62629699707031, 0.48879051208496094, 12.227371215820312, 26.504364013671875, 13.725807189941406, 10.625701904296875, 57.897491455078125, -3.1978912353515625, 53.43461608886719, 15.302154541015625, 69.5248794555664, 0.1903076171875, -1.7517471313476562, 48.05327606201172, 6.74371337890625, -34.2906494140625, -0.1091766357421875, -11.946746826171875, 100.87152099609375, 60.85333251953125, 31.576095581054688, -22.52105712890625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000199.npy"}
|
|
{"epoch": 0.4167539267015707, "step": 200, "batch_size": 128, "mean": 17.95825958251953, "std": 39.68683624267578, "min": -80.25822448730469, "p10": -35.36367492675781, "median": 13.548942565917969, "p90": 72.11383666992187, "max": 127.30682373046875, "pos_frac": 0.6796875, "sample": [54.74530029296875, -44.844268798828125, 92.510009765625, 17.11219024658203, -44.662109375, 10.729499816894531, -36.04487609863281, 47.25726318359375, -8.24847412109375, 5.028240203857422, -1.4127960205078125, 60.745361328125, 20.65960693359375, -1.4849700927734375, -70.10980224609375, 66.2054443359375, 46.51026153564453, -13.2572021484375, 15.814788818359375, 91.4691162109375, -51.132965087890625, -35.05224609375, 84.06228637695312, 23.521034240722656, 81.28515625, 56.81207275390625, -18.142662048339844, -5.143058776855469, 50.349517822265625, -4.2935791015625, -29.2486572265625, -58.853546142578125, 16.91939353942871, -37.863525390625, 13.89117431640625, 16.659454345703125, -39.05956268310547, 55.5640869140625, 40.32623291015625, -14.55938720703125, 0.0, 39.660179138183594, 34.5992431640625, 26.862022399902344, -6.826499938964844, -61.644317626953125, 71.86285400390625, 7.12457275390625, 100.02761840820312, -2.4195404052734375, 16.7568359375, 12.515625, 43.001708984375, 4.957511901855469, -15.353057861328125, -8.80926513671875, 19.218711853027344, -58.93357849121094, 77.71533203125, 2.307220458984375, 43.63665771484375, 7.360107421875, 0.031749725341796875, 28.08978271484375, 64.22250366210938, 46.26443862915039, 11.068859100341797, 72.699462890625, -36.915008544921875, 50.35040283203125, 79.92286682128906, 42.635009765625, -31.80267333984375, 39.988319396972656, 40.255340576171875, 19.08929443359375, 1.51739501953125, -8.448394775390625, 9.6014404296875, 56.41790771484375, -47.31866455078125, 77.1455078125, 127.30682373046875, -0.305145263671875, 11.628173828125, 37.927978515625, -6.5033111572265625, 45.83003616333008, -80.25822448730469, 49.461761474609375, 10.421524047851562, 94.51275634765625, -20.492347717285156, 67.054931640625, 48.0648193359375, 68.4381103515625, 5.2213897705078125, 79.8380126953125, -18.28192138671875, -28.36151123046875, 10.950927734375, 24.471832275390625, 58.77850341796875, 34.638694763183594, -12.68505859375, 59.33613586425781, 13.206710815429688, -35.07173156738281, -12.44512939453125, 1.7164306640625, 9.30535888671875, 49.862648010253906, 38.30560302734375, 81.6240234375, 25.182052612304688, 33.9888916015625, -4.13970947265625, -10.671134948730469, 49.01637268066406, 2.0302696228027344, 5.8380889892578125, -1.4605741500854492, 14.521347045898438, 3.67437744140625, 3.3362655639648438, 13.934028625488281, 24.321197509765625, 2.3874282836914062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000200.npy"}
|
|
{"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 22.49065399169922, "std": 32.54866027832031, "min": -59.05218505859375, "p10": -21.615504455566406, "median": 22.834529876708984, "p90": 66.88967895507812, "max": 107.88229370117188, "pos_frac": 0.7265625, "sample": [-26.080230712890625, -12.941070556640625, 40.557090759277344, 57.82611083984375, 43.384796142578125, 33.2000732421875, 28.44189453125, 27.26446533203125, 19.765167236328125, 48.33826446533203, 61.54991149902344, 3.191155433654785, -4.738304138183594, 21.88873291015625, -21.136505126953125, -30.787994384765625, 41.7181396484375, 71.93804931640625, 48.56224060058594, 26.806640625, 15.170562744140625, 43.22288513183594, 57.34381866455078, 0.0, 49.21363830566406, 25.86572265625, 55.35267639160156, 61.539886474609375, 16.08130645751953, 64.83917236328125, 73.6864013671875, 42.03228759765625, 27.615333557128906, -4.3241424560546875, 0.0, -53.4544677734375, 5.383674621582031, -1.686279296875, 6.418476104736328, 5.1511383056640625, 28.65240478515625, 33.16168212890625, 54.47740173339844, 54.466651916503906, -28.244308471679688, 11.060989379882812, 27.307701110839844, 45.330474853515625, -13.558692932128906, 36.77643585205078, 1.0528182983398438, 22.91046905517578, 20.4954833984375, 41.3154296875, -0.6645622253417969, 24.343917846679688, 30.09931182861328, 65.52830505371094, 30.99847412109375, 51.851531982421875, -3.8096923828125, 41.91015625, 71.45065307617188, 11.4920654296875, 69.2080078125, 49.938453674316406, 28.160125732421875, 0.0, 9.65667724609375, -59.05218505859375, 14.7589111328125, 10.782852172851562, 0.610748291015625, 89.4056396484375, 34.36566162109375, 22.758590698242188, 5.351596832275391, 7.575141906738281, -10.7044677734375, 25.62457275390625, 21.0286865234375, 39.90777587890625, -15.399925231933594, -12.151580810546875, -0.315887451171875, 72.92027282714844, -6.21466064453125, 56.5526123046875, -23.675567626953125, 107.88229370117188, 73.05419921875, 35.0599365234375, 43.820220947265625, -4.87646484375, -27.78057861328125, -13.058685302734375, 8.890907287597656, 69.12326049804688, 11.158557891845703, 68.0665283203125, 57.6097412109375, 51.875484466552734, -44.46307373046875, 52.367584228515625, 10.700714111328125, -5.529083251953125, -22.737289428710938, 46.83929443359375, 68.9796142578125, -42.14042663574219, 15.461471557617188, -0.693634033203125, 15.17645263671875, 6.481170654296875, -1.3141403198242188, 84.47320556640625, 5.3267364501953125, 30.15198516845703, 33.702735900878906, 66.38531494140625, 10.976322174072266, -40.4927978515625, 9.880416870117188, 71.51165771484375, -4.785896301269531, 30.588470458984375, -22.733169555664062, -33.833160400390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000201.npy"}
|
|
{"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 21.239116668701172, "std": 37.250240325927734, "min": -80.9591064453125, "p10": -19.575634002685547, "median": 14.641395568847656, "p90": 72.3318862915039, "max": 85.39950561523438, "pos_frac": 0.703125, "sample": [80.58306884765625, 25.21221160888672, 14.01959228515625, 15.533493041992188, 69.8759765625, 2.2695045471191406, -15.093475341796875, 72.15763854980469, -41.93043518066406, -10.96466064453125, -1.9195556640625, 77.81338500976562, 24.41131591796875, 55.72491455078125, 52.80220031738281, -1.0269927978515625, -17.857467651367188, 53.309539794921875, 4.814155578613281, 80.79660034179688, 23.116405487060547, 51.479888916015625, -7.631011962890625, 13.1575927734375, 1.6219024658203125, 8.2808837890625, 81.7017822265625, 1.2308568954467773, 46.61968994140625, 10.06658935546875, -30.812232971191406, 23.62188720703125, -17.7239990234375, 73.5657958984375, 12.221206665039062, 56.54750061035156, 51.671295166015625, 85.39950561523438, 9.083045959472656, -44.128875732421875, 2.7185897827148438, 80.52459716796875, 48.42127990722656, 48.077423095703125, 25.466827392578125, 50.03759765625, 63.65106201171875, 18.586746215820312, 55.71588134765625, -2.0867767333984375, -60.24507141113281, 44.806739807128906, 65.99015808105469, -11.436203002929688, -55.85052490234375, 40.216033935546875, 41.6734619140625, -11.29852294921875, 22.74334716796875, -20.51335906982422, 22.331527709960938, 46.76780700683594, 76.29624938964844, -19.12481689453125, -5.2581787109375, -17.703887939453125, 82.27951049804688, -3.594207763671875, 36.894073486328125, 3.9198150634765625, 1.2446746826171875, 63.838134765625, 56.255950927734375, 3.0985107421875, 45.196990966796875, 56.05865478515625, 56.49640655517578, 28.34906005859375, 67.57560729980469, 61.29302978515625, 0.0, 71.61227416992188, 1.0506858825683594, 7.95751953125, 14.323944091796875, -1.1282157897949219, 0.42685699462890625, -12.850814819335938, 6.078010559082031, -26.260498046875, 47.20489501953125, 56.116668701171875, 4.76141357421875, 14.848312377929688, 76.06381225585938, 40.822776794433594, -19.173751831054688, -22.468505859375, -9.57464599609375, -24.47509765625, -80.9591064453125, 72.73846435546875, 31.98779296875, 52.85813903808594, 32.1080322265625, -14.228271484375, -15.675018310546875, 14.434478759765625, 29.718917846679688, 2.444375991821289, 37.818756103515625, 62.191925048828125, -36.70269775390625, 11.373321533203125, 50.643768310546875, -79.28604125976562, 6.992828369140625, 49.4292106628418, -39.983673095703125, -13.524833679199219, -6.253364562988281, 64.50553894042969, 3.6249923706054688, 3.225139617919922, -0.6271438598632812, 79.32666015625, -8.018875122070312, 84.0731201171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000202.npy"}
|
|
{"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 16.912290573120117, "std": 28.84592056274414, "min": -63.82371520996094, "p10": -16.71855316162109, "median": 12.6558198928833, "p90": 55.50376281738281, "max": 94.47232055664062, "pos_frac": 0.703125, "sample": [2.2010364532470703, 9.643798828125, 38.241180419921875, 25.712181091308594, 39.76141357421875, 30.730087280273438, 12.781625747680664, 20.893417358398438, 27.364013671875, 54.94227600097656, 56.99188232421875, 22.004104614257812, 2.8708019256591797, 33.597869873046875, 34.83674621582031, 35.551025390625, -5.45831298828125, -19.37366485595703, 13.820289611816406, 9.121063232421875, 12.324752807617188, 0.0, -33.93035888671875, 45.95824432373047, -19.994430541992188, 10.726478576660156, -21.74530029296875, 27.83489990234375, -3.5277099609375, -1.39874267578125, 21.58477783203125, 37.743377685546875, 30.61016845703125, 63.65643310546875, 42.60106658935547, -5.737314224243164, 23.198455810546875, 88.36483764648438, 17.15875244140625, -10.75933837890625, 49.96864318847656, 67.154296875, 19.4132080078125, 46.895538330078125, 1.5607337951660156, -3.1247711181640625, -3.6322708129882812, 45.298858642578125, -0.387969970703125, 44.90003204345703, 2.492818832397461, 17.255447387695312, 22.573944091796875, 32.913169860839844, 94.47232055664062, -45.7044677734375, 40.246002197265625, -3.1124420166015625, 56.67767333984375, 59.63330078125, 5.5329132080078125, 34.41368103027344, 48.706573486328125, -5.638343811035156, -34.50701141357422, 65.2772216796875, -32.916473388671875, 63.97564697265625, 12.46126937866211, 1.77374267578125, 16.171905517578125, 0.01413726806640625, -63.82371520996094, 50.36663818359375, -7.96514892578125, -2.4283924102783203, -16.276824951171875, 2.63018798828125, -25.690704345703125, 5.288352966308594, 27.450347900390625, 7.5066680908203125, -43.906005859375, 42.98779296875, 53.69171905517578, 12.530014038085938, 41.19633483886719, 16.88836669921875, 74.34500122070312, 50.63922119140625, 10.633255004882812, -10.96954345703125, -5.779632568359375, 45.843597412109375, 5.185445785522461, -6.0883636474609375, 30.60162353515625, -4.1407318115234375, 55.100006103515625, 10.371810913085938, 45.47673034667969, 3.854705810546875, -12.58038330078125, 10.06787109375, 11.780046463012695, -39.033782958984375, -2.1295166015625, -10.788162231445312, 13.7017822265625, 69.45114135742188, 0.0, 10.80328369140625, 17.12958526611328, 21.067962646484375, -0.5201549530029297, -4.779388427734375, 18.076248168945312, 59.048370361328125, 6.546257019042969, -1.0878143310546875, 32.727203369140625, -41.71807861328125, 7.3404083251953125, 56.44586181640625, 30.544967651367188, -17.749252319335938, 11.251750946044922, 21.997007369995117], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000203.npy"}
|
|
{"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 24.392196655273438, "std": 33.55776596069336, "min": -64.73153686523438, "p10": -18.20499572753906, "median": 17.36118793487549, "p90": 68.31260223388671, "max": 111.84573364257812, "pos_frac": 0.7890625, "sample": [92.1717529296875, 51.67518615722656, 80.7869873046875, 8.6197509765625, 10.508750915527344, 67.75399780273438, 4.598766326904297, 39.57231903076172, 4.1133880615234375, 79.8070068359375, 12.07080078125, 6.4925537109375, 54.19775390625, -17.929779052734375, -31.820556640625, -64.73153686523438, 19.511917114257812, -27.003814697265625, 7.768165588378906, 21.028594970703125, 67.08090209960938, 7.986473083496094, 2.9720458984375, 52.52173614501953, 21.7579345703125, 4.3194580078125, 23.09233856201172, -21.502059936523438, 15.093399047851562, 30.781982421875, 27.739288330078125, 111.84573364257812, 6.8853302001953125, -23.179153442382812, 44.855560302734375, -13.235107421875, 17.293481826782227, 96.0330810546875, -2.9104766845703125, 62.49613952636719, 69.61601257324219, 8.835762023925781, 35.65234375, -9.34906005859375, 52.545562744140625, 56.91558837890625, 7.2902069091796875, 76.45379638671875, 63.346099853515625, 85.36001586914062, 54.240821838378906, 41.84332275390625, 1.7138671875, 48.70159912109375, 26.357948303222656, -4.191989898681641, 51.076690673828125, 36.137367248535156, 21.60919952392578, -20.64453125, 25.796478271484375, 53.869537353515625, 21.310791015625, -13.448013305664062, -4.67254638671875, 81.6197509765625, 17.236236572265625, 66.14495849609375, -52.615966796875, 58.779693603515625, -5.975038528442383, 61.60211181640625, -1.21759033203125, -9.070884704589844, 5.867768287658691, 13.534706115722656, 28.039203643798828, 63.47203826904297, -4.630086898803711, 17.42889404296875, 7.305580139160156, -3.7869873046875, -40.20054626464844, 13.48529052734375, 31.9256591796875, -3.5440673828125, 51.410552978515625, 14.54718017578125, 64.815673828125, 54.51136779785156, 75.15032958984375, 12.591873168945312, -35.978668212890625, 54.01153564453125, -22.352643966674805, 4.91455078125, 59.672821044921875, -26.82830810546875, 16.018402099609375, 50.425537109375, 7.317253112792969, -0.353546142578125, 8.63250732421875, 29.608673095703125, 48.36521911621094, 77.56207275390625, 26.27155303955078, 15.706642150878906, 38.833343505859375, 5.688629150390625, 44.34873962402344, 0.2108612060546875, 46.870697021484375, 54.541290283203125, 11.101959228515625, 23.722579956054688, 32.96928405761719, 2.004230499267578, 2.7621536254882812, 15.204498291015625, -21.219276428222656, 14.033210754394531, 70.98680114746094, 28.31053924560547, -18.84716796875, 92.52825927734375, 6.905387878417969, 2.33709716796875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000204.npy"}
|
|
{"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 22.578981399536133, "std": 34.152652740478516, "min": -55.3402099609375, "p10": -19.988572692871095, "median": 19.658763885498047, "p90": 66.34969177246093, "max": 110.142822265625, "pos_frac": 0.7734375, "sample": [21.109100341796875, 0.0, 1.398712158203125, -0.15766143798828125, 13.661201477050781, 58.629638671875, -19.890625, -14.836395263671875, -20.217117309570312, 0.22559165954589844, 110.142822265625, 49.329261779785156, 4.978961944580078, 9.274772644042969, -33.287200927734375, 12.19607925415039, 84.23785400390625, 76.63555908203125, 55.688873291015625, 7.358489990234375, 4.022247314453125, 16.195823669433594, -8.8951416015625, 57.5386962890625, 9.22796630859375, 0.4791259765625, 56.904327392578125, 11.123832702636719, 76.569091796875, 30.392654418945312, 24.671783447265625, 35.359092712402344, 2.398651123046875, -5.356597900390625, 10.287384033203125, 6.239736557006836, 39.675071716308594, 11.93023681640625, 33.50238800048828, 11.023284912109375, 29.147705078125, -0.47021484375, -42.53582763671875, 42.79193115234375, 9.455078125, 28.3603515625, 0.0, 27.84259033203125, -43.51103210449219, 21.862548828125, 4.948579788208008, 18.20842742919922, 68.96171569824219, 69.93777465820312, -40.923126220703125, 23.544036865234375, 62.10248565673828, 29.633132934570312, 2.19769287109375, 49.730987548828125, -49.763084411621094, -1.699462890625, 7.80218505859375, 30.59442138671875, 0.4290618896484375, 14.84112548828125, 66.1734619140625, -22.254638671875, -38.807861328125, 101.5233154296875, 56.6680908203125, 54.23699951171875, -41.429412841796875, 36.145782470703125, -5.840492248535156, 53.80073547363281, 2.04351806640625, 50.29585266113281, 52.32997131347656, -4.96917724609375, 24.322311401367188, 78.41682434082031, 48.329498291015625, 66.76089477539062, 24.451812744140625, 16.45205307006836, -35.290802001953125, 1.400115966796875, 49.55462646484375, 59.01999282836914, 67.85110473632812, 60.09259033203125, 87.7017822265625, -14.465545654296875, 27.61865234375, 70.32475280761719, 61.33567810058594, 13.367206573486328, 10.941543579101562, 2.4045562744140625, 84.992919921875, 52.246826171875, -5.139892578125, -37.430084228515625, 38.62762451171875, 44.5562744140625, 7.95159912109375, 38.4967041015625, 47.775840759277344, 39.044525146484375, 21.873802185058594, 7.04815673828125, -55.3402099609375, 12.191864013671875, 35.1552734375, -19.8575439453125, -43.31501007080078, 58.06498718261719, -7.6067657470703125, 53.796504974365234, 45.20135498046875, 50.21531677246094, 0.888397216796875, 44.79107666015625, 0.42034912109375, 64.643798828125, 0.0, 27.053504943847656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000205.npy"}
|
|
{"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 20.485063552856445, "std": 32.021121978759766, "min": -76.25244140625, "p10": -16.26652679443359, "median": 13.808441162109375, "p90": 61.38168029785156, "max": 85.21572875976562, "pos_frac": 0.7265625, "sample": [40.09550476074219, 3.578319549560547, -55.42756271362305, 55.133819580078125, 85.21572875976562, -3.1790847778320312, -15.887969970703125, 14.119140625, 13.49774169921875, 60.897064208984375, 20.213272094726562, 54.20429992675781, 45.6329345703125, -8.609466552734375, -24.428192138671875, 18.88409423828125, 44.31689453125, 0.8664932250976562, -2.6341018676757812, 65.01545715332031, 7.191556930541992, 60.9324951171875, -1.6952667236328125, -17.764617919921875, -2.4320068359375, 5.911041259765625, -24.3096923828125, 28.032135009765625, 34.50780487060547, 12.241737365722656, 1.24737548828125, -11.731201171875, -35.92573547363281, 55.573699951171875, 59.406524658203125, -14.078704833984375, 82.80950927734375, 66.06053161621094, 67.17068481445312, 48.90851593017578, 7.739105224609375, 21.30097198486328, -6.955169677734375, 28.43841552734375, 38.19378662109375, 28.812957763671875, -22.04339599609375, 68.17501831054688, 56.0833740234375, 62.429779052734375, 23.963348388671875, 47.432403564453125, 48.959991455078125, 38.60992431640625, 7.336658477783203, 32.626922607421875, 25.807205200195312, 81.08200073242188, -76.25244140625, 13.378997802734375, 28.665008544921875, -28.47325897216797, -0.40289306640625, 54.46122741699219, 46.59100341796875, 7.6428375244140625, 3.99786376953125, -11.1243896484375, -3.992002487182617, 39.519622802734375, 9.436660766601562, 56.965789794921875, 35.05316162109375, -36.160430908203125, 53.539794921875, 4.163330078125, -9.15374755859375, 15.068939208984375, 8.9166259765625, -0.13317489624023438, -17.149826049804688, -13.67138671875, 73.5849609375, 57.480010986328125, 47.57710266113281, 7.652862548828125, 30.059814453125, -33.93864822387695, 9.620904922485352, 0.7099876403808594, 47.001495361328125, 53.87567901611328, 45.142913818359375, 5.2640838623046875, -11.276824951171875, 39.395233154296875, 14.859130859375, -10.803375244140625, 4.84185791015625, -5.230375289916992, 55.20771789550781, -6.945831298828125, 75.837158203125, 67.37603759765625, 0.012725830078125, 54.4935302734375, -32.43243408203125, -13.9189453125, -22.12786865234375, 15.195343017578125, 3.0527496337890625, 79.87545776367188, 51.73899841308594, 5.5731201171875, 2.1063156127929688, 52.26373291015625, 49.39909362792969, 4.670356750488281, 66.51158142089844, 32.93902587890625, 9.10546875, 11.86419677734375, -2.461883544921875, 51.707763671875, 30.898178100585938, 10.91070556640625, -11.39483642578125, 2.372314453125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000206.npy"}
|
|
{"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 17.63336944580078, "std": 38.40909957885742, "min": -77.595458984375, "p10": -27.418765258789062, "median": 13.728622436523438, "p90": 69.77535858154296, "max": 96.75531005859375, "pos_frac": 0.6953125, "sample": [56.40960693359375, 39.985626220703125, 12.44891357421875, -7.564899444580078, 13.997451782226562, 5.2288055419921875, 23.835418701171875, 94.90237426757812, 20.75579833984375, -2.258352279663086, 20.653844833374023, 80.63623046875, -48.329345703125, 3.3172531127929688, -1.791015625, 66.40901947021484, 5.925800323486328, 20.35565185546875, 96.75531005859375, -8.887847900390625, 67.65299987792969, -8.011962890625, -75.23211669921875, -12.124168395996094, 53.88039779663086, 21.921981811523438, -11.510330200195312, 15.154129028320312, -56.294189453125, 45.8516845703125, 58.254852294921875, 13.166252136230469, 11.27423095703125, -28.853607177734375, 42.659423828125, 56.83712387084961, 14.24822998046875, -5.94157600402832, 18.00497817993164, 69.58541870117188, -10.074996948242188, 6.9336700439453125, 40.065216064453125, 5.64544677734375, 6.2567138671875, -20.013580322265625, 17.032424926757812, 2.118560791015625, -63.137237548828125, 11.531692504882812, -77.595458984375, -17.23321533203125, -32.48493957519531, 25.837646484375, 65.18212890625, 0.0, 10.034431457519531, -26.8038330078125, 31.08380126953125, -5.446441650390625, 81.827880859375, -9.223541259765625, -38.975425720214844, 23.97552490234375, 13.459793090820312, -72.541015625, 73.40110778808594, -51.35552978515625, 3.4799957275390625, -5.6082000732421875, 35.624969482421875, -2.281982421875, 10.708526611328125, 68.75814819335938, 73.57315063476562, 35.07945251464844, 45.49999237060547, -18.362716674804688, -3.1762847900390625, 72.18315124511719, 4.713239669799805, -51.42864990234375, 15.984832763671875, 28.146560668945312, 5.272895812988281, 94.0411376953125, 14.450859069824219, 59.446380615234375, -16.806976318359375, 1.9269638061523438, 6.189178466796875, 15.214309692382812, -23.466827392578125, 71.805908203125, 16.972869873046875, 53.90966796875, 37.14231872558594, 6.1612701416015625, 29.50214385986328, -67.4493408203125, 83.708984375, 68.83688354492188, -12.70599365234375, 54.67340850830078, 6.0617218017578125, 84.5853271484375, 23.512176513671875, 1.921051025390625, 15.842977523803711, 66.88034057617188, 6.870456695556641, 34.59405517578125, -1.220306396484375, 70.21855163574219, 37.33563232421875, -8.822517395019531, 41.514892578125, 65.30023193359375, 73.04861450195312, -13.92706298828125, 11.092002868652344, -41.08073425292969, 61.00172424316406, 60.36119079589844, -3.55914306640625, 15.440475463867188, 56.52448272705078, 9.0447998046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000207.npy"}
|
|
{"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 21.42392921447754, "std": 34.61748123168945, "min": -88.9034423828125, "p10": -12.073176383972168, "median": 15.172195434570312, "p90": 64.49066925048828, "max": 117.0745849609375, "pos_frac": 0.75, "sample": [9.721832275390625, 9.611892700195312, -25.279571533203125, 14.174751281738281, -55.904754638671875, 4.444038391113281, -6.516244888305664, 21.433807373046875, 4.2858734130859375, 52.79376220703125, -12.593835830688477, 9.92596435546875, 38.417572021484375, 36.247222900390625, 38.63181686401367, 93.82366943359375, 14.98114013671875, 10.159126281738281, 16.537109375, 97.40338134765625, -70.49261474609375, 63.85874938964844, -7.7236175537109375, 39.71728515625, 13.10162353515625, 56.442779541015625, 90.21914672851562, 52.55390167236328, 13.404556274414062, -19.18853759765625, 117.0745849609375, 44.37555694580078, 43.344146728515625, 38.6768798828125, -11.85003662109375, 73.58668518066406, -88.9034423828125, 11.019027709960938, 11.975128173828125, 4.439483642578125, 27.737119674682617, -47.04644775390625, 62.20182800292969, -7.644542694091797, 59.595947265625, 27.49468994140625, 18.776992797851562, 86.2061767578125, 0.0, 104.007568359375, -22.675384521484375, -0.20743179321289062, 8.443405151367188, -11.161422729492188, 2.6799087524414062, 106.0654296875, 10.896896362304688, 20.857040405273438, 21.888824462890625, 70.61572265625, 9.334392547607422, 28.814193725585938, -7.960113525390625, 43.50048828125, -5.086334228515625, 3.78564453125, 2.370941162109375, -0.5563278198242188, 21.648460388183594, 25.479698181152344, 104.097412109375, 0.0, 19.097991943359375, 47.30584716796875, 21.654052734375, 61.69635009765625, -5.691200256347656, 65.96514892578125, 13.50933837890625, 51.51972198486328, 21.304485321044922, 35.31256103515625, 53.606292724609375, 73.79806518554688, -5.960418701171875, 42.214874267578125, -10.703216552734375, -0.07172012329101562, 3.3390655517578125, 33.137847900390625, -9.644256591796875, 17.35455322265625, 37.139129638671875, 6.788177490234375, 54.740478515625, 28.39361572265625, 51.675865173339844, 12.043060302734375, 41.11639404296875, -9.21728515625, 3.8752593994140625, 9.716842651367188, 4.4169464111328125, 11.720787048339844, 15.363250732421875, 12.849742889404297, -0.6939811706542969, 59.16876220703125, 13.364883422851562, 22.086715698242188, 16.552169799804688, -24.14556884765625, 48.10253143310547, 30.2916259765625, 6.4642333984375, -31.17584228515625, 13.962104797363281, 2.2935562133789062, -24.962875366210938, 39.5501708984375, 49.063323974609375, 16.05145263671875, -13.568840026855469, 67.12751007080078, 28.60041046142578, -32.02398681640625, -2.24041748046875, 36.936767578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000208.npy"}
|
|
{"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 18.715049743652344, "std": 34.37697982788086, "min": -83.8011474609375, "p10": -21.536779022216795, "median": 16.886459350585938, "p90": 62.93718948364258, "max": 101.73307800292969, "pos_frac": 0.640625, "sample": [91.67768859863281, -8.090484619140625, 56.275001525878906, 56.963539123535156, -9.085205078125, -3.369171142578125, 23.7142333984375, -24.08868408203125, -31.06744384765625, 17.49761962890625, 61.10528564453125, -8.77880859375, 2.123260498046875, 8.390167236328125, 82.1630859375, 63.685569763183594, 17.25177001953125, 5.730953216552734, 4.933319091796875, 46.57470703125, 69.88821411132812, -12.444671630859375, 4.3766632080078125, -48.289093017578125, 17.52777099609375, 7.916259765625, 25.76616859436035, 9.815673828125, -17.81878662109375, 62.58802795410156, 64.69869995117188, -30.177566528320312, -3.1862335205078125, -35.51710510253906, -1.1396484375, -30.124008178710938, 91.34951782226562, 18.337554931640625, 0.0, 9.735275268554688, 6.645263671875, -13.591278076171875, 40.42567443847656, -30.573638916015625, 21.8790283203125, 38.402130126953125, 42.33333969116211, 4.19019889831543, 20.51641845703125, 81.26225280761719, 34.1544189453125, 26.335952758789062, 12.85406494140625, -12.206939697265625, 16.521148681640625, -2.1884326934814453, -7.369781494140625, 14.9698486328125, 53.096160888671875, -7.366386413574219, 46.737335205078125, -15.475555419921875, -5.698455810546875, 63.786407470703125, 64.45028686523438, 19.49073028564453, -13.091583251953125, -3.85546875, 75.43841552734375, 8.591766357421875, 18.887451171875, 55.9791259765625, 23.88092041015625, 30.431903839111328, -1.1910324096679688, 0.0, 44.5030517578125, 50.446685791015625, -20.874305725097656, 29.18609619140625, -20.22967529296875, -4.4866180419921875, 44.91529846191406, 0.1075592041015625, 0.6484222412109375, 62.616455078125, 53.15174865722656, -31.868057250976562, -13.25408935546875, 7.24566650390625, 21.59002685546875, -23.082550048828125, -16.083251953125, 51.23082733154297, -2.508331298828125, -1.34222412109375, 23.16180419921875, 36.803916931152344, 101.73307800292969, -83.8011474609375, -9.616241455078125, 5.538280487060547, -30.504913330078125, -6.059322357177734, -5.8406982421875, -59.78114700317383, -5.04248046875, 25.03685760498047, 50.668426513671875, 51.6490478515625, 43.99440002441406, 60.07118225097656, 45.9407958984375, 18.808746337890625, 43.23692321777344, 66.71990966796875, 32.09672546386719, 57.753448486328125, -2.94183349609375, 87.4100341796875, 28.98773193359375, -27.244964599609375, 43.39540100097656, 44.6383056640625, 49.29010009765625, 59.0703125, -8.127998352050781, 61.008056640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000209.npy"}
|
|
{"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 20.243694305419922, "std": 34.3052978515625, "min": -79.4013671875, "p10": -14.253237915039062, "median": 18.504133224487305, "p90": 62.05704345703125, "max": 126.8475341796875, "pos_frac": 0.6875, "sample": [-6.666618347167969, -0.46340179443359375, 31.569793701171875, -2.08135986328125, 38.328086853027344, 13.1846923828125, 84.4163818359375, 78.46282958984375, 30.283203125, 21.095809936523438, 62.7073974609375, 17.92684555053711, 3.95611572265625, 48.45295715332031, 37.57568359375, -1.728790283203125, 45.64866638183594, -3.4859886169433594, 43.25297546386719, 2.744720458984375, 28.544422149658203, -4.5575103759765625, -46.058990478515625, -2.367645263671875, 45.21930694580078, 33.662147521972656, 8.886749267578125, 126.8475341796875, -44.69866943359375, -4.1909942626953125, 40.091339111328125, 20.315155029296875, 12.656707763671875, -1.0943145751953125, 56.16119384765625, -15.226715087890625, 7.108301162719727, 40.255126953125, 1.2126274108886719, 24.766357421875, 3.198394775390625, 39.213043212890625, 9.977380752563477, -7.930809020996094, 10.066474914550781, 47.18017578125, 29.0478515625, 1.1077728271484375, 57.41365051269531, 12.38800048828125, 19.0814208984375, -2.46771240234375, 35.95404052734375, 78.31060791015625, 25.426055908203125, 37.224822998046875, 37.015228271484375, -22.260101318359375, 59.474365234375, -5.55340576171875, 48.408935546875, 47.253143310546875, 61.7783203125, 53.16816329956055, 20.390716552734375, 26.514511108398438, -64.6123046875, -7.40509033203125, 12.68865966796875, 25.9234619140625, -45.6217041015625, -24.906005859375, 70.22233581542969, 30.12652587890625, 78.13128662109375, 65.31498718261719, -0.1688098907470703, 6.402923583984375, -1.86651611328125, -79.4013671875, 54.16081237792969, 13.943923950195312, -1.199075698852539, 83.55364990234375, 26.159324645996094, 60.360931396484375, 92.81658935546875, 33.286407470703125, -1.1270751953125, -2.6083221435546875, -9.82501220703125, 12.261470794677734, 3.100078582763672, -19.681777954101562, 22.86654281616211, 23.8748779296875, -4.890655517578125, 85.90513610839844, -13.87835693359375, 39.57391357421875, 41.068267822265625, 91.325927734375, 7.285652160644531, -7.839447021484375, 77.99456787109375, 15.378486633300781, 59.344635009765625, -1.5231857299804688, 11.86700439453125, -15.127960205078125, 34.08677673339844, 20.395553588867188, -2.6258697509765625, 13.588043212890625, 13.38116455078125, -68.27090454101562, 11.06829833984375, 34.018707275390625, 23.17694091796875, -57.2203369140625, 51.0347900390625, -9.882720947265625, 49.36347961425781, 40.1854248046875, -21.16693115234375, -5.38690185546875, -2.8385009765625, 30.936721801757812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000210.npy"}
|
|
{"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 27.79233169555664, "std": 32.523651123046875, "min": -52.028480529785156, "p10": -10.075668334960938, "median": 23.08016586303711, "p90": 68.40928497314454, "max": 106.49636840820312, "pos_frac": 0.8046875, "sample": [25.274742126464844, 12.255233764648438, 86.0750732421875, 43.841949462890625, 56.97222900390625, 67.62359619140625, 10.699703216552734, 6.837024688720703, -10.490097045898438, 47.340576171875, 56.455078125, -49.4202880859375, 34.6898193359375, 37.816070556640625, -11.438735961914062, 66.21282958984375, 2.1139297485351562, -2.0153427124023438, 6.20050048828125, -14.120986938476562, 9.44683837890625, 48.67127227783203, 13.053909301757812, 17.359817504882812, 29.41541290283203, 20.885589599609375, 7.099761962890625, -40.01690673828125, 35.55705261230469, -6.532440185546875, 14.749828338623047, 57.89231872558594, -0.2737579345703125, 20.77880859375, 4.531288146972656, 31.434417724609375, 8.922521591186523, -33.44213104248047, 33.24561309814453, 68.64547729492188, 14.37908935546875, 44.354095458984375, 7.954864501953125, -1.2373046875, -3.9392242431640625, 95.6666259765625, 68.30805969238281, 46.721954345703125, -2.65924072265625, 10.403270721435547, 54.00433349609375, -9.1192626953125, 3.8314208984375, 0.6982574462890625, 106.49636840820312, 52.203125, 46.538177490234375, -13.731483459472656, 17.54241943359375, 52.018310546875, -6.8876953125, -14.316680908203125, -17.291671752929688, 94.8167724609375, -17.678466796875, 5.0478668212890625, -1.53802490234375, 53.869354248046875, 65.37393188476562, 10.645292282104492, -17.694854736328125, 0.6834716796875, -1.8440513610839844, 45.1993408203125, 4.809232711791992, 12.423601150512695, 55.57735824584961, 51.44989013671875, 54.959739685058594, 52.15240478515625, 37.9029541015625, 35.8258056640625, 2.5993118286132812, 48.4852294921875, 55.131103515625, 100.37945556640625, 10.72113037109375, 2.0872344970703125, 50.767539978027344, 11.304176330566406, -9.898056030273438, 86.51705932617188, 81.8531494140625, -25.900390625, 77.51324462890625, 18.536842346191406, 62.418670654296875, 70.36367797851562, 14.783988952636719, 0.679443359375, 53.171051025390625, 60.48370361328125, 34.92193603515625, 10.47543716430664, -52.028480529785156, 4.223552703857422, 39.4891357421875, 38.70562744140625, 75.63494873046875, 49.952850341796875, 36.61834716796875, 11.653383255004883, 62.26704406738281, 34.848358154296875, 7.141939163208008, 32.03753662109375, 81.73587036132812, 72.89596557617188, 51.73826599121094, 18.795257568359375, 17.769241333007812, 31.437942504882812, -2.855499267578125, 66.64901733398438, 37.721900939941406, 48.59355163574219, 2.284576416015625, 64.4463882446289], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000211.npy"}
|
|
{"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 19.482942581176758, "std": 32.616485595703125, "min": -75.59967041015625, "p10": -15.276610946655273, "median": 16.112327575683594, "p90": 61.654632568359375, "max": 98.74960327148438, "pos_frac": 0.734375, "sample": [6.125358581542969, -2.890655517578125, -5.7978515625, 19.639366149902344, -11.872406005859375, -15.010360717773438, 2.169036865234375, 13.362335205078125, 2.5481414794921875, -1.82537841796875, 63.62712860107422, 3.0396957397460938, 96.11468505859375, 3.4908199310302734, 65.398681640625, -23.93799591064453, 8.457393646240234, 55.130889892578125, 61.64085388183594, 26.101654052734375, 54.153533935546875, 8.406951904296875, 1.761688232421875, 18.91436767578125, 22.867767333984375, 11.277191162109375, 1.465118408203125, -51.51422119140625, 23.148509979248047, 21.63232421875, -15.89786148071289, -27.199851989746094, 5.0091552734375, 5.8656463623046875, 20.593406677246094, 38.2103271484375, 26.9854736328125, 72.51795959472656, 19.783538818359375, -1.4637908935546875, 41.337005615234375, -69.90475463867188, -2.72021484375, -5.002872467041016, -9.195648193359375, 55.32691955566406, 23.732025146484375, 58.071990966796875, -21.908905029296875, -37.18622970581055, 21.711769104003906, 8.634368896484375, -34.915008544921875, 56.242340087890625, 29.733749389648438, -7.21630859375, -6.7574920654296875, 52.1278076171875, 3.5765380859375, -7.07647705078125, -5.768230438232422, 3.675107955932617, 56.89820098876953, 2.7760009765625, -4.49749755859375, 56.898109436035156, -8.991626739501953, 43.256072998046875, 39.4449462890625, 70.55591583251953, 33.21989440917969, 59.45105743408203, -28.30908203125, 13.62542724609375, 61.68678283691406, 11.524917602539062, 40.098270416259766, 49.41265869140625, -1.9837760925292969, 79.76211547851562, 35.886474609375, 50.3189697265625, 17.924209594726562, 49.5462646484375, -60.40177917480469, 19.903533935546875, 19.330734252929688, -17.8328857421875, 13.061904907226562, 7.896934509277344, 38.85931396484375, 61.98931884765625, 6.040374755859375, 5.652332305908203, 96.69378662109375, 46.200225830078125, 20.16937255859375, 68.77165985107422, 43.502593994140625, 29.235610961914062, 26.919570922851562, 98.74960327148438, 50.44102478027344, 0.37511444091796875, 14.300445556640625, -75.59967041015625, 36.06793212890625, 56.741119384765625, -9.940383911132812, -9.486221313476562, -0.588134765625, -29.518310546875, 27.976806640625, 9.114240646362305, 24.291839599609375, -0.33664703369140625, 29.100372314453125, 9.664390563964844, 41.06232833862305, -0.6066665649414062, 52.352691650390625, 10.545989990234375, 38.398765563964844, 66.31641387939453, 10.3182373046875, 27.588943481445312, 84.14411926269531, 9.299072265625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000212.npy"}
|
|
{"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 18.59851837158203, "std": 33.81898880004883, "min": -59.754852294921875, "p10": -17.624322509765626, "median": 13.532459259033203, "p90": 70.98654708862304, "max": 111.89892578125, "pos_frac": 0.6796875, "sample": [75.42990112304688, 51.57524108886719, -12.081024169921875, 65.5189208984375, -17.48773193359375, 41.98539733886719, 56.4683723449707, 15.942779541015625, -3.73004150390625, 28.448486328125, -7.451629638671875, -2.2857666015625, 17.343551635742188, -14.474037170410156, 22.48678970336914, 96.79583740234375, 0.8156890869140625, 20.743255615234375, 20.83135986328125, 26.147613525390625, 17.21551513671875, -6.41387939453125, 91.4739990234375, -1.156494140625, 65.94172668457031, -10.5970458984375, -3.089019775390625, 14.545074462890625, 15.958763122558594, 13.217994689941406, 6.07391357421875, -19.94012451171875, -11.354522705078125, 22.13726806640625, 64.31907653808594, 6.0393829345703125, 55.959136962890625, -3.52935791015625, 7.9831390380859375, -50.42279052734375, 64.11978149414062, 29.941024780273438, 21.937850952148438, 36.037506103515625, -34.63543701171875, 60.88922119140625, 42.358184814453125, 7.418975830078125, -28.939178466796875, 12.47235107421875, 5.854896545410156, 40.16400146484375, 47.803009033203125, 3.9221725463867188, 53.61798095703125, -21.617755889892578, -3.8374099731445312, 53.03266906738281, -35.567352294921875, 77.10289001464844, 16.451202392578125, 14.95166015625, 17.674293518066406, -24.118804931640625, 7.5045166015625, 13.846923828125, 83.11459350585938, -17.237945556640625, -12.8375244140625, -7.3385467529296875, 54.99980163574219, -17.66888427734375, 22.544647216796875, -59.754852294921875, 20.299224853515625, 3.5527877807617188, 39.558738708496094, 72.31851196289062, 7.475902557373047, 33.3360595703125, -17.605224609375, 6.7145843505859375, -2.344240188598633, 0.09228515625, -0.22663116455078125, 52.924072265625, -26.942718505859375, -3.0325927734375, 13.85467529296875, 98.5450439453125, 49.663368225097656, 74.401123046875, 4.551017761230469, 7.81781005859375, 32.507049560546875, -5.837491989135742, 71.36529541015625, -12.212890625, -16.94921875, 111.89892578125, 1.9707107543945312, 31.82653045654297, 16.842864990234375, 70.82422637939453, -28.2781982421875, 94.89971923828125, 29.688888549804688, -4.2615966796875, 32.97138977050781, 0.0, 9.993852615356445, 11.713088989257812, 18.6715087890625, 71.771484375, 10.289548873901367, -43.396728515625, 29.640838623046875, 25.7778263092041, -4.824134826660156, 16.8594970703125, 2.565502166748047, 40.0137939453125, -21.145904541015625, 2.2786788940429688, 83.360595703125, 3.8502273559570312, -16.8653564453125, -3.7451095581054688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000213.npy"}
|
|
{"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 19.084556579589844, "std": 33.97089767456055, "min": -51.265594482421875, "p10": -22.841866302490235, "median": 19.330764770507812, "p90": 63.860560607910145, "max": 120.03384399414062, "pos_frac": 0.71875, "sample": [44.37785339355469, 31.300750732421875, -5.708549499511719, 7.1363525390625, 13.284744262695312, 38.4603271484375, 17.686630249023438, -10.857162475585938, -38.11907958984375, -14.86822509765625, 23.97760009765625, 40.96391296386719, -18.858367919921875, 31.527801513671875, 15.89898681640625, 51.05376434326172, -41.62823486328125, 40.230621337890625, -4.7012786865234375, 29.332916259765625, 1.7378616333007812, -9.953399658203125, 16.887893676757812, 42.11529541015625, 32.78167724609375, 46.29045104980469, 25.73931121826172, -16.6248779296875, 22.681884765625, 22.193443298339844, 5.630729675292969, 9.90899658203125, 7.36724853515625, 2.7359237670898438, 62.53163146972656, 17.05377197265625, 120.03384399414062, 22.985763549804688, 71.82406616210938, -48.9190673828125, 18.933975219726562, -20.18109130859375, 2.909128189086914, 13.63348388671875, 16.203243255615234, 3.7364540100097656, 1.173583984375, 70.19381713867188, 5.51922607421875, -6.425348281860352, 50.158363342285156, 57.5238037109375, -21.123886108398438, 39.472084045410156, 2.6612319946289062, 61.31852722167969, 19.9024658203125, 40.690643310546875, -8.336822509765625, 59.632232666015625, 46.492393493652344, 89.58026123046875, 30.918869018554688, 4.9813232421875, 4.8822174072265625, -21.894027709960938, -4.6543731689453125, -41.149070739746094, 5.39501953125, 0.0, -7.9454803466796875, -22.793533325195312, -3.608306884765625, 13.885589599609375, 54.50518798828125, -15.335906982421875, 37.17504119873047, -28.623035430908203, 8.910606384277344, 11.283430099487305, -12.219635009765625, -43.6910400390625, 81.718994140625, 37.37261962890625, 40.53562927246094, -27.88177490234375, 71.71022033691406, 31.36925506591797, 20.204315185546875, 14.502593994140625, -51.265594482421875, 25.426513671875, 23.3798828125, 46.57373046875, 74.41696166992188, 19.727554321289062, -5.155677795410156, 27.68792724609375, -0.44879150390625, 30.543960571289062, 62.29461669921875, 76.56817626953125, 32.668060302734375, 32.285552978515625, 27.6693115234375, 68.0736083984375, 97.901611328125, 26.48687744140625, -50.17918395996094, 30.95880126953125, 108.8494873046875, 48.044464111328125, 13.647857666015625, -27.91278076171875, -14.57171630859375, 66.96139526367188, 33.5989990234375, -45.78143310546875, -11.16143798828125, -25.332244873046875, 7.050567626953125, 40.324241638183594, 23.577072143554688, 20.715972900390625, -22.95464324951172, 76.785400390625, 22.319686889648438, 44.33203125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000214.npy"}
|
|
{"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 19.512264251708984, "std": 37.44340133666992, "min": -85.16974639892578, "p10": -26.003900146484373, "median": 13.555598258972168, "p90": 66.50846557617186, "max": 131.467529296875, "pos_frac": 0.6796875, "sample": [13.436239242553711, 24.357147216796875, 13.674957275390625, 52.714630126953125, 82.04898071289062, 50.2838134765625, 47.6204833984375, 11.260002136230469, 21.76348876953125, -5.82080078125, 83.36502075195312, 78.81838989257812, 50.1368408203125, -20.564605712890625, 1.35552978515625, 24.60595703125, -4.842987060546875, 102.90118408203125, 106.010986328125, 54.682342529296875, 53.17901611328125, -2.643707275390625, 0.0, -25.81121826171875, -3.0002822875976562, -15.033317565917969, 26.254318237304688, 27.586715698242188, 99.33840942382812, -9.343521118164062, 61.24789047241211, -0.88092041015625, -6.952667236328125, -37.20484924316406, 0.9429283142089844, 30.6817626953125, 61.378517150878906, 5.7997589111328125, 39.46721649169922, 98.13973999023438, -9.257293701171875, 26.7060546875, -26.4534912109375, -0.952606201171875, -0.1359405517578125, -0.2193450927734375, -16.341217041015625, 38.286651611328125, 53.162994384765625, 43.058311462402344, 3.677459716796875, 131.467529296875, 51.79962158203125, 20.21533203125, -36.050567626953125, 2.843780517578125, 8.109207153320312, -13.2855224609375, 0.0, 60.986968994140625, -3.4890289306640625, 3.114471435546875, 24.7279052734375, -19.998321533203125, 33.056060791015625, 20.608078002929688, -36.85484313964844, 1.6944580078125, 70.45758056640625, 48.8011474609375, 26.167068481445312, 7.829946517944336, -3.755767822265625, 65.0262451171875, -49.32470703125, 54.839935302734375, 5.498779296875, -85.16974639892578, 30.0374698638916, 36.6163330078125, 22.129714965820312, 9.404678344726562, 15.9658203125, -0.5089111328125, -24.92120361328125, -5.0755157470703125, 55.185516357421875, 11.592987060546875, 71.46148681640625, -29.766937255859375, -10.652923583984375, 17.650604248046875, 15.384819030761719, -3.80230712890625, 6.568000793457031, 12.648483276367188, -2.721162796020508, 20.88153076171875, 84.40031433105469, 122.02423095703125, -26.83697509765625, 22.925926208496094, -39.648887634277344, 0.7656822204589844, -59.340423583984375, -32.234130859375, 39.91990661621094, 64.93978118896484, -36.73516845703125, 45.924285888671875, 2.95281982421875, -7.143047332763672, 26.65160369873047, 27.0709228515625, 9.329132080078125, 23.140838623046875, 15.396759033203125, 37.52256774902344, -41.181396484375, 25.7552490234375, 12.876434326171875, 3.004425048828125, 7.699462890625, 69.96697998046875, 45.876190185546875, 43.65034484863281, 7.9576416015625, 61.05940246582031], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000215.npy"}
|
|
{"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 21.55377960205078, "std": 36.4968147277832, "min": -75.15408325195312, "p10": -22.014849853515617, "median": 20.076313018798828, "p90": 72.19607238769531, "max": 104.0152587890625, "pos_frac": 0.7578125, "sample": [24.788536071777344, 20.031478881835938, 46.38677978515625, 26.681655883789062, 0.4066162109375, 4.4031219482421875, -75.15408325195312, 9.545822143554688, 50.20994567871094, -3.1873397827148438, 59.031532287597656, 68.46087646484375, 30.378662109375, -3.805023193359375, 0.112640380859375, 47.63948059082031, 104.0152587890625, 8.194496154785156, 48.01373291015625, 83.447265625, 17.954742431640625, -65.70367431640625, 38.10816955566406, 21.849929809570312, -13.290531158447266, 18.66754150390625, -68.3580322265625, 17.375350952148438, 29.36818504333496, 40.1204833984375, -6.6468048095703125, -26.432937622070312, -41.9874267578125, -2.4164810180664062, 46.42575454711914, 8.833354949951172, 55.64337158203125, -41.99029541015625, 84.565185546875, 54.09827423095703, -11.73397445678711, 63.5538330078125, 72.17874145507812, 56.44275665283203, 52.24253845214844, 48.576019287109375, 32.91326904296875, 28.4534912109375, 3.5433349609375, 0.553619384765625, -20.267364501953125, -6.632270812988281, 83.19635009765625, 41.727996826171875, -11.546028137207031, 39.6009521484375, 2.964263916015625, 81.61398315429688, 26.693572998046875, 5.060050964355469, 0.0, 39.7327880859375, -17.19195556640625, 20.554901123046875, 73.65121459960938, 70.0960693359375, 72.23651123046875, -13.444915771484375, -7.6425628662109375, 47.07891845703125, 53.88804626464844, 53.11456298828125, -57.32427978515625, 13.50018310546875, 65.14500427246094, 78.15675354003906, 2.95867919921875, 43.219390869140625, 23.5281982421875, 8.868316650390625, 27.488677978515625, 5.45465087890625, 31.258209228515625, 22.146392822265625, 79.89114379882812, 35.08467102050781, -42.17987060546875, -6.0999755859375, 81.93017578125, 95.91534423828125, -34.230560302734375, -26.092315673828125, 4.75653076171875, 65.0374755859375, -0.4929962158203125, 13.8140869140625, 15.9498291015625, -12.44854736328125, 12.673118591308594, 9.122344970703125, 76.5025634765625, 10.726539611816406, 0.0, 33.64311981201172, 48.560340881347656, -48.17955017089844, 2.0044174194335938, 45.89897155761719, 48.93267822265625, 18.39733123779297, -0.28041648864746094, 41.27313232421875, 25.5743408203125, 19.99896240234375, 20.12114715576172, 16.06488037109375, 26.5333251953125, 15.20819091796875, 7.0540771484375, 72.63494873046875, 27.68609619140625, -68.45404052734375, 42.28212356567383, 50.07992172241211, 4.3277587890625, 0.4901123046875, -35.4072265625, 3.185150146484375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000216.npy"}
|
|
{"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 28.542221069335938, "std": 39.545806884765625, "min": -75.41329956054688, "p10": -16.451184082031247, "median": 20.853614807128906, "p90": 76.2656997680664, "max": 136.67657470703125, "pos_frac": 0.7734375, "sample": [8.02557373046875, -0.614013671875, -26.2750244140625, 29.54510498046875, 52.49769592285156, -51.10691833496094, 7.6389312744140625, 71.94827270507812, -0.33928680419921875, -0.74053955078125, 75.69636535644531, 1.00225830078125, 20.418167114257812, 42.7119140625, -0.8006591796875, 73.5099105834961, 55.4249382019043, 61.61326599121094, 59.62916564941406, 4.491302490234375, 66.99430847167969, 20.073883056640625, 2.542469024658203, 62.446746826171875, 136.67657470703125, 28.676605224609375, 60.4549560546875, 57.54845428466797, 62.045318603515625, 73.8908462524414, 23.92822265625, 64.27761840820312, -12.174774169921875, 99.4500732421875, 49.53446960449219, 20.6864013671875, 99.78268432617188, 57.55889892578125, 73.03012084960938, -6.60430908203125, -75.41329956054688, 67.5777587890625, 32.77996826171875, 119.58441162109375, -23.136764526367188, 42.37273406982422, 10.56671142578125, -39.566490173339844, 32.2506103515625, 16.09454345703125, 3.269428253173828, 3.3511962890625, -17.558563232421875, 32.86383056640625, 33.76329803466797, -41.5687255859375, 7.278778076171875, 62.325347900390625, 28.703956604003906, -7.556304931640625, 3.2759170532226562, 56.67926025390625, -3.593170166015625, 98.47164916992188, 1.8686676025390625, 13.358535766601562, 9.076875686645508, 77.59414672851562, 47.965660095214844, 4.7951202392578125, 55.19365692138672, 58.17316436767578, 33.330039978027344, -3.9762535095214844, 75.43927001953125, -9.17291259765625, -14.073898315429688, -21.63172149658203, 12.518646240234375, 130.56427001953125, 9.433853149414062, 49.632080078125, 17.272029876708984, -8.937801361083984, -69.90899658203125, 93.08035278320312, 79.56100463867188, 70.9969482421875, 7.8653564453125, -15.976593017578125, 13.993011474609375, 61.45208740234375, -4.40728759765625, 58.878936767578125, 7.5737457275390625, 85.4483642578125, 0.8711318969726562, 12.611488342285156, 7.880287170410156, 67.57005310058594, -33.19438171386719, 0.0, 20.158294677734375, 21.020828247070312, -22.93072509765625, 103.861328125, 54.217529296875, -23.945404052734375, 47.27349853515625, 90.09696960449219, 22.741012573242188, 20.42828369140625, 21.4991455078125, 78.9932861328125, 58.94465637207031, 37.42805480957031, 18.368385314941406, 17.75531005859375, -26.60113525390625, 8.131881713867188, 48.59638977050781, 9.725765228271484, 34.95989990234375, 15.763778686523438, 23.167694091796875, -15.231094360351562, 35.05712890625, 3.292217254638672], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000217.npy"}
|
|
{"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 22.96459197998047, "std": 35.16831588745117, "min": -60.16083526611328, "p10": -14.904643249511718, "median": 17.76748275756836, "p90": 66.72306060791016, "max": 109.8768310546875, "pos_frac": 0.765625, "sample": [28.2373046875, 8.217140197753906, 0.834991455078125, 66.55068969726562, 1.4961395263671875, -2.23828125, -12.83880615234375, 80.01812744140625, 11.3250732421875, 8.69671630859375, -45.539947509765625, 8.670761108398438, 27.536376953125, 75.82403564453125, 65.30667114257812, -47.627288818359375, -9.1671142578125, 59.674705505371094, 57.628929138183594, 21.856643676757812, -7.3953399658203125, 56.367095947265625, 49.5174560546875, 60.21394348144531, 12.027130126953125, 13.901847839355469, -1.8699951171875, -7.25103759765625, -36.849708557128906, 46.189300537109375, 50.133628845214844, 92.88848876953125, 4.7294158935546875, 66.08415985107422, -15.65313720703125, 9.05560302734375, 58.1771240234375, -51.77105712890625, 23.482681274414062, 28.771392822265625, 109.8768310546875, 26.332778930664062, 28.668548583984375, 1.7906379699707031, 1.6117706298828125, 66.33827209472656, 4.037532806396484, -11.732574462890625, -27.48956298828125, 59.96391296386719, 58.82002258300781, -14.601974487304688, 53.69059753417969, 19.34642791748047, 30.449127197265625, -0.3262367248535156, 67.93472290039062, 51.56634521484375, 69.56393432617188, -19.64202880859375, 9.288238525390625, -5.180335998535156, -25.867443084716797, 10.21832275390625, 0.0, 42.035430908203125, 13.857376098632812, 0.42205810546875, -14.486183166503906, 66.27923583984375, 4.266685485839844, 26.423690795898438, -41.2858772277832, 64.33085632324219, 29.3076171875, 77.92852783203125, 13.762008666992188, 28.450164794921875, 21.67041015625, 1.0098724365234375, 91.0999755859375, 45.825775146484375, 89.02001953125, 6.76580810546875, -2.68682861328125, 16.18853759765625, 12.609138488769531, 2.0542068481445312, 50.966033935546875, 29.050537109375, 10.672630310058594, 44.179473876953125, 13.1409912109375, -58.555877685546875, 29.42877960205078, 27.739913940429688, 67.12525939941406, 103.60784912109375, -35.25732421875, 33.5535888671875, 2.702911376953125, 34.3701171875, -6.311542510986328, -1.87744140625, 43.699188232421875, -11.89703369140625, -60.16083526611328, 58.21556091308594, 4.50555419921875, 27.574920654296875, 6.098052978515625, 49.54376220703125, 31.08160400390625, 37.307861328125, -15.610870361328125, 13.2042236328125, 87.59292602539062, 0.223907470703125, 10.58563232421875, -11.081085205078125, 52.942352294921875, 45.6199951171875, 25.850860595703125, 13.27508544921875, 35.255523681640625, 101.684326171875, 13.54718017578125, 21.15509033203125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000218.npy"}
|
|
{"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 24.179359436035156, "std": 34.60063934326172, "min": -85.76495361328125, "p10": -8.694397735595704, "median": 20.311267852783203, "p90": 68.49937438964844, "max": 105.67279052734375, "pos_frac": 0.75, "sample": [-76.3251953125, 1.2143707275390625, 32.34815979003906, 9.506240844726562, 63.67036437988281, 20.532691955566406, -11.61260986328125, 24.62042236328125, 59.85223388671875, 25.09307861328125, 31.840911865234375, -1.3226757049560547, 29.8480224609375, -6.20013427734375, 2.2970352172851562, 49.0509033203125, 64.68206787109375, -17.728363037109375, 48.80323791503906, 51.02468490600586, 37.730369567871094, 6.19635009765625, 34.22540283203125, 95.05349731445312, 5.7414703369140625, 33.562591552734375, 38.36714172363281, 4.716648101806641, 56.473236083984375, -11.215835571289062, -8.644004821777344, -3.3746280670166016, -31.009307861328125, 37.13710021972656, 2.447784423828125, 69.32203674316406, -27.199005126953125, 46.40566635131836, 29.895706176757812, 66.89483642578125, 85.99258422851562, 105.67279052734375, 6.750770568847656, 48.84269714355469, 7.3367767333984375, -22.445098876953125, 5.439422607421875, 33.40240478515625, 30.9935302734375, 65.56607055664062, 0.0, 18.76263427734375, 13.008895874023438, 24.851165771484375, -5.448982238769531, 75.04202270507812, -15.26300048828125, 73.14962768554688, 61.45819091796875, 12.470703125, 22.974319458007812, 22.780029296875, 4.3675689697265625, 2.4234619140625, 2.04750919342041, -6.422576904296875, -11.45172119140625, 9.82967758178711, 0.9559478759765625, 21.654525756835938, -8.811981201171875, 17.1297607421875, 10.262138366699219, 91.33139038085938, 10.597930908203125, -0.035125732421875, -44.76435852050781, -3.808380126953125, 16.484619140625, 52.439605712890625, -1.2475833892822266, 62.781982421875, 20.08984375, 22.52880859375, -4.802057266235352, -0.696197509765625, 83.99545288085938, 4.226591110229492, 85.2208251953125, -3.10260009765625, 57.54791259765625, 32.825439453125, 44.40240478515625, 26.38238525390625, 47.18254852294922, 68.14680480957031, 16.518077850341797, 38.48077392578125, 61.515289306640625, 66.4073486328125, -5.303466796875, 62.69425964355469, 54.3592529296875, -7.6416778564453125, 70.31158447265625, 44.592498779296875, 44.69036865234375, 7.1202545166015625, 22.987762451171875, 17.570053100585938, 5.947689056396484, -3.31292724609375, 67.12275695800781, -85.76495361328125, 95.968994140625, 77.3270263671875, 7.6537933349609375, -7.221534729003906, 26.683578491210938, 7.029609680175781, 6.298583984375, 34.67451477050781, 59.66400146484375, 8.723541259765625, 95.93115234375, -7.3903045654296875, -2.1307601928710938, -51.521728515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000219.npy"}
|
|
{"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 25.48628044128418, "std": 36.48053741455078, "min": -96.44732666015625, "p10": -19.49813079833984, "median": 27.998809814453125, "p90": 69.5083282470703, "max": 112.62075805664062, "pos_frac": 0.734375, "sample": [63.382781982421875, 81.87994384765625, 16.797271728515625, 48.988067626953125, 20.5085391998291, 52.307373046875, 4.196258544921875, 57.36712646484375, 15.084999084472656, 30.53302001953125, -11.333927154541016, 4.599151611328125, 53.17687225341797, -1.977294921875, 10.812019348144531, 43.916748046875, 51.429107666015625, 58.6229248046875, -13.36175537109375, 32.548553466796875, 37.37127685546875, 4.44970703125, 16.330169677734375, 2.592294692993164, 0.5533313751220703, 112.62075805664062, -22.290802001953125, 57.57624816894531, -18.474685668945312, 45.2562255859375, 13.894393920898438, 4.29461669921875, -8.3204345703125, -53.94036865234375, 24.8265380859375, 7.008514404296875, -3.1304931640625, 43.487396240234375, -0.4309062957763672, 89.1390380859375, 16.30834197998047, 27.43841552734375, 64.56103515625, 32.94622802734375, 84.37965393066406, 57.922218322753906, -5.687202453613281, -96.44732666015625, 46.26165771484375, 37.040771484375, 75.91595458984375, 39.802154541015625, -6.739898681640625, 82.88815307617188, -22.33892822265625, 14.984931945800781, -36.350921630859375, 10.503562927246094, 3.7677001953125, 35.221649169921875, -3.87554931640625, 44.469566345214844, 63.870452880859375, 42.05963134765625, 6.241020202636719, -10.142807006835938, 3.7142181396484375, 43.3616943359375, -23.6497802734375, 63.481781005859375, 41.14692687988281, 86.81947326660156, 27.055801391601562, 41.25433349609375, 59.318084716796875, 53.175048828125, -17.473602294921875, 45.82037353515625, 49.68446350097656, -13.9356689453125, -7.006477355957031, 10.697525024414062, 40.111026763916016, 4.62261962890625, 19.06121063232422, 90.69173431396484, -35.01751708984375, 42.817047119140625, 67.64959716796875, 96.9915771484375, -21.88616943359375, 14.7559814453125, -34.12896728515625, 0.6909370422363281, 12.613861083984375, 91.97727966308594, 56.10125732421875, 14.37225341796875, 38.606597900390625, -17.059722900390625, 65.87933349609375, 47.324790954589844, -30.717529296875, -36.78497314453125, -0.9374160766601562, -8.409530639648438, 55.0723876953125, -26.698074340820312, 75.75320434570312, 43.87953186035156, 55.93040466308594, 29.7821044921875, 46.5054931640625, -10.800018310546875, 92.2177734375, -5.9215087890625, 62.324127197265625, 45.6214599609375, 10.875591278076172, 58.38885498046875, -41.516754150390625, 35.39874267578125, 54.43383026123047, 31.301599502563477, 28.5592041015625, -10.10369873046875, 73.84536743164062, -2.6860809326171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000220.npy"}
|
|
{"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 19.218725204467773, "std": 35.565555572509766, "min": -76.54641723632812, "p10": -22.372429656982423, "median": 14.037109375, "p90": 67.91599578857421, "max": 91.34930419921875, "pos_frac": 0.71875, "sample": [32.819732666015625, 28.24932861328125, -12.5625, 13.973114013671875, 24.76471710205078, 38.339935302734375, 13.10797119140625, 32.82102966308594, -9.119644165039062, -9.146451950073242, 16.372154235839844, 59.912445068359375, 2.54254150390625, 53.3792724609375, -23.158309936523438, 21.882965087890625, 24.11236572265625, -23.648635864257812, 9.476886749267578, -12.556556701660156, 44.319793701171875, 20.304168701171875, 90.49755859375, -16.94921875, 23.095306396484375, 54.14012908935547, 49.86541748046875, 44.010223388671875, 60.372283935546875, 0.341644287109375, 13.93548583984375, 10.631546020507812, 87.1114501953125, -76.54641723632812, -14.633636474609375, 11.619415283203125, 9.544830322265625, 59.80419921875, -1.274566650390625, 48.0704345703125, -12.738800048828125, 5.95501708984375, -27.690170288085938, 24.64862060546875, -23.838607788085938, 2.2620849609375, 55.5308837890625, -8.620849609375, 33.3914794921875, 24.96722412109375, -0.02779388427734375, 62.08390808105469, 14.132080078125, -11.74261474609375, -7.178796768188477, 75.22662353515625, 46.23382568359375, -48.942352294921875, -0.790069580078125, 22.34320831298828, 5.4320068359375, 35.58049011230469, 37.254913330078125, 4.596698760986328, 30.67730712890625, 2.2116546630859375, 55.718902587890625, 48.96026611328125, -1.5065765380859375, 60.425743103027344, 44.92725372314453, -5.242401123046875, 35.278289794921875, 24.37714385986328, 70.33451843261719, 47.82586669921875, 83.10894775390625, 18.176803588867188, -48.48571014404297, -35.856658935546875, 65.1463623046875, 80.0838623046875, -9.986480712890625, 6.315582275390625, -22.106658935546875, 66.87948608398438, 13.070953369140625, -14.96954345703125, 52.404052734375, 16.00299072265625, 84.07760620117188, 80.45404052734375, 0.7558364868164062, 5.759988784790039, -12.64385986328125, 3.2039718627929688, -24.73522186279297, 91.34930419921875, -70.33526611328125, 14.101104736328125, 4.4066925048828125, -53.39508056640625, -22.99256134033203, 82.47557067871094, 13.083282470703125, 0.60009765625, 11.937278747558594, 81.38645935058594, 57.981964111328125, 10.368843078613281, 14.28094482421875, 51.144927978515625, -16.386077880859375, -16.289066314697266, 7.664161682128906, 4.475151062011719, 70.81986999511719, 25.269439697265625, 61.48394775390625, -7.9173583984375, 19.9940185546875, 55.307586669921875, -16.6494140625, 74.68075561523438, 13.5020751953125, 9.866567611694336, -65.8221435546875, 25.067840576171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000221.npy"}
|
|
{"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 24.013065338134766, "std": 34.229522705078125, "min": -85.607666015625, "p10": -10.95675048828125, "median": 15.421527862548828, "p90": 71.6750503540039, "max": 100.17041015625, "pos_frac": 0.7734375, "sample": [29.752182006835938, 11.08026123046875, 71.60575866699219, 54.295562744140625, -39.19416809082031, -49.30094909667969, 61.256591796875, 21.782583236694336, 14.548248291015625, 7.329437255859375, -0.540435791015625, 18.6334228515625, 64.18829345703125, 6.702423095703125, 20.1649169921875, 15.027656555175781, 28.866455078125, 92.23858642578125, 45.17547607421875, 7.361305236816406, 6.602325439453125, 2.2989501953125, -14.006820678710938, -28.1712646484375, 18.929168701171875, 100.17041015625, 17.15997314453125, -6.50775146484375, -54.53564453125, 7.4640960693359375, 5.5634002685546875, 36.85443115234375, 71.27899169921875, -6.851051330566406, 14.659507751464844, 8.746898651123047, 27.06951904296875, 42.26361083984375, 55.70719909667969, 61.961517333984375, 5.617034912109375, -9.996688842773438, 21.506683349609375, 0.9976654052734375, 58.10746765136719, 7.093090057373047, -7.191307067871094, 37.898658752441406, -85.607666015625, 37.94158935546875, 75.79994201660156, 50.53580856323242, -0.23936843872070312, 71.83673095703125, -7.6270599365234375, 75.0062255859375, 15.679634094238281, 78.3016357421875, 1.4965286254882812, 3.00433349609375, 16.636993408203125, 19.118408203125, 21.7362060546875, 50.94696044921875, 13.456939697265625, 12.84332275390625, 60.83495330810547, 92.6090087890625, 54.09844970703125, 32.216888427734375, 72.51924133300781, 59.110748291015625, 1.582000732421875, 54.54779052734375, -17.25921630859375, -6.444793701171875, 15.163421630859375, 7.107944488525391, 85.41845703125, -10.739990234375, 9.747665405273438, 9.930673599243164, -13.856943130493164, 14.2120361328125, -23.778656005859375, -0.2594451904296875, 65.0283203125, -17.511978149414062, -6.0166473388671875, 25.445571899414062, 65.70120239257812, 54.03724670410156, 7.2404632568359375, 23.15399169921875, 6.624467849731445, -14.660537719726562, 3.2068405151367188, 56.31394958496094, 22.999267578125, 81.63421630859375, 86.52044677734375, -6.52630615234375, -4.78887939453125, 28.909149169921875, 78.78834533691406, -2.7305068969726562, 67.45751953125, 27.230545043945312, 3.608551025390625, 70.77780151367188, 40.378997802734375, 44.02748107910156, -45.70579528808594, 7.349822998046875, 8.239288330078125, -4.155548095703125, 13.364959716796875, 46.704803466796875, -11.4625244140625, 55.17376708984375, 45.712158203125, 9.87396240234375, 85.318603515625, 57.05693054199219, 9.728446960449219, -0.34053611755371094, 12.340927124023438, 68.33238220214844], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000222.npy"}
|
|
{"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 27.175519943237305, "std": 37.30398178100586, "min": -61.67662811279297, "p10": -23.137763214111327, "median": 26.119552612304688, "p90": 72.82860107421875, "max": 114.11285400390625, "pos_frac": 0.7890625, "sample": [10.645561218261719, -19.672836303710938, 4.936119079589844, 69.13153076171875, 7.94647216796875, 11.612686157226562, 0.5757293701171875, -15.806680679321289, 13.647502899169922, 72.02264404296875, 19.06512451171875, 69.9697265625, 33.50274658203125, 9.33648681640625, 7.167022705078125, 102.51669311523438, 77.35295104980469, -61.607635498046875, 58.54627227783203, -26.40057373046875, 27.04736328125, 73.69381713867188, 3.0811767578125, 27.981201171875, 18.542739868164062, -24.028106689453125, 11.43463134765625, 52.37908935546875, 71.253662109375, -12.130157470703125, 51.5399169921875, 21.043914794921875, 45.77729797363281, 68.44586181640625, 53.31840515136719, 69.09452819824219, 103.1514892578125, -38.47747802734375, 46.271446228027344, -56.485870361328125, 13.310989379882812, 34.31036376953125, 8.662933349609375, 49.1051139831543, 76.13145446777344, 35.932525634765625, 85.62785339355469, 50.96630859375, 59.04449462890625, 52.164276123046875, -1.90728759765625, 18.9063720703125, 64.5450439453125, 94.13552856445312, 58.300933837890625, 32.50421142578125, 0.24395751953125, 75.27902221679688, 17.749366760253906, 41.36981201171875, 37.232635498046875, 23.757530212402344, 25.38018798828125, 114.11285400390625, 8.944068908691406, 59.42445373535156, 21.5133056640625, 52.37237548828125, 53.56621551513672, 1.2194061279296875, -10.795013427734375, 31.21026611328125, 29.723907470703125, 72.45779418945312, -23.85832977294922, 16.095806121826172, -27.143844604492188, -25.444198608398438, 33.45928955078125, -12.037834167480469, 17.287498474121094, -33.104766845703125, 1.118255615234375, 3.2699108123779297, 64.31336975097656, 45.6348876953125, -11.81903076171875, -24.625808715820312, 50.496368408203125, -8.481590270996094, -61.67662811279297, 41.39411163330078, 15.6710205078125, 77.5989990234375, 63.97503662109375, 87.57177734375, 64.85748291015625, 45.731834411621094, 46.218902587890625, 41.634246826171875, -0.46511077880859375, 46.24627685546875, 26.858917236328125, 51.43561553955078, 50.021240234375, 7.5068359375, 65.31304931640625, 97.2174072265625, -1.3738136291503906, 0.708099365234375, 79.4595947265625, -39.551788330078125, -8.1280517578125, 0.558197021484375, 72.4273681640625, 39.654991149902344, -46.279537200927734, 1.2947998046875, 23.901947021484375, 11.30328369140625, -21.945274353027344, 61.082000732421875, 11.619979858398438, 34.49517822265625, 16.32994842529297, 11.87579345703125, -22.828948974609375, -22.333770751953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000223.npy"}
|
|
{"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 26.484888076782227, "std": 34.56409454345703, "min": -61.17875671386719, "p10": -16.225338745117188, "median": 25.356443405151367, "p90": 73.81914520263672, "max": 109.94512939453125, "pos_frac": 0.8046875, "sample": [91.85284423828125, 109.94512939453125, 23.357955932617188, 35.63232421875, 3.0730934143066406, 4.955982208251953, -48.470306396484375, 29.763534545898438, 61.10603332519531, 11.039039611816406, 6.3694305419921875, 53.0218620300293, 16.729507446289062, 37.113609313964844, 25.90630340576172, 25.58074951171875, 8.845947265625, 56.815185546875, 1.9293594360351562, -32.86338806152344, 30.967742919921875, -11.245651245117188, 90.80372619628906, 84.80755615234375, 42.22962951660156, 42.889495849609375, 52.08202362060547, 13.685226440429688, 19.049819946289062, 62.199371337890625, 79.37350463867188, 64.84721374511719, -60.589385986328125, 51.31010437011719, 25.021255493164062, 20.97271728515625, 26.263290405273438, 46.74249267578125, 50.48480224609375, 25.56524658203125, 3.43841552734375, -10.994293212890625, 60.017616271972656, 33.90618896484375, 43.55561828613281, 27.969131469726562, 43.950103759765625, -19.999053955078125, 65.93014526367188, -2.4409046173095703, 35.25621032714844, 38.060546875, 0.0, 77.18814086914062, 62.75662612915039, 13.790081024169922, -6.728179931640625, -36.73356628417969, 4.9304351806640625, 26.2110595703125, 14.93035888671875, 24.754913330078125, 4.123331069946289, 57.161163330078125, -0.3893280029296875, 56.9932861328125, -15.949188232421875, -6.952178955078125, 14.65191650390625, -35.12550354003906, 0.071990966796875, 89.1868896484375, 9.254592895507812, 53.769622802734375, 12.6319580078125, 13.395782470703125, -4.7803192138671875, 45.84844970703125, 9.372161865234375, 80.30233764648438, -21.736923217773438, -10.986740112304688, -36.07456970214844, 10.269180297851562, 14.067901611328125, 66.62443542480469, 36.5316162109375, -16.86968994140625, 75.08795166015625, -61.17875671386719, 67.26399230957031, 18.440704345703125, 49.78477478027344, 73.35678100585938, 25.81182861328125, 16.623321533203125, 47.216949462890625, 29.78070068359375, 36.3441162109375, 6.33111572265625, -20.023887634277344, 28.151123046875, 6.275787353515625, 16.397216796875, 18.928604125976562, 25.147640228271484, 105.80218505859375, -30.80156707763672, 84.12429809570312, 43.320220947265625, 15.40899658203125, -33.275848388671875, 6.169223785400391, 3.7294921875, -6.9139404296875, 43.664825439453125, 52.487213134765625, 7.2030181884765625, 49.0823974609375, -5.988555908203125, 6.472400665283203, 72.40737915039062, 48.421112060546875, 44.689666748046875, 74.89799499511719, 55.11134338378906, 4.354473114013672, 85.6531982421875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000224.npy"}
|
|
{"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 20.501537322998047, "std": 42.31202697753906, "min": -75.40548706054688, "p10": -33.39769668579101, "median": 14.442035675048828, "p90": 79.885986328125, "max": 114.45120239257812, "pos_frac": 0.6640625, "sample": [79.3385009765625, -32.88787841796875, 54.75799560546875, 77.55758666992188, 68.78923034667969, 8.369903564453125, -41.489681243896484, -52.56787109375, 88.26776123046875, 81.52529907226562, 23.574691772460938, 71.06948852539062, 29.692184448242188, -7.924713134765625, 89.35885620117188, 24.9405517578125, 91.08816528320312, 62.264373779296875, 6.013149261474609, 10.424072265625, -9.9102783203125, -12.417190551757812, 19.43408966064453, -6.9698486328125, 77.53842163085938, -1.853302001953125, 87.94378662109375, -25.349044799804688, 2.0443115234375, -43.7489013671875, 2.8692550659179688, -51.622222900390625, 60.45013427734375, 48.77374267578125, 9.049072265625, 14.974510192871094, 56.9423828125, -3.9598655700683594, 1.383270263671875, -17.053268432617188, 42.88922119140625, 89.97418212890625, -34.58727264404297, -41.39033508300781, -55.971832275390625, -42.87999725341797, 46.57484436035156, 62.405731201171875, -10.440998077392578, 29.230010986328125, -25.245193481445312, 21.447998046875, 41.453407287597656, 67.0179443359375, -5.92706298828125, 12.4573974609375, -27.5032958984375, 27.884765625, 73.89462280273438, 53.66807556152344, 58.36933898925781, 84.29936218261719, 94.114501953125, -15.75762939453125, 24.003173828125, -8.533126831054688, -20.988739013671875, 48.659637451171875, 30.83416748046875, 91.74087524414062, 3.7931671142578125, 77.9820556640625, 28.01165771484375, 55.67491149902344, -16.188751220703125, 16.117141723632812, 72.28278350830078, 69.80496215820312, -21.966522216796875, 57.411407470703125, -9.666351318359375, -43.511962890625, -22.14228057861328, 7.465776443481445, -53.2774658203125, -23.15777587890625, 22.314943313598633, 0.019134521484375, 40.03785705566406, -75.40548706054688, 28.8685302734375, -3.76654052734375, -28.430908203125, -8.0684814453125, 44.064666748046875, -25.64605712890625, 1.0089111328125, 10.501106262207031, 114.45120239257812, 45.990875244140625, 3.2911758422851562, 52.20604705810547, 79.52947998046875, 80.71783447265625, -18.8426513671875, 43.65803527832031, -43.749778747558594, 13.561111450195312, 10.405059814453125, 101.5177001953125, -10.751152038574219, 28.993896484375, 60.991607666015625, 62.496551513671875, 3.09893798828125, 26.99627685546875, 13.909561157226562, -14.458663940429688, 28.850696563720703, -37.05790710449219, 8.2247314453125, 1.6921234130859375, -12.862388610839844, 5.91064453125, -24.296600341796875, 17.618057250976562, 30.383590698242188, 95.2158203125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000225.npy"}
|
|
{"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 18.484375, "std": 36.43391418457031, "min": -71.91799926757812, "p10": -29.34130630493164, "median": 12.569480895996094, "p90": 68.10122680664062, "max": 100.19888305664062, "pos_frac": 0.6953125, "sample": [12.1624755859375, 100.19888305664062, -10.71209716796875, 43.633331298828125, 70.41513061523438, 79.96426391601562, 8.394973754882812, -1.6693649291992188, 19.423675537109375, 64.17233276367188, 84.09529113769531, 41.07658004760742, 5.47528076171875, 80.916259765625, -0.040435791015625, 10.003856658935547, 28.75213623046875, -30.59051513671875, 66.50204467773438, 74.74240112304688, 23.185218811035156, -0.20130538940429688, 54.14497375488281, 47.295562744140625, 0.7649688720703125, 17.816055297851562, -12.697359085083008, -56.464202880859375, 57.28156280517578, 25.865509033203125, 0.50738525390625, 62.92375183105469, 40.7662353515625, 46.43756103515625, -7.3716583251953125, 11.6312255859375, 65.5938720703125, -71.91799926757812, -2.231109619140625, 40.38603210449219, -47.52870178222656, 12.976486206054688, -28.50072479248047, -2.4123191833496094, 14.762481689453125, 36.7957763671875, 14.143341064453125, -41.846405029296875, -36.0263671875, 47.03839111328125, 7.970947265625, -39.7242431640625, 69.40518188476562, 3.3827247619628906, 8.261093139648438, -45.632537841796875, 3.2506179809570312, 21.3348388671875, 0.29128265380859375, -1.6376419067382812, -11.632110595703125, 30.744781494140625, 18.11351776123047, -36.05741882324219, 33.542236328125, -5.245109558105469, 9.8326416015625, 67.54238891601562, 6.806423187255859, 7.85943603515625, -19.933746337890625, 5.569314956665039, 21.9818115234375, -16.600753784179688, 25.581268310546875, 4.17529296875, 6.338165283203125, 49.6148681640625, 6.988578796386719, 77.07534790039062, -52.00349426269531, 65.86307525634766, 80.29165649414062, -53.151390075683594, 83.465576171875, -0.521209716796875, 11.518623352050781, 11.265495300292969, -0.9117431640625, 65.81890869140625, 54.595733642578125, 65.51953125, 61.08726501464844, 26.62664794921875, 51.23504638671875, 17.555908203125, -11.601531982421875, 9.861251831054688, 54.545249938964844, 13.464591979980469, 86.56254577636719, 18.3397216796875, 78.85711669921875, 1.25079345703125, 5.93084716796875, 17.888809204101562, 56.94822692871094, 26.203125, 54.500030517578125, -13.9449462890625, 35.968505859375, -35.39619445800781, -1.5095024108886719, 50.65625, 9.014442443847656, -2.2238693237304688, 56.29804992675781, -14.327108383178711, -23.25482177734375, 82.9166259765625, 22.09747314453125, -28.805931091308594, -39.10687255859375, 18.192142486572266, -24.256439208984375, -5.20391845703125, 17.155670166015625, -8.509788513183594], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000226.npy"}
|
|
{"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 26.228845596313477, "std": 37.556243896484375, "min": -78.183837890625, "p10": -17.552810668945302, "median": 24.447097778320312, "p90": 78.81250686645507, "max": 108.71646118164062, "pos_frac": 0.7734375, "sample": [-52.50152587890625, 35.877899169921875, 13.40863037109375, 23.008087158203125, 53.67108154296875, 37.20696258544922, 48.485198974609375, 20.498565673828125, 1.1222915649414062, 24.3304443359375, 76.67254638671875, 45.22541809082031, 11.586105346679688, -10.6202392578125, 5.7325286865234375, 31.67339515686035, 0.0, 83.764892578125, -78.183837890625, -50.411136627197266, 31.90478515625, -49.7408447265625, 61.52914047241211, 80.39808654785156, 108.71646118164062, 105.715087890625, 62.8470458984375, -30.400726318359375, 64.00285339355469, -14.608154296875, -6.170318603515625, 68.18568420410156, 12.639450073242188, 14.438873291015625, 26.73681640625, -13.8212890625, 22.79230499267578, 30.094924926757812, 34.660736083984375, 81.53494262695312, 83.6695556640625, 67.787841796875, 12.94232177734375, 42.0972900390625, 30.1971435546875, -8.24249267578125, 30.74853515625, 3.672760009765625, -11.740463256835938, 22.830535888671875, 81.71939086914062, 23.29839324951172, -7.34820556640625, -24.423675537109375, 27.713043212890625, 96.60343933105469, 4.1591796875, 49.49603271484375, 84.20259094238281, 33.80062484741211, 85.82647705078125, -47.29280090332031, -8.87762451171875, 56.745941162109375, -69.65057373046875, 19.20184326171875, 22.949310302734375, 86.6004638671875, -27.638458251953125, 31.88189697265625, 52.88026428222656, 24.563751220703125, 5.071990966796875, 50.012969970703125, 74.71334838867188, 15.965606689453125, 40.095855712890625, 45.79557800292969, -35.17852783203125, 59.91758728027344, 85.13339233398438, 3.99493408203125, 17.737350463867188, 75.05403137207031, 1.7896728515625, 42.63697814941406, 12.4654541015625, 24.663970947265625, 15.823287963867188, 12.21490478515625, -26.103515625, 46.95478820800781, 15.087188720703125, -27.049720764160156, 71.19161987304688, 34.8939208984375, 61.4996337890625, 5.009765625, -11.717300415039062, 78.13297271728516, -1.2064208984375, 82.990478515625, 2.278047561645508, 65.796875, -32.802268981933594, 63.510589599609375, -2.767911911010742, -2.830963134765625, 2.508056640625, 6.6887054443359375, 37.29144287109375, -11.268966674804688, 70.07998657226562, 64.08409118652344, 12.371246337890625, 27.313873291015625, 30.053680419921875, 36.066680908203125, 49.461761474609375, -5.0699462890625, -9.147216796875, 14.607162475585938, 17.83856201171875, 41.68857955932617, 5.5316009521484375, 6.195356369018555, 50.131065368652344, 57.71290588378906], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000227.npy"}
|
|
{"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 26.50731658935547, "std": 36.516136169433594, "min": -47.60772705078125, "p10": -18.686927795410156, "median": 23.647384643554688, "p90": 72.7722152709961, "max": 106.3984375, "pos_frac": 0.71875, "sample": [-28.233489990234375, -29.44831085205078, 64.0296630859375, 67.96896362304688, 53.92900085449219, -5.086698532104492, 92.46197509765625, 24.584747314453125, 40.3197021484375, 72.85041809082031, 11.225914001464844, -39.486244201660156, -3.648101806640625, 86.158447265625, 16.122039794921875, 32.25103759765625, -2.741546630859375, 56.396995544433594, -23.245559692382812, 32.289093017578125, 64.28766632080078, 16.54582977294922, -47.181785583496094, -18.3333740234375, 12.231353759765625, 59.78056335449219, 22.71002197265625, 2.545318603515625, -11.69244384765625, 19.554664611816406, 11.381927490234375, 19.817672729492188, -19.511886596679688, 48.407958984375, 46.874481201171875, 69.21208190917969, 40.670806884765625, 18.682952880859375, -14.01837158203125, -6.909027099609375, 67.1280517578125, -4.680694580078125, -14.59588623046875, -8.6832275390625, 47.837554931640625, 6.984710693359375, -1.49810791015625, 16.84548568725586, 33.341094970703125, 64.14067077636719, 33.490966796875, 94.09573364257812, 45.668792724609375, 10.693901062011719, 85.56753540039062, 34.3299560546875, 59.226318359375, 28.516387939453125, 57.351043701171875, 86.6446533203125, 0.344146728515625, -34.07575988769531, -8.035362243652344, -2.210542678833008, 52.334625244140625, 2.3667449951171875, 0.0, 24.727783203125, 22.346710205078125, 11.754661560058594, 11.791412353515625, 92.28109741210938, 25.202293395996094, -2.6353302001953125, -23.159408569335938, 65.796630859375, -15.49835205078125, 71.65472412109375, -5.8159942626953125, 68.54666137695312, -14.937255859375, -37.75041198730469, 16.00115966796875, 51.845703125, 69.27059936523438, 82.92109680175781, 3.56866455078125, -23.046234130859375, 87.16502380371094, 25.445953369140625, -17.8099365234375, 72.74871826171875, 21.672576904296875, 69.00714111328125, 33.590972900390625, 57.0194091796875, -10.297393798828125, 3.456573486328125, 11.959613800048828, 72.82704162597656, 58.028900146484375, -45.50709533691406, -11.780265808105469, 32.455810546875, 57.57331848144531, 50.983123779296875, 62.280792236328125, -47.60772705078125, -15.740081787109375, -0.76922607421875, 15.670684814453125, 51.317535400390625, 7.03173828125, 37.17008972167969, 37.23028564453125, 94.1055908203125, 67.9661865234375, 41.9144287109375, 11.579479217529297, -33.77783203125, 46.5064697265625, 58.57598876953125, 81.70246887207031, 43.362091064453125, 106.3984375, 7.2073974609375, 31.194549560546875, 19.32836151123047], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000228.npy"}
|
|
{"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 34.63862228393555, "std": 39.183040618896484, "min": -62.02156066894531, "p10": -16.558622741699217, "median": 34.66752243041992, "p90": 85.77171630859374, "max": 137.43112182617188, "pos_frac": 0.8046875, "sample": [60.923828125, -8.597412109375, 52.130279541015625, 55.3582763671875, -7.86749267578125, -56.252410888671875, 33.23085021972656, 59.44044494628906, 37.11114501953125, -38.635704040527344, 107.146728515625, 93.1419677734375, 34.8839111328125, 2.467853546142578, 99.64329528808594, 71.72967529296875, 7.4069671630859375, 12.064788818359375, 23.618240356445312, -1.1075439453125, 23.808242797851562, 17.92706298828125, 67.75799560546875, -1.649169921875, 20.377853393554688, -24.414718627929688, 8.715934753417969, 7.7297515869140625, 83.46896362304688, 42.04951477050781, 31.67474365234375, 61.34698486328125, 93.27423095703125, 56.59466552734375, 32.02601623535156, 87.6766357421875, 19.888092041015625, 17.0526123046875, 105.09280395507812, -62.02156066894531, 18.5101318359375, -47.92535400390625, 40.88731384277344, 60.6527099609375, 36.80720901489258, 99.16815185546875, 137.43112182617188, 76.17271423339844, 59.33189392089844, 67.14663696289062, 45.335540771484375, -1.028533935546875, 31.52276611328125, 6.49102783203125, 61.61719512939453, 64.95108032226562, 70.40557861328125, 9.623321533203125, 8.483154296875, -37.992156982421875, 25.23961067199707, -43.92299270629883, 76.39425659179688, 67.60162353515625, -7.101005554199219, -22.7890625, 58.33062744140625, 11.340324401855469, 56.04302215576172, -2.254547119140625, 1.542694091796875, 34.333290100097656, -0.19406700134277344, 6.65692138671875, 64.01028442382812, 39.499542236328125, 23.47618865966797, 69.10391998291016, 81.49078369140625, 37.51654052734375, 73.77838134765625, 53.30662536621094, 28.27777099609375, 66.71229553222656, 24.77044677734375, 9.16693115234375, -35.96195983886719, 78.76451110839844, 98.93365478515625, 34.451133728027344, 98.69161987304688, 67.50294494628906, 8.654678344726562, 21.3485107421875, 13.298431396484375, 74.79641723632812, 34.981170654296875, 83.51614379882812, 26.060638427734375, 24.385238647460938, 46.64830017089844, -10.74530029296875, 35.454742431640625, 88.35934448242188, -3.814117431640625, 7.460723876953125, 37.890541076660156, -15.833831787109375, 10.040691375732422, 84.955322265625, 46.90721893310547, -10.015810012817383, 33.34222412109375, 43.567230224609375, 40.564056396484375, 33.008270263671875, 64.25244140625, -19.565673828125, 49.11652374267578, -29.1571044921875, 92.55851745605469, 25.93341064453125, 44.32209777832031, 47.31205749511719, 58.85650634765625, -37.43458938598633, 92.45028686523438, -18.249801635742188], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000229.npy"}
|
|
{"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 28.080293655395508, "std": 40.04191970825195, "min": -121.49603271484375, "p10": -14.113214874267577, "median": 22.17057991027832, "p90": 80.5010284423828, "max": 119.69171142578125, "pos_frac": 0.7890625, "sample": [77.74822998046875, 50.78422546386719, 32.86553955078125, 65.54407501220703, 119.69171142578125, 48.068328857421875, 90.51722717285156, -4.374420166015625, 83.61727905273438, 52.13865661621094, 14.353492736816406, 25.406585693359375, 16.929969787597656, -48.10950469970703, -121.49603271484375, 63.73356628417969, 72.031005859375, -2.805757522583008, -5.25927734375, -54.63836669921875, 0.436492919921875, 11.488903045654297, 91.67041015625, -23.098831176757812, 0.0, 79.93295288085938, -1.941192626953125, 10.47153091430664, 22.235015869140625, 64.35618591308594, 55.62323760986328, 71.63970947265625, 1.378936767578125, 34.49769592285156, -0.633148193359375, -35.0987548828125, 61.36053466796875, 61.67951965332031, 43.199951171875, 5.7500762939453125, 1.4509696960449219, -25.32489013671875, 9.35504150390625, 36.240081787109375, 14.323463439941406, 73.76983642578125, 40.83623504638672, 22.71600341796875, 68.3380126953125, -13.56121826171875, -3.777111053466797, 24.110992431640625, -27.791748046875, 46.534271240234375, 65.6622314453125, -17.985000610351562, 44.04808044433594, 5.34405517578125, 69.55014038085938, 21.302230834960938, -30.367141723632812, 46.107757568359375, 100.52935791015625, 56.661712646484375, 12.295318603515625, 4.199920654296875, -9.18438720703125, 69.19113159179688, 0.45831298828125, 55.49971008300781, 40.213623046875, 5.964019775390625, -9.883979797363281, -9.487991333007812, 23.1826171875, 14.67724609375, 6.96051025390625, 22.106143951416016, 95.32275390625, -0.2520751953125, -1.1009807586669922, 56.267242431640625, 20.707183837890625, -15.555885314941406, 14.524917602539062, -92.06739807128906, 4.9299468994140625, -15.401206970214844, 89.78512573242188, 62.335784912109375, 97.28955078125, 39.46501541137695, 13.683975219726562, 5.9315032958984375, 88.35317993164062, 82.38616943359375, 1.632162094116211, 12.792655944824219, 69.24356842041016, 57.10308837890625, 77.76390075683594, 2.039041519165039, 51.307228088378906, 10.034835815429688, -25.296852111816406, 56.39149475097656, -0.09368896484375, 81.8265380859375, 50.7828369140625, 4.892772674560547, 54.33149337768555, 77.68199157714844, 37.70947265625, 2.7676773071289062, 42.19706726074219, 31.28649139404297, 5.573215484619141, 91.82740783691406, 115.54962158203125, 13.999526977539062, 1.5051956176757812, 23.56658935546875, 10.408966064453125, 7.349395751953125, 26.7279052734375, 48.978370666503906, 7.481666564941406, 12.349586486816406], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000230.npy"}
|
|
{"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 22.426101684570312, "std": 35.4854850769043, "min": -69.32496643066406, "p10": -22.57915802001953, "median": 18.058555603027344, "p90": 64.6271484375, "max": 105.04000854492188, "pos_frac": 0.78125, "sample": [24.669219970703125, 15.934295654296875, -17.33905029296875, -32.6844482421875, 54.726966857910156, 62.580535888671875, 38.5413818359375, -22.287750244140625, 49.37646484375, -69.32496643066406, 72.21345520019531, -26.736541748046875, 50.573333740234375, 52.1544189453125, 0.6670684814453125, 50.2861328125, 9.324661254882812, 7.057376861572266, 19.21624755859375, -13.669708251953125, -23.259109497070312, -23.617813110351562, -8.3695068359375, 40.371490478515625, 10.722320556640625, 50.88987731933594, 16.179489135742188, 56.72315216064453, 1.7877197265625, -44.5885009765625, 51.16131591796875, 64.02420043945312, 47.5599365234375, 3.8389549255371094, 53.1549072265625, -10.4171142578125, 88.30899047851562, 42.02044677734375, 4.7713775634765625, 36.3780517578125, 0.618988037109375, 31.094482421875, -2.663299560546875, 1.8912200927734375, 79.2574462890625, 55.4033203125, -5.457275390625, 26.9693603515625, 9.4483642578125, 25.751708984375, 65.0733642578125, -9.22686767578125, 10.593124389648438, 12.679412841796875, 6.53509521484375, 13.254058837890625, 61.290771484375, -14.328125, 22.556854248046875, 96.204833984375, -22.211761474609375, 3.374317169189453, 0.902313232421875, 33.2593994140625, 1.3946456909179688, 43.59942626953125, 33.469818115234375, 26.254302978515625, 18.11016845703125, 27.897003173828125, 46.45320129394531, 34.8095703125, 97.8663330078125, -43.45965576171875, -56.66542053222656, 1.30218505859375, 16.5235595703125, -7.054534912109375, 59.4122314453125, 36.421142578125, 16.16180419921875, 11.597625732421875, 26.71875, 20.051300048828125, 1.7745285034179688, 14.886199951171875, 3.68975830078125, 79.80682373046875, 37.81060791015625, 11.85186767578125, 54.54380798339844, 50.48738098144531, 39.561004638671875, 50.2578125, -30.797622680664062, 48.61041259765625, 14.91672134399414, 41.42610168457031, 100.60552978515625, 0.894805908203125, -37.73040771484375, 22.0462646484375, 17.960922241210938, 105.04000854492188, 53.55308532714844, -6.792938232421875, 3.1812515258789062, 58.10054016113281, 10.9681396484375, 32.071044921875, -4.486236572265625, 15.339452743530273, 48.09458923339844, -49.55645751953125, 48.440460205078125, 92.84208679199219, 6.658355712890625, 80.91543579101562, 18.006942749023438, 41.88481140136719, -12.559776306152344, 5.30401611328125, 71.09510803222656, -45.49456787109375, 64.4359130859375, 0.0, -52.94075012207031, 67.78408813476562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000231.npy"}
|
|
{"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 25.189655303955078, "std": 38.52091979980469, "min": -56.648956298828125, "p10": -22.320690917968747, "median": 19.40985107421875, "p90": 74.238134765625, "max": 125.57980346679688, "pos_frac": 0.7109375, "sample": [-17.09295654296875, 74.19113159179688, -2.4681015014648438, 77.86569213867188, 25.887191772460938, 42.54669189453125, 63.673828125, 32.636322021484375, 6.097320556640625, -11.063003540039062, 33.81825256347656, 81.9061279296875, 26.108612060546875, -16.90570068359375, -24.168594360351562, 28.46698760986328, 9.896537780761719, 5.7245330810546875, 18.76336669921875, 16.504348754882812, 21.8580322265625, 8.762954711914062, -51.155426025390625, 45.34666442871094, 65.60479736328125, 53.71527099609375, 41.16437530517578, 11.369110107421875, 1.3654861450195312, 95.8648681640625, 23.153533935546875, -8.419303894042969, -33.67962646484375, 47.149749755859375, 58.475616455078125, 72.40509033203125, -32.915283203125, 28.841949462890625, 12.013916015625, 69.71405029296875, 23.668548583984375, -25.8173828125, 1.3792800903320312, 13.973152160644531, 13.712615966796875, 2.307830810546875, 54.82414245605469, 11.254486083984375, 64.55195617675781, 0.44915008544921875, 26.238616943359375, -23.220001220703125, 67.14923858642578, 22.790313720703125, 2.312774658203125, 61.77716064453125, 19.637863159179688, 72.69603729248047, -1.98834228515625, 64.80078125, 12.268646240234375, 103.53753662109375, 75.96806335449219, -23.6798095703125, 63.32777404785156, 56.34642028808594, -3.44140625, 11.86865234375, -5.291656494140625, -2.6010971069335938, 11.196357727050781, 74.01513671875, 8.567413330078125, -17.9566650390625, 93.3517837524414, 95.853515625, -3.6296539306640625, -21.935272216796875, 9.58770751953125, 45.84197998046875, 3.0133590698242188, 6.4197845458984375, -18.9130859375, 52.09752655029297, 69.51226806640625, -14.958606719970703, -5.6587982177734375, 48.75811767578125, -5.97967529296875, -26.15753173828125, 66.76263427734375, 41.405426025390625, 0.0, 86.005126953125, 40.36732482910156, 5.695075988769531, 58.362579345703125, 24.31824493408203, 65.59223937988281, 60.59100341796875, 64.93965148925781, 19.181838989257812, 99.69024658203125, -56.648956298828125, 58.82972717285156, 125.57980346679688, 12.180908203125, -16.89691162109375, 64.79843139648438, 31.684951782226562, -4.576904296875, 92.912109375, -44.684791564941406, -9.835159301757812, -19.40009307861328, -38.678863525390625, 40.229331970214844, -4.830604553222656, 38.3941650390625, -36.38734436035156, 75.37152099609375, -2.515960693359375, -49.54560852050781, 74.34780883789062, 52.86747741699219, 50.148399353027344, 8.995147705078125, -17.82562255859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000232.npy"}
|
|
{"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 27.052589416503906, "std": 38.67059326171875, "min": -57.53120422363281, "p10": -21.18304443359375, "median": 20.53303337097168, "p90": 81.42067413330078, "max": 114.39605712890625, "pos_frac": 0.7890625, "sample": [-36.18357849121094, 64.94820404052734, 2.673126220703125, -8.595458984375, 2.7186813354492188, 77.25140380859375, 16.038742065429688, 44.039398193359375, 8.323875427246094, 10.271812438964844, 35.903472900390625, 30.63840103149414, 75.7552490234375, 45.261260986328125, 30.829864501953125, 1.449432373046875, 43.50225830078125, 61.97021484375, 28.844085693359375, 104.43179321289062, 71.31484985351562, 70.29657745361328, 78.88052368164062, 62.11891174316406, 41.148048400878906, 82.85968017578125, 71.48507690429688, 16.666336059570312, 62.338287353515625, -26.888992309570312, 24.855316162109375, -11.289833068847656, 66.97515869140625, 3.1226959228515625, 81.69102478027344, 27.56939697265625, 18.203994750976562, -44.55931854248047, 50.458560943603516, 93.1759033203125, -16.229354858398438, -4.558294296264648, 13.73211669921875, 13.8997802734375, -17.516448974609375, 104.10052490234375, 36.01741027832031, 18.3680419921875, 19.21746826171875, 13.32403564453125, 86.22418212890625, 100.71458435058594, 17.41534423828125, -14.104934692382812, 25.6790771484375, -5.560150146484375, 107.42266845703125, 16.63836669921875, 65.37713623046875, 7.60797119140625, 33.50750732421875, 69.03765869140625, -40.249237060546875, 15.671588897705078, 7.871551513671875, -20.964599609375, -22.813720703125, 20.722652435302734, -55.255401611328125, -36.55780029296875, 68.86624145507812, 83.5162353515625, 81.3048095703125, 42.08650207519531, 99.1639404296875, 43.624664306640625, 18.25836181640625, 5.990196228027344, 14.943742752075195, 4.7991943359375, 8.941448211669922, 4.2054443359375, 12.69097900390625, 48.31575012207031, -21.6927490234375, 4.843667984008789, 114.39605712890625, 32.283416748046875, -30.247344970703125, 35.99786376953125, 30.768878936767578, -26.261016845703125, -52.69354248046875, 98.12640380859375, 19.475868225097656, -18.34136962890625, 2.010040283203125, 59.981231689453125, -0.509033203125, 71.86387634277344, -10.987396240234375, 4.639045715332031, -10.1470947265625, -57.53120422363281, 50.001800537109375, 60.26202392578125, 13.678573608398438, 49.804534912109375, 29.70123291015625, -48.40565490722656, 0.753936767578125, 57.43865966796875, 48.851043701171875, -3.8529815673828125, 35.62254333496094, 22.616256713867188, -18.77154541015625, 85.53140258789062, 25.73694610595703, 36.68206787109375, 32.798614501953125, 4.483980178833008, 0.53289794921875, 45.07257080078125, 53.83837890625, 18.161590576171875, 20.343414306640625, 19.93206787109375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000233.npy"}
|
|
{"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 27.95116424560547, "std": 43.84157943725586, "min": -72.23269653320312, "p10": -28.756335449218746, "median": 26.978557586669922, "p90": 84.37402038574217, "max": 133.9078369140625, "pos_frac": 0.7734375, "sample": [-31.360336303710938, -65.38507080078125, -21.4844970703125, 3.1041688919067383, 49.5819091796875, 71.20846557617188, 87.1080322265625, 41.545623779296875, -20.62713623046875, 92.41122436523438, 71.01319885253906, -64.96426391601562, 67.52288818359375, 53.062217712402344, -16.68524169921875, 47.818756103515625, 27.396263122558594, 17.8138427734375, 29.913345336914062, 91.92774963378906, 56.19712829589844, 30.2296142578125, 50.258209228515625, 96.9974365234375, 12.7615966796875, 1.4769973754882812, 50.67841339111328, 60.33769226074219, 66.09828186035156, 33.333892822265625, 17.26806640625, 46.78570556640625, 70.6263427734375, 103.64212036132812, 66.08206176757812, 16.668975830078125, 31.1229248046875, 5.413421630859375, 2.08355712890625, 29.46728515625, 50.659393310546875, 13.95965576171875, 5.389801025390625, -27.640335083007812, 20.322723388671875, -16.78375244140625, 56.743743896484375, 58.0484504699707, -10.006546020507812, 77.96723937988281, 16.502838134765625, 8.36077880859375, 97.741455078125, 44.07803726196289, -5.765602111816406, 114.41806030273438, 66.76669311523438, 39.251564025878906, 63.05251693725586, 46.032379150390625, 9.624702453613281, 9.573577880859375, 83.20230102539062, 9.9715576171875, 7.45770263671875, -1.005706787109375, -23.360595703125, 0.0, 28.68212890625, -8.456527709960938, 12.683578491210938, 33.60038757324219, -58.390045166015625, 8.57440185546875, 70.65115356445312, 105.36831665039062, 77.09112548828125, 18.6654052734375, -48.4434814453125, 45.536224365234375, 0.43395042419433594, 58.6845703125, 83.14512634277344, -18.88287353515625, 41.075775146484375, 26.56085205078125, 57.88972473144531, 14.2713623046875, -72.23269653320312, 101.196044921875, -45.831573486328125, 95.1334228515625, 12.50067138671875, 17.67059326171875, 68.12316131591797, 0.3864593505859375, 62.72096633911133, 31.940185546875, 0.18367385864257812, 7.3076171875, 56.1654052734375, 0.0, 13.879165649414062, 26.3084716796875, 133.9078369140625, 94.37445068359375, -60.12312316894531, -16.29083251953125, 93.44210815429688, 55.5538330078125, 52.62285614013672, 1.98101806640625, -58.04588317871094, 25.836212158203125, 51.73577880859375, 77.01638793945312, -32.153167724609375, -22.272979736328125, -32.14459228515625, 1.36773681640625, 38.698028564453125, 2.5703887939453125, -50.43798828125, 10.606361389160156, -56.780792236328125, 78.123779296875, -9.107025146484375, 80.06243896484375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000234.npy"}
|
|
{"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 24.840576171875, "std": 42.26842498779297, "min": -93.12371826171875, "p10": -29.08737392425537, "median": 23.77255630493164, "p90": 75.15671234130859, "max": 112.4466552734375, "pos_frac": 0.6953125, "sample": [18.068603515625, 10.784149169921875, -5.44793701171875, 1.9200057983398438, 10.803668975830078, -65.7662353515625, -31.970184326171875, -5.921348571777344, 58.36520004272461, 61.25343322753906, -1.6424484252929688, 107.26666259765625, 66.64776611328125, 42.122283935546875, 39.262115478515625, 32.67866897583008, 66.34313201904297, 38.87640380859375, -1.372161865234375, -4.91802978515625, 49.18983459472656, 20.92828369140625, 35.49755859375, 47.748382568359375, 97.02670288085938, 74.73947143554688, 65.87274169921875, -15.885635375976562, 65.77672576904297, -11.640768051147461, 81.13525390625, -1.8360595703125, 41.478790283203125, 73.36052703857422, 30.36920166015625, 58.83563995361328, 87.810546875, 72.69735717773438, 6.893394470214844, 106.105712890625, -5.2917022705078125, 52.281646728515625, -21.417884826660156, -30.5296630859375, 47.05519104003906, 102.9671630859375, -68.396240234375, -28.49004554748535, -51.811920166015625, 29.350723266601562, 51.933319091796875, -14.165008544921875, -2.9746971130371094, -22.19708251953125, 31.15728759765625, 42.41145324707031, -14.142791748046875, 64.71504211425781, -9.844024658203125, -39.91705322265625, 11.19351577758789, 17.528350830078125, 74.99223327636719, -16.543487548828125, 108.22967529296875, 35.19749450683594, -10.08306884765625, -70.73400115966797, 2.7286300659179688, 33.80986022949219, -30.811614990234375, 47.837066650390625, -19.6754150390625, 56.71855163574219, 7.194614410400391, 11.88287353515625, 8.26263427734375, 79.25630187988281, 5.7911376953125, 31.1036376953125, 34.646453857421875, 22.703582763671875, -71.67172241210938, 21.14019775390625, 21.475372314453125, -24.175113677978516, 61.523399353027344, 62.37472915649414, 60.95465087890625, 2.044342041015625, 112.4466552734375, 66.08450317382812, 43.31459045410156, 109.18441772460938, 77.99945068359375, 59.185455322265625, 24.39978790283203, 18.9058837890625, 65.63542175292969, -7.456338882446289, -2.2696990966796875, 75.54049682617188, -36.28920364379883, 41.136749267578125, 0.6895751953125, 63.727821350097656, -6.8259124755859375, 24.38580322265625, 59.32122802734375, -14.6817626953125, -4.719696044921875, 23.334877014160156, 9.814460754394531, -30.48114013671875, 20.569541931152344, 9.45608139038086, 73.15635681152344, -93.12371826171875, -36.19116973876953, 3.0645904541015625, 55.000518798828125, 57.78966522216797, 21.895263671875, 78.56219482421875, 47.087371826171875, 37.13011932373047, 24.210235595703125, -2.4428749084472656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000235.npy"}
|
|
{"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 28.853857040405273, "std": 40.23786544799805, "min": -89.62603759765625, "p10": -22.953589630126952, "median": 28.919742584228516, "p90": 77.07516326904296, "max": 121.8470458984375, "pos_frac": 0.78125, "sample": [36.87062072753906, 27.7744140625, -63.024627685546875, 31.414283752441406, 47.897125244140625, 55.566375732421875, 10.439620971679688, 70.26744079589844, -12.4688720703125, -35.86663818359375, -54.97088623046875, 44.70164489746094, -22.323081970214844, 37.09844970703125, 64.80670166015625, 64.13235473632812, -0.2194805145263672, 62.179359436035156, 41.91172790527344, 18.595367431640625, 23.922637939453125, 62.42096710205078, 31.026092529296875, -26.130168914794922, 36.78961944580078, 66.4134521484375, 63.899505615234375, 7.620992660522461, 41.977691650390625, 70.23658752441406, 21.510971069335938, 31.67755126953125, 29.41913604736328, 24.80089569091797, 8.870292663574219, 24.864479064941406, 7.890625, 93.97552490234375, 69.06143188476562, 4.512451171875, 8.431869506835938, 72.22230529785156, 78.9022216796875, 35.84169006347656, 80.70709228515625, 3.341266632080078, -34.97758483886719, 6.970794677734375, 13.434585571289062, 0.0, -4.63031005859375, 32.01258850097656, 65.432861328125, 61.38250732421875, 8.740694046020508, -22.0447998046875, -68.39431762695312, 74.48234558105469, 9.75244140625, -24.424774169921875, 47.670528411865234, 80.63107299804688, 50.79888916015625, 35.12500762939453, 1.29046630859375, -29.106109619140625, 56.04583740234375, -89.62603759765625, 45.347900390625, 7.335521697998047, -21.79944610595703, 121.8470458984375, 33.51313781738281, 108.41552734375, 0.2972869873046875, -8.06451416015625, -48.033111572265625, -3.604400634765625, 76.68147277832031, 51.87652587890625, 16.486404418945312, 56.3482666015625, -19.65771484375, 57.989166259765625, 60.11524963378906, 28.42034912109375, 84.91407775878906, 102.60113525390625, -2.4832229614257812, 54.32780456542969, 54.371856689453125, 70.63900756835938, 72.17279052734375, 11.464813232421875, 8.497833251953125, 80.94406127929688, 65.00949096679688, 39.324188232421875, 69.27227783203125, 77.9937744140625, 68.7176513671875, 26.527587890625, -12.680633544921875, 19.94451904296875, 0.0, 26.029937744140625, 13.28057861328125, 24.6331787109375, -65.71473693847656, 76.27597045898438, 26.693435668945312, -6.107868194580078, -51.30537414550781, -13.096885681152344, 42.30914306640625, 68.86296844482422, 79.28523254394531, 12.7393798828125, 5.8874969482421875, -24.902198791503906, 78.29158020019531, 63.26329040527344, 80.26458740234375, 18.52435302734375, 4.159149169921875, 70.49850463867188, 12.367843627929688, 26.424652099609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000236.npy"}
|
|
{"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 24.698917388916016, "std": 40.035614013671875, "min": -94.11602783203125, "p10": -19.098538208007806, "median": 22.601226806640625, "p90": 80.55861282348633, "max": 142.672607421875, "pos_frac": 0.75, "sample": [26.29974365234375, 9.468536376953125, 30.972999572753906, 88.56015014648438, -23.436798095703125, 30.031097412109375, 27.10413360595703, 54.64324188232422, 36.064453125, 21.996871948242188, -6.112518310546875, 25.200347900390625, 97.8433837890625, -31.02893829345703, -5.815589904785156, 1.5739974975585938, 15.4608154296875, 32.01678466796875, 142.672607421875, 59.238685607910156, 47.498199462890625, 27.802200317382812, 82.83309936523438, 49.71045684814453, 2.1769866943359375, -0.0557861328125, 46.4990234375, 0.0, 49.012123107910156, 57.242706298828125, 3.9276123046875, 16.36309814453125, -23.24566650390625, 16.92481231689453, -14.631317138671875, 34.78143310546875, 21.875640869140625, -0.649078369140625, -94.11602783203125, -24.93701934814453, 3.1648941040039062, -1.76654052734375, 2.41424560546875, 57.99256896972656, 3.4523162841796875, 0.278350830078125, 73.3118896484375, 28.22857666015625, 36.48616027832031, 8.161752700805664, 6.0030670166015625, 105.5718994140625, 40.67903137207031, 14.622184753417969, 4.6076202392578125, 58.09283447265625, 79.08660888671875, 50.260223388671875, 22.3834228515625, 30.999664306640625, 99.09103393554688, 29.242095947265625, 14.35626220703125, 29.424102783203125, 44.40491485595703, -5.4451141357421875, 88.27313232421875, 67.56700134277344, -29.371536254882812, -61.00921630859375, 0.30399513244628906, 96.571044921875, -61.014801025390625, -1.24554443359375, 68.80438232421875, 4.791206359863281, -34.1673583984375, 0.417449951171875, 100.25767517089844, -11.113525390625, 22.81903076171875, 24.798492431640625, 77.89279174804688, 26.847137451171875, 78.8692626953125, -43.56024169921875, -1.7222976684570312, 4.104888916015625, 79.89864349365234, 27.507476806640625, 18.074813842773438, 112.51126098632812, -11.438064575195312, -3.258819580078125, -30.8975830078125, 88.61651611328125, -28.4200439453125, 1.13519287109375, 61.41221618652344, 32.7966194152832, 9.7257080078125, 7.1282196044921875, 62.196746826171875, -1.7208023071289062, 8.930038452148438, 13.074634552001953, 68.31837463378906, -17.321197509765625, -15.23406982421875, -7.2208251953125, 1.7200927734375, 82.12742614746094, -9.933967590332031, 10.968009948730469, 30.69634246826172, 25.735565185546875, -79.523681640625, 26.974014282226562, 66.2010498046875, 82.09854125976562, 66.39797973632812, -11.9537353515625, 30.515945434570312, 55.07928466796875, 24.960662841796875, 24.39410400390625, 18.26129150390625, 54.94366455078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000237.npy"}
|
|
{"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 28.17261505126953, "std": 41.910430908203125, "min": -64.86918640136719, "p10": -18.496707916259762, "median": 24.391773223876953, "p90": 80.33461837768553, "max": 124.31008911132812, "pos_frac": 0.6953125, "sample": [-17.652008056640625, -2.6042404174804688, 4.914878845214844, 9.097198486328125, -43.10340881347656, 121.374755859375, 65.36752319335938, 36.60955810546875, 57.47967529296875, -39.92962646484375, -6.99658203125, 33.91015625, -13.720916748046875, 8.667526245117188, -4.574363708496094, -41.49620056152344, 85.07095336914062, 77.82269287109375, -8.494865417480469, 48.211761474609375, 70.05238342285156, 67.03106689453125, 31.853759765625, 71.76387786865234, 106.68901062011719, 40.570072174072266, -11.2276611328125, 4.388359069824219, 47.52325439453125, -64.86918640136719, -45.5072021484375, -58.52265930175781, 102.32177734375, -29.176361083984375, -3.081390380859375, 92.86453247070312, 27.11175537109375, 70.81109619140625, 8.49627685546875, 57.42266845703125, 109.4346923828125, 13.5377197265625, 47.31195068359375, 104.84521484375, 66.02749633789062, 75.91525268554688, 60.518707275390625, 61.30609130859375, -0.55364990234375, 35.9771728515625, 83.05055236816406, 118.86328125, 34.558929443359375, -20.467674255371094, 15.48016357421875, 14.2301025390625, -7.12823486328125, 0.9156742095947266, -12.576919555664062, 47.090492248535156, -1.1221160888671875, 2.587310791015625, -16.3123779296875, -55.0650634765625, 71.2408447265625, 40.03507995605469, 46.62245178222656, 75.87105560302734, 54.82305908203125, 79.17064666748047, -38.054534912109375, -12.56365966796875, 30.414955139160156, -1.991607666015625, -4.532867431640625, 43.602386474609375, 87.71270751953125, -12.1640625, 57.93693542480469, 66.5477294921875, 50.842132568359375, 91.84860229492188, -4.907596588134766, -13.455188751220703, 4.6051025390625, 23.913848876953125, 60.17676544189453, -8.384370803833008, 29.593994140625, 29.909454345703125, 3.8155555725097656, 2.533905029296875, -12.238357543945312, 16.640960693359375, 124.31008911132812, -7.582427978515625, 35.7060546875, 17.40209197998047, 15.597709655761719, 26.022159576416016, 23.594009399414062, 96.58193969726562, 78.96186065673828, 20.186767578125, 49.45961380004883, 4.1680145263671875, -26.387405395507812, 69.92459106445312, 23.35003662109375, -20.76129150390625, 19.438507080078125, 77.62582397460938, 15.63739013671875, 12.522979736328125, 58.0155029296875, -10.944366455078125, 26.396804809570312, 53.849853515625, -7.4544830322265625, -11.59954833984375, 77.90176391601562, 76.97210693359375, 32.015533447265625, -22.82421875, 24.86969757080078, 6.3568115234375, 59.9530029296875, -7.6329345703125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000238.npy"}
|
|
{"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 23.50127601623535, "std": 43.72100830078125, "min": -104.493896484375, "p10": -26.16448974609375, "median": 16.637725830078125, "p90": 90.438224029541, "max": 117.1083984375, "pos_frac": 0.703125, "sample": [-43.752769470214844, -25.715606689453125, 96.05419158935547, 58.7222900390625, -13.875762939453125, -8.5318603515625, -104.493896484375, 46.449729919433594, 100.38381958007812, 0.0226898193359375, 56.1533203125, -97.19000244140625, -60.166046142578125, -5.50848388671875, 61.841407775878906, 44.0924072265625, 63.18084716796875, 91.74272918701172, 32.498985290527344, -4.96484375, 0.29559326171875, 62.6612548828125, 12.028186798095703, 13.165573120117188, -1.9359970092773438, 36.53245544433594, 8.473995208740234, 32.51367950439453, 21.903411865234375, -13.218841552734375, -35.6326904296875, 68.3670654296875, 52.64661407470703, 5.109130859375, 107.49459838867188, -9.39080810546875, 23.519912719726562, 71.5831298828125, 117.1083984375, 86.97021484375, 64.7232894897461, -5.432861328125, 16.798187255859375, -1.360748291015625, 112.3939208984375, -12.462493896484375, -12.781494140625, 28.25384521484375, -47.97172546386719, 64.28829956054688, -50.62902069091797, 52.12234115600586, 8.406791687011719, 29.219558715820312, 41.55340576171875, -14.00372314453125, 43.737091064453125, 13.086654663085938, 6.40435791015625, -33.135589599609375, 16.477264404296875, 88.0445556640625, 5.495550155639648, 64.473388671875, 18.080345153808594, 29.5860595703125, 62.096763610839844, -2.8669052124023438, 8.83270263671875, 98.452392578125, 0.8172473907470703, 24.703277587890625, 10.113922119140625, 33.472869873046875, 65.48021697998047, -5.5106201171875, 45.34228515625, 9.559349060058594, -32.79615783691406, 20.839080810546875, -0.15289306640625, 35.432151794433594, 98.08064270019531, -18.582366943359375, -2.37347412109375, 89.879150390625, -56.342987060546875, 24.73479461669922, 6.9117889404296875, -41.56756591796875, 79.68435668945312, -3.7487525939941406, -4.210357666015625, 72.5414810180664, 2.084423065185547, 40.955379486083984, 10.454383850097656, 4.681396484375, 101.04339599609375, 30.909088134765625, -0.5133209228515625, 102.49160766601562, 9.26849365234375, -21.920318603515625, 45.217376708984375, 4.6001129150390625, -27.211883544921875, 23.207366943359375, 33.232086181640625, 97.15597534179688, 1.7933502197265625, 33.887664794921875, 44.789794921875, 103.40106201171875, 94.65603637695312, 0.0, 37.03289794921875, 2.3955860137939453, 78.56745910644531, 3.420867919921875, -23.855728149414062, 57.43962860107422, 14.5909423828125, 16.349151611328125, 30.09649658203125, -52.99021911621094, -6.344390869140625, 25.9495849609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000239.npy"}
|
|
{"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 30.09414291381836, "std": 38.97469711303711, "min": -75.47946166992188, "p10": -13.392478942871092, "median": 28.570823669433594, "p90": 76.84649581909179, "max": 128.07418823242188, "pos_frac": 0.75, "sample": [12.601913452148438, 68.9383544921875, 46.9310302734375, 65.51724243164062, -11.989662170410156, -12.009334564208984, 45.8126220703125, 41.15655517578125, 0.0, 104.1004638671875, 31.32391357421875, -23.4219970703125, -38.73091125488281, 10.707901000976562, -15.680778503417969, 76.3294448852539, 41.2800178527832, 43.212745666503906, 66.45210266113281, 29.454421997070312, 15.84466552734375, 2.338592529296875, 57.685577392578125, 72.57296752929688, 15.957550048828125, 69.61866760253906, 52.35542297363281, -25.112701416015625, 78.05294799804688, 82.78770446777344, 31.740753173828125, 20.189453125, 67.0611572265625, 62.353660583496094, 10.03753662109375, -9.457122802734375, 85.17184448242188, 30.227813720703125, 60.013946533203125, 5.968597412109375, 18.652053833007812, 44.248809814453125, 81.92886352539062, -2.9196014404296875, 61.0224609375, 43.489410400390625, 63.608734130859375, 44.461761474609375, 43.182594299316406, 60.15742492675781, 75.52200317382812, 38.329917907714844, 0.0, 36.352928161621094, -75.47946166992188, 23.933929443359375, -7.5872802734375, -0.6598358154296875, -15.529510498046875, 13.76666259765625, 36.58668518066406, 67.13571166992188, -8.03985595703125, -6.93682861328125, 85.48065185546875, 70.24542236328125, 42.5296630859375, 59.784706115722656, 75.23551940917969, 50.591522216796875, 96.48696899414062, -18.42333984375, 0.0, 15.018310546875, -14.082374572753906, -8.45806884765625, 108.85980224609375, 0.0723114013671875, 19.792633056640625, 24.901153564453125, -13.096809387207031, 35.52667236328125, 62.90362548828125, 23.677215576171875, 43.86207580566406, 128.07418823242188, -37.16748046875, 127.43600463867188, -69.31381225585938, 66.6031494140625, 15.4969482421875, 1.2265625, 16.646270751953125, 74.35050964355469, 62.65922546386719, 21.32769775390625, 1.6392364501953125, 57.58685302734375, 10.984640121459961, -10.304010391235352, -4.373626708984375, 3.199687957763672, 9.008060455322266, 17.421920776367188, 69.67323303222656, 33.4549560546875, 55.21180725097656, 84.8824462890625, 25.689361572265625, 6.2703857421875, 91.30303955078125, 29.824417114257812, 24.51708984375, -54.85578918457031, 27.687225341796875, -3.686786651611328, -25.8187255859375, 48.55558776855469, -9.990478515625, 64.58950805664062, 0.0, -1.5330657958984375, 35.496551513671875, 21.4747314453125, 19.19744873046875, -49.844146728515625, 5.3455810546875, 98.57373046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000240.npy"}
|
|
{"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 23.147201538085938, "std": 43.03947448730469, "min": -84.6661148071289, "p10": -31.535353088378905, "median": 14.863487243652344, "p90": 80.09872817993163, "max": 131.87078857421875, "pos_frac": 0.6953125, "sample": [3.71063232421875, 89.26385498046875, 58.59602355957031, 51.495208740234375, 24.355010986328125, -24.197265625, 18.71820068359375, 1.3029422760009766, 39.5372314453125, 65.99508666992188, 67.71707153320312, 0.0, -10.446533203125, 74.41250610351562, 61.78187561035156, 102.948486328125, 4.27301025390625, 4.2609100341796875, 14.9027099609375, 85.1705322265625, 52.455989837646484, 71.4262924194336, -10.6395263671875, -12.886962890625, 70.95901489257812, 13.357223510742188, -31.70635986328125, 20.267730712890625, 67.91777038574219, -11.4798583984375, 78.9175033569336, -31.504302978515625, 74.6571044921875, 59.36181640625, 15.269073486328125, 10.592254638671875, -9.07476806640625, -7.9153594970703125, 4.22566032409668, 78.26046752929688, -4.126747131347656, 14.824264526367188, 56.809837341308594, 107.93475341796875, 5.156951904296875, 62.7921142578125, 58.538604736328125, -77.64303588867188, 16.432510375976562, 22.48651123046875, 18.051979064941406, -2.2668685913085938, 5.094873428344727, 63.39532470703125, -23.670021057128906, -4.365264892578125, 78.65863037109375, 7.721099853515625, 6.313873291015625, -31.607803344726562, 28.770477294921875, 77.50302124023438, 35.018524169921875, 70.88931274414062, -4.993476867675781, 33.147491455078125, -14.584182739257812, 32.92802429199219, 5.754798889160156, 31.482391357421875, -54.89399719238281, 5.787689208984375, 92.00746154785156, 43.100555419921875, -18.102294921875, -37.6209716796875, 86.60467529296875, 51.596275329589844, 4.372833251953125, -1.268310546875, 5.140106201171875, -2.8156166076660156, 89.27761840820312, 24.5203857421875, 35.019317626953125, 23.550796508789062, -40.106719970703125, 74.04818725585938, -9.615524291992188, 82.85491943359375, 8.599456787109375, -48.36238098144531, 1.637603759765625, 4.8555755615234375, 60.827239990234375, 38.71858215332031, 3.4195194244384766, 3.520050048828125, 92.1353759765625, -32.828216552734375, 60.602569580078125, -38.41259765625, 18.9395751953125, -6.9337310791015625, 86.5921630859375, 76.1048583984375, -29.785995483398438, 6.815452575683594, 87.60018920898438, 49.20867156982422, -22.5206298828125, -84.6661148071289, -24.72802734375, -5.988277435302734, 63.52943420410156, -0.7814197540283203, 104.9329833984375, 40.47804260253906, -44.89170837402344, 29.628204345703125, 131.87078857421875, 2.861083984375, 10.33758544921875, -59.337005615234375, 10.37506103515625, -17.945556640625, 60.481658935546875, -40.212188720703125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000241.npy"}
|
|
{"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 27.870983123779297, "std": 38.77899169921875, "min": -80.67103576660156, "p10": -18.446325111389157, "median": 24.946348190307617, "p90": 82.31791381835937, "max": 107.78402709960938, "pos_frac": 0.765625, "sample": [63.08937072753906, 5.652153015136719, 82.8092041015625, 8.447063446044922, 13.517738342285156, 19.116493225097656, 40.278839111328125, 46.53570556640625, -22.06146240234375, 8.889347076416016, 95.90603637695312, 62.9498291015625, 19.633026123046875, -8.347137451171875, -59.73369598388672, 31.441879272460938, 13.79168701171875, 73.05582427978516, -80.67103576660156, 33.57566452026367, 12.039064407348633, 95.90867614746094, 85.3089599609375, 44.79539489746094, 43.4337158203125, 97.97036743164062, 3.816864013671875, 92.94805908203125, -0.4531402587890625, -17.878347396850586, 42.044403076171875, 70.33377075195312, -0.00177001953125, 83.60882568359375, 31.930816650390625, 107.78402709960938, 91.91731262207031, 3.3391494750976562, -27.056777954101562, 1.9206390380859375, 69.37844848632812, -19.7716064453125, -2.186279296875, 49.207977294921875, -17.655426025390625, -13.225793838500977, 28.872406005859375, 48.94207763671875, 68.6556396484375, 4.997802734375, -8.390609741210938, -9.313289642333984, -17.104816436767578, 44.922454833984375, -6.254356384277344, -11.048669815063477, 73.92138671875, 9.965225219726562, -8.61785888671875, 62.425811767578125, 31.076416015625, 1.3613033294677734, -21.065704345703125, 89.59841918945312, 9.237777709960938, 1.8737564086914062, 100.2266845703125, 49.445098876953125, 24.836471557617188, 17.172805786132812, 31.667083740234375, 57.1187744140625, -21.596328735351562, 9.056480407714844, 60.146728515625, 16.104469299316406, 49.27562713623047, 29.652484893798828, 20.48687744140625, -30.928268432617188, 35.16432189941406, 61.85271072387695, 20.227294921875, 46.359561920166016, -48.9764404296875, 13.33685302734375, 8.192276000976562, 23.786476135253906, 39.15948486328125, 66.89657592773438, 104.78836059570312, 5.586090087890625, 21.042724609375, 25.056224822998047, -1.7037429809570312, 17.405914306640625, 62.334381103515625, 70.52023315429688, 47.69403076171875, 19.164901733398438, 3.1358871459960938, 53.3826904296875, -36.79765701293945, -13.85504150390625, 52.82696533203125, 17.685752868652344, 34.313045501708984, -61.3956298828125, 33.29730224609375, 31.308380126953125, 26.23760986328125, 6.3175201416015625, 77.58561706542969, 92.11405944824219, 66.5665283203125, 58.5433349609375, 82.10736083984375, 21.71788787841797, -41.553306579589844, 62.527191162109375, -1.665679931640625, 35.70641326904297, 12.59478759765625, -36.238922119140625, 53.19999694824219, 70.7314453125, -14.486831665039062, 59.63677978515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000242.npy"}
|
|
{"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 32.57144546508789, "std": 40.055782318115234, "min": -73.17373657226562, "p10": -15.6472412109375, "median": 30.593482971191406, "p90": 85.6084991455078, "max": 118.613037109375, "pos_frac": 0.7578125, "sample": [-49.7783203125, -11.437393188476562, 32.43730163574219, 32.0263671875, 66.45819091796875, 48.4951171875, 27.34210205078125, 90.02251434326172, -43.298126220703125, 68.70606231689453, 24.08709716796875, 108.98638916015625, 4.706451416015625, -21.60797882080078, 49.47869873046875, -12.913558959960938, -35.09333801269531, 28.950836181640625, 96.43756103515625, -60.655052185058594, 13.789260864257812, 14.18194580078125, -21.210296630859375, -1.5094833374023438, 38.296173095703125, 81.873779296875, 118.613037109375, 81.9163818359375, -11.414199829101562, 48.315765380859375, 5.790990829467773, 65.3066177368164, 16.144981384277344, 41.72355651855469, 79.45770263671875, 4.41650390625, 90.62223815917969, 15.367767333984375, 51.330230712890625, 85.72323608398438, -15.938323974609375, 67.41229248046875, 37.98187255859375, 57.001007080078125, 43.633544921875, 54.6295166015625, 88.8729248046875, -15.1356201171875, 51.119384765625, 33.976341247558594, 37.534820556640625, -73.17373657226562, -6.094657897949219, 37.115081787109375, 4.0438232421875, 1.4959716796875, 24.24773406982422, 31.03192138671875, 10.472305297851562, -8.28253173828125, 110.79945373535156, 69.85052490234375, 63.832550048828125, -8.636993408203125, -3.527130126953125, 66.63396453857422, 18.529342651367188, 68.07403564453125, 61.227630615234375, -5.0704345703125, 33.00807189941406, -28.267120361328125, -9.8079833984375, 74.57032775878906, 62.244842529296875, 13.824188232421875, 5.004852294921875, 53.34062194824219, 3.9896240234375, 24.65618133544922, -5.471521377563477, 64.08061981201172, -21.24945831298828, 34.959442138671875, -0.316009521484375, 0.0, 8.576366424560547, 30.155044555664062, 65.33350372314453, 79.46987915039062, 36.93877410888672, 53.993255615234375, 58.72601318359375, -5.526939392089844, 78.72122192382812, 70.27578735351562, 91.4415283203125, 17.545578002929688, 9.861671447753906, 30.10723876953125, 19.8895263671875, 20.79290771484375, -10.78936767578125, 24.690399169921875, -15.55865478515625, 85.559326171875, 100.744384765625, 61.41534423828125, 4.400108337402344, 28.579345703125, 22.40738296508789, 13.43524169921875, 81.43031311035156, -16.019744873046875, -22.90765380859375, 115.99017333984375, 18.05162811279297, 82.88960266113281, -6.013671875, 21.247634887695312, 35.772979736328125, 66.86888122558594, 108.82369995117188, 61.513893127441406, 101.27325439453125, 43.70074462890625, 40.881805419921875, -15.85394287109375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000243.npy"}
|
|
{"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 33.17345428466797, "std": 42.60713577270508, "min": -90.27810668945312, "p10": -11.791299438476559, "median": 34.99452209472656, "p90": 85.65823516845703, "max": 161.933837890625, "pos_frac": 0.765625, "sample": [-90.27810668945312, 18.931549072265625, 47.662322998046875, 0.13741302490234375, 27.803390502929688, 7.375217437744141, 65.5994873046875, -23.351730346679688, 82.16790771484375, 52.27562713623047, 81.75668334960938, 63.592071533203125, 19.23504638671875, -5.9155426025390625, 8.397361755371094, 20.6402587890625, 57.890625, 64.27841186523438, 57.93736267089844, 56.13763427734375, 14.439132690429688, 68.98668670654297, 0.5675735473632812, -2.62445068359375, 65.69525146484375, -0.47052001953125, -23.05059814453125, -39.54302215576172, 82.24694061279297, -23.943695068359375, 7.93157958984375, -10.675628662109375, -4.246788024902344, -5.909263610839844, -75.63656616210938, 12.925239562988281, 93.14434814453125, 1.7579498291015625, 54.560516357421875, 67.74217987060547, -1.2855987548828125, 55.38848876953125, 88.75521850585938, 45.002044677734375, 105.1871337890625, 60.77293395996094, -0.817718505859375, 26.910293579101562, 31.2208251953125, 47.39115905761719, 59.25177001953125, 68.37738037109375, 0.0, 69.09780883789062, -8.30426025390625, -38.397796630859375, -14.39453125, 66.36839294433594, -9.71563720703125, 1.5218582153320312, 91.16888427734375, -24.57916259765625, 0.004730224609375, 54.45586395263672, 4.228004455566406, 82.93161010742188, 82.02098846435547, 92.85208129882812, 161.933837890625, 17.44110107421875, 36.55546569824219, 43.912994384765625, -27.5260009765625, 65.46954345703125, 35.326202392578125, 8.27789306640625, 99.19024658203125, -10.5189208984375, 100.470947265625, 59.26922607421875, 3.07586669921875, 67.67459106445312, 8.888839721679688, -55.15803527832031, 23.73919677734375, 54.4906005859375, -3.146484375, 67.62451171875, 66.71949768066406, 12.52264404296875, 7.060760498046875, -3.57708740234375, 34.662841796875, 55.171630859375, 104.46762084960938, 88.49404907226562, 78.4335708618164, 3.6490631103515625, 108.908935546875, 37.97853088378906, 4.949943542480469, 67.58474731445312, 40.2850341796875, 45.037803649902344, -3.2322235107421875, 56.03923034667969, 20.303070068359375, 18.41802978515625, 88.54109954833984, 35.614707946777344, -23.157379150390625, 41.26446533203125, 5.758247375488281, 84.44288635253906, 61.79693603515625, 114.549072265625, -8.138067245483398, 53.36680603027344, 53.74400329589844, -7.7249755859375, 21.814895629882812, 10.725509643554688, 75.08692932128906, -56.452972412109375, 27.093414306640625, 57.9189453125, 77.05804443359375, 0.447845458984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000244.npy"}
|
|
{"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 32.8387565612793, "std": 42.9328498840332, "min": -92.66413879394531, "p10": -18.545956420898435, "median": 29.202966690063477, "p90": 80.2447021484375, "max": 128.6387939453125, "pos_frac": 0.78125, "sample": [22.727294921875, -11.077987670898438, 25.790664672851562, 28.73145294189453, 7.562774658203125, 17.98712158203125, 63.512054443359375, 3.27215576171875, 4.650238037109375, 61.437591552734375, 32.92669677734375, -6.10400390625, 12.094589233398438, 94.61288452148438, 54.87290954589844, 14.408477783203125, 109.97831726074219, 105.50129699707031, 21.130752563476562, 123.27825927734375, -92.66413879394531, -19.385345458984375, -51.93841552734375, -24.4632568359375, 43.2615966796875, 49.75152587890625, 0.0, 28.99566650390625, 53.86920166015625, -5.6286773681640625, 64.04563903808594, 65.98747253417969, 56.005706787109375, 72.90530395507812, 1.636260986328125, 106.77520751953125, 2.4694747924804688, 18.521728515625, 69.88139343261719, 36.379119873046875, 97.59886169433594, 12.404449462890625, 32.71490478515625, 52.48420715332031, 11.420440673828125, 66.51654052734375, 24.57586669921875, 39.614105224609375, 25.908855438232422, -1.5727386474609375, -34.09149169921875, 28.742830276489258, -2.563812255859375, 76.17535400390625, 20.411468505859375, -6.908164978027344, 58.993621826171875, 38.57861328125, 48.020599365234375, 78.88917541503906, 113.396728515625, 0.8274574279785156, 0.9101791381835938, 48.32732391357422, 39.47447204589844, 65.14085388183594, 26.65716552734375, -31.8472900390625, 48.91650390625, -55.85406494140625, 23.73089599609375, 67.80377960205078, 55.88630676269531, 61.68510437011719, 74.1250228881836, 8.09930419921875, -44.573394775390625, 29.410266876220703, 126.35250854492188, 67.44172668457031, 6.773502349853516, 73.88658905029297, -31.336212158203125, 9.827110290527344, 9.725341796875, -46.53533935546875, 10.653167724609375, 80.75927734375, 78.04261779785156, 71.02757263183594, 44.84629821777344, 5.66558837890625, -2.1863021850585938, 115.16998291015625, 36.88360595703125, 111.77755737304688, 15.297744750976562, 32.406982421875, 45.290061950683594, 63.22480010986328, -61.47206115722656, 80.024169921875, 53.48997497558594, -15.71868896484375, -1.8039054870605469, 67.154541015625, 22.9412841796875, 69.4854736328125, 53.9014778137207, 63.132049560546875, 10.076934814453125, -18.18621826171875, 35.66278076171875, 26.948974609375, 69.40338897705078, 15.06768798828125, -1.044189453125, -16.371875762939453, 105.03114318847656, -2.1881542205810547, 75.25820922851562, 17.744720458984375, 66.99734497070312, 128.6387939453125, -50.333770751953125, -24.5052433013916, 0.0, 45.302459716796875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000245.npy"}
|
|
{"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 21.635652542114258, "std": 44.839027404785156, "min": -85.10780334472656, "p10": -28.776371765136716, "median": 12.932754516601562, "p90": 88.58099212646484, "max": 125.98681640625, "pos_frac": 0.6640625, "sample": [42.959930419921875, 61.944000244140625, 55.953834533691406, 2.300994873046875, 67.1751708984375, -8.454444885253906, 125.98681640625, 6.95892333984375, 8.163776397705078, 43.56294250488281, -45.96988296508789, 54.170310974121094, 21.212432861328125, 14.421295166015625, -21.13507080078125, -11.476760864257812, -11.61065673828125, 103.63314819335938, 103.31619262695312, -40.022552490234375, 86.83856201171875, 103.17935180664062, 15.79095458984375, 71.17777252197266, 37.562225341796875, -8.106033325195312, 14.54095458984375, -32.56611633300781, 24.507827758789062, -3.7908477783203125, 34.00225830078125, -27.58416748046875, 84.4293212890625, 9.148818969726562, 7.490936279296875, 2.7507171630859375, -54.82972717285156, 90.32275390625, 1.410430908203125, 1.6469001770019531, 13.4107666015625, -74.9954833984375, -15.151580810546875, 89.31182861328125, -42.667205810546875, -21.59710693359375, -47.299015045166016, -27.262344360351562, 2.959716796875, 42.943695068359375, 28.079559326171875, 104.154296875, -16.0751953125, 18.245330810546875, 91.23101806640625, -14.28635025024414, 27.24578857421875, -16.793121337890625, -85.10780334472656, 22.772262573242188, 97.09695434570312, 43.33221435546875, 42.28386688232422, 33.988616943359375, -45.702056884765625, 37.71833038330078, 103.95993041992188, 57.941192626953125, -14.460433959960938, 106.22369384765625, 86.10635375976562, 41.13250732421875, -27.375335693359375, 1.632171630859375, 18.582977294921875, -0.8619384765625, 77.25851440429688, 12.43328857421875, -20.767410278320312, -23.4119873046875, -48.960235595703125, -0.9691162109375, 70.05267333984375, 79.75885009765625, -18.048851013183594, -3.603240966796875, -53.8792724609375, -34.2642822265625, 17.35107421875, 6.8472442626953125, 35.561126708984375, 83.62306213378906, 48.067474365234375, -24.675094604492188, 41.903160095214844, 24.914451599121094, 5.313018798828125, -4.587188720703125, 0.0, 73.87164306640625, 88.26777648925781, 62.428466796875, -11.08258056640625, -18.363357543945312, 64.064697265625, 0.7299346923828125, 0.196868896484375, 53.762908935546875, 9.45627212524414, 54.41435241699219, 23.78089141845703, 1.6501007080078125, 80.76058959960938, 48.2509765625, 11.8106689453125, 90.43911743164062, -12.358951568603516, -31.558181762695312, -20.30499267578125, -19.395172119140625, 95.48773193359375, 17.254058837890625, 4.1269378662109375, 45.37541198730469, 3.1527481079101562, 12.454742431640625, -3.9474639892578125, 82.98870849609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000246.npy"}
|
|
{"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 27.41891860961914, "std": 40.35748291015625, "min": -87.35556030273438, "p10": -22.821493530273436, "median": 22.578338623046875, "p90": 87.01676177978516, "max": 111.92498779296875, "pos_frac": 0.734375, "sample": [74.77146911621094, -25.76849365234375, 55.906005859375, 52.09925079345703, -6.991912841796875, 30.677749633789062, 85.32254028320312, 88.39694213867188, 63.347412109375, 42.05085754394531, 0.0, 65.69779968261719, 87.85453796386719, 18.623382568359375, 6.873847961425781, 33.57501220703125, -20.900901794433594, 2.3702392578125, -37.6173095703125, 58.74229431152344, 2.4681396484375, 90.763916015625, -23.576629638671875, -36.93486785888672, 48.73541259765625, 68.48379516601562, 39.42889404296875, -0.57366943359375, 71.71774291992188, 86.65771484375, 58.0162353515625, 77.4184799194336, 11.52972412109375, 106.11279296875, 52.923744201660156, -26.68487548828125, 5.27801513671875, 94.55108642578125, -6.6075439453125, 8.67340087890625, -8.183837890625, -18.37824249267578, 22.05352783203125, -16.064773559570312, 4.010040283203125, 74.35791015625, -3.9007606506347656, 32.35198974609375, 4.61981201171875, -63.097320556640625, -1.0684986114501953, 38.07489013671875, 32.157203674316406, 45.38653564453125, -25.251602172851562, 74.8774185180664, 111.92498779296875, 16.111083984375, 2.33343505859375, -10.5693359375, -24.17194366455078, -11.518096923828125, -12.875518798828125, 98.7744140625, -0.2579345703125, -0.390594482421875, 52.00775146484375, -48.85711669921875, 10.295944213867188, -22.49786376953125, 15.831008911132812, 89.93978881835938, -87.35556030273438, -0.49346923828125, 60.6187744140625, 10.639297485351562, 23.581016540527344, 7.661346435546875, 23.04864501953125, 52.57081604003906, 30.2864990234375, 54.60945129394531, 97.4569091796875, 9.171295166015625, -23.66119384765625, -3.5033493041992188, 3.363372802734375, 35.35231018066406, 67.64657592773438, 21.172698974609375, 25.522216796875, 61.88058853149414, 34.47035217285156, 19.843414306640625, 10.671051025390625, 68.10603332519531, 94.92483520507812, 73.78639221191406, 102.39190673828125, -14.905807495117188, 29.097732543945312, 14.450164794921875, 50.31312561035156, 35.22625732421875, 61.05853271484375, 24.306076049804688, 22.1080322265625, -28.96026611328125, 97.82696533203125, 20.5029296875, -61.427032470703125, 68.35633850097656, 46.37122344970703, 0.296112060546875, 23.62875747680664, 3.470724105834961, 57.387115478515625, -2.08544921875, 2.90423583984375, -3.89239501953125, 62.57347869873047, 35.254852294921875, 93.16780090332031, 2.588031768798828, 67.07903289794922, 7.951873779296875, 20.73321533203125, 39.041259765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000247.npy"}
|
|
{"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 24.000213623046875, "std": 41.716102600097656, "min": -90.01162719726562, "p10": -26.554113006591795, "median": 19.768030166625977, "p90": 83.93261260986327, "max": 134.735107421875, "pos_frac": 0.7578125, "sample": [11.094532012939453, 74.29351806640625, -25.72675323486328, 4.834068298339844, 99.2064208984375, -20.227996826171875, 17.022293090820312, 13.70697021484375, 82.25250244140625, 14.606765747070312, 24.494243621826172, 30.219558715820312, 51.704833984375, 17.95642852783203, 7.797187805175781, 29.986541748046875, 101.51657104492188, 20.702362060546875, 6.2245941162109375, 58.832881927490234, -13.297607421875, 20.0438232421875, 43.32521057128906, 40.732749938964844, 89.82781982421875, 27.19720458984375, 14.57497787475586, 38.6656494140625, 4.4678955078125, 48.545257568359375, 14.486061096191406, -20.93891143798828, 95.6571044921875, -12.668289184570312, 3.64825439453125, 30.175048828125, 77.12770080566406, -90.01162719726562, 30.66533660888672, -3.890634536743164, -28.484619140625, 26.084930419921875, 28.621841430664062, -18.043243408203125, -5.597148895263672, 33.3702392578125, -7.944515228271484, -58.47041320800781, 18.612014770507812, 39.17613220214844, 13.833023071289062, 94.31913757324219, 26.82391357421875, 19.810535430908203, 112.9896240234375, -34.1578369140625, 85.69805908203125, 0.6554069519042969, -6.498199462890625, 97.5341796875, 37.80517578125, 67.41205596923828, -24.336441040039062, -33.04931640625, 14.648345947265625, 13.129150390625, 134.735107421875, 79.19085693359375, 48.36567687988281, 58.484222412109375, 56.458778381347656, -12.63592529296875, 1.6982536315917969, 51.45448684692383, 0.41229248046875, -56.971221923828125, 0.0, 29.094024658203125, 101.83123779296875, 85.293701171875, 66.15739440917969, 48.8883056640625, 43.58506774902344, 52.91357421875, 9.614917755126953, 55.92950439453125, -20.11932373046875, -46.34552001953125, -2.5987091064453125, 11.713966369628906, -31.690170288085938, 16.06488037109375, 87.94624328613281, 69.96194458007812, 83.34928894042969, -53.6795654296875, 28.890472412109375, 1.9598922729492188, 5.598075866699219, 70.08238220214844, 0.6453781127929688, -37.40675354003906, 20.836959838867188, -7.708282470703125, 59.08973693847656, 32.10528564453125, 51.22442626953125, 7.348541259765625, 11.226089477539062, -59.0565185546875, 19.72552490234375, 17.818115234375, 6.879600524902344, -5.2879180908203125, 111.36660766601562, 55.2935791015625, 49.44586181640625, 52.53587341308594, 31.168197631835938, -55.4146728515625, 3.359973907470703, -5.0597686767578125, 13.033889770507812, 21.760009765625, 62.34814453125, -59.73541259765625, 4.518697738647461, 21.55975341796875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000248.npy"}
|
|
{"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 26.178115844726562, "std": 41.382450103759766, "min": -93.5826416015625, "p10": -20.521512603759763, "median": 21.299137115478516, "p90": 75.30318298339844, "max": 126.95556640625, "pos_frac": 0.7265625, "sample": [-4.6451416015625, 9.449657440185547, -50.61109924316406, -37.21636962890625, 32.850189208984375, 3.870952606201172, -14.4603271484375, 5.84979248046875, 6.080047607421875, 12.325927734375, -2.6332855224609375, -14.465461730957031, 66.65994262695312, 92.28387451171875, -2.6875762939453125, 52.770111083984375, 31.83087158203125, 65.28677368164062, 23.956298828125, 65.53448486328125, 11.438278198242188, -93.5826416015625, 75.52297973632812, 22.84783935546875, -22.149261474609375, 109.40216064453125, 62.432220458984375, -19.09466552734375, 11.124755859375, 33.92616271972656, 16.943687438964844, -11.660491943359375, 36.55108642578125, -14.101631164550781, 7.484710693359375, 20.373489379882812, 93.1715087890625, 16.88365936279297, 67.72518920898438, -44.252838134765625, 71.05632019042969, 68.82843017578125, -7.3684844970703125, 82.81290435791016, 113.166015625, 9.548263549804688, -69.67794799804688, 48.49462890625, 49.4903564453125, 4.584724426269531, -6.91937255859375, 22.38324737548828, 119.88845825195312, 52.125343322753906, 67.7789306640625, -1.739654541015625, 10.949371337890625, 47.82563018798828, -29.6796875, 56.973876953125, 30.504989624023438, -2.86492919921875, 0.824920654296875, 4.088409423828125, 71.10797119140625, 8.411865234375, 62.47247314453125, 0.9332962036132812, 47.34962463378906, 49.63689422607422, -19.728134155273438, -19.82390594482422, 17.29559326171875, 57.63240051269531, 22.207443237304688, 59.54638671875, 126.95556640625, 100.10101318359375, 71.47782897949219, 63.084449768066406, 112.3284912109375, 50.192481994628906, 13.911300659179688, -25.71917724609375, 59.02972412109375, 99.85221862792969, -1.566232681274414, -17.454681396484375, -2.2773399353027344, 47.261871337890625, 29.617691040039062, -25.68975830078125, 61.17236328125, 55.3939208984375, 31.76788330078125, 53.615447998046875, 56.71728515625, 23.05523681640625, 0.617706298828125, -29.499969482421875, 86.45745849609375, 2.1104736328125, 23.638629913330078, 38.627166748046875, 34.975677490234375, 11.74249267578125, -3.8779754638671875, -68.74212646484375, -24.229156494140625, -11.428329467773438, -23.8892822265625, 5.17432975769043, 48.55126953125, 1.3117828369140625, 18.4130859375, -0.480804443359375, 54.5955810546875, 35.511627197265625, 70.83134460449219, 6.19677734375, 75.208984375, 57.59527587890625, 20.390830993652344, -17.983230590820312, -3.7521286010742188, 30.80926513671875, 10.022470474243164, 95.93949890136719], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000249.npy"}
|
|
{"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 26.823524475097656, "std": 45.56985092163086, "min": -79.1512451171875, "p10": -26.685540771484373, "median": 29.031879425048828, "p90": 85.71380767822265, "max": 124.516357421875, "pos_frac": 0.703125, "sample": [24.010009765625, 11.040733337402344, -60.00859069824219, 76.509765625, 53.246734619140625, 124.516357421875, 117.82916259765625, 62.67097473144531, 28.030059814453125, -64.30826568603516, 56.46209716796875, 20.72892189025879, 89.20767211914062, 30.75164794921875, 19.231903076171875, 97.13124084472656, 40.61187744140625, -1.770751953125, 77.51690673828125, 72.88134765625, 88.01246643066406, 56.2734375, 55.785980224609375, 55.40789794921875, -24.8072509765625, 1.100830078125, -79.1512451171875, 19.875320434570312, -5.25250244140625, 1.83837890625, 103.98370361328125, -60.00140380859375, 45.926300048828125, 115.8802490234375, -9.7808837890625, 6.983642578125, 22.2666015625, -2.2861862182617188, 89.51449584960938, -75.00799560546875, -1.6585655212402344, 39.29620361328125, 84.6458740234375, 29.795860290527344, 57.741851806640625, 43.84849548339844, -34.493194580078125, -53.7259521484375, -1.5930290222167969, 50.650787353515625, 61.271766662597656, -13.611534118652344, 41.0201416015625, -5.5472412109375, 28.267898559570312, 11.5487060546875, -10.579002380371094, -27.56195068359375, 45.668601989746094, 1.2321014404296875, 51.907928466796875, 89.7186279296875, 66.02792358398438, -0.2944660186767578, 32.526649475097656, 32.7266845703125, 110.10919189453125, 69.50157165527344, -22.059524536132812, 33.094573974609375, 62.98439025878906, 66.28416442871094, -16.023895263671875, 74.63731384277344, 31.2000732421875, -20.440658569335938, 51.22404479980469, 77.24746704101562, 84.72866821289062, 26.68863868713379, 14.422769546508789, -15.94732666015625, 32.33953857421875, -17.5897216796875, 80.88223266601562, 21.49472427368164, -2.79656982421875, -65.69029235839844, 60.281158447265625, 105.93505859375, -9.969879150390625, 98.46772766113281, -46.41363525390625, -25.9204158782959, -67.91705322265625, 3.732421875, 16.439781188964844, -51.31170654296875, -65.34112548828125, 68.79627990722656, 49.177886962890625, 58.84973907470703, 61.4993896484375, 38.86473083496094, 57.34516143798828, 53.16022491455078, -26.3099365234375, 23.500335693359375, 13.031795501708984, -2.1080379486083984, 8.153106689453125, 45.0571403503418, 102.26956176757812, -4.985687255859375, 34.768096923828125, 30.797330856323242, 69.81632995605469, 4.102996826171875, 2.7540283203125, -3.7070236206054688, -15.000282287597656, 66.02188873291016, 36.436126708984375, 0.384185791015625, 6.75537109375, 23.953216552734375, 41.95526123046875, -7.884651184082031], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000250.npy"}
|
|
{"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 27.281192779541016, "std": 41.109474182128906, "min": -95.41023254394531, "p10": -19.57956848144531, "median": 21.644119262695312, "p90": 82.1491912841797, "max": 145.1934814453125, "pos_frac": 0.7265625, "sample": [37.82896423339844, 32.90484619140625, 24.0908203125, -13.1922607421875, -25.97235107421875, -30.385345458984375, -46.596160888671875, 13.224685668945312, 21.943206787109375, 16.6839599609375, 9.326456069946289, 53.44342041015625, 54.7164306640625, 64.55121612548828, -0.303192138671875, -3.6053543090820312, 18.39483642578125, 21.976852416992188, 5.075006484985352, 79.85296630859375, 5.4500274658203125, -7.72998046875, -21.451141357421875, 81.49810791015625, 55.38600158691406, 93.88102722167969, -36.80657958984375, 65.59455871582031, 82.3089599609375, 21.34503173828125, 74.4671630859375, 6.638217926025391, -9.843574523925781, 32.801513671875, 28.767822265625, -15.453136444091797, 37.6600341796875, 0.7931652069091797, 6.94696044921875, -22.110023498535156, 50.89002990722656, 35.449859619140625, 73.21994018554688, 88.7349853515625, 89.81571960449219, 6.725433349609375, -1.5401611328125, -26.829750061035156, 3.012706756591797, -3.6404876708984375, -29.233642578125, 18.16900634765625, 145.1934814453125, 33.839111328125, 56.414794921875, 66.62027740478516, 11.9947509765625, 77.76632690429688, 10.406120300292969, 40.725868225097656, 63.549072265625, -95.41023254394531, 22.734355926513672, 59.9237060546875, 19.701675415039062, -21.57825469970703, 10.042144775390625, -3.3796157836914062, 20.24578857421875, -4.841165542602539, 72.15249633789062, 58.745361328125, -9.395050048828125, 3.920734405517578, 57.50567626953125, 1.155914306640625, 12.443359375, 26.437164306640625, 3.3145217895507812, 53.51948547363281, -0.19964599609375, 56.45930480957031, 1.677128791809082, 21.147750854492188, 53.41328430175781, 48.98302459716797, 91.95196533203125, -54.61235046386719, -0.447967529296875, -0.0373077392578125, 24.567588806152344, 92.27862548828125, 51.82568359375, 40.927490234375, 111.58480834960938, -2.6220436096191406, 124.81488037109375, 82.08071899414062, 63.36238098144531, -4.071613311767578, 47.02618408203125, 26.670166015625, 59.238555908203125, 37.17034912109375, 76.8609619140625, 25.3009033203125, 84.69303131103516, -0.32413482666015625, 40.19671630859375, -53.334014892578125, 0.080657958984375, 29.74114990234375, 11.163925170898438, -12.364494323730469, 3.6784896850585938, -42.61543273925781, -18.7774658203125, -16.071685791015625, 7.1077880859375, 106.48663330078125, 81.49838256835938, -10.8857421875, 93.50779724121094, 98.08123779296875, -13.37103271484375, 4.535060882568359, 49.6533203125, 61.33685302734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000251.npy"}
|
|
{"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 28.559059143066406, "std": 46.69532775878906, "min": -102.18682861328125, "p10": -30.274098205566403, "median": 29.119901657104492, "p90": 88.48354187011718, "max": 148.080810546875, "pos_frac": 0.7265625, "sample": [76.08139038085938, 98.2992172241211, 95.62506103515625, 51.55622863769531, -47.94419479370117, 29.243404388427734, -43.53730773925781, 15.031959533691406, 9.000801086425781, -15.7198486328125, 3.5419998168945312, -25.096664428710938, 15.78936767578125, 30.973419189453125, 132.18377685546875, 29.259368896484375, 86.86021423339844, 53.008079528808594, 79.10826110839844, 46.869659423828125, 2.6797657012939453, 60.54664611816406, -16.123291015625, 93.86338806152344, -102.18682861328125, 2.112030029296875, 4.898590087890625, 48.631752014160156, -29.863525390625, 15.24798583984375, 1.6613235473632812, 85.52474975585938, 82.0030517578125, 59.85699462890625, 56.439186096191406, 27.08184814453125, -53.8492431640625, -9.87309455871582, 21.804840087890625, 148.080810546875, 61.280975341796875, 1.9918022155761719, 40.81474304199219, 25.027923583984375, 35.10591125488281, 4.638031005859375, 29.649063110351562, -6.646759033203125, 5.94140625, 11.965545654296875, 69.54339599609375, 2.44366455078125, 95.20742797851562, 61.400794982910156, -18.50732421875, -3.0753173828125, 61.232147216796875, -33.654571533203125, 24.1148681640625, 7.790779113769531, 59.16966247558594, -17.39752960205078, 28.5185546875, 107.2303466796875, 29.472320556640625, 42.63557434082031, 79.48052978515625, 33.8406982421875, 57.93609619140625, 72.43463134765625, -6.663421630859375, 71.28007507324219, 102.996826171875, -3.336334228515625, -23.19287109375, 61.56890869140625, -52.65789794921875, 7.886268615722656, 51.79937744140625, 60.60780715942383, 60.73321533203125, 5.4974365234375, 42.8892822265625, 66.88277435302734, -7.353515625, -1.0746612548828125, 0.0, 63.925758361816406, -42.11323547363281, 78.51683807373047, -0.44495582580566406, 73.37487030029297, -1.064605712890625, -20.365234375, 28.99639892578125, 97.04006958007812, -4.97052001953125, 106.18112182617188, 52.417572021484375, 86.89300537109375, -9.49420166015625, 38.467872619628906, 42.36805725097656, 4.022453308105469, 92.76699829101562, -4.1348876953125, 63.281768798828125, -88.0888671875, 0.62176513671875, 7.7591400146484375, -10.51959228515625, -31.232101440429688, 31.450927734375, -79.06658935546875, -61.34051513671875, 2.913116455078125, 63.653076171875, 92.19479370117188, -43.63604736328125, 72.94769287109375, 45.46435546875, 101.75274658203125, 61.007659912109375, -36.7159423828125, 10.885101318359375, 53.75321960449219, 14.199321746826172, 75.77128601074219], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000252.npy"}
|
|
{"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 23.45456886291504, "std": 43.77859115600586, "min": -93.11859130859375, "p10": -28.649548339843747, "median": 15.340934753417969, "p90": 86.59424743652343, "max": 116.44830322265625, "pos_frac": 0.7265625, "sample": [106.56149291992188, -5.8179931640625, 8.210922241210938, 26.348556518554688, 17.70196533203125, 32.48905944824219, 1.5092182159423828, 78.07070922851562, 107.91522216796875, 22.92724609375, 27.89697265625, 88.84091186523438, 6.1282958984375, -7.4288330078125, 50.502105712890625, -70.72637939453125, 38.74578094482422, -38.000221252441406, 73.50395202636719, -29.67254638671875, 0.186431884765625, 13.54449462890625, 15.822385787963867, 15.677408218383789, -2.347900390625, -26.5167236328125, 27.588104248046875, 89.10150146484375, 68.80200958251953, 64.5316390991211, -8.035928726196289, -8.51883316040039, -54.985687255859375, 15.745132446289062, 116.44830322265625, 2.76434326171875, -23.499832153320312, 74.249755859375, -93.11859130859375, -25.04364013671875, 70.16946411132812, 16.7200927734375, 87.64593505859375, 22.014419555664062, 14.32305908203125, -32.27587890625, 83.26988220214844, 3.333709716796875, -8.65911865234375, 17.01141357421875, 83.95458984375, 26.96832275390625, 86.14352416992188, -5.307098388671875, -9.883071899414062, 78.987548828125, 6.668407440185547, 10.549766540527344, -30.729583740234375, 62.52644348144531, 43.07270812988281, -34.063079833984375, -2.4120941162109375, 74.34980010986328, 22.24310302734375, 79.97267150878906, -31.164215087890625, -13.007453918457031, 56.22418212890625, -28.21112060546875, 15.5738525390625, -9.565353393554688, -80.95935821533203, 2.39642333984375, -70.02584838867188, 10.771894454956055, -52.62356948852539, 3.8872241973876953, -3.3600616455078125, -17.809112548828125, 51.29126739501953, 2.5133819580078125, 43.8668212890625, 65.93341064453125, 29.1365966796875, -0.3428955078125, 26.811180114746094, 7.84033203125, -43.957855224609375, 104.07742309570312, -12.974395751953125, 7.837127685546875, 60.0528564453125, 2.853363037109375, 4.873291015625, 78.40887451171875, -20.27147674560547, 106.848876953125, 9.091682434082031, 49.88743591308594, 55.90808868408203, 11.234199523925781, -22.61041259765625, -1.542144775390625, 34.553955078125, 88.30023193359375, 10.837966918945312, 92.28445434570312, 7.4737396240234375, 11.881656646728516, 21.38043212890625, 99.27276611328125, 62.614524841308594, 2.8594436645507812, 67.69775390625, 11.203178405761719, 10.74664306640625, 15.108016967773438, 26.723419189453125, 91.152099609375, 4.5154266357421875, 31.972412109375, 94.2646484375, 69.93475341796875, 9.602203369140625, 60.98631286621094, 43.32940673828125, 69.89920043945312], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000253.npy"}
|
|
{"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 29.19921112060547, "std": 40.6039924621582, "min": -86.19866943359375, "p10": -18.540086364746088, "median": 22.071582794189453, "p90": 82.89877624511718, "max": 117.86457824707031, "pos_frac": 0.7578125, "sample": [96.11065673828125, -1.14501953125, -56.19296646118164, 21.977272033691406, -6.378501892089844, -2.8553905487060547, 20.5340576171875, 16.52142333984375, -0.9832897186279297, 17.973098754882812, -13.528205871582031, -11.57464599609375, 0.35406494140625, -4.940277099609375, 5.0637054443359375, 75.00679016113281, 7.0072021484375, 77.691162109375, 45.45002746582031, -26.124603271484375, -29.094879150390625, 87.36553955078125, 8.614326477050781, 9.4224853515625, -9.355072021484375, 12.8936767578125, 66.79499816894531, -35.4573974609375, 69.03970336914062, 47.144012451171875, 73.54144287109375, 97.96334838867188, -50.490020751953125, 48.99409484863281, 1.011281967163086, 14.190277099609375, -17.25616455078125, 11.332447052001953, 9.01641845703125, 6.95098876953125, 44.6207275390625, 44.887451171875, 59.00054931640625, 24.470481872558594, -53.93934631347656, 24.977447509765625, 46.84368896484375, 42.03118896484375, 53.68994140625, 69.92994689941406, 20.639251708984375, 4.085567474365234, 21.53965950012207, 75.81492614746094, 55.807861328125, 22.1658935546875, 64.1536865234375, -35.19134521484375, 21.771636962890625, 54.8807373046875, 2.4071121215820312, 10.450103759765625, 100.59780883789062, 8.624214172363281, 40.88316345214844, 88.22222137451172, 38.33074951171875, 80.06730651855469, 77.02743530273438, 26.171340942382812, 8.213668823242188, 33.288604736328125, 89.86767578125, 4.7689666748046875, 38.1273193359375, -16.360260009765625, -5.8565521240234375, 29.56072998046875, 2.5452003479003906, 8.233016967773438, -2.81427001953125, -8.052459716796875, 96.898193359375, 81.10150146484375, 38.9052734375, 107.64993286132812, 55.694488525390625, 72.55299377441406, 18.9190673828125, 84.8087158203125, 17.18795394897461, 78.56352233886719, 78.2987060546875, 67.50224304199219, 77.56168365478516, 25.86553955078125, 1.072652816772461, 69.6624984741211, -3.772186279296875, -5.8371429443359375, -21.790512084960938, 34.218353271484375, 25.8323974609375, -0.2897472381591797, 0.0, -21.535903930664062, 90.512451171875, 65.49800109863281, 28.72503662109375, 26.850494384765625, 7.605705261230469, 14.43402099609375, 32.91009521484375, 82.08023071289062, 53.293670654296875, -0.2441864013671875, -86.19866943359375, -22.23746681213379, 105.39141845703125, 21.63140106201172, -35.506683349609375, -36.242889404296875, 3.8029022216796875, 117.86457824707031, 109.08183288574219, 54.06266784667969, 55.874969482421875, 70.16816711425781], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000254.npy"}
|
|
{"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 32.955970764160156, "std": 47.45438766479492, "min": -80.48619079589844, "p10": -17.739190673828123, "median": 23.970260620117188, "p90": 95.2123550415039, "max": 137.01498413085938, "pos_frac": 0.765625, "sample": [4.410350799560547, -53.05841064453125, 105.77349853515625, 55.981964111328125, 11.067428588867188, -71.31669616699219, 1.24688720703125, 87.34271240234375, -10.00921630859375, 73.11039733886719, -9.4932861328125, 11.483070373535156, 19.675548553466797, 3.96795654296875, 112.30126953125, 98.756103515625, 6.421760559082031, 10.9122314453125, 102.08087158203125, 9.03387451171875, 82.66510009765625, -28.706024169921875, -11.41265869140625, 75.65545654296875, 33.584625244140625, 86.29542541503906, 41.119773864746094, 69.4381103515625, -18.104827880859375, 33.761474609375, 56.89691925048828, 12.992279052734375, 18.180313110351562, 11.4072265625, 24.75189971923828, 25.168991088867188, 86.05514526367188, 46.4451904296875, -59.021026611328125, -17.582489013671875, 137.01498413085938, 117.49560546875, 93.89027404785156, 16.832778930664062, -60.54560089111328, -10.033235549926758, -61.412689208984375, 72.2708740234375, 65.61659240722656, 58.10459899902344, 84.2281494140625, 88.06533813476562, 54.263214111328125, 62.457183837890625, 114.85154724121094, -3.2560577392578125, 44.481719970703125, -57.56956481933594, -37.15101623535156, 76.6572265625, 20.0906982421875, 86.85009765625, 72.25698852539062, 65.47117614746094, 46.368408203125, 55.1624755859375, 4.1382598876953125, -65.92938995361328, 92.18412780761719, 15.218795776367188, 100.61270141601562, 12.912811279296875, 21.64804458618164, -9.882698059082031, 5.5938720703125, 6.276001930236816, 22.77234649658203, 23.391815185546875, 8.135772705078125, 52.576133728027344, -3.740814208984375, 59.888671875, -80.48619079589844, 107.844970703125, 89.22433471679688, 65.41929626464844, -4.7152099609375, 57.59740447998047, 16.163528442382812, 11.617942810058594, 98.78421020507812, 65.739501953125, 48.250823974609375, 92.89899444580078, -13.382476806640625, 118.74130249023438, 85.72567749023438, 7.18341064453125, -3.0163726806640625, -17.089096069335938, 0.5644760131835938, -28.60296630859375, 43.64851379394531, 4.990020751953125, 98.29721069335938, 38.144439697265625, 4.627471923828125, 73.712646484375, 9.532341003417969, 40.312408447265625, 11.6480712890625, 13.294265747070312, -10.181892395019531, 86.14126586914062, 15.640365600585938, 34.91120910644531, -8.586685180664062, -10.629974365234375, 73.23341369628906, 37.01634216308594, 24.5487060546875, 76.29485321044922, -2.852294921875, 69.84002685546875, -42.7247314453125, 118.06204223632812, 20.59375, -7.150642395019531], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000255.npy"}
|
|
{"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 33.2393798828125, "std": 50.93164825439453, "min": -99.98800659179688, "p10": -33.96305885314941, "median": 33.65887451171875, "p90": 96.68278198242187, "max": 130.5147705078125, "pos_frac": 0.7421875, "sample": [1.7146759033203125, 82.14959716796875, 89.51626586914062, -29.3287353515625, 18.76605224609375, 32.858978271484375, 80.67884826660156, 71.9945068359375, 73.40106201171875, -18.363723754882812, 68.28642272949219, -47.33441162109375, -27.389022827148438, -65.7502670288086, 52.933570861816406, 84.88456726074219, 100.87030029296875, -21.776947021484375, 51.469730377197266, 85.09515380859375, 1.7202911376953125, 81.33602905273438, -20.647659301757812, 130.5147705078125, 17.94091796875, 89.68707275390625, 77.95819091796875, -52.59820556640625, 100.54208374023438, 14.885009765625, 104.76361083984375, 24.65170669555664, 29.49615478515625, 33.770599365234375, 36.231597900390625, 6.925445556640625, -38.943145751953125, -52.92645263671875, 86.45169067382812, 21.563682556152344, 108.442138671875, 24.44012451171875, 51.21354675292969, 94.15666198730469, 38.08805847167969, 39.3924560546875, -24.759769439697266, 17.73340606689453, -30.822509765625, 123.78091430664062, 5.636405944824219, -22.083694458007812, 100.56739044189453, 106.12857055664062, 0.8433380126953125, 3.457763671875, -99.98800659179688, -26.92291259765625, 83.14346313476562, 52.56712341308594, 72.9063720703125, 16.038330078125, 70.62155151367188, 104.51040649414062, -1.5440034866333008, 62.10331726074219, 50.34400939941406, -65.4531021118164, 71.27474975585938, -1.760284423828125, -45.61468505859375, -66.46734619140625, 68.25677490234375, 55.293701171875, 37.069976806640625, 77.02545166015625, -12.322418212890625, 89.000732421875, 21.61126708984375, 95.22549438476562, 115.77731323242188, 58.8052978515625, 68.26768493652344, -0.2873992919921875, 3.647186279296875, 51.739501953125, -56.20668029785156, 19.90313720703125, 13.16534423828125, 100.682861328125, -40.304168701171875, 94.14971923828125, 67.43106842041016, 98.737548828125, 10.01910400390625, 73.38048553466797, 35.369415283203125, -10.803924560546875, 20.41046905517578, 22.8514404296875, 33.547149658203125, 30.19178009033203, 98.664306640625, 55.16259765625, 29.032928466796875, 44.2451171875, 15.43206787109375, 87.37469482421875, -36.53878402709961, -97.28030395507812, 0.74786376953125, 92.13426208496094, 95.83355712890625, 13.949363708496094, -3.1653976440429688, -5.1110382080078125, 93.03787231445312, -28.9110107421875, -14.992965698242188, -32.85917663574219, 67.39273071289062, -6.076171875, 5.095848083496094, 38.562721252441406, 18.54742431640625, 59.00233840942383, 52.75518798828125, 80.99533081054688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000256.npy"}
|
|
{"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 30.097444534301758, "std": 44.13159942626953, "min": -76.77162170410156, "p10": -21.08968353271484, "median": 24.75043487548828, "p90": 89.39332122802733, "max": 131.25927734375, "pos_frac": 0.671875, "sample": [48.414947509765625, 34.61810302734375, 117.59294128417969, -6.037071228027344, -2.3284225463867188, 118.0506591796875, 22.76806640625, 90.56257629394531, 39.11944580078125, 75.80005645751953, 42.456459045410156, 110.47296142578125, 68.52996826171875, 58.93133544921875, -0.17266082763671875, 28.940284729003906, 23.790542602539062, -40.9212646484375, -15.817512512207031, 57.643310546875, 88.8922119140625, -27.00585174560547, 61.3089599609375, 122.87277221679688, -17.671295166015625, -27.732666015625, 19.029296875, 22.448623657226562, -1.55999755859375, -32.540428161621094, -9.333274841308594, -26.0965576171875, 85.099853515625, 25.7103271484375, 55.192962646484375, 94.22494506835938, 126.67523193359375, 8.2655029296875, 102.25039672851562, 66.80413055419922, -11.15509033203125, 2.345794677734375, -11.688911437988281, 75.52615356445312, 94.84219360351562, -6.0018310546875, -2.1594924926757812, 36.56599426269531, 17.027557373046875, -76.77162170410156, 57.60377502441406, 109.71627807617188, 52.416038513183594, 44.096160888671875, 82.77587890625, 26.86334228515625, 42.17231750488281, -6.3690338134765625, 39.555084228515625, 17.732269287109375, 2.7105560302734375, 96.09555053710938, 70.18539428710938, 15.933231353759766, -31.703262329101562, -10.709747314453125, 44.17242431640625, 4.23902702331543, 7.5420074462890625, 48.7745361328125, -4.024688720703125, 69.68023681640625, 54.25756072998047, 34.000762939453125, 13.259458541870117, 13.349517822265625, -53.02418518066406, -1.4160995483398438, 62.42994689941406, 83.63291931152344, 7.3709259033203125, 5.907096862792969, 59.3642578125, 27.771881103515625, 5.2494049072265625, -31.6258544921875, -7.57171630859375, 84.2176513671875, 43.68463134765625, -9.142578125, -15.2291259765625, 70.391357421875, -20.224945068359375, 87.24063110351562, 46.11240768432617, 74.603271484375, 72.30667114257812, 34.673095703125, -18.234756469726562, 7.200187683105469, 11.455596923828125, 12.0877685546875, 86.67578125, 26.149505615234375, -4.526153564453125, -12.6297607421875, -6.3037109375, -49.12046813964844, -23.253753662109375, 93.9168701171875, 23.31854248046875, 80.29458618164062, -18.52728271484375, 131.25927734375, -15.11566162109375, 0.4026336669921875, 59.69970703125, -23.107406616210938, 65.71755981445312, 31.1749267578125, 69.32415771484375, -7.334442138671875, -6.4207916259765625, 42.37702941894531, -2.388916015625, -4.3873291015625, -23.174057006835938, 77.11627197265625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000257.npy"}
|
|
{"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 32.08823013305664, "std": 41.19615173339844, "min": -99.4669189453125, "p10": -17.67815399169922, "median": 30.140106201171875, "p90": 85.19546356201171, "max": 123.89190673828125, "pos_frac": 0.78125, "sample": [76.93649291992188, 12.583763122558594, 8.626197814941406, -2.1522293090820312, 47.83634948730469, 16.884506225585938, 76.72427368164062, -17.7891845703125, 22.078540802001953, 97.7431640625, -26.180328369140625, 30.877532958984375, 42.79595947265625, 11.555831909179688, 79.75601196289062, 49.96666717529297, -17.990692138671875, 18.37939453125, 70.38116455078125, 62.80961608886719, 95.67454528808594, 37.065345764160156, -40.535980224609375, 8.94390869140625, 105.83551025390625, -9.546127319335938, -99.4669189453125, 83.01995086669922, 64.49005126953125, 14.694656372070312, 81.12783813476562, 54.1820068359375, -19.985595703125, 45.63421630859375, 56.46630859375, 31.61712646484375, 2.963970184326172, -21.50225830078125, 42.71851348876953, 71.18235778808594, 63.144317626953125, 36.02788543701172, 5.840900421142578, 33.75408935546875, 11.869613647460938, 20.572845458984375, 21.9891357421875, 17.61785888671875, 13.948760986328125, 12.06158447265625, 42.474388122558594, 7.404510498046875, 77.80941772460938, 21.27777099609375, -86.61849975585938, 56.722190856933594, 32.00401306152344, -5.00738525390625, -40.75311279296875, 8.737060546875, -17.630569458007812, 17.66680908203125, 29.402679443359375, 32.9014892578125, 88.48440551757812, 0.27392578125, 103.93423461914062, 63.89349365234375, -11.89276123046875, 73.54307556152344, 54.679656982421875, 58.9364013671875, 0.5767593383789062, -3.791778564453125, 0.7469367980957031, 50.583984375, 90.67645263671875, -39.21168518066406, 55.65167999267578, 22.362823486328125, -7.214710235595703, -23.6358642578125, 69.48759460449219, 77.62532043457031, -8.513839721679688, -0.3369293212890625, 60.70000076293945, 32.21417236328125, 84.77206420898438, -19.83642578125, 22.095195770263672, 5.35113525390625, 24.003997802734375, 8.378612518310547, -3.785797119140625, 33.5352783203125, 68.29753112792969, 86.18339538574219, 12.91976547241211, -12.5372314453125, 4.877838134765625, -2.6522979736328125, 48.228271484375, 1.4854202270507812, 123.89190673828125, -3.1684112548828125, 79.76446533203125, 100.78849792480469, 80.02694702148438, 68.57423400878906, 36.388153076171875, -12.314971923828125, 106.77865600585938, 59.844818115234375, 92.76519775390625, 101.61590576171875, 51.379783630371094, 12.992889404296875, 46.97259521484375, 62.08487319946289, -12.45733642578125, 72.34831237792969, 59.73274230957031, 107.21746826171875, 5.636474609375, 19.99371337890625, 24.90533447265625, -29.178924560546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000258.npy"}
|
|
{"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 30.10215950012207, "std": 44.95541000366211, "min": -83.13333129882812, "p10": -23.14134979248047, "median": 26.48339080810547, "p90": 87.92517318725585, "max": 123.90158081054688, "pos_frac": 0.7421875, "sample": [73.36770629882812, -27.563186645507812, 90.94317626953125, 10.335334777832031, -21.620101928710938, 9.598587036132812, 12.44281005859375, 102.82061004638672, 85.485595703125, -8.97906494140625, 49.482177734375, 13.895233154296875, 42.14567565917969, 78.12336730957031, 20.47834014892578, 36.18470764160156, 73.33258056640625, -71.28956604003906, 56.34112548828125, 80.58219909667969, 68.74188232421875, 107.41119384765625, -18.6292724609375, 22.529605865478516, 96.87295532226562, 13.707443237304688, 46.40686798095703, 59.38801574707031, -54.76361083984375, -2.8640899658203125, 2.7574310302734375, 18.191009521484375, -18.454986572265625, 80.78765869140625, 45.084014892578125, 37.0853271484375, 42.992919921875, 9.826423645019531, 15.758941650390625, 46.67803955078125, -29.366912841796875, 19.555938720703125, 91.70962524414062, 16.033782958984375, 5.6058349609375, -52.99658203125, 57.24866485595703, 101.77230834960938, -2.1399612426757812, 58.50042724609375, 63.97531509399414, 35.784820556640625, 78.17803955078125, 17.4365234375, 8.243385314941406, 14.411598205566406, -1.7565155029296875, 98.86407470703125, 57.824920654296875, 14.234817504882812, 120.9305419921875, 49.390594482421875, 14.17828369140625, 1.6898078918457031, -4.73272705078125, -68.17896270751953, 38.376922607421875, -22.707412719726562, -46.92826843261719, 85.659912109375, 59.13438415527344, -10.131591796875, -6.2755889892578125, 51.8048095703125, 93.64482116699219, -1.820159912109375, 37.412078857421875, 0.0, -8.909423828125, 89.40029907226562, 66.92828369140625, 76.31167602539062, -16.692855834960938, 12.88262939453125, 40.93408203125, 2.002960205078125, 99.71365356445312, -60.284812927246094, 0.9330215454101562, 44.904449462890625, -83.13333129882812, -0.62054443359375, 86.69432067871094, 123.90158081054688, 25.875091552734375, 50.5794677734375, 58.35468292236328, -16.6021728515625, 73.9636459350586, 3.39654541015625, 39.31671142578125, 44.774139404296875, -22.051025390625, 79.88150024414062, 15.27001953125, 27.44183349609375, 87.29297637939453, -52.10003662109375, 42.17034912109375, 84.21533203125, 25.028961181640625, 78.75679016113281, 8.711639404296875, -24.99371337890625, 7.9208984375, 14.561332702636719, 104.56452941894531, 6.2034912109375, 27.091690063476562, -20.24383544921875, 72.64778137207031, -57.437103271484375, -24.15386962890625, 69.84228515625, -0.729736328125, 51.062347412109375, 67.48805236816406, 59.827144622802734], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000259.npy"}
|
|
{"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 13.688793182373047, "std": 44.863677978515625, "min": -110.07919311523438, "p10": -39.38095703125, "median": 11.651004791259766, "p90": 71.65738220214844, "max": 113.88723754882812, "pos_frac": 0.6171875, "sample": [95.49029541015625, 1.5365791320800781, 62.06482696533203, 17.224884033203125, -82.13247680664062, 66.64935302734375, 34.824676513671875, 89.02398681640625, 31.182266235351562, 36.9254150390625, -11.706634521484375, 17.606781005859375, 0.0, -16.9332275390625, 22.756851196289062, 4.397846221923828, 3.654205322265625, 113.88723754882812, 22.652008056640625, 27.597145080566406, 72.177490234375, -110.07919311523438, -3.8769073486328125, -19.00225830078125, 103.67901611328125, -47.40782165527344, 13.415668487548828, 59.80398178100586, 26.58514404296875, 13.8497314453125, 28.296722412109375, -44.55842590332031, 16.665061950683594, -0.6198596954345703, 4.619333267211914, 104.37445068359375, -37.53144836425781, -59.891632080078125, -9.849960327148438, 23.333106994628906, 2.681671142578125, 16.582183837890625, 40.79339599609375, -37.12469482421875, -22.3814697265625, 69.82258605957031, -50.83918380737305, -18.528827667236328, -35.20684814453125, 14.761110305786133, 1.76708984375, 5.99322509765625, 89.160888671875, 80.10125732421875, -3.047313690185547, 12.3760986328125, 86.45602416992188, -94.84738159179688, 70.94744873046875, 57.84025573730469, -24.01104736328125, -3.341564178466797, 69.69615173339844, -8.205169677734375, 7.523906707763672, -53.76557922363281, 42.28704071044922, 71.43447875976562, -3.93994140625, 87.32150268554688, 18.645980834960938, 57.991973876953125, -28.468505859375, 8.08587646484375, 8.414009094238281, -13.643157958984375, -5.024147033691406, 37.699859619140625, 12.594146728515625, 14.145538330078125, -48.59210205078125, -14.167633056640625, 102.76376342773438, -2.75274658203125, 0.135833740234375, 28.5523681640625, 19.724136352539062, -18.86968994140625, -13.027847290039062, -69.30586242675781, -25.544891357421875, 3.506622314453125, -71.95948791503906, 100.7196044921875, 27.182098388671875, 15.43939208984375, -6.5647735595703125, 30.43688201904297, -23.90313720703125, 90.18769836425781, 58.412017822265625, 41.905975341796875, 19.062896728515625, 43.63349914550781, 32.55108642578125, 3.03448486328125, -4.404815673828125, 64.00130462646484, -29.445289611816406, 66.46258544921875, -13.321868896484375, 47.442626953125, 65.15435791015625, 10.8121337890625, -39.5634765625, -0.142303466796875, -21.631805419921875, -84.97906494140625, -13.609527587890625, 15.242706298828125, 36.2021484375, -39.302734375, 28.286102294921875, 10.925910949707031, 65.29839324951172, -3.4716567993164062, 53.879066467285156, -37.66259765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000260.npy"}
|
|
{"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 25.574077606201172, "std": 47.38594436645508, "min": -80.09149169921875, "p10": -32.63590316772461, "median": 26.406539916992188, "p90": 85.43171463012695, "max": 151.94970703125, "pos_frac": 0.6953125, "sample": [-34.62477111816406, -80.09149169921875, 50.47200012207031, -20.652328491210938, 33.795806884765625, 21.52789306640625, 61.481689453125, -16.092132568359375, 12.541458129882812, 82.25567626953125, 89.36495971679688, 0.0, -12.945760726928711, -28.960784912109375, -0.35631752014160156, 39.6893310546875, -19.81616973876953, 33.930908203125, 80.46693420410156, 32.16961669921875, 43.53533935546875, -55.478153228759766, 80.06668853759766, -73.5361328125, 57.11216735839844, -31.68771743774414, -11.3360595703125, 77.21633911132812, -31.111953735351562, -65.09942626953125, 59.10777282714844, 72.54154968261719, 87.89950561523438, 47.66822814941406, 97.12788391113281, 49.105133056640625, -18.59471893310547, 15.83333969116211, -39.192840576171875, 49.36236572265625, 12.178565979003906, 10.398468017578125, 99.05764770507812, -20.208892822265625, 1.3700942993164062, 7.7570343017578125, -0.23898887634277344, 85.44474029541016, -54.594383239746094, 66.40015411376953, 24.2177734375, -22.628204345703125, 35.74359130859375, 102.25563049316406, 82.10562133789062, 49.247467041015625, 41.5445556640625, -30.612533569335938, 72.75399780273438, 82.75413513183594, 83.046630859375, 85.42613220214844, 56.602474212646484, 12.39837646484375, 50.569175720214844, -2.087635040283203, 26.873863220214844, -30.494338989257812, 22.157089233398438, -2.8627548217773438, 64.05035400390625, 151.94970703125, 1.624359130859375, 105.684814453125, 35.25834655761719, 1.57781982421875, 5.41070556640625, 3.375162124633789, 7.81854248046875, 8.696121215820312, 29.890899658203125, -5.741180419921875, 72.98326110839844, -56.80877685546875, 111.5394287109375, 26.735382080078125, 34.92939758300781, 48.01007843017578, 14.413490295410156, -60.00909423828125, 9.740997314453125, 20.244049072265625, -67.5262451171875, -0.705535888671875, -44.793487548828125, 94.34928894042969, 66.32525634765625, 80.44854736328125, 84.78117370605469, 32.1231689453125, -40.81147766113281, 36.40022277832031, -2.36383056640625, 68.49990844726562, 20.118820190429688, 39.931396484375, -31.783531188964844, 115.9324951171875, 96.02487182617188, -14.647743225097656, 73.96188354492188, -5.091178894042969, 38.768333435058594, -75.26406860351562, 33.3367919921875, 8.871406555175781, 11.597366333007812, 26.07769775390625, 54.661590576171875, 26.94561767578125, 46.35985565185547, 109.471923828125, 9.369552612304688, -26.1927490234375, -0.5575294494628906, 27.758758544921875, 4.7685394287109375, 63.689735412597656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000261.npy"}
|
|
{"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 31.825603485107422, "std": 43.725669860839844, "min": -71.62831115722656, "p10": -22.39451904296875, "median": 28.07879638671875, "p90": 89.76217346191406, "max": 124.11676025390625, "pos_frac": 0.734375, "sample": [86.22238159179688, 36.88470458984375, 69.19781494140625, 17.202903747558594, 69.13710021972656, 2.6717987060546875, 104.9910888671875, 82.03732299804688, 21.333099365234375, 90.57913208007812, 115.02276611328125, -19.10601806640625, 19.59111785888672, 16.8704833984375, 14.029525756835938, 4.446804046630859, 71.6393814086914, 25.7017822265625, 2.1746826171875, 63.70176696777344, 82.82218933105469, -5.023956298828125, -32.12739562988281, 89.577392578125, 32.89398193359375, 105.69256591796875, 18.103660583496094, 31.461883544921875, 78.1009521484375, 101.0277099609375, 31.01580810546875, -71.62831115722656, 27.29779052734375, -9.689109802246094, 1.6352996826171875, -57.7265625, -25.991302490234375, 38.4847412109375, -2.5675811767578125, 85.67263793945312, 19.827713012695312, -6.515407562255859, -7.894744873046875, -19.113067626953125, 28.85980224609375, 36.98461151123047, 45.218170166015625, -20.7244873046875, 29.783355712890625, -10.794219970703125, 55.486907958984375, 124.11676025390625, 79.71841430664062, 54.646751403808594, -29.3837890625, -2.74969482421875, 47.425018310546875, 42.5235595703125, -22.989501953125, 14.683090209960938, 53.235809326171875, -14.633026123046875, -6.9242095947265625, 59.981048583984375, -2.9720840454101562, 10.7432861328125, 101.40579223632812, 9.770980834960938, 97.68499755859375, 33.784454345703125, -28.76824951171875, -0.9528961181640625, 97.061767578125, -1.971923828125, 33.6083984375, -32.74456787109375, -28.32483673095703, 87.36923217773438, 2.0633544921875, -28.38977813720703, -71.60092163085938, -12.585765838623047, 0.0, 61.82322692871094, 117.40264892578125, 35.8807373046875, 10.832405090332031, 65.51838684082031, 9.974380493164062, 22.672836303710938, 77.495361328125, 73.21636962890625, 78.74375915527344, 17.7733154296875, 90.19332885742188, 63.36383056640625, -3.702545166015625, 48.91131591796875, 29.995452880859375, 64.09478759765625, 2.7200355529785156, -6.830535888671875, 66.96112060546875, 80.06878662109375, -22.1395263671875, 6.695716857910156, 27.23590087890625, 25.162109375, 92.5777587890625, 57.56695556640625, -2.579578399658203, 18.707275390625, 111.72088623046875, 54.33165740966797, 77.16728210449219, 19.71136474609375, 88.00230407714844, 50.222259521484375, 81.66282653808594, 79.66275024414062, 19.278549194335938, -58.662818908691406, 28.9188232421875, 50.79188537597656, 7.400413513183594, 63.40223693847656, 0.925506591796875, -40.50640869140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000262.npy"}
|
|
{"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 27.036659240722656, "std": 47.08371353149414, "min": -95.98301696777344, "p10": -23.590091896057125, "median": 18.783235549926758, "p90": 91.42005462646485, "max": 144.68658447265625, "pos_frac": 0.6875, "sample": [91.24551391601562, 24.5213623046875, 93.27658081054688, 21.97283935546875, -4.58441162109375, 55.490447998046875, -25.488555908203125, 5.70941162109375, 56.457298278808594, 50.758544921875, -1.9342670440673828, 50.01722717285156, 25.472503662109375, 23.315185546875, -30.28955078125, 3.8991355895996094, 1.7701530456542969, -58.678955078125, 50.49200439453125, 11.06805419921875, 46.12261962890625, -8.099700927734375, 1.0523529052734375, 55.84346008300781, -10.15826416015625, 86.34663391113281, 35.092803955078125, 49.478370666503906, -2.40216064453125, -30.79351806640625, -14.866073608398438, 41.846343994140625, 96.662109375, 80.69915771484375, 108.455810546875, -15.0784912109375, -3.905242919921875, -7.86358642578125, -41.5428466796875, 89.38398742675781, 12.203369140625, -6.215576171875, -16.090866088867188, 101.86004638671875, 65.06390380859375, 11.411407470703125, 58.09356689453125, -0.6747894287109375, 81.72775268554688, 25.66943359375, 52.018829345703125, -19.90191650390625, -15.188156127929688, 66.67173767089844, 95.28646850585938, 61.29595947265625, -69.30693054199219, 7.3646240234375, 79.02413940429688, 84.43299865722656, 0.0, 24.800567626953125, -95.98301696777344, 18.455829620361328, 40.388427734375, 7.7196044921875, -0.3008270263671875, 1.5489654541015625, 7.621788024902344, 23.35418701171875, 87.50820922851562, 2.98590087890625, -46.25044250488281, -22.776464462280273, 26.5797119140625, 7.369140625, 57.618865966796875, 69.02618408203125, 59.5322265625, 3.137359619140625, 4.573112487792969, 5.3115234375, -40.36175537109375, 14.075149536132812, 21.347457885742188, 45.384220123291016, -61.321319580078125, 108.43832397460938, 93.90219116210938, 5.291099548339844, -18.83349609375, 103.22747802734375, 9.799934387207031, -80.32147216796875, -17.692848205566406, -5.787750244140625, -62.89167785644531, 89.88229370117188, 19.110641479492188, -38.03558349609375, 69.58216857910156, -16.07318115234375, 0.0, 5.720771789550781, 62.702606201171875, 50.684661865234375, 144.68658447265625, 41.052452087402344, 129.66824340820312, -4.873390197753906, 50.56365966796875, 13.799613952636719, 91.82731628417969, 0.0, 95.35638427734375, 75.7674789428711, 17.298049926757812, 32.24388885498047, 78.9301986694336, 9.75161361694336, 81.45325469970703, 85.36625671386719, 32.02839660644531, -7.51605224609375, 132.54598999023438, -6.775463104248047, 62.223388671875, -10.266448974609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000263.npy"}
|
|
{"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 33.16969299316406, "std": 44.060367584228516, "min": -88.98654174804688, "p10": -14.349726867675777, "median": 25.855886459350586, "p90": 96.6925308227539, "max": 138.8817138671875, "pos_frac": 0.7578125, "sample": [-3.3023452758789062, -12.132675170898438, 120.79647827148438, 28.95635986328125, 107.31622314453125, 75.82256317138672, 51.56791687011719, 65.83953857421875, 0.2857227325439453, 7.2081298828125, 4.859760284423828, 88.10357666015625, 36.787109375, 46.06114196777344, -18.21197509765625, 77.93692016601562, 53.57707214355469, 81.22319030761719, -6.854839324951172, 0.4733848571777344, 89.15176391601562, 79.20538330078125, -22.057098388671875, 47.640106201171875, -36.830299377441406, 113.812255859375, 82.1513671875, -0.6518516540527344, 41.16326904296875, 67.2314453125, -24.759109497070312, 50.11859893798828, 83.74334716796875, 82.94889831542969, 112.7806396484375, 0.0, 0.91973876953125, 78.269775390625, 44.39910125732422, 18.19464874267578, -19.492706298828125, 99.30740356445312, -17.635452270507812, 54.17724609375, 54.68297576904297, 94.4814453125, -22.469802856445312, 111.96347045898438, -52.2235107421875, 28.713973999023438, 4.8236083984375, 11.7308349609375, 84.82794189453125, 53.59913635253906, 84.0703125, -60.116607666015625, 91.21731567382812, 71.66960144042969, 19.112730026245117, 13.812522888183594, 43.753509521484375, 2.290050506591797, 29.999176025390625, 36.08782958984375, 11.4769287109375, -12.941558837890625, -4.8267822265625, 14.537994384765625, 14.712692260742188, 85.52099609375, -2.9111328125, 10.646484375, -88.98654174804688, -7.221412658691406, 27.281246185302734, -1.0287551879882812, 27.585834503173828, 8.498359680175781, 100.20800018310547, 97.91989135742188, 70.173828125, -0.6521835327148438, 1.4289894104003906, 12.207763671875, -2.4897994995117188, 138.8817138671875, 2.2694244384765625, 0.0, 22.94451904296875, 15.160480499267578, -36.91326904296875, 34.868865966796875, 77.61333465576172, 101.54251098632812, 19.39593505859375, 69.39476013183594, -4.349557876586914, 31.85797119140625, 60.09264373779297, 15.85595703125, 96.16651916503906, 49.33648681640625, -10.45904541015625, -12.171905517578125, 6.170074462890625, 5.053497314453125, 27.30084228515625, 8.477447509765625, -17.718536376953125, 65.62159729003906, 0.11053466796875, 38.983306884765625, 37.325286865234375, 113.51113891601562, 15.22662353515625, 105.00228881835938, -4.668304443359375, 18.38544464111328, 103.1885986328125, -10.254989624023438, -39.21051788330078, 11.077911376953125, 52.94500732421875, 24.430526733398438, 73.87509155273438, 1.7807025909423828, 31.18902587890625, 13.16180419921875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000264.npy"}
|
|
{"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 31.06427001953125, "std": 46.545738220214844, "min": -78.43125915527344, "p10": -31.381308746337886, "median": 31.13134765625, "p90": 91.98191833496092, "max": 120.77484130859375, "pos_frac": 0.6875, "sample": [5.072771072387695, 119.211669921875, 79.52078247070312, -40.927154541015625, 35.308807373046875, 6.703948974609375, 72.04297637939453, 52.04188537597656, 52.413726806640625, -48.4511604309082, 13.078338623046875, 32.59002685546875, 66.4819564819336, -12.0513916015625, 94.84895324707031, 78.99815368652344, 34.391082763671875, 102.64556884765625, 78.330810546875, 49.135765075683594, 115.6822509765625, 90.65277099609375, 111.044921875, 27.294334411621094, 120.77484130859375, 66.6319580078125, 67.1605224609375, 106.79913330078125, 87.08926391601562, -3.71490478515625, -33.107933044433594, 68.3828125, 109.13046264648438, 9.67193603515625, 79.63603210449219, 96.7120361328125, 51.548248291015625, 74.85922241210938, 54.91241455078125, 99.53244018554688, -1.6968994140625, 1.475067138671875, -42.596435546875, 31.33953857421875, -5.566070556640625, -18.723114013671875, -18.121337890625, -34.611541748046875, -3.5600814819335938, 66.4425048828125, -0.41943359375, -10.129653930664062, -26.188980102539062, 4.1017608642578125, -18.559738159179688, 86.0267333984375, 73.5196533203125, 8.86322021484375, 47.41145324707031, -78.43125915527344, 17.510005950927734, 76.04097747802734, 5.0601806640625, 68.05867004394531, 40.32000732421875, -40.87994384765625, 90.75318908691406, 17.471405029296875, 43.0609130859375, -21.755088806152344, -34.46331787109375, 90.54389953613281, 79.83259582519531, 11.902240753173828, 0.0, -6.409637451171875, -63.402435302734375, -63.8013916015625, 79.068359375, 86.45875549316406, 30.92315673828125, 4.4881591796875, 61.768272399902344, -0.8930206298828125, 73.18843841552734, 30.548370361328125, 0.0, -2.973480224609375, -17.623016357421875, 57.573974609375, 54.44061279296875, -27.2269287109375, -1.074615478515625, -0.41654205322265625, 41.4595947265625, 7.3159332275390625, 31.822021484375, 117.64346313476562, 34.746002197265625, -30.641326904296875, 17.700340270996094, 42.883026123046875, 48.96775817871094, 43.42301559448242, 5.152496337890625, 55.760986328125, -52.476898193359375, 101.6234130859375, 73.68939208984375, -37.06903076171875, -2.5022811889648438, 16.1307373046875, 6.6859130859375, -6.95916748046875, 43.02082824707031, -26.425323486328125, -0.5787162780761719, 103.0904541015625, 60.06613540649414, 63.780303955078125, 67.14306640625, -6.152130126953125, 27.04784393310547, 26.851409912109375, 20.073801040649414, -33.55615234375, 7.7937469482421875, 37.96734619140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000265.npy"}
|
|
{"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 32.78792953491211, "std": 44.197723388671875, "min": -96.90309143066406, "p10": -10.786685943603516, "median": 27.396056175231934, "p90": 86.87777099609374, "max": 120.92489624023438, "pos_frac": 0.8125, "sample": [39.6324462890625, 67.24819946289062, 120.92489624023438, 18.124786376953125, -5.376251220703125, -11.160179138183594, 67.90442657470703, -0.1068115234375, -16.33416748046875, 5.578392028808594, 7.43212890625, -0.681427001953125, -1.5560455322265625, 46.4747314453125, 43.64329528808594, -7.7200927734375, -10.626617431640625, 27.014175415039062, 27.165311813354492, -63.056884765625, 85.2213134765625, 53.70240020751953, 3.804168701171875, 28.244354248046875, 97.45156860351562, 50.968994140625, -1.00860595703125, 38.29624938964844, 104.4718017578125, 58.63218688964844, 0.4481925964355469, 118.84616088867188, -2.5617523193359375, 38.436065673828125, 5.322734832763672, 80.1068115234375, -36.947998046875, 7.271060943603516, 48.44415283203125, 57.831329345703125, 5.0081787109375, 23.3656005859375, 16.802627563476562, 4.169654846191406, 89.1193618774414, 79.5283203125, 66.34574890136719, 83.94608306884766, 10.99935531616211, 26.836776733398438, 10.91314697265625, 72.23947143554688, -73.92317199707031, -82.00294494628906, 1.176462173461914, -38.83903503417969, 4.831689834594727, 31.418487548828125, 29.680694580078125, 27.626800537109375, 85.66751098632812, 63.06761169433594, 13.993576049804688, 13.819610595703125, 18.789045333862305, 90.45404052734375, 58.29368591308594, 53.69293212890625, 7.97991943359375, 23.718673706054688, 30.0662841796875, 104.84591674804688, 49.910888671875, 39.162017822265625, 79.55220794677734, 118.2939453125, 29.6485595703125, -4.13653564453125, -53.657135009765625, -65.7332992553711, -96.90309143066406, 99.85441589355469, 12.321136474609375, 83.90837097167969, -12.47210693359375, 45.69342041015625, 33.364593505859375, 17.63128662109375, 86.75119018554688, 60.04853820800781, -48.578033447265625, 6.22308349609375, -11.965362548828125, 10.673004150390625, 83.42703247070312, 54.47930908203125, 60.10923767089844, 67.49969482421875, 26.67022705078125, 7.7468109130859375, 15.5555419921875, 51.33929443359375, 17.866424560546875, 111.66871643066406, 12.574655532836914, 9.6553955078125, -10.39327621459961, 2.4539413452148438, 30.200538635253906, -3.812103271484375, 87.17312622070312, 75.28863525390625, 110.83796691894531, 5.338783264160156, 85.8721923828125, 12.286376953125, 69.49496459960938, 59.76560974121094, 11.677631378173828, 17.403289794921875, 116.71063232421875, 9.997737884521484, 84.10475158691406, 7.817440032958984, 77.39645385742188, 68.50640869140625, 49.40850830078125, 86.00430297851562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000266.npy"}
|
|
{"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 26.779220581054688, "std": 48.879886627197266, "min": -93.6275634765625, "p10": -37.7837043762207, "median": 24.9840087890625, "p90": 85.7469268798828, "max": 134.4979248046875, "pos_frac": 0.71875, "sample": [4.235044479370117, 75.41816711425781, 69.58279418945312, 69.26971435546875, 123.490478515625, 29.498931884765625, 3.1851959228515625, 22.47076416015625, -46.247314453125, 3.795654296875, -0.04180145263671875, 23.319183349609375, -10.73260498046875, 54.63409423828125, 33.04042053222656, -20.712310791015625, -1.28173828125, 103.1781005859375, 101.75994873046875, 28.962799072265625, -56.14837646484375, 108.513427734375, 84.6494140625, 4.5186767578125, 21.408939361572266, -22.394439697265625, 46.43275451660156, 70.82139587402344, 59.25604248046875, -91.77276611328125, 49.666168212890625, 18.4266357421875, 24.512969970703125, -14.062103271484375, 11.2216796875, 62.3851318359375, -12.405143737792969, 101.61053466796875, 85.73959350585938, 62.47193145751953, -52.6605224609375, 82.88543701171875, 35.139732360839844, 63.18017578125, 11.599922180175781, 134.4979248046875, 11.34136962890625, -18.3111572265625, 82.46365356445312, 91.07090759277344, -70.62156677246094, -59.834869384765625, 59.443695068359375, 68.93070220947266, 8.34124755859375, 133.4256591796875, 61.95215606689453, 62.5894775390625, 100.45675659179688, 7.58680534362793, 58.07228088378906, 59.647438049316406, -72.5277099609375, 12.710662841796875, 33.7939453125, 1.0713958740234375, 58.69635009765625, -43.1457405090332, 68.80804443359375, 73.54165649414062, 10.074066162109375, 32.4267578125, 56.588592529296875, -93.6275634765625, 39.29136657714844, 70.56283569335938, 0.862762451171875, 56.140625, 4.248418807983398, 0.0818939208984375, -36.15166473388672, 56.1607666015625, -7.125587463378906, 35.30322265625, -7.099098205566406, 18.465805053710938, -34.316162109375, 56.86842346191406, 18.08349609375, 62.923828125, -14.15810775756836, -41.591796875, 25.455047607421875, -9.47412109375, 3.14422607421875, 78.65150451660156, 39.855812072753906, 52.26628494262695, 44.77293395996094, -8.796142578125, 85.7640380859375, 26.658416748046875, -20.37200927734375, 46.917724609375, -10.20819091796875, 23.91558837890625, 26.250656127929688, 76.50050354003906, -6.932579040527344, 128.7869873046875, -19.334060668945312, -45.23492431640625, 69.054931640625, 19.99200439453125, 68.81690979003906, 1.9957389831542969, 94.62225341796875, 76.11714172363281, -11.14996337890625, -21.4500732421875, 91.61289978027344, -76.58900451660156, -9.575439453125, 1.3213443756103516, 34.841758728027344, -55.85546875, -29.653244018554688, 5.214111328125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000267.npy"}
|
|
{"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 31.050168991088867, "std": 45.50742721557617, "min": -77.84950256347656, "p10": -15.892996978759754, "median": 22.933494567871094, "p90": 93.90908660888671, "max": 133.32110595703125, "pos_frac": 0.765625, "sample": [23.760513305664062, 5.45068359375, 60.39868927001953, 64.19271850585938, 88.83596801757812, 6.24554443359375, -0.990509033203125, -12.987083435058594, -35.552528381347656, -41.86238098144531, -6.305908203125, 116.8826904296875, 16.3759765625, 98.69965362548828, 1.0032958984375, 59.07989501953125, 133.32110595703125, -6.717555999755859, 1.943115234375, 71.79486083984375, 79.55206298828125, 80.77162170410156, 26.798988342285156, -77.84950256347656, 58.79425048828125, 29.900955200195312, -2.419626235961914, 37.454833984375, 36.314483642578125, 14.211006164550781, 6.84385871887207, 68.28453063964844, 51.20458984375, 71.12491607666016, 21.732620239257812, 1.39447021484375, -77.58499145507812, -53.877655029296875, -22.6734619140625, 127.15570068359375, 73.71243286132812, 27.678497314453125, 21.93890380859375, 96.865234375, 5.233642578125, 13.839706420898438, 54.34303283691406, -68.3818359375, -32.84080505371094, -6.725151062011719, 23.05694580078125, 68.60418701171875, 44.801025390625, 92.22073364257812, 4.501544952392578, 35.147918701171875, 108.5029296875, -7.0070648193359375, 45.318115234375, 6.131927490234375, 123.339111328125, 16.260154724121094, 87.2831039428711, -7.176727294921875, -3.4791259765625, 24.24371337890625, 92.94282531738281, 30.882110595703125, -26.095489501953125, -8.8836669921875, 65.68647766113281, 14.45513916015625, 81.18716430664062, 72.30067443847656, 18.942840576171875, -8.39385986328125, 43.51116943359375, 110.34780883789062, 96.1636962890625, 43.95574951171875, 52.320281982421875, 57.591712951660156, 59.380859375, -9.006019592285156, 55.626434326171875, 77.08999633789062, -43.15330505371094, 42.761444091796875, 35.684112548828125, 7.110321044921875, 21.100830078125, 67.83848571777344, 110.97763061523438, 50.54827880859375, 2.81884765625, 92.20156860351562, 14.888946533203125, 2.4857177734375, 17.184051513671875, 38.27972412109375, -5.9227294921875, 41.075927734375, 31.195556640625, 1.0102386474609375, 4.882345199584961, 75.46002197265625, 10.906707763671875, -31.668182373046875, -3.93603515625, 26.8853759765625, 122.52911376953125, 9.72281265258789, 60.810211181640625, 79.08218383789062, 0.3349609375, 100.21920776367188, -57.74187469482422, -9.174911499023438, 127.936279296875, 11.570083618164062, 1.473480224609375, -27.95758056640625, 15.557891845703125, 22.810043334960938, 8.8092041015625, 16.48821258544922, -6.783935546875, 0.0], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000268.npy"}
|
|
{"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 28.206409454345703, "std": 48.3121452331543, "min": -137.79922485351562, "p10": -24.470391845703126, "median": 26.192344665527344, "p90": 90.67009048461914, "max": 131.18377685546875, "pos_frac": 0.7109375, "sample": [-69.42323303222656, 37.15623474121094, 77.3038330078125, -41.51881408691406, 38.19268798828125, -7.57916259765625, 38.73268127441406, 62.84033203125, 20.266876220703125, -52.107879638671875, 67.57709503173828, 7.154640197753906, -7.989097595214844, 65.04639434814453, -24.400390625, 81.46697998046875, 5.362762451171875, -2.942840576171875, 90.48538970947266, 25.21092987060547, 131.18377685546875, 107.74076843261719, 19.917129516601562, 98.11660766601562, 103.87869262695312, -18.000015258789062, 24.45440673828125, 68.90785217285156, 0.9873046875, 50.136627197265625, 3.0309524536132812, 61.09571838378906, 79.13128662109375, -55.29634094238281, 77.85421752929688, 0.0, 28.36431884765625, -17.356338500976562, 53.524879455566406, 70.74038696289062, -36.16089630126953, 17.692779541015625, 102.38382720947266, 53.44584655761719, 7.135162353515625, 55.503578186035156, 108.17071533203125, 0.0, 60.48332214355469, 55.479583740234375, 74.4881362915039, -13.128631591796875, 89.41999816894531, -73.77713012695312, -55.63743591308594, 87.50160217285156, 67.90123748779297, -10.190193176269531, -42.41046142578125, -3.3149261474609375, 91.10105895996094, 53.84370422363281, 40.68060302734375, 14.832855224609375, 70.2669906616211, -24.63372802734375, 12.4979248046875, 61.128814697265625, 83.58903503417969, 14.696769714355469, 50.633819580078125, 61.903411865234375, 0.0, 92.66893005371094, 13.970108032226562, 16.57073974609375, -95.2987060546875, 2.653430938720703, 36.338134765625, 5.433277130126953, -5.97265625, -11.0789794921875, 96.821533203125, 16.59996795654297, 52.90557861328125, 24.968536376953125, 29.69403076171875, 46.746002197265625, 38.277862548828125, 93.87409973144531, 45.19921875, 27.614532470703125, 13.33477783203125, -16.953399658203125, 86.44876098632812, 19.6876220703125, -12.42449951171875, 70.98020935058594, 7.2416534423828125, 45.704185485839844, 61.14735412597656, 27.17375946044922, 2.755044937133789, -17.311973571777344, 2.502836227416992, -70.6024169921875, 57.26251220703125, 101.74478149414062, -4.894256591796875, 65.44556427001953, -22.627227783203125, 13.306358337402344, 61.37060546875, -15.939498901367188, 71.75294494628906, -49.77187728881836, -20.836883544921875, -7.315010070800781, 30.038803100585938, -6.057701110839844, 21.846282958984375, 96.91770935058594, 43.389862060546875, -137.79922485351562, -17.49224853515625, 11.883148193359375, 86.6536865234375, 109.09736633300781], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000269.npy"}
|
|
{"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 33.1375617980957, "std": 50.721832275390625, "min": -133.39956665039062, "p10": -27.17789611816406, "median": 31.145889282226562, "p90": 100.77420959472656, "max": 126.4278564453125, "pos_frac": 0.75, "sample": [-63.638427734375, 47.16465759277344, 69.4447021484375, 68.5098648071289, -30.97426986694336, 105.84890747070312, -7.918647766113281, 24.863922119140625, 31.913818359375, -41.8175048828125, 86.55990600585938, 15.037109375, 18.69097900390625, 27.757705688476562, 16.687149047851562, 74.09756469726562, -10.896636962890625, 101.97898864746094, 39.21460723876953, -40.54624938964844, -44.5631103515625, 38.404029846191406, 5.182243347167969, 36.28648376464844, 11.071701049804688, 3.610870361328125, 84.01970672607422, 82.47344207763672, -12.232818603515625, 0.9671363830566406, 37.72906494140625, 70.87107849121094, 36.50634765625, 3.9112319946289062, 5.446815490722656, 8.676704406738281, -35.85882568359375, 15.01495361328125, -92.055908203125, 31.297470092773438, -23.166915893554688, 106.00006103515625, 26.371566772460938, 10.820037841796875, 80.661865234375, 0.158935546875, -4.166500091552734, 98.62741088867188, 64.78671264648438, -85.8839111328125, -39.6080322265625, 85.90939331054688, 94.21499633789062, 90.29145812988281, 25.998062133789062, 57.977264404296875, 104.98699951171875, -26.22967529296875, 87.42648315429688, 4.9326171875, 21.1251220703125, -16.581787109375, 10.651824951171875, 15.405574798583984, 92.52108764648438, 86.60655212402344, 101.88389587402344, 3.5875892639160156, 89.4703369140625, 38.485015869140625, 68.25064086914062, -6.381744384765625, -87.53750610351562, 88.8894271850586, -24.201690673828125, 100.29862976074219, 40.80500793457031, 35.46704864501953, 68.03213500976562, 27.220916748046875, -15.829010009765625, 8.638458251953125, 29.55963134765625, 74.31564331054688, 30.247509002685547, -5.807502746582031, -1.45770263671875, 16.934951782226562, 1.9465408325195312, -2.8690872192382812, -29.390411376953125, -39.1983642578125, 78.87591552734375, 107.91778564453125, -6.8689727783203125, 18.1693115234375, 75.00985717773438, 104.35281372070312, 47.885467529296875, -23.20220947265625, 85.97756958007812, 109.89971923828125, 47.323638916015625, 57.78082275390625, 30.994308471679688, -133.39956665039062, -11.584136962890625, -7.60675048828125, 102.3583984375, 16.31103515625, 41.162689208984375, 69.407958984375, 37.69635009765625, 71.193359375, 102.32122802734375, 108.268798828125, 98.53753662109375, 34.48497009277344, 8.401947021484375, 104.70263671875, -18.041656494140625, 126.4278564453125, 87.31649780273438, 92.41534423828125, 59.810523986816406, 79.58123779296875, 60.08502197265625, -22.2938232421875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000270.npy"}
|
|
{"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 31.428512573242188, "std": 45.916893005371094, "min": -66.6171646118164, "p10": -18.575921630859376, "median": 22.307109832763672, "p90": 86.9860450744629, "max": 156.10321044921875, "pos_frac": 0.7578125, "sample": [75.7767562866211, -66.6171646118164, -8.07861328125, 25.21221923828125, 10.695816040039062, 0.141845703125, 11.598976135253906, -9.85867691040039, 30.0927734375, -4.808479309082031, -4.958251953125, 18.70599365234375, 87.96803283691406, 4.963043212890625, 26.768386840820312, 30.2880859375, 57.10296630859375, 20.697580337524414, 23.2747802734375, 22.14044952392578, 22.473770141601562, 8.487419128417969, 86.46249389648438, 78.59785461425781, 2.158111572265625, 2.9992809295654297, -49.77838134765625, 77.83425903320312, 11.512760162353516, -22.12322998046875, 80.15045166015625, 103.08917999267578, -2.1875, 31.391845703125, 4.007728576660156, 24.234100341796875, 27.3419189453125, 16.79174041748047, 57.73974609375, -43.91117858886719, 81.43023681640625, 17.567520141601562, -0.708648681640625, 84.24990844726562, 72.99954223632812, 13.84710693359375, 42.34356689453125, -0.7954025268554688, 52.412841796875, 10.616424560546875, -5.01568603515625, 66.78152465820312, 93.35530090332031, 44.92698669433594, 156.10321044921875, 116.605712890625, 0.9639301300048828, 97.169921875, 26.54010009765625, 73.12548828125, 68.97323608398438, 7.489013671875, -10.256080627441406, 143.00924682617188, 68.8995361328125, -58.110595703125, -18.80206298828125, 97.07937622070312, -10.058502197265625, -11.16619873046875, 86.56519317626953, 131.51705932617188, -22.559539794921875, 56.825313568115234, -9.726318359375, 16.765899658203125, -41.518760681152344, 27.2430419921875, 30.750091552734375, 61.653839111328125, -18.47900390625, -54.140380859375, 82.77798461914062, -8.22967529296875, 125.722900390625, 66.07940673828125, -3.3900909423828125, -35.442901611328125, 29.718738555908203, 19.3585205078125, 4.859661102294922, 95.925048828125, 74.63934326171875, -10.242843627929688, 1.8567962646484375, 73.94944763183594, 3.4192943572998047, 75.84909057617188, 79.66011047363281, 24.229652404785156, 9.67095947265625, 20.476272583007812, -3.78704833984375, 70.4542236328125, 110.54461669921875, 1.054901123046875, -43.78790283203125, 15.19287109375, 17.373550415039062, 63.04736328125, 31.1842041015625, 75.6898193359375, 83.82023620605469, 8.936904907226562, 52.65785217285156, 12.281890869140625, 59.74488830566406, 82.36309814453125, 1.2386016845703125, -9.207611083984375, 84.46194458007812, -57.93098449707031, 82.1280517578125, 105.041015625, 14.96807861328125, 3.4830169677734375, 25.173423767089844, -49.01698303222656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000271.npy"}
|
|
{"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 29.400623321533203, "std": 46.898921966552734, "min": -92.30078125, "p10": -22.31761322021484, "median": 23.364164352416992, "p90": 88.84014282226562, "max": 143.7344970703125, "pos_frac": 0.765625, "sample": [32.2342529296875, 0.0, -27.576019287109375, 41.46015930175781, -52.93572998046875, 50.6998291015625, 112.70437622070312, 23.84613037109375, 79.08865356445312, 2.5156097412109375, -24.542892456054688, 16.71971321105957, 42.5296630859375, 84.00953674316406, 56.03839111328125, 113.96768188476562, 111.55426025390625, 80.97219848632812, 32.439910888671875, 16.24072265625, 82.63664245605469, 8.6055908203125, -10.02618408203125, -4.348052978515625, 77.31712341308594, -9.928840637207031, 3.915212631225586, 117.98831176757812, 1.8403434753417969, 7.39019775390625, 86.61041259765625, 66.12041473388672, 56.661376953125, 3.724346160888672, 99.0218276977539, 58.69720458984375, 114.02932739257812, 96.62957763671875, 13.110443115234375, -1.9494857788085938, 55.45512390136719, -37.70599365234375, -28.351593017578125, 57.513946533203125, -10.33966064453125, 23.762802124023438, 76.50527954101562, -39.833648681640625, 1.9355239868164062, 2.192476272583008, 103.97607421875, 80.4728775024414, -12.078929901123047, 20.469390869140625, 60.93360900878906, 97.67861938476562, 44.45454406738281, 42.790740966796875, 68.92727661132812, 0.4039344787597656, -1.3323116302490234, 9.009796142578125, 8.008529663085938, 11.06494140625, 58.02845001220703, 22.965526580810547, 143.7344970703125, 22.27618408203125, 77.56637573242188, 11.39959716796875, 86.61512756347656, -9.771919250488281, 52.2760009765625, 10.021072387695312, -87.4815673828125, -35.672332763671875, -0.638824462890625, 29.310256958007812, 3.166717529296875, 59.59326171875, 52.80859375, 42.27398681640625, -6.141845703125, 115.23103332519531, 0.8612060546875, -92.30078125, 26.417638778686523, 13.443683624267578, 25.191162109375, 2.744874954223633, 32.586753845214844, 64.30609893798828, 5.0547332763671875, -66.12173461914062, 57.01271057128906, 39.52685546875, 46.441741943359375, 77.73973846435547, -15.08587646484375, 3.5066184997558594, 0.0, -21.363922119140625, 10.480184555053711, -66.37835693359375, -16.83502197265625, 13.03814697265625, 33.311614990234375, 0.9774398803710938, 0.31383514404296875, -2.53936767578125, 62.89019012451172, 143.24462890625, -77.48834228515625, 44.76372528076172, 87.4991455078125, -57.10821533203125, 18.041465759277344, 2.9017086029052734, 7.106405258178711, -3.465545654296875, 87.66912841796875, 40.29309844970703, 28.849212646484375, 14.9117431640625, 54.43475341796875, 91.572509765625, 50.99000549316406, 50.3586311340332], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000272.npy"}
|
|
{"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 32.34339904785156, "std": 49.05825424194336, "min": -105.52423095703125, "p10": -26.439694213867185, "median": 30.17559814453125, "p90": 94.83637084960938, "max": 158.27056884765625, "pos_frac": 0.6796875, "sample": [37.87603759765625, 118.6558837890625, 110.39215087890625, 11.442564010620117, 122.35033416748047, -5.28826904296875, 131.75726318359375, 19.7294921875, 26.547149658203125, 7.358589172363281, -21.01800537109375, 32.00129699707031, -1.5151176452636719, 57.7149772644043, 158.27056884765625, -74.4381103515625, 52.46124267578125, 122.11471557617188, 12.78643798828125, 40.084075927734375, -42.99287414550781, 61.32383728027344, 87.34455871582031, 26.082218170166016, 94.98577880859375, 22.834625244140625, -69.58535766601562, 28.35186767578125, 55.070343017578125, -7.387786865234375, 18.804561614990234, -28.95220947265625, -7.421760559082031, 83.72024536132812, -16.391357421875, -1.596710205078125, 6.74932861328125, 10.92266845703125, -5.20458984375, 79.12911987304688, -5.455928802490234, -4.609710693359375, 0.6421432495117188, 39.116485595703125, 41.7281494140625, 21.1878662109375, 73.67005157470703, -0.22011375427246094, -25.8770751953125, 8.283538818359375, 53.1793212890625, -20.678436279296875, 104.97402954101562, -16.95947265625, 52.368072509765625, -3.93365478515625, -0.56219482421875, 45.710845947265625, 15.724868774414062, 70.01200103759766, -39.9320068359375, 75.40642547607422, -0.6771011352539062, 69.10466003417969, -15.301025390625, 97.47235107421875, 64.8408203125, 0.856170654296875, 68.18199157714844, -1.0635986328125, 84.79721069335938, -7.135528564453125, -54.09144592285156, -15.310470581054688, 39.5721435546875, 93.977294921875, 26.942352294921875, 88.45956420898438, 77.44987487792969, -34.783477783203125, 70.50733947753906, 81.54145050048828, 27.43848419189453, -28.934425354003906, 48.301025390625, 80.66302490234375, 28.25640869140625, 36.33936309814453, 99.83575439453125, 35.93647766113281, 37.68487548828125, -37.92795181274414, -4.1563568115234375, 76.07058715820312, -105.52423095703125, 31.99932861328125, -68.421142578125, 80.94320678710938, 94.7723388671875, 5.234619140625, 107.496337890625, 34.582489013671875, 75.4619140625, 88.42892456054688, 89.096923828125, -4.2825822830200195, -46.345855712890625, 17.8154296875, 87.48739624023438, 70.09795379638672, 41.834869384765625, 45.234619140625, -3.3093109130859375, 43.75360107421875, 11.6041259765625, 57.228302001953125, -27.752471923828125, 97.02420043945312, 19.651336669921875, -5.7036285400390625, -8.430004119873047, 99.36856079101562, 36.931640625, 78.46016693115234, 81.738525390625, -24.555450439453125, 80.23272705078125, -15.892730712890625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000273.npy"}
|
|
{"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 34.89495849609375, "std": 47.98119354248047, "min": -101.67144775390625, "p10": -15.623516845703126, "median": 24.44341278076172, "p90": 92.22592239379883, "max": 133.86529541015625, "pos_frac": 0.796875, "sample": [84.8345718383789, 88.44851684570312, -40.32647705078125, 75.46076202392578, -2.79193115234375, 58.5946044921875, 45.200469970703125, 31.243804931640625, 67.64888763427734, 7.320014953613281, 30.034271240234375, 80.70323944091797, 117.00566101074219, 22.223861694335938, 6.426826477050781, 75.73257446289062, 129.32528686523438, 109.76731872558594, -60.1302490234375, 5.991706848144531, 13.30902099609375, 94.49105834960938, 85.39834594726562, 39.14642333984375, -6.930816650390625, -4.33746337890625, -57.341636657714844, -34.51496887207031, 9.862800598144531, 19.14117431640625, 20.9488525390625, 24.532745361328125, 74.03797912597656, 38.488525390625, 85.2630615234375, 26.78857421875, 25.274826049804688, 77.64094543457031, 14.784896850585938, 53.83984375, 107.74774169921875, -6.091026306152344, -3.323486328125, 8.938545227050781, -15.58380126953125, 91.27362060546875, 21.30084228515625, 13.07143783569336, 39.865264892578125, 116.59507751464844, -8.770103454589844, 8.95233154296875, 82.26455688476562, 15.834136962890625, -88.53048706054688, 22.47089385986328, -15.1646728515625, -19.065460205078125, 0.5758781433105469, 6.70513916015625, 19.906396865844727, 6.24559211730957, -21.73810577392578, 55.76817321777344, -27.948410034179688, -1.1682548522949219, 76.43226623535156, 0.5309791564941406, 62.335784912109375, 82.819091796875, 11.705474853515625, -96.27725219726562, 97.51458740234375, 24.354080200195312, 53.47216796875, 42.722625732421875, -4.66937255859375, 39.22560119628906, 133.86529541015625, 8.062118530273438, 131.88232421875, 76.36129760742188, 19.37841796875, 85.02406311035156, -1.9881439208984375, 53.604644775390625, 81.78315734863281, 88.70196533203125, 59.78448486328125, 94.44795989990234, 63.93927764892578, 22.109649658203125, -30.16399383544922, 62.334320068359375, 81.45343017578125, 76.2194595336914, 8.771499633789062, 97.49325561523438, 115.47543334960938, 20.610595703125, -61.14521789550781, 15.123199462890625, 10.4937744140625, -15.7161865234375, 36.47113037109375, 81.43862915039062, 4.990898132324219, 88.71183776855469, 20.193145751953125, 45.275634765625, 83.40963745117188, 74.74246978759766, 3.2015380859375, 15.750289916992188, 42.54905700683594, -10.017578125, 4.4761810302734375, 4.62103271484375, 8.507308959960938, 40.90838623046875, 7.804040908813477, -101.67144775390625, 79.95086669921875, -1.0392017364501953, 19.380611419677734, 74.01626586914062, 83.8330078125, 128.30923461914062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000274.npy"}
|
|
{"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 40.34345245361328, "std": 46.23505783081055, "min": -79.19363403320312, "p10": -7.188390350341794, "median": 32.202903747558594, "p90": 102.54137573242187, "max": 134.48440551757812, "pos_frac": 0.8046875, "sample": [134.48440551757812, -5.53790283203125, -2.4473228454589844, -64.21856689453125, 100.95828247070312, 47.58116149902344, 96.16622924804688, 19.89617919921875, 24.433837890625, -79.19363403320312, 16.163314819335938, 113.07891845703125, -29.84515380859375, 69.9112548828125, 0.0, 8.670623779296875, 13.42083740234375, 64.6473388671875, 3.4139366149902344, 2.7597732543945312, 0.0, 0.0, 80.48340606689453, 85.65289306640625, 68.66795349121094, 2.6613311767578125, 108.845947265625, 71.27987670898438, 18.25146484375, 63.50239562988281, 57.895263671875, 4.861394882202148, 70.93160247802734, 95.2506103515625, 109.99006652832031, 103.83920288085938, 0.7148590087890625, 0.72869873046875, -70.1514892578125, 39.899932861328125, 99.04278564453125, 27.942474365234375, 49.69921112060547, 20.838714599609375, 15.037612915039062, -4.1065826416015625, 7.3890380859375, 91.79006958007812, 103.09201049804688, 13.46234130859375, 99.8994140625, -36.89337158203125, 2.0504894256591797, 23.86322021484375, 65.4072265625, 16.13134765625, 66.54415893554688, 22.5928955078125, 93.97225952148438, 0.0, 17.0006103515625, 21.74211883544922, 27.60333251953125, 88.36135864257812, -5.553657531738281, 112.15567016601562, 112.69230651855469, 64.73373413085938, 31.245223999023438, 0.0, 5.027349472045898, 103.331787109375, 44.936798095703125, 77.850830078125, -4.4935150146484375, 38.9603271484375, 99.23159790039062, 121.44180297851562, 40.19383239746094, 5.728893280029297, 80.58026123046875, 89.27076721191406, 56.24749755859375, 67.85150146484375, 36.709808349609375, 2.2708187103271484, 20.550323486328125, 75.65234375, 56.270179748535156, 28.873779296875, -17.485076904296875, 40.411014556884766, 7.415374755859375, 103.5855712890625, 11.4578857421875, 87.63934326171875, -12.382354736328125, 3.0101470947265625, 126.8326416015625, 14.696609497070312, 33.62312316894531, 19.398193359375, 83.0428466796875, -6.379890441894531, 97.08717346191406, 86.94998931884766, 12.903816223144531, -36.42034912109375, -3.12677001953125, -9.07489013671875, 102.30538940429688, -47.87970733642578, 80.95741271972656, 3.045013427734375, 33.16058349609375, 4.432294845581055, 15.788543701171875, 70.5697021484375, 116.9796142578125, 84.02464294433594, -16.637672424316406, 64.12139892578125, 49.10118103027344, -9.631858825683594, 68.30067443847656, -14.0877685546875, 90.28216552734375, 84.07376098632812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000275.npy"}
|
|
{"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 36.630062103271484, "std": 45.97486114501953, "min": -68.66705322265625, "p10": -19.567381286621092, "median": 30.348499298095703, "p90": 99.78470764160156, "max": 150.06475830078125, "pos_frac": 0.7734375, "sample": [111.46295166015625, 24.491127014160156, 78.54674530029297, -4.527610778808594, 19.6524658203125, -1.866455078125, 5.351806640625, 44.902557373046875, 32.17181396484375, 43.888145446777344, 57.8416862487793, 2.888031005859375, 9.709457397460938, 2.609128952026367, 3.1218414306640625, 80.80863952636719, 40.042572021484375, 19.331268310546875, -28.84820556640625, 45.0985107421875, 106.41775512695312, 34.57490539550781, 8.413570404052734, 103.58285522460938, 59.473602294921875, -21.161468505859375, -19.08294677734375, -20.59515380859375, -7.9676513671875, 82.22280883789062, 35.79827880859375, 33.42671203613281, 60.9276123046875, 61.312889099121094, 59.38134765625, 12.00244140625, 30.294036865234375, -1.053152084350586, 37.73460388183594, 32.14912414550781, 18.74700927734375, 20.695892333984375, 92.80194091796875, 25.65911865234375, 5.725666046142578, 99.76754760742188, 58.5343017578125, 100.57781982421875, 14.33331298828125, -5.7784423828125, 5.8898468017578125, 62.794769287109375, 73.69961547851562, 26.195220947265625, 90.3866195678711, 11.048629760742188, 56.97538757324219, 24.690338134765625, 92.74166870117188, -32.451171875, 0.0, -42.832366943359375, 11.97503662109375, -55.334442138671875, 5.403099060058594, -29.7354736328125, 32.98362731933594, 120.70260620117188, -2.3333511352539062, -11.62786865234375, 33.84252166748047, -10.45623779296875, 76.537841796875, 88.68661499023438, 79.00713348388672, 97.6162109375, 86.28778076171875, -21.321990966796875, 30.40296173095703, 70.89814758300781, 90.98393249511719, 80.1201171875, 80.54501342773438, 39.6304931640625, 10.650733947753906, 49.298095703125, -68.66705322265625, 69.27519989013672, -4.2662811279296875, 28.6417236328125, -2.0140228271484375, 150.06475830078125, 53.1322021484375, 16.5604248046875, 8.709716796875, 6.9463653564453125, -3.3363494873046875, 114.0216064453125, 12.71990966796875, 99.80780029296875, 58.33990478515625, 21.617233276367188, 100.73199462890625, 2.0877685546875, -1.8880157470703125, 73.78793334960938, -13.703523635864258, -43.40340805053711, 23.797821044921875, 90.0797119140625, 4.9781036376953125, 74.13507080078125, 18.263832092285156, 113.84423828125, 94.81205749511719, -38.889556884765625, 99.77481079101562, 21.167755126953125, 72.2723388671875, -37.6724853515625, 113.96636962890625, 16.883819580078125, 72.20301818847656, -61.48188781738281, -19.126907348632812, 100.84649658203125, 130.06719970703125, 90.06623840332031], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000276.npy"}
|
|
{"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 26.359458923339844, "std": 51.31071090698242, "min": -112.760498046875, "p10": -35.47104110717773, "median": 24.2927188873291, "p90": 85.4694839477539, "max": 154.25604248046875, "pos_frac": 0.6953125, "sample": [1.5712890625, 4.453094482421875, -5.615488052368164, 80.3436279296875, 32.99371337890625, -19.160964965820312, -48.81317138671875, -6.41876220703125, 42.70704650878906, 29.5145263671875, 23.8739013671875, 32.03599548339844, 83.15516662597656, 85.22102355957031, -5.500518798828125, 68.14492797851562, 62.33456802368164, -77.31695556640625, 52.4989013671875, 0.78265380859375, 27.17706298828125, 7.860721588134766, 73.68975830078125, 45.17181396484375, 69.18206787109375, 74.98202514648438, -0.5891342163085938, -4.57427978515625, 82.24848937988281, 34.277320861816406, 55.504539489746094, -99.427734375, -8.740787506103516, -35.040626525878906, -9.801177978515625, 13.981582641601562, 81.71598815917969, 6.636085510253906, 59.01136016845703, 23.61260986328125, -112.760498046875, 2.1918792724609375, -20.488182067871094, -20.884231567382812, 8.6644287109375, 55.3424072265625, 30.5943603515625, 80.52284240722656, 2.2462615966796875, 35.416778564453125, 6.7593994140625, -13.1137056350708, 1.041046142578125, 154.25604248046875, 82.02995300292969, 0.0, 12.44036865234375, 71.2242660522461, 83.39622497558594, 24.482025146484375, 118.09805297851562, 76.75912475585938, -77.26023864746094, -3.0482311248779297, -0.193206787109375, 4.388092041015625, 64.09996795654297, -58.9482421875, 115.141357421875, 136.69064331054688, -49.8967399597168, 22.0416259765625, 113.73764038085938, 42.589324951171875, 5.646514892578125, -106.67433166503906, 45.12550354003906, -0.444915771484375, -24.919845581054688, 24.103412628173828, 18.953369140625, 68.9939956665039, 34.510955810546875, -7.371124267578125, 39.78515625, 72.90077209472656, 74.75627136230469, 87.5828857421875, -62.44465637207031, 93.09912109375, 1.5315589904785156, 6.712501525878906, -12.06341552734375, 26.815597534179688, 77.55137634277344, -8.751068115234375, 45.53486633300781, 63.786376953125, 41.42048645019531, 89.93264770507812, 23.951980590820312, 13.417396545410156, 82.90084838867188, 84.059814453125, 28.834800720214844, 86.04922485351562, 53.2298583984375, 97.8736572265625, -36.78167724609375, 0.0, 96.56845092773438, 133.248779296875, 5.029701232910156, 37.29158020019531, -36.475341796875, -8.840225219726562, -82.51522827148438, 86.67889404296875, 16.675369262695312, 49.8846435546875, 65.56031799316406, -72.97793579101562, -0.8189907073974609, -1.0582618713378906, -21.799007415771484, 36.066986083984375, 54.045799255371094, -33.40778350830078], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000277.npy"}
|
|
{"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 24.66567611694336, "std": 53.986751556396484, "min": -134.014404296875, "p10": -50.36451873779296, "median": 18.223705291748047, "p90": 91.80171966552733, "max": 156.72964477539062, "pos_frac": 0.71875, "sample": [84.99395751953125, 67.27880859375, -54.11175537109375, 21.43431854248047, 11.78729248046875, 30.168487548828125, 45.381256103515625, 12.2401123046875, 13.207584381103516, 97.88970947265625, 38.6064453125, 83.48046875, 65.59857177734375, 56.47441101074219, 67.56434631347656, 55.6539306640625, -82.1568603515625, -14.574813842773438, 21.987075805664062, -3.812164306640625, 14.165748596191406, 72.92617797851562, -2.9425048828125, -16.738510131835938, 14.969696044921875, 106.37228393554688, 106.4478759765625, 69.72612762451172, 5.4529876708984375, 1.8604736328125, -134.014404296875, 136.75567626953125, -64.08386993408203, 7.2542724609375, 47.51751708984375, 46.26570129394531, 29.459016799926758, -47.10662841796875, 111.98190307617188, 18.20142364501953, 96.97396087646484, 110.766845703125, 52.227020263671875, 12.08929443359375, -2.8957977294921875, 38.689208984375, -1.4390716552734375, 117.56632995605469, 49.376373291015625, 44.56438446044922, 9.318185806274414, 2.93072509765625, 2.935089111328125, -61.259918212890625, -15.389801025390625, -86.8087158203125, 3.4805431365966797, 117.027099609375, -48.75856018066406, -66.68536376953125, 18.245986938476562, 49.82080078125, 21.971710205078125, 6.73291015625, 18.3873291015625, 88.80545043945312, -32.34071350097656, 82.99258422851562, 85.37164306640625, 27.309280395507812, 124.76544189453125, 156.72964477539062, -6.004186630249023, -20.72686767578125, 64.1446533203125, -69.46343994140625, 76.82930755615234, 8.387924194335938, 66.96829223632812, 25.342758178710938, -90.76222229003906, 10.277641296386719, 2.672048568725586, 38.15754699707031, 76.53714752197266, -10.09383773803711, 13.0750732421875, 64.7924575805664, 56.53607177734375, 76.0177001953125, 1.9164581298828125, -4.058502197265625, 89.58504486083984, 131.2960205078125, -9.330291748046875, 89.30972290039062, -57.10784912109375, 20.09710693359375, 28.119247436523438, -2.3925628662109375, -9.783737182617188, 7.1865234375, 22.10333251953125, -68.0636215209961, -7.893829345703125, 19.633377075195312, 97.85054016113281, -14.99658203125, 15.707275390625, 17.014774322509766, 80.69070434570312, 4.204748153686523, -6.173858642578125, 76.11639404296875, 1.2970046997070312, -29.1663818359375, 20.83390235900879, -88.66905212402344, 59.57780075073242, -22.006607055664062, 8.485038757324219, 8.349411010742188, 12.51373291015625, 85.85513305664062, -83.87014770507812, 48.98069763183594, -16.080223083496094, 80.32583618164062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000278.npy"}
|
|
{"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 38.34954833984375, "std": 48.63117218017578, "min": -90.64124298095703, "p10": -10.499561691284178, "median": 30.649627685546875, "p90": 102.87936859130859, "max": 160.18411254882812, "pos_frac": 0.796875, "sample": [35.3505859375, 82.43775939941406, 59.31269836425781, -22.17719268798828, 11.050125122070312, 30.52593994140625, 18.20599365234375, 40.6156005859375, 4.4932098388671875, 84.23440551757812, 48.77642822265625, 120.67889404296875, 34.67230224609375, -4.76055908203125, -4.16070556640625, 29.764007568359375, 70.09475708007812, 74.2297134399414, 32.7056884765625, 35.2435302734375, 14.440994262695312, 30.045989990234375, -5.5657958984375, 30.023101806640625, 24.752227783203125, 24.38372802734375, 14.419551849365234, -9.7734375, -79.00367736816406, 0.32904815673828125, 43.441619873046875, 130.22528076171875, 93.74951171875, 93.43109130859375, 58.296234130859375, -16.136962890625, 3.6221847534179688, 71.15155029296875, -18.744491577148438, 10.966110229492188, 75.83379364013672, 84.01860046386719, 73.8486328125, 110.96739196777344, 8.949745178222656, 101.39006805419922, -3.6633758544921875, 8.070419311523438, -17.836898803710938, 64.29510498046875, 22.8699951171875, 52.46893310546875, 2.095733642578125, 28.13884162902832, 51.32191467285156, -10.096343994140625, 160.18411254882812, -90.64124298095703, 73.58395385742188, 120.55935668945312, -11.44040298461914, 7.10833740234375, 13.71075439453125, 24.8826904296875, -3.577178955078125, -9.15911865234375, -8.691543579101562, 87.16297912597656, -2.1886978149414062, 127.120849609375, -20.374481201171875, 140.78839111328125, 32.962127685546875, 59.50621032714844, 8.154193878173828, -85.26730346679688, 12.3507080078125, 83.0074462890625, 130.94845581054688, -69.52377319335938, 4.138641357421875, 23.752227783203125, 61.94293212890625, 77.93603515625, 51.177093505859375, 61.550811767578125, 25.41644287109375, 4.414951324462891, 103.14608764648438, 11.062583923339844, 71.59849548339844, -18.185104370117188, 82.8118896484375, 37.90362548828125, 22.107555389404297, 12.627593994140625, 102.76506042480469, 2.532651901245117, 38.834930419921875, 107.82551574707031, 61.056427001953125, 99.2812271118164, -7.810150146484375, 86.7384033203125, 113.7989501953125, 94.67132568359375, 11.49725341796875, 10.1531982421875, 11.930328369140625, 102.60050964355469, -8.960479736328125, 30.7733154296875, 45.04443359375, 134.5457763671875, 84.66313171386719, 45.72491455078125, 24.74267578125, 34.643310546875, -64.27180480957031, -16.627288818359375, 115.1148681640625, -3.1790771484375, 75.615966796875, 4.332275390625, 88.80267333984375, 27.07891845703125, 31.64788818359375, 60.586456298828125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000279.npy"}
|
|
{"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 34.51774978637695, "std": 49.956050872802734, "min": -103.85813903808594, "p10": -24.184711456298828, "median": 24.761898040771484, "p90": 105.06800537109375, "max": 136.95977783203125, "pos_frac": 0.7578125, "sample": [77.94775390625, 81.18836975097656, 2.73382568359375, 82.437744140625, -4.469856262207031, -98.18649291992188, 100.87416076660156, 90.66314697265625, 117.74856567382812, 37.63322448730469, -7.3595123291015625, 42.32725524902344, 23.33983612060547, 71.03152465820312, 17.37750244140625, -24.081932067871094, 100.39730834960938, 67.39764404296875, 104.71343994140625, -0.028839111328125, 14.30682373046875, 109.63826751708984, -51.6656494140625, 107.46035766601562, -22.626068115234375, 76.65110778808594, 23.763328552246094, 16.9246826171875, -1.1797027587890625, 86.12557983398438, 0.0, 17.690383911132812, 27.46075439453125, -24.932891845703125, 30.44219970703125, 136.95977783203125, -23.974746704101562, 103.97410583496094, 12.27591323852539, -3.05743408203125, -3.6153564453125, 54.464500427246094, 8.936206817626953, 19.764190673828125, 64.27094268798828, 28.119338989257812, 73.93017578125, -42.1343994140625, 112.382568359375, 16.226806640625, 32.92681884765625, 6.615562438964844, 6.0328369140625, 6.810384750366211, 75.5576171875, 114.610595703125, 37.07366943359375, -17.694976806640625, 4.956939697265625, 83.10433959960938, 93.066162109375, 27.066192626953125, 24.628501892089844, -50.280784606933594, -7.57122802734375, 7.1611480712890625, -50.38909912109375, 105.89532470703125, 7.871574401855469, 62.79359436035156, 94.75717163085938, 39.21405792236328, 115.2803955078125, 108.47760009765625, 117.8768310546875, -19.350830078125, 73.44358825683594, 53.299835205078125, 24.48419189453125, 44.506103515625, 66.49925231933594, 83.91024780273438, -24.424530029296875, 81.07908630371094, 102.97026062011719, -4.903343200683594, 6.9879150390625, 21.082794189453125, -34.0772705078125, 0.51519775390625, 71.06552124023438, 20.2921142578125, 54.2129020690918, -52.34208679199219, 43.52931594848633, 108.60201263427734, -5.3596343994140625, 69.30894470214844, 73.74705505371094, 24.895294189453125, 95.01185607910156, 17.27233123779297, 90.76126098632812, -19.474395751953125, 14.303543090820312, 11.077949523925781, 0.24736785888671875, 5.5947723388671875, 111.46869659423828, 2.384765625, 35.19244384765625, 31.798782348632812, 5.5571136474609375, 65.97514343261719, -31.167572021484375, 74.60379791259766, 84.72415161132812, 127.2845458984375, -12.652679443359375, -103.85813903808594, -12.85443115234375, 8.916095733642578, -31.944000244140625, 11.542179107666016, 31.05596923828125, 1.848785400390625, 63.2967529296875, -41.776763916015625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000280.npy"}
|
|
{"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 33.005699157714844, "std": 47.8970947265625, "min": -89.72198486328125, "p10": -15.370352172851556, "median": 21.5421142578125, "p90": 100.21504211425781, "max": 188.2669677734375, "pos_frac": 0.71875, "sample": [67.81817626953125, 37.43189239501953, -1.7156982421875, -5.741920471191406, 44.171478271484375, 9.343658447265625, 44.58038330078125, 46.3626708984375, 6.88299560546875, 106.982421875, 102.75418090820312, -66.17739868164062, -79.52783203125, 108.38320922851562, 15.882633209228516, 29.82183837890625, 72.72277069091797, 61.13604736328125, 3.2648773193359375, 7.122367858886719, 101.64161682128906, 8.830238342285156, -0.2225189208984375, -5.650604248046875, 55.733154296875, -3.8533477783203125, -37.891082763671875, 100.598388671875, 14.272705078125, 36.158233642578125, 0.0, 34.65606689453125, 1.28094482421875, 66.70956420898438, 17.121173858642578, 88.76365661621094, 35.879241943359375, -29.437042236328125, 107.04014587402344, 54.318878173828125, -89.72198486328125, 29.660614013671875, 94.06365966796875, -12.296945571899414, 0.0, 31.57818603515625, 18.951934814453125, 73.71804809570312, -20.2977294921875, 25.847061157226562, 37.43524169921875, -26.6829833984375, 9.8388671875, 32.67396545410156, 1.760009765625, 119.6351318359375, -7.55517578125, 26.929153442382812, -4.3457183837890625, 50.369293212890625, 12.103271484375, 27.59881591796875, -19.68402099609375, -2.2068328857421875, 74.30499267578125, 57.440582275390625, 19.047706604003906, -8.371749877929688, -1.5799560546875, 5.37896728515625, 16.4312744140625, 95.91996765136719, -4.916015625, 100.05075073242188, 90.41203308105469, 33.585113525390625, 99.982177734375, 6.506988525390625, 31.7847900390625, 41.640663146972656, 19.49169158935547, 68.6064453125, 188.2669677734375, 95.39300537109375, 88.389404296875, -27.271751403808594, -1.0930938720703125, 76.88543701171875, 29.660903930664062, 2.911449432373047, 128.7032470703125, 29.430450439453125, 69.7882080078125, -7.733707427978516, 21.275161743164062, -35.96270751953125, -5.953765869140625, 117.1370849609375, 89.8260498046875, 8.0985107421875, -21.900360107421875, 36.741546630859375, 0.0, 111.40863037109375, 4.4487152099609375, -13.521636962890625, 67.79849243164062, -39.410797119140625, 115.222412109375, 117.64688110351562, 88.53470611572266, 94.99651336669922, 93.94012451171875, -1.81787109375, 94.31866455078125, 11.47235107421875, -10.733108520507812, 46.66168212890625, 13.067939758300781, 20.491134643554688, 21.790771484375, -52.920318603515625, -8.052391052246094, 96.653564453125, 21.2572021484375, 18.106231689453125, 21.29345703125, -3.0521774291992188], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000281.npy"}
|
|
{"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 31.5324649810791, "std": 53.33699035644531, "min": -113.34329223632812, "p10": -28.79866333007812, "median": 25.251022338867188, "p90": 99.52498779296874, "max": 157.4486083984375, "pos_frac": 0.75, "sample": [97.98139953613281, -58.94642639160156, 25.566619873046875, 157.4486083984375, 49.556304931640625, 6.755699157714844, -113.34329223632812, 100.51397705078125, 57.37147521972656, 7.339935302734375, 6.8617095947265625, 63.3189697265625, 23.931884765625, -26.32452392578125, 9.11724853515625, 8.236785888671875, 86.1571044921875, -27.95672607421875, 86.10946655273438, -2.4801788330078125, -23.56396484375, 37.773651123046875, -57.80006790161133, 39.97552490234375, 9.109931945800781, 133.7222900390625, 94.18966674804688, 51.639892578125, -5.72479248046875, 9.212615966796875, -27.2197265625, 15.297531127929688, 89.7677001953125, 45.22520446777344, -4.300384521484375, 40.68757629394531, 66.505126953125, 10.799880981445312, 38.65472412109375, 18.55389404296875, 35.37312316894531, 54.23292541503906, 6.050811767578125, -46.552734375, 95.70654296875, -27.862594604492188, -72.91624450683594, 35.62782287597656, 110.1419677734375, 6.616357803344727, 26.35418701171875, 13.1593017578125, -22.359954833984375, 10.742256164550781, -30.76318359375, 126.922119140625, 33.28996276855469, 13.483993530273438, 2.3241119384765625, -26.427154541015625, 5.57806396484375, 2.190185546875, -13.989654541015625, 83.31082153320312, 103.35623168945312, 73.99258422851562, 71.4252700805664, -9.056411743164062, -3.1337203979492188, 93.9648208618164, -69.15631103515625, -0.615203857421875, 94.6445083618164, 36.66914367675781, 9.3704833984375, 126.5250244140625, 84.67364501953125, -38.30450439453125, 73.8585205078125, -19.868988037109375, -13.308624267578125, 9.622745513916016, 13.7503662109375, 112.80435180664062, 37.631378173828125, -46.159881591796875, 23.584678649902344, 14.144699096679688, 81.7152099609375, 99.10113525390625, 123.72149658203125, 57.276123046875, 11.555633544921875, 53.38311767578125, -80.7178955078125, 27.977035522460938, 75.75759887695312, 105.68421936035156, 7.7085113525390625, -3.2536773681640625, 97.64785766601562, 155.6248779296875, 84.47747802734375, 4.976016998291016, 83.57818603515625, 3.1062469482421875, -94.81417846679688, 19.296173095703125, 39.8802490234375, 27.740894317626953, 95.20916748046875, 26.751358032226562, 45.593902587890625, 55.125999450683594, 6.630043029785156, 98.944091796875, 107.75845336914062, 24.9354248046875, 104.07618713378906, 46.575958251953125, 33.10453796386719, -51.80324172973633, -1.914022445678711, -41.79278564453125, 88.33486938476562, -9.6224365234375, 10.968399047851562, 75.48724365234375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000282.npy"}
|
|
{"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 32.67304992675781, "std": 49.938262939453125, "min": -81.39755249023438, "p10": -31.874554443359372, "median": 27.366260528564453, "p90": 91.98029327392578, "max": 149.77133178710938, "pos_frac": 0.7578125, "sample": [-19.327972412109375, 22.648727416992188, 92.72247314453125, -23.342132568359375, 10.232004165649414, -44.079345703125, 88.22996520996094, -47.970947265625, 48.2655029296875, -1.8192596435546875, 17.52740478515625, 96.779296875, 58.68415069580078, 64.38322448730469, -1.0506668090820312, -46.40887451171875, -11.2490234375, 64.21002960205078, 9.203636169433594, -27.300201416015625, 138.51153564453125, 114.46826171875, 28.92791748046875, -32.61578369140625, 98.36590576171875, 81.95492553710938, 64.41139221191406, 1.353790283203125, 52.84199523925781, 11.083778381347656, 40.46336364746094, 41.343505859375, 28.340072631835938, 81.57969665527344, 38.80913543701172, 72.19902038574219, 1.1778717041015625, -0.15184783935546875, 47.140533447265625, 40.71528625488281, -13.83612060546875, 81.73800659179688, -25.106399536132812, -56.11531448364258, 75.46361541748047, 69.03497314453125, 14.201416015625, 2.801910400390625, -66.75639343261719, 80.04443359375, 64.156982421875, 97.87667846679688, 79.470947265625, 64.6417007446289, 114.85902404785156, 3.1630325317382812, 8.130977630615234, 46.61717224121094, 32.08270263671875, 73.09956359863281, 33.87469482421875, 76.671630859375, 110.85031127929688, 113.86279296875, 90.12849426269531, -4.29998779296875, 89.95138549804688, 14.6795654296875, 27.92554473876953, 14.694229125976562, -46.579734802246094, 84.92257690429688, -60.45074462890625, 79.93374633789062, 26.455215454101562, 22.356964111328125, 91.66221618652344, 76.9283447265625, 0.0, 1.150146484375, -43.62012481689453, 14.566177368164062, 74.32821655273438, -73.52439880371094, 93.23712158203125, 66.5003662109375, 56.049591064453125, 17.772872924804688, -30.898895263671875, 25.10107421875, 6.59375, -1.9892196655273438, 8.43853759765625, -16.495849609375, 26.030303955078125, 90.54052734375, 78.32913208007812, 51.777740478515625, -81.39755249023438, 75.269287109375, 1.3644428253173828, 26.53411865234375, 148.2579345703125, 0.8025989532470703, 29.89501953125, 22.27044677734375, 78.57229614257812, -73.27493286132812, 23.385223388671875, 5.9319915771484375, 111.49258422851562, 17.70355224609375, -0.7317295074462891, 0.6624355316162109, 0.0, 59.3416748046875, 83.17948150634766, 149.77133178710938, 15.089637756347656, 3.9766845703125, 36.90472412109375, 81.21497344970703, 44.939605712890625, 26.806976318359375, -31.556884765625, 75.13290405273438, -57.243804931640625, -26.42632293701172], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000283.npy"}
|
|
{"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 32.73715591430664, "std": 57.58705520629883, "min": -116.69252014160156, "p10": -42.42476119995116, "median": 27.603286743164062, "p90": 106.74989929199218, "max": 166.85955810546875, "pos_frac": 0.75, "sample": [-39.54851531982422, 56.79571533203125, -56.12030792236328, -116.69252014160156, 65.98194885253906, 29.0653076171875, 79.38681030273438, 106.1278076171875, -50.85984802246094, 28.20227813720703, 48.94270324707031, 108.49795532226562, 59.330772399902344, 15.796150207519531, 25.725738525390625, 69.23710632324219, 99.95191955566406, 58.66656494140625, 112.38247680664062, 6.746101379394531, 96.89620971679688, 135.66851806640625, -15.420318603515625, 70.298828125, -6.354055404663086, 12.984085083007812, 20.32464599609375, -48.318572998046875, 10.568061828613281, -36.517547607421875, -56.42543029785156, 30.929412841796875, 38.455780029296875, -23.481727600097656, 8.301406860351562, 32.30206298828125, 80.06112670898438, 52.83296203613281, 58.268760681152344, 5.4140777587890625, -87.62125396728516, 9.707687377929688, 64.41673278808594, 27.789276123046875, -5.205757141113281, -10.970611572265625, -35.86505126953125, 43.725341796875, 12.06689453125, 143.48316955566406, 96.74755859375, -70.57860565185547, 35.89422607421875, 20.951812744140625, 101.59051513671875, 68.37014770507812, 2.763916015625, -8.61468505859375, 108.20144653320312, 54.8046875, 66.22305297851562, 123.50263977050781, 19.9197998046875, -6.9621734619140625, 100.63511657714844, 45.947174072265625, 129.18377685546875, 23.90679931640625, 59.566070556640625, 10.307146072387695, 71.22682189941406, 10.999916076660156, -12.795877456665039, 43.90484619140625, 4.292392730712891, 166.85955810546875, 2.691925048828125, 9.714508056640625, 15.484878540039062, -4.6222991943359375, 124.46247863769531, 101.54324340820312, 73.0812759399414, 114.14495849609375, -22.7486572265625, -101.67703247070312, 10.386152267456055, -14.222930908203125, 39.72178649902344, -39.898841857910156, 73.47242736816406, 24.06463623046875, 75.49354553222656, 90.73279571533203, 0.5069770812988281, -1.0343780517578125, 31.3365478515625, -78.85289001464844, 1.142669677734375, 67.07003784179688, 103.84445190429688, 20.615997314453125, 69.80531311035156, 12.0020751953125, 0.49654388427734375, 91.51739501953125, 14.475570678710938, 20.111053466796875, 21.89959716796875, 81.79595947265625, 95.77156829833984, 148.1103515625, 45.50044250488281, -19.205474853515625, -85.69560241699219, -76.10723876953125, 70.96399688720703, -59.00370788574219, 3.5035572052001953, -18.999755859375, 128.08154296875, 27.41729736328125, 75.87118530273438, 141.20126342773438, 79.31045532226562, -2.839435577392578, -62.856842041015625, 67.99569702148438], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000284.npy"}
|
|
{"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 25.575075149536133, "std": 49.50269317626953, "min": -92.09133911132812, "p10": -36.713462066650386, "median": 20.807228088378906, "p90": 93.14958953857422, "max": 133.6053924560547, "pos_frac": 0.7109375, "sample": [10.928863525390625, 61.0972900390625, 69.53683471679688, 26.51123046875, 86.12351989746094, 99.77938842773438, 87.17990112304688, -81.68060302734375, 85.3887939453125, 34.09193420410156, 27.675018310546875, -13.239578247070312, 10.215980529785156, -65.24726104736328, -25.00054931640625, 31.123336791992188, 16.795623779296875, -46.6671142578125, -18.55206298828125, -7.791290283203125, -6.058349609375, 63.1700439453125, -68.33319854736328, 19.176589965820312, 27.643653869628906, 51.71998596191406, 113.70086669921875, 45.395904541015625, 43.466278076171875, 32.652862548828125, 133.6053924560547, -48.14239501953125, 12.045562744140625, 1.822235107421875, 99.8079833984375, 102.56095886230469, 88.7703857421875, -21.896804809570312, -36.00831604003906, 6.479736328125, -39.7158203125, -27.16845703125, -17.98675537109375, -4.090240478515625, 0.7162380218505859, 58.93072509765625, 29.64996337890625, 52.890411376953125, 0.7030715942382812, 7.0885467529296875, 97.27552795410156, 10.549552917480469, 6.2443695068359375, -22.797386169433594, 106.59103393554688, 2.6459274291992188, -5.246543884277344, -5.407068252563477, 80.89405822753906, 12.502677917480469, 20.235794067382812, 57.5732421875, 88.55009460449219, 49.548431396484375, 39.49774169921875, 26.166748046875, -57.761505126953125, 12.390703201293945, -59.83610534667969, 41.71307373046875, -65.17990112304688, 92.2103271484375, 21.270584106445312, 12.127113342285156, 70.19941711425781, -4.76708984375, 47.764068603515625, 25.238597869873047, 31.508514404296875, 15.248291015625, 62.86077880859375, 85.59164428710938, 10.211166381835938, 0.0695648193359375, -13.648757934570312, 4.89630126953125, 26.446563720703125, 94.81333923339844, 110.027587890625, 79.32252502441406, 72.59211730957031, 86.1063232421875, -2.4215850830078125, -19.574951171875, -21.443572998046875, 22.17138671875, 32.9886474609375, 45.92633056640625, -30.345199584960938, 14.50152587890625, 67.88322448730469, -38.358802795410156, 61.21513366699219, -31.7679443359375, -7.4996337890625, -21.631881713867188, 12.89508056640625, -15.287677764892578, 107.52951049804688, 63.740203857421875, -92.09133911132812, 13.706298828125, 29.590576171875, -29.0733642578125, 3.47625732421875, -85.97210693359375, 88.80986785888672, 3.8440170288085938, 92.43655395507812, 37.8988037109375, 20.3438720703125, 60.10517883300781, 59.37605285644531, 95.3295669555664, -47.346435546875, 85.16656494140625, 115.10073852539062, 105.2830810546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000285.npy"}
|
|
{"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 27.724712371826172, "std": 53.57719039916992, "min": -97.876220703125, "p10": -44.33184204101562, "median": 27.130542755126953, "p90": 97.54082946777343, "max": 148.93536376953125, "pos_frac": 0.671875, "sample": [-9.65606689453125, -19.17742919921875, 46.1134033203125, 24.473342895507812, 13.347991943359375, 28.26824951171875, 13.7841796875, 3.6824798583984375, 9.6007080078125, 55.8367919921875, 47.473411560058594, 16.046630859375, -7.178733825683594, 50.50663757324219, -84.65005493164062, 117.17857360839844, 9.829360961914062, 106.7696533203125, 78.04341125488281, 100.40350341796875, -32.222381591796875, 63.782867431640625, 12.3585205078125, 27.656997680664062, 16.081199645996094, 18.13604736328125, -10.750411987304688, -5.7189178466796875, 50.56132507324219, -6.7545166015625, -97.876220703125, -3.2291107177734375, 91.26168060302734, 86.18754577636719, -22.77349853515625, 34.42955780029297, 14.7044677734375, 53.469024658203125, 7.934810638427734, 66.58198547363281, -10.09521484375, 89.03314208984375, 81.31253814697266, -60.33690643310547, 69.82821655273438, -46.09820556640625, 31.76702880859375, 148.93536376953125, 21.186080932617188, -11.345703125, 97.4632568359375, -89.38311767578125, 99.99496459960938, 70.46818542480469, 29.82769775390625, 93.101318359375, -0.370697021484375, 87.42306518554688, -51.51025390625, 119.60598754882812, -7.071403503417969, 0.0, -5.93212890625, -72.88075256347656, 37.005287170410156, 45.74761962890625, -5.163448333740234, -35.42274475097656, -8.958847045898438, 38.6787109375, -40.238433837890625, 79.38711547851562, 82.88737487792969, 40.831756591796875, -31.956195831298828, 31.125885009765625, 92.26797485351562, 52.902099609375, -49.789920806884766, 17.068119049072266, 46.34521484375, 7.959510803222656, 7.897123336791992, 138.40869140625, 95.34234619140625, 31.077621459960938, 20.186614990234375, -57.8282470703125, 80.62236022949219, -74.46575927734375, 52.829917907714844, 114.22348022460938, 27.59893798828125, 44.743927001953125, 97.72183227539062, 56.0032958984375, 52.58563232421875, 66.15562438964844, 71.7720718383789, 68.43008422851562, 120.034423828125, 20.228912353515625, 102.64007568359375, -6.063648223876953, 49.08038330078125, -38.40875244140625, 119.52879333496094, 44.526519775390625, -18.965789794921875, -31.834178924560547, -47.09104919433594, 91.33203125, -2.662933349609375, -43.5748291015625, -71.9507064819336, 140.18209838867188, -53.087890625, -41.8973388671875, 26.662147521972656, 13.676513671875, 43.58953857421875, 70.12246704101562, -2.476886749267578, 16.640182495117188, 41.48138427734375, -19.885299682617188, 78.71853637695312, 4.7964935302734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000286.npy"}
|
|
{"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 30.367271423339844, "std": 50.32646942138672, "min": -112.02166748046875, "p10": -34.263816833496094, "median": 30.291744232177734, "p90": 91.95387573242188, "max": 134.08770751953125, "pos_frac": 0.7109375, "sample": [76.114990234375, -14.422950744628906, -49.00567626953125, -9.709768295288086, -21.427230834960938, -48.5032958984375, 92.9737548828125, 85.04518127441406, 58.646080017089844, 100.3704833984375, 80.50186157226562, 13.615447998046875, 8.09625244140625, 94.90693664550781, 6.71929931640625, 12.9688720703125, -22.11126708984375, -47.4119873046875, -14.611030578613281, 26.540611267089844, -21.564071655273438, 134.08770751953125, 86.87850952148438, -0.45806884765625, 45.135528564453125, -15.136703491210938, -10.209403991699219, 29.70123291015625, 73.0797119140625, 33.965057373046875, 1.3702774047851562, 91.51678466796875, 70.11248016357422, 69.19088745117188, -7.619659423828125, 4.2488250732421875, 51.66749572753906, 78.3006591796875, -1.1526298522949219, 85.62459564208984, 79.61693572998047, 69.92720031738281, 11.914520263671875, 37.46820068359375, 101.90711975097656, 3.294069290161133, -4.699951171875, 90.49555969238281, 47.76019287109375, 117.01943969726562, 29.88836669921875, 84.8577880859375, 79.69107055664062, 49.36848068237305, 17.969940185546875, 37.5635986328125, 27.81340789794922, -21.7735595703125, 58.23614501953125, -17.189453125, 86.66178131103516, 54.34220886230469, 38.373016357421875, 28.00127410888672, 2.2914676666259766, 54.231353759765625, 67.65756225585938, 73.40509033203125, 56.77471923828125, 54.5404052734375, -89.74188232421875, -3.2260875701904297, 67.74539184570312, 102.52545928955078, 44.8223876953125, 83.6265869140625, 79.39373779296875, 74.78756713867188, 15.375244140625, 68.82109069824219, 119.74261474609375, -46.067352294921875, -70.39462280273438, 7.6231842041015625, -43.180389404296875, 55.43197250366211, -26.00299072265625, 40.32759094238281, -34.1121826171875, 28.209136962890625, 76.76008605957031, 106.06298828125, -112.02166748046875, -11.41363525390625, 39.51686096191406, -14.752716064453125, 49.0189094543457, 6.090084075927734, -82.06527709960938, 30.69512176513672, 4.993408203125, 14.351959228515625, 5.6450042724609375, 38.84342956542969, -26.05078125, -34.61763000488281, 80.84761810302734, -93.13800048828125, 109.54837036132812, -7.830543518066406, 94.384521484375, -4.174690246582031, 85.27622985839844, 43.53291320800781, 99.67218017578125, 7.784881591796875, 13.840179443359375, 72.84938049316406, 57.69708251953125, -0.67108154296875, 19.31956672668457, 6.9027862548828125, -57.23382568359375, -39.891021728515625, 26.445556640625, 99.43716430664062, -3.10595703125, 63.311431884765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000287.npy"}
|
|
{"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 29.317310333251953, "std": 51.28274154663086, "min": -98.41708374023438, "p10": -27.71379241943359, "median": 22.42428970336914, "p90": 92.23759613037109, "max": 153.24749755859375, "pos_frac": 0.734375, "sample": [21.90087890625, 1.8202438354492188, -16.7860107421875, -17.073089599609375, 63.70267868041992, 1.4102783203125, -11.85284423828125, -17.911056518554688, 22.35511016845703, 70.77059936523438, 22.49346923828125, 122.13861083984375, 1.49884033203125, 72.21810913085938, 55.150543212890625, -8.574798583984375, -4.5058746337890625, 44.3056640625, 22.636474609375, 0.0, 20.325958251953125, -9.9569091796875, 28.036102294921875, 8.131824493408203, 50.39087677001953, -7.4254150390625, -27.026519775390625, 4.4539794921875, -44.864105224609375, 16.6689453125, 127.30191040039062, 67.74526977539062, 36.76445770263672, 63.55266571044922, 6.8191680908203125, 0.0, -14.585166931152344, 17.331146240234375, 15.93865966796875, 18.40313720703125, -13.6142578125, 66.54339599609375, -60.628692626953125, -59.562469482421875, 11.260202407836914, 82.92034912109375, 119.38458251953125, 106.26411437988281, -98.41708374023438, 1.4179458618164062, 86.41607666015625, -8.260986328125, 55.0731201171875, 14.271896362304688, 36.42529296875, -85.38894653320312, 0.6306686401367188, 38.2705078125, 83.13499450683594, -71.94198608398438, 153.24749755859375, -58.272178649902344, -18.45672607421875, 65.34089660644531, 134.698974609375, 13.4398193359375, -13.984207153320312, -41.220306396484375, 91.51988220214844, 14.919624328613281, 86.65118408203125, 26.420135498046875, 1.4251251220703125, 54.488739013671875, 10.652587890625, 17.023040771484375, 14.678466796875, 48.39936828613281, -24.64361572265625, 41.243370056152344, 64.76568603515625, 35.14006042480469, -15.791725158691406, -62.1475830078125, -19.45246124267578, -61.15962219238281, 16.39841079711914, 111.48165893554688, 91.08110046386719, 23.7506103515625, 55.228271484375, 4.3574676513671875, 70.50445556640625, -3.9139251708984375, 79.60818481445312, 81.39437866210938, 67.81202697753906, -55.5128173828125, 101.37362670898438, 23.73883056640625, 47.119720458984375, 52.96368408203125, -71.67022705078125, 132.77130126953125, 13.170822143554688, 32.9163818359375, 22.14962387084961, 29.47735595703125, 40.090911865234375, -11.57879638671875, -29.317428588867188, 59.46910095214844, 49.524383544921875, 82.69021606445312, 64.56436157226562, 87.96060180664062, 29.43603515625, 9.166259765625, 18.699310302734375, 87.03019714355469, 31.311119079589844, 149.93563842773438, 56.03376770019531, 122.08523559570312, 93.91226196289062, 4.447479248046875, 114.99639892578125, 79.12924194335938], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000288.npy"}
|
|
{"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 22.02238655090332, "std": 51.748321533203125, "min": -95.40603637695312, "p10": -44.7134765625, "median": 15.819297790527344, "p90": 92.80844268798828, "max": 128.04690551757812, "pos_frac": 0.6328125, "sample": [-71.83074951171875, -14.261810302734375, 13.752685546875, -51.57440948486328, -9.53826904296875, 2.526702880859375, 12.832794189453125, -17.53314208984375, 60.99078369140625, -43.409454345703125, 91.1431884765625, -2.4453964233398438, 1.4238471984863281, 41.723388671875, -10.305984497070312, 52.457427978515625, -27.317672729492188, 98.990234375, 13.084976196289062, -47.756195068359375, 128.04690551757812, 96.18966674804688, 80.54019165039062, 16.074417114257812, -8.2974853515625, -14.251632690429688, -10.920135498046875, 3.45965576171875, 9.502349853515625, 61.31404113769531, -21.091110229492188, 83.088623046875, -3.167083740234375, 20.70965576171875, -0.5404510498046875, 0.0, 6.671058654785156, 55.657257080078125, -17.222457885742188, 13.74798583984375, 18.08733367919922, -9.950439453125, -73.08807373046875, -91.23245239257812, 61.9404296875, -23.995574951171875, 43.1982421875, 76.14019775390625, -2.80975341796875, -1.05743408203125, -1.9858245849609375, 75.76568603515625, -95.40603637695312, -71.07647705078125, 12.0123291015625, 22.843849182128906, -5.562395095825195, -6.3085784912109375, 37.197227478027344, 116.60433959960938, -56.29559326171875, 83.62257385253906, 52.049560546875, 62.57511901855469, 38.8328857421875, 88.74430847167969, -29.698089599609375, -83.72050476074219, -35.12110137939453, 2.650177001953125, 94.68508911132812, -26.348388671875, 6.282745361328125, 78.01417541503906, 22.1766357421875, 17.140914916992188, 16.871322631835938, 42.15423583984375, 50.37603759765625, 8.315338134765625, 54.911590576171875, -19.0203857421875, -82.46627807617188, 100.42880249023438, -16.368988037109375, 64.4666748046875, -8.61541748046875, 91.99711608886719, -26.622238159179688, 98.39933776855469, -12.452606201171875, 112.05828857421875, 86.04536437988281, 9.9332275390625, 26.041046142578125, 46.551727294921875, 78.3819580078125, 13.7645263671875, 61.81615447998047, 45.204193115234375, 97.33724975585938, 49.863731384277344, 105.20635986328125, 72.66224670410156, -19.007415771484375, 83.73741149902344, 18.52496337890625, 21.425704956054688, 67.6953125, 122.27182006835938, -70.77978515625, 81.33770751953125, 92.00416564941406, -23.272720336914062, 95.42713928222656, -49.66900634765625, -74.91278076171875, 29.53716278076172, -27.071861267089844, 89.84972381591797, 29.66485595703125, 29.256103515625, 113.08137512207031, 1.021942138671875, 15.564178466796875, -6.833164215087891, 24.29010009765625, 19.11041259765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000289.npy"}
|
|
{"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 26.736392974853516, "std": 49.07054138183594, "min": -105.20755004882812, "p10": -28.337313842773437, "median": 16.017070770263672, "p90": 95.70802764892578, "max": 135.90023803710938, "pos_frac": 0.6796875, "sample": [104.84823608398438, 11.48236083984375, -4.31304931640625, 76.19013977050781, 51.82611083984375, -105.20755004882812, 0.0616912841796875, 69.44235229492188, -4.7823486328125, -4.8409423828125, 125.792724609375, 87.08343505859375, 39.7305908203125, -12.655204772949219, 90.31814575195312, 57.76673889160156, 14.735206604003906, 33.8787841796875, 99.15518188476562, 79.98176574707031, 88.69229125976562, 0.9962387084960938, 1.2305908203125, 6.7309722900390625, 135.90023803710938, 3.881683349609375, -60.02796936035156, -76.33984375, -14.269195556640625, 0.667266845703125, 108.0994873046875, 3.4124107360839844, -0.20849609375, 45.38922119140625, 91.76947021484375, 116.42706298828125, 95.32322692871094, 46.010581970214844, -1.0809326171875, -52.1734619140625, 76.89734649658203, 39.837310791015625, 0.1856689453125, 83.25131225585938, 36.634552001953125, -1.1192779541015625, 0.8486328125, -15.517608642578125, 32.26526641845703, 53.087982177734375, -11.646331787109375, 3.296905517578125, 33.99884033203125, 35.78601837158203, 16.35413360595703, -26.66339111328125, 12.281059265136719, 67.11448669433594, 9.628326416015625, 14.871780395507812, 8.385284423828125, 14.307762145996094, -6.540672302246094, 80.574462890625, 24.5733642578125, 30.5389404296875, 20.263633728027344, 66.5333251953125, 15.680007934570312, 85.69049072265625, -27.219482421875, 31.0687255859375, -4.557769775390625, -2.6269378662109375, -6.9394683837890625, 113.23101806640625, -8.332969665527344, -41.00439453125, 0.6153564453125, 49.88787841796875, -3.020111083984375, -12.935272216796875, -27.871795654296875, 91.25991821289062, 81.58186340332031, 83.5767822265625, 76.48846435546875, -17.687835693359375, -97.2425537109375, 6.051300048828125, 61.961395263671875, -30.750320434570312, 43.590911865234375, -10.585357666015625, -56.24188232421875, 4.609184265136719, -19.531509399414062, -7.4053955078125, -46.13890075683594, 77.81111145019531, 8.896537780761719, -47.228370666503906, 37.11517333984375, 99.241943359375, 99.1531982421875, 34.868408203125, 96.60589599609375, 98.90545654296875, 102.59356689453125, 16.707130432128906, -40.313690185546875, 18.54205322265625, 78.1796875, 37.36407470703125, 33.29638671875, -31.990875244140625, -19.13951873779297, 14.092681884765625, -29.42352294921875, 82.40570831298828, -8.17657470703125, -3.8035430908203125, 17.076385498046875, 54.38068389892578, -16.273117065429688, 36.8370361328125, 104.12643432617188, 94.25225830078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000290.npy"}
|
|
{"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 38.273136138916016, "std": 52.89626693725586, "min": -116.47409057617188, "p10": -23.012519836425778, "median": 36.57782745361328, "p90": 105.6833251953125, "max": 143.90374755859375, "pos_frac": 0.7421875, "sample": [105.69985961914062, -0.7065658569335938, 132.38265991210938, 9.265594482421875, 133.60107421875, 38.48359680175781, 111.03457641601562, 90.91018676757812, 33.30810546875, -32.4085693359375, 3.341583251953125, 75.59915161132812, 19.091995239257812, -1.561309814453125, 109.640869140625, 9.456985473632812, 105.67623901367188, -13.894317626953125, 71.5889892578125, 21.364477157592773, 71.56423950195312, 41.14524841308594, 74.31399536132812, 23.51654052734375, -7.5217437744140625, 91.70703125, 37.75086975097656, 4.9573974609375, 70.45364379882812, -21.8306884765625, -97.09202575683594, -11.10074234008789, -34.81379699707031, 20.310489654541016, -116.47409057617188, 24.684173583984375, 77.00616455078125, 56.819580078125, 34.42333984375, 10.04681396484375, 107.19146728515625, -25.770126342773438, -54.85651397705078, 42.86638641357422, 103.98649597167969, 22.946319580078125, 1.4778976440429688, 63.007049560546875, 69.250244140625, 22.370407104492188, 84.62371826171875, 56.241310119628906, 65.59820556640625, -18.51763916015625, 109.19384765625, 81.91426086425781, 45.221954345703125, -21.4246826171875, -46.12060546875, 87.33782958984375, 114.56185913085938, 8.3836669921875, -7.1439208984375, -7.05145263671875, 84.58432006835938, -3.64117431640625, 7.2877197265625, 43.574462890625, -45.94654846191406, 101.26443481445312, -26.145233154296875, 18.13348388671875, 20.561939239501953, 143.90374755859375, 47.11933898925781, 26.9407958984375, 48.78325653076172, 88.80673217773438, 100.93768310546875, -5.1984100341796875, 59.602447509765625, 1.8246612548828125, -10.046539306640625, -20.9317626953125, 40.24549865722656, 112.95108032226562, 61.19911575317383, -62.170135498046875, 0.0, -10.324607849121094, -37.63191223144531, 87.96551513671875, 20.501876831054688, -72.8746109008789, 139.21221923828125, 78.93397521972656, 86.93801879882812, 30.177459716796875, 104.19070434570312, -20.197265625, 63.670379638671875, 93.27377319335938, 27.42791748046875, 88.07681274414062, 0.400970458984375, 56.21575927734375, 13.86639404296875, 22.70989990234375, 137.2293701171875, -2.07696533203125, -54.331878662109375, 87.21737670898438, 4.153800964355469, 42.545814514160156, 82.64866638183594, 35.40478515625, 83.829345703125, -0.9800701141357422, 116.517578125, 97.39295959472656, 55.578033447265625, 15.929595947265625, 0.14678955078125, -11.649797439575195, 104.61436462402344, 75.68940734863281, 76.51942443847656, 69.37901306152344], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000291.npy"}
|
|
{"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 31.43490219116211, "std": 49.99909973144531, "min": -136.30615234375, "p10": -29.99264678955078, "median": 26.047744750976562, "p90": 93.90523834228516, "max": 165.19775390625, "pos_frac": 0.7421875, "sample": [21.35406494140625, 3.9986724853515625, 79.4977798461914, -49.741973876953125, 88.2394027709961, 69.56539154052734, 94.70524597167969, 124.93777465820312, 20.277969360351562, 83.71865844726562, 139.663330078125, 67.07955932617188, -24.712646484375, -3.6303939819335938, 44.43377685546875, 24.18072509765625, -45.07734680175781, -8.493673324584961, 11.407867431640625, -3.48748779296875, 22.23406982421875, 18.14928436279297, 28.7196044921875, 57.83502197265625, 64.7232666015625, 49.180450439453125, 74.06269836425781, 22.573638916015625, 21.740333557128906, -3.861661911010742, 63.16534423828125, 75.20301818847656, 1.6740608215332031, 36.57545471191406, 72.53228759765625, 20.13815689086914, 29.738967895507812, -41.4476318359375, -115.45083618164062, 12.717742919921875, 51.08659362792969, 33.95972442626953, 49.090850830078125, 22.802452087402344, -19.073486328125, 7.3642120361328125, 40.57061004638672, -64.86129760742188, -3.1339187622070312, 97.120361328125, -25.4656982421875, 111.03167724609375, -28.941635131835938, -14.212188720703125, -9.0377197265625, 35.54779052734375, 2.1617431640625, -10.685104370117188, 20.81817626953125, 11.00360107421875, 77.55841827392578, 26.19305419921875, 93.5623779296875, -33.76373291015625, -3.092926025390625, 9.90536117553711, 14.398635864257812, 54.819122314453125, 165.19775390625, 45.148681640625, 5.6962890625, 30.52984619140625, 52.155059814453125, 33.26348876953125, 74.67127990722656, 124.74090576171875, 21.26919174194336, 100.90901184082031, 24.453685760498047, -4.257749557495117, -42.297943115234375, -136.30615234375, 121.57827758789062, 62.12004089355469, 92.79108428955078, 65.33706665039062, 70.96539306640625, 25.902435302734375, 5.279485702514648, 96.38406372070312, -32.44500732421875, 5.5194244384765625, -20.720928192138672, 19.37451171875, 43.11228942871094, -34.674713134765625, -4.460075378417969, -23.773910522460938, 21.130678176879883, 100.3819580078125, 63.903167724609375, 69.63838958740234, 13.71551513671875, 15.94511604309082, 24.95482635498047, 65.34375, 50.03633117675781, -0.670745849609375, 6.5801239013671875, 30.029296875, -37.075408935546875, 59.60615539550781, 43.225555419921875, -0.92083740234375, 93.23562622070312, 92.72653198242188, 131.16802978515625, -54.277587890625, 60.023468017578125, 42.6776123046875, 111.37332153320312, -53.13494873046875, -23.876869201660156, 68.89730834960938, 77.62428283691406, 81.59242248535156, 48.6983642578125, 38.805999755859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000292.npy"}
|
|
{"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 32.117252349853516, "std": 50.07157897949219, "min": -91.98068237304688, "p10": -25.02652893066406, "median": 22.44120979309082, "p90": 96.06907653808594, "max": 145.68585205078125, "pos_frac": 0.7265625, "sample": [96.95002746582031, -12.019012451171875, 93.88381958007812, 108.15103149414062, 21.06396484375, -61.09722900390625, 16.114715576171875, 70.160400390625, 4.2889251708984375, 21.880958557128906, -5.9439239501953125, 63.10071563720703, 31.34747314453125, 22.676315307617188, 42.53965759277344, 11.176132202148438, -33.872825622558594, -0.978118896484375, -39.01097106933594, -24.093231201171875, 41.04404067993164, -3.0998382568359375, 107.53427124023438, 87.75558471679688, 12.525665283203125, 0.0, 96.433837890625, 36.69721984863281, 112.61820220947266, 21.201148986816406, 81.89617919921875, -27.2042236328125, -11.660202026367188, 9.798614501953125, 17.1710205078125, -4.1371002197265625, 18.39556884765625, 82.13414001464844, -0.544891357421875, 22.127059936523438, 87.11026763916016, -0.585205078125, 91.91635131835938, 86.39944458007812, 19.99505615234375, -88.89508056640625, 93.06634521484375, 94.31414794921875, 125.58636474609375, 22.206104278564453, 44.15769958496094, 37.517333984375, 83.888427734375, 73.03973388671875, -2.714691162109375, 79.44789123535156, 28.863861083984375, -12.486328125, 6.904052734375, 16.025604248046875, -8.175315856933594, 105.53150939941406, 48.186309814453125, -38.603782653808594, -71.02971649169922, 35.767616271972656, 9.869049072265625, -22.5994873046875, 47.910797119140625, -8.677162170410156, -91.98068237304688, -30.75079345703125, 65.33409118652344, 140.60552978515625, -59.1986083984375, 7.726806640625, -10.204345703125, 82.42227172851562, -17.252090454101562, 33.965911865234375, 20.051910400390625, 13.1617431640625, 38.474151611328125, 12.681007385253906, 95.26318359375, 19.7103271484375, 41.030670166015625, 90.80404663085938, 95.91275024414062, -20.223892211914062, -17.51922607421875, 117.19647216796875, 9.693305969238281, 91.96821594238281, 42.586212158203125, 38.665557861328125, 34.601078033447266, -81.24629211425781, 36.349212646484375, 19.866756439208984, 56.774749755859375, 85.49057006835938, 120.6348876953125, 107.40399169921875, 53.13714599609375, 92.77711486816406, 21.51555633544922, 26.699172973632812, -35.199066162109375, 1.0603561401367188, 3.9237213134765625, -5.781097412109375, -0.71185302734375, 62.5234375, 138.63677978515625, 145.68585205078125, 23.26922607421875, 23.902023315429688, 80.83812713623047, -58.20574951171875, 65.01766967773438, -6.684959411621094, 11.41357421875, 55.30342102050781, 22.06303596496582, 31.246368408203125, 36.312591552734375, 21.3238525390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000293.npy"}
|
|
{"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 28.069772720336914, "std": 50.15009689331055, "min": -87.84457397460938, "p10": -31.492757415771482, "median": 18.15705108642578, "p90": 96.45736923217773, "max": 150.26629638671875, "pos_frac": 0.703125, "sample": [80.32473754882812, 84.565673828125, 86.06283569335938, -5.0406341552734375, -4.962890625, -2.0826187133789062, 0.8756484985351562, 4.058197021484375, 14.270347595214844, 95.96224975585938, 41.3916015625, 91.16354370117188, 80.00167846679688, -12.700729370117188, -7.066070556640625, -0.36296844482421875, 92.80989837646484, 113.31477355957031, 94.84571075439453, 17.894424438476562, 32.685943603515625, 74.9134521484375, 60.372406005859375, -30.49597930908203, 12.075187683105469, 91.6925048828125, 12.3519287109375, -10.221237182617188, 5.2947998046875, 19.63733673095703, 116.53976440429688, 16.290771484375, 7.887298583984375, 66.75531005859375, -13.651336669921875, 18.419677734375, -14.86346435546875, -65.50100708007812, 55.55622863769531, 19.942855834960938, 48.329986572265625, 118.5205078125, -42.077728271484375, -44.778953552246094, 104.86854553222656, 36.988182067871094, 68.41890716552734, -4.831352233886719, -4.327064514160156, 34.83308410644531, 106.63667297363281, -8.992572784423828, 108.73966217041016, 97.6126480102539, 39.418033599853516, 98.21348571777344, 101.55265808105469, -24.282745361328125, 14.938079833984375, 10.906883239746094, 66.5701904296875, 22.84857177734375, 25.7008056640625, 87.48597717285156, -82.34442138671875, 64.37312316894531, 76.70805358886719, 45.603759765625, 19.1119384765625, 6.46466064453125, 9.204681396484375, 121.07536315917969, 73.45297241210938, 0.0, 94.3561782836914, 150.26629638671875, -42.77351379394531, -2.2557525634765625, -7.980995178222656, 32.98236083984375, -7.26848030090332, 81.96145629882812, 6.757434844970703, -87.84457397460938, -15.524810791015625, 47.880523681640625, -28.675048828125, 4.312385559082031, 66.01129150390625, -46.36793518066406, 24.647674560546875, 10.556632995605469, -3.8131027221679688, -86.14990234375, 9.898094177246094, 61.69635772705078, 21.393600463867188, 1.2491683959960938, 10.45660400390625, -9.547897338867188, 16.88232421875, 45.376914978027344, 14.292449951171875, 16.880233764648438, 73.79653930664062, 52.59220886230469, -55.6676025390625, -71.2318115234375, 25.386474609375, -0.54150390625, -44.019371032714844, 63.381134033203125, 64.88253784179688, 10.74200439453125, 71.8515625, 31.8646240234375, 1.7046127319335938, -19.379905700683594, 68.68876647949219, -16.955955505371094, 8.5994873046875, -33.818572998046875, 49.030303955078125, 2.1102676391601562, 100.79220581054688, -87.6973876953125, 78.3008041381836, 100.91312408447266], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000294.npy"}
|
|
{"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 32.6781005859375, "std": 53.5423469543457, "min": -94.50833129882812, "p10": -28.643719482421872, "median": 23.422500610351562, "p90": 101.08257751464843, "max": 172.08203125, "pos_frac": 0.75, "sample": [110.22860717773438, 19.211395263671875, -30.141815185546875, -70.64642333984375, -6.35296630859375, 44.47681427001953, -13.770980834960938, 89.15660858154297, -66.71847534179688, -1.3956451416015625, -13.0838623046875, 25.72895050048828, 101.25663757324219, 17.061187744140625, -10.54547119140625, -10.823318481445312, -83.08683776855469, 68.77789306640625, -70.37301635742188, 26.722274780273438, 53.74864959716797, 133.2418212890625, -16.11627197265625, -94.50833129882812, -12.1343994140625, 74.32657623291016, 15.7523193359375, -26.78192138671875, 61.760772705078125, -1.9576568603515625, -8.3624267578125, 21.290771484375, 42.83953857421875, 63.57000732421875, 78.81529235839844, 9.63739013671875, 76.84262084960938, 12.984649658203125, 51.82012939453125, 56.30279541015625, 136.44337463378906, -13.18881607055664, 58.02783203125, 74.347412109375, 20.41607666015625, 60.83935546875, 7.249473571777344, 126.66441345214844, 95.94882202148438, 80.74739074707031, -79.868408203125, -63.18181610107422, 98.42694091796875, 172.08203125, 98.58314514160156, 141.393798828125, -15.966094970703125, 69.3331527709961, 44.625518798828125, 12.157108306884766, -8.674285888671875, 4.259521484375, 25.356521606445312, 1.6285552978515625, 81.73361206054688, 41.965293884277344, 21.322410583496094, 44.4373779296875, 81.920166015625, 15.564010620117188, 4.169158935546875, 10.828094482421875, 0.3398017883300781, 99.75991821289062, -28.001678466796875, 44.02259826660156, 101.00798034667969, 83.29629516601562, 13.367401123046875, 67.4180908203125, 7.124412536621094, 25.576751708984375, 20.773399353027344, 31.328277587890625, 7.9704742431640625, 25.3043212890625, 0.0, 63.392581939697266, 24.654815673828125, 13.972076416015625, 88.06742858886719, 127.38128662109375, 37.36944580078125, 28.568450927734375, 98.74581909179688, 3.0784339904785156, -58.000091552734375, 115.2532958984375, 37.53031921386719, 102.15789794921875, 72.91548156738281, -13.519187927246094, 11.719894409179688, 18.04974365234375, 22.190185546875, -66.41324615478516, 85.36434173583984, 7.860349655151367, 1.4155426025390625, 50.46681213378906, -23.631175994873047, 93.27949523925781, 109.5614013671875, 100.42377471923828, 2.8592987060546875, 121.84121704101562, 3.389049530029297, 6.607170104980469, 21.00469207763672, -41.02093505859375, -33.987701416015625, 1.02972412109375, 110.97340393066406, -12.272064208984375, 81.38104248046875, 98.15678405761719, 34.92376708984375, -31.5765380859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000295.npy"}
|
|
{"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 44.2476806640625, "std": 53.95270538330078, "min": -112.58993530273438, "p10": -21.554269409179685, "median": 45.753177642822266, "p90": 114.7228012084961, "max": 139.900146484375, "pos_frac": 0.78125, "sample": [80.11334991455078, 34.58517837524414, 21.84149169921875, 80.49161529541016, 85.06086730957031, -13.601242065429688, -7.0078277587890625, -0.7156658172607422, 48.627235412597656, 7.159069061279297, 66.91705322265625, -50.397308349609375, 31.03363037109375, 35.406494140625, 9.142303466796875, 94.02703857421875, 92.11421203613281, 62.44659423828125, 85.01461791992188, -17.28118896484375, 96.81170654296875, 47.0943603515625, -0.1669921875, 86.64503479003906, 52.3955078125, 108.91696166992188, -0.3817424774169922, 15.247509002685547, 107.15423583984375, -32.230255126953125, 15.670684814453125, 34.92254638671875, 118.17889404296875, 1.1170539855957031, -89.54443359375, 121.83175659179688, 44.98095703125, 80.67533874511719, 103.22538757324219, 26.68122100830078, -22.663848876953125, 23.316436767578125, 118.7142333984375, 93.907470703125, 4.9889068603515625, -21.0787353515625, 31.043792724609375, -25.1676025390625, 135.787109375, 18.52435302734375, 121.92996215820312, 49.95310974121094, 15.071380615234375, 106.189697265625, 117.37371826171875, 131.24136352539062, 9.664924621582031, 126.15799713134766, 21.02667236328125, 73.67774963378906, -8.65478515625, -38.346961975097656, 76.68609619140625, -43.377960205078125, 47.059814453125, 58.25807189941406, 26.49127197265625, 108.88427734375, -84.52482604980469, 75.562255859375, 100.99478149414062, 11.88720703125, -39.529998779296875, 87.40258026123047, 61.929656982421875, 31.051849365234375, 127.802001953125, 134.77291870117188, -5.76164436340332, 75.77923583984375, 12.9791259765625, 93.4169921875, 28.945388793945312, -112.58993530273438, 115.57595825195312, 74.37187194824219, 29.982742309570312, -34.897369384765625, 101.80157470703125, -6.897735595703125, 85.88389587402344, 12.151931762695312, 59.54345703125, -37.748809814453125, 51.7554931640625, 0.0, 84.71969604492188, 81.4957275390625, 84.17333984375, 61.1214599609375, 18.645599365234375, 98.30145263671875, 61.933563232421875, 88.93978881835938, 16.528854370117188, 42.313232421875, 3.93011474609375, 8.152069091796875, 32.885009765625, 88.00430297851562, 121.4912109375, 61.99683380126953, 20.876861572265625, -12.334014892578125, -2.4694290161132812, 94.656494140625, 91.17782592773438, -3.495758056640625, 3.25860595703125, 114.35716247558594, 26.259048461914062, 63.87811279296875, 18.1151123046875, 112.70733642578125, 46.52539825439453, 139.900146484375, -95.0546875, -1.7900390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000296.npy"}
|
|
{"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 33.73333740234375, "std": 52.00026321411133, "min": -108.10174560546875, "p10": -21.746096801757812, "median": 26.111740112304688, "p90": 107.98010559082032, "max": 142.34771728515625, "pos_frac": 0.6875, "sample": [126.2088623046875, 9.272285461425781, 92.86296081542969, 107.93280029296875, -12.081100463867188, 0.5555419921875, 96.23095703125, 43.62480163574219, 109.63092041015625, -13.816635131835938, -0.55877685546875, -3.34820556640625, -23.713958740234375, 3.591644287109375, 114.63259887695312, -50.81999588012695, -9.329388618469238, 31.300750732421875, 58.89912414550781, 67.00603485107422, 9.840202331542969, 28.698486328125, -10.154411315917969, 34.4903564453125, -39.586212158203125, -31.954315185546875, -22.584716796875, -108.10174560546875, 15.50836181640625, -11.17868423461914, -36.730613708496094, 75.20487976074219, 19.171295166015625, 57.57701110839844, -11.723968505859375, 4.81243896484375, 105.7000732421875, 8.560760498046875, 142.34771728515625, 10.804412841796875, -3.5564422607421875, 29.935935974121094, 56.207489013671875, -21.386688232421875, 5.446624755859375, -19.353179931640625, -0.930755615234375, 15.980712890625, -79.15252685546875, 89.36505126953125, 17.096282958984375, -20.780975341796875, 2.939910888671875, -29.54034423828125, -0.65087890625, 136.10540771484375, 96.5344009399414, 76.26712036132812, -4.40472412109375, -21.0316162109375, 69.88336944580078, 89.05947875976562, -80.30502319335938, 25.0401611328125, 113.87747192382812, 31.26409912109375, 22.649871826171875, -46.39494323730469, 115.40716552734375, -12.540496826171875, 30.321197509765625, 77.48318481445312, 113.7525405883789, 104.43013000488281, 125.1556396484375, -44.817840576171875, 33.111732482910156, 9.453811645507812, 70.97119140625, -6.727909088134766, -12.259063720703125, 55.35466003417969, 52.609375, -6.491737365722656, 1.5933074951171875, 25.567276000976562, 26.656204223632812, -4.303131103515625, 96.75314331054688, 6.34130859375, 85.92010498046875, 1.523101806640625, 49.089935302734375, 86.42414855957031, 59.6287841796875, 62.06536865234375, 7.5368804931640625, 101.56806945800781, -4.219818115234375, 121.05926513671875, 83.7994384765625, 77.95803833007812, 42.90972900390625, 95.98724365234375, -9.460708618164062, 68.06658935546875, 14.91046142578125, 30.63433837890625, 75.0150375366211, 73.40057373046875, 126.00732421875, 86.85302734375, 10.7296142578125, 26.995895385742188, -1.04339599609375, -21.04681396484375, -3.981842041015625, 27.48919677734375, 88.30162048339844, 27.2705078125, 17.945755004882812, 91.80096435546875, 114.76275634765625, -35.98541259765625, -5.700775146484375, 49.714508056640625, 89.07758331298828, 108.09048461914062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000297.npy"}
|
|
{"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 38.104705810546875, "std": 55.32673645019531, "min": -100.4727783203125, "p10": -29.552261352539062, "median": 32.14582824707031, "p90": 109.70939941406249, "max": 156.37252807617188, "pos_frac": 0.734375, "sample": [89.11466979980469, 92.08221435546875, -13.29632568359375, 27.962692260742188, -29.4444580078125, -45.386077880859375, -32.131683349609375, -11.7611083984375, -21.154327392578125, 118.51260375976562, 38.621490478515625, -8.651519775390625, 8.270751953125, 34.377471923828125, 78.14048767089844, -48.325225830078125, 84.07293701171875, 60.254669189453125, 20.520118713378906, 68.41252136230469, 101.38925170898438, 88.48165893554688, 36.175994873046875, 150.84735107421875, 3.2777862548828125, -28.251739501953125, 93.095703125, -67.75422668457031, 97.71821594238281, 36.57463073730469, 75.14593505859375, -28.9173583984375, 90.27222442626953, 60.55766296386719, 28.87213134765625, -45.108123779296875, 6.983787536621094, 57.4312744140625, 100.7868881225586, -33.45545959472656, 85.59687805175781, -12.04443359375, 87.18572998046875, -21.224212646484375, 35.75091552734375, 118.13626098632812, 37.362762451171875, 20.608901977539062, -56.54637145996094, 4.6977691650390625, 58.433074951171875, 96.7005615234375, 15.5, 106.84939575195312, -55.37921142578125, 34.74000549316406, 86.70359802246094, 106.12216186523438, 8.110870361328125, 77.71051025390625, 130.02413940429688, -100.4727783203125, 89.75643920898438, -11.59893798828125, -19.192840576171875, 148.01556396484375, 111.78713989257812, 43.78802490234375, -29.803802490234375, 0.0, 3.8378448486328125, 5.311210632324219, 138.360595703125, 47.064605712890625, 24.938751220703125, -3.3675880432128906, 83.52558898925781, 24.678932189941406, 5.436866760253906, 87.33917236328125, -22.514785766601562, -10.820955276489258, 11.695384979248047, 65.0809326171875, 86.3646469116211, 7.294780731201172, 108.81893920898438, 93.29373168945312, 94.33198547363281, 4.380462646484375, 156.37252807617188, 33.83636474609375, -39.134620666503906, 0.4661865234375, 10.10894775390625, 29.44403076171875, 54.574615478515625, 14.085662841796875, -6.929832458496094, -3.812255859375, -4.353179931640625, 87.51385498046875, 30.229270935058594, 94.70648193359375, 138.5733642578125, 115.9716796875, -14.8746337890625, 64.85975646972656, 153.04693603515625, 43.282920837402344, 104.6287841796875, 30.455291748046875, 2.716991424560547, 23.774688720703125, 113.38226318359375, 16.2918701171875, -13.6824951171875, 84.280029296875, 36.7042236328125, 125.4193115234375, 3.9150142669677734, 47.670989990234375, 12.764434814453125, 13.0526123046875, -41.35040283203125, -22.26934814453125, 43.71174621582031, -48.71632385253906], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000298.npy"}
|
|
{"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 30.603721618652344, "std": 47.361602783203125, "min": -96.94235229492188, "p10": -24.85395050048828, "median": 20.580341339111328, "p90": 94.25793075561522, "max": 133.29025268554688, "pos_frac": 0.7578125, "sample": [62.88172912597656, 99.90095520019531, 15.85604476928711, 61.127830505371094, 20.7452392578125, 6.119781494140625, 91.8662338256836, -51.967041015625, 70.92535400390625, -96.94235229492188, 116.68966674804688, 45.7386474609375, 20.415443420410156, 78.10324096679688, -24.65557861328125, 98.11123657226562, 3.081939697265625, 19.583152770996094, 11.170989990234375, -90.73143005371094, 92.52165985107422, -3.0660400390625, 60.23779296875, -26.469879150390625, 105.3448486328125, -2.5096588134765625, -46.052734375, -20.664260864257812, 75.62171936035156, 19.149320602416992, 50.76904296875, -5.747772216796875, 104.67668151855469, 1.35845947265625, -61.835235595703125, 33.3751106262207, 84.38006591796875, -7.9290313720703125, -47.4271240234375, 79.9473876953125, 9.95538330078125, 117.14927673339844, 21.14112091064453, 8.955537796020508, -6.942939758300781, -1.41229248046875, 109.43325805664062, -5.150815963745117, 2.0173492431640625, -51.3270263671875, -50.5035400390625, 3.9609222412109375, 4.975744247436523, 17.03643798828125, 76.55975341796875, -31.08013916015625, 103.71212768554688, -4.027557373046875, 119.16339111328125, 30.55762481689453, 69.28580474853516, 7.47564697265625, 12.148696899414062, 47.446929931640625, -5.122779846191406, 133.29025268554688, 6.0234375, 6.229806900024414, 57.583984375, 66.57768249511719, 69.47019958496094, 7.335224151611328, 52.02435302734375, 7.93170166015625, 84.56120300292969, 74.00355529785156, 6.048820495605469, -22.464183807373047, -6.5803070068359375, 92.428955078125, 4.370849609375, 50.249237060546875, 39.05303955078125, 15.73638916015625, 24.003814697265625, -11.069198608398438, 98.88693237304688, 44.326873779296875, 46.11323547363281, 60.665283203125, 9.198394775390625, 63.056610107421875, 91.44670104980469, 0.1094970703125, 10.746814727783203, -6.723297119140625, 31.664077758789062, 1.9728927612304688, 83.4417953491211, -16.06903839111328, -30.097625732421875, 39.91957092285156, 23.7000732421875, -31.97357177734375, 82.82969665527344, 127.50912475585938, 12.736251831054688, 92.60651397705078, 22.3856201171875, 2.437774658203125, 15.355712890625, 1.8086700439453125, 84.0210952758789, -25.316818237304688, 5.26849365234375, 9.18316650390625, -5.178009033203125, 28.638702392578125, 43.43370056152344, 41.444549560546875, -11.097930908203125, 62.249969482421875, 103.07524108886719, 83.4111328125, 36.4654541015625, 88.83798217773438, 23.71246337890625, 65.15444946289062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000299.npy"}
|
|
{"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 31.876262664794922, "std": 53.2100830078125, "min": -89.36904907226562, "p10": -33.41575927734375, "median": 35.497793197631836, "p90": 98.274658203125, "max": 142.30984497070312, "pos_frac": 0.6953125, "sample": [59.676422119140625, 73.85542297363281, 18.494548797607422, -61.66883850097656, 25.54930877685547, -19.826171875, 60.729034423828125, 62.53582000732422, 21.59764289855957, -33.37701416015625, 97.1890869140625, 94.70394897460938, 98.24691772460938, 6.067707061767578, 2.4148330688476562, 100.173095703125, -66.99591064453125, 11.0992431640625, 29.314102172851562, -6.981842041015625, 109.1624755859375, 48.3388671875, 41.588592529296875, 62.12139892578125, 55.31121826171875, 13.350830078125, 35.712677001953125, -3.564544677734375, 13.46087646484375, -20.213584899902344, 76.92910766601562, 77.7974853515625, 97.71033477783203, 50.60906982421875, 84.24771118164062, -33.50616455078125, -83.91238403320312, 17.877532958984375, 4.055139541625977, 142.30984497070312, 51.653717041015625, -17.8372802734375, -29.329681396484375, 3.422607421875, 75.881103515625, 93.30654907226562, -24.936767578125, 93.16822814941406, 80.31980895996094, 123.74655151367188, 22.243148803710938, -72.72288513183594, -77.79354858398438, 41.21354675292969, 35.28290939331055, -6.502872467041016, 43.8944091796875, 41.72344970703125, 97.10757446289062, -9.669830322265625, 92.95182037353516, 88.8111572265625, -40.8472900390625, 117.34811401367188, 42.81219482421875, 4.718994140625, 1.507598876953125, 66.2933349609375, 67.24728393554688, 94.81080627441406, 48.57940673828125, 68.07933044433594, -1.0152702331542969, 12.88751220703125, -17.744232177734375, -74.99723815917969, 122.53486633300781, 52.2735595703125, -44.72589111328125, 100.14480590820312, 98.932861328125, 12.257099151611328, 41.240509033203125, 0.2579803466796875, 59.23729705810547, 19.549407958984375, 0.99224853515625, 0.0, -5.8898162841796875, 24.16522216796875, 119.03842163085938, -11.32257080078125, -89.36904907226562, 40.070465087890625, -39.41943359375, 98.33938598632812, 53.52006530761719, -0.251129150390625, 77.1693115234375, 81.22857666015625, 15.25457763671875, 125.659423828125, 5.3836669921875, 65.59782409667969, -38.650726318359375, 71.79017639160156, 75.5484619140625, -69.601806640625, -29.987945556640625, -15.911346435546875, 75.31320190429688, -2.278188705444336, 36.351043701171875, -22.388992309570312, 72.03564453125, 83.35932922363281, 41.539398193359375, 37.008636474609375, -25.452865600585938, -3.21319580078125, 132.12847900390625, -3.13726806640625, -32.34410095214844, -23.7508544921875, 93.68817138671875, 108.59228515625, -6.9442901611328125, 6.800567626953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000300.npy"}
|
|
{"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 31.463499069213867, "std": 53.577674865722656, "min": -126.1019287109375, "p10": -33.83384399414062, "median": 24.847625732421875, "p90": 102.33712463378905, "max": 165.724609375, "pos_frac": 0.7265625, "sample": [12.606277465820312, 124.51361083984375, 95.97811889648438, 56.75152587890625, 20.259807586669922, 92.25863647460938, 39.51795196533203, -45.77360534667969, 59.787353515625, 7.895591735839844, 4.955928802490234, -2.27679443359375, 87.72555541992188, 15.676605224609375, 4.8759765625, 1.5556182861328125, -51.13243103027344, -37.408782958984375, 90.00312805175781, 122.5181884765625, 15.872589111328125, 18.732627868652344, 52.73114013671875, 21.246368408203125, 26.123931884765625, 17.39898681640625, -0.101593017578125, 48.26806640625, 12.91839599609375, 44.85821533203125, 64.16259765625, -5.324798583984375, 0.0, 122.625732421875, 87.00286865234375, 60.629356384277344, -11.801483154296875, 32.832275390625, -39.61157989501953, 85.35159301757812, -18.27326011657715, 83.67410278320312, 115.54618835449219, 24.601638793945312, -111.7274169921875, 1.631744384765625, -45.45587158203125, 78.6124267578125, -26.780059814453125, 24.25701904296875, 15.769588470458984, -13.31610107421875, -32.301727294921875, 98.49494934082031, 129.904541015625, 44.029014587402344, 28.113510131835938, 111.50411987304688, 3.4615478515625, 165.724609375, 22.558250427246094, 81.38252258300781, 20.769744873046875, 37.72248840332031, -48.631317138671875, -46.679805755615234, 10.1702880859375, -126.1019287109375, -76.22775268554688, 59.17604064941406, 44.66139221191406, 45.25042724609375, -11.353271484375, 15.589262008666992, 37.20569610595703, 24.97198486328125, 105.81817626953125, 10.10137939453125, 109.41390991210938, -49.82061767578125, -4.59490966796875, -17.995208740234375, 127.36647033691406, 82.14395141601562, 89.11395263671875, 24.7232666015625, -5.364938735961914, -28.256454467773438, 83.07666015625, -73.385986328125, 97.09329223632812, 28.4156494140625, -11.484649658203125, 4.5541839599609375, 85.878662109375, -15.77557373046875, 49.86485290527344, 38.60420227050781, 64.67826080322266, 81.86956787109375, 78.31571960449219, 96.32057189941406, 111.95718383789062, -4.6947479248046875, 25.493064880371094, 100.84524536132812, 53.02008056640625, -16.724685668945312, 119.269287109375, -16.732986450195312, 19.630096435546875, 32.14288330078125, -4.052978515625, 21.1171875, 3.2146377563476562, 71.23462677001953, -29.873016357421875, 28.44025230407715, 1.2785224914550781, 94.01814270019531, 119.80743408203125, -56.88177490234375, -16.2237548828125, 44.860015869140625, 28.395339965820312, 44.79441833496094, 73.32872009277344, 6.852142333984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000301.npy"}
|
|
{"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 36.25926208496094, "std": 53.244728088378906, "min": -108.68727111816406, "p10": -21.220734405517565, "median": 36.38495635986328, "p90": 105.3808380126953, "max": 137.7884521484375, "pos_frac": 0.7578125, "sample": [112.28010559082031, 48.675811767578125, 79.86827087402344, 87.49765014648438, 20.215106964111328, 70.97994232177734, 0.0, -0.0336456298828125, -2.5849761962890625, -33.68467712402344, -95.50668334960938, 78.14395141601562, 61.40044403076172, 105.7841796875, -14.92889404296875, 7.519691467285156, 34.69085693359375, 44.86224365234375, -16.822555541992188, -29.243812561035156, 108.01699829101562, 66.54736328125, 124.25225830078125, -12.99151611328125, 84.969970703125, 97.46343994140625, 137.7884521484375, 9.847663879394531, -72.806884765625, 74.41729736328125, 10.849807739257812, 83.73110961914062, 87.21977996826172, 75.24713134765625, 44.21413040161133, 72.79312133789062, 128.32760620117188, 53.31977844238281, 49.3939208984375, 40.09197998046875, 48.266204833984375, 0.0607757568359375, -106.4183349609375, -0.48583984375, 55.574501037597656, 60.01997375488281, 124.93380737304688, 32.29011535644531, 120.15411376953125, 98.148193359375, -8.093475341796875, 78.45005798339844, 7.726844787597656, 3.1401138305664062, 5.588645935058594, -58.041534423828125, 65.85848999023438, 37.91796875, 77.77499389648438, 32.19696044921875, 59.4227294921875, 57.08539581298828, 36.885986328125, 9.046920776367188, 36.86944580078125, 3.221038818359375, 1.7862548828125, 10.362253189086914, -5.314666748046875, -57.180625915527344, 87.58075714111328, 115.42840576171875, -31.087677001953125, -38.89390563964844, 100.99420166015625, 60.06097412109375, -2.0181961059570312, 58.491607666015625, 89.67294311523438, 40.30487060546875, 22.128662109375, 33.50347900390625, 95.25985717773438, -73.72711181640625, 105.20797729492188, -5.550262451171875, -108.68727111816406, 22.37554168701172, -16.863998413085938, -54.447235107421875, 11.39996337890625, -64.919189453125, 1.97662353515625, 23.128875732421875, 7.478116989135742, 41.71063232421875, 66.85530090332031, 105.86859130859375, 103.87030029296875, -17.782272338867188, 4.618560791015625, 14.702545166015625, 35.90046691894531, -14.635787963867188, 75.32898712158203, 0.6411724090576172, 22.714141845703125, 115.7852783203125, 33.83486557006836, 79.96757507324219, -2.5777587890625, 25.63147735595703, 39.202392578125, -12.315750122070312, 85.74984741210938, 34.4112548828125, 132.73516845703125, 2.819011688232422, 88.71591186523438, 11.778793334960938, 19.744674682617188, 130.760986328125, 38.98692321777344, 95.85011291503906, 67.12889099121094, -5.532012939453125, 99.667236328125, -8.80364990234375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000302.npy"}
|
|
{"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 34.14532470703125, "std": 54.90463638305664, "min": -111.52041625976562, "p10": -27.571405029296873, "median": 36.135833740234375, "p90": 97.71784591674805, "max": 158.5516357421875, "pos_frac": 0.7578125, "sample": [13.164825439453125, 21.59349822998047, -97.33749389648438, -79.58583068847656, 80.22775268554688, -14.330024719238281, 72.86495971679688, 58.600006103515625, 91.85894775390625, 65.17572021484375, -29.208740234375, 84.97663879394531, 111.12100219726562, 93.49714660644531, 38.041595458984375, -50.384033203125, -3.89404296875, 79.02275085449219, -8.17074203491211, 113.4913330078125, 69.94595336914062, 28.997406005859375, 82.35460662841797, 32.2833251953125, 41.27423095703125, -26.67010498046875, 46.00349426269531, -8.62994384765625, 119.08242797851562, 8.543746948242188, 96.53173828125, 94.53711700439453, -26.86968994140625, 97.80796813964844, -7.13812255859375, 9.13922119140625, 4.4877777099609375, 42.58770751953125, 0.0, 16.589698791503906, -63.75128936767578, 25.82984161376953, -9.499671936035156, 2.591400146484375, 7.904811859130859, 3.366668701171875, 63.10792541503906, 35.398529052734375, 62.509368896484375, 80.85714721679688, 93.10379028320312, -77.64015197753906, 88.48712158203125, -104.39886474609375, 101.64126586914062, 99.37651062011719, 105.35662841796875, 58.63513946533203, 26.568206787109375, 83.32186889648438, -0.065399169921875, -3.74609375, 16.86407470703125, -9.785400390625, -55.502410888671875, 6.642730712890625, 65.041748046875, 56.937255859375, 81.97994995117188, 43.91704559326172, 124.45623779296875, -41.498382568359375, 22.806365966796875, 2.328887939453125, 93.31075286865234, 80.54450225830078, -2.2523956298828125, 61.754241943359375, 97.6792221069336, 143.4290771484375, 78.96031188964844, -102.81903076171875, 66.6978988647461, -85.42689514160156, 1.9467849731445312, 4.830535888671875, 83.49676513671875, 32.278106689453125, 42.906768798828125, 69.0216064453125, 6.985504150390625, -6.732177734375, 103.74739074707031, 3.454315185546875, 94.45367431640625, 2.468536376953125, 3.83453369140625, 41.264251708984375, -25.73931884765625, 28.06243896484375, -25.677581787109375, -111.52041625976562, 43.3939208984375, -1.1966400146484375, 3.9595108032226562, 72.41132354736328, 98.1241455078125, 60.781158447265625, 96.49209594726562, 72.0514144897461, 38.676605224609375, 46.54595947265625, 36.873138427734375, 91.00677490234375, -1.69195556640625, 158.5516357421875, 48.99261474609375, -65.16867065429688, 74.11541748046875, 12.939697265625, 117.76513671875, 15.657196044921875, 92.01409912109375, 3.3293609619140625, 29.701324462890625, 7.85357666015625, 29.64886474609375, 52.08766174316406], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000303.npy"}
|
|
{"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 30.861858367919922, "std": 51.303836822509766, "min": -98.25357055664062, "p10": -30.701791381835935, "median": 22.44183349609375, "p90": 104.59653091430664, "max": 139.20022583007812, "pos_frac": 0.703125, "sample": [0.0, 27.44024658203125, -10.687347412109375, 34.826812744140625, -21.44671630859375, 0.9791259765625, -98.25357055664062, 62.77984619140625, 81.39950561523438, -14.1715087890625, 65.03191375732422, 84.9569091796875, 55.34812927246094, 53.4112548828125, -59.57575988769531, 34.859619140625, 12.89093017578125, 104.53252410888672, 24.244354248046875, 12.3121337890625, 12.651840209960938, 20.286487579345703, 17.555007934570312, -6.6307830810546875, 80.14276123046875, 93.01966857910156, 82.84344482421875, 14.44281005859375, 21.640411376953125, 3.660125732421875, -57.532135009765625, -1.9029407501220703, 17.007522583007812, 2.1018600463867188, 21.294265747070312, 16.51812744140625, 31.178993225097656, 7.594429016113281, 81.76544189453125, 68.60330200195312, -6.6343536376953125, 66.47285461425781, 82.43263244628906, -30.123809814453125, 122.51055908203125, 23.317764282226562, -2.3561935424804688, 54.93365478515625, -35.924163818359375, 32.7701416015625, 33.52037048339844, -21.442764282226562, 27.723068237304688, -82.77239990234375, 0.0, 105.2425537109375, -49.93634033203125, 54.20891571044922, 4.719024658203125, -10.95623779296875, -34.788299560546875, 18.074920654296875, -58.742523193359375, 72.85943603515625, 10.816986083984375, 25.9857177734375, -28.253997802734375, 42.7374267578125, 133.25457763671875, 86.648681640625, -76.16415405273438, 22.3857421875, 28.700454711914062, -32.0504150390625, -40.940765380859375, 112.74810791015625, 109.20155334472656, -2.8030242919921875, 111.28085327148438, -58.684783935546875, 95.90423583984375, 81.51742553710938, -10.590797424316406, -5.182598114013672, 139.20022583007812, 17.281768798828125, 104.74588012695312, -4.836395263671875, 93.13641357421875, -87.458984375, 13.603408813476562, -0.7145175933837891, 22.4979248046875, 108.29147338867188, 111.81544494628906, 12.311351776123047, -6.758453369140625, 82.70640563964844, 99.74189758300781, 106.0103759765625, 28.9295654296875, 27.14288330078125, 37.18421936035156, 6.6886749267578125, 38.886619567871094, 100.37017822265625, 17.968597412109375, 120.22006225585938, 83.50680541992188, 61.06671142578125, 75.02516174316406, 38.8262939453125, -18.664642333984375, -4.46588134765625, 53.82847595214844, 15.427581787109375, -1.8919677734375, -2.5137252807617188, 76.82392883300781, 23.115066528320312, 20.545623779296875, 78.36577606201172, -0.595672607421875, 94.65240478515625, -16.668548583984375, 99.70709228515625, 18.178314208984375, 118.3450927734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000304.npy"}
|
|
{"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 43.94454574584961, "std": 54.669715881347656, "min": -135.0882568359375, "p10": -10.58779602050781, "median": 39.35625076293945, "p90": 112.0309844970703, "max": 177.45489501953125, "pos_frac": 0.828125, "sample": [7.888755798339844, 20.74753189086914, -4.92877197265625, 147.27713012695312, 23.9246826171875, 1.759490966796875, 75.45449829101562, -37.01521301269531, 84.51315307617188, 22.356529235839844, 41.82184600830078, 26.659225463867188, -2.156890869140625, 33.177764892578125, 16.573486328125, -56.77223205566406, -52.96661376953125, 23.776870727539062, 42.4710693359375, 0.7918891906738281, 48.367462158203125, 99.5819091796875, 3.3411331176757812, 45.62394714355469, 6.5353851318359375, 17.994308471679688, -77.76687622070312, -2.95758056640625, 25.829147338867188, 77.75991821289062, 21.04443359375, 69.2159423828125, 77.48602294921875, 53.5682373046875, 112.1395263671875, 52.82658386230469, 0.17069244384765625, -3.0841293334960938, 96.37873077392578, 2.102691650390625, -95.08621215820312, 47.56074523925781, -3.631732940673828, 74.41797637939453, 110.31268310546875, 94.55845642089844, 43.8939208984375, 113.64886474609375, 99.662841796875, 27.447525024414062, 0.31209754943847656, -79.06024169921875, 125.77658081054688, 116.035400390625, 30.148086547851562, 19.383636474609375, 177.45489501953125, 58.386932373046875, 137.45404052734375, 80.4937744140625, 27.57293701171875, 3.8245582580566406, -2.6844482421875, 111.98446655273438, 36.4080810546875, 74.85902404785156, 113.92443084716797, 81.76626586914062, 87.45733642578125, 91.48358154296875, -12.412628173828125, 83.20606231689453, -17.991256713867188, 32.24729919433594, 90.73069763183594, -135.0882568359375, 24.0975341796875, 72.52119445800781, 2.102874755859375, 77.33670043945312, 55.02197265625, 108.0721435546875, 119.30351257324219, 8.084701538085938, 46.310821533203125, 79.01947021484375, -25.356475830078125, 134.62545776367188, 2.010650634765625, -9.80572509765625, 110.7623291015625, 71.025634765625, 44.728729248046875, 36.97959899902344, 32.745147705078125, 85.17440795898438, 49.742462158203125, 109.84844970703125, 132.92013549804688, 41.73290252685547, 29.905242919921875, -2.8427734375, 88.53411865234375, 84.51614379882812, 137.70236206054688, 17.821746826171875, 28.19024658203125, 23.80279541015625, 19.455490112304688, 26.9354248046875, 101.31124877929688, 0.3508434295654297, 54.85063934326172, 3.333984375, 17.637802124023438, -56.92505645751953, 104.90652465820312, 3.3942337036132812, 11.853240966796875, -27.672698974609375, 106.00875854492188, 43.4561767578125, 145.28167724609375, 93.35775756835938, 102.0628662109375, -27.983444213867188, -4.82244873046875, 103.50428771972656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000305.npy"}
|
|
{"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 38.153221130371094, "std": 52.90965270996094, "min": -103.83001708984375, "p10": -16.36453323364258, "median": 32.465003967285156, "p90": 111.12107543945312, "max": 141.6358642578125, "pos_frac": 0.75, "sample": [6.186676025390625, -62.991058349609375, -63.50481414794922, 131.8775634765625, -75.31680297851562, 129.42291259765625, -10.183443069458008, 32.119842529296875, -8.812774658203125, 101.02630615234375, 7.268312454223633, 22.495651245117188, 101.41172790527344, 50.916595458984375, 8.701499938964844, 8.413543701171875, -30.795989990234375, 87.27629089355469, -4.064582824707031, 19.73565673828125, 40.1241455078125, 25.499420166015625, -38.79986572265625, 59.47900390625, -103.83001708984375, 128.0345458984375, 131.6328125, 92.1033935546875, 82.26712036132812, -0.6816368103027344, 52.51177978515625, 58.330718994140625, 18.9833984375, 86.19912719726562, 111.78390502929688, 84.8631362915039, 122.9986572265625, 90.83932495117188, 52.03692626953125, 44.200164794921875, 5.9875640869140625, 82.65164947509766, -7.58770751953125, 8.574142456054688, 41.431365966796875, 21.965606689453125, 32.81016540527344, 64.61691284179688, 54.791900634765625, -63.7474365234375, -9.976287841796875, 100.6475830078125, -16.89697265625, 133.476318359375, 76.86318969726562, -100.91900634765625, 26.643226623535156, 61.722686767578125, -15.168777465820312, 123.85760498046875, -2.7833023071289062, 29.60540771484375, 65.45986938476562, 45.88824462890625, 88.43875122070312, 3.714794158935547, 7.138214111328125, 3.1096763610839844, 71.44400024414062, -25.879165649414062, 70.26126098632812, -38.53045654296875, 80.5238037109375, 22.67877197265625, 134.25299072265625, -0.6824512481689453, -0.3292236328125, 25.130035400390625, 16.811508178710938, 63.89862060546875, 16.912033081054688, 43.3826904296875, 79.79531860351562, 50.73846435546875, 112.6136474609375, -13.302536010742188, -14.079833984375, 50.90753173828125, 55.623138427734375, 141.6358642578125, 55.99932861328125, 40.680328369140625, 1.3595733642578125, -20.05914306640625, 5.208953857421875, 64.92449951171875, 81.95179748535156, -7.5005340576171875, 81.049072265625, -16.13634490966797, 101.2920150756836, 65.59030151367188, 94.26557922363281, 99.23344421386719, 54.43424987792969, 19.8411865234375, 30.68254852294922, 1.5233306884765625, 2.37347412109375, -2.7172393798828125, -8.095478057861328, 5.436767578125, 98.98954772949219, 88.39053344726562, 12.44732666015625, 131.8951416015625, 15.011007308959961, -2.309417724609375, -7.626312255859375, 3.212554931640625, 69.49575805664062, 82.6874771118164, -0.60009765625, -61.705345153808594, 48.658836364746094, 30.996002197265625, 110.83700561523438, 116.01397705078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000306.npy"}
|
|
{"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 40.947059631347656, "std": 52.33745574951172, "min": -127.83920288085938, "p10": -15.990860748291015, "median": 40.473018646240234, "p90": 104.88334655761719, "max": 146.14923095703125, "pos_frac": 0.7734375, "sample": [104.77078247070312, 97.21894836425781, 2.1608123779296875, 112.97308349609375, -48.3179931640625, -5.216585159301758, 60.913490295410156, 71.43601989746094, 47.838134765625, 94.3717041015625, -123.822998046875, -4.008575439453125, 28.485137939453125, 26.156394958496094, 1.2974853515625, 79.53001403808594, 92.37396240234375, 82.41783142089844, 0.47657012939453125, 15.436065673828125, 50.73193359375, -29.5048828125, 62.66560363769531, 9.18528938293457, 125.66644287109375, -21.779083251953125, 40.714111328125, 8.532958984375, 87.28315734863281, 124.669677734375, 73.25555419921875, 21.529510498046875, 23.4735107421875, 47.12639617919922, 48.02720642089844, -12.105621337890625, 97.88233947753906, 57.611785888671875, 12.855133056640625, -0.19969940185546875, 18.9573974609375, 40.23192596435547, 87.3475570678711, -10.50311279296875, 76.52737426757812, 69.10574340820312, 101.76443481445312, -11.718246459960938, -23.446319580078125, 22.260772705078125, 54.46699523925781, 3.8728160858154297, 106.27787780761719, 7.763860702514648, 49.004302978515625, 89.65596008300781, 81.20658874511719, 112.35179138183594, 14.594146728515625, 3.3341445922851562, 15.72900390625, 57.77215576171875, -55.0787353515625, 82.56414794921875, -4.282131195068359, -3.1293983459472656, 92.53489685058594, 54.63642120361328, -5.2994537353515625, 6.388294219970703, -5.58122444152832, 100.80245971679688, -30.827850341796875, -25.86932373046875, 28.16650390625, -11.10675048828125, 26.55499267578125, 96.6143798828125, 80.22821044921875, 19.178558349609375, 105.93281555175781, -17.105369567871094, 38.2372932434082, 118.74835205078125, 123.88046264648438, 94.82992553710938, -11.524131774902344, 85.5975341796875, 12.404436111450195, 76.01565551757812, 19.40239715576172, -33.258514404296875, 66.71694946289062, 130.86129760742188, 71.55555725097656, 10.95770263671875, 88.64468383789062, -96.7591323852539, -15.513214111328125, 89.37432861328125, 3.579986572265625, 132.89791870117188, 27.76922607421875, 108.27273559570312, -127.83920288085938, 12.709442138671875, -0.1419677734375, 77.38107299804688, 27.423431396484375, 67.02993774414062, -23.21820068359375, 27.4197998046875, 146.14923095703125, 75.9372329711914, 16.26739501953125, 69.70294189453125, -1.634033203125, -11.462875366210938, 45.279083251953125, 96.48934936523438, 71.17529296875, 5.226978302001953, 84.47769165039062, 105.14599609375, 81.06527709960938, 10.63693618774414, 99.83123779296875, 77.49012756347656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000307.npy"}
|
|
{"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 29.131240844726562, "std": 52.04854202270508, "min": -109.34994506835938, "p10": -27.30687789916991, "median": 16.204010009765625, "p90": 94.27293472290039, "max": 140.76174926757812, "pos_frac": 0.75, "sample": [105.99562072753906, 52.93975067138672, -36.23565673828125, 100.08280944824219, 79.2215576171875, -65.06257629394531, -40.247833251953125, 41.69354248046875, 7.57843017578125, 22.30520248413086, 6.93499755859375, 80.31365966796875, -1.1658248901367188, -83.36151123046875, -2.903778076171875, 3.236328125, 89.05209350585938, -4.02813720703125, 86.76141357421875, 65.162841796875, 88.10102081298828, 59.228912353515625, 9.798324584960938, -45.16643524169922, 9.712188720703125, 93.13592529296875, -109.34994506835938, -20.560951232910156, 15.75054931640625, 105.51605224609375, 74.46490478515625, 6.432098388671875, 26.5003662109375, 56.82781982421875, -14.2874755859375, 30.168853759765625, 84.16212463378906, 2.1919403076171875, 75.60324096679688, 16.17132568359375, 0.3756256103515625, 33.957313537597656, 138.72052001953125, -1.28839111328125, -23.385894775390625, -3.70556640625, 1.861053466796875, 104.86669921875, -4.915874481201172, 86.15455627441406, 6.174091339111328, 13.397514343261719, 2.4800262451171875, 75.62264251708984, -1.4694137573242188, 25.67449951171875, 93.24432373046875, 12.696052551269531, -85.88720703125, 93.076416015625, 40.564430236816406, 42.050262451171875, -7.19482421875, -12.21539306640625, 140.76174926757812, 88.05979919433594, -3.3354740142822266, -83.02169799804688, 13.389434814453125, 1.1331939697265625, 73.49418640136719, -59.82636260986328, 96.67302703857422, 56.03068542480469, 8.90936279296875, 17.056610107421875, 109.58383178710938, 134.94378662109375, 111.10980224609375, -1.2021522521972656, 79.05768585205078, 6.83795166015625, 1.1751289367675781, 6.052978515625, 60.63494873046875, -100.67620849609375, 14.460472106933594, 7.994903564453125, -19.540557861328125, 73.86198425292969, 16.2366943359375, 9.923042297363281, 35.29156494140625, 8.530181884765625, 84.49064636230469, 79.80779266357422, 104.27532958984375, 97.80384063720703, 28.609634399414062, 63.2452392578125, 30.991226196289062, 63.4913330078125, 135.82098388671875, -23.509803771972656, 24.09180450439453, 12.414543151855469, 14.432861328125, -18.67348289489746, 80.21282958984375, 82.20559692382812, -59.566436767578125, 13.10943603515625, -21.405670166015625, 3.1467742919921875, 3.7512130737304688, 72.41925048828125, -5.3035888671875, 21.008636474609375, 2.096343994140625, 4.747802734375, -36.166717529296875, 70.42604064941406, 49.844482421875, 47.747283935546875, -59.49049758911133, 42.95574951171875, 81.96157836914062, 30.679031372070312], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000308.npy"}
|
|
{"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 40.73405456542969, "std": 54.98590087890625, "min": -85.95144653320312, "p10": -27.08402366638183, "median": 33.406700134277344, "p90": 113.43986053466796, "max": 159.36517333984375, "pos_frac": 0.7421875, "sample": [33.29933166503906, 85.0361328125, 91.98553466796875, 35.548370361328125, 0.835052490234375, -32.98014831542969, -1.30718994140625, -64.53397369384766, 100.51083374023438, 38.94677734375, 123.7672119140625, 2.7675247192382812, -9.842926025390625, 4.9217681884765625, -20.727394104003906, 71.1837387084961, 12.32708740234375, 23.59601593017578, -8.306549072265625, 31.009857177734375, -7.3557586669921875, 79.4676742553711, -2.982025146484375, -60.79066467285156, 10.14434814453125, 68.9947509765625, -0.101837158203125, 159.36517333984375, 69.9783935546875, 33.105072021484375, -1.3196868896484375, 115.31396484375, 83.59370422363281, 3.9615478515625, -1.738037109375, -7.874725341796875, 68.39279174804688, -19.239486694335938, 153.13720703125, -69.25, 102.048095703125, -34.2158203125, -4.608268737792969, 23.241729736328125, 86.60894775390625, -26.27826690673828, 0.0433349609375, -40.349334716796875, -39.210113525390625, 25.590774536132812, 96.92964172363281, -9.079063415527344, 17.695220947265625, -75.84222412109375, 93.86578369140625, 90.99923706054688, 85.27314758300781, 20.73895263671875, 53.8018798828125, 15.451446533203125, 116.318115234375, 0.507568359375, 39.95378875732422, 139.339599609375, -9.040184020996094, 117.5699462890625, 9.126266479492188, 86.4738998413086, 24.44957733154297, 36.667694091796875, 20.137115478515625, 24.027793884277344, 8.230003356933594, 132.906005859375, 2.3843841552734375, 104.85887145996094, 100.50204467773438, 74.03779602050781, 11.347007751464844, 119.7528076171875, 77.48680114746094, 93.95124053955078, 4.1006622314453125, 15.316688537597656, 90.42447662353516, 107.71408081054688, -2.9106292724609375, 34.24150848388672, 29.50054931640625, -28.964122772216797, 83.28240966796875, -60.44432067871094, 49.26786804199219, -16.336591720581055, 112.63667297363281, 60.038116455078125, 17.955230712890625, 95.16302490234375, 52.900604248046875, 8.98916244506836, -63.66375732421875, 69.43331909179688, 94.26873779296875, 120.79522705078125, 108.18411254882812, -1.03155517578125, 142.24630737304688, 26.496337890625, 29.8341064453125, -85.95144653320312, 110.27651977539062, 54.32781982421875, 52.695556640625, 102.74871826171875, 62.7459716796875, 52.062744140625, 47.3477783203125, 124.44522094726562, -17.99017333984375, 103.60334777832031, 120.01898193359375, 76.80408477783203, 77.11006164550781, -32.06573486328125, 72.64813232421875, 33.514068603515625, -3.8321609497070312, 77.4788818359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000309.npy"}
|
|
{"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 33.20099639892578, "std": 50.126060485839844, "min": -78.91558837890625, "p10": -24.86973419189453, "median": 23.55352020263672, "p90": 99.78018493652343, "max": 148.62722778320312, "pos_frac": 0.734375, "sample": [147.97125244140625, 9.606048583984375, 1.4326457977294922, 136.62860107421875, 39.46710205078125, 24.906906127929688, 20.774932861328125, 5.383735656738281, 89.51901245117188, 40.24163818359375, 79.58482360839844, 130.9326171875, 56.39017868041992, 82.6905517578125, 21.270111083984375, 32.717864990234375, 11.86407470703125, -3.772136688232422, 32.44488525390625, -34.50910949707031, -24.425094604492188, 44.49114990234375, 16.462478637695312, 43.6243896484375, 35.087791442871094, -62.4439697265625, 62.87718200683594, -37.6204833984375, 87.90284729003906, -3.950927734375, -6.3473052978515625, -6.948688507080078, -19.71820068359375, 96.68759155273438, -1.822662353515625, 0.2135009765625, 63.9547119140625, 125.861328125, -69.22903442382812, -57.66400146484375, 78.79145812988281, -2.2097015380859375, 12.260711669921875, 18.69622802734375, 37.634490966796875, -9.72119140625, 65.2078857421875, 26.67706298828125, 21.880203247070312, -1.7831268310546875, -7.91900634765625, 32.06745910644531, 109.75830078125, 79.77398681640625, 59.049285888671875, 98.98516845703125, 127.01763916015625, 42.841217041015625, -4.044197082519531, -78.91558837890625, -20.818862915039062, 113.66769409179688, 45.790016174316406, 22.20013427734375, -0.400054931640625, -60.40283966064453, 9.055526733398438, 25.751739501953125, 88.67376708984375, 3.7941131591796875, -12.436851501464844, 7.232452392578125, 13.291152954101562, 27.415618896484375, 63.360382080078125, 71.09353637695312, 88.23931884765625, 124.02374267578125, 53.27764892578125, -2.2069244384765625, -24.17792510986328, 65.58372497558594, 77.66708374023438, 12.379226684570312, -23.237701416015625, 86.65396881103516, 85.25260162353516, 4.936065673828125, 81.65872955322266, -3.5833740234375, 7.943084716796875, -18.54351806640625, 82.977294921875, -53.19378662109375, 42.92107009887695, 1.1209030151367188, 8.787094116210938, 70.27903747558594, 58.57470703125, 18.312362670898438, 41.346641540527344, 110.66925048828125, 7.63922119140625, -25.9072265625, 96.86141967773438, -5.114990234375, 5.6888580322265625, -36.755126953125, -50.26007080078125, 63.308013916015625, -27.128807067871094, 101.63522338867188, 0.1402740478515625, 57.642578125, 10.83172607421875, -39.2977294921875, 14.52581787109375, 14.959392547607422, 9.03851318359375, 65.79356384277344, 40.700653076171875, 6.126232147216797, 98.393798828125, 104.63705444335938, 148.62722778320312, 87.27406311035156, 102.35221862792969, 84.50083923339844], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000310.npy"}
|
|
{"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 36.676483154296875, "std": 58.6807746887207, "min": -131.9580078125, "p10": -30.948071289062483, "median": 31.79486083984375, "p90": 108.2218017578125, "max": 167.88870239257812, "pos_frac": 0.7265625, "sample": [6.425621032714844, -5.4459686279296875, 89.72451782226562, 74.91991424560547, 105.096435546875, 16.348270416259766, 19.476486206054688, 0.3209686279296875, 105.01263427734375, 80.45086669921875, 96.53114318847656, 4.368335723876953, -20.774860382080078, -20.245101928710938, 29.834457397460938, 2.15972900390625, 95.35488891601562, 38.07148742675781, -44.88240051269531, 27.156982421875, 135.48275756835938, 101.34752655029297, 4.2833099365234375, 80.7203369140625, -5.5074005126953125, 0.568206787109375, 112.36215209960938, 90.45791625976562, 7.381320953369141, -1.4300537109375, -60.80463409423828, -131.9580078125, -104.46554565429688, 88.90992736816406, -17.909149169921875, 113.276611328125, -87.3458251953125, -43.259765625, 93.72166442871094, 41.200164794921875, -50.902130126953125, 55.53217697143555, 47.16987609863281, 7.04119873046875, -9.943758010864258, 42.209320068359375, -13.213798522949219, 74.31414794921875, -69.44770812988281, 16.987754821777344, 65.41748809814453, 26.313720703125, -1.250518798828125, 28.20623779296875, 82.86961364746094, -10.598983764648438, 55.0386848449707, 8.390869140625, -19.7430419921875, 108.94427490234375, 118.79925537109375, 100.2437973022461, 103.06561279296875, 81.35306549072266, 89.5416259765625, -54.94476318359375, 30.98529052734375, -0.38287353515625, 29.281082153320312, 36.115455627441406, -69.94688415527344, -9.696460723876953, 118.99664306640625, 67.679931640625, 17.716567993164062, 32.60443115234375, 103.51776123046875, 68.11663818359375, 27.967708587646484, 72.55078125, 167.88870239257812, 8.2611083984375, -67.83375549316406, 37.769493103027344, -0.194610595703125, -10.09613037109375, 122.00613403320312, 109.25189208984375, 97.1421127319336, 22.750381469726562, 65.92500305175781, 69.78440856933594, 73.47114562988281, 36.8692626953125, 123.78466796875, -78.73143005371094, 129.00128173828125, -5.511383056640625, -86.95219421386719, 82.7989273071289, 7.665214538574219, 107.91217041015625, -20.495697021484375, 52.239349365234375, 10.949371337890625, -3.9263877868652344, 109.82493591308594, 102.50143432617188, 9.13981819152832, 39.19580078125, 80.37043762207031, 73.53387451171875, 88.73858642578125, 27.102386474609375, -25.671630859375, 7.417716979980469, 68.7262191772461, 104.71940612792969, 101.24932861328125, 2.77313232421875, -17.824588775634766, -1.81512451171875, 3.958465576171875, 149.7520751953125, 61.72114944458008, 90.27311706542969, 66.3759765625, -23.037582397460938], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000311.npy"}
|
|
{"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 36.08184051513672, "std": 54.70176315307617, "min": -117.025390625, "p10": -31.548757171630857, "median": 29.871261596679688, "p90": 104.73174743652343, "max": 152.32058715820312, "pos_frac": 0.75, "sample": [61.3367919921875, -3.1070404052734375, -3.721588134765625, 90.48397827148438, 89.66529846191406, -117.025390625, 93.30178833007812, 90.63841247558594, -38.62743377685547, -19.318740844726562, -22.525787353515625, -33.663917541503906, 64.17996215820312, 76.72286987304688, 46.454833984375, 115.15911865234375, 57.31171417236328, 97.08680725097656, -20.548858642578125, -61.97674560546875, 4.96771240234375, 84.94186401367188, 93.70428466796875, 1.583465576171875, 7.345855712890625, 15.479263305664062, 42.63726806640625, 152.32058715820312, -36.655853271484375, -13.379646301269531, 94.35629272460938, -4.796070098876953, 6.631591796875, 117.76800537109375, 41.58428955078125, 26.930465698242188, 55.97825622558594, 0.061908721923828125, 59.14414596557617, 59.271820068359375, 129.583251953125, 105.75611877441406, 25.778114318847656, 84.960205078125, 68.49564361572266, 77.15962219238281, 45.62615966796875, 87.637451171875, 88.84474182128906, 22.98992919921875, 99.10787963867188, 93.43940734863281, 20.378585815429688, 3.80609130859375, -19.06524658203125, 20.097076416015625, 34.522491455078125, 27.1796875, -2.6369171142578125, 115.65341186523438, 6.1439056396484375, -20.27838134765625, 104.47189331054688, -76.53120422363281, 27.520294189453125, -30.963348388671875, 126.71572875976562, 60.406402587890625, -25.825820922851562, 86.0513916015625, 23.017333984375, 97.06817626953125, 23.543066024780273, 0.6441364288330078, 70.89505004882812, 78.41497802734375, 4.388267517089844, 106.35195922851562, -8.6954345703125, 78.02197265625, 88.77978515625, 109.52371215820312, -59.256500244140625, 90.80144500732422, 0.47967529296875, 0.0, 50.71490478515625, -32.914710998535156, -82.29559326171875, -110.86653137207031, 15.896728515625, 54.37109375, 7.761253356933594, 43.87188720703125, 78.85134887695312, 0.9438629150390625, -76.95608520507812, 17.104633331298828, 75.3829345703125, 3.0797195434570312, 32.22222900390625, 20.601455688476562, 86.73220825195312, 72.99603271484375, -9.7786865234375, 64.11820983886719, 83.30902099609375, 2.83477783203125, 126.38320922851562, 27.1629638671875, 105.33807373046875, 14.9554443359375, 40.0382080078125, -3.557819366455078, -37.00993347167969, 60.77825927734375, 56.10107421875, 21.86602783203125, -18.929290771484375, 121.36148071289062, -50.98719787597656, 119.89071655273438, -25.319915771484375, 93.324462890625, 71.75264739990234, 22.033538818359375, 23.383869171142578, -10.77435302734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000312.npy"}
|
|
{"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 34.682308197021484, "std": 59.10423278808594, "min": -116.2630615234375, "p10": -38.35760841369629, "median": 28.300186157226562, "p90": 108.56500396728515, "max": 175.8089599609375, "pos_frac": 0.7265625, "sample": [119.69606018066406, 98.02264404296875, -29.2474365234375, -13.416839599609375, -39.05902099609375, -5.623161315917969, 67.84503173828125, 95.45223999023438, 111.89328002929688, -13.93231201171875, -11.800521850585938, 99.67446899414062, -74.76806640625, 102.29084777832031, 64.3047866821289, 109.82695007324219, 98.36740112304688, 36.746490478515625, -23.026611328125, -21.654632568359375, 9.907882690429688, 16.10174560546875, 3.1066856384277344, 103.7252197265625, 77.47323608398438, -6.821197509765625, -3.9031982421875, 31.796844482421875, 73.50192260742188, 137.2052001953125, 132.58648681640625, 158.9295654296875, 46.67205810546875, 42.5709228515625, -34.90814208984375, 53.34173583984375, 48.807823181152344, 59.090179443359375, 130.6090087890625, 41.811866760253906, -25.139678955078125, 134.56854248046875, 107.39323425292969, 13.916748046875, -65.04881286621094, 27.585037231445312, 17.5343017578125, -45.946258544921875, 80.30131530761719, 95.49407958984375, 38.61235046386719, 57.665809631347656, -80.56564331054688, 11.230321884155273, 9.18377685546875, -3.78887939453125, 105.264404296875, 67.23992919921875, 23.85279083251953, 3.9798583984375, 87.50004577636719, 4.7349090576171875, 88.240966796875, 11.393043518066406, 12.459312438964844, 47.31230163574219, -34.40776062011719, 72.35415649414062, -38.057003021240234, -45.3397216796875, 121.18583679199219, -10.6190185546875, 21.88568878173828, 108.024169921875, 16.66680908203125, -5.21044921875, 18.87261962890625, 104.29426574707031, -63.993770599365234, 73.68595886230469, 11.4661865234375, -80.33209228515625, 72.96823120117188, -35.02360534667969, 34.0330810546875, 92.36479187011719, 51.8118896484375, -13.560211181640625, 13.154154777526855, 6.1012115478515625, 23.40185546875, 175.8089599609375, 80.43057250976562, 52.49114990234375, 89.22935485839844, 104.14552307128906, 20.92498016357422, 121.28131103515625, -30.993682861328125, -105.01144409179688, 48.93084716796875, 50.903053283691406, 7.25483512878418, 92.43341064453125, 29.015335083007812, -52.15394592285156, -0.7995338439941406, 4.2769775390625, 99.8023681640625, 89.10791015625, 14.839157104492188, 55.523887634277344, 26.253387451171875, 2.574554443359375, -0.8895549774169922, -7.994422912597656, 81.14450073242188, 21.572479248046875, -70.97775268554688, -79.45222473144531, 4.982933044433594, -116.2630615234375, 74.09686279296875, 35.508934020996094, 30.40911865234375, 114.28836059570312, 25.587158203125, 117.15469360351562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000313.npy"}
|
|
{"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 35.343257904052734, "std": 55.6815299987793, "min": -115.76173400878906, "p10": -37.221361541748045, "median": 34.30933380126953, "p90": 101.04348754882812, "max": 151.43435668945312, "pos_frac": 0.71875, "sample": [26.327621459960938, 37.76568603515625, 84.04197692871094, 39.167999267578125, -17.677017211914062, 21.971588134765625, -47.28572082519531, 75.04585266113281, -46.06621551513672, -76.31028747558594, -12.728759765625, 48.96617126464844, 94.79977416992188, 97.01689147949219, 118.5286865234375, -48.45556640625, -56.97148132324219, 124.86189270019531, 90.52155303955078, 106.03244018554688, 51.156829833984375, -4.307281494140625, 51.34833908081055, 73.96772766113281, 8.08148193359375, 30.169532775878906, 7.555625915527344, 117.36224365234375, 8.870452880859375, 126.04681396484375, 12.684585571289062, -15.709503173828125, 8.5634765625, -37.78474426269531, 96.406005859375, 51.84637451171875, 42.33172607421875, 96.25466918945312, 30.900390625, 107.2215576171875, -16.235977172851562, 76.0521240234375, -43.487274169921875, 83.83784484863281, 4.138496398925781, 32.36518859863281, 128.65643310546875, 46.000732421875, 19.62847900390625, 6.8941650390625, 93.64715576171875, -0.02413177490234375, 10.954055786132812, 75.98046875, -5.517578125, 63.1646728515625, 25.4295654296875, 18.31085205078125, -0.7083282470703125, -63.64679718017578, 91.94256591796875, 98.22586059570312, 77.41401672363281, 0.0, -10.070068359375, -53.09510803222656, -114.584716796875, 91.43411254882812, 22.39569091796875, -13.01409912109375, -10.116424560546875, 18.40667724609375, 59.75390625, 85.97320556640625, 100.6898193359375, -36.97991180419922, 6.575965881347656, 123.61495971679688, -29.32183837890625, -115.76173400878906, 80.6238784790039, 44.082122802734375, 94.17990112304688, 72.63656616210938, 0.0, 101.86871337890625, -6.442626953125, 75.29583740234375, 95.5850830078125, 28.266448974609375, 24.534080505371094, 85.88690185546875, 90.82870483398438, 32.546600341796875, 52.944244384765625, 102.55972290039062, 51.36254119873047, 23.789718627929688, 36.75384521484375, 11.0257568359375, 8.208778381347656, -23.368515014648438, -3.55133056640625, -35.122711181640625, 45.457489013671875, -3.254119873046875, 74.19368743896484, 83.83119201660156, 46.914886474609375, 81.89481353759766, 122.59130859375, 36.07206726074219, -58.49853515625, 1.7532196044921875, 0.492034912109375, 151.43435668945312, 89.44190979003906, -10.118354797363281, 11.842498779296875, 60.369422912597656, 96.6673583984375, 59.57667541503906, -14.010284423828125, 66.96469116210938, -114.746337890625, 126.5440673828125, 60.71485900878906, -8.12822151184082], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000314.npy"}
|
|
{"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 40.05291748046875, "std": 55.146854400634766, "min": -108.52108764648438, "p10": -19.014492034912106, "median": 26.127220153808594, "p90": 112.36215515136718, "max": 138.77032470703125, "pos_frac": 0.7890625, "sample": [4.01947021484375, 46.31248474121094, 2.7062740325927734, 1.9277191162109375, 36.005794525146484, -30.864501953125, -23.68353271484375, 10.722671508789062, 106.51116943359375, -10.816986083984375, 27.2994384765625, -11.02325439453125, 81.87971496582031, 19.54779052734375, 87.08673095703125, -25.844985961914062, 73.07551574707031, 107.20718383789062, 43.69989013671875, -14.782730102539062, -47.215145111083984, -17.597816467285156, 4.9050445556640625, 1.1234970092773438, 134.33432006835938, 9.04473876953125, 77.62673950195312, 122.97946166992188, 134.01959228515625, 2.0673675537109375, 24.805156707763672, 35.08416748046875, 113.59982299804688, 110.91748046875, 44.2352294921875, 138.77032470703125, 119.87200927734375, 102.1888427734375, 72.50982666015625, -4.84759521484375, 5.96014404296875, -26.94732666015625, 24.54351806640625, 100.18765258789062, -16.012542724609375, 13.730300903320312, 6.28759765625, 7.181976318359375, 16.416259765625, 6.49139404296875, 85.4742431640625, -15.89223861694336, 101.49423217773438, 121.47842407226562, 100.7591552734375, 103.83262634277344, 13.886474609375, -6.657814025878906, 20.138916015625, 44.4696044921875, 9.939987182617188, 1.295013427734375, -7.956974029541016, 42.01640319824219, 95.92892456054688, 99.2318115234375, 56.889678955078125, 76.74665069580078, 25.46405029296875, 135.4920654296875, -57.75323486328125, 26.790390014648438, 74.64080810546875, 26.803192138671875, 100.28741455078125, -33.533203125, -108.52108764648438, 123.43292236328125, 6.177133560180664, 111.00946044921875, 111.83172607421875, 7.30889892578125, -2.1689605712890625, 95.0191650390625, 137.61505126953125, 5.8222503662109375, 31.242462158203125, 82.81251525878906, 18.406951904296875, 131.30328369140625, 15.669754028320312, 16.14910888671875, -99.50640869140625, -22.320068359375, 106.50048828125, 119.60513305664062, 57.78369140625, 15.875377655029297, -84.39659118652344, 103.86083984375, 29.48503875732422, -2.4718017578125, -0.988739013671875, 48.02552032470703, 4.119415283203125, -76.6536865234375, -39.552764892578125, 49.4127197265625, 14.47396469116211, 12.24334716796875, 107.50808715820312, 14.527862548828125, 91.38468933105469, 130.18333435058594, 17.693115234375, 76.11502075195312, 108.82305908203125, 10.259315490722656, -0.493743896484375, 16.674850463867188, 61.29600524902344, -6.4333648681640625, 98.23959350585938, 24.22918701171875, 89.7029800415039, 78.873046875, 49.26628875732422, 51.803260803222656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000315.npy"}
|
|
{"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 41.92375946044922, "std": 57.260494232177734, "min": -92.60012817382812, "p10": -27.314866256713866, "median": 33.22209167480469, "p90": 118.01501922607422, "max": 180.00125122070312, "pos_frac": 0.734375, "sample": [109.65167236328125, 23.26348876953125, 33.12469482421875, 33.75299072265625, -10.82904052734375, -27.680160522460938, -8.21221923828125, 123.60708618164062, 34.44969177246094, 128.93362426757812, 24.5135498046875, -29.250213623046875, 105.01808166503906, 26.466339111328125, -39.41664123535156, 30.399564743041992, 80.32076263427734, -35.718650817871094, 73.56069946289062, 15.520660400390625, 92.43646240234375, 113.12248229980469, -29.1007080078125, 83.4107666015625, 95.59751892089844, 147.6505126953125, 38.82781982421875, 96.57877349853516, 24.971435546875, 138.5918426513672, -3.8856658935546875, 102.18412780761719, -20.37469482421875, 83.8885498046875, 109.73980712890625, 4.414421081542969, 66.65159606933594, 106.18206787109375, 118.12397766113281, 33.319488525390625, 5.42059326171875, 65.36236572265625, 8.62774658203125, 55.180389404296875, 12.693817138671875, 57.801246643066406, -27.532337188720703, 114.1295166015625, 17.28106689453125, 0.194244384765625, 32.79498291015625, 4.600555419921875, -92.60012817382812, -21.46026611328125, -23.087692260742188, 117.96832275390625, 78.6611328125, 65.59153747558594, 6.49041748046875, -1.1923980712890625, -18.228496551513672, 43.07867431640625, 139.68865966796875, 104.23226928710938, 29.99505615234375, 21.94500732421875, -30.69256591796875, 180.00125122070312, 32.70514678955078, 77.63177490234375, 4.8287811279296875, -1.244354248046875, -27.221664428710938, -44.437774658203125, 0.3375244140625, 94.17533111572266, 45.41864013671875, 103.93482971191406, 7.539703369140625, 102.04727172851562, 90.91259002685547, 90.64164733886719, 126.57501220703125, -4.5145721435546875, -4.189321517944336, 43.79302978515625, 38.726165771484375, 137.84405517578125, -24.700706481933594, 103.6024169921875, -76.3240966796875, 57.7537841796875, 0.0, -62.23504638671875, 32.96771240234375, 120.01251220703125, 131.91659545898438, 12.708213806152344, -1.656585693359375, 141.37261962890625, 67.35397338867188, 14.268753051757812, -74.80770874023438, 16.879150390625, -20.09967041015625, 77.20355224609375, 111.95462036132812, -4.2801513671875, 54.95726013183594, 75.76898193359375, 131.6219482421875, 12.73138427734375, 115.9583740234375, -20.838531494140625, 18.226806640625, 51.95172119140625, -12.053688049316406, -90.10838317871094, 29.63104248046875, -3.3782958984375, 57.601593017578125, 70.1190185546875, 84.44126892089844, 36.769195556640625, 81.2073745727539, 10.169143676757812, 89.97779083251953, -8.657913208007812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000316.npy"}
|
|
{"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 37.55297088623047, "std": 53.376216888427734, "min": -100.99488830566406, "p10": -19.587624359130857, "median": 30.182353973388672, "p90": 107.03553314208985, "max": 183.67413330078125, "pos_frac": 0.7265625, "sample": [87.6702880859375, -89.0650634765625, 0.0, 6.403984069824219, 132.880859375, -20.56427001953125, 119.51737976074219, 59.42115783691406, 10.546367645263672, 115.10858154296875, 13.32843017578125, 106.99562072753906, 57.17408752441406, -15.771560668945312, 95.00650024414062, -51.036163330078125, -15.939468383789062, 98.1058349609375, 76.998046875, 10.838554382324219, 49.863525390625, 11.36553955078125, 84.26226806640625, 76.91702270507812, 7.897102355957031, 101.4778060913086, 3.4011459350585938, 0.8098068237304688, -8.27838134765625, 25.59747314453125, -13.370525360107422, -26.61224365234375, 25.35205078125, -6.112548828125, -19.240760803222656, 69.09202575683594, 80.07830810546875, 97.01654052734375, 82.01808166503906, -29.59130859375, -21.486541748046875, 29.219528198242188, 119.88113403320312, -87.1207046508789, -84.7144775390625, 103.58551025390625, 135.119384765625, 96.4647216796875, 14.851776123046875, 0.0, 60.030853271484375, 54.12786865234375, 43.126617431640625, 137.95138549804688, -38.054534912109375, -15.898246765136719, 6.392570495605469, 31.783905029296875, -0.14878082275390625, -1.126800537109375, 15.06201171875, 111.76687622070312, 83.51264953613281, 70.0768051147461, 0.0, 5.46295166015625, 18.940155029296875, 10.994613647460938, 29.99913787841797, 63.926849365234375, 90.76931762695312, -5.6107635498046875, -11.509353637695312, -100.99488830566406, 120.56942749023438, 78.01963806152344, 107.128662109375, -54.86016845703125, 55.97425079345703, -10.193405151367188, 29.616500854492188, 84.89588928222656, 79.49565124511719, -16.539573669433594, 42.58538818359375, 61.4515380859375, 35.788909912109375, -20.39697265625, 80.39401245117188, 103.93065643310547, -1.6310272216796875, 10.280975341796875, 120.04254150390625, 10.651390075683594, 13.048553466796875, -2.640043258666992, 8.03765869140625, 82.93707275390625, 67.3160400390625, 183.67413330078125, 5.1372833251953125, 78.5789794921875, 55.723907470703125, -17.163970947265625, -22.675430297851562, 30.365570068359375, 20.886383056640625, 33.47119140625, 0.5193920135498047, 40.3284912109375, 76.76099395751953, 5.374746322631836, 101.99053955078125, 5.564765930175781, 61.350830078125, 91.05209350585938, 79.67546844482422, 24.786500930786133, 117.21923828125, 33.713687896728516, -3.736757278442383, 41.0386962890625, -13.425018310546875, 111.73652648925781, 31.196609497070312, 91.79734802246094, -3.749614715576172, 83.74053955078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000317.npy"}
|
|
{"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 37.215614318847656, "std": 52.597713470458984, "min": -115.61090087890625, "p10": -21.90780334472656, "median": 37.84385681152344, "p90": 104.19092178344727, "max": 147.23980712890625, "pos_frac": 0.7421875, "sample": [53.69874572753906, 19.927566528320312, 103.9692611694336, 93.02709197998047, 14.042266845703125, -1.036651611328125, 31.121307373046875, 21.57879638671875, 102.99905395507812, 38.36656188964844, 1.152984619140625, -12.281597137451172, 110.2271728515625, 19.610239028930664, 125.03433227539062, 13.237701416015625, 86.25381469726562, -11.757171630859375, 66.73072814941406, 39.48216247558594, -19.8563232421875, -19.878265380859375, 81.74234008789062, 114.75656127929688, -14.653900146484375, 46.10591125488281, 80.8614501953125, 93.65766906738281, -78.77867126464844, 24.13409423828125, 1.70977783203125, -32.07139587402344, 33.42401123046875, 27.027488708496094, 31.493545532226562, -17.272613525390625, 7.529746055603027, 40.664276123046875, 89.69729614257812, 138.33441162109375, 41.534881591796875, 75.35061645507812, 52.55609130859375, 75.45693969726562, -7.8188934326171875, 5.37457275390625, -93.58016967773438, 75.1165771484375, 3.4208984375, 83.56695556640625, -115.61090087890625, 14.11556625366211, 57.02679443359375, 119.63555908203125, 85.74463653564453, 37.32115173339844, -13.054344177246094, 85.50901794433594, 99.9368896484375, -7.73004150390625, 62.30454635620117, 74.17373657226562, 123.2392578125, 34.603790283203125, -54.95155715942383, 40.10723876953125, -23.6956787109375, 3.78924560546875, 107.48455810546875, -3.2083587646484375, 90.74113464355469, 39.21092224121094, 7.03643798828125, 55.13097381591797, 83.66360473632812, 89.34878540039062, 51.49613952636719, 88.41192626953125, 51.7470703125, -8.407257080078125, 32.85838317871094, 30.565460205078125, 29.308990478515625, 100.27357482910156, 53.79740905761719, -29.13409423828125, 29.98980712890625, 78.71319580078125, -40.504600524902344, -9.132003784179688, -2.261383056640625, 41.49357604980469, 28.1844482421875, 71.14048767089844, 51.4329833984375, 40.628265380859375, 63.7718505859375, -23.99749755859375, 84.48458862304688, -14.31964111328125, 82.6797866821289, 147.23980712890625, 118.6834716796875, 30.862030029296875, -41.93310546875, 65.17706298828125, 40.045013427734375, 9.457015991210938, -5.162109375, -33.66339111328125, 109.802734375, 28.85760498046875, 47.51025390625, 117.70127868652344, 26.906932830810547, 103.78909301757812, 43.7365608215332, 104.7081298828125, -21.141571044921875, 71.90985107421875, -84.82177734375, 35.18804931640625, -4.415779113769531, 24.138626098632812, 115.1873779296875, -0.9033126831054688, -115.345703125, 0.0], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000318.npy"}
|
|
{"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 40.303741455078125, "std": 48.359153747558594, "min": -116.58760070800781, "p10": -20.20059280395508, "median": 32.25794219970703, "p90": 106.02826690673828, "max": 149.88619995117188, "pos_frac": 0.8203125, "sample": [73.8214111328125, 3.0700912475585938, 40.61334228515625, 41.5284423828125, 20.99189567565918, 146.26483154296875, 70.094970703125, -41.92681884765625, 61.10838317871094, 20.00592041015625, -21.866783142089844, 14.491537094116211, 55.00541687011719, 114.22064208984375, 72.4952392578125, 108.3892822265625, 3.07440185546875, 0.0, 28.126495361328125, 63.126678466796875, 8.969047546386719, 25.76049041748047, 67.39588928222656, 7.9084014892578125, 14.5989990234375, 87.18844604492188, 6.784576416015625, 26.623733520507812, -27.207122802734375, 67.39028930664062, -26.384735107421875, 9.386505126953125, 58.9576416015625, 10.15557861328125, 109.44855499267578, 27.531463623046875, 7.6405181884765625, 99.49053955078125, 99.74102783203125, 27.826522827148438, 4.414405822753906, 95.64599609375, 65.4134521484375, -71.46966552734375, -0.7450332641601562, 59.59649658203125, 108.89913940429688, 6.2172698974609375, -43.458709716796875, 105.71563720703125, 85.9815673828125, 80.20945739746094, 13.385543823242188, 18.846221923828125, 70.1544189453125, 26.17620849609375, 20.043060302734375, -34.607643127441406, 32.314125061035156, 101.15959167480469, -116.58760070800781, 38.22277069091797, 71.93640899658203, -52.60176086425781, 58.56861877441406, -13.004993438720703, 43.16070556640625, 22.048429489135742, 26.0118408203125, 68.11636352539062, 5.879981994628906, 63.398231506347656, 79.48443603515625, 106.75773620605469, -3.622650146484375, -8.444427490234375, 85.06256103515625, 84.3665771484375, 107.62281036376953, -20.475296020507812, 13.576072692871094, 80.5736083984375, 88.8171157836914, 27.041107177734375, 22.42889404296875, 0.44542694091796875, 88.03903198242188, 149.88619995117188, 63.66778564453125, 9.68597412109375, -10.443450927734375, 53.194305419921875, 46.831756591796875, -9.06689453125, 81.02166748046875, 14.010757446289062, 21.57928466796875, 110.00155639648438, -35.01995849609375, 124.2138671875, 37.39990234375, 10.776016235351562, 113.68536376953125, 20.777191162109375, -41.62672424316406, 9.822418212890625, 78.38922119140625, 50.10240173339844, 103.79081726074219, -5.880195617675781, 35.40374755859375, 32.201759338378906, -20.082862854003906, 95.99964141845703, -14.336090087890625, 79.36199188232422, 80.46136474609375, 127.40380859375, -26.204025268554688, 31.811508178710938, 112.005859375, 8.95782470703125, 58.44126510620117, 5.5191192626953125, 88.68527221679688, 45.52874755859375, 23.617919921875, 104.74794006347656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000319.npy"}
|
|
{"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 38.60157775878906, "std": 59.81398010253906, "min": -108.811279296875, "p10": -38.43306732177734, "median": 36.133758544921875, "p90": 112.1398956298828, "max": 139.942626953125, "pos_frac": 0.7421875, "sample": [72.82841491699219, 42.07123565673828, -45.83563232421875, -100.3347396850586, 123.30911254882812, 110.30401611328125, -48.5743408203125, 4.145751953125, 103.80221557617188, 109.87615966796875, 44.2666015625, 86.37680053710938, 29.4991455078125, 6.69586181640625, 107.14788818359375, 86.6434326171875, -16.40167236328125, -1.5889739990234375, 14.311767578125, 21.396774291992188, 11.203414916992188, 102.44386291503906, 42.359619140625, 20.854476928710938, 2.9446258544921875, -48.10569763183594, 130.01470947265625, 77.77033996582031, 36.163543701171875, 35.36554718017578, 63.50923156738281, 83.7723388671875, 113.54055786132812, 40.66889953613281, 41.677947998046875, 98.27328491210938, -12.41583251953125, 99.25392150878906, 13.9449462890625, -1.2400436401367188, 5.950307846069336, 20.011917114257812, 77.54010009765625, -22.254287719726562, 119.71638488769531, 130.89306640625, 60.65240478515625, -2.698974609375, 62.05694580078125, 60.25068664550781, -29.510467529296875, 16.407073974609375, -77.8975830078125, 0.69818115234375, 110.9012451171875, -9.76617431640625, 18.3427734375, 5.744659423828125, 58.986114501953125, 109.38778686523438, 91.72479248046875, 115.49606323242188, 22.37994384765625, 111.53961181640625, 110.49581909179688, -29.827354431152344, 105.5270767211914, -16.11376953125, 121.17611694335938, 100.13590240478516, 39.46271514892578, -11.85711669921875, 23.50872802734375, 18.657623291015625, -108.811279296875, 53.78727340698242, 9.959991455078125, 111.07794189453125, 10.90093994140625, 94.02085876464844, -86.088134765625, 30.643150329589844, 11.760019302368164, -7.611968994140625, -25.109451293945312, 47.89141082763672, 57.55645751953125, 91.95639038085938, 53.15191650390625, -0.9227523803710938, -57.71772766113281, 36.103973388671875, 133.35986328125, -3.1539306640625, 84.53134155273438, -19.846172332763672, 11.616546630859375, 63.86736297607422, -23.543670654296875, 120.9984130859375, 72.130126953125, -36.19073486328125, 68.74200439453125, 116.28456115722656, 97.20526123046875, 139.942626953125, 96.94369506835938, -43.66517639160156, 64.84721374511719, -83.3641357421875, 27.407424926757812, 77.39215087890625, -93.414306640625, -5.8489532470703125, 122.62286376953125, 28.822525024414062, 103.88143157958984, 30.206100463867188, 87.32377624511719, 74.96165466308594, -17.97011375427246, 14.927490234375, 26.059707641601562, -90.50211334228516, 131.74624633789062, 32.64323425292969, -93.14839172363281, 80.97927856445312], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000320.npy"}
|
|
{"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 32.93797302246094, "std": 53.38371658325195, "min": -108.8106689453125, "p10": -30.1796630859375, "median": 25.812850952148438, "p90": 109.31161499023436, "max": 135.8026123046875, "pos_frac": 0.7109375, "sample": [96.45443725585938, -52.632537841796875, 58.1053466796875, -106.227783203125, 23.16290283203125, 123.79522705078125, 15.27069091796875, -7.914031982421875, -43.70967102050781, 131.56573486328125, 86.22303771972656, 29.441421508789062, -40.363731384277344, 42.385169982910156, 59.93940734863281, 8.091796875, 2.663665771484375, 93.87075805664062, 120.90118408203125, 12.3350830078125, 53.94700622558594, -30.48138427734375, 7.738460540771484, -4.924774169921875, 111.64950561523438, 68.31245422363281, -1.2434234619140625, 13.150299072265625, 69.12571716308594, -0.995391845703125, 88.16590881347656, 63.0098876953125, 62.34907150268555, 11.535842895507812, 32.710723876953125, -41.88165283203125, -30.05035400390625, 40.16984558105469, 7.0074920654296875, 82.58660888671875, 40.88558578491211, 31.10345458984375, -5.6705169677734375, 54.76788330078125, 10.515594482421875, 121.83805847167969, -8.9920654296875, 104.62724304199219, 135.8026123046875, 49.4556884765625, -7.70611572265625, 27.554183959960938, -17.35140609741211, 39.479248046875, -7.690864562988281, 2.8705596923828125, 108.30966186523438, -4.760894775390625, 102.07376098632812, -33.1441650390625, 18.414154052734375, 71.6085433959961, 33.981117248535156, 83.35625457763672, 56.11543273925781, 112.60565185546875, -11.041748046875, 48.64697265625, -38.20196533203125, -5.94696044921875, -7.869049072265625, 49.362884521484375, -22.122833251953125, 120.39337158203125, -108.8106689453125, 83.55645751953125, 119.18035888671875, 133.73275756835938, 23.935699462890625, 78.11279296875, 20.10577392578125, -25.135009765625, 47.4176025390625, 72.54139709472656, 8.91162109375, 98.13595581054688, -24.786087036132812, 114.07276916503906, -69.52635192871094, 93.04208374023438, 23.48095703125, 14.281534194946289, 75.32806396484375, 4.472282409667969, 12.249568939208984, 124.91641235351562, -1.72039794921875, 45.538299560546875, 28.37908935546875, 77.8251953125, 78.25244903564453, -5.288726806640625, 66.57861328125, 16.994171142578125, 2.794403076171875, 70.17881774902344, -2.6418533325195312, 0.1781768798828125, 5.125526428222656, -0.5758209228515625, -6.093658447265625, -40.153594970703125, 8.353918075561523, -17.427169799804688, 29.618919372558594, 133.45010375976562, -100.16943359375, 101.59434509277344, 31.695526123046875, 32.20500183105469, 2.05364990234375, 53.316986083984375, 24.071517944335938, 5.573005676269531, 100.70123291015625, 99.32781982421875, 0.0, -51.40069580078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000321.npy"}
|
|
{"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 41.066307067871094, "std": 54.45328140258789, "min": -133.1024169921875, "p10": -23.563648986816403, "median": 39.455657958984375, "p90": 113.57094421386718, "max": 147.86312866210938, "pos_frac": 0.75, "sample": [-49.317138671875, 9.28533935546875, -47.45115661621094, 66.79834747314453, 35.5233154296875, 10.781291961669922, 39.20977783203125, -32.646484375, 98.21226501464844, 60.62890625, -10.7830810546875, -12.183303833007812, 81.16702270507812, 11.81610107421875, 98.00164031982422, 81.250244140625, 39.7015380859375, -3.832275390625, 82.93801879882812, 63.95465087890625, 44.45269775390625, -13.773223876953125, 129.8057861328125, 2.4154510498046875, 4.617439270019531, 32.567413330078125, -22.104949951171875, 75.3609619140625, 6.456573486328125, 77.37109375, 45.162750244140625, 98.14791870117188, -3.1911544799804688, -17.112457275390625, 92.03048706054688, 28.38238525390625, 20.22516632080078, 104.9019546508789, -52.74821472167969, 88.21111297607422, 100.93182373046875, 83.83055114746094, -40.24237060546875, 17.983642578125, 2.3521728515625, 0.0986328125, 69.87728118896484, -26.283981323242188, 95.85580444335938, -88.58695983886719, 87.67359924316406, 5.7083740234375, 93.16229248046875, 135.81524658203125, -7.8447265625, 96.11674499511719, 67.05213928222656, -21.295761108398438, 37.61259460449219, 84.89659881591797, 121.3687744140625, 14.292388916015625, 138.48251342773438, 147.86312866210938, -22.825042724609375, 83.12519073486328, 61.66831970214844, 37.26629638671875, 80.14492797851562, 13.269790649414062, 131.23678588867188, 113.41165161132812, 80.896728515625, 50.301055908203125, 10.691009521484375, 0.0, -53.28936767578125, 46.037506103515625, 59.312103271484375, -25.287063598632812, 113.942626953125, 64.21646881103516, 96.74267578125, 108.54132080078125, -9.723358154296875, 47.643341064453125, 52.915557861328125, -32.07666015625, -2.449981689453125, 55.65373229980469, 125.30911254882812, -4.905029296875, 8.48052978515625, -32.852783203125, -6.91119384765625, 2.6717967987060547, 49.092559814453125, -0.1495513916015625, -133.1024169921875, 63.07708740234375, 86.02497863769531, 23.95630645751953, 52.831298828125, 116.42756652832031, 76.6915283203125, -5.780189514160156, 2.2845306396484375, 144.17042541503906, 138.900390625, 5.570533752441406, -13.9517822265625, 88.41508483886719, 23.532150268554688, 26.240249633789062, 1.048675537109375, 9.840290069580078, -28.98681640625, 12.650634765625, 24.64263916015625, 71.18760681152344, 80.49368286132812, 134.23233032226562, 77.38395690917969, -1.697998046875, 14.48077392578125, 109.05877685546875, 56.47333526611328, 117.3359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000322.npy"}
|
|
{"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 28.739612579345703, "std": 58.52621841430664, "min": -104.65277099609375, "p10": -37.46557922363281, "median": 19.696393966674805, "p90": 102.4390182495117, "max": 141.83441162109375, "pos_frac": 0.671875, "sample": [7.4307098388671875, -16.75946044921875, -89.36848449707031, -82.9267578125, 89.05011749267578, 90.27882385253906, -19.9359130859375, 47.395538330078125, 4.73394775390625, 133.05307006835938, -30.022735595703125, 88.95217895507812, 22.18145751953125, -18.50079345703125, 9.667984008789062, 74.1123275756836, 97.45414733886719, -7.7698822021484375, 53.323028564453125, -19.8355712890625, 140.49224853515625, 74.63919067382812, 18.778472900390625, 89.8822021484375, 17.32550048828125, 123.32810974121094, 5.9964599609375, 99.57897186279297, -48.89817810058594, 100.70803833007812, -15.998046875, -88.71575927734375, -14.471343994140625, 35.26568603515625, 0.23606300354003906, 71.387451171875, 17.269485473632812, 19.13414764404297, 57.95660400390625, 21.388486862182617, 21.635696411132812, 86.66567993164062, -42.379058837890625, 81.254150390625, 94.10037231445312, -22.86175537109375, -12.494537353515625, -104.65277099609375, 101.21905517578125, 26.435401916503906, 5.484712600708008, 20.25864028930664, -26.566055297851562, 7.837604522705078, 43.73597717285156, 63.53082275390625, 40.443145751953125, 98.6124038696289, 121.27757263183594, -20.347320556640625, 71.07806396484375, 75.3568115234375, 76.76515197753906, -23.293212890625, 119.04425048828125, 17.61505126953125, 89.33596801757812, -21.828155517578125, -35.35980224609375, 74.43328857421875, -2.33563232421875, 4.055469512939453, -23.24481201171875, 6.93365478515625, 88.69744873046875, 41.298553466796875, 129.37728881835938, -14.765060424804688, 122.08755493164062, 3.6961669921875, 87.57523345947266, -89.37687683105469, 20.33983612060547, 23.2764892578125, 15.00408935546875, 129.41195678710938, -51.3994140625, 106.94815063476562, -86.74969482421875, 32.52265548706055, 100.23553466796875, 33.37184143066406, -26.8538818359375, 73.97423553466797, 2.3411865234375, -9.492271423339844, -52.49884796142578, -19.00341033935547, 7.354217529296875, -2.180023193359375, 38.2469482421875, -62.886932373046875, -8.424652099609375, -4.4929962158203125, 27.851470947265625, 62.806884765625, 13.682703018188477, -16.33177947998047, 95.12152099609375, 128.7767333984375, -2.8588790893554688, 39.81459045410156, 105.28559875488281, 98.77427673339844, -89.08924102783203, -99.26956176757812, 11.120590209960938, 18.01153564453125, 69.95608520507812, 27.578292846679688, 16.667213439941406, -5.707672119140625, 65.711181640625, 108.06918334960938, 83.76031494140625, -14.15496826171875, -5.918590545654297, 141.83441162109375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000323.npy"}
|
|
{"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 45.114830017089844, "std": 50.59831619262695, "min": -57.59288787841797, "p10": -12.731085205078124, "median": 32.27495002746582, "p90": 108.74315490722655, "max": 156.0760498046875, "pos_frac": 0.78125, "sample": [40.31394958496094, 122.2061767578125, 17.268707275390625, 92.91007995605469, 23.4769287109375, 84.30517578125, 3.7194671630859375, 5.460023880004883, 102.75973510742188, 23.919639587402344, 26.66253662109375, -1.1229248046875, -34.71882629394531, 121.67074584960938, -2.299835205078125, 99.01986694335938, 87.3615493774414, 14.573921203613281, -14.830718994140625, 0.0, 10.45196533203125, 0.3101654052734375, 89.52001953125, -1.5756454467773438, 2.5870094299316406, 29.873077392578125, 84.29046630859375, -35.178688049316406, 86.79513549804688, -38.00440979003906, 5.513372421264648, 92.40201568603516, 101.87501525878906, 118.25299072265625, 94.21406555175781, 87.01322174072266, 17.39691162109375, 46.079986572265625, 70.28225708007812, 0.0, -3.0444908142089844, 103.63641357421875, 114.13421630859375, 86.16937255859375, 77.91389465332031, 93.01049041748047, 0.0, -3.43023681640625, 73.76007080078125, -6.982940673828125, -41.56501770019531, 30.88318634033203, 54.086944580078125, 20.81964111328125, 75.8355712890625, 62.6663932800293, 6.2414398193359375, 90.91928100585938, 120.44818115234375, 16.5570068359375, -38.30469512939453, 108.01632690429688, 8.517181396484375, 89.58335876464844, 74.39249420166016, -9.110990524291992, 69.53096771240234, 82.797607421875, 20.208099365234375, -51.80352783203125, 116.77415466308594, 46.44098663330078, -57.59288787841797, 92.69827270507812, 27.78081512451172, -1.2401275634765625, 23.126068115234375, 30.833282470703125, 7.8798370361328125, 156.0760498046875, 140.0545654296875, 5.123632431030273, -12.52423095703125, 40.61810302734375, 135.88153076171875, 103.4744873046875, 16.849945068359375, 18.288848876953125, -17.48370361328125, 24.31219482421875, 95.47088623046875, 36.336822509765625, 47.979896545410156, -30.43585205078125, -13.2137451171875, 50.421630859375, 33.66671371459961, 115.2845458984375, 91.142578125, 22.27520751953125, -32.69818878173828, -4.105003356933594, 8.279083251953125, 43.584503173828125, -36.257049560546875, 110.4390869140625, 137.76296997070312, 99.98434448242188, 26.526840209960938, 5.6134033203125, 107.76376342773438, 44.49026107788086, -10.285186767578125, 91.42192077636719, 25.460296630859375, 96.08918762207031, 111.88177490234375, 92.11262512207031, -0.2968921661376953, 107.30790710449219, 94.23538208007812, 1.2293663024902344, 90.75990295410156, 100.08668518066406, 101.26593017578125, 9.809661865234375, 2.0631103515625, 3.2305450439453125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000324.npy"}
|
|
{"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 38.161865234375, "std": 64.09121704101562, "min": -104.962646484375, "p10": -29.774526977539054, "median": 23.671432495117188, "p90": 123.38196716308593, "max": 184.671630859375, "pos_frac": 0.671875, "sample": [63.247283935546875, 102.88516235351562, -68.67523193359375, -16.92919921875, 69.79833984375, 24.10943603515625, 54.320831298828125, 110.15057373046875, 33.71531677246094, -3.8482666015625, 1.1707763671875, 21.790008544921875, 103.48054504394531, 103.27284240722656, -4.423248291015625, 3.88592529296875, 118.75770568847656, 88.6915283203125, 63.99334716796875, 124.04507446289062, 105.64804077148438, 126.99465942382812, 69.28790283203125, -4.091545104980469, 85.89215087890625, 8.361900329589844, 13.896705627441406, 61.71006774902344, 89.09394836425781, 12.679840087890625, -63.5906982421875, 9.102203369140625, 44.44546890258789, 122.94963073730469, 115.08612060546875, 43.408050537109375, 19.117273330688477, 120.86802673339844, 25.358062744140625, -21.90119171142578, 90.22030639648438, -99.72012329101562, 112.19282531738281, 0.26153564453125, -63.41508483886719, -1.25775146484375, 171.03271484375, -88.78810119628906, 135.76548767089844, -6.80145263671875, -2.7453155517578125, 31.9293212890625, 122.6798095703125, 8.84075927734375, 4.6892852783203125, -63.00836944580078, -8.780166625976562, -82.02349090576172, -4.250946044921875, -40.35589599609375, 12.627487182617188, 75.89984130859375, 0.2685699462890625, -1.226409912109375, 65.9771728515625, 43.592567443847656, -13.781272888183594, -77.27731323242188, 150.93157958984375, 19.780609130859375, 68.39529418945312, 21.113189697265625, 184.671630859375, -7.274505615234375, -8.45343017578125, 0.78961181640625, -11.195266723632812, -15.2869873046875, 55.417083740234375, 99.70782470703125, -40.12982940673828, -8.427398681640625, -18.950775146484375, 135.114501953125, -60.90826416015625, 58.23223876953125, 95.29879760742188, 43.0418701171875, 34.03465270996094, -27.807891845703125, -15.256256103515625, 83.33741760253906, 129.95538330078125, -16.887176513671875, 3.1589813232421875, -104.962646484375, -16.432159423828125, 123.0977783203125, 44.65455627441406, 92.75957489013672, 132.15646362304688, 23.233428955078125, 71.01911926269531, 154.146484375, 117.89996337890625, -34.36334228515625, 5.184183120727539, 63.80626678466797, 17.593887329101562, -1.15625, -17.933349609375, 106.68133544921875, 36.19111633300781, 97.90009307861328, 6.024360656738281, 90.30011749267578, 81.48141479492188, 140.84652709960938, 6.656158447265625, -1.7226409912109375, -16.62127685546875, -26.945404052734375, 161.79241943359375, -21.942779541015625, -15.190505981445312, 150.15493774414062, 61.64154052734375, 104.06329345703125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000325.npy"}
|
|
{"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 37.380882263183594, "std": 57.9418830871582, "min": -111.58282470703125, "p10": -35.04396820068359, "median": 36.3743839263916, "p90": 108.47629394531249, "max": 135.37283325195312, "pos_frac": 0.7265625, "sample": [89.91751098632812, 86.04074096679688, 90.91426086425781, 124.78184509277344, 51.84596252441406, 50.335113525390625, -33.35772705078125, -13.74798583984375, 97.43138885498047, -38.97853088378906, 66.8004150390625, -16.926971435546875, 99.40663146972656, 102.03843688964844, 20.153526306152344, 1.3477249145507812, 5.975303649902344, 109.60833740234375, 135.2615203857422, 9.298187255859375, 46.0826416015625, 106.79502868652344, 5.4776611328125, -14.727386474609375, 51.787811279296875, 60.387542724609375, 89.15040588378906, 74.7720947265625, 37.852725982666016, 14.894599914550781, -84.13418579101562, 110.86309814453125, 1.7940673828125, 62.49449157714844, -83.99040222167969, -60.087005615234375, 33.07240295410156, 47.373172760009766, -21.5474853515625, 89.36773681640625, 51.123382568359375, 59.06724548339844, -6.695709228515625, 106.25457763671875, 4.372772216796875, 69.92381286621094, -22.948318481445312, 102.03386688232422, -25.53717041015625, 119.61085510253906, 49.773681640625, 57.847137451171875, 104.81161499023438, 13.392692565917969, 108.1571044921875, -87.39301300048828, -32.177276611328125, 12.81573486328125, 87.00155639648438, -8.983421325683594, 89.12362670898438, 84.94255828857422, 19.6053466796875, 76.787353515625, 0.0, -111.58282470703125, 13.294410705566406, 2.735595703125, 85.09580993652344, -30.582489013671875, 3.5766143798828125, 119.04672241210938, 135.37283325195312, 59.769439697265625, 32.408775329589844, -44.73199462890625, 27.849349975585938, -46.28684997558594, 29.646934509277344, 90.18609619140625, -0.666473388671875, 121.03411865234375, -8.4697265625, 15.246734619140625, 9.410835266113281, 99.34873962402344, -28.096832275390625, -64.0777587890625, 74.21343994140625, 11.73858642578125, 97.46797943115234, 26.01275634765625, 123.6138916015625, -9.725067138671875, 59.27154541015625, 32.0152587890625, 63.937339782714844, 109.2210693359375, 21.0975341796875, 79.26449584960938, 112.57333374023438, 0.0, 95.61309814453125, 34.89604187011719, 51.0072021484375, 95.48675537109375, -17.444183349609375, 105.68931579589844, 69.58766174316406, 0.0, 85.30618286132812, -82.39422607421875, -48.307342529296875, -69.26093292236328, 5.352996826171875, -80.22674560546875, 111.57229614257812, -13.1778564453125, 102.63580322265625, 26.999698638916016, 47.740501403808594, 12.642669677734375, -16.58966064453125, -19.8974609375, 99.74490356445312, 110.67517852783203, 21.26727294921875, 106.86865234375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000326.npy"}
|
|
{"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 42.49372863769531, "std": 59.08407211303711, "min": -112.43112182617188, "p10": -21.11616821289062, "median": 30.738380432128906, "p90": 118.64846496582031, "max": 164.39892578125, "pos_frac": 0.7734375, "sample": [20.12071990966797, 90.93397521972656, 81.66831970214844, 19.88690185546875, 155.59884643554688, -97.3065185546875, 68.9747314453125, 104.60343170166016, 4.8743896484375, 82.06340026855469, -4.001169204711914, 143.478271484375, -6.80780029296875, -98.54827880859375, 35.94719696044922, -6.0311279296875, -46.689453125, 1.9441986083984375, 129.52676391601562, 127.54045104980469, 74.39685821533203, 51.391944885253906, 20.85601806640625, 26.287399291992188, 51.158111572265625, -37.1968994140625, 12.660369873046875, 76.95362854003906, 14.247100830078125, 91.66808319091797, 46.50927734375, 83.94424438476562, -14.591238021850586, -9.217933654785156, 43.88096618652344, 84.10784912109375, 119.54702758789062, 7.1502685546875, 164.39892578125, 74.76234436035156, -23.42889404296875, -11.774585723876953, -1.532562255859375, 84.50935363769531, 27.305381774902344, -17.995391845703125, 115.766845703125, 30.3499755859375, 78.74504089355469, 15.60980224609375, 111.65103149414062, 4.069549560546875, -5.805755615234375, -112.43112182617188, 92.44525909423828, 2.56292724609375, 100.83625030517578, 81.34492492675781, 139.5372314453125, 37.0535888671875, 35.33595275878906, 64.96063232421875, 77.84797668457031, 29.6002197265625, 115.68917846679688, 105.14727783203125, 121.99627685546875, 11.910430908203125, 58.791839599609375, -12.489837646484375, 28.9012451171875, 11.946197509765625, -9.345458984375, -23.53185272216797, 68.03138732910156, 20.266510009765625, 30.831314086914062, 45.962646484375, 132.83352661132812, 8.074493408203125, -20.125, 8.756561279296875, 28.4197998046875, 92.47259521484375, 19.819381713867188, 124.93218994140625, 14.47723388671875, 0.0, 109.52630615234375, 9.559272766113281, 26.015777587890625, 121.4822998046875, 141.88565063476562, 13.085174560546875, 87.71269226074219, 19.606491088867188, 95.44680786132812, 28.393081665039062, 104.56248474121094, 114.89599609375, 70.52166748046875, 3.2422943115234375, -110.09709167480469, -2.5619354248046875, 62.95416259765625, 80.94356536865234, -65.21517944335938, 30.64544677734375, 60.34429931640625, 97.5894775390625, -90.83414459228516, 69.776611328125, 118.26336669921875, 8.29315185546875, -31.4508056640625, 90.35235595703125, -3.190143585205078, 121.7796630859375, 0.23686790466308594, -8.264602661132812, 107.00216674804688, 97.08353424072266, 106.5460205078125, 6.858673095703125, -72.459228515625, 97.85012817382812, 21.58740234375, -31.79541015625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000327.npy"}
|
|
{"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 36.246707916259766, "std": 54.14553451538086, "min": -106.66757202148438, "p10": -19.26215515136719, "median": 24.7254638671875, "p90": 103.76636962890625, "max": 149.9337158203125, "pos_frac": 0.765625, "sample": [13.573684692382812, 62.4742317199707, 113.2623291015625, 23.93695068359375, 94.572998046875, 24.22149658203125, 2.961212158203125, 85.29457092285156, 77.22343444824219, 81.39657592773438, 54.098480224609375, 47.65192413330078, 7.240734100341797, 82.23095703125, 22.410911560058594, -1.3245906829833984, 10.06103515625, -10.872955322265625, 37.0518798828125, 63.00340270996094, 33.47991943359375, 76.25975799560547, 32.34912109375, -18.259593963623047, 139.0958709716797, 37.65277099609375, 4.7597808837890625, 30.95703125, 48.831787109375, 87.06461334228516, -2.2143096923828125, 91.76821899414062, 30.7364501953125, 117.66531372070312, 149.9337158203125, 74.88256072998047, 121.25456237792969, 3.8676681518554688, 124.42254638671875, -4.293115615844727, 7.91729736328125, 107.89814758300781, 77.60739135742188, 4.704833984375, 9.213653564453125, -7.535211563110352, -19.3055419921875, 19.90692138671875, -18.136505126953125, 91.90325927734375, 30.716033935546875, 14.119873046875, -19.243560791015625, 121.3404541015625, 9.4078369140625, 31.421142578125, 9.98516845703125, 8.987007141113281, 14.42156982421875, -31.619346618652344, 13.252510070800781, 22.399978637695312, 92.62362670898438, 107.98046875, 57.95506286621094, 120.1142578125, 88.41913604736328, -61.7391357421875, 93.6275863647461, -65.45806884765625, 12.590988159179688, 77.8843994140625, 90.8255615234375, 88.28404235839844, 16.24835205078125, 98.87823486328125, 135.06387329101562, 11.333343505859375, -14.17742919921875, 90.58872985839844, -106.66757202148438, 6.6555023193359375, 105.7305908203125, 25.22943115234375, -78.54464721679688, 5.598175048828125, 84.57925415039062, 94.93326568603516, 86.20402526855469, -21.064453125, -7.364013671875, 92.21551513671875, 14.291786193847656, -1.49835205078125, 9.09808349609375, -20.291717529296875, -4.4759674072265625, 16.40936279296875, -79.65763092041016, 23.94258689880371, 46.19572448730469, 29.968841552734375, 83.20199584960938, -1.11090087890625, -7.7219390869140625, 100.84759521484375, 102.924560546875, -105.06976318359375, -2.7086029052734375, -24.120025634765625, 37.94720458984375, 14.602569580078125, 102.11528778076172, -1.54443359375, 29.442626953125, 6.4485931396484375, 2.9799652099609375, 97.33355712890625, -4.497373580932617, -94.94248962402344, 0.4392547607421875, 101.31572723388672, 47.08506774902344, -66.30453491210938, 59.998779296875, 125.2120361328125, 90.10948181152344, 13.016242980957031], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000328.npy"}
|
|
{"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 32.165748596191406, "std": 63.115867614746094, "min": -125.29916381835938, "p10": -40.40132751464843, "median": 21.114036560058594, "p90": 114.2643341064453, "max": 167.8336181640625, "pos_frac": 0.6640625, "sample": [-17.477508544921875, 2.55511474609375, -3.7839202880859375, 40.7662353515625, -0.28643798828125, -5.170085906982422, 11.565505981445312, 21.879058837890625, 3.7003860473632812, 29.81793212890625, 105.484375, -18.138343811035156, -77.96337890625, 60.951873779296875, 62.81227111816406, 137.5615997314453, 3.888519287109375, -54.855438232421875, 96.85153198242188, 69.29901123046875, 41.69573974609375, 14.31475830078125, 103.81253051757812, -15.240264892578125, 89.703369140625, 111.55418395996094, -94.61034393310547, 80.392822265625, 127.57444763183594, 45.945526123046875, 1.081146240234375, 15.30859375, 11.6624755859375, 110.43212890625, -34.009521484375, 37.5174560546875, 73.95664978027344, 43.02496337890625, 20.349014282226562, 76.72415161132812, 3.6263427734375, 129.138427734375, -2.58404541015625, -24.91265869140625, 67.35931396484375, -44.3802490234375, -4.033599853515625, 113.05209350585938, -4.091949462890625, -125.29916381835938, -4.047855377197266, 130.611328125, -34.53173828125, -10.312652587890625, 98.80279541015625, 44.5821533203125, 53.71075439453125, 36.34965515136719, 36.15663146972656, -70.48832702636719, 9.609138488769531, 108.1317138671875, -2.07855224609375, -38.23497009277344, -0.09604644775390625, -4.227619171142578, 0.03090667724609375, -38.696075439453125, 147.68161010742188, 13.457008361816406, 36.29002380371094, 0.67974853515625, 80.935302734375, 159.3284912109375, 12.457542419433594, 81.77850341796875, 55.883514404296875, -20.0340576171875, -71.5866928100586, 107.26934814453125, 102.58076477050781, -3.087432861328125, -109.35256958007812, 75.5977783203125, -10.427734375, 130.75294494628906, 54.33197021484375, 6.192138671875, 84.974365234375, 1.2989349365234375, 143.1668701171875, 3.6412887573242188, -65.562255859375, 133.59982299804688, 96.13003540039062, -88.44097900390625, 134.0186309814453, 117.0928955078125, -2.1544933319091797, 85.971435546875, -9.716110229492188, 14.72723388671875, 76.85073852539062, 34.37025451660156, -1.969146728515625, -4.308326721191406, 160.614501953125, 41.8668212890625, 108.0015869140625, 167.8336181640625, -8.99176025390625, 71.58622741699219, 94.8197250366211, -13.254669189453125, 34.1800537109375, -73.19210052490234, 18.270965576171875, -19.23785400390625, 46.693328857421875, -11.1522216796875, 40.893585205078125, 86.80825805664062, 1.7552547454833984, 52.18656921386719, 96.26690673828125, 37.57594299316406, -70.404052734375, -104.11821746826172], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000329.npy"}
|
|
{"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 28.790977478027344, "std": 57.606346130371094, "min": -96.01448059082031, "p10": -37.77921295166014, "median": 24.25836181640625, "p90": 112.33580627441405, "max": 182.98834228515625, "pos_frac": 0.734375, "sample": [10.3292236328125, 102.42588806152344, 18.7960205078125, -83.147705078125, 52.38683319091797, 77.26953125, 17.378021240234375, 125.79986572265625, 86.08311462402344, 30.598526000976562, -19.338134765625, 37.441009521484375, 2.580413818359375, -31.125, 35.93597412109375, -13.530441284179688, 113.77447509765625, 27.991668701171875, 19.84814453125, 0.5003662109375, 106.53916931152344, -74.41136169433594, -20.159164428710938, 5.139862060546875, 1.1142578125, 82.17230224609375, 15.7647705078125, -15.083824157714844, 8.331283569335938, -6.488525390625, 46.96124267578125, 40.580596923828125, 133.96148681640625, 69.65261840820312, 17.528282165527344, -4.03118896484375, -4.84503173828125, -63.529052734375, -21.9591064453125, 18.53460693359375, 44.116455078125, 127.04949951171875, 30.378402709960938, 36.695465087890625, 47.12469482421875, 39.92913818359375, 18.33041000366211, 145.150146484375, 8.764808654785156, 63.16986083984375, 5.58552360534668, 112.00906372070312, 75.32772827148438, -86.66033935546875, 8.398681640625, 22.641830444335938, -86.37158966064453, 25.876983642578125, 48.819091796875, 34.07453155517578, 6.0217132568359375, 125.31764221191406, 9.898811340332031, 70.589599609375, -96.01448059082031, 1.443756103515625, 92.61228942871094, -58.8258056640625, 5.776226043701172, 22.141036987304688, 38.186553955078125, -48.149169921875, 129.70101928710938, 42.09820556640625, -0.485992431640625, 86.81790161132812, -92.06109619140625, 138.4488525390625, -21.532806396484375, 1.213836669921875, 25.271728515625, 86.62025451660156, -3.453887939453125, -12.617767333984375, 24.239349365234375, 2.4767303466796875, 30.40777587890625, 113.09820556640625, 67.94802856445312, 36.26707458496094, 135.66033935546875, 9.46795654296875, 64.16974639892578, 28.815887451171875, 4.04119873046875, -33.33494567871094, 12.117340087890625, 26.561697006225586, 31.03778076171875, 95.1800765991211, 131.6649169921875, 28.83148193359375, -80.61387634277344, 60.17548370361328, -80.85317993164062, 106.91937255859375, -10.12591552734375, 19.949737548828125, -28.592754364013672, 38.1925048828125, 94.27218627929688, -77.17626190185547, -4.272621154785156, -63.958251953125, 1.77374267578125, 142.20782470703125, 38.378082275390625, 37.803321838378906, -0.677490234375, 100.44908142089844, -21.836639404296875, 182.98834228515625, 49.31842041015625, -9.593017578125, -22.39801025390625, 99.21209716796875, 63.574806213378906, 24.277374267578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000330.npy"}
|
|
{"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 36.597900390625, "std": 56.02269744873047, "min": -128.17788696289062, "p10": -24.077442169189453, "median": 24.268903732299805, "p90": 116.6013931274414, "max": 152.7247314453125, "pos_frac": 0.78125, "sample": [5.957183837890625, 141.77230834960938, 67.70758056640625, -56.810943603515625, 152.7247314453125, 60.06492614746094, -4.923370361328125, 0.312652587890625, 40.74481201171875, 63.2071533203125, 2.9943389892578125, -25.841148376464844, 52.48089599609375, -35.14720153808594, 139.7042236328125, -5.540008544921875, -1.2730712890625, 2.848785400390625, -8.377532958984375, -25.071563720703125, 11.270797729492188, 39.023956298828125, -28.941329956054688, 23.97845458984375, 24.55935287475586, 116.38856506347656, 94.43289184570312, 37.46119689941406, 7.6244049072265625, 96.71490478515625, 20.339813232421875, 32.60882568359375, 0.868896484375, 28.505767822265625, 36.91852569580078, 117.09799194335938, 15.860305786132812, 47.26612854003906, 126.2154541015625, 91.21019744873047, -31.59069061279297, 7.07244873046875, -7.9030609130859375, -117.50105285644531, 4.691572189331055, 7.8471527099609375, 118.03169250488281, 63.561431884765625, 61.65057373046875, 0.0, 33.7373046875, -1.1249008178710938, 77.94805908203125, -1.863037109375, 8.469146728515625, 52.12312316894531, 18.0888671875, 105.22552490234375, 34.450653076171875, 49.2896728515625, 15.938690185546875, 45.89569091796875, -35.151763916015625, -12.294723510742188, 140.75872802734375, 49.044246673583984, 125.92422485351562, 74.55564880371094, -1.2258453369140625, 41.35833740234375, 2.826141357421875, 14.261199951171875, 110.30694580078125, 12.831695556640625, 38.35187911987305, 17.338436126708984, -25.449737548828125, 4.001800537109375, 10.681312561035156, -23.651390075683594, 20.149566650390625, 28.812255859375, 94.35958862304688, 20.088485717773438, -128.17788696289062, 136.72100830078125, 4.31103515625, 114.57174682617188, -11.279525756835938, 32.2996826171875, 138.0650634765625, 80.90896606445312, 74.7820816040039, 136.74996948242188, 29.702346801757812, 111.05831909179688, 93.77510070800781, 13.217720031738281, -12.9666748046875, -1.9254608154296875, -9.8807373046875, 12.30511474609375, 111.83424377441406, 69.90092468261719, -102.78150177001953, 26.722686767578125, 147.4730224609375, 3.581521987915039, 92.59588623046875, 2.1443099975585938, 6.1984100341796875, 16.001632690429688, 98.87289428710938, 11.167755126953125, 86.32103729248047, 72.754638671875, 6.12249755859375, 72.014892578125, 16.268802642822266, 91.17394256591797, -105.72659301757812, 22.25598907470703, 135.09036254882812, 93.39390563964844, 49.891693115234375, -29.53265380859375, 97.16515350341797, 20.532058715820312], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000331.npy"}
|
|
{"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 42.15357971191406, "std": 59.70602035522461, "min": -131.1736602783203, "p10": -29.43467712402343, "median": 32.77445602416992, "p90": 113.59139404296874, "max": 161.91522216796875, "pos_frac": 0.7734375, "sample": [31.19369888305664, 47.9390869140625, 3.9587631225585938, 140.79779052734375, 87.28921508789062, 99.7593765258789, 58.90580749511719, 64.55338287353516, 148.77899169921875, 96.11962890625, 5.5439910888671875, 11.647867202758789, 5.744470596313477, -7.6437225341796875, 30.095687866210938, 78.92413330078125, -24.308746337890625, 18.451004028320312, 112.67327880859375, -36.05000305175781, 81.498046875, 2.88818359375, 33.61676025390625, 31.932151794433594, 0.5618133544921875, 0.432769775390625, -26.9287109375, 4.911468505859375, -5.836029052734375, 5.454345703125, 99.535888671875, 124.91314697265625, 1.84503173828125, 2.740814208984375, 115.84762573242188, 79.60967254638672, 98.17471313476562, 110.52708435058594, 43.61474609375, 65.76914978027344, 57.92573547363281, 11.99664306640625, 87.74786376953125, -3.319561004638672, 98.67364501953125, 31.495513916015625, 114.90850830078125, 93.35389709472656, -0.3820648193359375, 24.85466766357422, 111.65823364257812, 138.61468505859375, -77.34654235839844, -57.6539306640625, 138.40554809570312, 90.02581024169922, -62.603759765625, 51.01686096191406, 113.02691650390625, -94.77099609375, 31.87781524658203, 110.97686767578125, 98.10337829589844, 0.48329925537109375, 11.668937683105469, 9.93609619140625, 63.47978973388672, 19.0093994140625, 12.85130500793457, -33.172760009765625, -10.093292236328125, 102.37640380859375, 66.80625915527344, 8.132217407226562, 2.1146240234375, 131.23635864257812, 68.31689453125, 72.11193084716797, -27.8326416015625, 51.539215087890625, -39.09342956542969, 76.31364440917969, -19.013717651367188, 4.308074951171875, 83.534423828125, -65.78889465332031, 26.221771240234375, 17.57427978515625, 8.744384765625, 87.00631713867188, 110.085693359375, 77.9835205078125, -61.70368194580078, 97.17300415039062, 59.9794921875, -0.179443359375, 96.07431030273438, -23.30298614501953, -14.056045532226562, 58.47106170654297, 121.6661376953125, 87.39484405517578, -85.6422119140625, -46.61907958984375, 161.91522216796875, 145.41241455078125, 91.93025207519531, 20.220794677734375, 9.439041137695312, 31.918563842773438, 107.22105407714844, 56.818695068359375, 85.43605041503906, 139.71994018554688, 50.3634033203125, 97.1090087890625, -19.103546142578125, 109.18109130859375, -11.156806945800781, 86.69548797607422, 81.59725952148438, 135.29620361328125, -11.745803833007812, -36.78498840332031, 0.0, 8.91864013671875, -131.1736602783203, 26.265869140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000332.npy"}
|
|
{"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 33.01744842529297, "std": 52.41979217529297, "min": -85.78118896484375, "p10": -37.609205627441405, "median": 32.350425720214844, "p90": 101.32052001953124, "max": 135.75006103515625, "pos_frac": 0.7109375, "sample": [76.94274139404297, -60.625885009765625, -63.2491340637207, 107.86474609375, 103.6920166015625, -33.566192626953125, 54.55755615234375, 32.1524658203125, 50.861541748046875, 84.744384765625, -37.87904357910156, 12.177978515625, -13.369155883789062, 108.17206573486328, 102.85283660888672, 98.40859985351562, 96.25012969970703, 9.390213012695312, 74.03565216064453, 33.21826171875, -20.520217895507812, 32.54838562011719, 3.3324432373046875, 108.6947021484375, 99.22004699707031, 45.85259246826172, 45.31072998046875, 65.70574951171875, 100.71090698242188, 9.062225341796875, 26.585906982421875, 121.14218139648438, -12.59405517578125, 63.3236083984375, -66.32763671875, 7.695075988769531, 40.834930419921875, -9.082977294921875, 98.634765625, 86.94718933105469, 43.33258056640625, 23.037822723388672, 8.779449462890625, 5.006195068359375, 44.79864501953125, 102.7410888671875, -2.5276355743408203, 101.19046020507812, 22.954345703125, -65.1702880859375, -56.1314697265625, 65.14535522460938, 0.0, -5.15191650390625, 101.02157592773438, 25.990966796875, 81.89862060546875, 121.41650390625, 99.62238311767578, 62.78216552734375, 0.966156005859375, 48.43577575683594, -51.79835510253906, 101.62399291992188, 54.72296142578125, -26.05230712890625, 59.898712158203125, 135.75006103515625, 36.812232971191406, 29.32183837890625, -77.16383361816406, 62.966705322265625, 41.28314208984375, -51.739959716796875, 9.454063415527344, 89.91737365722656, -1.739776611328125, -1.92279052734375, 17.04718017578125, 88.56781005859375, 32.93074035644531, 38.68658447265625, 42.664154052734375, 109.49212646484375, -37.493560791015625, 99.39836120605469, 98.4345703125, -7.78076171875, -17.72857666015625, 109.32173156738281, 21.103252410888672, 66.59423828125, -10.625839233398438, 68.38252258300781, -25.752365112304688, -19.239990234375, 23.12744140625, 12.563232421875, -85.78118896484375, 76.74758911132812, -8.098464965820312, 83.52364349365234, -56.107696533203125, -7.2260284423828125, 73.69964599609375, 93.70097351074219, -0.110137939453125, 52.2308349609375, 101.63534545898438, 15.5709228515625, 91.4398422241211, 3.489765167236328, 89.3353271484375, 12.52008056640625, 33.482139587402344, -9.14306640625, -49.94309997558594, 75.05938720703125, -11.63525390625, 31.133819580078125, -7.587310791015625, 12.01617431640625, -83.447509765625, 12.959983825683594, -7.4514312744140625, 37.13822937011719, 20.414276123046875, 1.7922897338867188], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000333.npy"}
|
|
{"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 38.479148864746094, "std": 62.37233352661133, "min": -138.7500762939453, "p10": -42.07706298828125, "median": 34.006134033203125, "p90": 122.3806884765625, "max": 162.81712341308594, "pos_frac": 0.71875, "sample": [86.2494125366211, 78.87969970703125, -21.904205322265625, -2.11334228515625, 0.0, -5.08905029296875, 18.269088745117188, -28.576553344726562, 29.235380172729492, 2.3762550354003906, 33.17169189453125, 35.97850036621094, 39.2225341796875, 88.67411804199219, -67.39254760742188, 86.41609191894531, -1.248748779296875, 128.89041137695312, 37.29264831542969, -10.075439453125, 79.52595520019531, 53.608795166015625, 162.81712341308594, 124.6572265625, -43.4041748046875, 32.455810546875, 19.873992919921875, 19.27587890625, -0.5506057739257812, 17.865127563476562, 68.3695068359375, -9.767845153808594, 0.3656940460205078, 84.23931884765625, 95.04190826416016, 33.04833984375, -1.4991302490234375, 83.12391662597656, -114.30007934570312, 83.16864013671875, -2.8225021362304688, 5.428436279296875, 18.128021240234375, 14.8060302734375, 50.961151123046875, 57.5325927734375, 2.9769287109375, -47.824951171875, 99.42770385742188, -6.6422119140625, 101.73573303222656, -75.68035888671875, 110.61070251464844, 75.6195068359375, 148.36312866210938, 98.2017822265625, -5.543659210205078, 65.3031005859375, 7.5449066162109375, 23.779525756835938, 19.07683563232422, 6.694812774658203, 152.11761474609375, 56.294586181640625, -48.05747985839844, -83.0479965209961, 42.050445556640625, -89.003173828125, 12.533843994140625, -41.50830078125, 121.405029296875, 0.860382080078125, -14.56793212890625, 136.22247314453125, 99.4632568359375, 94.82305145263672, 60.88898468017578, 19.19274139404297, 9.95770263671875, 83.525634765625, 34.06201171875, 42.9163703918457, -8.903228759765625, 92.99845886230469, 83.32662963867188, 134.9943084716797, 53.92139434814453, 48.327117919921875, 84.68212890625, 23.780242919921875, 86.934814453125, 91.10052490234375, 104.50794982910156, -12.31976318359375, 2.141643524169922, 13.918693542480469, -3.1062088012695312, -21.592010498046875, 110.12078857421875, 129.38583374023438, 125.95976257324219, 9.589263916015625, -0.4294281005859375, -67.84378051757812, 113.5848617553711, -49.0919189453125, 63.70533752441406, 88.96160888671875, 126.04232788085938, 105.37521362304688, -138.7500762939453, 145.30990600585938, 111.80364990234375, 0.44157981872558594, 81.71731567382812, -112.16390991210938, -5.061553955078125, 61.1593017578125, -26.77886199951172, 37.77366638183594, 33.95025634765625, 161.9735107421875, 101.44595336914062, 50.080963134765625, 52.40907287597656, -9.621292114257812, -61.62579345703125, 137.2166748046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000334.npy"}
|
|
{"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 33.88789367675781, "std": 62.43171310424805, "min": -136.44256591796875, "p10": -48.559683227539054, "median": 24.721961975097656, "p90": 116.04530029296873, "max": 153.61138916015625, "pos_frac": 0.6953125, "sample": [-60.770240783691406, -106.21969604492188, 19.709945678710938, -32.03855895996094, 63.756744384765625, -53.746856689453125, 125.18621826171875, -10.063751220703125, 24.993301391601562, 99.6826171875, 72.11141967773438, -105.81707763671875, 36.99237060546875, -3.7331790924072266, -1.6035728454589844, 16.81878662109375, -10.481010437011719, -0.83782958984375, -21.97308349609375, -46.33660888671875, -14.55242919921875, 127.41726684570312, -0.9901275634765625, 10.401725769042969, 11.26104736328125, 1.111328125, 54.334442138671875, 50.048866271972656, -31.895355224609375, 101.26055908203125, -136.44256591796875, 151.36489868164062, -17.97705078125, 81.48382568359375, 37.62762451171875, 144.35269165039062, 9.442306518554688, 83.30093383789062, 110.7376480102539, 22.42249298095703, 118.91763305664062, 98.59939575195312, 101.763916015625, 84.70518493652344, 139.1553955078125, 114.81430053710938, 52.81895446777344, 95.96285247802734, 89.14425659179688, 120.93182373046875, -6.734855651855469, 70.88130187988281, 7.827409744262695, 76.95909881591797, 93.1367416381836, 86.82466125488281, 149.24050903320312, 58.532257080078125, 2.5247726440429688, 35.927734375, 29.540924072265625, -77.90042114257812, -7.2258453369140625, -24.7171630859375, -77.16482543945312, 106.0296630859375, 50.879005432128906, 112.12295532226562, 60.93766784667969, 24.45062255859375, 20.861602783203125, -3.0933837890625, 0.8771820068359375, 89.33946228027344, 147.04837036132812, 101.191162109375, 6.1391448974609375, -63.96598815917969, 7.5168609619140625, 18.698223114013672, 49.13728332519531, 57.454742431640625, -7.620391845703125, 85.6383056640625, 122.69767761230469, -12.728767395019531, -45.098876953125, 2.4403533935546875, 55.8345947265625, 15.7503662109375, -33.062957763671875, 1.0355987548828125, 82.15740966796875, 3.6550140380859375, -3.1193618774414062, -10.12017822265625, 14.152297973632812, 16.97687530517578, 64.60894775390625, -4.73138427734375, 131.74472045898438, -68.50650024414062, 27.387481689453125, 51.589447021484375, 100.43875122070312, 53.49009704589844, 0.0, 112.26776123046875, 124.053466796875, 97.70873260498047, -71.59393310546875, 75.43705749511719, 99.84251403808594, -83.53353881835938, -68.80632781982422, 153.61138916015625, 2.1795482635498047, -2.5399856567382812, 12.749143600463867, -26.396942138671875, 11.220169067382812, 73.05259704589844, 22.231689453125, 73.35545349121094, -55.63235092163086, 26.22125244140625, 33.51263427734375, 63.670074462890625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000335.npy"}
|
|
{"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 32.742488861083984, "std": 56.04511642456055, "min": -115.54605102539062, "p10": -28.064735412597653, "median": 25.082202911376953, "p90": 108.85775299072264, "max": 194.3865966796875, "pos_frac": 0.7109375, "sample": [50.051727294921875, 18.66546630859375, 62.88104248046875, -19.468292236328125, 90.59187316894531, 38.56976318359375, -15.54833984375, 95.27037048339844, 92.26666259765625, 76.31857299804688, 1.8497161865234375, 26.98773193359375, 44.8514404296875, 39.60369873046875, 3.1581459045410156, 42.81689453125, 0.24365234375, 134.48593139648438, 6.139533996582031, 110.90584564208984, 123.97398376464844, 23.176673889160156, 76.86778259277344, 96.03203582763672, -50.708099365234375, -14.013687133789062, 8.309036254882812, 10.827705383300781, 17.099624633789062, 34.164642333984375, 103.28677368164062, 34.98149108886719, -0.8979415893554688, 100.88238525390625, 89.53253173828125, 110.14439392089844, 46.6925048828125, 155.50064086914062, 13.552177429199219, 8.305023193359375, -62.90904235839844, 97.43037414550781, -34.722930908203125, 108.30633544921875, 86.19757080078125, 0.392822265625, -0.9537506103515625, 79.822509765625, -30.871795654296875, 20.39371109008789, 69.40959930419922, 1.759979248046875, 16.735076904296875, 67.53396606445312, 32.743865966796875, -40.901580810546875, -10.750732421875, -16.803741455078125, 55.541534423828125, -10.353309631347656, 64.72561645507812, 32.0206298828125, -67.30828857421875, 59.95404052734375, -24.5328369140625, 64.07295989990234, 35.923583984375, -54.42718505859375, 63.68312072753906, 105.0922622680664, 3.704315185546875, -7.0697021484375, 107.84930419921875, 49.011871337890625, 13.195684432983398, -32.87554931640625, -2.34912109375, 10.458389282226562, 16.2235107421875, -26.861709594726562, 3.899730682373047, 36.68878173828125, 30.873367309570312, -84.14181518554688, 92.37260437011719, 30.084930419921875, 79.9664306640625, 59.5712890625, 17.889854431152344, 108.05857849121094, -14.883026123046875, -82.18482208251953, 119.97270202636719, 118.68075561523438, 60.7344970703125, 58.875518798828125, 126.73294067382812, 16.078857421875, -4.716705322265625, 58.0616455078125, -3.501861572265625, 119.99159240722656, 4.549102783203125, -17.897064208984375, -52.54583740234375, 194.3865966796875, 9.7169189453125, 99.33619689941406, 148.22828674316406, -115.54605102539062, 5.293415069580078, -4.254234313964844, -0.514373779296875, 110.2625732421875, 28.966522216796875, 9.652828216552734, 52.9788818359375, -15.23907470703125, -84.302734375, 0.0, -13.185546875, 21.224212646484375, 31.450439453125, 123.82355499267578, -25.836700439453125, -14.225540161132812, -18.215133666992188, 36.982696533203125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000336.npy"}
|
|
{"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 37.97222137451172, "std": 58.30208206176758, "min": -98.20527648925781, "p10": -34.196215820312496, "median": 33.869773864746094, "p90": 117.35119171142577, "max": 151.6580810546875, "pos_frac": 0.7265625, "sample": [26.850311279296875, -30.761383056640625, 6.44049072265625, 8.550582885742188, 11.929489135742188, -2.02777099609375, -10.586822509765625, 136.06845092773438, 59.40899658203125, 151.6580810546875, 1.0643463134765625, 55.18085861206055, 20.45330810546875, 42.9805908203125, 133.966064453125, 116.61117553710938, -52.169219970703125, 12.177970886230469, -1.44537353515625, 15.374649047851562, -52.495849609375, -52.424156188964844, -26.120391845703125, -37.17437744140625, 62.576759338378906, 1.6852569580078125, 11.4788818359375, 111.40684509277344, 0.4936256408691406, 105.29852294921875, 22.701904296875, 122.62628173828125, 45.364837646484375, 64.19671630859375, 109.17330932617188, 76.8714599609375, 34.057708740234375, 4.869041442871094, 69.2813720703125, 78.74427795410156, -9.767631530761719, -25.265213012695312, -14.618072509765625, 107.8807373046875, 54.41748046875, 104.21717834472656, -6.67498779296875, -77.09669494628906, -21.955810546875, 32.128997802734375, 12.914703369140625, 96.86961364746094, -6.8598175048828125, 68.78982543945312, 69.706298828125, -56.239349365234375, 36.17506408691406, 47.44700622558594, 47.8572998046875, 19.81756591796875, 1.1529998779296875, 2.1823348999023438, 62.720069885253906, 125.34178924560547, 91.7766342163086, -25.1226806640625, 95.00126647949219, 90.73236083984375, -98.20527648925781, 22.248443603515625, 81.458740234375, 125.48434448242188, 30.0550537109375, 29.006988525390625, 83.28770446777344, 2.267852783203125, -3.353729248046875, -19.89057159423828, 119.07789611816406, 75.72467041015625, 14.336299896240234, 79.6444091796875, 56.15178680419922, 112.9764404296875, 92.94967651367188, 112.6322021484375, -15.804086685180664, 71.34393310546875, -26.316543579101562, -62.186256408691406, 44.36787414550781, 108.92539978027344, 33.68183898925781, 139.02847290039062, -58.94116973876953, 127.53778076171875, -11.860015869140625, -3.62115478515625, 78.7844009399414, 75.063720703125, 54.350502014160156, -11.37353515625, -73.60838317871094, 123.59451293945312, -31.94500732421875, -54.72765350341797, 137.72589111328125, 51.80564880371094, 1.7155647277832031, 1.6965351104736328, -21.69781494140625, 36.79307556152344, 109.50933837890625, 120.08489990234375, -52.753997802734375, 125.0606689453125, 1.33001708984375, 3.4353713989257812, -32.91986083984375, 101.17300415039062, 92.23538970947266, 75.93820190429688, -41.856658935546875, 39.02812194824219, 12.635833740234375, 112.8388900756836, 112.23432922363281, 74.4185791015625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000337.npy"}
|
|
{"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 44.19725799560547, "std": 58.31208801269531, "min": -98.74943542480469, "p10": -19.623168945312496, "median": 32.013465881347656, "p90": 125.02816772460938, "max": 169.924560546875, "pos_frac": 0.796875, "sample": [124.36148071289062, -14.042804718017578, 6.39910888671875, 86.66206359863281, 117.45858764648438, 128.6126708984375, 89.90286254882812, 85.0028076171875, -6.924736022949219, -29.2305908203125, 28.823192596435547, 102.68031311035156, 32.30303955078125, 4.702045440673828, 112.8323974609375, 16.650466918945312, 94.63925170898438, 37.66864013671875, -12.490249633789062, -18.38885498046875, 121.09307861328125, 71.37510681152344, -0.9249629974365234, 137.812744140625, 11.30047607421875, 43.24632263183594, 105.04367065429688, 70.19762420654297, -8.084259033203125, 1.2859992980957031, 40.60577392578125, 8.191009521484375, -2.462310791015625, 18.6636962890625, 53.24922180175781, 89.82376098632812, 43.88397216796875, 122.18096923828125, 21.245193481445312, 30.339065551757812, 58.661956787109375, 27.560028076171875, 51.88763427734375, 114.0526123046875, 127.426025390625, 26.7310791015625, -28.7276611328125, 108.90991973876953, 65.99362182617188, 31.598388671875, 42.4445686340332, -22.50323486328125, 126.55838012695312, -83.2857437133789, 27.1729736328125, -0.8012924194335938, 10.680709838867188, 128.77706909179688, -59.70855712890625, 124.93798828125, 97.61248779296875, 169.924560546875, 105.56211853027344, 40.2373046875, 60.4111328125, 26.25634765625, 14.942771911621094, -10.1580810546875, -96.69473266601562, 1.4007186889648438, 8.32357406616211, 86.28121948242188, 127.817626953125, 9.730239868164062, -24.263595581054688, 0.9122905731201172, 17.94706153869629, 115.18441772460938, 114.88821411132812, -5.584928512573242, 71.2532958984375, 1.3248653411865234, 86.09120178222656, -2.8807525634765625, 89.88731384277344, 20.23187255859375, -16.49456787109375, 6.7641754150390625, 68.50601196289062, 98.53338623046875, 74.39309692382812, 135.511474609375, 17.539016723632812, 0.765869140625, -85.16600036621094, 111.94296264648438, 18.2015380859375, 148.11618041992188, -74.79393005371094, 49.838539123535156, 22.416839599609375, -49.25550842285156, 13.030097961425781, 54.672119140625, 26.044525146484375, 124.634765625, 78.46189880371094, 31.723892211914062, -35.579627990722656, 143.0086669921875, 40.957489013671875, 13.133735656738281, 0.6084461212158203, 33.48822021484375, -6.9774169921875, 125.23858642578125, 30.4404296875, 78.29194641113281, 78.00369262695312, 38.15814208984375, -98.74943542480469, 5.08294677734375, 28.61151123046875, 1.0863571166992188, 143.88043212890625, 143.67416381835938, -32.958709716796875, 101.77008056640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000338.npy"}
|
|
{"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 44.218299865722656, "std": 60.325984954833984, "min": -135.65725708007812, "p10": -33.927050781249996, "median": 48.69752502441406, "p90": 112.42944030761718, "max": 178.30941772460938, "pos_frac": 0.7734375, "sample": [57.790550231933594, -36.18402099609375, 66.3392562866211, 80.4354248046875, 76.14579010009766, -4.7136688232421875, 28.528717041015625, -28.0572509765625, 81.48471069335938, -3.37725830078125, 130.49172973632812, 17.44732666015625, 1.3714599609375, 2.707061767578125, 84.37228393554688, 0.600433349609375, 42.04144287109375, 65.01388549804688, 52.4371337890625, -54.61322021484375, 97.40342712402344, 10.379058837890625, 50.637725830078125, 84.61575317382812, 0.0, 112.2352294921875, -32.95977783203125, 42.18968200683594, 0.0, 54.97005844116211, 56.131874084472656, -40.14134216308594, 126.54376220703125, 100.35252380371094, 3.932861328125, -96.54875183105469, 99.64031982421875, 125.0166015625, 37.49540710449219, 18.320541381835938, 61.3184814453125, 106.8985595703125, -26.37237548828125, 85.79397583007812, 90.05035400390625, 46.75732421875, 81.0283203125, 1.7006759643554688, 2.4282455444335938, 102.53121948242188, 93.79122924804688, 14.599609375, 168.752197265625, 70.19204711914062, 98.516357421875, 99.80145263671875, 23.324066162109375, 8.057289123535156, 105.30776977539062, 86.57955932617188, 98.43203735351562, -0.75482177734375, -20.548004150390625, -62.03643798828125, -1.923666000366211, 132.99432373046875, 9.119964599609375, 88.56254577636719, 94.314697265625, 30.89593505859375, 33.494598388671875, 15.974456787109375, -81.66947937011719, 64.28399658203125, 95.5743408203125, 24.731689453125, 113.941650390625, 108.30697631835938, -57.103759765625, 87.60574340820312, 3.4884567260742188, 88.70179748535156, 118.81564331054688, -9.025421142578125, 111.47909545898438, 178.30941772460938, 50.829345703125, 0.0, 62.9326171875, 61.1787109375, 80.99298095703125, 95.10159301757812, 97.20777893066406, 113.75567626953125, 105.88104248046875, -9.812713623046875, 26.577346801757812, -3.296630859375, 4.58428955078125, 112.88259887695312, 107.59539794921875, -42.38417053222656, 26.597900390625, 16.201904296875, 10.78741455078125, 100.6427001953125, 103.48783874511719, 1.7471923828125, -38.68341064453125, -14.945198059082031, 124.27497100830078, 54.4090576171875, -135.65725708007812, 101.44395446777344, 24.08856201171875, 12.121353149414062, 13.797576904296875, -3.08538818359375, 121.51422119140625, -68.4501953125, -119.53700256347656, 138.8792724609375, 40.349212646484375, 31.76153564453125, 106.88862609863281, 2.4196929931640625, -58.45496368408203, 105.81759643554688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000339.npy"}
|
|
{"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 37.374969482421875, "std": 59.568359375, "min": -103.21548461914062, "p10": -34.51472930908203, "median": 28.376907348632812, "p90": 113.0009780883789, "max": 169.4649658203125, "pos_frac": 0.71875, "sample": [154.11534118652344, 58.779014587402344, -27.317733764648438, 101.41241455078125, 2.173999786376953, 108.99208068847656, -37.86114501953125, -15.879951477050781, -46.8590087890625, 113.24266052246094, -69.25518798828125, 90.48762512207031, -59.668487548828125, 4.595428466796875, 0.0, 107.7012710571289, 26.606231689453125, -33.08055114746094, 82.37606048583984, 120.75675964355469, 109.0467758178711, 15.38763427734375, -26.303482055664062, 84.3360595703125, -46.91326904296875, -6.0713653564453125, -23.356658935546875, 98.71123504638672, -4.6117401123046875, 112.89739990234375, 145.7889404296875, 2.399656295776367, 59.735870361328125, 113.43046569824219, 24.59051513671875, -10.586181640625, 1.880117416381836, -53.673095703125, 54.854949951171875, 12.364891052246094, 105.76611328125, 96.579833984375, -69.94366455078125, 36.16363525390625, 48.36504364013672, 114.64982604980469, -21.40618896484375, 51.73500061035156, 13.50006103515625, -14.459823608398438, 69.39288330078125, 138.37548828125, -11.60882568359375, -63.48333740234375, 8.675664901733398, -8.006256103515625, 100.0882568359375, 83.17791748046875, 31.916839599609375, -20.369186401367188, 5.9209442138671875, -103.21548461914062, 120.55055236816406, 0.0, 147.41525268554688, 12.479278564453125, 53.14105224609375, 13.972686767578125, 104.74710083007812, 30.1475830078125, 9.90594482421875, -3.477771759033203, 98.85072326660156, -70.9178695678711, 92.11891174316406, 33.6405029296875, 54.2186279296875, -50.72686767578125, 118.69775390625, 10.98248291015625, 99.87808227539062, 169.4649658203125, -63.65704345703125, 20.7559814453125, -12.726776123046875, 77.33438110351562, 49.738983154296875, 17.975387573242188, 86.96728515625, 22.047012329101562, 46.307373046875, -8.784128189086914, 52.944854736328125, -10.41412353515625, 94.9827880859375, 82.46856689453125, -14.856094360351562, 4.0443115234375, 40.03558349609375, 46.848602294921875, 166.96929931640625, 7.721590042114258, 104.7796630859375, 1.4434051513671875, 101.83535766601562, -25.663360595703125, 20.13488006591797, -89.38992309570312, 103.98388671875, 88.223876953125, 97.63196563720703, 35.964027404785156, 8.785701751708984, 75.40054321289062, 118.84521484375, 111.57221984863281, 14.5281982421875, 23.504966735839844, 13.605113983154297, 70.43212890625, 38.244873046875, 41.11871337890625, -21.323760986328125, 21.524303436279297, 36.71320343017578, 14.455001831054688, 98.20956420898438, -9.396747589111328], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000340.npy"}
|
|
{"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 30.348173141479492, "std": 62.4902458190918, "min": -107.45327758789062, "p10": -54.05557861328125, "median": 22.864105224609375, "p90": 111.77181091308591, "max": 169.27532958984375, "pos_frac": 0.6796875, "sample": [5.064537048339844, -11.51409912109375, 66.42831420898438, 104.63053894042969, 82.07911682128906, -35.177825927734375, 66.45849609375, 15.31536865234375, 18.780288696289062, -1.3138427734375, 92.90960693359375, 33.45562744140625, 107.38720703125, 11.396228790283203, 140.33111572265625, 124.58189392089844, -42.209686279296875, 39.44854736328125, 38.2630615234375, 44.087608337402344, 101.97735595703125, 21.63385009765625, 95.25253295898438, -75.98112487792969, 135.1715087890625, 40.303466796875, 8.820289611816406, 16.22564697265625, 7.5822906494140625, -9.58135986328125, 1.812225341796875, 32.66249084472656, -11.632492065429688, -5.53204345703125, -80.70904541015625, 17.03363037109375, -55.50958251953125, -63.512420654296875, -0.20273590087890625, 34.20063781738281, 12.623870849609375, 45.020511627197266, 117.18161010742188, 34.17933654785156, 48.44285583496094, 6.7479248046875, 109.58004760742188, 67.19412231445312, -5.167593002319336, 89.15846252441406, 71.74847412109375, 102.00747680664062, 144.30316162109375, -17.057861328125, 104.12014770507812, 83.23455810546875, 77.0869140625, 34.170623779296875, 18.318954467773438, -17.93585205078125, 54.992919921875, -85.51699829101562, 169.27532958984375, -7.131183624267578, 95.76834106445312, -86.62628936767578, -83.5643310546875, 106.89701080322266, 120.113037109375, 3.0360107421875, 35.58905029296875, 107.62368774414062, 76.692138671875, -11.630119323730469, 89.73411560058594, 107.8014907836914, 66.23464965820312, 5.2105865478515625, 24.357162475585938, -94.08195495605469, 53.565673828125, 1.586639404296875, -12.263290405273438, -30.46319580078125, 152.60797119140625, 24.593826293945312, -15.49322509765625, 46.9295654296875, 11.713546752929688, 116.98883056640625, -60.47944641113281, 60.63812255859375, 130.56991577148438, 90.87489318847656, -63.44358825683594, -14.654541015625, -26.38971710205078, -1.082113265991211, -10.212432861328125, 24.0943603515625, -13.760162353515625, 104.09271240234375, 17.6783447265625, -2.0831298828125, -95.7103271484375, 25.789047241210938, 27.132354736328125, -39.11865234375, -107.45327758789062, -27.42690658569336, 11.486419677734375, -53.43243408203125, -14.52850341796875, 119.917236328125, -106.64614868164062, 3.5929412841796875, 98.37521362304688, 116.88592529296875, 97.04190063476562, 53.33610534667969, 102.76593017578125, 5.2155914306640625, 25.844459533691406, -33.007080078125, 144.03668212890625, 14.8787841796875, -12.492218017578125, 12.323898315429688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000341.npy"}
|
|
{"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 37.798404693603516, "std": 55.81844711303711, "min": -139.0428466796875, "p10": -21.021900939941407, "median": 21.941715240478516, "p90": 112.28545837402342, "max": 160.31890869140625, "pos_frac": 0.7421875, "sample": [-6.026092529296875, 18.2125244140625, 27.58453369140625, 18.06317138671875, 60.794891357421875, 149.29318237304688, 13.945230484008789, 48.14421844482422, -0.26320648193359375, 34.9998779296875, -19.147933959960938, 10.877349853515625, 0.3119964599609375, 92.78457641601562, 75.60595703125, 98.78231811523438, -5.039665222167969, 48.64617919921875, -45.590553283691406, 3.3963165283203125, 42.66169738769531, 5.664947509765625, 11.549072265625, 24.096824645996094, -37.867919921875, 10.554954528808594, 14.442070007324219, -41.479248046875, 42.156036376953125, 94.65301513671875, 8.812446594238281, 84.81488037109375, 68.65995788574219, 63.298187255859375, -6.5417022705078125, 11.863250732421875, 14.757514953613281, -13.313468933105469, -26.304763793945312, 24.741058349609375, 48.38230895996094, 136.04718017578125, 16.74664306640625, -29.080368041992188, -14.740310668945312, -17.52399444580078, -32.310791015625, 52.8106689453125, 46.9830322265625, 8.028350830078125, 86.22494506835938, 1.255157470703125, -20.775482177734375, 117.90628051757812, 10.996963500976562, 125.62899780273438, 109.99990844726562, -1.05706787109375, 62.298828125, 84.56427001953125, 119.00827026367188, 117.618408203125, 81.16921997070312, 88.46391296386719, 81.21800231933594, -81.09725952148438, -12.776779174804688, 0.0, -9.12115478515625, 36.8017578125, 128.18084716796875, 2.47491455078125, 128.53857421875, 4.583044052124023, 44.23497772216797, 91.42169189453125, 109.05502319335938, 31.218894958496094, 125.66061401367188, 3.6818294525146484, -13.3682861328125, 93.63812255859375, 19.786605834960938, 75.52609252929688, -64.2138671875, -3.073322296142578, -139.0428466796875, -17.004241943359375, 77.46673583984375, 74.47237396240234, 104.27958679199219, 86.31491088867188, 0.3891754150390625, 11.998046875, 12.446941375732422, 84.8564682006836, 85.713134765625, 24.135986328125, 13.896835327148438, -61.62274169921875, 15.7674560546875, 93.72323608398438, -21.596878051757812, 160.31890869140625, 14.337799072265625, 95.03886413574219, -25.034820556640625, 11.8814697265625, 75.6864013671875, -68.19872283935547, 119.1036376953125, 87.3773193359375, 35.06257629394531, 100.60272216796875, 91.43328857421875, 109.181396484375, 18.871734619140625, 19.08966064453125, 93.86712646484375, 107.95515441894531, -12.510887145996094, -11.869888305664062, -18.004791259765625, 145.249755859375, 86.47940826416016, -12.126716613769531, 143.010498046875, 11.59051513671875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000342.npy"}
|
|
{"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 33.91915512084961, "std": 58.759883880615234, "min": -136.590576171875, "p10": -33.81996154785156, "median": 23.119098663330078, "p90": 110.49217071533202, "max": 161.63116455078125, "pos_frac": 0.71875, "sample": [-26.8221435546875, 86.67819213867188, 104.07701110839844, 3.6442394256591797, 36.91729736328125, 41.99920654296875, 39.618682861328125, 20.25933837890625, -38.193634033203125, 26.0299072265625, 108.84201049804688, -34.76983642578125, -11.741943359375, -28.3193359375, 73.05908203125, 123.9319076538086, 112.77348327636719, 37.349273681640625, -8.888404846191406, -5.83296012878418, 135.28628540039062, 16.962844848632812, 117.132568359375, 109.51446533203125, 70.7441635131836, 62.92039489746094, 2.450775146484375, 65.80032348632812, 19.7755126953125, 93.77362823486328, 45.6090087890625, -75.32006072998047, 25.319061279296875, -7.3384857177734375, -77.31689453125, 22.56195831298828, 11.30133056640625, 108.23947143554688, -34.974395751953125, 2.324951171875, -26.219406127929688, 134.34268188476562, 31.902587890625, 47.6788330078125, 0.0, -69.289306640625, 22.3277587890625, 101.44183349609375, 50.516944885253906, 153.32745361328125, -93.64459228515625, 9.895133972167969, 29.465757369995117, 12.997478485107422, 107.64697265625, -33.412872314453125, 78.75291442871094, 59.91664123535156, -6.838203430175781, 45.30853271484375, 92.82162475585938, 3.4368438720703125, -11.57509994506836, 2.6973114013671875, 87.39128112792969, 107.03962707519531, -13.5128173828125, 107.78079223632812, 14.2965087890625, 60.6409912109375, 60.647377014160156, -15.152679443359375, 49.132781982421875, -20.321945190429688, 1.07794189453125, 30.43158721923828, 2.5423583984375, 161.63116455078125, 91.21029663085938, 106.14970397949219, 27.263504028320312, 43.3114013671875, -27.609771728515625, 10.0919189453125, 92.74742126464844, 9.97943115234375, 3.1209716796875, 30.967758178710938, 10.312957763671875, -24.21856689453125, 10.638824462890625, 107.4337158203125, 104.12991333007812, 156.42205810546875, 14.2337646484375, -8.180435180664062, 92.95161437988281, 50.66107177734375, 103.16400146484375, 18.52936553955078, 118.87977600097656, 102.94947814941406, 4.291015625, 85.20965576171875, -136.590576171875, 116.90460205078125, -16.879486083984375, -4.643955230712891, -47.34228515625, -38.77984619140625, 8.00234603881836, 90.44397735595703, 114.7051010131836, 4.9580078125, -53.611541748046875, -10.373748779296875, -8.496292114257812, -56.469696044921875, 28.668487548828125, 134.36032104492188, 60.414756774902344, -0.07247543334960938, 23.676239013671875, 8.351913452148438, 7.645263671875, 121.45687866210938, -10.078704833984375, -81.73954010009766], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000343.npy"}
|
|
{"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 39.381263732910156, "std": 59.39867401123047, "min": -112.14691162109375, "p10": -27.071836853027342, "median": 27.951736450195312, "p90": 121.15880126953125, "max": 154.49722290039062, "pos_frac": 0.7421875, "sample": [20.1866455078125, -7.0450592041015625, 89.7017822265625, 62.14080810546875, -11.810272216796875, 101.85211181640625, 106.52035522460938, 27.065582275390625, 100.27203369140625, 7.17315673828125, 86.45892333984375, 17.457107543945312, 86.64356994628906, -25.136348724365234, 68.34788513183594, 131.88348388671875, 21.155975341796875, 42.46647644042969, -112.14691162109375, 81.98080444335938, -62.664794921875, 3.341827392578125, 2.12017822265625, -22.189971923828125, 141.010498046875, 59.352813720703125, -13.068479537963867, 133.00350952148438, 113.53620910644531, 11.40521240234375, 116.21200561523438, -32.218505859375, -0.14923095703125, 100.88247680664062, -110.79592895507812, 41.47454833984375, 99.72747802734375, 82.94436645507812, 10.510833740234375, -8.669235229492188, -108.74909973144531, 15.316619873046875, 13.733543395996094, -15.310394287109375, -4.691505432128906, 14.296737670898438, -54.321044921875, 50.460205078125, 130.041259765625, 41.563873291015625, -26.6956787109375, -27.949539184570312, -0.365814208984375, -42.91522979736328, 65.88348388671875, 68.7833251953125, 125.05027770996094, -3.970550537109375, 149.56028747558594, 129.0067138671875, 108.65608215332031, 103.96022033691406, 150.35971069335938, 86.90347290039062, 113.99595642089844, -3.1928176879882812, 14.708251953125, 95.20077514648438, -5.9926300048828125, -7.256988525390625, -43.806365966796875, 14.876174926757812, 13.58258056640625, 154.49722290039062, 48.161590576171875, 48.68971252441406, 16.4580078125, 62.0975341796875, 13.74267578125, 70.85009765625, 41.33647155761719, 15.3470458984375, 46.895751953125, 103.27679443359375, 2.731231689453125, 48.2999267578125, 1.84869384765625, 106.22994232177734, 16.225433349609375, 32.32488250732422, 145.11126708984375, 63.74462127685547, 14.387649536132812, -37.02618408203125, -10.685646057128906, 124.5320053100586, 57.1104736328125, -2.4209861755371094, 2.6480865478515625, -31.215774536132812, -6.100788116455078, 30.739883422851562, 15.2713623046875, 139.777099609375, -12.1485595703125, -48.9189453125, 110.21051025390625, 26.575775146484375, 10.769523620605469, 10.65185546875, 4.5540771484375, 78.0777587890625, 5.9327545166015625, 119.8330078125, 70.03147888183594, 35.72566223144531, 51.606201171875, 23.37432861328125, 28.837890625, 120.89309692382812, 94.29035949707031, 25.940345764160156, 52.955902099609375, 115.65135955810547, 55.87615966796875, -19.969322204589844, 121.77877807617188, -98.26986694335938], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000344.npy"}
|
|
{"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 35.097618103027344, "std": 51.30451965332031, "min": -111.45166015625, "p10": -19.57052307128906, "median": 22.032840728759766, "p90": 107.559326171875, "max": 137.64501953125, "pos_frac": 0.7578125, "sample": [35.033447265625, 74.96212005615234, 65.94052124023438, -80.29039001464844, 132.94952392578125, 70.58634948730469, 1.89776611328125, -35.57781982421875, 13.536224365234375, -2.2035598754882812, 64.80433654785156, 9.018852233886719, -3.9189376831054688, 1.9225234985351562, 49.65861511230469, 29.466598510742188, 117.62265014648438, 11.64459228515625, 8.154541015625, -12.97186279296875, 18.798065185546875, -20.98577880859375, 87.54859161376953, 104.07328796386719, 3.1126575469970703, 47.386810302734375, 99.33139038085938, 23.337188720703125, -74.43777465820312, 1.5417594909667969, -7.565460205078125, 11.588104248046875, -4.304393768310547, 34.10333251953125, -0.569915771484375, 49.4671630859375, 54.7626953125, 5.542327880859375, 16.04058837890625, 22.57061767578125, 35.080963134765625, -7.168478012084961, -27.89590072631836, 96.32525634765625, 8.6405029296875, 102.48681640625, 19.929153442382812, 0.38696861267089844, 40.53070068359375, 137.64501953125, 94.03504180908203, 90.18804931640625, 50.680389404296875, 102.31266784667969, 124.47920227050781, 30.004241943359375, 25.18255615234375, -28.837005615234375, 13.345565795898438, -6.425048828125, 116.59429931640625, 41.36865234375, 44.122344970703125, 27.08905029296875, 17.8651123046875, -46.67547607421875, -111.45166015625, 11.606014251708984, -43.29449462890625, -19.251068115234375, 126.55636596679688, 132.2996826171875, 107.08512878417969, 18.843246459960938, 34.151283264160156, -15.25006103515625, 136.89700317382812, -35.81715393066406, 19.831573486328125, -2.760528564453125, 108.2685546875, 119.798828125, 74.7724838256836, -68.13095092773438, 66.35446166992188, 25.249156951904297, -26.415176391601562, 131.27215576171875, 48.193359375, 65.0479507446289, 123.25032043457031, 92.71495056152344, 17.967315673828125, 8.469482421875, 68.93058776855469, 13.228057861328125, 49.54242706298828, 99.6932144165039, 97.68914794921875, -15.593170166015625, 108.5096435546875, -20.31591796875, 92.35762786865234, 7.3308868408203125, 16.201873779296875, 37.05381774902344, 1.4269332885742188, 20.288909912109375, -1.2389984130859375, -11.665374755859375, -9.953475952148438, 85.62792205810547, 16.61614990234375, 55.33995056152344, 107.25537109375, 98.1021728515625, 10.990875244140625, 24.706317901611328, 0.0, 21.49506378173828, -3.0331192016601562, 4.2821197509765625, 8.333091735839844, -0.279327392578125, 103.35588073730469, 5.33331298828125, 69.26911926269531, 60.48779296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000345.npy"}
|
|
{"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 38.70262908935547, "std": 57.26865005493164, "min": -133.5858154296875, "p10": -28.67510223388671, "median": 34.287052154541016, "p90": 114.95774688720702, "max": 149.3201904296875, "pos_frac": 0.7265625, "sample": [-40.585052490234375, -34.34663391113281, -0.38985443115234375, -43.58119201660156, 47.05638122558594, 52.8486328125, 125.57174682617188, 127.68115234375, 70.80049896240234, -2.9873504638671875, 98.13612365722656, -16.186416625976562, -19.383453369140625, 27.809967041015625, 101.11294555664062, -84.37286376953125, -45.85404968261719, 6.557098388671875, 21.03741455078125, 119.53875732421875, 12.942916870117188, 4.4567413330078125, 18.360610961914062, -6.237762451171875, 42.38467025756836, 91.33024597167969, 7.64581298828125, 144.85552978515625, -12.769500732421875, -10.63165283203125, 30.487884521484375, 24.588661193847656, -74.13948822021484, 19.4439697265625, -0.195892333984375, 117.9252700805664, 46.41789245605469, 119.2200927734375, 15.749515533447266, 20.75970458984375, 63.14764404296875, 47.104217529296875, -96.61907958984375, 5.546234130859375, -38.13581848144531, 52.99169921875, -1.86846923828125, -6.1385040283203125, -6.6354827880859375, 51.76144027709961, 92.92160034179688, 16.32574462890625, 93.10610961914062, -26.24444580078125, 22.538970947265625, 68.10577392578125, 134.03408813476562, 34.67120361328125, -2.5245590209960938, 86.19131469726562, 127.2161865234375, 130.3458251953125, 31.54443359375, 10.222877502441406, 26.729476928710938, 33.90290069580078, 130.17422485351562, -11.490631103515625, 74.06204223632812, 76.4041748046875, 81.10052490234375, 54.1160888671875, 100.83880615234375, 134.66436767578125, 109.922607421875, 107.822998046875, -49.83796691894531, 108.9990234375, 49.26106262207031, 35.199951171875, 70.29583740234375, -88.85595703125, 113.91879272460938, 6.889984130859375, -6.0068511962890625, -8.879684448242188, -21.99030303955078, 4.153696060180664, -39.48114013671875, 76.83697509765625, 74.854248046875, 32.9429931640625, -10.985824584960938, 113.29461669921875, 0.1127777099609375, -9.692474365234375, 88.36573791503906, -133.5858154296875, 0.2520294189453125, 106.9930419921875, 2.5576629638671875, 7.6048583984375, 86.49578857421875, 82.95051574707031, 117.38197326660156, 61.633445739746094, 84.13861083984375, 58.31842041015625, 13.09710693359375, 58.187103271484375, 0.0, 41.71324157714844, -3.643463134765625, 50.49336242675781, 98.69961547851562, 81.29754638671875, 2.8953857421875, 99.05050659179688, 39.44551086425781, -35.123138427734375, 95.05537414550781, 38.24749755859375, -20.388931274414062, 91.48880004882812, 110.67613220214844, 91.71449279785156, 6.657957077026367, 149.3201904296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000346.npy"}
|
|
{"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 33.67393112182617, "std": 56.4010009765625, "min": -153.45687866210938, "p10": -28.494104003906248, "median": 24.88736343383789, "p90": 109.36051025390624, "max": 134.32843017578125, "pos_frac": 0.734375, "sample": [8.147674560546875, 64.26348876953125, 98.04362487792969, 10.3509521484375, 36.68768310546875, 56.61297607421875, 5.17364501953125, 115.31758880615234, 11.751930236816406, 92.35188293457031, -153.45687866210938, 54.4364013671875, 105.66865539550781, 28.046737670898438, 122.90260314941406, -57.43084716796875, -27.55682373046875, -1.8937530517578125, 99.95458984375, 121.51301574707031, 71.82713317871094, -33.226654052734375, 72.3995361328125, 9.28811264038086, 27.00829315185547, 33.39434814453125, 108.53128051757812, 19.77032470703125, 21.234054565429688, -2.4542922973632812, 1.515228271484375, 44.918212890625, 123.32832336425781, 18.753005981445312, 107.54299926757812, 32.6588134765625, 105.00582885742188, -8.502647399902344, 42.208099365234375, 23.793739318847656, 28.16954803466797, 108.44366455078125, 132.85537719726562, 22.935577392578125, 16.5355224609375, 67.56243896484375, -13.552459716796875, 13.5408935546875, 19.782119750976562, -71.38937377929688, -2.9337310791015625, -3.5250701904296875, 52.61408996582031, -44.00999450683594, 49.57403564453125, 73.67965698242188, -3.8248443603515625, 100.12535858154297, 65.755126953125, -48.20295715332031, 82.13262939453125, 40.803192138671875, -1.0516128540039062, -8.564779281616211, 11.312118530273438, 72.31208801269531, 73.66754150390625, 38.67041015625, 132.81234741210938, 44.83770751953125, 8.471893310546875, -99.62667846679688, -34.26121520996094, 10.79817008972168, 83.60159301757812, -11.38859748840332, -20.139625549316406, -8.256546020507812, 19.78057098388672, 5.3067626953125, 5.61468505859375, 73.3970947265625, 119.90554809570312, -13.272064208984375, 12.269367218017578, -109.02022552490234, 76.4371109008789, 134.32843017578125, 36.75067138671875, 55.16436767578125, -14.404129028320312, 71.76487731933594, -48.8507080078125, 100.6438980102539, 12.701469421386719, -6.7356109619140625, 50.16961669921875, -2.7169647216796875, 101.14132690429688, -43.776611328125, -132.23028564453125, -24.59844970703125, 122.85595703125, 111.29537963867188, 0.0, 14.923004150390625, 15.284637451171875, 29.5418701171875, 71.18580627441406, 118.175048828125, 25.980987548828125, 7.3002777099609375, 101.09275817871094, 91.15953063964844, -9.056884765625, 13.402999877929688, 23.39794921875, -7.1902313232421875, 97.00517272949219, 21.452850341796875, 39.54302978515625, 123.88519287109375, 120.92607116699219, 54.120635986328125, 3.1727371215820312, -30.68109130859375, 69.19573974609375, 12.378517150878906], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000347.npy"}
|
|
{"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 30.387466430664062, "std": 54.600223541259766, "min": -146.32327270507812, "p10": -23.745132446289062, "median": 11.071807861328125, "p90": 108.3810821533203, "max": 146.0135955810547, "pos_frac": 0.6640625, "sample": [24.2742919921875, -2.25958251953125, 0.06739616394042969, 2.8099365234375, -4.14501953125, 3.779022216796875, 100.62313842773438, -7.285541534423828, 60.265350341796875, 88.84860229492188, 2.552886962890625, 107.62750244140625, 37.626708984375, 93.15966796875, 30.800689697265625, 12.2734375, 138.98895263671875, 90.4580078125, -48.138633728027344, 0.0, 17.0474853515625, 42.43577575683594, 82.90282440185547, 7.415252685546875, -29.10247802734375, 130.25718688964844, -3.96734619140625, -2.7795562744140625, -7.7097930908203125, 72.54867553710938, 58.89752197265625, 77.42111206054688, -8.93646240234375, -12.975044250488281, 67.7276611328125, -13.44329833984375, 111.98388671875, -4.770671844482422, 83.3603515625, 125.80059814453125, 2.8092174530029297, -13.70159912109375, -61.33099365234375, 22.814117431640625, -2.9493408203125, 105.29908752441406, 32.610748291015625, 6.72528076171875, 23.115631103515625, 64.9469223022461, 31.83653450012207, 74.39694213867188, -23.69757080078125, 1.2912673950195312, -7.227508544921875, 23.773040771484375, 102.97528076171875, 139.66387939453125, 119.07846069335938, -67.45867919921875, -3.8350830078125, -9.58953857421875, -30.5079345703125, 84.21939086914062, 0.7996673583984375, 58.116912841796875, 27.918853759765625, -1.8353614807128906, 6.41229248046875, 36.2130126953125, 89.08759307861328, -57.90478515625, -96.84262084960938, -4.251003265380859, -44.403350830078125, -1.013275146484375, -25.86505126953125, 12.605209350585938, 70.48171997070312, 6.4853515625, 130.1165771484375, 4.31805419921875, 73.63955688476562, 146.0135955810547, 0.97845458984375, 127.04586791992188, 16.770721435546875, -21.8841552734375, -24.51153564453125, 19.05810546875, -6.772558212280273, 112.26495361328125, 93.150390625, -11.545326232910156, 145.77096557617188, -23.856109619140625, 8.738883972167969, 3.756776809692383, -21.206771850585938, 7.799560546875, 103.01058959960938, 10.156234741210938, 79.72718811035156, -10.178548812866211, 9.841773986816406, -0.16509246826171875, 107.0809326171875, 51.840911865234375, 10.245025634765625, 110.13943481445312, 67.72967529296875, -14.684066772460938, 56.08335494995117, -35.009674072265625, 6.872779846191406, -1.328704833984375, 118.05317687988281, 31.81634521484375, 106.06551361083984, 29.268081665039062, 90.5379638671875, 3.560619354248047, -6.80413818359375, 11.898590087890625, -9.864410400390625, 33.13471984863281, -146.32327270507812, 79.5413818359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000348.npy"}
|
|
{"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 41.677852630615234, "std": 60.252410888671875, "min": -103.28781127929688, "p10": -30.469946289062495, "median": 33.360870361328125, "p90": 120.40115051269531, "max": 171.14813232421875, "pos_frac": 0.7734375, "sample": [72.96549987792969, 103.85865783691406, 1.396270751953125, -103.28781127929688, -69.11447143554688, 1.7658271789550781, 171.14813232421875, 71.39962768554688, -60.002166748046875, 49.9613037109375, 36.8516845703125, -55.15351104736328, 85.27601623535156, 126.4503173828125, 52.72625732421875, 107.92532348632812, 132.61981201171875, 5.6948394775390625, 76.85545349121094, 132.8486328125, 61.0948486328125, 4.73968505859375, -23.3973388671875, -23.02459716796875, -4.7967529296875, 22.817047119140625, -39.71826171875, -4.20770263671875, 136.6341552734375, 116.55264282226562, 43.42588806152344, 96.48634338378906, 95.9617919921875, -39.9383544921875, 30.482955932617188, 35.78020477294922, 137.68785095214844, 1.90191650390625, 15.423492431640625, 44.042083740234375, 12.642234802246094, -1.6539726257324219, 95.22625732421875, -0.763214111328125, 13.406051635742188, 50.07501220703125, -10.679229736328125, 162.57913208007812, 38.53118896484375, 7.9039764404296875, -46.904266357421875, -2.0938282012939453, -32.897216796875, 113.62628173828125, 107.802978515625, 50.48985290527344, 5.379203796386719, 17.217803955078125, -73.45339965820312, 52.9112548828125, 0.0, 151.19334411621094, -4.9210205078125, 54.833740234375, 99.32086181640625, 24.46337890625, 106.36479187011719, 120.4197998046875, 101.01571655273438, -3.154935836791992, 112.76673889160156, -29.4296875, -77.34071350097656, 163.4588623046875, 0.69659423828125, 33.440521240234375, -0.870452880859375, 88.49152374267578, 5.9479827880859375, 33.24229431152344, 18.68450927734375, 78.40189361572266, 120.61602783203125, 81.8419189453125, 132.20538330078125, 8.9310302734375, 46.37828063964844, 16.428386688232422, 109.25616455078125, 37.5645751953125, 46.802001953125, 33.281219482421875, 108.59574890136719, 6.319591522216797, -22.308822631835938, 5.8312835693359375, 108.46966552734375, 72.24707794189453, 21.385101318359375, -84.59449768066406, 27.89495849609375, 31.750877380371094, 36.76838684082031, 3.456451416015625, -98.139892578125, 14.646484375, 0.948333740234375, 5.333225250244141, 16.539276123046875, 101.5394287109375, 14.89044189453125, 103.66180419921875, 27.496429443359375, 94.26438903808594, 107.40769958496094, 120.39315795898438, 113.42291259765625, -14.34442138671875, 69.19813537597656, 139.6734619140625, 9.159942626953125, 35.69319152832031, 91.62094116210938, 89.71221923828125, -53.63970947265625, -1.1318359375, 12.259872436523438, 98.53336334228516], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000349.npy"}
|
|
{"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 37.28400421142578, "std": 62.04899597167969, "min": -116.54000854492188, "p10": -46.83425750732421, "median": 34.05951690673828, "p90": 121.82469329833984, "max": 154.09661865234375, "pos_frac": 0.78125, "sample": [119.76620483398438, 27.94463348388672, -0.1806488037109375, -1.1956939697265625, 9.060726165771484, 45.36146545410156, 15.730667114257812, 142.73529052734375, 3.925863265991211, 89.39942932128906, 30.38677978515625, 13.151641845703125, 110.59881591796875, -4.8864593505859375, 23.596710205078125, 117.60516357421875, 111.82333374023438, 121.53677368164062, 14.046211242675781, -39.30388641357422, 109.68965148925781, 13.883743286132812, 5.073822021484375, 83.85737609863281, 75.37638092041016, 100.00518798828125, 39.75926208496094, 39.818817138671875, 151.67462158203125, 131.8167724609375, 70.83224487304688, 50.893218994140625, 16.225650787353516, 5.346668243408203, -39.47337341308594, 86.23323822021484, 12.108879089355469, 12.041412353515625, 52.797821044921875, 31.59832763671875, 18.53289794921875, 66.94400024414062, 74.4412841796875, -92.06607055664062, 112.204833984375, -8.54705810546875, 6.24822998046875, 38.85813903808594, -116.54000854492188, 10.579086303710938, 36.79216003417969, 94.992431640625, 151.33126831054688, 71.09603881835938, 22.212982177734375, 37.80540466308594, 123.59249877929688, 100.16574096679688, 45.965484619140625, 144.58428955078125, 76.40898132324219, -39.94207763671875, 122.49650573730469, 7.134307861328125, -9.571563720703125, -44.32258605957031, 35.74540710449219, 35.74934387207031, -88.80841064453125, 7.452524185180664, 86.39771270751953, 52.363037109375, 2.1737442016601562, -62.400482177734375, -10.245819091796875, 92.70184326171875, -8.352821350097656, 23.820205688476562, 21.250762939453125, 52.261474609375, 137.09906005859375, 107.75198364257812, 66.7562484741211, 33.10554504394531, 26.072113037109375, 100.8179702758789, 17.8212890625, 129.52069091796875, -62.119537353515625, 6.200897216796875, 135.57606506347656, 34.41816711425781, 19.794097900390625, 33.70086669921875, 102.88592529296875, 77.09280395507812, 24.066497802734375, 104.14815521240234, 154.09661865234375, 105.44622802734375, 102.94943237304688, -61.4246826171875, 46.213836669921875, 36.94251251220703, -102.38031005859375, -74.10713958740234, -52.69482421875, 11.431350708007812, -89.2962646484375, 36.971221923828125, -103.736572265625, -62.90576171875, 12.133392333984375, 123.42672729492188, 48.20527648925781, 36.668701171875, 4.169586181640625, 14.673355102539062, 126.05474853515625, -70.16204833984375, -3.1751708984375, 40.717411041259766, -13.644287109375, -10.1192626953125, 118.9725341796875, -34.98271942138672, 40.504913330078125, 4.526145935058594], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000350.npy"}
|
|
{"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 27.6404972076416, "std": 62.90945816040039, "min": -151.80776977539062, "p10": -34.400761413574216, "median": 14.540878295898438, "p90": 117.58723831176758, "max": 174.09918212890625, "pos_frac": 0.6484375, "sample": [0.0, 100.903076171875, 9.529205322265625, 147.78756713867188, -1.1192779541015625, 137.35470581054688, 14.535934448242188, -38.30724334716797, 48.249359130859375, -2.133636474609375, 29.428939819335938, -31.547454833984375, 11.5341796875, 14.545822143554688, 112.42455291748047, -70.25202941894531, 27.012340545654297, 77.42256927490234, -33.36994171142578, 17.99664306640625, 22.643600463867188, 96.94009399414062, -10.865936279296875, 150.82354736328125, 81.11219787597656, 117.14415740966797, 7.231760025024414, -13.572601318359375, 8.335968017578125, 3.7023468017578125, 120.85208129882812, 132.93788146972656, 35.867218017578125, 20.730518341064453, -5.8515167236328125, -44.2244873046875, -1.3219966888427734, 99.658447265625, -25.18769073486328, -38.81973648071289, -70.3338623046875, 86.54830932617188, 9.301651000976562, 45.289306640625, -27.656982421875, 88.58842468261719, 7.273689270019531, 19.1693115234375, -19.641357421875, 19.02667236328125, -29.95355224609375, -21.205535888671875, 15.980224609375, -6.69744873046875, 6.59814453125, -1.627044677734375, 108.72225952148438, -19.0762939453125, -9.421905517578125, 17.169769287109375, 125.03904724121094, 160.46694946289062, 17.190114974975586, 154.038330078125, 102.60820007324219, -17.742218017578125, 68.87472534179688, 9.79510498046875, 99.698486328125, -59.66943359375, 0.2187957763671875, 89.90874481201172, -22.493682861328125, 7.67340087890625, 124.8529281616211, -133.60012817382812, 33.665802001953125, -8.56884765625, 30.466278076171875, 29.28582763671875, 0.0, -6.96435546875, 20.638427734375, 113.61483764648438, 102.96528625488281, -0.7133512496948242, 1.1226043701171875, 21.950538635253906, 85.0714111328125, -8.514251708984375, 28.1732177734375, 5.2693328857421875, 70.97476196289062, -11.621065139770508, 0.0, 118.62109375, 121.91958618164062, -5.05517578125, 19.558856964111328, 13.9114990234375, 6.4669036865234375, -151.80776977539062, 83.5899887084961, 107.45890808105469, 68.362060546875, -11.119888305664062, 174.09918212890625, 25.113983154296875, 113.26052856445312, -2.9947891235351562, 0.9494152069091797, 13.271133422851562, -10.97332763671875, 30.895034790039062, 8.235818862915039, 110.70223999023438, 17.215469360351562, -42.5927734375, -103.27334594726562, -66.392822265625, 21.153854370117188, 62.350738525390625, -128.5826416015625, 32.758628845214844, -30.745407104492188, -36.806007385253906, 168.50152587890625, 28.070465087890625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000351.npy"}
|
|
{"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 41.51838302612305, "std": 60.199440002441406, "min": -103.5390625, "p10": -31.351498413085938, "median": 39.055572509765625, "p90": 124.69747924804686, "max": 150.22991943359375, "pos_frac": 0.75, "sample": [87.61215209960938, 20.6485595703125, 127.833740234375, -41.027008056640625, 9.414825439453125, 144.9410400390625, 74.27140045166016, 67.22557067871094, 80.07125854492188, 1.6162490844726562, -28.93609619140625, 95.01419067382812, -93.55825805664062, 19.535675048828125, 107.37391662597656, 5.032253265380859, 28.3297119140625, 129.37762451171875, 106.75729370117188, 75.80996704101562, 150.22991943359375, -34.837799072265625, -31.897369384765625, 129.22628784179688, 39.639892578125, -103.5390625, -15.182098388671875, 59.01026916503906, 18.4034423828125, 85.27787017822266, -1.1631126403808594, 9.344345092773438, 126.62457275390625, 14.495071411132812, 98.02786254882812, -65.08501434326172, 10.547019958496094, 65.74853515625, -9.955764770507812, 2.00311279296875, 52.35380554199219, -90.76043701171875, -18.387939453125, 67.47744750976562, -22.487991333007812, 25.233016967773438, 0.0, 64.917724609375, 36.7813720703125, 123.87158203125, 4.2103271484375, 138.3323211669922, 88.29595947265625, 26.76226806640625, 110.5559310913086, 63.30523681640625, 13.858016967773438, 0.0, 53.44671630859375, 72.67822265625, 76.85984802246094, 0.495635986328125, 18.440467834472656, 71.14666748046875, -95.11557006835938, -23.43988037109375, 11.911766052246094, 38.47125244140625, -57.1051025390625, 8.485916137695312, 74.49188232421875, 138.87112426757812, 54.793853759765625, 29.335477828979492, 10.451725006103516, -86.65243530273438, 133.29391479492188, 43.16904830932617, -71.5606689453125, -57.368865966796875, 142.49781799316406, 127.33802795410156, 123.00303649902344, 18.012123107910156, -6.001739501953125, 104.94183349609375, 11.970684051513672, -31.1175537109375, -0.6890106201171875, 92.22640991210938, 28.6322021484375, -7.272430419921875, 46.185791015625, 26.087299346923828, 43.40205383300781, 43.059967041015625, 104.44854736328125, 96.18915557861328, -6.16162109375, 137.38209533691406, 44.64579772949219, 80.79231262207031, 95.11262512207031, 59.222442626953125, 104.89451599121094, -19.326171875, 96.23330688476562, 103.57080078125, 27.648338317871094, -5.679740905761719, 26.583831787109375, 86.39935302734375, 117.93768310546875, 2.885711669921875, -29.033462524414062, -17.563491821289062, 27.33597183227539, 116.04776000976562, 96.71627044677734, 133.99635314941406, -23.038436889648438, -35.9830322265625, 101.83174133300781, 94.64727783203125, 44.06110382080078, 71.41155242919922, 29.19387435913086, 95.99267578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000352.npy"}
|
|
{"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 41.240718841552734, "std": 58.81978225708008, "min": -120.28121948242188, "p10": -23.99073181152343, "median": 33.52961730957031, "p90": 114.22848205566406, "max": 180.63433837890625, "pos_frac": 0.765625, "sample": [24.508071899414062, 103.74859619140625, 17.571571350097656, 2.162261962890625, 22.401702880859375, 125.29534912109375, -76.40482330322266, 104.23681640625, 65.23202514648438, 14.2662353515625, 36.7510986328125, 14.815444946289062, 5.799835205078125, 68.0057373046875, -8.190673828125, 12.4951171875, -14.937347412109375, -21.6531982421875, 38.237823486328125, -10.74761962890625, 34.33056640625, 75.26643371582031, 2.5366439819335938, -120.28121948242188, 102.30223083496094, -5.10418701171875, 64.70230102539062, 72.54562377929688, 174.26910400390625, -1.7154159545898438, 122.747802734375, 94.68378448486328, 106.787353515625, -72.75315856933594, 103.04669189453125, -57.78448486328125, 138.81307983398438, 81.7674560546875, 62.296142578125, 180.63433837890625, 13.417037963867188, -62.79046630859375, 32.728668212890625, 116.19985961914062, 20.488754272460938, -2.79473876953125, 113.38360595703125, 168.92755126953125, 145.44198608398438, 89.7369384765625, 98.96763610839844, 52.37147521972656, -46.461082458496094, 79.54083251953125, 12.898544311523438, 155.65292358398438, 6.070003509521484, -8.289253234863281, 89.49777221679688, 41.668853759765625, 140.33143615722656, 51.90380859375, -33.97126770019531, -68.92034912109375, 3.1863555908203125, 49.467559814453125, -83.71796417236328, -13.20123291015625, 42.062255859375, 24.940109252929688, 64.08053588867188, 75.91905212402344, 92.03382873535156, 104.79073333740234, 103.81474304199219, 92.14128112792969, 14.028190612792969, 98.27719116210938, 52.30522155761719, 8.661598205566406, 99.9078369140625, 13.915397644042969, 19.340469360351562, 19.817474365234375, 44.316436767578125, 116.21527099609375, 32.18769836425781, 94.25547790527344, 87.75823974609375, 3.7512245178222656, -3.8486576080322266, -32.49639892578125, 94.110595703125, 49.618865966796875, 10.554832458496094, -18.30621337890625, -3.3734874725341797, 29.786102294921875, 15.960777282714844, -10.2205810546875, 63.129638671875, 5.8460845947265625, 26.2421875, 4.884052276611328, 117.94712829589844, 37.279510498046875, 65.76815795898438, 50.716064453125, 103.84541320800781, -33.77264404296875, 24.31940460205078, -0.09912872314453125, 23.981342315673828, 54.56414794921875, 5.505571365356445, 0.0, 146.33294677734375, 55.137451171875, -29.444976806640625, -6.83380126953125, 3.03082275390625, -85.48049926757812, -10.00030517578125, 16.43621826171875, 111.98298645019531, 69.06768798828125, 96.47473907470703, 81.22515869140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000353.npy"}
|
|
{"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 24.66413116455078, "std": 63.13558578491211, "min": -127.28582763671875, "p10": -52.48608856201171, "median": 13.467658996582031, "p90": 111.47233886718749, "max": 156.7537841796875, "pos_frac": 0.640625, "sample": [3.3677711486816406, -38.11470031738281, -103.34242248535156, 109.19284057617188, 39.56419372558594, 153.270751953125, 20.382537841796875, 10.990829467773438, 10.511627197265625, 1.8377113342285156, 30.23162841796875, 37.70050048828125, -58.66279602050781, 13.823989868164062, 113.99057006835938, -6.769981384277344, -14.280128479003906, -0.1246337890625, 1.364715576171875, -0.85968017578125, 74.26327514648438, 70.88212585449219, 82.37387084960938, -81.381103515625, 22.669227600097656, 15.2138671875, -7.9786529541015625, 1.8488311767578125, 103.98837280273438, 151.46875, -33.86116027832031, -1.20465087890625, 0.4764385223388672, 47.6336669921875, 132.69711303710938, 19.4249267578125, 11.85501480102539, -1.582305908203125, 110.39309692382812, 102.6138687133789, 2.2911376953125, -47.95556640625, 47.02473449707031, -19.64215087890625, 98.88539123535156, 80.58882904052734, -2.483154296875, 6.737091064453125, -106.09324645996094, 5.1352081298828125, 13.111328125, 150.35662841796875, 44.679840087890625, 128.66741943359375, 127.31063842773438, 124.0819091796875, 92.77186584472656, 51.972320556640625, 43.2650146484375, 49.285980224609375, 30.635162353515625, -11.692977905273438, -14.486747741699219, -44.883331298828125, 5.538368225097656, 66.06108093261719, 12.151412963867188, 156.7537841796875, -4.0723876953125, -2.8638687133789062, 116.1981201171875, -49.325950622558594, 89.19419860839844, 6.635004043579102, 151.39306640625, 45.216064453125, 87.75330352783203, -81.7442626953125, -9.165374755859375, 0.7578372955322266, -39.89959716796875, -49.83892822265625, 120.53182983398438, 36.06494140625, -17.068206787109375, 40.72998046875, 0.0789337158203125, 104.7235336303711, -22.604827880859375, -61.42539978027344, 22.29034423828125, -14.154808044433594, 107.783935546875, 6.4454345703125, 58.689605712890625, 26.225296020507812, -8.695571899414062, -17.622299194335938, -76.87252807617188, -84.42245483398438, 48.17216491699219, -4.638763427734375, -9.119674682617188, 76.184326171875, 82.94979858398438, 45.495269775390625, -0.88812255859375, -4.80706787109375, 19.356475830078125, 139.9708251953125, 97.89144897460938, 55.23333740234375, -49.279380798339844, 105.949462890625, -88.08341979980469, 63.44476318359375, -127.28582763671875, 73.12159729003906, -100.27069091796875, -40.6917724609375, 68.52497863769531, 24.521665573120117, -81.56723022460938, 29.78802490234375, -5.14923095703125, -70.33815002441406, 76.91909790039062, 14.73223876953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000354.npy"}
|
|
{"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 42.04008483886719, "std": 60.914512634277344, "min": -116.51104736328125, "p10": -21.203376770019524, "median": 32.06433868408203, "p90": 119.64524841308592, "max": 157.24563598632812, "pos_frac": 0.7734375, "sample": [-1.0153312683105469, -116.51104736328125, 84.70550537109375, 55.033416748046875, 3.524932861328125, 9.54776382446289, 45.64306640625, 4.52001953125, 65.94012451171875, 25.0777587890625, 118.08894348144531, 9.02419662475586, -13.590240478515625, 93.01054382324219, 1.4873943328857422, 110.39411926269531, 129.68756103515625, 113.35847473144531, -3.4717788696289062, -11.98187255859375, -59.352081298828125, 12.7259521484375, 32.999786376953125, 136.54595947265625, 9.93280029296875, 123.27662658691406, -10.841636657714844, -4.89813232421875, 32.30555725097656, 40.812957763671875, 21.810089111328125, 50.984832763671875, 65.00613403320312, 141.1864013671875, 134.50799560546875, 39.13044738769531, 144.65109252929688, 28.783306121826172, 9.147705078125, -42.85699462890625, 82.33072662353516, 108.17521667480469, 61.928924560546875, 71.17292785644531, -50.54161071777344, 9.02691650390625, -5.4346923828125, 28.81098175048828, -1.7757492065429688, 25.86785888671875, 102.79861450195312, 90.65403747558594, 79.0782470703125, 11.791839599609375, 0.729736328125, 84.396728515625, 1.8021354675292969, 33.452789306640625, 25.0159912109375, -17.67782211303711, 123.66110229492188, 8.894876480102539, -25.77239990234375, -13.879043579101562, 89.09104919433594, -112.5718994140625, 20.803543090820312, 157.24563598632812, 78.67489624023438, 100.18678283691406, 2.8744144439697266, 117.6982421875, -113.025634765625, 108.67889404296875, 1.3697700500488281, 32.3514404296875, 101.07025146484375, 44.49293518066406, 139.35379028320312, 72.3497314453125, 70.39547729492188, 105.603271484375, -1.434051513671875, 4.078338623046875, 24.542266845703125, -50.1817626953125, 31.8231201171875, -39.92431640625, 25.055068969726562, 28.534942626953125, 98.24356079101562, -16.297882080078125, -104.36166381835938, 0.8502197265625, -7.661651611328125, 34.3739013671875, 22.955230712890625, 130.80287170410156, 4.24932861328125, 116.9459228515625, 38.874725341796875, -46.002777099609375, 149.56243896484375, 84.90220642089844, 133.84341430664062, 106.09197235107422, 118.07484436035156, 80.92153930664062, -51.87451171875, -9.363861083984375, 29.015045166015625, 30.229354858398438, -19.245223999023438, 111.50741577148438, 88.2315673828125, 91.2564926147461, 111.19613647460938, 92.16087341308594, 101.699462890625, 82.76496887207031, 13.189605712890625, -0.54827880859375, 20.536895751953125, 129.32223510742188, 54.41082763671875, -85.19650268554688, 115.82353210449219, 27.670257568359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000355.npy"}
|
|
{"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 33.856292724609375, "std": 54.37367630004883, "min": -125.5513916015625, "p10": -28.629641723632812, "median": 22.781814575195312, "p90": 108.10395660400391, "max": 135.03018188476562, "pos_frac": 0.703125, "sample": [1.150146484375, -2.40863037109375, 108.23526000976562, -4.57086181640625, 84.85104370117188, -33.9718017578125, 17.551605224609375, -26.441452026367188, 2.4607677459716797, 6.8373870849609375, -5.681732177734375, 10.131179809570312, 22.64459228515625, 74.12106323242188, -7.46733283996582, 1.8836040496826172, 38.726219177246094, -35.609710693359375, -25.9527587890625, 27.55908203125, -35.670440673828125, 106.65034484863281, 38.95136260986328, 31.9873046875, -0.23721981048583984, 134.2701416015625, -75.4698486328125, 93.95361328125, 3.708038330078125, 70.4334716796875, 28.367340087890625, 68.008056640625, 13.10247802734375, 73.8041763305664, 87.05093383789062, 108.95481872558594, 72.91500854492188, 77.2193603515625, 131.78585815429688, 81.78535461425781, -62.43785095214844, 103.68362426757812, 2.567779541015625, 5.6687469482421875, 21.400062561035156, 22.13629150390625, -5.497955322265625, 132.60958862304688, 69.05462646484375, -4.935157775878906, 19.18323516845703, 110.28456115722656, 114.72021484375, 121.31158447265625, -0.49362945556640625, 30.98065185546875, -61.06675720214844, 101.58779907226562, 80.03401184082031, 38.519683837890625, 100.99786376953125, -96.71980285644531, 88.15350341796875, 77.4190673828125, 91.11817932128906, 98.88431549072266, 62.420623779296875, -6.97125244140625, 1.5583877563476562, 2.1967010498046875, -13.766098022460938, 20.368148803710938, 81.89566040039062, -19.194007873535156, 5.027099609375, 64.23465728759766, 77.88087463378906, 0.3597373962402344, 108.04768371582031, -29.790359497070312, 35.21592712402344, -6.480922698974609, 135.03018188476562, 30.270904541015625, -29.61993408203125, 84.73147583007812, 95.59942626953125, 82.90409851074219, 28.561729431152344, -1.7922172546386719, -57.820396423339844, -3.3999671936035156, -6.003143310546875, -125.5513916015625, 25.385955810546875, 13.4931640625, 38.41455078125, 0.0, 30.884490966796875, 22.919036865234375, 65.75918579101562, 16.488670349121094, -7.268390655517578, -17.6197509765625, 0.5908660888671875, 15.6663818359375, 19.53757095336914, 81.59756469726562, -2.1294097900390625, -60.94355773925781, 108.92002868652344, 133.09559631347656, 101.70974731445312, 80.83447265625, -14.173606872558594, 113.23684692382812, 92.89437866210938, 1.0150299072265625, 48.5086669921875, 97.15774536132812, -28.205230712890625, 70.79194641113281, -0.8209381103515625, -55.939544677734375, 46.63365173339844, -15.839042663574219, 20.946487426757812, 109.36724090576172], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000356.npy"}
|
|
{"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 37.787376403808594, "std": 61.59958267211914, "min": -130.9688720703125, "p10": -28.26114807128906, "median": 23.497102737426758, "p90": 124.7119400024414, "max": 186.41448974609375, "pos_frac": 0.734375, "sample": [55.244415283203125, 112.02557373046875, -10.19143295288086, -1.14923095703125, 113.7265625, 37.77294921875, 0.0, -2.846282958984375, 57.36705017089844, 4.3619384765625, 99.32450866699219, -86.61065673828125, 0.029478073120117188, 186.41448974609375, 149.05108642578125, 152.162841796875, 26.515453338623047, 2.83404541015625, 69.70576477050781, 20.501708984375, 12.432479858398438, 8.55963134765625, 53.01417541503906, 111.78079223632812, 126.25663757324219, -62.367462158203125, 168.011962890625, 19.79686737060547, 24.40237808227539, 138.78363037109375, -27.144927978515625, 11.875762939453125, -17.869762420654297, 7.611026763916016, 160.9005126953125, 30.135761260986328, -14.370941162109375, -48.89734649658203, 3.08221435546875, 66.97232055664062, 80.84765625, 153.90228271484375, 22.591827392578125, 78.110107421875, 39.21775817871094, -33.009613037109375, 4.3861083984375, -78.83612060546875, 119.1099853515625, 16.555206298828125, -19.491714477539062, 29.64795684814453, 3.2530670166015625, 40.45526123046875, 72.50926208496094, 85.58674621582031, 130.77667236328125, 108.48684692382812, 29.38214111328125, -15.007659912109375, 105.36581420898438, 24.8623046875, 19.14959716796875, -21.853057861328125, 84.60345458984375, -7.54815673828125, 105.28472900390625, 99.1986083984375, 37.83636474609375, 68.78018188476562, 33.669189453125, -28.622833251953125, -24.239364624023438, 4.52325439453125, 2.14056396484375, 105.94088745117188, 61.46778869628906, -50.594329833984375, 16.504806518554688, -31.773773193359375, -44.26795959472656, 11.731094360351562, 14.091171264648438, 19.2257080078125, 131.37332153320312, 24.553436279296875, 116.05792236328125, -28.10614013671875, 3.1558990478515625, 0.1689300537109375, -2.5254135131835938, 17.6341552734375, 37.914947509765625, -38.71693420410156, 39.625701904296875, 101.52861022949219, 0.155426025390625, -72.85294342041016, 0.0, 127.18475341796875, -95.07658386230469, 0.0, 34.779388427734375, -13.847869873046875, 124.0499267578125, 7.788425445556641, 18.14429473876953, 93.60365295410156, 116.91848754882812, 68.46774291992188, 137.76486206054688, -130.9688720703125, 115.52957153320312, 56.8057861328125, 2.5628719329833984, 68.74232482910156, -5.6685791015625, 92.60453796386719, 69.04627990722656, 103.15283203125, 144.04501342773438, 17.003276824951172, -7.178985595703125, -18.0789794921875, 57.3245849609375, -1.0531082153320312, 1.5487518310546875, 88.46749877929688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000357.npy"}
|
|
{"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 42.124786376953125, "std": 63.81361389160156, "min": -121.979736328125, "p10": -25.864596557617187, "median": 31.7124605178833, "p90": 123.67252273559569, "max": 160.00836181640625, "pos_frac": 0.7734375, "sample": [0.267974853515625, 2.435108184814453, 68.61196899414062, -45.6630859375, 0.314727783203125, 31.931955337524414, -22.791290283203125, 84.17214965820312, 133.52777099609375, 160.00836181640625, 6.196006774902344, 2.478076934814453, 40.42903137207031, 110.92646789550781, 126.0804443359375, 111.5278091430664, 20.912460327148438, 121.95236206054688, 126.82127380371094, 33.2623291015625, -99.84786987304688, 102.6201400756836, -20.758010864257812, -83.14710998535156, 102.72036743164062, 31.492965698242188, 101.82357788085938, -2.73260498046875, 129.70022583007812, -9.90618896484375, 5.8139801025390625, 61.89784240722656, 12.722801208496094, 91.68365478515625, 47.05999755859375, 26.13275146484375, 154.61514282226562, -61.085906982421875, 30.8983154296875, 92.22410583496094, -81.17768859863281, 83.63626098632812, 113.50981140136719, 131.33877563476562, 30.16608428955078, 114.32382202148438, 88.63760375976562, 113.61360168457031, 118.27570343017578, 6.5665283203125, -25.688751220703125, -8.078453063964844, -5.235282897949219, 20.502517700195312, 80.58918762207031, -13.427566528320312, 112.46041870117188, 117.52163696289062, 12.567626953125, 151.56524658203125, -22.870513916015625, -121.979736328125, 138.6453857421875, -13.425613403320312, 6.273262023925781, 118.59359741210938, 66.55364990234375, 20.56831169128418, 2.210784912109375, 9.89971923828125, 32.216468811035156, 12.137908935546875, 12.990066528320312, 16.966873168945312, 3.5228271484375, -93.33877563476562, 103.712158203125, 77.20401000976562, -10.511283874511719, 139.74685668945312, 34.367027282714844, -54.802825927734375, 95.24195861816406, 46.2935791015625, 122.64055633544922, 28.13897705078125, 81.51507568359375, 33.6279296875, 27.022579193115234, 119.98126220703125, 8.5369873046875, 134.79928588867188, 100.21124267578125, 80.55360412597656, 74.71273803710938, 108.33432006835938, -26.27490234375, 11.184356689453125, 6.13677978515625, 62.44131851196289, 150.1248779296875, 6.612213134765625, 57.14912414550781, 127.10787200927734, 59.740478515625, 120.16537475585938, 114.1678466796875, -111.44921875, -13.421600341796875, -90.01101684570312, -60.24827194213867, 6.145408630371094, 93.78704833984375, 7.44281005859375, 51.343292236328125, -23.205772399902344, 8.794624328613281, 12.899101257324219, -1.06597900390625, 26.46722412109375, 0.0, 46.69793701171875, 78.3428955078125, 38.404815673828125, 109.70581817626953, -33.02130126953125, 27.97564697265625, -1.45361328125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000358.npy"}
|
|
{"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 39.27888488769531, "std": 56.2483024597168, "min": -133.08685302734375, "p10": -22.827980041503906, "median": 31.683910369873047, "p90": 109.35556335449218, "max": 159.85498046875, "pos_frac": 0.7734375, "sample": [53.277381896972656, -45.4930419921875, 98.77305603027344, 130.98489379882812, 91.43707275390625, -22.2027587890625, 56.41956329345703, 13.281547546386719, 30.94677734375, 37.291900634765625, 26.41508674621582, 62.61639404296875, 25.12961196899414, 17.707889556884766, -2.40460205078125, -18.943878173828125, 82.3050308227539, 30.725311279296875, 18.85565185546875, 18.506637573242188, 79.17925262451172, 5.9034271240234375, 107.8006591796875, 130.99505615234375, -26.588104248046875, 99.44840240478516, 6.43084716796875, 30.54315185546875, 65.207763671875, -130.1231689453125, -2.558349609375, 154.3201904296875, 6.012786865234375, 34.197364807128906, 3.7581424713134766, 55.483612060546875, -7.037322998046875, 100.23251342773438, 116.42922973632812, 92.74481201171875, -21.469863891601562, -84.35893249511719, 103.80633544921875, 3.4482498168945312, 95.66667175292969, 72.79354858398438, 43.635589599609375, 58.4744873046875, 94.49674987792969, 87.38386535644531, 65.38674926757812, 2.66534423828125, 159.85498046875, -22.623794555664062, 70.45257568359375, 0.3313140869140625, 46.283355712890625, -124.10430908203125, 94.58399200439453, 38.2513427734375, 143.50079345703125, 62.925811767578125, -133.08685302734375, 66.46182250976562, -0.946990966796875, 0.01564788818359375, -7.33599853515625, 42.030181884765625, 90.62066650390625, 78.14508056640625, -4.61981201171875, 27.460052490234375, -31.12476348876953, 32.700714111328125, 67.2004165649414, 65.30184936523438, -39.481842041015625, 92.998291015625, 12.108741760253906, 104.62373352050781, 54.1156005859375, -3.3473358154296875, 7.087677001953125, 57.90032958984375, 37.49919128417969, 11.057044982910156, 13.505645751953125, 97.43698120117188, 25.790252685546875, 17.027694702148438, 119.80799865722656, 13.555709838867188, 70.03094482421875, 93.09831237792969, 24.200927734375, -24.844772338867188, -21.638931274414062, 77.35321044921875, -27.8363037109375, -0.7675685882568359, 25.961395263671875, 125.64218139648438, 31.0889892578125, -35.85723876953125, 17.609756469726562, -23.304412841796875, 125.22085571289062, -15.7816162109375, 47.757110595703125, 112.98367309570312, 96.86515808105469, 32.278831481933594, 12.564422607421875, 29.995363235473633, -11.34429931640625, 133.86480712890625, 94.3629150390625, 54.549713134765625, 80.62020874023438, 0.7204208374023438, -5.962547302246094, 142.02005004882812, 21.459152221679688, 7.729902267456055, 130.1077880859375, 13.828048706054688, 87.66273498535156, -28.444305419921875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000359.npy"}
|
|
{"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 34.79438018798828, "std": 54.755523681640625, "min": -155.88037109375, "p10": -17.881803894042967, "median": 18.8167724609375, "p90": 112.02478027343749, "max": 169.50872802734375, "pos_frac": 0.7578125, "sample": [70.6490478515625, 75.53553771972656, -33.582794189453125, 118.21708679199219, -100.48182678222656, -7.347072601318359, -77.26754760742188, 49.0506591796875, 169.50872802734375, -37.52363586425781, 31.07391357421875, 59.92889404296875, -17.663436889648438, 91.99041748046875, 48.8709716796875, 68.36776733398438, 16.039337158203125, 91.0787353515625, 17.822509765625, 38.33094787597656, 122.68377685546875, 5.4075164794921875, 13.0372314453125, 87.41082763671875, 90.31879425048828, -3.0757293701171875, 4.482460021972656, 19.54376220703125, -11.156524658203125, 115.53045654296875, 30.724639892578125, 136.44606018066406, 49.600433349609375, 17.736785888671875, 99.24508666992188, 4.51275634765625, -7.09161376953125, 97.38168334960938, -29.08935546875, 47.35881042480469, 96.8697509765625, 60.1015625, 64.54501342773438, 45.415679931640625, 32.876708984375, 16.099712371826172, 13.978973388671875, 101.34944152832031, -6.699493408203125, 32.717132568359375, -10.988296508789062, 99.28401184082031, 57.826446533203125, 61.892547607421875, 141.51119995117188, -7.7650604248046875, 124.32841491699219, -20.3343505859375, 109.02841186523438, 6.361305236816406, 25.00958251953125, 5.2974853515625, 110.673828125, 35.98919677734375, 18.08978271484375, 2.7413482666015625, 122.24008178710938, 13.6007080078125, 61.49281311035156, -92.62747192382812, 6.65130615234375, 89.01226806640625, 1.3088130950927734, 95.79655456542969, -12.0810546875, 20.43814468383789, 56.47937774658203, 127.03778076171875, 108.30764770507812, 100.67439270019531, -10.599258422851562, 86.41387939453125, 6.979938507080078, 0.6618518829345703, -155.88037109375, 83.97550964355469, 42.83294677734375, 14.802764892578125, -18.391326904296875, 10.338859558105469, 115.177001953125, -18.552276611328125, 130.15798950195312, 14.03070068359375, -0.2733001708984375, 17.22149658203125, 4.277130126953125, -79.75833129882812, 8.447174072265625, 6.926761627197266, 87.90553283691406, 0.0, 8.994598388671875, 15.449905395507812, 116.71600341796875, -0.9813823699951172, 59.6248779296875, 35.4842529296875, 4.627899169921875, 4.694047927856445, 11.402679443359375, -6.18841552734375, -13.276039123535156, 88.41567993164062, 8.953567504882812, 42.162864685058594, -25.81175994873047, -11.34405517578125, 0.0, 9.100700378417969, 25.24273681640625, 26.781108856201172, 29.265853881835938, -31.321189880371094, -11.370830535888672, 16.576614379882812, 121.61799621582031, 104.03070068359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000360.npy"}
|
|
{"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 46.260047912597656, "std": 61.547786712646484, "min": -122.24407958984375, "p10": -23.96061096191406, "median": 47.13816452026367, "p90": 125.43896179199218, "max": 174.86676025390625, "pos_frac": 0.7265625, "sample": [59.280250549316406, -8.720123291015625, -68.22697448730469, 31.2403564453125, 27.31591796875, -7.290130615234375, 0.324310302734375, -21.256072998046875, 104.13101196289062, 39.357452392578125, 80.50971221923828, 12.162628173828125, 8.14407730102539, 132.77597045898438, -2.1056861877441406, 116.86573791503906, -3.581270217895508, 11.981468200683594, -23.130401611328125, 2.9203033447265625, 107.4796142578125, 140.19424438476562, -33.567291259765625, 111.92274475097656, -4.389198303222656, 136.02752685546875, 88.61814880371094, 97.31654357910156, 21.380630493164062, 148.72940063476562, 11.3365478515625, -8.127212524414062, 74.84803009033203, 47.80310821533203, 12.402427673339844, 105.7418212890625, -69.13900756835938, 79.90545654296875, 89.9561538696289, 26.06085205078125, 137.36793518066406, 124.42672729492188, 22.242507934570312, 62.863529205322266, 155.3048095703125, 25.765350341796875, -73.34649658203125, -9.863006591796875, 18.9805908203125, -15.085086822509766, -2.5410289764404297, 69.857177734375, 76.14410400390625, 0.0, 36.10174560546875, 11.691986083984375, 105.40264892578125, 48.730621337890625, 4.447166442871094, -9.5274658203125, 46.47322082519531, 99.12760925292969, -14.46605110168457, 127.80084228515625, 63.925018310546875, -43.9073486328125, 34.369873046875, 70.18921661376953, 28.92034912109375, -44.232513427734375, 67.84539794921875, -5.61578369140625, -25.89776611328125, 3.075592041015625, 57.80038070678711, 0.0, 84.509521484375, -41.141082763671875, 45.29664611816406, -6.338653564453125, 159.09027099609375, 115.33734893798828, 122.16688537597656, 56.037445068359375, -122.24407958984375, 105.25324249267578, -2.874542236328125, -31.259033203125, 174.86676025390625, 102.4005126953125, -44.85601806640625, 51.086883544921875, -91.692138671875, 115.21666717529297, 131.85556030273438, 95.59239196777344, 76.3474349975586, 12.765625, 115.08164978027344, 90.51138305664062, 121.45883178710938, -14.764848709106445, 151.3932647705078, 59.705902099609375, 117.22511291503906, 112.417236328125, 54.87315368652344, 29.0384521484375, 55.377166748046875, 77.67901611328125, 120.7001953125, 59.56212615966797, -11.01104736328125, 20.669662475585938, 131.10157775878906, 107.84707641601562, -5.86328125, 115.48916625976562, 127.81106567382812, 56.00553894042969, 13.661819458007812, 86.50357055664062, -19.616653442382812, 49.28683090209961, 12.402557373046875, 15.3153076171875, 112.91114044189453, -80.4749755859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000361.npy"}
|
|
{"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 36.898582458496094, "std": 57.868045806884766, "min": -128.02471923828125, "p10": -33.35668563842773, "median": 25.82550048828125, "p90": 107.00929946899413, "max": 202.13037109375, "pos_frac": 0.7734375, "sample": [10.398246765136719, 67.64337158203125, -0.4660224914550781, 36.99774169921875, 92.6766357421875, 19.037353515625, 93.99446105957031, 7.671234130859375, 23.88909912109375, 11.681304931640625, 35.56573486328125, -128.02471923828125, -96.10594177246094, -61.6951904296875, 108.08218383789062, 11.379587173461914, 120.06422424316406, -41.86663818359375, 17.011878967285156, 83.05287170410156, 69.34297180175781, 106.54949188232422, 202.13037109375, 42.93023681640625, 63.90802001953125, 114.85826110839844, 90.90045166015625, 3.16973876953125, -98.05065155029297, 36.135284423828125, -17.617874145507812, -22.700618743896484, 76.05497741699219, -7.1670989990234375, -39.36261749267578, 100.48922729492188, 17.367164611816406, -12.029651641845703, 105.63288879394531, 14.894554138183594, 154.2308349609375, 28.609466552734375, 134.58892822265625, 94.41981506347656, 74.3349609375, 109.16593170166016, 100.0719223022461, -2.734710693359375, -10.083251953125, 22.202072143554688, -30.78271484375, 78.97830200195312, 61.4815673828125, -30.298065185546875, 1.747091293334961, 6.821380615234375, 16.62115478515625, 122.64834594726562, 2.08953857421875, 101.92927551269531, 67.09877014160156, -15.916748046875, 27.400115966796875, 88.04779052734375, 70.15385437011719, 80.881591796875, 17.497970581054688, 17.637542724609375, 80.12290954589844, 57.956390380859375, 125.84060668945312, 68.55130004882812, 43.575462341308594, 9.9603271484375, -1.5829391479492188, 148.258056640625, 64.83396911621094, 6.162193298339844, 76.56427001953125, 88.27345275878906, -43.26622009277344, 88.13973999023438, -41.800968170166016, 10.02325439453125, -43.12103271484375, -50.770233154296875, -16.091367721557617, 132.13905334472656, 7.408924102783203, 10.630271911621094, 21.82403564453125, 4.4634857177734375, -7.8735504150390625, 12.189361572265625, 32.8072509765625, 69.46995544433594, 89.17151641845703, 19.323532104492188, 12.856048583984375, -14.4517822265625, -10.938278198242188, 94.8729019165039, 105.46786499023438, 19.24462890625, 4.4649658203125, 27.7635498046875, -109.11187744140625, 15.011688232421875, 73.2371826171875, 15.533866882324219, 105.44743347167969, 17.524364471435547, 87.00047302246094, 33.33709716796875, 24.250885009765625, -9.531803131103516, 63.657501220703125, 126.712158203125, 16.992156982421875, 27.54790496826172, -48.8692626953125, 72.6738510131836, 88.96255493164062, 113.39291381835938, 3.1553955078125, -71.6002197265625, 72.1341552734375, 55.8319091796875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000362.npy"}
|
|
{"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 44.07638168334961, "std": 61.219852447509766, "min": -146.2645263671875, "p10": -17.072113037109375, "median": 30.35101318359375, "p90": 122.39682159423828, "max": 175.59600830078125, "pos_frac": 0.7734375, "sample": [105.98053741455078, -2.7597198486328125, -11.208251953125, 126.3682861328125, 106.77392578125, 78.357666015625, -0.46523094177246094, -81.06217956542969, 20.1043701171875, 76.8104019165039, -17.038787841796875, 7.219825744628906, 104.43366241455078, 71.63807678222656, 1.9973602294921875, 56.38825988769531, 79.03909301757812, 145.87620544433594, -0.071563720703125, -22.84018898010254, 20.602752685546875, 88.56997680664062, 140.48460388183594, 58.98944091796875, 22.694107055664062, 136.6903076171875, 142.0032958984375, 172.45709228515625, 10.799156188964844, 4.016357421875, 111.22265625, -1.9524078369140625, -8.631439208984375, -2.0557098388671875, 85.75604248046875, 1.4916553497314453, -1.9604415893554688, 75.9765625, 111.60990905761719, 2.9095458984375, -16.195762634277344, 18.434524536132812, 26.889556884765625, 87.64087677001953, 72.21684265136719, 67.90728759765625, -4.243408203125, 73.39602661132812, 123.38725280761719, 13.4573974609375, 101.560302734375, 2.466409683227539, 143.321044921875, 114.18270874023438, -91.45394897460938, 75.5484619140625, 5.722808837890625, 74.78823852539062, 28.3609619140625, 69.6915283203125, 24.097381591796875, 35.02064514160156, 109.65086364746094, 95.62277221679688, 121.92633056640625, -0.01064300537109375, 133.16775512695312, 50.34552001953125, 32.341064453125, 4.514556884765625, 102.49771118164062, 4.131866455078125, 140.0030517578125, 175.59600830078125, 6.989166259765625, 118.99969482421875, -51.079498291015625, -5.644287109375, 28.25253677368164, 44.444854736328125, 6.260322570800781, 13.572463989257812, -17.149871826171875, 9.30999755859375, 55.90550231933594, -11.746902465820312, -2.53076171875, -7.035675048828125, -81.476806640625, 19.36029052734375, 64.48614501953125, -96.80280303955078, 93.30230712890625, 9.876365661621094, 74.15158081054688, 119.947265625, 97.64995574951172, 3.891387939453125, 110.24124145507812, 75.67123413085938, 19.780303955078125, 66.02474975585938, -106.30999755859375, 86.58367919921875, 107.84197998046875, 89.60504150390625, 24.325042724609375, 50.11614990234375, 123.91285705566406, 53.03228759765625, -146.2645263671875, -48.86433410644531, -26.520767211914062, 1.5935745239257812, 16.877685546875, 32.74700927734375, 7.622039794921875, 7.184288024902344, 95.02426147460938, 102.68186950683594, 121.97235107421875, 111.51861572265625, 16.93097686767578, 127.25579833984375, -18.052597045898438, 13.646636962890625, 18.50628662109375, -19.039093017578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000363.npy"}
|
|
{"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 33.19866943359375, "std": 61.74570083618164, "min": -156.677978515625, "p10": -26.731150817871093, "median": 21.44457244873047, "p90": 113.17855682373045, "max": 193.79693603515625, "pos_frac": 0.6796875, "sample": [60.224090576171875, -2.6491241455078125, 84.34808349609375, 21.255142211914062, 50.811676025390625, 12.990325927734375, 92.59368896484375, 103.82901000976562, 89.42750549316406, 11.507865905761719, 10.372467041015625, 72.5723876953125, -8.60809326171875, 30.19244384765625, -14.1561279296875, 32.527984619140625, 46.981353759765625, -7.66705322265625, 6.513519287109375, 63.252288818359375, -112.57624053955078, -2.3903656005859375, 48.70521926879883, -113.31643676757812, 21.634002685546875, -64.35171508789062, 139.18710327148438, 102.72911834716797, -12.006477355957031, 98.69937133789062, 7.6343994140625, 14.386428833007812, 42.84283447265625, 6.3646697998046875, -20.800079345703125, 2.4280548095703125, -2.0460052490234375, 97.36479187011719, 44.666412353515625, 108.12507629394531, 82.0943603515625, 8.8544921875, -136.77496337890625, 48.70021057128906, 97.27294921875, 128.1608428955078, -30.388452529907227, 37.656005859375, -21.068084716796875, -21.324234008789062, 94.47784423828125, 117.57879638671875, 7.6749267578125, 55.856658935546875, -9.733978271484375, -2.8299713134765625, 129.8380126953125, -1.2015552520751953, -156.677978515625, 165.53163146972656, 36.46966552734375, -11.7872314453125, -7.00244140625, -26.221023559570312, 30.602645874023438, 87.90914154052734, -27.92144775390625, 124.89730834960938, -8.14569091796875, 4.403167724609375, 36.18829345703125, 111.29273986816406, 70.08636474609375, -13.585884094238281, 92.36563873291016, 9.0889892578125, 125.325927734375, 64.05339813232422, 18.8870849609375, 99.72691345214844, 89.14190673828125, -13.192169189453125, 22.690994262695312, -6.15618896484375, -3.535858154296875, 22.305313110351562, 17.1265869140625, -8.49041748046875, 33.54154968261719, 142.34002685546875, 89.76556396484375, -1.6706619262695312, -17.25909423828125, 8.781143188476562, 104.12954711914062, 95.95980834960938, 21.041690826416016, -2.521465301513672, -31.05389404296875, 193.79693603515625, 30.632904052734375, 124.83119201660156, -65.95028686523438, 160.0921630859375, 88.99333190917969, -48.82720947265625, 48.782562255859375, 67.94638061523438, 2.449981689453125, -5.970123291015625, 30.323150634765625, 6.05352783203125, -4.7802276611328125, 137.1298370361328, 36.77776336669922, 23.63653564453125, 2.122589111328125, -48.44441223144531, 84.8294677734375, -65.82379150390625, 3.7470855712890625, 53.80900573730469, 10.484710693359375, -5.296966552734375, -35.081939697265625, 110.837158203125, 155.44764709472656, 20.005401611328125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000364.npy"}
|
|
{"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 42.541847229003906, "std": 58.82757568359375, "min": -93.64239501953125, "p10": -20.522888183593746, "median": 24.446121215820312, "p90": 124.99593048095701, "max": 161.761962890625, "pos_frac": 0.75, "sample": [0.0, 2.514516830444336, 88.0526123046875, 5.518104553222656, 123.04583740234375, 135.61767578125, 24.610870361328125, 54.523345947265625, 0.0, 24.2813720703125, 11.20654296875, 6.207672119140625, 20.547317504882812, 131.94161987304688, 9.180473327636719, 9.983474731445312, 16.10760498046875, 21.356857299804688, 17.528488159179688, 117.19416809082031, 39.24041748046875, 84.59033203125, 0.2299957275390625, 15.870162963867188, 39.07307434082031, 23.483642578125, -19.92376708984375, 109.30929565429688, 94.59718322753906, 114.66317749023438, 115.40303802490234, -6.66717529296875, -26.192962646484375, 88.84439849853516, -28.20233154296875, -62.05218505859375, 100.30477905273438, 60.83495330810547, 85.14878845214844, 42.171234130859375, -10.06930160522461, 102.84234619140625, 135.9052734375, 75.83403015136719, -36.39996337890625, 140.68923950195312, -35.703521728515625, 83.49397277832031, 38.417999267578125, 120.5323486328125, 97.34927368164062, -18.592254638671875, 11.40875244140625, -3.9748687744140625, 12.920684814453125, 8.182830810546875, 77.29551696777344, 123.03114318847656, 108.60296630859375, -0.8488922119140625, 19.81243896484375, 103.52554321289062, -29.757415771484375, 3.2545166015625, 4.704242706298828, 39.5596923828125, 95.66542053222656, 101.81788635253906, 20.594482421875, 88.26141357421875, 146.6036376953125, -9.201507568359375, 146.77133178710938, 1.9130401611328125, -1.0511474609375, 148.98883056640625, 71.33993530273438, -93.64239501953125, 123.68914794921875, 3.2920074462890625, 136.26876831054688, 2.43170166015625, 138.06124877929688, -10.607673645019531, 14.606325149536133, -14.804458618164062, 107.95712280273438, 52.80354309082031, 0.4608573913574219, -13.062492370605469, 54.829654693603516, 25.089908599853516, 33.35688400268555, 128.09033203125, -0.48540496826171875, 8.621414184570312, 12.208465576171875, -16.420806884765625, 95.54254913330078, -4.26348876953125, 19.480819702148438, -82.69290161132812, -12.16693115234375, -4.0882568359375, 99.34526062011719, 0.0, 12.491355895996094, 6.5204315185546875, 97.87913513183594, 142.40829467773438, 59.87420654296875, 66.87596130371094, 7.907527923583984, -21.92083740234375, -40.92962646484375, -86.67822265625, 95.91436767578125, 128.0450897216797, -30.884765625, 65.99691772460938, 31.2066650390625, -85.7462158203125, 161.761962890625, 78.00216674804688, 37.46343994140625, 103.74250793457031, 81.97404479980469, 55.68632507324219], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000365.npy"}
|
|
{"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 42.01689910888672, "std": 58.88408660888672, "min": -146.2764892578125, "p10": -19.823968505859362, "median": 35.4332275390625, "p90": 120.77010498046874, "max": 166.10012817382812, "pos_frac": 0.7578125, "sample": [10.002342224121094, 30.715011596679688, 166.10012817382812, 96.32762145996094, 89.14461517333984, 97.79388427734375, 128.00787353515625, 51.8826904296875, 57.799072265625, 121.46194458007812, 125.6318359375, 91.86796569824219, 77.26329803466797, -36.5296630859375, 24.603628158569336, 84.44436645507812, 45.05570983886719, 61.549560546875, -63.86566162109375, 44.56437683105469, -146.2764892578125, 151.1051025390625, -92.47376251220703, 16.335052490234375, 41.48687744140625, 101.67434692382812, 70.88398742675781, 65.4288330078125, 112.88995361328125, 69.55352783203125, -13.934539794921875, 100.68658447265625, 21.8504638671875, -30.16241455078125, 117.65350341796875, -87.50942993164062, -0.6279296875, 78.71197509765625, 31.37921142578125, 35.913970947265625, 4.731510162353516, -59.74536895751953, -8.386871337890625, -11.135932922363281, -1.766754150390625, 31.214492797851562, 107.89117431640625, 24.25921630859375, 84.79940032958984, -12.34906005859375, 55.364471435546875, 28.575607299804688, 7.83392333984375, 46.24516677856445, 64.65837097167969, -8.97808837890625, 30.594146728515625, 13.838615417480469, 37.48712158203125, 19.438148498535156, 3.3485107421875, 108.85295104980469, 30.983505249023438, 3.187580108642578, 2.4428329467773438, 128.8203125, 31.700862884521484, 88.66571044921875, 66.16313934326172, -57.875579833984375, 107.2093505859375, 125.36276245117188, -78.01182556152344, 0.0, -15.704498291015625, 123.01033020019531, 53.8780517578125, 76.46798706054688, 91.49978637695312, -2.5426807403564453, 1.2431640625, 39.64836120605469, 128.88870239257812, 55.096710205078125, -5.218994140625, 48.251434326171875, 34.952484130859375, 147.5479736328125, 1.63397216796875, -12.018783569335938, -4.1744384765625, 12.74298095703125, 76.81996154785156, 14.432144165039062, 33.10448455810547, 99.57156372070312, 38.03521728515625, -6.71783447265625, 43.562255859375, 120.47360229492188, 100.07894897460938, 18.149444580078125, 3.713409423828125, 149.38485717773438, 95.751220703125, 15.48602294921875, 92.12135314941406, -12.9091796875, 150.890380859375, 3.7398853302001953, 1.75653076171875, 92.88947296142578, 90.16461181640625, 20.6656494140625, -2.50677490234375, -29.436065673828125, 116.98103332519531, -76.72982788085938, 98.62547302246094, 31.09442138671875, 100.52334594726562, 19.42059326171875, -3.4449462890625, 100.53675079345703, -42.4364013671875, -9.2061767578125, 125.4783935546875, -30.9117431640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000366.npy"}
|
|
{"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 43.71631622314453, "std": 55.84779739379883, "min": -87.204345703125, "p10": -21.4219970703125, "median": 39.28451347351074, "p90": 121.36809539794922, "max": 153.23104858398438, "pos_frac": 0.765625, "sample": [97.10873413085938, 27.709991455078125, -10.458709716796875, 134.13185119628906, -2.807819366455078, 15.451400756835938, 30.934410095214844, -87.204345703125, -7.63873291015625, 43.67399597167969, 108.1138916015625, 22.527036666870117, 15.687812805175781, -3.0317535400390625, 76.11483764648438, 140.547607421875, 24.455703735351562, 122.5968017578125, -1.07720947265625, 28.071182250976562, -3.52642822265625, 144.21875, 76.49896240234375, -40.27581787109375, 52.94751739501953, 137.02456665039062, -85.2659912109375, -10.06610107421875, 42.01171875, -9.916046142578125, -25.609237670898438, 60.67744445800781, 2.5701560974121094, -11.453155517578125, 107.32632446289062, -4.610382080078125, 49.5849609375, 93.123046875, 70.07461547851562, -55.45147705078125, 118.17926025390625, 45.738121032714844, 6.819459915161133, 86.200439453125, 99.17578125, 3.7508163452148438, -58.92291259765625, 66.29437255859375, -4.189727783203125, -31.01934814453125, 45.56976318359375, 10.413467407226562, 38.820350646972656, 40.366943359375, 103.1495361328125, 116.93414306640625, 89.62307739257812, 51.36144256591797, 51.163787841796875, -21.0008544921875, -52.458221435546875, -22.4046630859375, 143.1612548828125, 92.11865234375, 12.409904479980469, 131.89822387695312, -15.875076293945312, 96.42501831054688, 153.23104858398438, 48.61223220825195, 58.595428466796875, 13.73126220703125, 25.296585083007812, 56.885040283203125, 117.84518432617188, 116.04702758789062, 28.826995849609375, 24.418869018554688, 15.426116943359375, -51.916778564453125, 36.19915771484375, 120.84150695800781, 115.9870376586914, 126.15618896484375, 15.5030517578125, 44.93157958984375, 107.90199279785156, 6.62945556640625, -23.152252197265625, 69.73138427734375, 108.96914672851562, 89.7701416015625, 19.5009765625, 6.46246337890625, 15.69462776184082, -6.087089538574219, -61.73015594482422, 125.0755615234375, 38.98661422729492, 127.23355102539062, 2.3172340393066406, 137.345458984375, 111.95736694335938, 46.0377197265625, -11.348320007324219, 63.226165771484375, 56.88494873046875, 102.0966796875, 10.65203857421875, 77.80621337890625, 39.58241271972656, 61.39173126220703, 25.734466552734375, -55.56324005126953, 49.047698974609375, 5.860870361328125, -0.4535369873046875, 103.31153869628906, 57.0184326171875, 103.29743957519531, 14.797590255737305, 30.994842529296875, 12.23699951171875, 8.515625, 143.35882568359375, 79.9251708984375, -18.401336669921875, 37.96039581298828], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000367.npy"}
|
|
{"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 49.60432434082031, "std": 57.979095458984375, "min": -106.19573974609375, "p10": -12.781765747070311, "median": 47.8033447265625, "p90": 123.11361312866211, "max": 167.51498413085938, "pos_frac": 0.8046875, "sample": [9.12115478515625, 139.20767211914062, 75.02831268310547, 114.534423828125, 16.706085205078125, 116.27706909179688, 10.671043395996094, 0.987457275390625, 3.490753173828125, -5.4240875244140625, 108.57839965820312, -27.365188598632812, 67.046630859375, 148.21728515625, 81.33804321289062, 48.7371826171875, 159.48150634765625, 17.3748779296875, 65.09463500976562, 73.08173370361328, -22.678192138671875, -0.87957763671875, 45.638427734375, -13.456825256347656, 4.85943603515625, 20.86297607421875, -9.419551849365234, 4.755741119384766, -3.2754669189453125, 122.5244140625, 0.8410110473632812, 55.28656005859375, -22.867759704589844, 66.84886169433594, 12.87152099609375, -106.19573974609375, -0.058013916015625, 167.51498413085938, 97.48521423339844, 94.3944091796875, 23.234466552734375, 124.92364501953125, 114.269287109375, 95.37136840820312, 23.275970458984375, 23.29705810546875, 99.376220703125, -8.345657348632812, 16.054994583129883, -78.75777435302734, 6.72113037109375, 92.69995880126953, 6.352008819580078, 106.26316833496094, -9.614616394042969, -1.655059814453125, 133.71005249023438, 95.04497528076172, -19.49017333984375, 77.43148803710938, 25.93359375, 56.560089111328125, 59.169036865234375, 110.13081359863281, 124.63653564453125, 86.56280517578125, -41.255279541015625, 59.08195495605469, 28.287017822265625, -5.41754150390625, 10.54766845703125, 128.948974609375, 122.56478881835938, 32.14605712890625, 24.463150024414062, 91.70121765136719, 107.19390869140625, 69.44378662109375, 116.48477172851562, 84.780029296875, 2.99395751953125, 85.6571044921875, 118.17613220214844, 46.8695068359375, -9.059455871582031, 116.787353515625, 9.381866455078125, 43.87549591064453, 2.0674896240234375, 53.476898193359375, 0.23664474487304688, 9.417182922363281, 124.39420318603516, 23.020233154296875, 79.56723022460938, 22.99858856201172, 52.95508575439453, 82.074462890625, -12.492454528808594, 87.07645416259766, 126.63552856445312, -48.9952392578125, 89.22605895996094, 101.03692626953125, -42.19873046875, 28.1624755859375, -7.8985595703125, 76.35577392578125, 158.71823120117188, 121.83950805664062, 25.56573486328125, 103.15110778808594, 45.199615478515625, 7.145683288574219, 13.979080200195312, 106.12460327148438, 133.13507080078125, 114.19061279296875, 79.3221435546875, 65.92071533203125, 36.080108642578125, -61.38999938964844, -88.01007080078125, 5.775177001953125, 98.84398651123047, 125.95756530761719, -64.972412109375, 111.61553955078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000368.npy"}
|
|
{"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 35.18191909790039, "std": 60.66558837890625, "min": -129.919189453125, "p10": -39.150543212890625, "median": 24.271438598632812, "p90": 111.93545684814453, "max": 172.88479614257812, "pos_frac": 0.7421875, "sample": [83.99229431152344, 33.19812774658203, 97.23941040039062, -40.933013916015625, 23.05670166015625, -7.848297119140625, 20.5023193359375, 28.015899658203125, 58.018394470214844, 3.616363525390625, 57.830047607421875, 34.5625, -12.87032699584961, 62.55108642578125, 142.50408935546875, 136.35675048828125, 86.22621154785156, 89.3469009399414, 5.35906982421875, 93.69273376464844, 141.53448486328125, 2.822620391845703, 67.46531677246094, 34.934814453125, -30.48779296875, -110.45697021484375, 93.7740478515625, -16.57708740234375, 69.70278930664062, 6.6144256591796875, 97.30892944335938, 21.76971435546875, 107.665771484375, 86.8525390625, 35.4188232421875, 105.87093353271484, -8.256332397460938, 49.999114990234375, 102.20018005371094, 54.967132568359375, 7.2221832275390625, 13.071990966796875, 115.048828125, -32.4876708984375, 7.7025146484375, -69.08447265625, -11.703418731689453, 98.93629455566406, -79.578369140625, 139.74908447265625, -16.9666748046875, -84.26860809326172, 49.926361083984375, -65.41357421875, 112.47183227539062, 89.46057891845703, -68.3824234008789, -6.191978454589844, 99.84616088867188, 5.7589111328125, -88.18400573730469, 3.034027099609375, -71.46182250976562, 107.90144348144531, 126.75247192382812, 32.36803436279297, 85.44120788574219, -2.0701751708984375, 65.787109375, -14.471389770507812, 172.88479614257812, -13.057754516601562, 25.205841064453125, 23.3370361328125, 2.055389404296875, 152.533447265625, 59.4385986328125, 10.4373779296875, -57.208335876464844, 89.4494400024414, -0.5164718627929688, -129.919189453125, 4.909950256347656, 11.60784912109375, 19.63232421875, 89.32022857666016, 112.74116516113281, 73.41085815429688, -48.908355712890625, 0.0, 29.89105224609375, 104.39126586914062, 6.3870086669921875, 94.4508056640625, -38.386627197265625, 3.5863494873046875, 84.81259155273438, -20.071075439453125, 39.30580139160156, 112.77023315429688, 13.355087280273438, 111.70558166503906, -0.54388427734375, 5.419921875, 53.706634521484375, -26.49711799621582, 102.05770874023438, 16.066146850585938, 81.9090576171875, -57.08152770996094, 49.77549743652344, 12.290519714355469, 12.2369384765625, 110.5634765625, 111.5157470703125, 0.83355712890625, 12.694000244140625, 126.60884094238281, -12.822280883789062, 42.37115478515625, 13.009010314941406, 129.376953125, 11.54376220703125, 14.767807006835938, 36.171722412109375, -8.930892944335938, 63.7742919921875, 5.1607666015625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000369.npy"}
|
|
{"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 43.855621337890625, "std": 61.964603424072266, "min": -149.54058837890625, "p10": -11.997535705566401, "median": 31.73918914794922, "p90": 120.50880432128906, "max": 183.36972045898438, "pos_frac": 0.796875, "sample": [54.15374755859375, -107.73739624023438, 108.941650390625, 29.037017822265625, 42.522613525390625, -0.054412841796875, -6.5877685546875, 3.610595703125, 80.52987670898438, 45.96931457519531, 86.08997344970703, 60.40483093261719, -9.782546997070312, 16.0262451171875, 79.19448852539062, 13.6265869140625, 0.397125244140625, 151.21307373046875, -3.3204574584960938, 29.2900390625, 97.3658447265625, 29.348297119140625, 81.115966796875, 103.74539184570312, 8.97491455078125, -2.4233856201171875, -10.28792953491211, 101.20413208007812, 3.6059913635253906, 27.6646728515625, 23.694366455078125, -15.31707763671875, 96.45463562011719, 23.055843353271484, 32.55912780761719, 4.3983612060546875, 167.19952392578125, 62.05615234375, 3.248577117919922, 134.72996520996094, 75.35479736328125, -113.83927917480469, 98.97482299804688, 7.45585823059082, 79.47632598876953, 70.493896484375, 64.35426330566406, 33.234130859375, 30.91925048828125, 0.49426841735839844, 88.79124450683594, 11.664215087890625, 8.292617797851562, 83.90693664550781, 81.11407470703125, 6.022735595703125, 89.26960754394531, -5.6043701171875, 122.91057586669922, -80.11719512939453, -20.55303192138672, 121.59176635742188, -10.574874877929688, -43.073097229003906, 6.261867523193359, 112.10148620605469, 157.31964111328125, 50.40870666503906, 89.5658187866211, -32.273406982421875, 1.6507568359375, -8.934768676757812, 48.556640625, 119.78680419921875, 95.79833984375, -19.8466796875, 128.14111328125, 126.7791748046875, 44.9678955078125, 117.80679321289062, 125.66064453125, 99.24433898925781, 16.15643310546875, 160.0416259765625, 104.9088363647461, 111.20449829101562, -21.17131805419922, 6.803802490234375, -115.74514770507812, 113.31008911132812, 92.73751068115234, 15.48579216003418, -65.46798706054688, 25.616943359375, 183.36972045898438, -5.351001739501953, 25.86993408203125, 19.667144775390625, 70.52601623535156, 114.41293334960938, 114.3817138671875, 126.565673828125, 45.723785400390625, 120.044677734375, -9.120803833007812, 77.37812042236328, 60.6558837890625, 52.75077438354492, 26.3946533203125, 61.50921630859375, 142.73486328125, -149.54058837890625, 10.827651977539062, -7.83001708984375, 0.8607559204101562, 111.18145751953125, 10.864593505859375, 7.228904724121094, 83.97454833984375, 14.235977172851562, 11.832815170288086, -75.17002868652344, 21.52196502685547, 111.57577514648438, 0.06754302978515625, 5.8923187255859375, 85.37854766845703, -10.249176025390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000370.npy"}
|
|
{"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 35.69501876831055, "std": 57.67400360107422, "min": -144.98526000976562, "p10": -27.362841796874996, "median": 30.273391723632812, "p90": 111.49268341064453, "max": 157.01345825195312, "pos_frac": 0.75, "sample": [3.967386245727539, 65.74711608886719, 38.3494873046875, 103.1549301147461, -11.262994766235352, 1.7097930908203125, 137.0081787109375, 9.97955322265625, 21.566818237304688, 16.103759765625, 2.1179733276367188, -136.65924072265625, 17.0633544921875, 5.240081787109375, 22.69110107421875, -8.177734375, -43.078369140625, 65.97828674316406, 6.630096435546875, -7.085052490234375, -15.3829345703125, 31.68035888671875, -9.517946243286133, 127.325927734375, 41.97895812988281, 78.55709838867188, -0.5542335510253906, 53.19281005859375, -68.88444519042969, 34.31573486328125, 88.7197265625, 122.56761169433594, -31.32452392578125, 34.53065490722656, -144.98526000976562, 56.294342041015625, -33.56341552734375, -0.3186798095703125, 86.99238586425781, 27.2645263671875, 115.55409240722656, 8.834442138671875, -55.681304931640625, 3.98388671875, 32.41650390625, 31.543914794921875, 9.987579345703125, 0.0, -19.863983154296875, 0.14009857177734375, 131.158203125, 113.16722106933594, 63.29345703125, -54.68853759765625, -99.10745239257812, -4.65606689453125, 45.345306396484375, 136.12115478515625, 110.57858276367188, 10.010978698730469, 18.39013671875, 77.01861572265625, 29.492538452148438, 152.92831420898438, 41.3394775390625, 32.95269775390625, 48.757354736328125, 90.92357635498047, 121.18919372558594, -21.989715576171875, 28.259029388427734, 100.00433349609375, 99.0407943725586, -63.99066162109375, 42.346046447753906, 94.98847198486328, 85.0697021484375, 31.054244995117188, 91.33612060546875, 18.940093994140625, -21.385955810546875, -57.387237548828125, 19.409149169921875, 157.01345825195312, 47.5950813293457, -25.66497802734375, -55.940269470214844, 95.2017593383789, 85.6749267578125, 36.30347442626953, 0.89190673828125, -14.95477294921875, 69.05419921875, 56.844146728515625, 27.806167602539062, -1.6420364379882812, 96.64849853515625, 55.33441162109375, 20.950531005859375, 143.20718383789062, 21.133304595947266, -3.8540916442871094, 66.42489624023438, 25.67517852783203, 103.9654541015625, 101.2373046875, -1.0791778564453125, 25.943603515625, 19.61157989501953, 47.536590576171875, 84.88050079345703, -36.015869140625, 46.4627685546875, 41.583900451660156, 3.081573486328125, 110.69660949707031, 119.04730224609375, 6.199619293212891, 110.7750244140625, 110.38592529296875, 7.41778564453125, -18.903564453125, 13.104095458984375, -3.7673797607421875, 35.410152435302734, 98.04269409179688, 81.90577697753906, 130.97930908203125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000371.npy"}
|
|
{"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 50.7540283203125, "std": 61.18595504760742, "min": -113.3973388671875, "p10": -16.310258483886717, "median": 50.83433532714844, "p90": 131.6912078857422, "max": 177.91021728515625, "pos_frac": 0.7890625, "sample": [135.3603515625, -6.577056884765625, 95.35171508789062, 50.93094253540039, -1.2930278778076172, 100.3966064453125, 90.398681640625, -5.731536865234375, 88.36675262451172, 127.72117614746094, 45.12825012207031, 71.31114196777344, 98.4788818359375, 91.65927124023438, -15.937057495117188, 87.3521728515625, 19.776832580566406, 121.04473876953125, -0.7900009155273438, 16.509002685546875, 64.81277465820312, 26.59423065185547, 97.56787109375, 20.04248046875, -17.181060791015625, 85.4931640625, 146.32809448242188, 63.09553527832031, 107.5731201171875, 98.53739929199219, -18.89569091796875, -34.6082763671875, 132.199951171875, -13.84375, 71.07412719726562, 84.730712890625, 142.6522216796875, -23.238082885742188, 56.172203063964844, 61.833290100097656, -14.422294616699219, 135.51080322265625, 60.702392578125, 74.38941192626953, 144.07769775390625, -105.15615844726562, 133.086181640625, 76.77938842773438, 4.7005157470703125, 48.6221923828125, 19.541114807128906, 17.802703857421875, 76.78866577148438, 0.218994140625, 15.046539306640625, 28.207275390625, 79.19393920898438, 114.9949951171875, 84.97006225585938, 34.995147705078125, 76.37324523925781, 29.558670043945312, 88.68495178222656, 16.581039428710938, 5.247978210449219, 83.38784790039062, 104.92813110351562, 15.35260009765625, 127.6201171875, -1.3370132446289062, -0.894622802734375, 44.883087158203125, 13.690078735351562, 115.80621337890625, 131.47317504882812, 142.92626953125, -63.932830810546875, 139.78274536132812, 32.89361572265625, -108.92889404296875, 109.49310302734375, 56.621307373046875, 42.615447998046875, 52.822357177734375, -94.51121520996094, 41.779815673828125, 49.3994140625, 18.967758178710938, -5.974199295043945, 119.56227111816406, -14.25143814086914, 0.353179931640625, 111.40736389160156, 53.13519287109375, 37.980712890625, -105.97940063476562, -33.85089111328125, 31.44647216796875, 91.80119323730469, -24.212158203125, 113.53753662109375, -2.6626739501953125, 6.111907958984375, 22.537750244140625, 120.95233154296875, 98.82315063476562, -113.3973388671875, 132.5787353515625, 177.91021728515625, 140.87042236328125, 23.787384033203125, 151.3697509765625, 123.64564514160156, 2.7556686401367188, 50.85076904296875, 124.62741088867188, -1.7094612121582031, 42.00732421875, -53.2200927734375, 107.220458984375, 59.963043212890625, 102.20392608642578, 19.442537307739258, 3.50335693359375, -1.1313285827636719, 50.817901611328125, 41.93524169921875, 28.033935546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000372.npy"}
|
|
{"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 44.65777587890625, "std": 61.25331115722656, "min": -153.67327880859375, "p10": -26.65781269073486, "median": 37.64350128173828, "p90": 127.88752288818358, "max": 152.34796142578125, "pos_frac": 0.796875, "sample": [8.711395263671875, 35.950416564941406, 116.88436889648438, 30.129745483398438, 0.6592216491699219, -2.958282470703125, 19.97991943359375, 138.67889404296875, 17.2772216796875, 104.80267333984375, 100.63168334960938, -11.84710693359375, 21.438690185546875, 120.95343017578125, 83.95736694335938, 55.1234130859375, 0.8738327026367188, 47.944366455078125, -26.94134521484375, 79.1468505859375, 45.89373779296875, 4.308681488037109, 63.01115417480469, 52.95159912109375, -36.2796630859375, 30.870025634765625, -5.596916198730469, -16.642623901367188, 33.27056884765625, 147.5113525390625, 5.566169738769531, -19.285980224609375, 42.451934814453125, 32.26116943359375, 33.222625732421875, 92.04730224609375, 116.40274047851562, -13.76123046875, 101.999267578125, 58.737213134765625, 45.318634033203125, 105.66353607177734, 90.84811401367188, 37.19134521484375, 12.2603759765625, 99.013671875, 9.956634521484375, -54.05027770996094, 38.942169189453125, 137.02056884765625, -27.890995025634766, 107.98468017578125, -69.61193084716797, -34.111328125, 73.1317138671875, -33.946533203125, 149.97329711914062, 0.436676025390625, 27.67236328125, 44.71592712402344, 9.11003303527832, -153.67327880859375, 152.34796142578125, 29.454055786132812, 114.09152221679688, 12.428863525390625, -64.79986572265625, 121.9217529296875, 17.707305908203125, 93.68316650390625, 94.96337890625, 116.23866271972656, 9.711677551269531, 58.28141784667969, 56.86993408203125, 40.808677673339844, 78.0640869140625, 2.958984375, 57.91937255859375, 5.938568115234375, -109.5750732421875, -11.595191955566406, 138.95809936523438, 46.04908752441406, 34.5184326171875, 125.22750854492188, 89.36581420898438, 141.39932250976562, 144.13717651367188, 74.29908752441406, -3.03863525390625, 129.6416778564453, 4.4912109375, 17.120315551757812, 120.01911926269531, 127.1357421875, 19.944305419921875, 137.703369140625, -75.6492919921875, -81.55967712402344, 46.90093231201172, 115.60055541992188, 143.5289306640625, 114.13601684570312, 39.67620849609375, 14.169281005859375, 91.88873291015625, -11.2724609375, 149.82632446289062, 16.27978515625, 12.845447540283203, 10.388648986816406, 15.5166015625, 38.09565734863281, -31.806262969970703, -4.484649658203125, 78.56283569335938, 1.53179931640625, -26.536298751831055, 83.12640380859375, -2.920469284057617, 118.2391357421875, -18.037281036376953, 7.521209716796875, 123.72555541992188, 130.55242919921875, 57.984580993652344, 7.6827545166015625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000373.npy"}
|
|
{"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 33.73068618774414, "std": 60.24182891845703, "min": -137.1243896484375, "p10": -37.46261901855468, "median": 21.403656005859375, "p90": 118.32016143798828, "max": 167.43209838867188, "pos_frac": 0.7265625, "sample": [160.0537109375, 0.0, 10.213470458984375, 166.19683837890625, 4.81884765625, -59.150482177734375, -15.824577331542969, 23.00860595703125, 126.15666198730469, 84.8392333984375, 17.441431045532227, 156.1396484375, 8.901229858398438, -39.68449401855469, 34.16529846191406, 34.402557373046875, 68.80696868896484, 88.42379760742188, 17.65985107421875, 8.58575439453125, -83.45516967773438, -38.9359130859375, 74.9446792602539, -50.314666748046875, 118.29106140136719, 32.085670471191406, 13.225494384765625, 8.52294921875, 5.204128265380859, -21.60449981689453, 1.3525238037109375, 10.778457641601562, -22.78363037109375, 47.13109588623047, -9.7254638671875, 26.998435974121094, 13.90472412109375, -24.084991455078125, 136.03024291992188, 106.08236694335938, 37.208709716796875, 13.203369140625, 167.43209838867188, -0.16962432861328125, 55.078704833984375, -0.236968994140625, 1.8718643188476562, 43.13690185546875, -137.1243896484375, -15.264854431152344, 39.633934020996094, -36.831207275390625, 114.16683959960938, 34.03455352783203, 16.92138671875, 51.589385986328125, 19.7987060546875, 54.50640869140625, 69.08607482910156, -0.4944305419921875, 13.583854675292969, 138.295166015625, 138.59735107421875, -9.111328125, -0.84490966796875, 42.18048095703125, 52.169189453125, 49.10015869140625, -20.759857177734375, 75.66331481933594, 17.13348388671875, 47.594757080078125, 98.32991790771484, 81.0380859375, -88.32476806640625, 26.053733825683594, -41.9970703125, 17.4263916015625, 54.946990966796875, 104.62464904785156, 87.92733001708984, 11.900543212890625, 33.30230712890625, 107.82550811767578, 0.0, 112.60328674316406, 15.800468444824219, -30.1246337890625, -32.024688720703125, -61.03363037109375, 59.712066650390625, 83.21647644042969, 76.30514526367188, 112.83772277832031, 18.8375244140625, -17.80645751953125, 150.78579711914062, 15.134857177734375, 139.44537353515625, 50.70111083984375, 13.90936279296875, -68.1630859375, -63.19935607910156, 32.715545654296875, 13.169401168823242, 4.696990966796875, 70.52471923828125, 134.03167724609375, -4.1055450439453125, -44.88641357421875, 3.437286376953125, 39.068511962890625, 6.8139190673828125, -0.2781219482421875, 118.3880615234375, 0.793609619140625, 67.37959289550781, 29.907501220703125, -30.770904541015625, 93.77363586425781, 163.19631958007812, 104.1595230102539, 111.97370910644531, 36.409149169921875, 27.66204833984375, -3.157001495361328, 58.47419738769531, -95.7916259765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000374.npy"}
|
|
{"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 39.60626220703125, "std": 62.89511489868164, "min": -130.59088134765625, "p10": -24.238883209228515, "median": 27.31468963623047, "p90": 123.3516357421875, "max": 189.72052001953125, "pos_frac": 0.75, "sample": [51.69718933105469, 17.29071044921875, 78.94915771484375, 4.187957763671875, 38.09814453125, 4.1558990478515625, 21.270782470703125, -81.23606872558594, 6.046417236328125, 27.484283447265625, 132.97576904296875, 0.7239551544189453, 112.8031005859375, 122.83700561523438, 19.5345458984375, -10.82427978515625, 63.734405517578125, 79.27449035644531, 12.060077667236328, -1.04449462890625, 121.2652587890625, 127.18234252929688, -3.1104278564453125, 27.145095825195312, 51.72605895996094, 108.67562866210938, -23.583702087402344, 8.033233642578125, -22.50048065185547, 69.9324722290039, 13.133872985839844, 28.76012420654297, -3.8907623291015625, -79.30911254882812, 5.781394958496094, 114.72979736328125, 77.91262817382812, 122.77960205078125, 18.89862823486328, -3.48345947265625, 32.756683349609375, 10.0606689453125, -72.70758056640625, -0.004119873046875, -130.59088134765625, 95.22731018066406, 95.88287353515625, 99.07369995117188, 32.196510314941406, -100.55117797851562, -92.84605407714844, -11.701416015625, 63.980438232421875, 3.7026119232177734, 45.43104553222656, -0.8907623291015625, 101.90594482421875, -9.360153198242188, -25.76763916015625, -84.58163452148438, 121.94284057617188, 122.9447021484375, -11.74591064453125, 160.05508422851562, 19.18265151977539, 24.26953125, -9.498794555664062, 9.509613037109375, 189.72052001953125, -5.163068771362305, 31.3380126953125, 10.640575408935547, 18.701187133789062, 70.39592742919922, 34.5843505859375, 141.0096435546875, 53.081756591796875, -27.436691284179688, 41.77490234375, 32.76188659667969, 84.72467041015625, 70.71685791015625, -20.686431884765625, -8.505523681640625, 0.0918426513671875, 90.26651000976562, 8.431121826171875, 91.26614379882812, 49.889007568359375, 114.8446044921875, -96.0028305053711, 129.27772521972656, 154.91790771484375, -10.315322875976562, 149.4710693359375, 170.66879272460938, -39.70880126953125, 114.61416625976562, 75.82516479492188, 125.8292236328125, -9.063270568847656, -38.821197509765625, 124.3011474609375, 0.149688720703125, 106.5755615234375, 4.865804672241211, 88.06134033203125, 128.0281982421875, 30.57843017578125, 4.404541015625, -3.1461334228515625, 136.84234619140625, 69.5225830078125, 39.32275390625, 117.97039794921875, -26.10729217529297, 3.1737442016601562, 119.9701156616211, 2.3313751220703125, 19.341094970703125, 19.921600341796875, 18.21141815185547, 77.67645263671875, 59.67762756347656, 9.541351318359375, 100.30155944824219, 7.970306396484375, 63.00139617919922], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000375.npy"}
|
|
{"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 37.23324966430664, "std": 63.420570373535156, "min": -113.37705993652344, "p10": -41.749609375, "median": 33.45977783203125, "p90": 122.10614624023437, "max": 179.2484130859375, "pos_frac": 0.7109375, "sample": [54.55986785888672, 27.2481689453125, 126.2808837890625, 172.250732421875, 88.94210815429688, 179.2484130859375, 41.196868896484375, 72.43643188476562, 48.69432067871094, -68.09078979492188, 87.97802734375, -24.047775268554688, -70.19302368164062, 29.707855224609375, -14.62481689453125, 35.306243896484375, -6.464637756347656, 38.492034912109375, 87.8302001953125, 117.7130126953125, 77.08172607421875, -108.7161865234375, 2.6079788208007812, 13.409698486328125, 53.921356201171875, 22.95978546142578, -15.265960693359375, 129.834716796875, -2.3041839599609375, 90.81109619140625, 88.756591796875, 19.748977661132812, 120.5842514038086, 40.32708740234375, -0.2006378173828125, -42.5328369140625, 108.36563110351562, 59.155235290527344, 129.7567138671875, 8.209716796875, 0.0, 17.552719116210938, 15.214767456054688, 81.5605239868164, -41.4139404296875, 143.03402709960938, -66.17184448242188, 95.83065795898438, 42.498958587646484, 83.5240478515625, -91.26210021972656, 76.34881591796875, 37.63214111328125, 9.38134765625, 110.36770629882812, 5.126678466796875, 15.81402587890625, 60.86662292480469, -34.45823669433594, -13.288818359375, 82.01033020019531, -2.772430419921875, 154.06976318359375, 145.07403564453125, 16.800201416015625, 144.7738037109375, 126.0257568359375, 30.671855926513672, -22.166168212890625, -24.2513427734375, 71.27728271484375, -34.4931640625, 27.064361572265625, -20.191490173339844, -4.142313003540039, 1.458038330078125, -78.63621520996094, 2.6910629272460938, -66.25375366210938, 123.71614074707031, 87.51739501953125, 56.07794189453125, 121.69842529296875, 32.55120849609375, 117.34573364257812, 87.24189758300781, 6.42469596862793, -113.37705993652344, 12.448760986328125, -69.4149169921875, 44.7369384765625, 2.6886329650878906, 106.13631439208984, -15.594379425048828, 2.5639476776123047, 98.4183120727539, 63.98289489746094, -8.232086181640625, -69.32086181640625, 123.0574951171875, -4.03399658203125, 89.20947265625, -45.94972229003906, 101.31414031982422, 26.153564453125, 24.854469299316406, 74.13846588134766, 54.27998352050781, 45.05267333984375, 106.569091796875, -13.237762451171875, -35.4998779296875, -32.05656051635742, 33.83148193359375, 86.45901489257812, -2.2108078002929688, 33.08807373046875, -22.892478942871094, 55.4222412109375, 1.3451004028320312, 4.983497619628906, 50.50923156738281, 61.13812255859375, 35.472900390625, 120.14450073242188, -77.1787109375, 147.20828247070312, 118.931396484375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000376.npy"}
|
|
{"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 34.466339111328125, "std": 60.10675048828125, "min": -109.81634521484375, "p10": -28.194918823242187, "median": 19.890289306640625, "p90": 116.78132476806638, "max": 164.34408569335938, "pos_frac": 0.6796875, "sample": [135.70947265625, 76.35758972167969, 88.086181640625, -11.643928527832031, 46.745086669921875, 13.500946044921875, 11.200164794921875, 72.9437255859375, 98.81941223144531, -9.439544677734375, 19.6451416015625, 17.7252197265625, 82.95977783203125, -51.957916259765625, 85.20933532714844, 84.06150817871094, 110.3568115234375, -21.2921142578125, 0.8433189392089844, -29.27642822265625, 104.08001708984375, 1.6283645629882812, 9.435592651367188, 31.422199249267578, -11.968664169311523, -17.518882751464844, 50.94921875, -24.456451416015625, 121.74766540527344, 52.29821014404297, 75.14010620117188, -8.1488037109375, 77.33335876464844, -89.22705078125, -5.4324493408203125, 104.36166381835938, 10.227649688720703, -4.97564697265625, -10.74859619140625, 29.86956787109375, 20.13543701171875, 69.00082397460938, 11.519752502441406, -61.9249267578125, 82.54315185546875, 7.57012939453125, 147.56591796875, 57.4510498046875, -36.94856262207031, -75.03659057617188, 9.17633056640625, 143.98785400390625, -17.892333984375, -0.8883056640625, 133.5780029296875, 1.8510971069335938, -22.94317626953125, -11.350887298583984, 97.85076904296875, 108.3901596069336, 114.65289306640625, 29.38641357421875, 93.83319091796875, 71.34866333007812, -27.731414794921875, 109.49267578125, 74.87849426269531, 0.0, -27.42547607421875, 48.933319091796875, 98.85305786132812, -10.110771179199219, 64.55317687988281, 98.35690307617188, -4.293113708496094, 10.2919921875, 8.474334716796875, -14.384033203125, -0.598388671875, 96.41085815429688, 12.72357177734375, 158.00125122070312, 6.58428955078125, -24.9095458984375, -14.761444091796875, -30.861000061035156, 46.246551513671875, 20.460311889648438, 103.78417205810547, 28.80767822265625, 45.5565185546875, 25.2728271484375, 8.289535522460938, 94.613037109375, -31.91351318359375, 130.6707305908203, 12.867195129394531, 22.373260498046875, 41.97637939453125, 98.17201232910156, 37.702728271484375, -75.28079223632812, -10.749130249023438, 42.173736572265625, 128.66668701171875, -41.78765869140625, 154.5507049560547, 28.351852416992188, 2.617340087890625, 107.44622802734375, 15.907798767089844, 10.350006103515625, 21.730178833007812, 10.8026123046875, 131.78216552734375, 96.94406127929688, 130.38499450683594, -97.26605224609375, -11.7701416015625, 164.34408569335938, 17.07280731201172, -5.6890869140625, -81.07548522949219, 140.386474609375, -18.093643188476562, -2.173797607421875, -109.81634521484375, 55.096038818359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000377.npy"}
|
|
{"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 33.95673370361328, "std": 57.26261901855469, "min": -105.01425170898438, "p10": -30.833350372314445, "median": 23.047775268554688, "p90": 116.69701309204102, "max": 157.23919677734375, "pos_frac": 0.7109375, "sample": [85.01272583007812, 5.46429443359375, 20.28436279296875, 91.62968444824219, -18.133819580078125, 5.372060775756836, -24.507080078125, 9.63214111328125, 118.33270263671875, 56.522796630859375, -9.918701171875, 5.3720855712890625, 18.31866455078125, 98.40934753417969, 94.56863403320312, -40.031982421875, -105.01425170898438, 135.62643432617188, 144.34515380859375, -4.752166748046875, 35.724029541015625, 38.623077392578125, 10.014129638671875, -35.87187957763672, 73.94293212890625, 90.55450439453125, 149.28945922851562, 108.74974822998047, 23.023590087890625, 83.74331665039062, -5.0300140380859375, 72.77828979492188, 78.44166564941406, 28.012542724609375, 11.8948974609375, -80.26156616210938, 1.6706790924072266, 2.2641448974609375, 41.78861999511719, 79.52658081054688, 0.0, 116.40328216552734, -7.210906982421875, 94.77734375, 132.88418579101562, -44.27392578125, -4.080699920654297, 157.23919677734375, -17.44933319091797, 126.5054931640625, 54.19073486328125, 79.89459228515625, -0.03426361083984375, 8.64825439453125, 1.49334716796875, 7.598121643066406, -24.32232666015625, 88.46047973632812, -20.34710693359375, 101.23043823242188, 79.68218994140625, -2.415304183959961, 97.7592544555664, 95.83834838867188, 35.427642822265625, 61.8233642578125, 12.270889282226562, -53.029205322265625, 117.38238525390625, 50.46453857421875, 40.476593017578125, 13.921142578125, -1.55499267578125, -11.616073608398438, -82.49666595458984, 119.83746337890625, -20.415206909179688, -54.88177490234375, 30.472442626953125, -2.5596694946289062, 27.17229461669922, 3.6259307861328125, 81.7366943359375, 55.77370834350586, 6.211822509765625, 58.96075439453125, 100.25955200195312, 6.461029052734375, 13.035247802734375, 38.666839599609375, 110.61419677734375, 104.76531982421875, -16.53582763671875, 4.180419921875, 46.32135009765625, -37.37842559814453, -63.7086181640625, 15.358116149902344, 10.71630859375, -23.65264892578125, -45.475128173828125, 129.54571533203125, -9.391082763671875, 25.4429931640625, -97.23005676269531, -1.1318817138671875, 23.07196044921875, 82.04051971435547, -9.8170166015625, 27.123260498046875, 27.6944580078125, 38.909942626953125, 1.5901412963867188, -3.897705078125, 147.72662353515625, 69.31465148925781, 123.22531127929688, 103.52909851074219, 130.7735595703125, 16.011611938476562, -61.17828369140625, 52.87134552001953, 96.47064208984375, 29.460205078125, 44.68511962890625, -28.673980712890625, 12.515960693359375, 5.2642822265625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000378.npy"}
|
|
{"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 36.51750946044922, "std": 62.69717025756836, "min": -125.9893798828125, "p10": -37.810371398925774, "median": 27.797405242919922, "p90": 123.49373931884766, "max": 159.51666259765625, "pos_frac": 0.765625, "sample": [-13.011425018310547, 5.2138824462890625, 4.8227081298828125, 13.715866088867188, -1.1719894409179688, -11.11297607421875, 10.159019470214844, 112.44711303710938, 27.947433471679688, 2.7192611694335938, 105.39947509765625, 73.394287109375, 41.56941223144531, 9.559234619140625, 10.521484375, 10.05218505859375, 47.227657318115234, -78.14373779296875, 126.18173217773438, 29.08887481689453, 86.40084838867188, 66.55706787109375, 113.57566833496094, -32.60906982421875, 8.519882202148438, 2.7880401611328125, 112.79377746582031, 0.36749267578125, 60.27777099609375, 3.14239501953125, 34.97247314453125, 123.13755798339844, 130.18896484375, 57.033111572265625, 43.0487060546875, 8.20404052734375, 24.998947143554688, 133.07196044921875, 95.88529968261719, 11.591117858886719, -11.08660888671875, -35.527923583984375, 22.915863037109375, 16.656925201416016, 3.9700927734375, 143.29742431640625, 17.88990020751953, -56.805816650390625, -21.200103759765625, 15.227523803710938, 85.466552734375, -10.599194526672363, 35.495849609375, -104.46026611328125, 6.870574951171875, 16.36846923828125, 119.90493774414062, 108.784423828125, 97.08418273925781, 82.92764282226562, 65.86900329589844, -57.86543273925781, 18.17902374267578, 30.128677368164062, 102.25436401367188, 159.51666259765625, 16.73486328125, 125.2601318359375, 7.9945526123046875, -15.654327392578125, -36.30778503417969, 0.0, -0.1102142333984375, 39.06791687011719, 15.876251220703125, 133.80154418945312, 35.32373046875, 110.46295166015625, 129.020751953125, 112.48281860351562, 4.77337646484375, 90.30717468261719, 39.750640869140625, 13.511360168457031, 91.71272277832031, 75.90360260009766, -7.089530944824219, 79.6507568359375, -19.257034301757812, 80.61168670654297, 50.55810546875, -12.554702758789062, 137.33670043945312, 4.420499801635742, 79.40902709960938, -6.39764404296875, 3.0228424072265625, 83.33840942382812, -84.62094116210938, 150.78814697265625, 116.71490478515625, -125.9893798828125, 11.26898193359375, 34.4483642578125, 124.3248291015625, 68.12619018554688, 50.826141357421875, -41.31640625, 79.93084716796875, 112.400390625, 133.49212646484375, 53.779388427734375, 113.76095581054688, 47.14167785644531, -83.0740966796875, -100.30027770996094, 27.647377014160156, 4.1880950927734375, -9.70068359375, 111.41455078125, -45.12115478515625, 84.364501953125, -100.15652465820312, -67.30433654785156, -99.5068359375, 21.083059310913086, 36.17120361328125, 124.71066284179688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000379.npy"}
|
|
{"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 37.90099334716797, "std": 61.50221633911133, "min": -95.37911987304688, "p10": -36.04061431884765, "median": 29.602401733398438, "p90": 122.2230743408203, "max": 206.662353515625, "pos_frac": 0.7265625, "sample": [99.08157348632812, 39.628211975097656, 70.80035400390625, 30.157943725585938, 85.62136840820312, 15.7193603515625, -80.22868347167969, -1.346771240234375, 1.0795059204101562, -65.37408447265625, 44.20306396484375, 68.52932739257812, 106.515869140625, -74.22681427001953, 17.10302734375, 11.202323913574219, 96.6002197265625, 40.4254150390625, 103.98748016357422, -4.285770416259766, 0.0, 41.14271545410156, 18.424102783203125, 21.83050537109375, 29.36859130859375, 102.927001953125, 9.13885498046875, 63.66839599609375, -1.187896728515625, 6.765045166015625, 42.72991943359375, 145.596435546875, 15.2215576171875, 88.57386779785156, 5.080108642578125, -69.16675567626953, 91.30014038085938, -0.6057319641113281, 30.335670471191406, -20.76495361328125, 43.122772216796875, 46.41239929199219, 123.63894653320312, -29.859161376953125, -16.08301544189453, -5.00543212890625, 20.571182250976562, 133.8890380859375, 137.2127685546875, -43.813751220703125, -45.80955505371094, 40.57191467285156, 24.907913208007812, 4.66363525390625, 22.216217041015625, -57.643409729003906, -19.237354278564453, 11.295501708984375, 139.82699584960938, 89.05233001708984, -2.349365234375, 117.17160034179688, 66.98126220703125, 35.6461181640625, 94.481201171875, 128.55303955078125, -68.64374542236328, -4.9798736572265625, 16.3568115234375, 18.24593734741211, 101.07266235351562, 142.77545166015625, 2.432464599609375, 89.5640869140625, 121.61627197265625, 0.32568359375, 37.024505615234375, -35.582305908203125, 38.869842529296875, -95.37911987304688, 64.94942474365234, 206.662353515625, -23.162139892578125, 78.77165222167969, 79.05203247070312, 39.88008499145508, 107.364501953125, 107.611328125, 117.52658081054688, 79.97740173339844, 29.836212158203125, 169.18679809570312, -61.077606201171875, 50.38916015625, 3.998889923095703, 47.76800537109375, 4.48321533203125, -35.12384033203125, 138.01739501953125, 190.34503173828125, 32.282169342041016, 43.45829772949219, 19.558486938476562, 113.53289794921875, 32.78880310058594, 14.778465270996094, 23.73016357421875, 32.51788330078125, -23.316192626953125, -41.326812744140625, -8.83935546875, -37.11000061035156, 12.431472778320312, 14.429967880249023, 22.57989501953125, 89.86386108398438, -21.14404296875, 65.58609008789062, 68.06570434570312, 182.774658203125, 26.4853515625, -15.31146240234375, -23.102935791015625, 93.929931640625, -2.79034423828125, -51.20851135253906, 152.31747436523438, -11.77447509765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000380.npy"}
|
|
{"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 40.4888801574707, "std": 58.20570373535156, "min": -86.84284973144531, "p10": -24.615679931640624, "median": 30.308382034301758, "p90": 120.45962677001953, "max": 148.34173583984375, "pos_frac": 0.7109375, "sample": [84.30859375, 144.04241943359375, 118.63507080078125, 30.203678131103516, 127.03663635253906, -24.18646240234375, 120.47315979003906, 129.67630004882812, 81.60968017578125, 4.6400604248046875, 136.9964599609375, 30.4130859375, 9.9840087890625, 109.5302734375, 86.35589599609375, 118.05642700195312, 0.0, 11.07940673828125, 40.72566223144531, 62.183807373046875, 90.87251281738281, 56.22088623046875, 9.448410034179688, 148.34173583984375, 49.278594970703125, 23.115554809570312, 61.720497131347656, -86.84284973144531, 0.0, -73.97659301757812, 71.231689453125, 70.51475524902344, -16.10894775390625, 107.22323608398438, -5.869274139404297, 25.068649291992188, 10.107574462890625, 59.298004150390625, 9.98199462890625, 78.3765869140625, 141.2279052734375, 51.60111999511719, -14.990997314453125, 93.77374267578125, 104.73797607421875, 8.09710693359375, 123.6805419921875, -9.859170913696289, -59.361968994140625, 92.55931091308594, 111.21366882324219, 117.31295776367188, 37.75508117675781, -2.9570770263671875, 79.660888671875, 113.46710205078125, 9.1441650390625, 88.63602447509766, 127.59539794921875, 51.45436096191406, -54.631256103515625, 20.4044189453125, 53.28107452392578, 9.67230224609375, -1.774932861328125, 35.811031341552734, 117.01236724853516, -0.165374755859375, -7.1367645263671875, -10.408843994140625, -18.294219970703125, 3.4168224334716797, -10.467437744140625, 105.431396484375, -14.6041259765625, 22.702239990234375, -7.071739196777344, 58.27613830566406, 33.24632263183594, -52.31085205078125, 120.45382690429688, -29.36992645263672, 103.99351501464844, 35.5489501953125, 128.475341796875, -5.15289306640625, -25.6171875, -6.3224029541015625, 121.42277526855469, 11.62872314453125, -1.548858642578125, -44.196929931640625, 104.15213775634766, 88.39976501464844, 78.50125122070312, -45.73760986328125, 2.4851036071777344, 2.597412109375, 9.616485595703125, -66.2453384399414, 66.11651611328125, 6.098640441894531, -66.35873413085938, 71.54186248779297, -54.69647216796875, 85.84771728515625, -22.5823974609375, 6.390520095825195, -16.882064819335938, 96.07968139648438, -2.7991943359375, 137.59449768066406, -9.4075927734375, 142.28421020507812, 19.684234619140625, 31.097183227539062, 25.50263214111328, 7.518913269042969, 91.82411193847656, 21.23406982421875, 115.88958740234375, 117.820556640625, 37.056236267089844, -20.689208984375, -59.555816650390625, 74.21694946289062, 26.04559326171875, 15.716552734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000381.npy"}
|
|
{"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 50.793601989746094, "std": 61.08102798461914, "min": -91.57325744628906, "p10": -25.10886535644531, "median": 57.54072189331055, "p90": 125.84639739990234, "max": 172.87261962890625, "pos_frac": 0.765625, "sample": [-50.68243408203125, 25.816452026367188, 60.0821533203125, 36.416839599609375, 81.51373291015625, 113.89604949951172, 95.07229614257812, 3.989349365234375, 10.644464492797852, 87.06704711914062, 85.06692504882812, -69.80120849609375, 88.59840393066406, 12.368080139160156, 142.52252197265625, 69.53781127929688, 32.84278106689453, 97.78338623046875, 94.674560546875, -11.574348449707031, 40.64984130859375, -91.57325744628906, 115.55625915527344, 1.697906494140625, -9.460540771484375, 132.26712036132812, 33.20953369140625, -21.388526916503906, 65.3223876953125, 74.25446319580078, -7.5576171875, 72.47331237792969, 53.021209716796875, 28.671058654785156, 95.13726806640625, 71.81841278076172, -1.8745346069335938, 56.039939880371094, -6.583270072937012, 126.44839477539062, 31.55892562866211, 20.79522705078125, 126.38104248046875, 19.85723876953125, 123.86856079101562, 122.17282104492188, 134.6318817138672, 41.5740966796875, 23.68194580078125, 7.660400390625, 128.13143920898438, 75.44569396972656, 15.903236389160156, 72.73526000976562, -3.9393310546875, 78.99444580078125, 172.87261962890625, 156.08505249023438, -32.4183349609375, -6.16265869140625, -24.62738037109375, -90.79912567138672, 169.00546264648438, 39.672210693359375, -1.3020477294921875, 80.64653015136719, 129.34628295898438, 47.850372314453125, 131.41818237304688, 21.727081298828125, 142.33377075195312, 17.362060546875, 24.105133056640625, -9.646942138671875, 123.18185424804688, 69.53828430175781, 97.55126953125, 16.990097045898438, 70.17312622070312, 120.10577392578125, 33.22430419921875, -31.63433837890625, 48.45977783203125, 85.33053588867188, -49.418060302734375, 4.731475830078125, -70.40757751464844, 117.2470703125, -9.637710571289062, 114.99972534179688, 120.6993408203125, -26.232330322265625, 11.058822631835938, 125.61726379394531, 124.29092407226562, 39.47010803222656, 80.4006118774414, -76.00518798828125, 5.797576904296875, -15.391357421875, 37.541534423828125, 59.04150390625, 69.96015930175781, 98.69802856445312, 79.01205444335938, -50.074737548828125, 30.93414306640625, 89.00082397460938, -85.73965454101562, 88.15934753417969, 94.8402099609375, -2.1155319213867188, 100.06878662109375, 81.08741760253906, 55.2689094543457, 84.24449157714844, -62.735198974609375, 62.61199951171875, -21.872718811035156, 88.11079406738281, 123.16690063476562, -16.474117279052734, 146.1632080078125, 93.58006286621094, 125.25711059570312, 81.59002685546875, 118.15736389160156, -16.9285888671875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000382.npy"}
|
|
{"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 38.70289611816406, "std": 60.1933479309082, "min": -107.59638977050781, "p10": -37.194281005859374, "median": 28.631725311279297, "p90": 115.65857543945312, "max": 162.07089233398438, "pos_frac": 0.75, "sample": [11.64141845703125, 45.53875732421875, 97.76093292236328, 96.0733642578125, -41.33515167236328, 20.162872314453125, -26.886810302734375, 16.852142333984375, -1.163421630859375, 27.78612518310547, 97.81511688232422, -90.35829162597656, 103.42935943603516, 21.496612548828125, 0.0, -8.685447692871094, 7.762214660644531, -49.779876708984375, 44.803009033203125, -39.007171630859375, 16.2802734375, 133.69168090820312, 104.73326110839844, 64.631103515625, -71.82965087890625, 133.93589782714844, -36.417327880859375, -3.884807586669922, -45.43864440917969, 10.218276977539062, 57.48427200317383, 15.244140625, -16.05358123779297, -32.638023376464844, 104.99446105957031, -14.11676025390625, 138.5550537109375, 117.1845703125, 33.20381164550781, 93.66860961914062, 68.806396484375, 7.420196533203125, 64.6640625, 162.07089233398438, -107.59638977050781, 0.2441558837890625, 43.21916198730469, 113.4844970703125, 73.8675537109375, 6.9921875, 119.86471557617188, 1.830230712890625, -1.9304943084716797, 105.61869049072266, 29.56304931640625, 20.6800537109375, 2.8921966552734375, -79.62303924560547, -0.13059234619140625, -81.70547485351562, 93.63465881347656, 84.3367919921875, 7.8804168701171875, 124.08539581298828, -21.07708740234375, 18.1898193359375, 125.12982177734375, 0.401092529296875, 1.770751953125, 4.599456787109375, 70.8343276977539, 128.3941192626953, 99.2193603515625, 101.3756103515625, 86.04238891601562, 41.13876724243164, -8.036834716796875, 2.8832855224609375, 30.985183715820312, 105.27375793457031, 1.7140350341796875, 84.21197509765625, 91.75257873535156, 46.50410461425781, 96.272216796875, 29.477325439453125, 24.099334716796875, 18.89004135131836, 30.644256591796875, -105.46356201171875, 115.00457763671875, 23.27637481689453, 48.37486267089844, 74.08414459228516, 49.410125732421875, 5.938152313232422, 10.876274108886719, 142.86868286132812, 97.85992431640625, 94.77205657958984, 105.23910522460938, -39.33203125, 158.45108032226562, -49.83259582519531, -19.2132568359375, 17.247650146484375, 50.132354736328125, 17.290252685546875, 0.0, 94.52591705322266, 99.63971710205078, -2.490224838256836, 63.03863525390625, 72.25750732421875, 48.36639404296875, -16.612197875976562, -26.322601318359375, 11.427810668945312, 128.32681274414062, 84.58367919921875, 9.170089721679688, 103.53821563720703, 138.06454467773438, -73.54579162597656, 14.02874755859375, -3.993804931640625, 103.06903839111328, 101.70266723632812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000383.npy"}
|
|
{"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 29.294429779052734, "std": 62.67204666137695, "min": -126.5545654296875, "p10": -42.75892868041992, "median": 18.38509750366211, "p90": 112.86517639160157, "max": 167.18670654296875, "pos_frac": 0.65625, "sample": [144.39617919921875, -126.5545654296875, 51.477325439453125, -17.52374267578125, -14.104248046875, 87.24601745605469, 47.1092529296875, 167.18670654296875, 18.078948974609375, -7.801078796386719, 89.12272644042969, -15.710079193115234, -21.919082641601562, -61.492950439453125, 26.8814697265625, -80.70819091796875, 113.7674560546875, -3.078968048095703, 135.39443969726562, -14.80364990234375, 7.226284027099609, 133.05325317382812, 94.79521179199219, 101.33366394042969, -2.2498931884765625, 86.416748046875, 21.472640991210938, 81.88278198242188, 17.202392578125, -26.525390625, 2.8380184173583984, -30.87054443359375, -2.964935302734375, 55.0823974609375, -27.050537109375, 84.08032989501953, 15.419647216796875, 27.421905517578125, 102.97883605957031, -19.22460174560547, 10.02911376953125, -83.87600708007812, 6.5519866943359375, 18.75989532470703, 97.31350708007812, 109.5126953125, 112.47848510742188, -108.61072540283203, 2.56732177734375, 18.691246032714844, 89.22848510742188, 141.37173461914062, -42.23059844970703, -1.55401611328125, -94.10800170898438, 42.95387268066406, 45.96533203125, 51.686767578125, -83.05802917480469, -19.901611328125, 30.4888916015625, -83.55264282226562, 92.87472534179688, 87.41970825195312, 138.01504516601562, 38.429595947265625, 32.120849609375, 49.49017333984375, 62.220977783203125, -1.0888175964355469, -8.549816131591797, 59.055137634277344, 113.78829956054688, 9.477310180664062, 80.72503662109375, 106.00735473632812, -12.659393310546875, 15.287567138671875, -14.179218292236328, 70.64825439453125, 53.44325256347656, -54.98030090332031, 56.64205551147461, 67.52197265625, 108.31607055664062, 149.27255249023438, 24.6446533203125, 8.03582763671875, 53.377716064453125, 36.24261474609375, 135.9864501953125, -4.217536926269531, 109.84383392333984, 68.94061279296875, -3.8907012939453125, 3.54302978515625, 3.9429931640625, 124.58567810058594, 7.80633544921875, -15.258358001708984, -4.599578857421875, -40.073760986328125, 20.98773193359375, -61.698394775390625, -107.50042724609375, 15.225517272949219, -2.25164794921875, 27.48100471496582, 13.612386703491211, 39.00567626953125, 141.13275146484375, -90.11663818359375, 58.94879150390625, -3.0090904235839844, -3.2710189819335938, -13.70892333984375, 121.11279296875, 107.98853302001953, -36.514190673828125, 93.77037048339844, 102.03633880615234, -34.5892333984375, 1.6668014526367188, 14.815643310546875, 16.428741455078125, 46.5291748046875, 17.397705078125, -43.99169921875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000384.npy"}
|
|
{"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 37.49778366088867, "std": 59.61392593383789, "min": -116.02398681640625, "p10": -27.669074249267577, "median": 30.659019470214844, "p90": 119.60561065673828, "max": 171.78082275390625, "pos_frac": 0.734375, "sample": [-35.612701416015625, 63.93883514404297, 122.66104125976562, 106.25115966796875, 37.2276611328125, 24.6766357421875, 116.8242416381836, 24.4696044921875, 98.88058471679688, 104.52815246582031, 27.475921630859375, 2.04559326171875, -46.07923889160156, 19.592987060546875, 25.823318481445312, 30.394210815429688, 38.0546875, 88.72599792480469, -53.656005859375, -15.2548828125, 93.0008544921875, -44.747833251953125, 93.00332641601562, -49.23756408691406, 57.412078857421875, 120.69242858886719, 82.20931243896484, 8.593177795410156, 93.09039306640625, 23.041259765625, 0.6774749755859375, 159.1676025390625, 97.32489776611328, -7.43682861328125, 126.15615844726562, 171.78082275390625, 94.74847412109375, 51.99443054199219, 62.812278747558594, 161.8198699951172, 5.096343994140625, -27.419052124023438, 49.553955078125, 63.256011962890625, 42.26556396484375, -85.49623107910156, 98.83001708984375, 90.98255920410156, -89.48764038085938, 59.72882080078125, 156.70700073242188, 132.19790649414062, 168.81207275390625, -6.967857360839844, 90.02471160888672, 17.367019653320312, 33.398529052734375, 2.2342662811279297, -116.02398681640625, 18.226661682128906, 17.440185546875, 134.06561279296875, -17.412841796875, 119.13983154296875, 14.4775390625, 0.315032958984375, -26.59393310546875, 86.211181640625, -5.6035308837890625, 76.73446655273438, 65.78707885742188, 2.2385482788085938, -27.895904541015625, 8.541866302490234, 20.6534423828125, -19.838821411132812, 68.12779235839844, 104.30581665039062, 77.88809204101562, -59.168243408203125, 8.302001953125, 71.3934097290039, 104.56780242919922, 30.923828125, 74.31292724609375, 9.708389282226562, -1.429595947265625, 25.627838134765625, -111.2801513671875, 5.80241584777832, 41.913360595703125, -3.2627296447753906, -5.219207763671875, 41.09423065185547, 86.06051635742188, 39.144500732421875, 0.0, 35.14411926269531, 59.9739990234375, 22.50543212890625, -2.8687896728515625, 13.801589965820312, 43.81011962890625, -26.1038818359375, 139.79721069335938, -12.052520751953125, -65.530517578125, 80.55242919921875, -11.34503173828125, 133.53167724609375, 46.02545166015625, 0.0, -18.925506591796875, -48.659889221191406, 91.98949432373047, 1.018585205078125, 63.520050048828125, -27.571861267089844, -3.788543701171875, 2.377685546875, -14.897079467773438, 10.575359344482422, 52.021873474121094, 33.594268798828125, 85.31498718261719, 34.1591796875, 132.7747802734375, 11.542144775390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000385.npy"}
|
|
{"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 47.13800048828125, "std": 63.51978302001953, "min": -164.25918579101562, "p10": -15.470426177978515, "median": 41.038299560546875, "p90": 129.1845748901367, "max": 174.9000244140625, "pos_frac": 0.78125, "sample": [75.1563720703125, -1.7652130126953125, 3.1416473388671875, 91.17906951904297, 56.794189453125, 11.7823486328125, 71.6856689453125, 126.59713745117188, 73.84622192382812, 97.25372314453125, 107.5206069946289, 168.4908447265625, -30.880615234375, 34.89483642578125, 134.7264404296875, 29.906295776367188, 32.3387451171875, 27.9027099609375, -1.0498180389404297, 85.6986083984375, 74.5970458984375, 104.31002807617188, 17.5791015625, 51.01202392578125, -8.307861328125, -104.71162414550781, 36.42047119140625, 15.081771850585938, 105.20277404785156, 55.06317138671875, 29.14599609375, -89.21327209472656, 136.78518676757812, 33.40496826171875, -9.062057495117188, 17.66925048828125, 17.862045288085938, 18.584854125976562, 43.886138916015625, 110.51190185546875, 174.9000244140625, 85.31210327148438, 52.85420227050781, 26.699386596679688, 77.61007690429688, 65.62353515625, 137.1160888671875, -54.521636962890625, 55.93157958984375, 143.90118408203125, 51.3580322265625, -17.52313232421875, -64.4444580078125, -33.264137268066406, 145.70712280273438, 91.77879333496094, -7.06915283203125, 0.0, -42.469757080078125, -16.18370819091797, -81.21533203125, -54.23516845703125, 76.57830810546875, 39.035552978515625, 0.8012313842773438, -8.705474853515625, 43.6048583984375, -2.4723777770996094, 116.68522644042969, -133.8282928466797, 131.31695556640625, 2.0042724609375, 86.06048583984375, 152.10516357421875, 151.69815063476562, 123.08037567138672, 7.678424835205078, -6.46832275390625, 124.25596618652344, 104.45752716064453, 101.17185974121094, 8.273345947265625, -164.25918579101562, 50.23724365234375, 9.631866455078125, 126.67489624023438, 104.18212890625, 10.817138671875, 97.7740478515625, 2.5927276611328125, 37.677215576171875, 16.612346649169922, -2.3685302734375, 0.0, 134.40093994140625, 54.46742248535156, -7.208892822265625, 5.055656433105469, 69.4677734375, 93.05511474609375, 106.8150634765625, 125.53497314453125, 29.28460693359375, 120.93312072753906, 9.759025573730469, 122.22894287109375, 20.43701171875, -15.16473388671875, -0.428955078125, 111.15423583984375, 141.83172607421875, 43.041046142578125, 128.69151306152344, 70.85408782958984, 7.599365234375, 44.60443115234375, 25.742034912109375, 8.98773193359375, 37.463951110839844, 108.25216674804688, 22.47481918334961, 130.33505249023438, 10.84476089477539, 4.840843200683594, 117.98016357421875, 0.0, 95.79556274414062, 58.72076416015625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000386.npy"}
|
|
{"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 42.879966735839844, "std": 64.25191497802734, "min": -134.11419677734375, "p10": -29.727487182617185, "median": 34.127933502197266, "p90": 132.29425201416015, "max": 157.85025024414062, "pos_frac": 0.7109375, "sample": [40.491668701171875, 3.068817138671875, 97.23336791992188, 25.285572052001953, 4.855224609375, 127.994384765625, 147.3362274169922, -45.947418212890625, -96.091796875, -54.34588623046875, 0.0, 133.42471313476562, -10.05596923828125, 135.70944213867188, 84.06588745117188, 100.22845458984375, 8.717628479003906, -29.588958740234375, -10.703598022460938, 43.876953125, -12.739486694335938, 4.699836730957031, 106.65736389160156, -33.60302734375, 28.303733825683594, 125.75363159179688, -67.72584533691406, -12.0299072265625, 123.38194274902344, 0.7657947540283203, 113.93763732910156, -15.910125732421875, 108.665283203125, 38.18943786621094, 102.83335876464844, 105.83615112304688, -11.93316650390625, 25.97570037841797, 50.79484558105469, -9.1627197265625, -31.398193359375, 101.36344909667969, 8.622894287109375, -25.042083740234375, 134.84140014648438, 34.86109161376953, 150.3876953125, 36.521026611328125, -27.571870803833008, -3.7950515747070312, -1.3605422973632812, 133.93283081054688, -32.853424072265625, -11.937103271484375, 29.91192626953125, 24.135604858398438, 11.36444091796875, -63.38059997558594, 106.73504638671875, 103.23114013671875, 71.440185546875, 33.59735107421875, 48.92742919921875, 21.65765380859375, 104.81944274902344, -103.1724853515625, 37.339996337890625, 40.585662841796875, 8.28594970703125, -11.676441192626953, -4.9178619384765625, -12.02520751953125, 43.90252685546875, 135.50811767578125, 109.35113525390625, 78.91688537597656, 61.41557312011719, 34.65851593017578, -8.672073364257812, 73.800537109375, 15.011680603027344, 0.32781219482421875, 125.83514404296875, 14.007598876953125, 58.88275146484375, 131.8097686767578, 92.08301544189453, 2.989410400390625, 63.3306884765625, 105.2006607055664, 5.350189208984375, 124.81197357177734, 38.76953125, 21.236724853515625, 155.84674072265625, 77.2130126953125, 101.86154174804688, -30.05072021484375, 111.364990234375, 121.46534729003906, 8.648687362670898, 29.57696533203125, -25.70489501953125, 106.1983871459961, -5.851404190063477, -55.296966552734375, -134.11419677734375, -19.598621368408203, 105.28982543945312, 157.85025024414062, 20.02734375, 14.793655395507812, 50.965789794921875, 72.96041870117188, 144.11767578125, 105.60404968261719, 102.20054626464844, 23.24151611328125, 149.91421508789062, 13.945365905761719, -12.548858642578125, -74.55587768554688, 133.77169799804688, 141.958984375, -2.0900344848632812, 109.57818603515625, -22.319122314453125, 92.16650390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000387.npy"}
|
|
{"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 32.755374908447266, "std": 62.855831146240234, "min": -135.015869140625, "p10": -41.597119140625, "median": 20.370580673217773, "p90": 120.88406372070312, "max": 160.38018798828125, "pos_frac": 0.703125, "sample": [57.8984375, -59.165802001953125, 98.9619140625, 0.0, 136.3502197265625, 108.93794250488281, -4.73974609375, 132.85256958007812, 18.234664916992188, 112.8082504272461, -88.74069213867188, 101.24690246582031, -88.04721069335938, 0.47864341735839844, 7.0363006591796875, 21.568023681640625, 24.052101135253906, 84.28802490234375, 152.19485473632812, -19.44097900390625, 88.76707458496094, 160.38018798828125, -69.79434204101562, -11.530059814453125, -1.78118896484375, -23.383651733398438, 50.64807891845703, 88.15191650390625, 126.38436889648438, 123.6798095703125, 30.171707153320312, 14.379005432128906, 0.0, 5.765106201171875, 16.092666625976562, -50.4122314453125, -11.3140869140625, 13.629302978515625, 73.09255981445312, 59.05769348144531, -42.16136169433594, 17.28398895263672, 66.56181335449219, 105.57987976074219, -21.901443481445312, -39.525543212890625, 128.32513427734375, -15.762039184570312, -6.089763641357422, 2.7280731201171875, 71.55419921875, 4.607208251953125, 87.68820190429688, -30.037124633789062, 10.49822998046875, 87.6523208618164, -8.156295776367188, 125.20516967773438, 57.93548583984375, -35.1168212890625, -15.922332763671875, 39.87678527832031, 104.20018005371094, 13.574356079101562, 15.79636001586914, -46.6800537109375, 63.365753173828125, -130.46273803710938, 139.17941284179688, -63.2496337890625, 46.32525634765625, 112.47946166992188, 81.86929321289062, 69.3708267211914, 25.587493896484375, -1.205718994140625, 1.83489990234375, 120.8262939453125, 6.773628234863281, -100.2619400024414, -13.649017333984375, 130.08419799804688, 121.01885986328125, -13.452301025390625, 19.173137664794922, 6.2307891845703125, 30.494598388671875, 93.88619995117188, 59.33396911621094, 31.084552764892578, -55.94415283203125, 24.713890075683594, 16.787155151367188, -135.015869140625, 30.896636962890625, 134.01495361328125, 114.014404296875, 100.95250701904297, 30.641258239746094, 91.85244750976562, 6.745567321777344, 86.6376953125, 119.1881103515625, 146.23419189453125, -31.510833740234375, 11.92388916015625, 43.378692626953125, 22.657073974609375, -25.760406494140625, 1.7515735626220703, 112.58340454101562, 11.833587646484375, -17.766647338867188, 13.408416748046875, 14.677831649780273, -25.595382690429688, 92.4765625, 7.470367431640625, 94.818359375, 41.483795166015625, -57.830848693847656, -41.35530090332031, 31.1685791015625, 6.05963134765625, 28.677688598632812, 97.88198852539062, 77.12024688720703, -21.69317626953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000388.npy"}
|
|
{"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 43.54994583129883, "std": 60.39097595214844, "min": -110.13250732421875, "p10": -27.292977905273435, "median": 43.27998352050781, "p90": 118.54799118041993, "max": 148.17401123046875, "pos_frac": 0.7578125, "sample": [0.188720703125, -57.65873718261719, -12.177879333496094, 78.39260864257812, 92.27775573730469, 107.69189453125, 30.77642822265625, 26.14349365234375, 0.6320877075195312, 123.7950439453125, 42.79681396484375, 121.68360900878906, 31.15838623046875, 148.17401123046875, 35.80833435058594, 108.62500762939453, -0.0492095947265625, 106.2242431640625, 69.42639923095703, 144.85296630859375, -49.84330749511719, 91.74898529052734, 126.6910400390625, 70.92037963867188, 2.8008041381835938, 105.69036865234375, -87.342041015625, 135.2720947265625, 30.573593139648438, 130.23870849609375, 22.480712890625, -110.13250732421875, 90.24862670898438, -33.32941436767578, -17.38458251953125, -4.258270263671875, -30.54499053955078, 88.61882781982422, 135.52059936523438, 109.69491577148438, 105.24333190917969, 0.0, -70.41867065429688, 5.158531188964844, 84.99691772460938, -55.276123046875, 118.66928100585938, 99.6768798828125, 110.07102966308594, 19.179901123046875, -6.259956359863281, 109.72303771972656, 23.762908935546875, 10.685401916503906, 88.47317504882812, 62.67578125, 23.201034545898438, 25.355194091796875, -29.577911376953125, -3.422332763671875, 32.702056884765625, 23.65584945678711, 50.27961730957031, 96.34568786621094, -41.9134521484375, 96.45279693603516, 74.36520385742188, 94.15037536621094, 73.97406005859375, 54.301483154296875, 105.14532470703125, -6.43096923828125, 38.1929931640625, 118.49600982666016, 3.974029541015625, 137.16476440429688, 55.38047790527344, 56.2479248046875, -26.313720703125, -2.322509765625, 63.849212646484375, 80.59136962890625, 12.094429016113281, 43.763153076171875, 7.94866943359375, 116.48876953125, -100.57154846191406, 93.69905090332031, 75.68588256835938, -21.3260498046875, 131.6015625, 38.6788330078125, 4.2127685546875, 2.587421417236328, 1.5995712280273438, 78.73450469970703, 6.532958984375, -103.25370788574219, -22.488555908203125, 126.08900451660156, -21.361114501953125, 83.574951171875, -15.41265869140625, -26.279403686523438, 65.13946533203125, 117.59765625, 0.0, 55.27848815917969, 83.61643981933594, 137.2802276611328, 8.769800186157227, 16.284194946289062, -3.751678466796875, 44.67181396484375, 66.88902282714844, 110.39297485351562, 79.848876953125, 16.24163818359375, -15.143829345703125, 13.978309631347656, 58.91813659667969, 103.82203674316406, -100.40304565429688, 9.818954467773438, 77.09454345703125, 7.414794921875, 94.25033569335938, 107.15338134765625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000389.npy"}
|
|
{"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 42.125064849853516, "std": 65.95372009277344, "min": -120.09664916992188, "p10": -46.81476440429687, "median": 42.445167541503906, "p90": 123.28185729980468, "max": 152.161865234375, "pos_frac": 0.75, "sample": [120.443603515625, 2.1603240966796875, 30.548065185546875, 14.25640869140625, 29.0316162109375, 96.014892578125, 1.502664566040039, 111.06887817382812, 78.61083984375, 42.365020751953125, -29.219314575195312, 7.8749237060546875, 3.6251602172851562, -24.10211181640625, 27.583831787109375, 86.62332153320312, 11.925365447998047, 130.89584350585938, 113.29426574707031, 75.031494140625, 53.87384033203125, -36.0323486328125, 122.95503234863281, 5.1475372314453125, 56.7325439453125, 53.623382568359375, -80.33477783203125, 108.588623046875, 101.072021484375, 54.91937255859375, 119.86699676513672, 86.02487182617188, -74.37855529785156, 27.44219207763672, 81.48861694335938, 39.00494384765625, 61.99085998535156, 23.883468627929688, 1.0778350830078125, 62.16911315917969, -76.33160400390625, 95.98344421386719, -52.138336181640625, 87.75189208984375, 61.124671936035156, -72.61953735351562, -19.229278564453125, -120.09664916992188, 92.50827026367188, -15.82855224609375, 49.035125732421875, -14.412628173828125, -92.68858337402344, 30.975433349609375, 146.15103149414062, 17.426166534423828, 48.267669677734375, 124.6821517944336, 83.15286254882812, -110.57601928710938, -44.533233642578125, 80.39578247070312, 124.04444885253906, 28.39105987548828, 141.31314086914062, -100.5186767578125, 152.161865234375, 70.5618896484375, -12.702301025390625, 99.45257568359375, 56.88258361816406, 136.65142822265625, 85.70556640625, 0.0, -6.895843505859375, -4.686370849609375, 120.2509765625, -7.8387908935546875, 103.28665924072266, 91.69810485839844, -20.728851318359375, 93.40536499023438, 120.24981689453125, -113.743896484375, -22.005157470703125, 107.99465942382812, 148.90093994140625, 111.39289855957031, 2.911376953125, 6.945774078369141, 7.0874786376953125, 133.63851928710938, 57.927215576171875, 56.542388916015625, 33.05058288574219, 88.33185577392578, -11.088676452636719, 131.545166015625, 25.297027587890625, -55.85089111328125, 1.0176048278808594, -57.809600830078125, 120.19085693359375, 108.9832763671875, -14.180183410644531, 35.3516845703125, 137.1744384765625, 10.0040283203125, 2.749645233154297, 5.0775146484375, -42.677520751953125, 0.9875030517578125, -2.5597000122070312, 91.7239990234375, 42.52531433105469, 27.896392822265625, 121.56777954101562, 113.3089828491211, 61.86631774902344, 122.94561767578125, 105.11238098144531, 35.76995849609375, -1.92669677734375, 130.6016845703125, 132.0638427734375, -75.36117553710938, 75.24479675292969, 27.14679718017578], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000390.npy"}
|
|
{"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 39.538551330566406, "std": 57.69470977783203, "min": -112.88255310058594, "p10": -24.49851379394531, "median": 29.285350799560547, "p90": 116.6016731262207, "max": 201.2723388671875, "pos_frac": 0.734375, "sample": [8.600395202636719, -42.283897399902344, 0.3728351593017578, 40.014129638671875, 13.098669052124023, 8.93096923828125, -4.4566650390625, 7.448173522949219, 17.23563003540039, 75.51074981689453, 4.631767272949219, 10.683868408203125, -25.437835693359375, 72.67527770996094, 129.91180419921875, 38.79345703125, 28.92498779296875, 66.22557067871094, -59.45323181152344, -12.192771911621094, -15.70501708984375, 116.32787322998047, 64.063720703125, 88.12496948242188, -37.877197265625, 47.20793151855469, 115.76422119140625, 47.158538818359375, 10.893928527832031, 26.246429443359375, -4.1300048828125, -26.3402099609375, 87.89593505859375, -9.198753356933594, 36.7574462890625, 72.71246337890625, 76.70228576660156, 8.459009170532227, -11.051895141601562, 44.36395263671875, 10.410961151123047, 99.2514419555664, 103.97111511230469, 162.591552734375, 109.73757934570312, -7.2235107421875, 123.3820571899414, 2.50311279296875, -9.849788665771484, -0.01983642578125, 146.17291259765625, 62.906272888183594, -26.33539581298828, 51.57646179199219, -14.795578002929688, 81.59780883789062, 84.98524475097656, 80.48240661621094, 95.02340698242188, -28.10302734375, 49.21197509765625, -72.93941497802734, 101.8285903930664, 29.645713806152344, 27.103271484375, 11.82623291015625, 78.48388671875, 8.266986846923828, 55.639015197753906, 4.9812164306640625, -1.6822509765625, 125.20108032226562, 24.50592041015625, 61.12872314453125, 79.38116455078125, -6.784393310546875, 104.962158203125, 50.24542236328125, -7.7036285400390625, 110.69400024414062, 1.9612579345703125, 40.0924072265625, 110.020263671875, 201.2723388671875, 134.39892578125, 20.411041259765625, -6.00750732421875, -5.7937164306640625, 113.28837585449219, 11.835968017578125, 14.7579345703125, 2.2477188110351562, -112.88255310058594, 117.24053955078125, 36.998931884765625, 46.5362548828125, -24.08465576171875, 90.43400573730469, 133.42791748046875, 129.10040283203125, -38.0101318359375, -38.562591552734375, -17.377426147460938, -44.0340576171875, 1.399261474609375, -2.7566070556640625, 26.997459411621094, 73.47209167480469, 94.03656005859375, 110.93389892578125, 13.226287841796875, 85.7020492553711, 99.38582611083984, -5.2058258056640625, 40.605377197265625, 58.018768310546875, 132.00961303710938, 105.00927734375, 163.96337890625, 6.921234130859375, -24.095947265625, 8.587799072265625, 29.7437744140625, 6.448738098144531, -7.981925964355469, 36.749298095703125, -109.5684814453125, 120.19068145751953], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000391.npy"}
|
|
{"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 51.860862731933594, "std": 59.20311737060547, "min": -113.13604736328125, "p10": -11.873070526123046, "median": 39.005714416503906, "p90": 137.24613647460936, "max": 167.73919677734375, "pos_frac": 0.7734375, "sample": [118.2918701171875, 90.12871551513672, 81.48466491699219, 38.890411376953125, -13.119857788085938, 4.428794860839844, 16.24853515625, 91.927734375, 70.96661376953125, 78.15951538085938, 17.609085083007812, 64.98397827148438, 76.68374633789062, -2.96600341796875, -11.166534423828125, 4.392974853515625, -113.13604736328125, 167.73919677734375, 39.12101745605469, -15.803047180175781, 136.6204833984375, -8.063129425048828, 5.233673095703125, 91.49920654296875, 111.5118408203125, 127.4600830078125, 0.0, -30.23974609375, 106.32827758789062, 100.53839111328125, 6.282989501953125, 9.55126953125, 145.58786010742188, 25.179840087890625, 49.250144958496094, 144.41131591796875, 139.01129150390625, 120.67391967773438, 105.15451049804688, 44.73735046386719, 103.16363525390625, 123.15060424804688, 15.619636535644531, 15.1314697265625, -8.788116455078125, 17.68115234375, -11.586494445800781, 2.6046142578125, 83.73629760742188, 112.98169708251953, 68.490478515625, 124.1998291015625, 148.9834747314453, 68.75663757324219, 139.76193237304688, 97.40814208984375, 13.39593505859375, 65.95468139648438, 134.9451141357422, -5.3414764404296875, -2.7726364135742188, 76.31416320800781, -39.209503173828125, 135.0040283203125, 21.822216033935547, 39.915184020996094, -23.402389526367188, 83.99053955078125, 152.54544067382812, -1.62933349609375, 46.025543212890625, 134.06533813476562, 138.70599365234375, 87.25279998779297, 22.584503173828125, 142.30279541015625, 36.289093017578125, -40.34422302246094, -14.521980285644531, -3.5619850158691406, 16.348255157470703, 118.59254455566406, 1.11248779296875, 57.19903564453125, 31.18170166015625, -12.541748046875, -11.24298095703125, 3.514892578125, 19.216720581054688, -20.467514038085938, 2.0184326171875, 0.0, 102.01622772216797, 24.68695068359375, 121.16900634765625, 143.50738525390625, 51.582275390625, 74.46044921875, 147.60488891601562, 10.086456298828125, 129.08441162109375, -0.15891075134277344, 1.4783477783203125, 1.4461078643798828, 30.8489990234375, 164.25143432617188, -2.9934005737304688, 144.4100341796875, 18.692657470703125, 60.37665557861328, -70.55006408691406, 117.32745361328125, -2.8717193603515625, 14.47314453125, 100.28579711914062, 30.0882568359375, 53.41015625, -36.1473388671875, 12.135009765625, 83.26689147949219, 20.067520141601562, 102.52288818359375, -6.119964599609375, 110.7962646484375, -19.4857177734375, 20.233787536621094, 111.28543090820312, 32.770782470703125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000392.npy"}
|
|
{"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 37.73059844970703, "std": 61.105655670166016, "min": -144.92977905273438, "p10": -31.69817199707031, "median": 31.947669982910156, "p90": 115.9107177734375, "max": 195.9263916015625, "pos_frac": 0.7421875, "sample": [39.431522369384766, -31.31201171875, 4.402252197265625, 55.065887451171875, 70.53363800048828, -54.74847412109375, 50.085105895996094, 46.689361572265625, 106.63409423828125, 21.160980224609375, 8.2435302734375, -4.144805908203125, 38.98097229003906, -3.0317306518554688, 96.65499114990234, 27.828460693359375, -86.18763732910156, 70.48368835449219, 34.64514923095703, 74.03094482421875, 115.44949340820312, 50.107421875, 5.8773040771484375, 39.92686462402344, -38.9586181640625, 126.2700424194336, 115.17173767089844, 111.32544708251953, 31.112411499023438, 14.96563720703125, 105.517578125, 4.3777923583984375, 33.51104736328125, 110.01535034179688, 32.782928466796875, 36.202789306640625, -3.7420654296875, 124.13607788085938, 20.571868896484375, 104.48269653320312, 22.10394287109375, -9.82086181640625, 85.5592041015625, 119.40933227539062, -64.93148803710938, 120.8726806640625, -20.094680786132812, -28.368011474609375, -63.68199157714844, 75.06332397460938, 195.9263916015625, 44.687252044677734, 9.587112426757812, 17.05382537841797, 13.873809814453125, 55.1986083984375, 0.7708797454833984, -38.675384521484375, 108.36843872070312, 58.641204833984375, -112.76913452148438, 14.967620849609375, 18.532089233398438, -13.291046142578125, 19.9488525390625, 111.74649047851562, 76.5782470703125, 117.58810424804688, -63.879638671875, -25.53546142578125, -32.599212646484375, 3.380939483642578, 8.126556396484375, 47.45576477050781, 14.04892349243164, 59.32635498046875, 147.67495727539062, -19.145477294921875, 49.80072021484375, -9.103446960449219, 7.678661346435547, -6.8345489501953125, 101.21514892578125, -118.10063171386719, 25.022375106811523, 71.52549743652344, 109.32746887207031, 87.57777404785156, 102.78939819335938, 114.13313293457031, -22.19952392578125, 119.78292846679688, 97.95484924316406, 123.13217163085938, 117.09974670410156, 33.265167236328125, 27.6959228515625, -33.929443359375, 106.49440002441406, -144.92977905273438, 14.604896545410156, 29.394485473632812, 5.99658203125, -3.1595230102539062, 161.612548828125, 59.4190673828125, 50.851715087890625, 0.0, 10.8294677734375, 51.74102783203125, 155.05044555664062, -37.57012939453125, 5.160285949707031, 8.394699096679688, -12.23297119140625, 76.26173400878906, 109.58274841308594, -11.867464065551758, -20.500595092773438, -9.08563232421875, 0.8764915466308594, -20.525543212890625, 107.91151428222656, 116.98690795898438, 1.3171920776367188, 55.71234130859375, 76.09432983398438, 109.01378631591797], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000393.npy"}
|
|
{"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 35.871803283691406, "std": 59.444576263427734, "min": -102.93437194824219, "p10": -31.705602264404295, "median": 20.071548461914062, "p90": 123.71177978515624, "max": 163.78598022460938, "pos_frac": 0.6796875, "sample": [127.41262817382812, 15.352874755859375, 128.62026977539062, 82.22418212890625, 46.32977294921875, 98.94483947753906, -8.27374267578125, -4.418537139892578, 9.821990966796875, 126.6993408203125, 65.97161865234375, -5.6379852294921875, 27.8531494140625, 76.24261474609375, -11.63922119140625, -2.0319862365722656, 110.43901062011719, -32.32910919189453, 103.3563232421875, -22.43552017211914, 7.9718017578125, 112.88616943359375, 154.08648681640625, -20.36944580078125, 1.348886489868164, -69.39347839355469, 13.494293212890625, 72.34999084472656, 56.53466796875, 38.4569091796875, -21.337432861328125, 0.0, -34.20686340332031, -51.863037109375, 129.95361328125, 11.66485595703125, 16.803558349609375, 77.22220611572266, 3.463623046875, 28.2979736328125, 61.267333984375, 73.44154357910156, 45.63446044921875, -102.93437194824219, 23.133636474609375, 85.93032836914062, 116.40071868896484, 28.659854888916016, 1.8017578125, 25.12152099609375, -5.765960693359375, -34.202667236328125, 50.016326904296875, -14.54034423828125, -23.589599609375, 131.4254150390625, 21.021728515625, 13.7110595703125, 7.321174621582031, 48.2493896484375, 52.963409423828125, -43.626495361328125, -0.5740509033203125, -6.610321044921875, 121.92724609375, 4.051153182983398, 0.9603652954101562, 71.16146850585938, 16.241294860839844, 55.09619140625, 107.42903137207031, 107.49125671386719, 130.02719116210938, 6.007915496826172, 154.23736572265625, -3.60107421875, -56.042022705078125, -22.53082275390625, -7.1819915771484375, 154.17007446289062, 18.289730072021484, -82.85894775390625, 96.1314697265625, -17.9722900390625, -41.338226318359375, 19.121368408203125, 35.48486328125, 27.860397338867188, 5.439159393310547, 63.490570068359375, -2.45391845703125, -3.9819412231445312, 47.225982666015625, 16.153594970703125, -25.098175048828125, -31.335662841796875, 10.6373291015625, 31.86517333984375, 66.39265441894531, 47.23541259765625, -79.56901550292969, 106.79736328125, 14.90216064453125, -19.966171264648438, 128.79306030273438, 163.78598022460938, -16.917083740234375, 37.13629150390625, 80.40802001953125, 128.44464111328125, 88.781494140625, 94.61422729492188, 113.752197265625, 111.1563720703125, 122.431396484375, -41.171722412109375, 140.205322265625, 17.11474609375, -23.771270751953125, 103.90182495117188, 11.6583251953125, 98.87078857421875, -44.720123291015625, 78.11515045166016, -25.92534637451172, -6.3222808837890625, -31.438385009765625, 106.66644287109375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000394.npy"}
|
|
{"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 38.918617248535156, "std": 58.24336624145508, "min": -94.44607543945312, "p10": -23.92043762207031, "median": 26.475631713867188, "p90": 128.24209899902343, "max": 159.19970703125, "pos_frac": 0.7265625, "sample": [40.98272705078125, 6.33685302734375, -6.879112243652344, -1.7831878662109375, -0.940338134765625, -16.053131103515625, 1.49993896484375, 151.6204833984375, 145.07257080078125, 29.718093872070312, 26.135589599609375, 55.87579345703125, 26.815673828125, -18.845458984375, 1.3203086853027344, 24.163375854492188, -4.060577392578125, -30.650604248046875, 40.06343078613281, 7.71282958984375, 22.129852294921875, 112.1776123046875, -16.630531311035156, -40.417236328125, 133.60342407226562, 69.35635375976562, 6.872650146484375, -5.219478607177734, 83.88079833984375, 46.08935546875, 95.54177856445312, 72.57183837890625, 79.49441528320312, 73.9681396484375, 25.2803955078125, 123.43153381347656, 41.40350341796875, 37.34124755859375, 0.6431350708007812, -12.537670135498047, 55.2812385559082, 12.037101745605469, -12.27294921875, -16.614654541015625, 104.80905151367188, 74.61454772949219, -6.3177032470703125, -49.085968017578125, 93.09099578857422, 129.70046997070312, 141.02374267578125, -72.73025512695312, 56.422576904296875, 110.47201538085938, 128.06918334960938, 76.10049438476562, 51.737030029296875, 128.64556884765625, 10.01849365234375, 61.62580871582031, -27.284133911132812, -57.24761962890625, 16.112258911132812, -3.9007415771484375, 109.82550048828125, -20.24810791015625, 24.49681854248047, 51.71141815185547, -14.064483642578125, 9.454010009765625, -25.26055908203125, 19.3663330078125, -3.78009033203125, -34.619842529296875, 31.058837890625, 83.37495422363281, 86.92289733886719, -2.5231781005859375, 54.054779052734375, 18.1539306640625, 22.6607666015625, 15.519706726074219, 159.19970703125, 125.99542236328125, 116.86802673339844, -23.346099853515625, -1.0896759033203125, -60.31494140625, 15.0469970703125, 36.2452392578125, 6.262022018432617, 44.82176971435547, -86.14117431640625, 64.96394348144531, -7.06024169921875, 29.383392333984375, 139.072265625, 135.48081970214844, 108.85664367675781, 128.73532104492188, -4.647804260253906, 70.27218627929688, 17.607574462890625, -6.0520172119140625, 80.87918090820312, 57.0626220703125, 120.95646667480469, 7.863861083984375, 114.19944763183594, -83.09400939941406, 5.84674072265625, 107.24739074707031, 132.5145263671875, -50.914794921875, 2.917572021484375, -94.44607543945312, 29.1058349609375, 7.378835678100586, 128.65869140625, 132.5843505859375, 124.31488037109375, 126.71128845214844, 28.38397216796875, 2.9033203125, 23.089126586914062, 47.582481384277344, 8.276840209960938, 53.90478515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000395.npy"}
|
|
{"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 40.69496154785156, "std": 61.933101654052734, "min": -151.0516357421875, "p10": -41.56984405517578, "median": 41.14861297607422, "p90": 121.8006591796875, "max": 167.70269775390625, "pos_frac": 0.7265625, "sample": [94.90372467041016, 19.026519775390625, -7.3897705078125, 122.83245849609375, 102.58627319335938, 10.44622802734375, -6.5407257080078125, 14.004436492919922, -40.4989013671875, 133.05648803710938, 58.37335205078125, -23.762283325195312, 48.31535339355469, 40.710205078125, 3.0325775146484375, 51.688232421875, 138.78265380859375, -45.2005615234375, 7.601142883300781, 82.95608520507812, -33.46659851074219, -9.610496520996094, -0.681121826171875, 167.70269775390625, -3.9531402587890625, 14.498077392578125, 72.24029541015625, 105.67868041992188, 68.84968566894531, 78.29940032958984, 58.126953125, -73.42387390136719, 56.20770263671875, 96.79745483398438, -86.80517578125, 116.12600708007812, 22.29168701171875, -3.10699462890625, -4.846385955810547, 52.81025695800781, 25.5341796875, 101.66979217529297, 63.829437255859375, 93.53466796875, 137.05108642578125, 17.87835693359375, 6.785708427429199, 10.434234619140625, -5.0927734375, 91.72003936767578, 144.50112915039062, 131.3739013671875, 152.68045043945312, 5.91375732421875, 112.1243667602539, 49.121185302734375, 113.18433380126953, 5.532661437988281, 13.80731201171875, 14.267263412475586, 7.0446624755859375, -21.8505859375, -44.11237335205078, -44.06871032714844, 73.31063842773438, -50.82585144042969, 49.873504638671875, -99.59382629394531, 14.025070190429688, 28.219802856445312, 104.77843475341797, -5.48760986328125, -29.226959228515625, 47.88201141357422, 105.9810791015625, 80.09246826171875, 50.89508056640625, 95.84860229492188, 20.193023681640625, -111.44786071777344, 51.89642333984375, 79.78815460205078, 45.88678741455078, 98.51861572265625, 55.58221435546875, 111.62432861328125, 23.436023712158203, 136.9871063232422, -51.724571228027344, 108.6202392578125, -9.18341064453125, 89.6419677734375, -36.02545166015625, 44.206459045410156, 41.58702087402344, 78.55024719238281, 103.50604248046875, 38.46746826171875, 8.561004638671875, 38.021728515625, 123.61273193359375, 96.35499572753906, -1.611785888671875, -151.0516357421875, -7.1856842041015625, 125.20393371582031, -6.549064636230469, 131.90377807617188, 121.35845947265625, 127.92843627929688, 0.0, 19.690155029296875, 114.16175842285156, 28.607208251953125, 25.663375854492188, 7.911476135253906, 0.0, 42.214630126953125, 66.87794494628906, -6.2382659912109375, 40.57786560058594, -72.63836669921875, 59.67205810546875, 116.56852722167969, 105.21420288085938, -52.68359375, -46.23179626464844, 85.63478088378906], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000396.npy"}
|
|
{"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 38.529685974121094, "std": 62.74658203125, "min": -115.18157958984375, "p10": -41.53666687011717, "median": 36.45622253417969, "p90": 122.7382598876953, "max": 156.00363159179688, "pos_frac": 0.703125, "sample": [-6.820404052734375, 7.6778106689453125, -10.519760131835938, 12.634262084960938, 130.88232421875, 48.141876220703125, -62.113807678222656, 41.50531005859375, -6.36207389831543, 34.90545654296875, 7.626739501953125, 20.3922119140625, 48.750762939453125, 6.51580810546875, 27.555099487304688, 125.79290771484375, 43.80260467529297, 86.8113021850586, 147.011962890625, 95.73680114746094, -77.21177673339844, 22.547454833984375, -26.7435302734375, 64.55198669433594, 91.79186248779297, 110.94097900390625, 0.0, -6.07049560546875, 92.651611328125, 39.493560791015625, 0.0, -21.46063232421875, 91.46304321289062, -29.171188354492188, -1.44903564453125, 2.749237060546875, 11.65594482421875, 110.55720520019531, -69.9990234375, 105.63290405273438, 131.3489990234375, 23.74927520751953, 98.369873046875, 125.04254150390625, -24.547760009765625, 3.5033187866210938, 97.68814086914062, 114.6280517578125, 154.75982666015625, -26.879852294921875, 156.00363159179688, -50.437591552734375, 130.65478515625, -115.18157958984375, -53.095550537109375, -37.72198486328125, 50.05328369140625, -93.09129333496094, -13.934635162353516, -105.69403076171875, -31.282958984375, 46.46730041503906, 118.01080322265625, 104.31263732910156, 52.19932556152344, 28.816009521484375, 88.10162353515625, -7.86065673828125, 25.831069946289062, -1.4350051879882812, 49.02867126464844, 136.21337890625, 68.20169067382812, 76.56610107421875, 109.2594223022461, 104.3892822265625, 99.9273452758789, -58.100341796875, 56.375732421875, 45.101226806640625, 108.46513366699219, 31.208251953125, 130.76873779296875, 126.34832763671875, 37.124420166015625, 0.9970207214355469, 102.82852172851562, -0.4583740234375, 12.452239990234375, 47.505218505859375, 97.170166015625, 17.750885009765625, 68.35777282714844, 11.371929168701172, 114.33819580078125, 106.19256591796875, 32.10581970214844, 25.390701293945312, 124.68673706054688, 88.9239501953125, 13.9842529296875, -51.18536376953125, 37.5230712890625, -12.983657836914062, 133.1983642578125, 110.510498046875, 78.13615417480469, 2.40020751953125, -12.199363708496094, -16.839828491210938, -22.60528564453125, 121.9031982421875, 95.72650146484375, -22.74945068359375, 36.989288330078125, 35.92315673828125, -74.96401977539062, -12.611968994140625, 98.34059143066406, 17.20599365234375, -36.97418212890625, -51.995391845703125, 57.41864013671875, 59.44732666015625, 25.92669677734375, -91.71746826171875, 86.776611328125, 56.48963928222656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000397.npy"}
|
|
{"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 28.934682846069336, "std": 68.76363372802734, "min": -132.9288330078125, "p10": -50.682092285156244, "median": 16.098514556884766, "p90": 123.79283142089844, "max": 189.4251708984375, "pos_frac": 0.65625, "sample": [13.7596435546875, 23.096832275390625, 109.77003479003906, -10.591949462890625, 189.4251708984375, 1.99151611328125, 47.767303466796875, -39.07470703125, 38.08464050292969, 122.22210693359375, -92.67892456054688, -49.8896484375, -117.140869140625, 24.387664794921875, 4.571098327636719, -5.38226318359375, 30.78607177734375, -3.4842453002929688, 117.8106689453125, 129.01077270507812, 118.52537536621094, -100.48187255859375, -71.6630859375, 96.23388671875, 34.010498046875, 84.71469116210938, 12.980758666992188, 87.99371337890625, 108.32717895507812, 22.40667724609375, -62.45640563964844, 11.67251205444336, 37.685394287109375, -132.9288330078125, -52.5311279296875, 62.147483825683594, 2.07366943359375, 98.26249694824219, 5.1004638671875, -75.91680908203125, 1.474893569946289, -47.38958740234375, 36.54534912109375, 97.88644409179688, -0.3502197265625, -47.3369140625, -12.51068115234375, 107.59660339355469, 16.497314453125, -13.326202392578125, 37.470672607421875, -17.34588623046875, 58.26664733886719, -19.266616821289062, 80.75955200195312, 36.1417236328125, 17.50503921508789, -87.22314453125, 24.533706665039062, -11.284517288208008, 9.77154541015625, 42.698699951171875, 124.07150268554688, -19.71258544921875, 132.11981201171875, 11.270965576171875, 15.699714660644531, -100.0870361328125, 91.23277282714844, 131.0711669921875, 33.709991455078125, 151.94354248046875, 4.2972564697265625, 157.668701171875, -12.820785522460938, -18.381332397460938, 8.966299057006836, 127.0855712890625, -24.97747802734375, -11.150115966796875, 131.41494750976562, 136.86019897460938, 6.6726837158203125, 32.67942810058594, 34.550537109375, 3.9266815185546875, -37.49314880371094, 108.05436706542969, 33.78729248046875, -38.88958740234375, 95.880615234375, 108.86297607421875, -14.467575073242188, -108.37814331054688, 11.5740966796875, -6.7987518310546875, 99.09130859375, -22.271026611328125, -6.7774658203125, 75.24374389648438, -17.16632843017578, -6.2806396484375, 103.70428466796875, 81.26237487792969, 10.25, 5.040557861328125, 81.01338195800781, 123.67340087890625, 121.52603149414062, -24.139495849609375, -101.69759368896484, 133.96029663085938, 97.0576171875, -41.514366149902344, 114.18023681640625, -7.7278594970703125, -2.837615966796875, -122.72732543945312, 12.636878967285156, 134.01034545898438, 69.64654541015625, -6.95556640625, 6.563377380371094, 160.64260864257812, 51.34625244140625, 24.95648193359375, 87.17405700683594, 38.798187255859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000398.npy"}
|
|
{"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 35.5391845703125, "std": 59.63016128540039, "min": -122.20791625976562, "p10": -30.60943603515624, "median": 25.265548706054688, "p90": 117.5293212890625, "max": 158.7437744140625, "pos_frac": 0.7109375, "sample": [-9.159500122070312, 103.89242553710938, 105.4652099609375, 103.72659301757812, 8.5657958984375, -23.628665924072266, 8.555267333984375, 27.8843994140625, 10.501220703125, 130.8248291015625, 52.16625213623047, -37.04951477050781, -40.90000915527344, 113.17778015136719, -2.847442626953125, 33.4185791015625, 66.31866455078125, 5.202728271484375, 122.43562316894531, 33.67115783691406, -0.92364501953125, -70.85041809082031, -15.854522705078125, 118.98358154296875, 9.20172119140625, -1.75030517578125, 21.293258666992188, -90.96527862548828, -121.46310424804688, 38.21881103515625, -3.01263427734375, 114.77828979492188, 16.76885986328125, 123.9586181640625, 16.579261779785156, -7.50201416015625, 107.54226684570312, -7.632209777832031, 16.67218017578125, 130.06304931640625, 29.29656982421875, 91.38296508789062, 158.7437744140625, 116.90606689453125, 33.4727783203125, -9.88458251953125, 114.74472045898438, -56.074676513671875, 11.597108840942383, 104.39813232421875, -36.91259765625, 44.81536102294922, 77.25131225585938, 76.7904052734375, 142.19635009765625, -5.312255859375, 17.220287322998047, 12.25189208984375, -8.472282409667969, -20.046382904052734, 39.95353698730469, 123.06048583984375, 39.96247863769531, 21.925392150878906, 99.90506744384766, 15.279071807861328, 106.51930236816406, 22.447599411010742, 55.44880676269531, 39.317138671875, 51.5062255859375, 70.85281372070312, -11.676055908203125, -85.91111755371094, -122.20791625976562, -69.95962524414062, -10.82452392578125, 35.5550537109375, 127.5081787109375, 114.98529052734375, 19.826461791992188, 105.78741455078125, 65.1195068359375, 87.94329833984375, 90.92184448242188, 53.19891357421875, 14.143646240234375, -13.523544311523438, 58.1246337890625, 67.58697509765625, 15.234575271606445, -24.66802978515625, 12.046550750732422, 64.73895263671875, 9.476081848144531, 22.083541870117188, -51.00639343261719, -1.6788749694824219, 125.43130493164062, 7.45428466796875, 68.6317138671875, 7.231719970703125, 46.90997314453125, -47.52520751953125, 23.57342529296875, 124.99349975585938, -18.091461181640625, 150.01324462890625, 91.38027954101562, 26.957672119140625, 131.85194396972656, -93.14364624023438, -27.9080810546875, 32.10331726074219, 22.48296546936035, 22.536941528320312, 62.08917236328125, 89.46401977539062, -15.581558227539062, 101.05766296386719, -21.464599609375, 42.782958984375, 89.02462768554688, 74.68955993652344, -21.78955078125, -2.76739501953125, 76.82986450195312, 20.1021728515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000399.npy"}
|
|
{"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 41.25928497314453, "std": 61.06866455078125, "min": -127.66395568847656, "p10": -31.165643310546873, "median": 30.055095672607422, "p90": 121.78238830566406, "max": 150.1201171875, "pos_frac": 0.7578125, "sample": [7.233001708984375, -27.684036254882812, 16.479461669921875, 37.19140625, 55.80645751953125, 19.112464904785156, 86.21294403076172, 86.7470703125, 18.6229248046875, -68.9475326538086, 10.760639190673828, 72.50039672851562, -127.66395568847656, 42.173919677734375, 126.27287292480469, 10.77606201171875, 7.490425109863281, 97.82685852050781, 88.42414855957031, 71.427734375, 148.14056396484375, 13.507568359375, 29.682762145996094, 59.095001220703125, 115.2078857421875, 72.71182250976562, -39.401641845703125, 71.38807678222656, -17.3421630859375, 74.42608642578125, 41.2435302734375, 5.7774658203125, 20.48863983154297, 92.40428161621094, 35.080322265625, 122.68914794921875, 121.4254150390625, 92.05670166015625, 21.528305053710938, 27.572547912597656, -94.64935302734375, 103.46465301513672, 94.25953674316406, 120.53646850585938, 13.0418701171875, 106.81964111328125, -0.174560546875, 49.06121826171875, 13.206085205078125, 13.109405517578125, 30.42742919921875, 99.30713653564453, 129.56253051757812, -79.7454833984375, 2.96630859375, -26.576705932617188, 12.008197784423828, 24.343421936035156, 124.64144897460938, 2.479677200317383, -1.1825408935546875, 66.88992309570312, 111.60317993164062, 96.93017578125, 20.4622802734375, -30.156402587890625, 17.12518310546875, 114.2575912475586, 81.9769287109375, 135.24147033691406, -13.77606201171875, 11.586982727050781, -20.125946044921875, 11.474014282226562, -35.2003173828125, -67.44448852539062, 122.61532592773438, 134.8466033935547, 5.608551025390625, -35.806671142578125, 61.90263366699219, 95.36886596679688, 95.51031494140625, 72.18218994140625, -3.3453025817871094, 126.45416259765625, 97.85835266113281, -21.521820068359375, 35.27754211425781, 37.772621154785156, 67.52425384521484, -17.97967529296875, -8.722915649414062, 5.130455017089844, 0.682373046875, 104.58108520507812, -16.728012084960938, 87.79280090332031, 2.878899574279785, -71.4771728515625, 26.5076904296875, 122.77778625488281, 98.72940826416016, 23.307342529296875, -33.520538330078125, 14.0555419921875, 96.1341552734375, -1.6114311218261719, -5.992462158203125, 117.06977844238281, 106.777099609375, 0.15612030029296875, 114.97225952148438, -9.89495849609375, 114.10723876953125, 106.75755310058594, 114.20831298828125, -12.196304321289062, 150.1201171875, 123.23420715332031, 127.26863098144531, 113.3106689453125, 0.0, -61.19134521484375, -64.65472412109375, 6.21563720703125, -95.078857421875, 35.01789093017578], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000400.npy"}
|
|
{"epoch": 0.837696335078534, "step": 401, "batch_size": 128, "mean": 50.1112060546875, "std": 59.30738830566406, "min": -122.32929992675781, "p10": -16.047034454345702, "median": 48.00958251953125, "p90": 122.13275756835937, "max": 180.77264404296875, "pos_frac": 0.75, "sample": [-10.705062866210938, 1.116363525390625, 32.89447021484375, -4.843719482421875, 19.32476806640625, 77.99552917480469, 55.250213623046875, 80.93949890136719, 69.89324951171875, -4.3260650634765625, 108.1519775390625, 20.722625732421875, 40.505828857421875, 123.16848754882812, 67.21188354492188, 33.89234924316406, 25.04864501953125, 101.3184814453125, -7.807708740234375, 95.32577514648438, -4.646331787109375, 86.55097961425781, -20.346694946289062, 25.373046875, -8.756011962890625, 121.68887329101562, 67.60891723632812, 1.5717506408691406, 103.8084716796875, 33.76336669921875, 138.02813720703125, 161.18450927734375, 105.68463134765625, 57.74200439453125, 12.93896484375, 12.65606689453125, 82.7135009765625, 102.10416412353516, 17.002700805664062, 104.90894317626953, 180.77264404296875, -9.649322509765625, 117.04255676269531, 126.26754760742188, 145.696044921875, 106.01296997070312, 57.09881591796875, 8.739166259765625, 101.61004638671875, 105.65890502929688, -52.02435302734375, 107.11961364746094, 48.13092041015625, 11.623062133789062, -121.191650390625, 97.58370971679688, 10.94671630859375, 95.218017578125, 71.79731750488281, -25.099899291992188, -33.34318542480469, 96.361328125, 121.52423095703125, 99.6148681640625, 81.72802734375, 71.77767944335938, -65.472412109375, -55.579307556152344, 115.4490966796875, -12.80450439453125, -2.7288970947265625, 0.8273696899414062, 148.0406494140625, 44.30499267578125, 35.81434631347656, 5.467216491699219, 108.44255065917969, 77.171875, 47.88824462890625, 51.409149169921875, 65.44775390625, -25.024017333984375, 112.03699493408203, 129.9154052734375, 103.29486083984375, 112.36126708984375, 40.825714111328125, -4.6519317626953125, 56.35235595703125, -21.964202880859375, 66.37176513671875, 25.05914306640625, -7.626251220703125, -14.069610595703125, 39.095706939697266, -15.368026733398438, -20.028106689453125, 135.63787841796875, 153.53102111816406, 24.454345703125, 26.6895751953125, -4.18231201171875, 127.953125, 113.91862487792969, 131.68801879882812, 21.933349609375, 10.275970458984375, 20.762542724609375, -19.296730041503906, 58.916595458984375, 15.645416259765625, -0.424072265625, 28.699600219726562, 4.886726379394531, -11.656097412109375, -1.2257461547851562, 52.926788330078125, 104.89982604980469, 98.39144897460938, -17.631385803222656, 94.63218688964844, 110.87773132324219, 115.78775024414062, -1.2716598510742188, 151.2727508544922, -122.32929992675781, -10.628585815429688, 111.166748046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000401.npy"}
|
|
{"epoch": 0.8397905759162304, "step": 402, "batch_size": 128, "mean": 42.022254943847656, "std": 62.76262664794922, "min": -114.08297729492188, "p10": -29.38353576660156, "median": 34.47246551513672, "p90": 120.28209686279297, "max": 182.626708984375, "pos_frac": 0.75, "sample": [-7.708778381347656, -0.5308303833007812, -89.2470703125, 40.35044860839844, 113.87686157226562, -14.983322143554688, -69.91375732421875, -18.458984375, 72.99209594726562, -111.880859375, -93.92962646484375, 97.41439819335938, 23.10394287109375, -37.528621673583984, 4.1568756103515625, 164.13973999023438, 95.16268920898438, 91.53406524658203, 42.216339111328125, -21.87640380859375, 98.10069274902344, 1.8282928466796875, 23.335960388183594, 138.2109375, -96.5885009765625, 2.6355056762695312, -4.924102783203125, 111.93575286865234, -114.08297729492188, 35.9046630859375, 83.43701171875, 19.40948486328125, 46.11656188964844, 95.07717895507812, 9.27557373046875, 0.0, 171.1103515625, 106.52023315429688, -34.285491943359375, 107.49176788330078, 47.18394470214844, 27.995880126953125, 76.62196350097656, 8.709182739257812, 69.880126953125, 109.72286224365234, -30.134185791015625, -6.70361328125, 137.82342529296875, 98.63847351074219, 23.9327392578125, -31.864837646484375, -25.992950439453125, 68.23245239257812, 83.94107055664062, 20.325592041015625, 112.98698425292969, 54.69197082519531, 2.6336441040039062, 66.72895050048828, 11.308677673339844, 151.235595703125, 123.5025634765625, 10.66241455078125, 88.57528686523438, -16.655410766601562, -44.50721740722656, 59.18548583984375, 26.70233154296875, -96.94940948486328, 118.93258666992188, 107.17888641357422, 33.04026794433594, -0.51824951171875, 102.50761413574219, -3.4587554931640625, 182.626708984375, 6.328807830810547, 60.19903564453125, 124.07872009277344, 119.64056396484375, 116.9505615234375, 10.08026123046875, 82.3742446899414, 49.62794494628906, 32.63275146484375, 130.50958251953125, 84.50311279296875, -23.298980712890625, 98.98736572265625, 55.23664093017578, 28.70983123779297, 11.265663146972656, -0.7660446166992188, 39.65863037109375, 59.61669921875, 6.802742004394531, 71.72412109375, 0.8240966796875, 121.6103515625, 123.68484497070312, 119.71284484863281, 92.26846313476562, -29.06182861328125, 153.995361328125, -10.170028686523438, 37.15303039550781, 10.51837158203125, 31.78594970703125, 116.63789367675781, 30.605079650878906, 99.42672729492188, 83.7744140625, 81.11721801757812, -39.98419189453125, 5.93963623046875, -2.3650054931640625, 60.050048828125, 24.536041259765625, 7.301849365234375, 124.1298828125, 1.1282386779785156, 6.536693572998047, -28.711944580078125, 1.86260986328125, 58.12451171875, 84.74395751953125, -1.310232162475586], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000402.npy"}
|
|
{"epoch": 0.8418848167539267, "step": 403, "batch_size": 128, "mean": 54.67933654785156, "std": 55.570499420166016, "min": -90.46023559570312, "p10": -10.470465087890625, "median": 50.18894577026367, "p90": 125.82680816650391, "max": 175.94940185546875, "pos_frac": 0.8125, "sample": [-13.705917358398438, 96.7969970703125, -3.56982421875, 114.24324035644531, 90.5560302734375, 51.51080322265625, 125.04301452636719, 65.00862121582031, 25.634490966796875, 21.400634765625, 124.82644653320312, 107.98004150390625, 39.1356201171875, 32.291015625, 9.04339599609375, 59.739044189453125, 17.151077270507812, 85.70431518554688, 32.347808837890625, 10.064285278320312, 125.95988464355469, -14.609371185302734, -31.6085205078125, 29.01300048828125, 115.19783020019531, 94.69795227050781, 117.33938598632812, 33.3485107421875, 62.2625732421875, 136.19647216796875, 151.92916870117188, 38.19818115234375, -0.37180328369140625, 81.33194732666016, -8.823644638061523, 66.34129333496094, 12.465469360351562, 13.66748046875, 0.0, -14.925628662109375, -0.5288352966308594, 117.03123474121094, 149.00885009765625, 69.33582305908203, 120.17672729492188, 116.6912841796875, 113.47683715820312, 14.031707763671875, 98.10830688476562, 157.88397216796875, 5.26171875, 35.834259033203125, 19.553955078125, 9.31558609008789, 130.73733520507812, 3.974822998046875, 27.491729736328125, -43.69580078125, 80.28680419921875, 8.990127563476562, -10.523101806640625, 43.51751708984375, -4.829833984375, 135.06134033203125, 109.67323303222656, 11.111465454101562, 152.8406524658203, 1.0446052551269531, 35.38593292236328, 23.778045654296875, 24.54644775390625, 44.10693359375, 12.789566040039062, 72.8712158203125, -10.447906494140625, 56.443206787109375, 80.09274291992188, 129.03427124023438, 6.333290100097656, 57.80621337890625, -0.3262672424316406, 53.9356689453125, 135.606689453125, 12.902435302734375, 95.68348693847656, -18.242034912109375, 1.5103302001953125, 108.14547729492188, 90.1572265625, 62.700897216796875, 175.94940185546875, 63.36077880859375, 174.584716796875, 125.769775390625, 38.82862854003906, 119.60000610351562, -17.829193115234375, 95.55868530273438, 104.94236755371094, 88.0821304321289, 88.96575927734375, 60.001869201660156, 102.59217834472656, 38.24870300292969, -90.46023559570312, 126.35035705566406, -25.59625244140625, 110.24613189697266, -1.1267471313476562, 105.37088012695312, -47.3057861328125, 8.745246887207031, 25.9254150390625, 48.867088317871094, -23.49310302734375, 38.95127868652344, 63.28668212890625, -3.560577392578125, 71.5540771484375, -3.3383407592773438, -84.6900634765625, 112.01983642578125, 110.037353515625, 45.3721923828125, 85.13853454589844, 88.83797454833984, 94.68328857421875, 39.9971923828125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000403.npy"}
|
|
{"epoch": 0.8439790575916231, "step": 404, "batch_size": 128, "mean": 38.28358459472656, "std": 65.43445587158203, "min": -154.89657592773438, "p10": -34.75929260253906, "median": 27.421680450439453, "p90": 125.76537628173828, "max": 177.8426513671875, "pos_frac": 0.71875, "sample": [128.28082275390625, 125.52052307128906, 32.048583984375, 112.51823425292969, -96.48417663574219, 112.34764099121094, 140.6569061279297, 43.82087707519531, -66.41529083251953, 0.0, -114.8572998046875, 50.46881866455078, 15.001976013183594, -11.67279052734375, -70.40426635742188, 87.61952209472656, 21.6881103515625, 128.06060791015625, -2.3284225463867188, -45.73455810546875, 124.88760375976562, 23.482879638671875, 18.64837646484375, -8.29766845703125, -14.03509521484375, 92.58132934570312, -154.89657592773438, 115.44774627685547, -28.79383087158203, 10.472236633300781, 10.176422119140625, -36.637359619140625, 29.167510986328125, 9.362762451171875, 72.6041259765625, -6.46173095703125, 3.3817691802978516, 6.952728271484375, 177.8426513671875, 80.68894958496094, 116.40634155273438, -2.71075439453125, 2.325836181640625, -65.70278930664062, 103.0513687133789, 52.85774230957031, 56.550445556640625, 27.034881591796875, 4.722623825073242, 14.542724609375, 80.75148010253906, 7.253173828125, 116.76712036132812, -23.952911376953125, 23.4970703125, 146.32476806640625, -83.91891479492188, 15.732376098632812, -10.533370971679688, 13.153160095214844, 155.65789794921875, 140.15573120117188, -82.08403015136719, 93.28302001953125, 72.03195190429688, 110.2977523803711, 108.40591430664062, 8.402496337890625, -5.574737548828125, 110.35777282714844, 100.5682144165039, 7.189451217651367, 7.25372314453125, -7.536088943481445, -3.484405517578125, 85.215087890625, 45.6021728515625, -9.7779541015625, 141.59732055664062, 35.64019775390625, 49.1153564453125, 21.418533325195312, 23.338409423828125, 59.2132568359375, 10.1683349609375, -27.642745971679688, -18.8642578125, -12.59393310546875, 124.13064575195312, -78.31375122070312, -4.844440460205078, 54.089813232421875, 102.87655639648438, 41.2083740234375, -95.7760009765625, -33.95440673828125, 61.394256591796875, 14.312110900878906, 128.79104614257812, 60.82135009765625, -11.814447402954102, 3.4659423828125, 87.2984619140625, 52.81536865234375, 126.33670043945312, 66.6609878540039, 27.80847930908203, 66.18409729003906, -82.94001007080078, 129.51861572265625, 26.439666748046875, 147.92376708984375, 100.02114868164062, 46.12184143066406, 17.57366943359375, -4.1265869140625, 47.3416748046875, -8.595218658447266, 70.80487060546875, 104.91506958007812, 28.94378662109375, -16.137451171875, 21.712677001953125, 97.2374267578125, 106.48553466796875, 81.68424987792969, 117.79733276367188, 145.867919921875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000404.npy"}
|
|
{"epoch": 0.8460732984293193, "step": 405, "batch_size": 128, "mean": 43.96354293823242, "std": 59.08167266845703, "min": -100.78277587890625, "p10": -27.483645629882812, "median": 33.75153350830078, "p90": 129.44965209960938, "max": 160.73046875, "pos_frac": 0.765625, "sample": [13.9217529296875, 7.18609619140625, 7.0023193359375, 0.0, -38.387908935546875, -0.694732666015625, 7.7972412109375, 111.19457244873047, 83.08456420898438, 94.64089965820312, 2.4395751953125, 15.58538818359375, 66.05697631835938, 35.86784362792969, 106.56970977783203, -25.24974822998047, -100.78277587890625, 23.024017333984375, 34.85914611816406, 134.49261474609375, 56.820369720458984, 118.5980224609375, -1.2614479064941406, -34.241912841796875, 105.95550537109375, 39.313323974609375, -20.81854248046875, 58.46015930175781, 129.2155303955078, 25.250839233398438, 90.46243286132812, 110.41088104248047, 6.71868896484375, -28.593414306640625, 59.437767028808594, 106.16840362548828, 35.890289306640625, -39.512969970703125, -49.14387512207031, 85.1208267211914, -1.9336261749267578, 144.2876434326172, 99.33760070800781, 1.0493431091308594, -3.273193359375, 160.73046875, 92.76409912109375, 76.10359191894531, 96.39730834960938, 29.22216796875, 2.741607666015625, 71.22964477539062, 129.75180053710938, 60.001373291015625, 16.409286499023438, 0.20361328125, 98.21841430664062, 109.40841674804688, 94.59332275390625, -24.593658447265625, -63.723609924316406, 49.55528259277344, 65.72235107421875, -15.646438598632812, 15.794891357421875, -3.965728759765625, 48.003753662109375, -6.612579345703125, 19.89208984375, 148.69322204589844, 28.549819946289062, 81.33065795898438, 110.78350830078125, 78.28543090820312, 130.2220458984375, 151.33148193359375, -28.15765380859375, 76.40434265136719, 7.985694885253906, 32.461700439453125, 144.33917236328125, 9.058032989501953, -27.20855712890625, -81.77362060546875, 96.42196655273438, 23.504955291748047, 1.6382064819335938, 153.5103759765625, 32.353759765625, 71.49610900878906, 106.045166015625, 25.696144104003906, -51.79133605957031, 6.77960205078125, 124.74154663085938, 109.104736328125, 10.450668334960938, 14.58262825012207, 101.42221069335938, 85.7871322631836, -56.452484130859375, 129.32015991210938, 2.148284912109375, 113.57415771484375, 52.73966979980469, 44.12548828125, 9.448257446289062, 70.03913879394531, 61.9510498046875, 39.9669189453125, 32.6439208984375, 1.4544754028320312, 134.68136596679688, -6.309413909912109, 0.4081268310546875, -11.735748291015625, -19.59850311279297, 7.6101837158203125, -26.095077514648438, 2.1149425506591797, -28.125518798828125, 137.90216064453125, 141.93130493164062, -13.8870849609375, -44.464698791503906, 130.5797119140625, 64.23040771484375, 88.55538940429688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000405.npy"}
|
|
{"epoch": 0.8481675392670157, "step": 406, "batch_size": 128, "mean": 35.168479919433594, "std": 63.027069091796875, "min": -139.65243530273438, "p10": -37.24791107177734, "median": 25.726173400878906, "p90": 115.04280090332031, "max": 159.74472045898438, "pos_frac": 0.6796875, "sample": [106.25685119628906, -20.78868865966797, -31.301712036132812, 20.775634765625, -65.80364990234375, 152.85784912109375, 84.22137451171875, 114.29949951171875, 71.25577545166016, 114.06196594238281, -7.579860687255859, 0.592498779296875, -17.98504638671875, 125.74331665039062, 49.486419677734375, -17.039886474609375, 105.21807861328125, -16.43372344970703, 39.92131042480469, 19.928359985351562, -86.339599609375, -39.02095031738281, 8.3302001953125, 63.195770263671875, 103.6292724609375, 10.075180053710938, 71.10014343261719, 117.18682861328125, 20.217864990234375, 83.74072265625, 44.484100341796875, 115.88381958007812, -108.59906005859375, -7.654815673828125, 36.58697509765625, -98.51905822753906, -22.037277221679688, -10.284416198730469, -8.880088806152344, 104.2728271484375, -36.488037109375, -34.72828674316406, -63.42083740234375, -18.26422119140625, 23.336624145507812, 114.69110107421875, 14.956268310546875, 92.74410247802734, 90.82186889648438, 102.42279815673828, -7.7361907958984375, 18.5789794921875, 104.48335266113281, 84.34537506103516, -139.65243530273438, -8.172821044921875, 93.53094482421875, 118.74298095703125, 79.92522430419922, 14.80511474609375, 3.796905517578125, 115.86343383789062, -50.2373046875, 132.45477294921875, 16.768692016601562, 31.08709716796875, 6.24151611328125, 4.01336669921875, -9.39324951171875, 101.809814453125, 150.08511352539062, 109.21089172363281, 9.4820556640625, 35.02532958984375, 127.62816619873047, 35.08502197265625, 5.7942962646484375, 108.7205810546875, 13.265193939208984, 128.5191650390625, 100.24240112304688, -12.409433364868164, -107.74268341064453, 7.672367095947266, 113.07740783691406, 91.52783203125, 78.36581420898438, -8.11517333984375, 29.0648193359375, 34.7899169921875, 28.11572265625, -12.215408325195312, 83.14817810058594, 89.97572326660156, -53.757713317871094, 155.77487182617188, 72.95936584472656, -0.54925537109375, 93.96755981445312, 33.99725341796875, 59.75428771972656, 88.86053466796875, 32.687408447265625, 101.25811767578125, -22.989227294921875, 34.35893249511719, -68.05203247070312, 141.37594604492188, -7.8839569091796875, 29.503265380859375, -14.825653076171875, 58.054931640625, -24.08050537109375, 77.531982421875, 159.74472045898438, -0.9461212158203125, 2.457744598388672, 21.758384704589844, -29.808197021484375, -8.3961181640625, -41.335296630859375, 20.427398681640625, 15.49737548828125, -0.16351318359375, 94.43588256835938, -45.97210693359375, 46.17453384399414, 19.04949188232422], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000406.npy"}
|
|
{"epoch": 0.8502617801047121, "step": 407, "batch_size": 128, "mean": 41.05936050415039, "std": 61.676448822021484, "min": -98.32713317871094, "p10": -28.682760620117186, "median": 31.826316833496094, "p90": 124.91752166748046, "max": 160.35287475585938, "pos_frac": 0.7109375, "sample": [-28.519989013671875, 23.81005859375, 63.94451904296875, -13.855880737304688, 112.09246826171875, 13.439533233642578, 116.84552001953125, 60.55766296386719, -29.06256103515625, 22.06717872619629, -1.764007568359375, -35.724029541015625, -83.59805297851562, -6.56585693359375, -98.32713317871094, 80.19141387939453, -0.0847320556640625, 0.07263946533203125, 14.36090087890625, -23.86505126953125, 96.97657012939453, -9.135055541992188, 3.5080108642578125, 45.01689147949219, 122.20123291015625, -18.207382202148438, 56.721214294433594, 86.70745849609375, -6.107948303222656, -10.87615966796875, -64.64994049072266, 54.77461242675781, -36.13104248046875, 32.4716796875, 74.37179565429688, 124.36079406738281, 121.72918701171875, 129.30902099609375, -33.43190002441406, -8.511764526367188, 60.991943359375, 22.7376708984375, 31.501983642578125, 3.7926864624023438, -33.145233154296875, 142.53030395507812, 28.78417205810547, 85.4996337890625, 148.89334106445312, 15.72662353515625, 44.099639892578125, 106.6942138671875, 11.125091552734375, 147.54379272460938, 17.862686157226562, 111.15774536132812, -24.45928192138672, -8.643119812011719, 25.331146240234375, 32.773223876953125, 101.00482940673828, 32.15065002441406, 5.4100341796875, 36.7607421875, 114.0721435546875, 43.6650390625, -27.838226318359375, -13.162567138671875, 74.23562622070312, 126.216552734375, 22.397918701171875, -13.100479125976562, 104.93756103515625, 126.42648315429688, 137.34234619140625, 20.644336700439453, 4.40570068359375, 19.41625213623047, 0.0, 160.35287475585938, 117.01310729980469, 6.913536071777344, 58.72697448730469, -25.3924560546875, 99.25192260742188, -76.0111083984375, 141.143798828125, 54.629581451416016, 11.494903564453125, 99.69970703125, 116.40676879882812, -4.114715576171875, 8.67706298828125, 78.53062438964844, 65.49002075195312, 11.749214172363281, -31.89306640625, 36.722930908203125, 38.652191162109375, 102.64175415039062, 111.97933959960938, -13.513031005859375, 154.80374145507812, -89.73492431640625, 60.76605224609375, -64.26629638671875, 86.63385009765625, 55.49244689941406, 127.22088623046875, 0.43695068359375, 114.21854400634766, 121.26290893554688, 47.2503662109375, -88.15475463867188, 136.94454956054688, -9.983673095703125, 35.53019714355469, -22.638227462768555, 7.095550537109375, 106.5540771484375, 106.35487365722656, 98.87774658203125, 8.3719482421875, 7.407449722290039, -3.263580322265625, 147.4505615234375, 108.91986083984375, 0.0], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000407.npy"}
|
|
{"epoch": 0.8523560209424084, "step": 408, "batch_size": 128, "mean": 41.28568649291992, "std": 61.41910171508789, "min": -140.46432495117188, "p10": -29.014813232421872, "median": 35.01323699951172, "p90": 123.26127014160156, "max": 150.30499267578125, "pos_frac": 0.7265625, "sample": [-3.2393951416015625, 12.777862548828125, -45.331016540527344, 54.50494384765625, 15.767715454101562, 116.9776611328125, 68.04254150390625, 135.375, 122.96627807617188, 109.34406280517578, 3.2088623046875, 6.831825256347656, 50.80572509765625, 84.59915161132812, -44.58990478515625, 108.37993621826172, 27.073486328125, -10.57177734375, -2.6260833740234375, 26.75581169128418, 1.0171356201171875, 18.965499877929688, -29.739593505859375, -5.271728515625, 139.9688720703125, 105.41275787353516, 139.96484375, 123.9495849609375, -11.4661865234375, -19.07135009765625, -22.650436401367188, 119.81867980957031, 94.92514038085938, 84.2623291015625, -25.44140625, 24.76076889038086, 150.30499267578125, 91.57861328125, 118.50324249267578, 48.707672119140625, -10.33148193359375, -127.7884521484375, -140.46432495117188, 8.995002746582031, 125.12359619140625, 92.51217651367188, 3.8759002685546875, 29.034278869628906, -30.69287109375, 114.84884643554688, 28.154205322265625, 74.08349609375, -21.580276489257812, -8.521713256835938, 57.64857482910156, 39.76190185546875, 75.8734130859375, 101.9925537109375, -28.704193115234375, 54.09922790527344, 120.89518737792969, 102.16899108886719, 145.92120361328125, 9.163421630859375, -18.6484375, 120.750244140625, -38.49200439453125, -0.0162506103515625, 100.28936767578125, -51.81817626953125, 134.186279296875, -18.012374877929688, 100.16806030273438, 12.4237060546875, 55.9592170715332, 125.22174072265625, -14.560211181640625, 53.86885070800781, -7.92962646484375, 24.893508911132812, 107.0782470703125, 47.68238830566406, -35.06059265136719, 36.593231201171875, -22.28759765625, 6.52783203125, 99.29696655273438, 83.97039794921875, 74.22573852539062, -52.92680358886719, 2.22332763671875, 7.976295471191406, 51.28961181640625, 107.75605010986328, 109.8662109375, -2.7930755615234375, -104.4346923828125, 54.815093994140625, 129.77305603027344, 60.67901611328125, -30.854354858398438, 116.04328918457031, 15.338348388671875, 41.759796142578125, 101.18769836425781, -10.4285888671875, 13.12786865234375, 17.727691650390625, 17.201416015625, 46.307708740234375, 45.93931579589844, 7.6576385498046875, 25.149059295654297, 0.0, 7.666473388671875, 134.86492919921875, 33.43324279785156, 2.0174427032470703, 87.03860473632812, -23.94451904296875, -85.66424560546875, 27.63226318359375, 125.9251480102539, 55.457275390625, 125.3172607421875, 85.67718505859375, 83.30873107910156, 77.526123046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000408.npy"}
|
|
{"epoch": 0.8544502617801047, "step": 409, "batch_size": 128, "mean": 43.666587829589844, "std": 66.88552856445312, "min": -119.7735595703125, "p10": -41.8455810546875, "median": 27.859695434570312, "p90": 132.38057861328124, "max": 189.55258178710938, "pos_frac": 0.75, "sample": [106.01068115234375, 86.08500671386719, 107.93069458007812, 18.472076416015625, 137.22833251953125, -49.03778076171875, 87.7357177734375, 137.75250244140625, 3.5281982421875, 31.564178466796875, 18.88591766357422, -8.524612426757812, 15.499038696289062, 6.9423828125, 21.11505126953125, 152.70632934570312, 47.433128356933594, 20.1728515625, 117.5719985961914, -102.71365356445312, -8.385231018066406, -48.721954345703125, -1.2445907592773438, 127.58251953125, 103.12884521484375, -47.71441650390625, 18.825958251953125, -78.16004943847656, 120.10237121582031, -119.7735595703125, 134.09130859375, 109.47769165039062, -55.72657775878906, 15.823074340820312, -4.14300537109375, 148.44921875, 17.9534912109375, 32.577728271484375, 140.1710205078125, 130.1063232421875, 77.61233520507812, 123.507568359375, -77.49368286132812, 24.396743774414062, 2.4628334045410156, 37.0599365234375, 24.46124267578125, 121.58041381835938, 28.13690185546875, 134.98495483398438, 95.443359375, 101.42594909667969, 67.56787109375, 101.42169189453125, 189.55258178710938, -34.897216796875, -40.35853576660156, 64.26266479492188, 17.308563232421875, 7.3426513671875, 159.06698608398438, 82.82980346679688, 12.495635986328125, 34.82830810546875, 116.10009765625, 50.51869201660156, 27.582489013671875, -28.740875244140625, -2.991546630859375, -32.0350341796875, 116.71322631835938, -19.466171264648438, -95.26416015625, 135.0916748046875, -3.759876251220703, 122.383544921875, 138.81307983398438, 132.28887939453125, 22.327545166015625, 26.0802001953125, 7.776004791259766, -5.416980743408203, 62.91796875, 0.0, 65.29098510742188, 168.8992919921875, 32.40974426269531, 132.59454345703125, 101.32589721679688, -81.9398193359375, -22.30755615234375, 120.7351303100586, 30.2354736328125, 8.575218200683594, 15.671493530273438, -5.640220642089844, 98.21580505371094, 118.57754516601562, 107.0704345703125, 105.46769714355469, 21.628128051757812, 4.201499938964844, 72.27658081054688, 64.22235107421875, 43.98895263671875, -5.56585693359375, 18.8994140625, -41.258544921875, -2.4388580322265625, -11.190071105957031, 25.660137176513672, 3.945943832397461, -112.86981201171875, 31.21635627746582, 95.63720703125, 50.49114990234375, 11.628973007202148, 101.2672119140625, 25.044265747070312, -43.21533203125, 114.33676147460938, 6.116630554199219, 6.538429260253906, 88.55131530761719, -45.2333984375, 99.24388122558594, 127.00912475585938, 27.312591552734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000409.npy"}
|
|
{"epoch": 0.856544502617801, "step": 410, "batch_size": 128, "mean": 38.79985046386719, "std": 65.40264892578125, "min": -110.04473876953125, "p10": -36.36678466796875, "median": 37.786109924316406, "p90": 124.07662963867188, "max": 180.8616943359375, "pos_frac": 0.6953125, "sample": [48.3052978515625, -39.83525848388672, 13.395492553710938, -29.731231689453125, -9.705253601074219, 116.3934326171875, 16.519311904907227, 31.862945556640625, -33.638336181640625, 103.90406799316406, 161.40328979492188, -58.642181396484375, 54.79472351074219, -36.36334228515625, 58.34851837158203, -110.04473876953125, -29.171142578125, 2.438720703125, 124.60556030273438, 52.391387939453125, 124.68173217773438, -92.25714111328125, 51.805397033691406, 10.497978210449219, 116.85385131835938, -2.5162200927734375, 87.43705749511719, 103.42483520507812, 82.7099609375, 103.983642578125, 46.06475830078125, 50.279815673828125, 17.87548828125, 111.47232818603516, -4.013145446777344, 88.91004943847656, 27.46038818359375, 69.02395629882812, 47.6983642578125, 123.84994506835938, 122.07891845703125, 40.65814208984375, -3.8937225341796875, 121.63763427734375, -92.75384521484375, 91.996826171875, -102.76904296875, 24.945068359375, -4.3684539794921875, -24.363693237304688, 90.3935546875, 103.6192626953125, 85.54689025878906, -28.86871337890625, -21.389060974121094, 49.79547119140625, -4.807037353515625, 115.04255676269531, 74.40687561035156, 42.60064697265625, -42.594215393066406, 144.9068603515625, 32.944122314453125, 125.77694702148438, -1.4615058898925781, 5.3802642822265625, -25.766937255859375, -1.25445556640625, 58.170654296875, 10.603012084960938, 92.02359008789062, 109.03563690185547, 50.7745361328125, -13.66668701171875, -54.07177734375, 0.0, 41.15045928955078, 13.493545532226562, 129.20297241210938, 86.45245361328125, 156.8809814453125, 2.496612548828125, 15.2235107421875, -35.59529113769531, -29.49346160888672, 107.35592651367188, 16.99486541748047, 25.6943359375, -23.91668701171875, 24.433433532714844, -103.57295227050781, 34.55756759643555, -36.37481689453125, 85.10530090332031, -26.950439453125, 155.3934326171875, -88.7484130859375, 109.48068237304688, 8.733589172363281, 27.947006225585938, 40.166786193847656, 19.19219207763672, 58.66831970214844, -101.30825805664062, 38.49554443359375, 22.841781616210938, 86.97982788085938, 105.58233642578125, 170.52975463867188, 77.51199340820312, 122.952880859375, -16.809814453125, 80.49668884277344, 44.9765625, -0.57904052734375, 75.14907836914062, 139.968994140625, -48.687713623046875, 37.07667541503906, 12.083328247070312, 3.4202842712402344, 133.58935546875, 145.56451416015625, 44.73613739013672, 50.67500305175781, -23.547500610351562, 180.8616943359375, -2.9336166381835938], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000410.npy"}
|
|
{"epoch": 0.8586387434554974, "step": 411, "batch_size": 128, "mean": 45.31635665893555, "std": 60.83454895019531, "min": -114.13742065429688, "p10": -20.1291648864746, "median": 33.10625457763672, "p90": 125.68688278198242, "max": 197.3782958984375, "pos_frac": 0.78125, "sample": [88.74580383300781, 127.18607330322266, 33.400146484375, 102.01983642578125, 19.1976318359375, 2.0281524658203125, -1.2406082153320312, 32.81236267089844, 45.97050857543945, 101.89799499511719, 108.13568115234375, 76.45318603515625, -1.7345962524414062, -18.327972412109375, -56.46502685546875, -49.113372802734375, 152.70693969726562, -13.99456787109375, 109.5919189453125, 41.898956298828125, 37.0948486328125, 53.017120361328125, 110.55928039550781, 45.223114013671875, 92.97309875488281, 59.2481689453125, 48.00732421875, 197.3782958984375, 96.51834106445312, -14.751312255859375, 131.52316284179688, 13.049589157104492, 27.258628845214844, 20.344589233398438, 98.35330200195312, -17.635345458984375, 136.77938842773438, 100.23139953613281, 30.12188720703125, 16.315074920654297, 6.500152587890625, 28.473220825195312, 0.6557159423828125, 66.03437805175781, 7.5430450439453125, -53.641326904296875, 8.02008056640625, 12.914291381835938, -17.149871826171875, 97.78445434570312, 106.30997467041016, 25.24298095703125, -4.84967041015625, 3.5720748901367188, 13.651959419250488, 125.04437255859375, -24.331947326660156, 73.1536865234375, 24.05731201171875, 24.198646545410156, 0.01238250732421875, 24.72222900390625, -4.837677001953125, -76.892578125, 103.84707641601562, 89.76214599609375, 73.80673217773438, 117.77886962890625, 16.529312133789062, 148.785888671875, 9.11614990234375, 96.75323486328125, 10.614410400390625, 122.34515380859375, 42.807090759277344, 138.75091552734375, 15.884201049804688, 8.41326904296875, 118.73555755615234, 103.8287582397461, 15.112525939941406, -114.13742065429688, 132.43829345703125, 5.561908721923828, 13.455215454101562, -91.07315826416016, 50.82708740234375, -0.37163543701171875, 107.84841918945312, 140.45111083984375, 75.57455444335938, -34.49406433105469, 29.023773193359375, 29.72021484375, 19.21649169921875, 35.11640167236328, 77.32073974609375, 160.33877563476562, 74.17456817626953, -55.152862548828125, 89.72817993164062, -51.218719482421875, -62.69720458984375, -15.6002197265625, 11.579080581665039, 23.0975341796875, 72.53921508789062, 35.59197998046875, 37.075897216796875, 92.83759307861328, -30.017379760742188, 74.999755859375, 117.39982604980469, -7.335685729980469, 13.678672790527344, -13.883255004882812, 63.00727844238281, 166.9703369140625, 105.80377197265625, 103.41734313964844, 94.692138671875, 149.97535705566406, 90.4964599609375, 150.32733154296875, 14.322647094726562, -8.209396362304688, -49.915924072265625, -5.84564208984375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000411.npy"}
|
|
{"epoch": 0.8607329842931937, "step": 412, "batch_size": 128, "mean": 39.39039993286133, "std": 66.35340118408203, "min": -140.452880859375, "p10": -33.22868499755859, "median": 28.262226104736328, "p90": 126.83700256347656, "max": 191.51910400390625, "pos_frac": 0.703125, "sample": [-87.08761596679688, 128.73684692382812, 5.1997528076171875, 134.03497314453125, 22.783447265625, 63.360626220703125, 29.47613525390625, 98.360107421875, -29.45953369140625, -18.846954345703125, 37.204559326171875, 31.438140869140625, 0.9780654907226562, -18.17234230041504, 26.324188232421875, -140.452880859375, 0.139434814453125, -23.437042236328125, -60.532073974609375, 22.712867736816406, -0.256195068359375, 1.4172229766845703, 16.92999267578125, 72.64813232421875, 125.52091979980469, 116.84503936767578, 133.1072998046875, 27.048316955566406, -18.917327880859375, 50.718971252441406, 87.54779815673828, 88.9324951171875, -11.93212890625, 127.67147827148438, 137.60928344726562, 91.43377685546875, 125.91769409179688, 16.24408721923828, 102.13299560546875, 62.779937744140625, 94.94181823730469, -23.549041748046875, 116.988525390625, -85.3714599609375, 62.99299621582031, -8.873504638671875, -3.74420166015625, -18.437744140625, -94.37750244140625, 36.157562255859375, 74.5308837890625, 133.6404571533203, 148.81475830078125, -17.582427978515625, 130.48968505859375, 5.7886505126953125, -11.209060668945312, 56.044189453125, -52.011924743652344, 14.671783447265625, -48.84442138671875, -6.275094985961914, 158.09716796875, 60.542724609375, -7.3622589111328125, 126.4793701171875, 94.81647491455078, 191.51910400390625, 46.62335205078125, 58.97021484375, -63.94537353515625, 73.33580017089844, -38.622314453125, -25.89483642578125, 46.983245849609375, 31.606582641601562, 145.56336975097656, 124.6151351928711, 9.346588134765625, -118.4063720703125, 51.01727294921875, 60.296417236328125, 110.34803771972656, 2.782562255859375, 120.01941680908203, 6.265167236328125, 120.9525146484375, 110.57856750488281, 138.67434692382812, -44.85020446777344, 15.097808837890625, 124.11962890625, 99.05918884277344, -76.93667602539062, -4.990730285644531, 94.41819763183594, -14.963485717773438, 105.44891357421875, 0.06500244140625, 95.90509033203125, -7.69921875, 0.11572074890136719, 148.64944458007812, 14.511207580566406, 94.83538818359375, 5.465892791748047, -28.372756958007812, 58.322021484375, 10.542083740234375, 62.16217041015625, 115.3876953125, 25.651809692382812, 0.0, 80.10154724121094, -10.24737548828125, 91.397216796875, -1.3211212158203125, 12.142921447753906, -75.85275268554688, 4.621068954467773, 14.285476684570312, 84.82159423828125, -30.917129516601562, 55.47930908203125, -8.451828002929688, 2.367767333984375, 93.0762939453125, 122.37637329101562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000412.npy"}
|
|
{"epoch": 0.86282722513089, "step": 413, "batch_size": 128, "mean": 36.01963424682617, "std": 57.157222747802734, "min": -95.896484375, "p10": -25.97510643005371, "median": 22.879575729370117, "p90": 115.08328857421874, "max": 184.92022705078125, "pos_frac": 0.7265625, "sample": [16.20472526550293, -6.9894561767578125, -0.9229907989501953, -14.061332702636719, -2.088226318359375, 106.07136535644531, 45.314849853515625, 20.99542236328125, -95.896484375, -25.58428955078125, 126.0098876953125, 35.7076416015625, 132.03689575195312, -41.77178955078125, 22.600982666015625, 11.790458679199219, 8.377593994140625, 131.319580078125, 51.81439208984375, -38.24433898925781, 41.41375732421875, 11.053516387939453, 71.28672790527344, 15.386932373046875, 24.905563354492188, -1.185638427734375, 7.825813293457031, 148.2108154296875, 1.6156902313232422, 73.45051574707031, 96.17573547363281, -7.12066650390625, -37.564117431640625, 65.26950073242188, -24.59375, 112.11143493652344, 114.09059143066406, 109.02896118164062, 111.62972259521484, 3.8031539916992188, 73.88169860839844, 2.5966796875, -56.410400390625, 1.10906982421875, 29.73486328125, -24.350189208984375, -26.97443389892578, 49.15354919433594, 100.51712036132812, 62.017059326171875, -38.77687072753906, -2.599884033203125, -18.120361328125, 27.686492919921875, 20.7572021484375, 91.13801574707031, 123.9029541015625, -21.059463500976562, 37.96076202392578, 47.21977233886719, 127.59097290039062, 18.966506958007812, 12.914398193359375, 24.446311950683594, 29.59625244140625, 156.1239013671875, -26.887012481689453, 59.490966796875, 114.2091064453125, -34.70722961425781, 99.15869140625, 49.49440002441406, -11.348854064941406, 184.92022705078125, 23.15816879272461, 1.5609130859375, 2.1945037841796875, 23.930130004882812, -7.468112945556641, 70.84881591796875, 8.336334228515625, 104.3648681640625, -40.1876220703125, 93.47308349609375, 104.36103820800781, -17.311874389648438, 3.12353515625, 73.4122314453125, 7.000932693481445, -47.372528076171875, 117.123046875, 26.315881729125977, -19.160049438476562, 94.26678466796875, 19.5550537109375, -23.14197540283203, -3.714447021484375, -24.921844482421875, 108.19866943359375, 13.134401321411133, 6.019504547119141, -23.10832977294922, 5.184844970703125, 102.432373046875, 87.734375, 65.135009765625, 6.139825820922852, 9.34649658203125, 82.71295166015625, 28.352325439453125, 25.26678466796875, 11.359619140625, 69.42143249511719, 111.95809936523438, -16.00946807861328, -61.7679443359375, 47.8446044921875, 118.97369384765625, -93.13455200195312, 161.35110473632812, 35.12730407714844, 128.10153198242188, 11.558090209960938, -0.013700485229492188, 127.97183227539062, 112.47589111328125, 12.232254028320312, 23.9359130859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000413.npy"}
|
|
{"epoch": 0.8649214659685864, "step": 414, "batch_size": 128, "mean": 40.48951721191406, "std": 67.21144104003906, "min": -151.862060546875, "p10": -34.45513916015625, "median": 24.171615600585938, "p90": 130.72134246826172, "max": 179.40167236328125, "pos_frac": 0.7109375, "sample": [81.83135986328125, -97.04716491699219, 56.449432373046875, 86.53570556640625, 1.6460113525390625, 103.5506591796875, 32.04608154296875, 67.55987548828125, 22.47113037109375, -45.3311767578125, -4.12939453125, 6.995307922363281, 131.40475463867188, 11.388092041015625, -1.6377716064453125, -1.2413654327392578, -33.100128173828125, -69.12200927734375, 110.76571655273438, -19.281230926513672, -151.862060546875, -41.18280029296875, 149.95538330078125, 6.0570068359375, 80.72451782226562, -76.93548583984375, -36.217315673828125, 5.719158172607422, 98.52969360351562, 104.447998046875, -32.656959533691406, -19.400970458984375, 80.97203063964844, 113.66152954101562, -50.826873779296875, 116.89488220214844, 107.14058685302734, 4.1175384521484375, 150.24981689453125, 9.008132934570312, 22.900299072265625, 121.09293365478516, 28.019866943359375, -7.6466064453125, 86.6239242553711, 96.13369750976562, 5.0389404296875, 12.789085388183594, 61.48199462890625, 148.93728637695312, -6.3765716552734375, -20.61969757080078, -2.6233043670654297, 159.76431274414062, 83.17251586914062, -3.289581298828125, 110.01504516601562, 126.12635803222656, 133.84225463867188, -73.99618530273438, 68.69476318359375, 4.03611946105957, -0.114837646484375, 112.32478332519531, 4.63592529296875, 58.69752502441406, 37.22637176513672, 74.29057312011719, 52.52264404296875, -4.08642578125, 106.94598388671875, -6.3588104248046875, -19.61810302734375, 146.07403564453125, 105.8938980102539, 101.6396484375, 113.23858642578125, 3.133392333984375, 86.50015258789062, 91.23119354248047, 22.714706420898438, -19.6463623046875, -85.1520004272461, 150.51364135742188, 177.52008056640625, 127.04632568359375, 130.42845153808594, 63.28015899658203, 1.9271087646484375, 103.77902221679688, -130.9111328125, 6.9451904296875, 25.746551513671875, 98.174072265625, 7.435699462890625, 127.4859619140625, -29.006439208984375, 0.293212890625, -11.831619262695312, 20.09941864013672, -37.624412536621094, 56.1712646484375, 179.40167236328125, 19.88286781311035, 9.405952453613281, 8.211593627929688, 109.86042785644531, 61.47068786621094, 69.22611999511719, 17.930694580078125, -33.699920654296875, 48.25836944580078, 149.9107666015625, 44.013824462890625, -3.122285842895508, 9.73297119140625, 86.2410888671875, 139.99508666992188, 20.4183349609375, 34.14414978027344, -20.310531616210938, 2.5284423828125, -52.9259033203125, -19.6224365234375, 25.44293212890625, 59.40673828125, 134.91864013671875, -1.8946075439453125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000414.npy"}
|
|
{"epoch": 0.8670157068062827, "step": 415, "batch_size": 128, "mean": 48.794334411621094, "std": 63.157569885253906, "min": -120.120849609375, "p10": -20.7078125, "median": 44.75798034667969, "p90": 129.23824768066405, "max": 178.901611328125, "pos_frac": 0.7421875, "sample": [18.278038024902344, 87.53544616699219, 13.793403625488281, 20.591604232788086, -29.71563720703125, 0.6682281494140625, 105.17533111572266, -0.14794921875, 102.75545501708984, 10.664169311523438, 30.601890563964844, -2.3465957641601562, 115.30517578125, 74.7767333984375, -23.383811950683594, 2.5781631469726562, 120.06765747070312, 3.47930908203125, 103.95266723632812, 36.95487976074219, 107.37527465820312, 15.21923828125, 133.48434448242188, 3.214019775390625, 85.05772399902344, 25.076385498046875, 66.26499938964844, 126.20654296875, 123.03633880615234, 72.0552978515625, -89.42994689941406, 141.1044921875, 123.7752685546875, 12.0732421875, -84.19830322265625, 20.76239013671875, 55.08123779296875, 132.16452026367188, -7.5126190185546875, 73.528564453125, 74.74340057373047, 8.537994384765625, -1.0894393920898438, 18.214996337890625, 27.500389099121094, -15.561721801757812, 123.02197265625, 38.81103515625, 137.622802734375, 104.99490356445312, -11.461051940917969, 34.272796630859375, -68.8525390625, 71.96810913085938, 106.87773132324219, 125.36141967773438, -16.35162353515625, 93.83062744140625, 36.89239501953125, -25.90985107421875, 48.38751220703125, 111.64994812011719, -30.088897705078125, 59.01423645019531, 178.901611328125, -21.88897705078125, 22.28246307373047, 46.64888381958008, -4.49212646484375, -0.8991527557373047, -4.4124298095703125, 142.64501953125, 92.78901672363281, 44.628814697265625, -8.345489501953125, -3.9464263916015625, 116.42791748046875, 150.62371826171875, 87.51229858398438, -39.4766845703125, 68.181884765625, 1.5455322265625, 38.192626953125, -46.34246826171875, 127.984130859375, -20.20159912109375, -95.6498794555664, 53.37648010253906, 73.47933959960938, -18.3919677734375, -7.147857666015625, -3.0892333984375, 63.45451354980469, 117.17205810546875, 107.2642822265625, -16.562591552734375, 49.138458251953125, 49.173736572265625, 114.00408935546875, -120.120849609375, 153.6927490234375, 33.217201232910156, 64.78326416015625, 59.09880065917969, 24.73663330078125, 138.48138427734375, 44.88714599609375, -18.319992065429688, 122.404541015625, 94.32402038574219, 112.24887084960938, 43.23365020751953, 109.20829772949219, 88.87709045410156, -72.80888366699219, 164.73944091796875, 17.642776489257812, 132.29563903808594, 138.21328735351562, -5.0853271484375, 3.6949615478515625, 19.98773193359375, 108.54530334472656, 5.0997161865234375, 119.26980590820312, 85.05409240722656, -20.07476043701172, 165.4583740234375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000415.npy"}
|
|
{"epoch": 0.8691099476439791, "step": 416, "batch_size": 128, "mean": 47.499176025390625, "std": 65.03706359863281, "min": -144.54644775390625, "p10": -31.960984039306638, "median": 42.501190185546875, "p90": 125.61352691650389, "max": 173.570556640625, "pos_frac": 0.734375, "sample": [123.36019897460938, 160.87844848632812, 133.37899780273438, 26.248931884765625, 3.8371200561523438, 2.7616729736328125, -38.57598876953125, 49.01947021484375, -10.217559814453125, 89.40806579589844, -36.61566162109375, 173.570556640625, 47.989471435546875, 25.081375122070312, 27.32373046875, 16.61402130126953, -38.0733642578125, 139.80438232421875, 166.6297607421875, 57.92774963378906, 78.71450805664062, 97.37100219726562, 39.40992736816406, 130.4556427001953, -10.011505126953125, 31.656478881835938, -34.16667938232422, 104.39923095703125, -41.7296142578125, 18.748916625976562, 85.42927551269531, 117.05826568603516, 111.6424560546875, -2.979034423828125, -2.4859161376953125, 136.05377197265625, -95.11727905273438, 114.21298217773438, 18.343246459960938, 15.49688720703125, 19.6365966796875, 22.187450408935547, -19.380416870117188, -144.54644775390625, -11.344074249267578, 117.44766235351562, 25.267776489257812, 94.75543212890625, 97.7320556640625, 124.05352783203125, 42.865966796875, 129.25352478027344, 14.502716064453125, 7.821083068847656, -69.73977661132812, 120.39266204833984, 120.6890869140625, 75.49609375, -7.0145111083984375, -6.929443359375, 81.66549682617188, 66.58671569824219, -92.14956665039062, -31.01568603515625, 143.31219482421875, 67.58303833007812, -3.8355789184570312, 96.8197021484375, -15.67230224609375, 102.18399047851562, -0.6300888061523438, -7.6042938232421875, 41.899169921875, 98.93692779541016, 162.85305786132812, 99.61334228515625, 84.47689819335938, 109.06988525390625, -13.35321044921875, 9.4228515625, 55.9512939453125, 31.7120361328125, 123.61474609375, -28.69104766845703, -55.53521728515625, 69.55522155761719, 18.456298828125, 84.26898193359375, -5.369991302490234, 23.94153594970703, 39.47441864013672, 97.11369323730469, 51.676368713378906, 80.77606201171875, -111.9461669921875, 110.96253967285156, 51.40765380859375, 0.832977294921875, 82.31356811523438, 122.34757995605469, -63.146942138671875, 89.778564453125, 42.13641357421875, 11.689849853515625, 51.706329345703125, -11.0123291015625, 144.81451416015625, 83.1186294555664, 157.0260009765625, -16.581939697265625, 1.6622238159179688, 155.90057373046875, 41.48443603515625, 17.174072265625, -16.022705078125, 119.50303649902344, 96.11129760742188, 82.84130859375, 41.63813018798828, -3.649829864501953, 49.227386474609375, -39.24395751953125, 123.26318359375, -17.178035736083984, 115.26852416992188, 7.618499755859375, 73.9580078125, 111.78334045410156], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000416.npy"}
|
|
{"epoch": 0.8712041884816754, "step": 417, "batch_size": 128, "mean": 40.44239807128906, "std": 64.1326904296875, "min": -120.189453125, "p10": -37.433980560302736, "median": 36.896942138671875, "p90": 125.39529724121094, "max": 174.00796508789062, "pos_frac": 0.734375, "sample": [-0.8437385559082031, 45.76335144042969, 25.608718872070312, 62.76567077636719, -24.042984008789062, 92.335205078125, 24.275711059570312, 8.331573486328125, 138.59117126464844, 146.37548828125, 12.713272094726562, 54.70745849609375, 3.4728965759277344, 76.27449035644531, -98.02611541748047, 7.033893585205078, 9.243682861328125, -7.6146240234375, -10.89276123046875, 112.71286010742188, 123.037353515625, -21.765045166015625, 137.01873779296875, -61.745697021484375, -37.196266174316406, -100.99488067626953, 58.4771728515625, 4.742713928222656, -28.452774047851562, 150.55245971679688, 52.51542663574219, -6.635284423828125, 5.7470703125, 66.08377838134766, 75.35002136230469, 125.99118041992188, 15.8377685546875, -78.4365234375, 102.47514343261719, 93.66683959960938, 115.91436767578125, 23.350982666015625, 65.74855041503906, -13.101776123046875, 89.1201171875, 74.335205078125, 13.81976318359375, 72.126708984375, 79.80682373046875, 51.12736511230469, 25.493806838989258, -1.0160274505615234, 8.290863037109375, -57.34672546386719, -60.524688720703125, 37.246246337890625, -9.9312744140625, 118.13027954101562, 33.79237365722656, -10.34326171875, 24.981353759765625, 28.513946533203125, 2.747211456298828, 14.16510009765625, 111.62722778320312, 6.6056365966796875, 97.52667999267578, -25.8251953125, -120.189453125, 1.9223175048828125, -17.772247314453125, 112.0784912109375, 86.70779418945312, 53.767486572265625, 43.997802734375, -37.9886474609375, 8.438690185546875, 122.4085693359375, 0.0, -1.6961669921875, 74.72367858886719, 0.0, 48.950469970703125, 104.1898193359375, 79.24905395507812, 88.92393493652344, 35.1656494140625, 4.8416290283203125, 97.64645385742188, -18.67181396484375, -52.249725341796875, 125.63897705078125, 105.83396911621094, 117.4527587890625, 124.44293212890625, 57.109344482421875, -101.2867431640625, 53.34913635253906, 166.828125, 0.013671875, 87.93011474609375, -70.3614730834961, -59.322418212890625, 53.603633880615234, 114.3125, 127.73297119140625, 142.73574829101562, -26.688751220703125, -68.02561950683594, 16.019287109375, 36.547637939453125, -4.8764801025390625, 15.895587921142578, 125.29086303710938, 140.809814453125, 117.22189331054688, -19.5765380859375, 42.54383850097656, 132.16506958007812, 6.482418060302734, 49.306854248046875, 42.27418518066406, 42.462646484375, 9.9014892578125, 75.68212890625, 85.77760314941406, 174.00796508789062, 151.51165771484375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000417.npy"}
|
|
{"epoch": 0.8732984293193717, "step": 418, "batch_size": 128, "mean": 35.890098571777344, "std": 61.430843353271484, "min": -132.91799926757812, "p10": -40.12512359619141, "median": 31.491676330566406, "p90": 119.25803833007812, "max": 178.7879638671875, "pos_frac": 0.71875, "sample": [-49.56634521484375, 109.4363021850586, 27.491607666015625, 46.113372802734375, -54.629852294921875, -66.1903076171875, 2.1282958984375, 139.82562255859375, -32.208587646484375, -132.91799926757812, 37.754180908203125, 60.10575866699219, 30.746826171875, -40.95512390136719, 53.117919921875, 46.27006530761719, 69.5761947631836, -53.320709228515625, 92.35101318359375, 9.575576782226562, 109.60122680664062, 78.58526611328125, -14.87060546875, 0.046253204345703125, -2.4332275390625, 7.50628662109375, -25.725906372070312, 100.32131958007812, 49.86712646484375, 147.3970947265625, 114.90613555908203, 45.66510009765625, 119.51669311523438, 88.22817993164062, 16.946548461914062, 67.74545288085938, -20.714645385742188, 101.52227783203125, -38.1607666015625, -13.326881408691406, 2.0455322265625, 74.99295043945312, 23.406280517578125, 61.94842529296875, -10.862197875976562, 25.113494873046875, -5.388271331787109, 54.69679260253906, 27.962554931640625, -15.164016723632812, 0.72943115234375, 125.591064453125, 91.98686981201172, 24.673751831054688, 33.025482177734375, -9.782562255859375, 8.246429443359375, 63.48974609375, 53.268463134765625, 42.20947265625, -68.9822998046875, 87.5567626953125, 115.04025268554688, 36.41212463378906, 31.4285888671875, 7.661094665527344, 92.71717071533203, -32.26593780517578, -79.76670837402344, 3.1045379638671875, -47.943267822265625, 14.663200378417969, 96.29544067382812, 43.492431640625, 67.614013671875, 141.5487823486328, 107.95404052734375, -6.5452423095703125, 72.3607177734375, -56.165740966796875, 41.79840087890625, 104.72181701660156, -7.271881103515625, 44.199798583984375, 55.637603759765625, 120.77320098876953, 112.4005126953125, -30.274688720703125, 1.926666259765625, -66.11981201171875, 10.16387939453125, -25.98175048828125, -18.32891845703125, 90.2138671875, 108.55062866210938, 12.155357360839844, 131.29296875, 75.07777404785156, 22.77965545654297, 114.80752563476562, -39.7694091796875, 0.8537254333496094, 10.290321350097656, 31.554763793945312, -75.19677734375, 27.367034912109375, 60.0059814453125, 102.3182144165039, -93.378173828125, 124.38766479492188, 21.306896209716797, 16.537994384765625, 43.11164855957031, -0.7511787414550781, 119.14718627929688, 49.58699035644531, 178.7879638671875, 124.54638671875, -19.331863403320312, 171.28512573242188, 55.227935791015625, -18.65662384033203, -8.92962646484375, 59.372283935546875, -37.58711624145508, 126.15621185302734, 140.99197387695312, 0.4780120849609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000418.npy"}
|
|
{"epoch": 0.875392670157068, "step": 419, "batch_size": 128, "mean": 48.81062698364258, "std": 67.89590454101562, "min": -164.44334411621094, "p10": -33.91221008300781, "median": 42.3680534362793, "p90": 131.46337280273437, "max": 172.52655029296875, "pos_frac": 0.7734375, "sample": [36.40460205078125, 14.475631713867188, 0.386260986328125, 126.10089111328125, 22.51073455810547, 23.817474365234375, 5.6575775146484375, 16.811172485351562, 55.093048095703125, -58.151580810546875, -64.11679077148438, 69.34234619140625, -91.48843383789062, 109.84843444824219, 17.28814697265625, 27.6878662109375, 155.34024047851562, 15.6903076171875, 68.6568603515625, 117.18905639648438, 123.20588684082031, 110.38229370117188, 134.429443359375, 50.36418914794922, 136.0738525390625, 9.13726806640625, 34.84452819824219, 3.79254150390625, -26.43157958984375, -71.90945434570312, 124.63993072509766, 115.24517822265625, 98.84956359863281, -3.0883407592773438, 41.216400146484375, -13.974472045898438, 74.47950744628906, -0.023893356323242188, 110.46417236328125, -92.04095458984375, 122.53494262695312, 5.9135589599609375, 121.16873168945312, 91.73263549804688, 126.04119873046875, 8.38479232788086, 149.43824768066406, -49.664886474609375, 48.89544677734375, 148.40615844726562, 85.31915283203125, 79.0460205078125, -19.641090393066406, 58.92200469970703, 42.151123046875, 115.29985046386719, 156.86431884765625, 125.03471374511719, 15.667236328125, -16.782257080078125, -69.77973937988281, 9.810943603515625, -31.596641540527344, 3.6461219787597656, 170.60614013671875, -38.8599853515625, 116.22673034667969, 135.97950744628906, -0.1830902099609375, 35.562469482421875, -18.672813415527344, 104.03471374511719, 98.03237915039062, 19.75787353515625, 85.07937622070312, 0.0, -68.5838623046875, 6.384796142578125, 10.6915283203125, 0.0, 31.178436279296875, 101.36060333251953, 108.28982543945312, 127.47998046875, -23.642173767089844, 115.6075439453125, 8.51837158203125, 133.66302490234375, 71.57406616210938, 16.769210815429688, 132.29705810546875, 31.297607421875, 132.5335693359375, 54.677337646484375, 0.739776611328125, 10.2244873046875, 97.5315170288086, 49.27105712890625, 115.96377563476562, 131.1060791015625, 123.64605712890625, 13.89544677734375, 172.52655029296875, 15.943344116210938, 114.49246215820312, -31.795501708984375, 108.05331420898438, 113.23760986328125, 49.88433837890625, 103.55599975585938, -38.8511962890625, -52.810546875, 134.27774047851562, 111.90849304199219, -164.44334411621094, 52.105438232421875, -104.89252471923828, -11.765419006347656, 102.00729370117188, -24.31439208984375, 75.56439208984375, 105.82080078125, -11.14105224609375, 20.4281005859375, 14.930496215820312, 42.584983825683594, 113.99642944335938, 41.405731201171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000419.npy"}
|
|
{"epoch": 0.8774869109947644, "step": 420, "batch_size": 128, "mean": 45.99137878417969, "std": 65.10261535644531, "min": -146.6979217529297, "p10": -28.594003295898435, "median": 36.141761779785156, "p90": 124.21033325195312, "max": 177.9388427734375, "pos_frac": 0.7734375, "sample": [-72.620849609375, 22.439407348632812, 25.441726684570312, -130.334228515625, -22.023406982421875, -24.561439514160156, 31.96563720703125, 117.72677612304688, 10.041961669921875, 71.46075439453125, 121.5639877319336, 1.71160888671875, -78.72787475585938, 78.15206909179688, 97.17121887207031, 8.891433715820312, 107.29409790039062, -61.78337097167969, 105.0076904296875, 84.8238296508789, 33.834747314453125, 54.07182312011719, -5.503932952880859, 36.348419189453125, 4.4058990478515625, 128.76812744140625, 11.17230224609375, 26.43529510498047, 75.12547302246094, 23.560760498046875, -8.5848388671875, 145.364990234375, 42.99583435058594, 94.10684204101562, 11.441925048828125, -4.81219482421875, 97.18926239013672, 116.00714111328125, 145.750244140625, 8.5235595703125, 140.1846923828125, 46.47686767578125, 27.484886169433594, 113.26925659179688, 13.776123046875, 35.0028076171875, 58.42384338378906, 0.0421142578125, -3.6651611328125, 67.87467956542969, -9.90478515625, 75.47918701171875, 98.75743103027344, 123.9439697265625, 96.45137786865234, 86.46734619140625, 0.0, 35.588165283203125, -28.151153564453125, -98.64533996582031, 98.96028900146484, 100.64681243896484, 8.93682861328125, 18.874080657958984, 16.202911376953125, 31.053253173828125, 8.91961669921875, 140.0428466796875, 158.98898315429688, 108.31958770751953, 100.77816772460938, -42.78160095214844, 24.27008056640625, 85.45201110839844, -37.092041015625, 133.26089477539062, 72.48222351074219, 17.78363037109375, 123.4966812133789, 34.41035461425781, -6.3176422119140625, 10.689983367919922, 177.9388427734375, -40.2591552734375, 114.55697631835938, 66.03472900390625, -23.834197998046875, -22.981273651123047, 120.88432312011719, 124.83184814453125, 55.1876220703125, 18.979888916015625, 133.2396240234375, 127.49945068359375, 10.8609619140625, 13.66455078125, 41.76175308227539, 60.906707763671875, 158.0247802734375, -0.7467041015625, 148.18142700195312, 88.6707763671875, -146.6979217529297, -9.419891357421875, 35.93510437011719, 114.962646484375, 95.59059143066406, 9.169097900390625, -63.723236083984375, -55.74852752685547, 78.27191162109375, 93.17213439941406, 84.11671447753906, 122.95327758789062, 110.30709838867188, 19.021881103515625, 76.34017944335938, -115.437744140625, 16.102783203125, 115.28277587890625, 90.57302856445312, 117.15109252929688, -29.6273193359375, -9.64794921875, 2.6036605834960938, 90.90151977539062, -3.4795494079589844, 58.7435302734375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000420.npy"}
|
|
{"epoch": 0.8795811518324608, "step": 421, "batch_size": 128, "mean": 47.29153060913086, "std": 59.663909912109375, "min": -105.57568359375, "p10": -22.050331115722653, "median": 42.439544677734375, "p90": 123.14722900390625, "max": 159.7841796875, "pos_frac": 0.7421875, "sample": [159.7841796875, 59.88050842285156, 86.55924224853516, 41.558868408203125, 90.25119018554688, 77.891357421875, 91.565185546875, 9.831436157226562, -104.73980712890625, -2.2932205200195312, 110.1929931640625, -12.20587158203125, 3.6163330078125, 87.5966796875, 28.619155883789062, 87.7608413696289, 36.41583251953125, -97.03277587890625, 36.966461181640625, 80.49812316894531, -15.772308349609375, -12.877059936523438, 29.188339233398438, 17.06415557861328, 87.66240692138672, 152.56097412109375, 111.80093383789062, 38.15496826171875, 8.677570343017578, 69.44811248779297, 0.0, -75.68585205078125, 72.7703857421875, 28.843856811523438, 13.588226318359375, 54.58061218261719, 124.5252685546875, 101.62042236328125, 25.646408081054688, 117.4647216796875, 135.27923583984375, 111.18016815185547, 38.82098388671875, 19.579357147216797, -2.8843116760253906, 14.276893615722656, 92.1998519897461, -20.38232421875, 136.5208740234375, 93.77340698242188, 99.43421936035156, -10.829002380371094, 43.962432861328125, 57.450439453125, 13.8466796875, 134.44403076171875, -72.01110076904297, 117.6925048828125, 12.7974853515625, 71.03257751464844, 33.88262939453125, 21.880294799804688, 149.48797607421875, 136.65621948242188, 116.66619873046875, 47.39904022216797, 80.40379333496094, 114.73895263671875, 120.78276062011719, 106.13008117675781, 112.10664367675781, 122.556640625, 65.8416748046875, 110.78884887695312, 51.148834228515625, 43.320220947265625, -17.74615478515625, 30.475067138671875, 117.00961303710938, 11.019630432128906, 81.25796508789062, -5.683380126953125, 38.62261962890625, 134.1846923828125, -3.30389404296875, 24.388519287109375, 32.10975646972656, -21.22052001953125, 102.06369018554688, -43.90874481201172, 88.0426254272461, 86.22703552246094, 142.40322875976562, 50.4422607421875, -31.89166259765625, 44.2001953125, 71.8302001953125, -40.288639068603516, -105.57568359375, 49.05464172363281, -34.34283447265625, -26.376556396484375, 39.7047119140625, -8.413116455078125, 38.31378173828125, 128.39596557617188, 8.346878051757812, -7.46697998046875, 80.36263275146484, -43.295440673828125, -23.986557006835938, 103.82901000976562, 41.167083740234375, 94.5560302734375, -20.799453735351562, 109.50596618652344, 52.64067077636719, 0.0, -31.7886962890625, -19.936534881591797, 154.46157836914062, 71.216064453125, 28.57464599609375, 0.0, 0.64874267578125, -1.82720947265625, 146.25555419921875, -2.096923828125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000421.npy"}
|
|
{"epoch": 0.881675392670157, "step": 422, "batch_size": 128, "mean": 36.96299743652344, "std": 66.10303497314453, "min": -136.08560180664062, "p10": -36.558935546875, "median": 31.999984741210938, "p90": 119.16610870361328, "max": 158.5972900390625, "pos_frac": 0.703125, "sample": [13.373504638671875, -14.50146484375, 111.85508728027344, -37.98345947265625, 102.50704956054688, -13.878852844238281, 118.54640197753906, -103.14353942871094, 35.272857666015625, -0.22406005859375, 92.29855346679688, 4.857177734375, 63.699920654296875, 110.72286987304688, 82.504150390625, -136.08560180664062, 8.96142578125, 13.761962890625, -26.579437255859375, 91.80944061279297, 81.20446014404297, 39.737396240234375, 74.95745849609375, 18.391494750976562, -35.94842529296875, 100.714599609375, -103.33403015136719, 80.02522277832031, 95.68351745605469, 139.15447998046875, 56.78471374511719, 2.6436767578125, -17.402938842773438, 87.47836303710938, 7.481983184814453, 101.17507934570312, 0.8365211486816406, -67.93399047851562, 136.4904327392578, 4.160552978515625, 17.904708862304688, 118.34394836425781, -30.86639404296875, 44.29644775390625, 45.56573486328125, -127.22198486328125, -12.222076416015625, 158.5972900390625, 58.509857177734375, 114.60293579101562, 46.52117919921875, 127.8466796875, 92.69381713867188, 21.727901458740234, 30.6708984375, 1.54736328125, -2.328216552734375, 12.653757095336914, 29.9398193359375, 46.01373291015625, 99.0716552734375, -32.00984191894531, -22.316635131835938, -98.74490356445312, 140.612548828125, 2.791900634765625, 73.2230453491211, 13.024642944335938, 1.4438819885253906, 140.99041748046875, 85.86911010742188, -131.85452270507812, 115.68653869628906, -5.1379241943359375, 56.7855224609375, -24.110393524169922, 98.35436248779297, -59.08177185058594, 134.1635284423828, 7.029064178466797, 134.7867431640625, -3.0190963745117188, 91.22442626953125, 33.06809997558594, 135.6668701171875, 52.59674072265625, -25.80804443359375, 15.740570068359375, 120.61209106445312, -1.213033676147461, -12.674503326416016, -23.695068359375, 60.526336669921875, -16.9224853515625, 95.4822769165039, 84.55551147460938, -26.76837158203125, -1.3577117919921875, 54.0196533203125, 80.56317138671875, 108.94715881347656, 12.556396484375, 140.52078247070312, 116.68206787109375, -66.67431640625, 101.26092529296875, 50.625732421875, -56.170867919921875, -20.22943115234375, 80.9105224609375, 0.0, -65.07760620117188, 113.13507080078125, 105.36787414550781, -66.47674560546875, 43.08966064453125, 25.6856689453125, 99.4854507446289, 30.592435836791992, -0.183624267578125, 12.249069213867188, 59.996986389160156, 3.4535980224609375, 131.3123321533203, 32.22566223144531, 31.774307250976562, -0.7517967224121094, 148.90972900390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000422.npy"}
|
|
{"epoch": 0.8837696335078534, "step": 423, "batch_size": 128, "mean": 48.92060089111328, "std": 65.7524185180664, "min": -110.78265380859375, "p10": -32.188943481445314, "median": 42.98701477050781, "p90": 131.72694702148436, "max": 160.05068969726562, "pos_frac": 0.75, "sample": [1.74298095703125, 137.4704132080078, -14.126663208007812, 51.033905029296875, 121.82064819335938, 107.85833740234375, 27.39547348022461, -13.82208251953125, 1.99969482421875, 109.39112091064453, 52.135498046875, 95.11294555664062, 30.439208984375, 94.98310852050781, -52.546417236328125, -11.392974853515625, 117.4783935546875, -47.76275634765625, 99.55853271484375, 9.046356201171875, 6.993959426879883, -24.15301513671875, 20.571701049804688, 102.5604019165039, -97.9075927734375, 79.79176330566406, 84.75373840332031, 10.77691650390625, 119.77561950683594, 141.12335205078125, 6.166595458984375, 23.0877685546875, 42.35870361328125, 9.815216064453125, 83.22532653808594, 97.77037048339844, 117.06771850585938, -2.31634521484375, -12.853912353515625, 64.76541137695312, 84.68109130859375, 129.72918701171875, -18.094512939453125, 111.26373291015625, -4.1460418701171875, 22.566444396972656, 5.8590087890625, 149.59014892578125, -32.72772216796875, -24.80303955078125, 140.91607666015625, -33.822784423828125, -39.375518798828125, 49.77862548828125, 123.22161865234375, -31.958038330078125, -71.77685546875, -70.10820770263672, 138.92758178710938, 131.01089477539062, 23.028648376464844, -4.9128265380859375, 88.78871154785156, 14.598381042480469, -9.886131286621094, 91.14454650878906, 4.257545471191406, 111.59396362304688, 5.8871002197265625, 29.366058349609375, 100.4995346069336, 120.7532730102539, 86.87631225585938, 98.36282348632812, 43.615325927734375, 52.828887939453125, -34.833404541015625, 121.35250091552734, 160.05068969726562, -7.132598876953125, 16.094085693359375, 133.39773559570312, 10.1197509765625, 111.68704223632812, 151.48306274414062, 48.57646179199219, 80.3538818359375, 16.81481170654297, 18.671096801757812, 140.42758178710938, 106.63136291503906, 106.61871337890625, 128.93634033203125, 4.436187744140625, 151.89254760742188, 104.4205322265625, 2.600332260131836, -24.8743896484375, 137.6145782470703, -8.13836669921875, 7.4137420654296875, 88.87896728515625, 17.096588134765625, 142.61801147460938, 152.99468994140625, -23.91632080078125, -7.2990570068359375, 97.90682220458984, 121.5345458984375, -18.704437255859375, -96.92953491210938, 20.701576232910156, -58.725341796875, 52.748046875, 15.235416412353516, 125.3817138671875, 87.96414184570312, 109.34589385986328, 128.84027099609375, 22.09320068359375, 91.08551025390625, 99.61538696289062, 127.37812042236328, -64.42605590820312, 16.7430419921875, 33.483795166015625, -2.3626537322998047, -110.78265380859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000423.npy"}
|
|
{"epoch": 0.8858638743455497, "step": 424, "batch_size": 128, "mean": 46.769691467285156, "std": 58.72008514404297, "min": -119.28182983398438, "p10": -22.26686096191406, "median": 44.42597961425781, "p90": 118.84484558105468, "max": 166.1990966796875, "pos_frac": 0.7578125, "sample": [69.45576477050781, 44.3917236328125, -0.5044403076171875, -4.35284423828125, 33.94419479370117, 14.194046020507812, -4.781280517578125, -0.309814453125, 29.501121520996094, 0.76629638671875, 6.8162689208984375, -51.271728515625, 166.1990966796875, 60.51231384277344, -1.483917236328125, 76.62581634521484, 3.84210205078125, 140.49234008789062, 128.12203979492188, 109.1003189086914, -6.347299575805664, 7.614013671875, 145.90145874023438, 105.65713500976562, 71.739990234375, 55.782981872558594, -2.7933578491210938, 18.366485595703125, 83.53433227539062, 93.05069732666016, 87.3835220336914, 131.79931640625, 117.38287353515625, -36.67487716674805, 88.7540283203125, 92.29048156738281, 27.73370361328125, 103.93536376953125, 120.6927490234375, -18.7991943359375, -16.444305419921875, 112.70854187011719, 62.04400634765625, -119.28182983398438, -1.288055419921875, 153.899169921875, 137.26150512695312, 66.122314453125, -106.58113098144531, 44.460235595703125, -18.507747650146484, -3.51654052734375, 0.902374267578125, 118.05288696289062, 55.54840087890625, 115.47372436523438, 13.594863891601562, 112.16143798828125, 98.96035766601562, 53.73521423339844, 37.244232177734375, -6.60540771484375, 109.78536987304688, 48.8212890625, 143.10205078125, 98.50585174560547, 40.9024658203125, 28.90496826171875, -55.57318115234375, -39.42626953125, -29.350830078125, 57.269012451171875, 99.64661407470703, 146.58294677734375, 142.46463012695312, 123.14451599121094, 6.1461181640625, 70.25647735595703, 88.59454345703125, 34.10943603515625, 41.29913330078125, 72.92642211914062, -1.7212371826171875, 2.8422698974609375, 103.72309875488281, 45.58707046508789, -58.66731262207031, 0.0, -34.54008483886719, 30.300384521484375, 12.41162109375, 85.06987762451172, 103.54183959960938, -7.266326904296875, 65.24410247802734, 105.89878845214844, 84.417724609375, 101.57009887695312, 108.93232727050781, 84.42005157470703, 14.768081665039062, -62.43232727050781, 81.25506591796875, 100.35112762451172, 77.34207153320312, 6.488813400268555, -19.04943084716797, 8.975341796875, -21.462066650390625, 5.544454574584961, 92.92539978027344, -54.999290466308594, 97.02337646484375, 108.64674377441406, -24.14471435546875, 7.175067901611328, 12.258062362670898, 37.75651168823242, 22.639404296875, 58.306854248046875, 24.08526611328125, 75.07030487060547, -27.655197143554688, 32.79188537597656, 93.7498779296875, 11.864433288574219, 9.145450592041016, 140.0157012939453], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000424.npy"}
|
|
{"epoch": 0.8879581151832461, "step": 425, "batch_size": 128, "mean": 40.6321907043457, "std": 63.5363655090332, "min": -152.23458862304688, "p10": -27.1300874710083, "median": 27.05596923828125, "p90": 123.37235107421874, "max": 176.15625, "pos_frac": 0.734375, "sample": [91.92515563964844, 107.38552856445312, 48.4788818359375, 48.13981628417969, -15.88922119140625, 17.662872314453125, 38.348175048828125, 97.78733825683594, 14.836395263671875, -14.958358764648438, 118.11599731445312, 11.9677734375, 122.119140625, 68.24691772460938, 98.1496353149414, 1.6497688293457031, 86.20793151855469, 110.94381713867188, 1.1685848236083984, 19.790283203125, 36.337646484375, 100.38227081298828, 34.41899108886719, -5.387420654296875, 115.06805419921875, 55.216461181640625, 3.5862550735473633, 19.80511474609375, -76.00971984863281, -3.287149429321289, 35.07536315917969, -5.2359619140625, -15.162399291992188, 26.45587158203125, 132.94158935546875, -11.999235153198242, 34.050201416015625, 140.08108520507812, 5.783416748046875, 15.079811096191406, -17.1302490234375, 143.45803833007812, 98.09269714355469, 26.186370849609375, -2.89208984375, 10.093315124511719, -26.972885131835938, -4.161773681640625, -91.5575942993164, 118.45648193359375, 67.6253662109375, 95.439208984375, -74.53546142578125, 45.995269775390625, 127.710693359375, 17.196044921875, 108.56474304199219, 176.15625, 35.488677978515625, 21.910064697265625, -25.311080932617188, 32.58685302734375, 7.945270538330078, 30.992843627929688, 140.97393798828125, -37.194427490234375, 60.212310791015625, -3.807210922241211, 121.217529296875, 99.9787826538086, 101.12567138671875, 8.373809814453125, 145.98626708984375, 5.91868782043457, -19.050994873046875, -8.390754699707031, -8.665889739990234, 25.954803466796875, -14.695053100585938, 70.23990631103516, 0.7028427124023438, 126.2965087890625, 18.924560546875, 72.029296875, 136.18299865722656, -74.71995544433594, 17.921470642089844, -40.34346008300781, 39.22041320800781, -34.660400390625, 19.7076416015625, -27.49689292907715, 13.871650695800781, -91.40232849121094, -8.86846923828125, -63.051849365234375, 85.16796875, 24.716644287109375, 99.980224609375, -73.9957275390625, 38.84877014160156, 139.10882568359375, 27.65606689453125, 119.17430114746094, 67.33900451660156, 70.88154602050781, 121.94393920898438, 0.709564208984375, 4.312328338623047, 109.2357177734375, 0.0, -19.008758544921875, 119.00177001953125, 115.543701171875, 165.24923706054688, -5.16258430480957, 18.934268951416016, 118.0889892578125, -152.23458862304688, 36.087646484375, 131.89749145507812, 134.1575164794922, 88.11773681640625, 112.09626770019531, 112.04360961914062, -49.89958953857422, 9.635147094726562, 8.188369750976562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000425.npy"}
|
|
{"epoch": 0.8900523560209425, "step": 426, "batch_size": 128, "mean": 44.746498107910156, "std": 59.67395782470703, "min": -115.38031005859375, "p10": -19.848880195617674, "median": 25.27789306640625, "p90": 124.49174194335937, "max": 161.142333984375, "pos_frac": 0.765625, "sample": [107.79904174804688, -2.6082000732421875, 1.3770408630371094, -32.76812744140625, 115.840087890625, 18.523269653320312, 91.46994018554688, 3.310823440551758, -88.3577880859375, 14.194007873535156, -115.38031005859375, -38.629638671875, -42.33265686035156, 83.90586853027344, 114.15744018554688, 124.44364929199219, 102.93115234375, 23.3746337890625, 104.54139709472656, 19.148834228515625, -35.27685546875, 55.701416015625, 138.75552368164062, 16.23912811279297, 2.1305084228515625, 20.1431884765625, 125.50619506835938, 135.90142822265625, 106.38140869140625, 129.30169677734375, 82.88530731201172, 2.431926727294922, 90.34600830078125, 94.3565902709961, 110.409912109375, 8.065298080444336, 2.5707168579101562, 20.01392364501953, 124.95248413085938, 131.41986083984375, 30.02410888671875, 131.2209014892578, 7.253044128417969, -5.8768768310546875, 111.0384521484375, 115.0230712890625, -1.074188232421875, 79.89720153808594, 13.122467041015625, 3.7176513671875, -62.64573669433594, 24.17803955078125, 37.710350036621094, 6.56182861328125, 19.880367279052734, 118.95648193359375, -53.5682373046875, 54.925514221191406, 120.89691162109375, 90.35087585449219, 88.20755004882812, 103.78384399414062, -26.210693359375, -1.9874305725097656, -0.3213653564453125, 79.15972900390625, 46.44773864746094, 85.0294189453125, 3.284515380859375, -5.6156005859375, 63.74517822265625, 18.91241455078125, -2.848724365234375, -3.363922119140625, -0.47431182861328125, 161.142333984375, -17.551162719726562, 18.543045043945312, 40.5042724609375, 97.39618682861328, -3.7498855590820312, 112.89964294433594, 70.35369873046875, 75.22456359863281, 23.691619873046875, 25.02081298828125, 25.53497314453125, 14.493186950683594, 19.010461807250977, 102.08584594726562, 120.26226806640625, 12.936111450195312, -0.10281753540039062, 6.668243408203125, 137.15206909179688, 37.19151306152344, 53.0025634765625, 131.53836059570312, -20.934417724609375, -19.383649826049805, 65.43962860107422, 11.14508056640625, 0.0, 9.811660766601562, -8.129974365234375, 148.72280883789062, 70.4459228515625, 11.606109619140625, 11.35552978515625, 56.70323181152344, 95.18585205078125, 34.392333984375, -79.52978515625, 2.56915283203125, -5.2565460205078125, 101.56084442138672, -29.779693603515625, 124.44146728515625, 114.62045288085938, 108.15887451171875, 100.15006256103516, 3.5830373764038086, -99.62130737304688, 79.23577880859375, 144.875, 62.16337585449219, -10.34747314453125, 124.60395812988281], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000426.npy"}
|
|
{"epoch": 0.8921465968586387, "step": 427, "batch_size": 128, "mean": 59.37547302246094, "std": 61.85523986816406, "min": -116.943359375, "p10": -17.316107177734374, "median": 62.615760803222656, "p90": 135.01754455566405, "max": 207.36517333984375, "pos_frac": 0.8046875, "sample": [17.546409606933594, 129.08230590820312, 14.047134399414062, -22.99462890625, -45.595794677734375, 10.15625, 75.65707397460938, 83.60137939453125, 86.607177734375, 29.210708618164062, 86.37640380859375, 48.28962707519531, 139.7425537109375, 26.033950805664062, 117.24151611328125, 118.66688537597656, 105.82853698730469, 7.794206619262695, 107.53219604492188, 148.81137084960938, 150.192138671875, 44.72550964355469, -23.977035522460938, 74.947998046875, 114.77044677734375, 84.04811096191406, 100.39971923828125, 114.76422119140625, 54.52912521362305, 49.95336151123047, 148.78317260742188, 114.023193359375, 65.472900390625, 32.987213134765625, 2.1787567138671875, 3.2887020111083984, 89.59297180175781, 12.526926040649414, -3.39996337890625, -30.867218017578125, 128.07847595214844, 22.351531982421875, -66.42207336425781, 59.75862121582031, -83.316162109375, 88.85040283203125, -3.81390380859375, 137.7272186279297, -15.854209899902344, 108.96916961669922, 133.185791015625, 20.272430419921875, 111.1710205078125, 70.1826171875, 151.71026611328125, 139.033447265625, 166.49639892578125, -7.485748291015625, 20.689361572265625, -8.144523620605469, 92.86328125, 81.76022338867188, 141.91555786132812, 17.5458984375, 27.737579345703125, 48.21806335449219, -16.80401611328125, 112.04875946044922, 112.27122497558594, -4.384239196777344, -116.943359375, 129.51947021484375, 90.94402313232422, -18.510986328125, 95.419921875, 17.89434814453125, -26.86663818359375, 110.52584838867188, 85.25439453125, 44.96873474121094, 137.67080688476562, -60.365325927734375, 109.60697174072266, -39.3741455078125, 27.721298217773438, 84.6744384765625, 107.42679595947266, 50.852294921875, 53.639068603515625, 132.622802734375, 102.68087768554688, 129.46099853515625, -29.15093994140625, 11.591949462890625, 120.27194213867188, 100.4349136352539, 138.60104370117188, 54.5279541015625, 10.551834106445312, 86.4007568359375, -3.77362060546875, 14.475814819335938, 38.07209777832031, -26.993606567382812, 13.395965576171875, 207.36517333984375, -3.9595489501953125, 113.00933837890625, 6.324920654296875, 120.9827880859375, 133.88043212890625, 83.17103576660156, 91.1689453125, 8.7567138671875, 39.33650207519531, -0.8178253173828125, 106.69775390625, 78.18501281738281, 1.9354248046875, 1.8780517578125, 127.21099853515625, 26.312820434570312, -0.6065673828125, -2.2500877380371094, 24.273788452148438, 109.65155029296875, 153.3496856689453, 127.78691101074219], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000427.npy"}
|
|
{"epoch": 0.8942408376963351, "step": 428, "batch_size": 128, "mean": 54.84465789794922, "std": 59.04267501831055, "min": -101.68716430664062, "p10": -13.222320365905759, "median": 53.59957504272461, "p90": 125.14943542480468, "max": 206.42129516601562, "pos_frac": 0.8203125, "sample": [24.700164794921875, 0.97491455078125, -38.19593048095703, 121.5543212890625, 124.47470092773438, -12.608062744140625, 50.1910400390625, 45.4208984375, -36.62062072753906, 15.927337646484375, 206.42129516601562, 21.432373046875, 89.20733642578125, 93.87225341796875, -2.16693115234375, 129.4220428466797, 19.6136474609375, -4.7088623046875, 12.762039184570312, -1.432647705078125, 92.0592041015625, 1.88177490234375, -86.31149291992188, 155.55465698242188, 103.27105712890625, 133.25982666015625, 87.5733642578125, 100.84839630126953, 74.08940124511719, 168.93084716796875, 23.63604736328125, 131.93890380859375, 1.8258991241455078, 38.670143127441406, -15.10821533203125, 14.346878051757812, -1.0724945068359375, 96.50003051757812, 32.92579650878906, 90.94973754882812, 50.3343505859375, -8.358573913574219, -101.68716430664062, 94.06636047363281, 120.82925415039062, 161.63580322265625, -28.795166015625, 52.82714080810547, 15.877189636230469, 12.60040283203125, 111.01838684082031, 97.88064575195312, 86.5743408203125, 75.00129699707031, 23.111114501953125, 76.34408569335938, 112.45832824707031, 77.43658447265625, -43.03973388671875, 166.73641967773438, -92.04292297363281, 149.63119506835938, 110.15896606445312, 35.6444091796875, 96.99627685546875, 8.074844360351562, 102.8919677734375, 92.34062194824219, 80.18539428710938, 45.80194091796875, 15.310104370117188, 32.653472900390625, 116.72102355957031, 108.1884765625, 140.26290893554688, 15.14263916015625, 44.78411865234375, 7.88519287109375, 55.275238037109375, 22.800338745117188, 52.39794158935547, 23.503402709960938, 91.58360290527344, 63.111297607421875, 2.567291259765625, 8.612548828125, -17.20053482055664, 14.72821044921875, 48.266693115234375, 69.09605407714844, 185.77301025390625, 126.72381591796875, 90.87322998046875, -1.3720703125, 78.43048095703125, -82.96865844726562, -22.378173828125, 126.9765625, 89.88531494140625, 0.0, 19.856353759765625, 122.33467102050781, 106.09425354003906, 19.92961883544922, 112.42098999023438, 58.11833953857422, -17.731239318847656, 78.01190185546875, 78.427490234375, 58.951812744140625, 54.837646484375, 100.86849975585938, 23.52954864501953, 25.144920349121094, 122.09901428222656, 54.37200927734375, 37.2725830078125, 66.314697265625, 86.02261352539062, -3.19854736328125, 20.073516845703125, 26.389892578125, 112.5025634765625, -14.655588150024414, -11.457427978515625, 112.859375, 95.04876708984375, 79.49922180175781], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000428.npy"}
|
|
{"epoch": 0.8963350785340314, "step": 429, "batch_size": 128, "mean": 44.08338165283203, "std": 72.27555084228516, "min": -139.51734924316406, "p10": -45.29193038940429, "median": 36.06581497192383, "p90": 134.6236358642578, "max": 164.295166015625, "pos_frac": 0.6953125, "sample": [104.51252746582031, -126.33511352539062, -1.98614501953125, 91.923583984375, 123.27406311035156, 118.13189697265625, 30.301620483398438, 28.978073120117188, -48.10801696777344, 94.04132080078125, 7.1890716552734375, -93.80921936035156, 17.81414794921875, 118.01657104492188, 131.10748291015625, 118.11505126953125, 3.9194793701171875, 122.69769287109375, -5.881439208984375, 21.49078369140625, 80.01152801513672, 101.37769317626953, 11.288030624389648, 120.19662475585938, -49.78558349609375, 121.3438720703125, 84.59857177734375, 148.08265686035156, 22.329803466796875, 72.5528564453125, -89.97915649414062, 101.98411560058594, 107.77508544921875, 122.51046752929688, 19.666793823242188, 21.306961059570312, -89.87881469726562, -13.8294677734375, 121.2030029296875, 164.295166015625, 107.65960693359375, 115.13972473144531, -20.17327880859375, 4.200859069824219, 68.07907104492188, -2.861053466796875, 157.19921875, 66.09814453125, 6.524139404296875, -26.09765625, 5.960710525512695, 141.52719116210938, 146.40023803710938, 125.4482650756836, 15.443832397460938, 46.0242919921875, 152.53463745117188, -63.89198303222656, -79.61349487304688, 112.15960693359375, -9.4403076171875, 49.266937255859375, 125.32437133789062, 136.90237426757812, -28.80859375, 41.83000946044922, 47.8563232421875, 19.92431640625, 17.80133819580078, 65.34780883789062, -17.835205078125, -45.069175720214844, -24.195953369140625, -45.81169128417969, 113.3369140625, -139.51734924316406, 133.64703369140625, -25.948768615722656, 21.656890869140625, 126.04241943359375, -4.035028457641602, 0.0, 122.85548400878906, 156.08639526367188, 107.31631469726562, 53.023406982421875, 132.0824737548828, 10.841300964355469, -1.285736083984375, 3.613208770751953, 158.53567504882812, -31.548309326171875, 90.43815612792969, -7.137550354003906, 12.566970825195312, 44.253936767578125, 20.058250427246094, 6.5062103271484375, -0.187347412109375, 89.70237731933594, -46.56684875488281, -9.679962158203125, -17.00494384765625, 7.0188140869140625, -31.995361328125, 77.08798217773438, 62.19976806640625, 145.10015869140625, -112.63967895507812, 54.40339660644531, 143.93679809570312, 115.3455810546875, -44.28173828125, 124.24751281738281, 147.726806640625, -8.200363159179688, 8.721136093139648, 2.8619232177734375, 112.63067626953125, 141.82293701171875, -10.898590087890625, 102.98681640625, -14.5478515625, -23.354141235351562, 57.962554931640625, 45.32000732421875, 78.88995361328125, -102.62155151367188], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000429.npy"}
|
|
{"epoch": 0.8984293193717278, "step": 430, "batch_size": 128, "mean": 38.029666900634766, "std": 65.87508392333984, "min": -106.64251708984375, "p10": -40.681396484375, "median": 26.529441833496094, "p90": 129.6006072998047, "max": 174.63302612304688, "pos_frac": 0.7109375, "sample": [-47.42010498046875, -9.798843383789062, -72.17892456054688, 4.381141662597656, 89.02999877929688, 97.42340087890625, -63.07377624511719, 93.5218505859375, -19.92193603515625, 5.183536529541016, 96.37060546875, 12.854869842529297, 28.13153076171875, 14.02215576171875, 44.04301452636719, -106.64251708984375, 145.58251953125, 16.186981201171875, 1.763101577758789, 106.4066162109375, 91.36888122558594, -64.295166015625, -16.005935668945312, 99.94387817382812, 16.204345703125, 23.81170654296875, -28.5341796875, 0.0, 134.51583862304688, 64.125244140625, 65.93968963623047, -19.0645751953125, 104.99053955078125, 46.94160461425781, 15.974205017089844, 0.0, 35.108795166015625, -2.6094894409179688, -0.383056640625, 93.81470489501953, -72.3990478515625, 136.98190307617188, -33.129302978515625, 83.51205444335938, 141.44485473632812, 77.81233215332031, 59.393211364746094, 13.82574462890625, -20.82373046875, 174.63302612304688, 83.99067687988281, 26.519683837890625, 96.5723876953125, 16.7252197265625, 125.41983032226562, 132.89759826660156, 7.837158203125, -37.2093505859375, 110.17938232421875, -106.28695678710938, 110.78053283691406, 123.71232604980469, 131.14956665039062, 120.33863830566406, 50.153045654296875, 51.45648193359375, 105.59384155273438, 104.98623657226562, 6.695880889892578, 144.5966796875, 21.62109375, 26.539199829101562, 151.94564819335938, 109.27389526367188, -41.71159362792969, -27.12310028076172, 2.9571533203125, 9.402252197265625, -0.9265899658203125, 46.81591796875, 104.839111328125, 26.038223266601562, -40.23988342285156, 11.458534240722656, -13.243988037109375, 45.80914306640625, -22.190032958984375, 161.27920532226562, 0.0, 3.1071929931640625, 28.828292846679688, -15.50079345703125, 71.64212036132812, 14.926620483398438, 143.11837768554688, 73.39187622070312, 27.3072509765625, 78.65643310546875, 3.7138214111328125, -86.2449722290039, 148.51429748535156, 139.8017578125, 49.00068664550781, -39.037628173828125, -75.73382568359375, 74.27505493164062, 112.8594970703125, 40.59825897216797, 128.936767578125, 59.4754638671875, -88.43109130859375, 2.2885894775390625, 55.169342041015625, -11.766983032226562, 24.778884887695312, 110.16729736328125, 9.358078002929688, -17.755218505859375, 0.287353515625, -93.38241577148438, -8.87261962890625, -91.44775390625, 51.192626953125, 93.91799926757812, 88.58106994628906, 17.488433837890625, -6.95001220703125, 113.92129516601562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000430.npy"}
|
|
{"epoch": 0.900523560209424, "step": 431, "batch_size": 128, "mean": 36.97406768798828, "std": 63.52135467529297, "min": -106.80490112304688, "p10": -31.62142791748046, "median": 27.637542724609375, "p90": 123.78009643554685, "max": 183.75244140625, "pos_frac": 0.6796875, "sample": [-10.186187744140625, -106.80490112304688, 54.39979553222656, 99.79873657226562, 94.15263366699219, 58.76715087890625, 6.2927703857421875, 68.12039184570312, 34.43280029296875, -53.232696533203125, -3.683837890625, 27.2357177734375, 11.30767822265625, 35.88787841796875, -37.22393798828125, 90.72756958007812, -16.423370361328125, 95.54551696777344, 77.25666809082031, 13.07596206665039, -7.8515625, 84.54643249511719, 49.29481506347656, -64.60978698730469, 58.5986328125, 108.28695678710938, 64.13395690917969, -43.419097900390625, 114.87837219238281, 159.54733276367188, 69.79933166503906, 59.383544921875, 36.6468505859375, -88.38480377197266, -41.04045104980469, 106.19534301757812, 78.73001861572266, -15.89788818359375, -2.9486312866210938, 80.04629516601562, -24.550983428955078, -93.3843994140625, 10.303068161010742, -4.863037109375, 98.84494018554688, 56.76416015625, 139.0321502685547, -16.301654815673828, -26.325559616088867, -106.19306182861328, 8.194883346557617, 28.03936767578125, 1.723337173461914, 105.0308837890625, 0.9791259765625, 105.39654541015625, 39.4256591796875, 141.713134765625, 119.55572509765625, -22.131500244140625, -0.252044677734375, 34.77833938598633, 108.51139831542969, 23.20965576171875, 16.99005126953125, 37.14154052734375, 54.09503173828125, 128.72833251953125, -17.567550659179688, -17.2813720703125, 121.659423828125, -20.317581176757812, 0.264404296875, 140.98309326171875, -3.319091796875, 178.154052734375, 0.328582763671875, 145.4915771484375, 41.29327392578125, 34.18470764160156, 137.611328125, 183.75244140625, 50.52935791015625, -43.31929016113281, -1.1553287506103516, -9.6136474609375, 96.53884887695312, 22.849929809570312, -42.89691162109375, 24.26758575439453, 157.580322265625, 105.46376037597656, 134.0327911376953, -29.220352172851562, 2.7137680053710938, 172.8587646484375, -20.207763671875, 107.24627685546875, 1.862447738647461, 13.871292114257812, 29.903305053710938, 36.88206481933594, -1.3021697998046875, 11.9127197265625, 25.317626953125, 0.46605682373046875, 110.3333740234375, 48.15153503417969, 89.34597778320312, -9.174575805664062, 33.36473083496094, -8.558975219726562, 59.532318115234375, 106.65728759765625, 23.2073974609375, 71.26046752929688, -0.686859130859375, -25.278594970703125, -84.64353942871094, 33.50189208984375, -7.9656829833984375, -3.936767578125, 13.764961242675781, -9.251007080078125, -68.26998901367188, 156.61048889160156, 96.53216552734375, 26.528472900390625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000431.npy"}
|
|
{"epoch": 0.9026178010471204, "step": 432, "batch_size": 128, "mean": 46.408103942871094, "std": 63.92136001586914, "min": -126.06378173828125, "p10": -23.976696777343747, "median": 34.060386657714844, "p90": 130.02332916259766, "max": 176.07806396484375, "pos_frac": 0.7265625, "sample": [-12.774322509765625, 122.36776733398438, 2.8010330200195312, 109.66312408447266, 161.98577880859375, 141.48275756835938, 102.36972045898438, 94.76541137695312, -47.89111328125, -112.179443359375, 141.813232421875, -10.647857666015625, 119.41265869140625, 33.07733154296875, 34.19865417480469, 14.908302307128906, 30.799652099609375, 81.63420104980469, -25.695571899414062, 0.09693145751953125, 89.48762512207031, -126.06378173828125, 36.55290222167969, 2.488189697265625, 14.001739501953125, 6.054986953735352, -20.966796875, 117.32524108886719, 51.60690689086914, -8.4964599609375, 21.2315673828125, 125.07640075683594, 144.46597290039062, 127.83926391601562, -17.823211669921875, -8.871620178222656, 89.263671875, -31.490386962890625, 131.54660034179688, 24.484039306640625, 96.99304962158203, 95.42300415039062, 35.15643310546875, 2.383411407470703, 4.639717102050781, -7.79974365234375, 54.97540283203125, 19.677261352539062, -101.85150146484375, -3.061676025390625, 87.90564727783203, 29.812850952148438, 8.854225158691406, 132.64596557617188, -8.206298828125, 9.5770263671875, -56.34913635253906, 110.88849639892578, -89.74397277832031, 14.875175476074219, -7.850494384765625, 18.316131591796875, 108.87516784667969, -14.656036376953125, 101.5401611328125, 5.818763732910156, 144.65472412109375, 97.80264282226562, 90.00770568847656, 99.01701354980469, -31.324203491210938, 46.164154052734375, 6.525390625, -6.248664855957031, 17.35820770263672, 147.58810424804688, 99.49771118164062, 157.8482666015625, 11.248046875, 113.68118286132812, 39.65960693359375, -1.0513076782226562, 176.07806396484375, -13.621994018554688, 80.32347106933594, 70.35955810546875, 129.37049865722656, 97.03653717041016, -25.654022216796875, 97.8837890625, 6.4878692626953125, -11.241451263427734, 0.0, 119.13958740234375, 120.68464660644531, -23.257843017578125, 8.9775390625, 18.9373779296875, -4.448272705078125, 43.59681701660156, 36.51629638671875, -5.130886077880859, -25.724992752075195, 96.3095703125, 80.33995819091797, 77.17921447753906, 152.51980590820312, 109.2733154296875, 157.55810546875, 125.6905517578125, -5.71624755859375, 100.575439453125, 33.922119140625, 148.43426513671875, -2.700777053833008, 29.473709106445312, 62.347503662109375, 79.66493225097656, 3.9683189392089844, 89.448974609375, 72.12738037109375, 109.35012817382812, -33.10563659667969, 5.958526611328125, 68.08119201660156, 80.50727844238281, -1.5934104919433594, -26.857696533203125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000432.npy"}
|
|
{"epoch": 0.9047120418848168, "step": 433, "batch_size": 128, "mean": 37.8322868347168, "std": 62.746856689453125, "min": -141.00628662109375, "p10": -30.294538879394526, "median": 35.965049743652344, "p90": 105.27192764282226, "max": 183.453857421875, "pos_frac": 0.7421875, "sample": [91.85121154785156, -113.03659057617188, -136.78085327148438, 111.44216918945312, 36.60978698730469, 22.792495727539062, 88.85382080078125, -10.359375, -20.389257431030273, 33.702911376953125, 30.20166015625, 24.5125732421875, 102.03060150146484, 62.42022705078125, 55.790435791015625, 17.7803955078125, -14.500640869140625, 91.8289566040039, 52.54803466796875, 16.043716430664062, 36.88018798828125, 8.684814453125, 98.22169494628906, -19.669830322265625, -0.6245307922363281, 34.584651947021484, 136.95852661132812, 2.445262908935547, 146.14495849609375, 102.67578125, 105.29012298583984, -12.339447021484375, 96.732421875, 46.66455078125, 69.94779968261719, 146.65298461914062, 101.40127563476562, 93.75879669189453, -82.28553771972656, 87.643310546875, -8.678001403808594, 27.16424560546875, 6.356536865234375, 44.70530700683594, 26.808761596679688, 85.91976928710938, 62.3612060546875, -1.3219833374023438, 20.17181396484375, 90.00261688232422, 83.08863830566406, 50.917137145996094, 97.80584716796875, -7.39483642578125, 20.81744384765625, -75.28497314453125, 35.3203125, 102.89997863769531, 15.70217514038086, 8.803153991699219, -12.919219970703125, 38.268409729003906, -90.62996673583984, 77.84910583496094, 129.07321166992188, 55.32914733886719, 81.48992919921875, -141.00628662109375, 71.29905700683594, 95.60037231445312, -11.374359130859375, 17.288238525390625, 113.64543151855469, -5.849277496337891, 3.82745361328125, 23.0823974609375, 85.42926788330078, 97.9444808959961, 104.35432434082031, -73.82138061523438, -74.40536499023438, 115.88536071777344, 66.29731750488281, -53.053802490234375, 23.6229248046875, -1.983154296875, 112.71631622314453, -1.2135200500488281, 53.67578125, 31.733428955078125, 98.75180053710938, -1.0335540771484375, -131.38702392578125, 8.589546203613281, 19.71185302734375, 55.11568069458008, 73.24505615234375, 123.19905090332031, 91.16378784179688, -58.5528564453125, 54.606719970703125, -3.5132217407226562, 2.2281455993652344, 183.453857421875, -19.04168701171875, 80.05796813964844, 118.330322265625, 86.47125244140625, 105.26412963867188, 1.199838638305664, 89.12744140625, 26.0618896484375, 48.40992736816406, -29.172500610351562, 31.499176025390625, 70.91604614257812, 43.236549377441406, -4.789524078369141, 88.609130859375, 146.31988525390625, 27.70416259765625, 81.5662841796875, -109.30000305175781, 30.526386260986328, -15.840324401855469, -32.912628173828125, 97.20732116699219, 4.0717010498046875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000433.npy"}
|
|
{"epoch": 0.9068062827225131, "step": 434, "batch_size": 128, "mean": 43.442710876464844, "std": 63.89311599731445, "min": -133.9219970703125, "p10": -20.786094284057615, "median": 28.522804260253906, "p90": 124.7652099609375, "max": 188.09893798828125, "pos_frac": 0.71875, "sample": [21.168075561523438, 5.1830596923828125, -4.53863525390625, 4.31622314453125, 125.14797973632812, 17.04400634765625, 78.21155548095703, 90.24691772460938, -27.197509765625, 178.82855224609375, -10.849681854248047, 116.08003234863281, -39.32769775390625, 5.487274169921875, -21.904178619384766, 6.29315185546875, -18.404754638671875, 92.23703002929688, 57.09571075439453, -2.1326904296875, 17.09735107421875, 177.16522216796875, -6.31927490234375, 188.09893798828125, 123.95175170898438, -69.270263671875, 0.0, -12.725669860839844, -0.35051727294921875, 80.93653869628906, 137.72393798828125, 103.44879150390625, -19.604278564453125, 102.9188232421875, 87.2291259765625, 102.50584411621094, -11.92242431640625, 35.339691162109375, 22.7265625, 137.17269897460938, 41.02119445800781, 72.60159301757812, 124.60116577148438, -35.75299072265625, 27.945892333984375, -68.14178466796875, 83.7735595703125, 89.35598754882812, 26.2694091796875, 79.39324951171875, 126.96549987792969, 107.561767578125, 83.55534362792969, 175.60748291015625, 18.493255615234375, 70.69474029541016, 115.58537292480469, 14.4068603515625, -6.5853271484375, 110.64639282226562, 23.571449279785156, 67.92669677734375, 158.0731201171875, -19.149776458740234, 13.245758056640625, 40.55426025390625, 79.36997985839844, 121.98745727539062, 15.065460205078125, 42.99168395996094, 141.7797088623047, 11.594512939453125, -15.809158325195312, 131.02151489257812, -54.91644287109375, 6.204074859619141, -26.012847900390625, 141.40219116210938, 7.410064697265625, 86.7860107421875, 119.57955932617188, -28.388214111328125, 53.5379638671875, 21.7799072265625, 36.357879638671875, 105.41513061523438, 114.2074203491211, -133.14202880859375, 79.14445495605469, -20.306915283203125, 59.503326416015625, 10.281570434570312, -80.79635620117188, 0.0, 80.8458251953125, 29.099716186523438, 9.937255859375, 124.0396728515625, 128.0208740234375, 6.985561370849609, -3.13543701171875, 0.6609649658203125, -18.173954010009766, 83.01361083984375, -7.5349578857421875, 8.802600860595703, -5.689424514770508, -15.340774536132812, 108.83905029296875, 51.55457305908203, 38.21649169921875, 5.324594497680664, 0.0, 85.9124755859375, 51.899871826171875, -3.425018310546875, 19.261428833007812, 102.6722412109375, 53.928924560546875, 70.2943115234375, -133.9219970703125, 13.201812744140625, 26.8740234375, -7.506744384765625, -79.86334228515625, 121.7400894165039, 66.4100341796875, 110.34752655029297], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000434.npy"}
|
|
{"epoch": 0.9089005235602095, "step": 435, "batch_size": 128, "mean": 39.6096076965332, "std": 61.12794494628906, "min": -128.21063232421875, "p10": -25.72955322265625, "median": 31.369586944580078, "p90": 117.35295257568359, "max": 165.6763916015625, "pos_frac": 0.765625, "sample": [83.39065551757812, 9.7169189453125, 4.309051513671875, 31.59369659423828, 28.243927001953125, 135.8972930908203, 113.16322326660156, -87.05929565429688, 96.06109619140625, -3.942962646484375, -5.715381622314453, -15.097381591796875, -128.21063232421875, 39.42790985107422, 5.873189926147461, -15.548431396484375, 59.69610595703125, 63.55145263671875, 114.98785400390625, 69.00973510742188, 14.516815185546875, 16.302337646484375, 106.79629516601562, 0.0, 3.900829315185547, -1.877349853515625, 50.690673828125, 19.741962432861328, 96.49066162109375, 20.808700561523438, 85.5359878540039, 111.77494812011719, 74.8271484375, -26.25127410888672, -10.260276794433594, 153.92166137695312, 55.833709716796875, 99.0289306640625, 34.98797607421875, 10.949676513671875, 126.9210205078125, 128.51629638671875, 127.28305053710938, 24.731155395507812, 106.53213500976562, 19.94476318359375, 110.03622436523438, -24.939666748046875, 116.02101135253906, 10.8690185546875, 112.72491455078125, 61.992095947265625, 132.88375854492188, 102.994140625, 20.38726234436035, 4.215614318847656, 49.28668212890625, 32.37605285644531, 60.837127685546875, -63.883941650390625, 10.471145629882812, 137.93475341796875, 7.626239776611328, -114.35893249511719, 12.285224914550781, 9.50653076171875, 39.13433837890625, 132.38671875, -87.62651824951172, -61.70854187011719, 23.075515747070312, -1.9851837158203125, 1.233154296875, 139.36843872070312, 135.8900146484375, 7.062713623046875, 102.68446350097656, 114.45906066894531, -47.15135955810547, 70.06330871582031, -9.436614990234375, 104.50601196289062, 51.62744140625, 95.54005432128906, -86.3814697265625, 8.089996337890625, 69.052001953125, 4.226015090942383, 0.47034454345703125, -4.69537353515625, -9.256292343139648, 102.29425048828125, 71.53085327148438, 165.6763916015625, 79.16757202148438, 120.4608154296875, 63.46461486816406, 75.52374267578125, -25.505958557128906, 53.76837158203125, -6.5407257080078125, 39.26298522949219, 44.73406982421875, 3.4884090423583984, 39.165283203125, 21.07879638671875, 31.145477294921875, 9.708450317382812, 92.50788879394531, 6.39727783203125, -7.996337890625, -37.866851806640625, -82.80032348632812, -4.829811096191406, -2.371002197265625, 160.14749145507812, 7.75872802734375, 21.158966064453125, 72.74784851074219, -88.66494750976562, 40.1627197265625, 91.23001098632812, 88.05671691894531, 1.5249786376953125, -28.0850830078125, 101.51432037353516, 30.77801513671875, 89.34669494628906], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000435.npy"}
|
|
{"epoch": 0.9109947643979057, "step": 436, "batch_size": 128, "mean": 28.622535705566406, "std": 61.361045837402344, "min": -121.16293334960938, "p10": -46.68823165893554, "median": 15.226219177246094, "p90": 120.38086242675782, "max": 172.6651611328125, "pos_frac": 0.671875, "sample": [55.426544189453125, -16.747314453125, 135.1546630859375, 93.88557434082031, 3.61737060546875, 27.15118408203125, 113.98814392089844, 36.538360595703125, -31.3897705078125, 88.3812255859375, -58.09332275390625, 45.69842529296875, -50.38328552246094, 41.81414794921875, 15.029754638671875, 136.52374267578125, 26.420654296875, -10.710502624511719, -59.076141357421875, -75.1353759765625, 5.8476715087890625, 91.60845184326172, 93.52247619628906, 93.16363525390625, 136.3466796875, -39.60791015625, 120.5469970703125, -0.909454345703125, 101.53545379638672, 27.12457275390625, 59.629005432128906, 23.16815185546875, -2.403127670288086, 20.9224853515625, 120.73963928222656, 33.25995635986328, -4.444221496582031, 29.621620178222656, 6.829620361328125, -42.57588195800781, 87.99341583251953, 9.87371826171875, 14.979835510253906, 66.47087097167969, 122.54811096191406, 31.01129150390625, -2.2383880615234375, 7.41802978515625, -10.638763427734375, 119.05059814453125, 29.529510498046875, -17.929840087890625, 11.954315185546875, -12.389404296875, 69.38290405273438, -17.95514678955078, 130.31536865234375, -118.24467468261719, -0.157012939453125, 120.30966186523438, 109.52587890625, 27.201828002929688, 10.43524169921875, 64.54067993164062, 41.976470947265625, 101.93600463867188, -24.7991943359375, -86.72833251953125, 6.001190185546875, 98.64801025390625, -4.091705322265625, 172.6651611328125, 74.06201171875, -81.09011840820312, 24.351837158203125, 48.698089599609375, 15.422683715820312, 87.0934829711914, 12.6904296875, -3.825164794921875, 64.1661376953125, -7.8878936767578125, 12.274993896484375, -4.911529541015625, 49.52508544921875, 97.73883056640625, -0.6607208251953125, -58.16754150390625, 28.342750549316406, -52.011016845703125, 126.67425537109375, 4.3033294677734375, -3.1717453002929688, -59.875762939453125, 14.595245361328125, 27.791259765625, 166.6591796875, 5.7554473876953125, 4.922657012939453, -27.886489868164062, -4.179443359375, -121.16293334960938, 13.238155364990234, 60.01409912109375, 52.42747497558594, 10.858642578125, -85.3509521484375, 96.23196411132812, 164.5362548828125, -19.579116821289062, 2.8478164672851562, 26.68951416015625, -34.992218017578125, 91.58029174804688, 6.3545989990234375, 89.44985961914062, 4.46026611328125, 139.52691650390625, -45.104637145996094, 11.60076904296875, -22.8424072265625, 39.793487548828125, -83.5277099609375, 25.18377685546875, -26.113525390625, 123.09112548828125, 33.79069519042969, -1.3336181640625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000436.npy"}
|
|
{"epoch": 0.9130890052356021, "step": 437, "batch_size": 128, "mean": 47.54484939575195, "std": 64.89299774169922, "min": -110.031494140625, "p10": -21.293243408203125, "median": 46.62831115722656, "p90": 131.6446563720703, "max": 174.86724853515625, "pos_frac": 0.75, "sample": [26.1058349609375, 126.55990600585938, 54.20036315917969, 100.55029296875, -99.75360107421875, 105.22772216796875, 4.715087890625, 8.529083251953125, 69.12322998046875, 10.13055419921875, 125.45172119140625, 16.821334838867188, 90.66609191894531, 62.518280029296875, 4.7673492431640625, 100.36114501953125, -21.418319702148438, -2.300872802734375, 36.92950439453125, 106.75299072265625, 3.02496337890625, 134.35513305664062, 152.60821533203125, 123.79267883300781, 4.662345886230469, 23.54461669921875, -3.5530014038085938, 101.76998901367188, 106.51519775390625, 65.16244506835938, -17.734619140625, 13.800872802734375, 7.388458251953125, 70.42951965332031, -1.855438232421875, 18.025381088256836, -92.43960571289062, -16.051727294921875, 7.0539398193359375, 27.005615234375, 103.12199401855469, 49.189544677734375, 131.27362060546875, 120.41848754882812, -34.745330810546875, 2.025299072265625, 26.392822265625, 18.9930419921875, 136.09060668945312, 80.6585693359375, 96.30715942382812, 151.41075134277344, 92.54536437988281, 118.12179565429688, 10.420852661132812, 16.430206298828125, -29.5570068359375, -13.655632019042969, 144.40020751953125, 136.22552490234375, 106.39920806884766, -1.9901123046875, 122.25811767578125, 136.56729125976562, 164.647216796875, 56.18975067138672, 10.224895477294922, 56.0355224609375, -21.239639282226562, -15.847846984863281, 76.67124938964844, 5.094970703125, -30.14544677734375, 132.51040649414062, -110.031494140625, 117.02957153320312, 3.1236696243286133, 61.56452941894531, -76.19189453125, 0.0, 118.27117919921875, -58.20979309082031, -0.7589263916015625, 1.5227375030517578, 135.31619262695312, 93.456298828125, -50.193145751953125, 44.2772216796875, -90.7442626953125, 65.2474365234375, 144.8505859375, 63.098236083984375, -3.3626708984375, 91.71481323242188, 19.544921875, -2.896728515625, 83.33139038085938, 21.672988891601562, 74.07659912109375, 80.17774200439453, 126.18820190429688, 140.403564453125, 85.410888671875, 127.3416748046875, 36.71563720703125, -17.029449462890625, 48.979400634765625, 78.77739715576172, 123.66738891601562, -87.67965698242188, 3.0729827880859375, -0.935394287109375, 104.44741821289062, -102.434814453125, -13.276092529296875, 112.71485137939453, 36.8687744140625, 113.37908935546875, 94.85073852539062, -5.144287109375, 71.96334838867188, -10.721466064453125, 24.217376708984375, 65.80377197265625, 0.0, 18.477508544921875, 102.04120635986328, 174.86724853515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000437.npy"}
|
|
{"epoch": 0.9151832460732985, "step": 438, "batch_size": 128, "mean": 22.82562255859375, "std": 63.92110061645508, "min": -117.67965698242188, "p10": -64.64688110351562, "median": 11.198746681213379, "p90": 111.57598724365234, "max": 174.6400146484375, "pos_frac": 0.640625, "sample": [13.557472229003906, -117.67965698242188, 81.502197265625, -9.442779541015625, -15.741622924804688, 73.09765625, 41.3385009765625, 27.929061889648438, 0.0, 11.47416877746582, 134.13986206054688, -69.68275451660156, 16.32036590576172, 110.09039306640625, 38.139129638671875, 9.276657104492188, 18.46099853515625, 112.80039978027344, 78.34334564208984, -15.068107604980469, -33.082489013671875, 5.22747802734375, 47.87933349609375, 1.082489013671875, -62.995819091796875, -48.436676025390625, -18.85052490234375, 94.813720703125, 9.4224853515625, 33.369873046875, 12.100488662719727, 97.55804443359375, 74.9539794921875, -4.6179046630859375, 26.954833984375, 113.0574951171875, -90.77755737304688, -27.91027069091797, 25.3912353515625, -116.683349609375, -109.05780029296875, -73.8447265625, 58.21807861328125, 87.7933578491211, -13.454055786132812, 6.0131683349609375, 3.678466796875, 23.138885498046875, -99.02175903320312, -3.4983558654785156, 1.3989715576171875, -97.39239501953125, -12.510688781738281, 17.371780395507812, 7.0107421875, 22.014419555664062, 75.564453125, -0.5238037109375, 64.11898803710938, 77.62596893310547, 9.692108154296875, -21.99724006652832, -2.0267887115478516, -106.77205657958984, 53.4954833984375, -9.187910079956055, 1.1309814453125, 17.406265258789062, 117.33102416992188, -83.19845581054688, 89.79415893554688, -49.842979431152344, 120.36099243164062, -10.177017211914062, 32.104339599609375, 71.26287841796875, 84.75474548339844, -68.49935913085938, 104.27960205078125, 111.05123901367188, -35.171783447265625, 109.4515380859375, -49.977378845214844, 1.451202392578125, 148.85791015625, -32.47462463378906, 98.1796875, 98.23757934570312, -95.354736328125, 4.3737945556640625, 65.50238037109375, 10.923324584960938, 1.3386783599853516, 26.55987548828125, 126.45681762695312, 103.92529296875, -5.9712371826171875, 120.45703125, 165.187255859375, -19.77520751953125, -15.185211181640625, -27.643157958984375, 35.56353759765625, 79.30560302734375, 91.15731048583984, -16.9183349609375, 18.285110473632812, 12.485603332519531, -18.936676025390625, 107.56939697265625, 130.77252197265625, 10.383512496948242, -34.11328125, 22.805034637451172, 0.0, -2.44964599609375, 54.630428314208984, 7.230842590332031, -6.214515686035156, 82.80192565917969, 121.25811767578125, -70.61082458496094, 8.449934005737305, 116.88581848144531, 174.6400146484375, -7.417213439941406, 6.22967529296875, 25.618911743164062], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000438.npy"}
|
|
{"epoch": 0.9172774869109948, "step": 439, "batch_size": 128, "mean": 40.22080612182617, "std": 62.27999496459961, "min": -135.8629150390625, "p10": -32.07881622314452, "median": 27.843647003173828, "p90": 120.92185058593749, "max": 159.8377685546875, "pos_frac": 0.703125, "sample": [-39.68218994140625, -64.25775146484375, -102.70907592773438, 150.73284912109375, -53.398193359375, 119.47259521484375, 20.392608642578125, 17.63884735107422, 37.01344299316406, 85.84927368164062, -2.0734806060791016, -1.322113037109375, 24.2391414642334, 0.201934814453125, 68.3817138671875, -79.68878173828125, 8.296028137207031, 79.78378295898438, -2.640777587890625, 37.83258056640625, 141.965087890625, 112.10687255859375, 6.101654052734375, 45.13018798828125, 1.74481201171875, 102.53089141845703, 28.554397583007812, 108.32733917236328, 120.46194458007812, 18.675617218017578, -15.589141845703125, 65.61196899414062, 7.559043884277344, 150.90463256835938, -22.292572021484375, 79.18011474609375, 152.2314453125, 51.84844970703125, 159.8377685546875, 21.359649658203125, -27.982757568359375, 15.101802825927734, 107.43453979492188, 108.34629821777344, 9.093597412109375, 94.52569580078125, 101.94210815429688, 13.590545654296875, 3.9513092041015625, -7.3556365966796875, 0.447509765625, 13.22353744506836, -36.99627685546875, 13.411285400390625, 71.82774353027344, 121.330810546875, 96.01521301269531, -18.954307556152344, 54.35679626464844, 0.0, 22.903472900390625, 0.0, -26.88525390625, 20.81981658935547, -26.3450927734375, 119.38833618164062, 23.1566162109375, -36.74958801269531, 11.193655014038086, -30.077056884765625, 1.211517333984375, 122.261474609375, 30.99163818359375, -23.697174072265625, -7.892719268798828, 27.132896423339844, -8.439754486083984, 78.43828582763672, -38.417724609375, 74.48760986328125, -42.09001159667969, 71.3281478881836, 86.38969421386719, 24.739364624023438, 55.46031188964844, 73.58517456054688, -26.191818237304688, -5.6808929443359375, -50.91349792480469, -7.462158203125, 118.56866455078125, -97.99075317382812, 0.0, 124.28048706054688, 81.5963134765625, 51.56914520263672, 148.09605407714844, 108.82644653320312, 35.960418701171875, -15.451446533203125, 36.91357421875, -38.53582763671875, 87.03440856933594, 35.55303955078125, 44.12086486816406, 23.364776611328125, 75.57598876953125, 86.1799545288086, 99.54241180419922, -23.34722900390625, 117.86515808105469, 65.93292236328125, 56.39182662963867, 120.74658203125, 108.45791625976562, -1.6260509490966797, -135.8629150390625, 137.29505920410156, 86.05300903320312, 134.2364501953125, 117.14096069335938, 98.19755554199219, 12.525993347167969, -19.2969970703125, 135.68765258789062, 0.0, 131.9763641357422, 114.41656494140625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000439.npy"}
|
|
{"epoch": 0.9193717277486911, "step": 440, "batch_size": 128, "mean": 31.307819366455078, "std": 64.8756332397461, "min": -133.59613037109375, "p10": -49.29643936157226, "median": 21.776885986328125, "p90": 120.68885650634763, "max": 190.44952392578125, "pos_frac": 0.703125, "sample": [-31.144195556640625, 0.0, 31.4586181640625, -98.55914306640625, -16.527618408203125, 55.543853759765625, -92.23147583007812, 82.22834777832031, -3.4511947631835938, 62.97201156616211, 22.0389404296875, 8.213134765625, -67.98333740234375, 2.167144775390625, 47.96673583984375, -76.49163818359375, 69.95024108886719, 96.23837280273438, 95.92814636230469, 85.4228515625, 41.259918212890625, 139.39877319335938, -2.149566650390625, 16.803558349609375, -50.60772705078125, -2.4742889404296875, -133.59613037109375, 85.92475891113281, 45.435028076171875, 84.3865966796875, 138.391357421875, 152.832275390625, 6.145420074462891, -14.6795654296875, -17.59503173828125, 3.645111083984375, 96.16452026367188, -109.95220947265625, 0.0, 45.9267578125, -124.1884765625, 0.879669189453125, 9.47998046875, 48.87639617919922, 126.21702575683594, 2.2243576049804688, -66.36358642578125, -0.2187042236328125, 13.649238586425781, 102.04310607910156, 154.94064331054688, -13.196525573730469, 95.0169677734375, 0.753692626953125, 31.138671875, 135.93682861328125, -2.6475448608398438, -87.39572143554688, -43.306182861328125, -22.308685302734375, -30.994781494140625, 2.5501708984375, 11.881454467773438, -31.75958251953125, 190.44952392578125, 54.01556396484375, 146.62747192382812, 99.40939331054688, 86.1304931640625, 48.4510498046875, -1.8753623962402344, 18.84661865234375, 10.892288208007812, 94.83383178710938, 111.72055053710938, 15.595794677734375, 1.898468017578125, -60.364864349365234, 21.464691162109375, 14.399078369140625, 27.523117065429688, 66.3804931640625, 103.03600311279297, 46.83489990234375, 132.24884033203125, -69.08114624023438, 21.51483154296875, 135.947021484375, 38.410888671875, -62.32282257080078, 142.71755981445312, -42.01835632324219, 115.54155731201172, -9.120429992675781, 96.51904296875, 22.738449096679688, 70.02412414550781, 30.007102966308594, 77.85040283203125, 37.11895751953125, 118.31964111328125, -10.39984130859375, 22.47607421875, 16.92139434814453, 41.127349853515625, 62.6021728515625, 5.9414825439453125, 130.34210205078125, 54.97809600830078, 59.746368408203125, 12.17755126953125, 46.3466796875, 18.32574462890625, 110.77243041992188, -41.314910888671875, 105.91180419921875, 62.0484619140625, 19.694580078125, 10.35394287109375, 67.531494140625, -24.138671875, 138.60304260253906, 24.214950561523438, -45.18321228027344, -15.793426513671875, 11.70465087890625, -48.734458923339844, 106.25265502929688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000440.npy"}
|
|
{"epoch": 0.9214659685863874, "step": 441, "batch_size": 128, "mean": 47.051849365234375, "std": 58.32120895385742, "min": -90.602783203125, "p10": -15.440638732910154, "median": 40.85673522949219, "p90": 124.81094512939453, "max": 160.411376953125, "pos_frac": 0.7734375, "sample": [-12.131912231445312, 32.1292724609375, -90.602783203125, 83.9476318359375, -0.5837001800537109, -9.9423828125, 139.5751953125, -14.707763671875, 6.848388671875, 0.581787109375, -47.09300231933594, 160.05520629882812, 26.535987854003906, 103.06890869140625, 29.427093505859375, 103.97213745117188, 107.80770874023438, 113.66891479492188, 89.3591079711914, 76.05767822265625, 51.501953125, 96.1429443359375, 13.756711959838867, 14.65997314453125, 92.3939208984375, 51.42008972167969, 96.44134521484375, 124.95137023925781, 64.448486328125, 6.010833740234375, 23.25872802734375, 101.0093994140625, 64.26161193847656, -9.926445007324219, 104.5255126953125, 90.88336181640625, 14.0806884765625, 124.75076293945312, 92.93760681152344, -46.81114959716797, 115.73196411132812, 97.80079650878906, -17.574607849121094, -9.37493896484375, 98.14175415039062, -0.9574909210205078, 40.176727294921875, 31.969375610351562, 76.45074462890625, 102.608642578125, 116.42875671386719, 10.0037841796875, 39.8497314453125, 11.774925231933594, 15.4886474609375, 10.65240478515625, 0.516265869140625, 100.325927734375, 11.786636352539062, -12.398834228515625, 6.55938720703125, 23.202301025390625, 3.495574951171875, 106.18289184570312, -3.6566162109375, 0.0, 3.3895950317382812, 56.18583679199219, 42.77379608154297, 129.85107421875, -42.397003173828125, -35.36796188354492, -6.4194488525390625, 21.073501586914062, 0.0, 97.85014343261719, 41.9873046875, 44.808746337890625, 106.09271240234375, 64.61117553710938, 8.865562438964844, 114.6103515625, 24.191360473632812, -58.013031005859375, 42.367279052734375, 73.02261352539062, 102.71755981445312, 75.71575927734375, -2.466217041015625, 18.87884521484375, 3.8133544921875, 25.0423583984375, 33.44940185546875, 74.46673583984375, 92.52522277832031, 8.750762939453125, 130.54869079589844, -74.83935546875, 137.08056640625, 41.5367431640625, 133.84686279296875, -87.33187866210938, 58.071258544921875, 11.398651123046875, -70.39337158203125, -17.150680541992188, 144.70254516601562, -75.92658996582031, 115.86033630371094, -3.70648193359375, 24.23426055908203, 131.71942138671875, 131.39932250976562, 129.13536071777344, 140.15237426757812, 77.63690185546875, 66.35313415527344, -3.8887939453125, -27.17694091796875, 7.5623779296875, 160.411376953125, -3.5345916748046875, 105.79586029052734, 110.35296630859375, 109.43475341796875, 83.31208801269531, 97.69080352783203, 32.1156005859375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000441.npy"}
|
|
{"epoch": 0.9235602094240838, "step": 442, "batch_size": 128, "mean": 35.86796569824219, "std": 70.29293823242188, "min": -145.811767578125, "p10": -54.25686492919922, "median": 26.73849868774414, "p90": 128.41692504882812, "max": 198.09405517578125, "pos_frac": 0.734375, "sample": [104.91458129882812, 12.273223876953125, 20.3543701171875, 32.60117721557617, 160.9638671875, 119.7891845703125, 1.0223064422607422, 1.00506591796875, 132.26458740234375, 3.0762481689453125, 64.42585754394531, 19.673660278320312, -56.207542419433594, -7.9459686279296875, 99.50546264648438, -108.70947265625, 21.683090209960938, 21.510772705078125, 8.691650390625, -93.16958618164062, 15.321533203125, 37.10528564453125, -26.885719299316406, -8.089303970336914, 87.59176635742188, -8.846572875976562, 91.58447265625, -38.767669677734375, 143.4917755126953, 135.00799560546875, -100.31527709960938, 5.7964935302734375, -42.40771484375, 111.22932434082031, 77.77384948730469, 108.68887329101562, 12.026779174804688, 11.166534423828125, -89.39605712890625, 4.5730133056640625, 20.406784057617188, -53.420860290527344, 56.34437561035156, 6.840370178222656, -23.167999267578125, -108.9234619140625, 17.006973266601562, 34.99652099609375, 6.4005584716796875, -2.66265869140625, 14.96484375, -3.587982177734375, 72.2701416015625, 147.77081298828125, 6.774715423583984, 54.38189697265625, 99.78834533691406, 0.0, 55.924224853515625, -73.37930297851562, -45.775848388671875, 1.8171005249023438, -111.54095458984375, 93.62484741210938, -102.4349365234375, 143.56997680664062, 75.77603149414062, -6.4355926513671875, 44.734588623046875, 96.1765365600586, 0.1682567596435547, -145.811767578125, 137.88812255859375, -7.6055145263671875, 83.55886840820312, 51.919403076171875, 154.44467163085938, -19.359375, 94.28256225585938, 27.372116088867188, 31.064682006835938, 60.036163330078125, -6.031642913818359, 149.6897735595703, 113.18325805664062, 116.377685546875, -49.4932861328125, -98.2476806640625, -30.409652709960938, -50.863258361816406, 15.646026611328125, 48.68915557861328, 11.903511047363281, 92.3896484375, 5.715118408203125, 100.31678009033203, 92.12767028808594, 104.0727310180664, 89.33135223388672, 108.52825164794922, 137.00149536132812, 126.97967529296875, 115.29935455322266, 131.7705078125, 25.462371826171875, 198.09405517578125, 17.16644287109375, -29.723472595214844, -26.524429321289062, 157.78713989257812, 97.86529541015625, 64.38955688476562, 40.123382568359375, 96.58522033691406, 4.8942718505859375, 23.917037963867188, 26.104881286621094, 38.55059814453125, 95.43194580078125, -56.733154296875, 56.25160217285156, 102.44902801513672, 114.72077178955078, 29.97161865234375, -92.11497497558594, 56.80670166015625, 98.49374389648438, 56.5828857421875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000442.npy"}
|
|
{"epoch": 0.9256544502617801, "step": 443, "batch_size": 128, "mean": 36.676849365234375, "std": 62.4803581237793, "min": -127.63427734375, "p10": -28.69862289428711, "median": 27.184965133666992, "p90": 122.3864059448242, "max": 175.9586181640625, "pos_frac": 0.7109375, "sample": [51.680389404296875, 11.175445556640625, 84.4718017578125, 17.071548461914062, -33.11248779296875, 49.892822265625, 144.3563995361328, 9.060089111328125, 132.216796875, -10.191169738769531, 23.177268981933594, 68.48516845703125, 1.2971572875976562, 86.5831298828125, -9.978759765625, -67.22297668457031, -11.075042724609375, 74.79674530029297, 119.54920959472656, -5.349842071533203, 34.19902038574219, 82.1280517578125, 60.69697570800781, 30.5665283203125, -25.589736938476562, 23.190223693847656, 78.619384765625, 111.82998657226562, 26.806060791015625, -28.145912170410156, -15.3675537109375, 59.5355224609375, -101.32785034179688, 7.324634552001953, -9.2325439453125, 33.308380126953125, 104.66046142578125, 30.329071044921875, 118.74561309814453, 10.891185760498047, 135.02732849121094, -15.475494384765625, -6.6417999267578125, 3.5594940185546875, 14.967636108398438, 24.000572204589844, 175.9586181640625, 4.8064727783203125, 55.155059814453125, -32.59111022949219, 56.8812255859375, -127.62034606933594, -16.34234619140625, -25.6427001953125, 55.76483154296875, 98.5987319946289, 74.75303649902344, 164.77157592773438, 29.314773559570312, 137.13583374023438, 88.66790771484375, 31.732818603515625, -7.665718078613281, 69.74136352539062, 77.257568359375, 9.379508972167969, -6.4316558837890625, 58.934547424316406, 94.77833557128906, 112.14913940429688, -6.0957183837890625, -3.352783203125, 134.39434814453125, 9.225078582763672, 121.39125061035156, 8.22418212890625, -0.19002532958984375, 100.23912048339844, -69.59405517578125, 8.830078125, -15.981422424316406, 124.70843505859375, 112.0096435546875, 67.26488494873047, 46.3287353515625, -49.385398864746094, 14.692642211914062, -51.51301574707031, 97.57421875, 159.12469482421875, 127.61138916015625, 72.51780700683594, 31.361328125, 23.319442749023438, -45.63529968261719, 93.31874084472656, 9.773178100585938, 151.40960693359375, 96.20530700683594, 95.79977416992188, 67.4769287109375, -8.33319091796875, -116.63162231445312, 99.26687622070312, 6.7325439453125, 2.56402587890625, 46.25762939453125, 73.92083740234375, 54.59368896484375, 6.1803131103515625, 44.71040344238281, -17.37603759765625, -29.98828125, -7.030342102050781, -0.2632904052734375, -8.451332092285156, 16.275054931640625, 131.4569549560547, 27.56386947631836, -127.63427734375, 108.57720947265625, 98.66844177246094, 126.76377868652344, 7.304893493652344, -0.15069580078125, -94.33950805664062, 8.169326782226562, 9.799530029296875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000443.npy"}
|
|
{"epoch": 0.9277486910994764, "step": 444, "batch_size": 128, "mean": 39.89689636230469, "std": 65.93734741210938, "min": -153.41387939453125, "p10": -30.041015624999996, "median": 23.78229522705078, "p90": 127.12097778320313, "max": 185.4971923828125, "pos_frac": 0.6953125, "sample": [-1.7376136779785156, 14.6383056640625, 132.97946166992188, 17.95745849609375, 75.62905883789062, 68.18302917480469, -31.308013916015625, 1.83062744140625, 19.27448272705078, 13.631271362304688, -9.52374267578125, 25.2130126953125, -25.46990966796875, 31.6307373046875, 94.37997436523438, 34.2374267578125, 96.80052185058594, 39.002044677734375, -22.10858154296875, 22.23406219482422, 81.61088562011719, 26.16485595703125, 91.38876342773438, -25.3275146484375, 17.21038818359375, -29.498016357421875, 112.2965087890625, 15.42987060546875, 4.976234436035156, 24.303253173828125, 14.741806030273438, 107.85912322998047, 85.78752136230469, -11.572677612304688, 125.45472717285156, -22.295913696289062, -28.363571166992188, 47.432281494140625, 124.4561767578125, 102.1922607421875, 98.8563232421875, 132.47714233398438, -13.55499267578125, -5.614171981811523, 35.43463134765625, 108.10714721679688, 17.1788330078125, -77.9693374633789, -88.08798217773438, -18.792236328125, 173.4156494140625, 88.87954711914062, -48.828948974609375, 159.37518310546875, -3.68963623046875, 66.4141845703125, 185.4971923828125, 130.0989990234375, -9.739959716796875, -6.373565673828125, 0.7639675140380859, 159.41571044921875, 103.26365661621094, -18.323562622070312, 55.20185852050781, -17.062469482421875, 139.68899536132812, 82.2857666015625, -75.13955688476562, 157.38323974609375, 96.26060485839844, 48.67271423339844, -24.39166259765625, 127.16891479492188, -13.780914306640625, 54.0902099609375, 126.64202880859375, 124.9981689453125, 101.95904541015625, 21.950942993164062, 127.10043334960938, -8.708623886108398, 2.4078102111816406, 131.111572265625, -8.03769302368164, 117.22518920898438, -6.486595153808594, 107.39373779296875, 74.63285064697266, -92.62360382080078, 1.241455078125, 143.30770874023438, -5.714267730712891, 80.88497924804688, 30.21819305419922, 23.261337280273438, 48.282989501953125, -35.884796142578125, 7.2327117919921875, -45.08380126953125, -14.07183837890625, 5.908416748046875, 122.23471069335938, 4.403717041015625, -31.342361450195312, 42.2103271484375, -32.060279846191406, 106.15179443359375, -35.16117858886719, 98.650146484375, 4.554071426391602, 66.60506439208984, 9.77960205078125, 82.11183166503906, 68.11640930175781, 22.49911117553711, 131.3406982421875, 103.77459716796875, 99.10540008544922, 100.07736206054688, -2.9335174560546875, 19.388622283935547, -24.991485595703125, 0.7990245819091797, -129.60736083984375, 22.77410888671875, 92.28531646728516, -153.41387939453125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000444.npy"}
|
|
{"epoch": 0.9298429319371728, "step": 445, "batch_size": 128, "mean": 33.944793701171875, "std": 63.69799041748047, "min": -117.56668090820312, "p10": -34.55117492675781, "median": 24.71011734008789, "p90": 125.16314086914062, "max": 172.91336059570312, "pos_frac": 0.6953125, "sample": [9.845146179199219, -117.55455017089844, 138.625244140625, 54.1644287109375, 36.069061279296875, 136.60562133789062, -3.6093063354492188, 35.650115966796875, 130.7833251953125, -77.23052978515625, 9.004608154296875, 69.78215026855469, 112.87203979492188, -52.479766845703125, -31.585662841796875, 4.734954833984375, 40.875579833984375, 16.958908081054688, 150.3516845703125, 31.026260375976562, 128.9188232421875, -1.564208984375, 25.359359741210938, -63.016448974609375, 26.453125, 91.125244140625, 28.266769409179688, -1.7893714904785156, 84.57478332519531, 20.06330108642578, 144.869384765625, 74.75741577148438, 37.76300048828125, 4.742195129394531, -21.600860595703125, -9.255905151367188, 21.073978424072266, 24.060874938964844, 92.14502716064453, 64.97015380859375, 8.4493408203125, 128.62969970703125, 142.19705200195312, -0.56005859375, 88.56610870361328, 3.7104969024658203, 12.343559265136719, 26.150177001953125, 51.050048828125, -90.260986328125, -33.80162811279297, 51.817169189453125, 90.13165283203125, -6.8336334228515625, 98.4642562866211, 0.12090301513671875, 39.83367919921875, 125.49908447265625, 10.548736572265625, 112.01075744628906, 69.25608825683594, 32.573760986328125, 172.91336059570312, 32.534027099609375, 112.89570617675781, 87.01763916015625, 82.69229125976562, -107.97871398925781, 43.446380615234375, 148.12188720703125, 42.92694091796875, 72.536865234375, -20.110137939453125, 99.2833251953125, 96.38151550292969, -8.026931762695312, -11.275238037109375, -98.05081176757812, 121.03887939453125, 68.46919250488281, -1.065826416015625, -14.789764404296875, -32.168548583984375, -89.84466552734375, 5.347080230712891, 33.6800537109375, 16.206954956054688, -33.06597900390625, 74.52648162841797, 1.3857345581054688, 43.92529296875, 100.79815673828125, 148.44296264648438, 110.37115478515625, 39.4639892578125, 80.4981689453125, -2.5602359771728516, -104.4759521484375, 133.30445861816406, -4.89727783203125, 6.017463684082031, 56.186256408691406, 119.34954833984375, 21.559104919433594, -6.6243896484375, 15.32720947265625, 8.546432495117188, 15.624412536621094, 125.0191650390625, 3.8040771484375, -34.4837646484375, -4.2105865478515625, 0.0, 10.006580352783203, 11.022186279296875, -34.708465576171875, -2.575836181640625, -5.130035400390625, 16.297622680664062, 121.57662963867188, -41.2080078125, -36.723388671875, -117.56668090820312, -0.9666042327880859, 47.16583251953125, 103.54104614257812, -29.52771759033203, 111.01484680175781], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000445.npy"}
|
|
{"epoch": 0.9319371727748691, "step": 446, "batch_size": 128, "mean": 44.44334030151367, "std": 64.5469741821289, "min": -143.82470703125, "p10": -20.80267639160156, "median": 37.3518009185791, "p90": 125.4720573425293, "max": 180.51065063476562, "pos_frac": 0.75, "sample": [91.82269287109375, 148.30279541015625, 41.14134979248047, 7.1396484375, 0.662841796875, 37.94789123535156, 109.83726501464844, 47.094764709472656, 83.93525695800781, -13.193161010742188, 31.824020385742188, -55.85270690917969, 72.51324462890625, 27.12926483154297, 11.511383056640625, 36.75571060180664, 153.53802490234375, 15.684417724609375, -18.52545166015625, -6.5109405517578125, 18.52984619140625, 92.71136474609375, 124.12640380859375, 100.75674438476562, -6.177970886230469, -68.65823364257812, 121.64950561523438, 5.0049896240234375, 5.099639892578125, -27.044418334960938, 10.547332763671875, 11.7294921875, -24.263092041015625, 27.090126037597656, -45.42559814453125, 17.833106994628906, 124.90119171142578, 1.3708114624023438, 20.595977783203125, 53.09949493408203, 96.78021240234375, 6.738471984863281, 70.03661346435547, 25.643234252929688, 41.287933349609375, 98.63912963867188, 126.8040771484375, 103.8095703125, 115.67461395263672, -143.82470703125, 30.045745849609375, 66.25178527832031, 156.32830810546875, 2.4336471557617188, 97.30055236816406, 99.78778076171875, 22.2137451171875, 36.06193542480469, -139.7440185546875, 78.27313232421875, 133.97396850585938, 0.24354934692382812, -16.07623291015625, 0.0, 35.855560302734375, 116.72511291503906, 42.899070739746094, 71.84559631347656, 86.56884765625, 108.97815704345703, -4.563751220703125, 106.29249572753906, 90.03927612304688, 88.62127685546875, 90.7603759765625, 10.124267578125, 122.09844970703125, 98.48381042480469, -16.606597900390625, 0.0, 76.49789428710938, -19.31964111328125, 146.47500610351562, -11.61529541015625, 22.285049438476562, -39.226806640625, 60.86529541015625, 150.93600463867188, 48.41485595703125, 54.6702880859375, -25.27825927734375, 180.51065063476562, 158.68862915039062, 121.43084716796875, 85.67202758789062, 48.6865234375, 62.59747314453125, 96.59600830078125, -11.483272552490234, 103.4295654296875, 16.647907257080078, -14.503326416015625, 147.2056884765625, 138.62222290039062, -132.07101440429688, -100.81710815429688, -10.95068359375, 18.0318603515625, -8.919326782226562, 2.4827880859375, 43.31446838378906, 0.0, -26.965560913085938, -4.5464324951171875, 101.42793273925781, -43.55731201171875, 146.706298828125, 26.3431396484375, 157.67178344726562, 16.4443359375, 6.70208740234375, 48.09373474121094, -13.243606567382812, -10.478439331054688, 118.3834228515625, 43.823204040527344, 100.09049987792969, 68.93606567382812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000446.npy"}
|
|
{"epoch": 0.9340314136125655, "step": 447, "batch_size": 128, "mean": 46.969024658203125, "std": 57.49828338623047, "min": -125.82522583007812, "p10": -17.92784271240234, "median": 38.748619079589844, "p90": 121.76341705322265, "max": 167.18328857421875, "pos_frac": 0.7734375, "sample": [-32.94493103027344, 9.117008209228516, 104.98214721679688, 105.72575378417969, 33.57264709472656, 9.17817497253418, 140.7137451171875, 9.233474731445312, -32.65765380859375, 116.50848388671875, 36.50480651855469, 86.16679382324219, 40.992431640625, 125.54876708984375, 110.89411926269531, 51.801822662353516, 0.0, 45.123809814453125, 131.62020874023438, 9.89291763305664, -1.5898590087890625, 43.94732666015625, 105.12973022460938, 18.92059326171875, 36.38189697265625, -2.18658447265625, 1.683258056640625, 77.14753723144531, -97.71340942382812, 99.5738525390625, -75.23780822753906, 21.11865234375, 82.39483642578125, 64.8006591796875, -9.405731201171875, 27.96881866455078, 100.04096984863281, 102.58206176757812, 20.510581970214844, 99.55744934082031, 6.7375640869140625, 93.68914794921875, 134.02975463867188, 131.51004028320312, 20.503646850585938, 0.724395751953125, 103.24777221679688, 108.93801879882812, 117.964111328125, 0.0, 123.27139282226562, 12.346923828125, 45.111572265625, 98.69728088378906, -25.595474243164062, -36.45796203613281, -6.744482040405273, 6.8743896484375, -32.322776794433594, 123.56349182128906, 6.126373291015625, 36.021949768066406, 20.612152099609375, 23.946914672851562, -42.15167236328125, 139.8253173828125, 70.08760070800781, -21.835723876953125, -14.6148681640625, 13.690238952636719, 41.98089599609375, -17.534957885742188, -18.844573974609375, 10.47259521484375, 59.867950439453125, 1.1319389343261719, 32.33559036254883, -16.232940673828125, 52.3829345703125, 119.2105712890625, 162.714599609375, -1.99334716796875, -3.140888214111328, 64.64608764648438, -125.82522583007812, 0.0, 6.611030578613281, 110.98503875732422, 32.958770751953125, 115.07464599609375, 47.8861083984375, 30.935760498046875, 28.881866455078125, 93.3072509765625, 161.35284423828125, 84.37591552734375, 113.35504913330078, 167.18328857421875, 0.0, 121.11714172363281, 62.341217041015625, 108.0326919555664, 25.4228515625, 26.899200439453125, 32.7337646484375, 68.68548583984375, 26.463607788085938, -63.37359619140625, 113.16371154785156, 77.93582153320312, -8.0625, 91.56195068359375, 50.66230773925781, 74.87103271484375, 104.98260498046875, 61.01481628417969, 6.829620361328125, 91.81636047363281, -7.876487731933594, -15.47352123260498, 57.836639404296875, 128.79986572265625, 103.365234375, 49.01161193847656, 19.666015625, 130.64651489257812, -23.600067138671875, 67.08582305908203], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000447.npy"}
|
|
{"epoch": 0.9361256544502617, "step": 448, "batch_size": 128, "mean": 41.08341979980469, "std": 60.350467681884766, "min": -108.0516357421875, "p10": -24.05663604736328, "median": 31.748207092285156, "p90": 123.39021606445311, "max": 181.2852783203125, "pos_frac": 0.7734375, "sample": [4.2767333984375, 8.571807861328125, -55.33253479003906, 12.865020751953125, 63.407012939453125, 134.601806640625, 2.133636474609375, -78.96051025390625, 67.11155700683594, 112.85652160644531, -6.453399658203125, -40.294281005859375, 40.325531005859375, 33.48260498046875, 3.0413818359375, 61.71466064453125, 97.73863983154297, 106.81619262695312, 63.19384765625, -6.784463882446289, 46.244529724121094, 131.6590576171875, 116.69660949707031, 5.092384338378906, 31.126007080078125, 24.30506134033203, 39.282958984375, -23.861953735351562, 7.47509765625, 3.778778076171875, -8.480842590332031, 37.89300537109375, 122.15359497070312, -69.50479125976562, 97.24533081054688, -26.79058837890625, 119.7213134765625, 128.53973388671875, 44.894744873046875, 48.157562255859375, 63.540496826171875, 30.736740112304688, 140.89822387695312, 1.426544189453125, 43.70402526855469, 10.960060119628906, 32.81511688232422, 91.42559051513672, -4.7884979248046875, 162.27117919921875, -7.209716796875, -3.684600830078125, -4.82843017578125, 72.79986572265625, 4.472602844238281, -13.748077392578125, 46.799339294433594, 87.40313720703125, -39.05975341796875, 126.27566528320312, 100.1326904296875, 56.339508056640625, 14.467559814453125, 1.4804420471191406, 128.2650146484375, 143.65716552734375, 20.315418243408203, 9.980728149414062, 114.82455444335938, 141.65777587890625, 114.47416687011719, 81.29798889160156, 83.47720336914062, 34.040283203125, 10.814834594726562, 59.7105712890625, 132.04769897460938, 109.05473327636719, 108.18787384033203, 53.371795654296875, -27.275909423828125, 127.17862701416016, -4.6713714599609375, 119.13505554199219, 10.022262573242188, -99.6282958984375, -96.8121337890625, -108.0516357421875, 108.80191040039062, 22.600189208984375, -10.194976806640625, -0.7170257568359375, 105.70756530761719, 120.62092590332031, 24.782760620117188, -90.343505859375, 20.531661987304688, -2.119020462036133, 32.78631591796875, 86.24235534667969, 108.0279541015625, 12.357770919799805, 129.75845336914062, 5.992374420166016, 60.690338134765625, -24.510894775390625, -64.96067810058594, 72.36668395996094, 32.37040710449219, 27.027008056640625, 121.43916320800781, -17.496917724609375, -7.299896240234375, 17.05133056640625, -17.699047088623047, 19.28582763671875, 8.01214599609375, 16.738800048828125, 15.94720458984375, 0.9276390075683594, 75.59485626220703, 2.660675048828125, 9.056295394897461, 121.33716583251953, 181.2852783203125, 75.32844543457031, 48.94903564453125, 30.097579956054688], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000448.npy"}
|
|
{"epoch": 0.9382198952879581, "step": 449, "batch_size": 128, "mean": 46.20602798461914, "std": 61.05792999267578, "min": -133.3062744140625, "p10": -21.719542312622067, "median": 44.239315032958984, "p90": 126.44100341796874, "max": 163.0101318359375, "pos_frac": 0.734375, "sample": [58.57855224609375, 61.1907958984375, -38.42152404785156, 96.9103775024414, 48.94805908203125, -12.8831787109375, -1.009490966796875, 138.225830078125, -5.8959197998046875, 14.124942779541016, 14.649520874023438, 20.07415771484375, 71.07077026367188, 45.14769744873047, -4.0137939453125, 104.26412200927734, 141.1173858642578, 126.40277099609375, 16.13702392578125, -4.023517608642578, -57.87849426269531, 98.43717956542969, 96.8306884765625, 108.96875762939453, 43.3309326171875, 70.25115203857422, -50.543731689453125, 63.692626953125, 5.7800798416137695, 131.9954833984375, 152.7557373046875, 131.15762329101562, 120.69775390625, 135.03109741210938, 24.23931884765625, 81.28768920898438, 17.5350341796875, 68.17091369628906, 69.43377685546875, 88.78387451171875, 126.53021240234375, 38.51104736328125, 96.80624389648438, 12.201332092285156, 163.0101318359375, 151.17477416992188, 4.212799072265625, -3.4895572662353516, 136.8369140625, 115.28521728515625, -26.292755126953125, 62.69966506958008, 21.086807250976562, 41.86189270019531, 6.770940780639648, 100.24653625488281, 18.926116943359375, 45.76214599609375, 2.90411376953125, 61.10040283203125, -14.194412231445312, 41.622283935546875, 109.8414306640625, 17.2744140625, 75.27153778076172, -1.1800384521484375, 39.664642333984375, 49.23240661621094, -1.362060546875, 101.0045166015625, -51.3697509765625, 53.90480041503906, 98.38943481445312, 106.81039428710938, -133.3062744140625, -1.7361297607421875, 4.917816162109375, 104.1644287109375, 41.72007751464844, 120.39630126953125, 107.06991577148438, 83.10023498535156, 161.93096923828125, 22.396652221679688, 80.74376678466797, 68.16500854492188, 120.27600860595703, 40.57371520996094, 87.95210266113281, -23.843379974365234, -7.001869201660156, 33.104034423828125, -6.267333984375, 141.6607666015625, 79.524169921875, 131.5072021484375, -52.38026428222656, -13.359046936035156, 60.4022216796875, 52.46051025390625, -32.4534912109375, 99.44519805908203, -25.254234313964844, -15.913787841796875, 55.445648193359375, -7.53387451171875, -48.717010498046875, -8.32464599609375, 103.05255126953125, 19.157424926757812, 6.5543060302734375, 99.61160278320312, -10.139068603515625, -2.9428863525390625, 31.449737548828125, -20.033103942871094, 105.91802978515625, 46.69898223876953, 6.3265533447265625, 124.22576904296875, 30.713592529296875, -121.15216064453125, 125.7430419921875, 22.884132385253906, 74.74652099609375, -116.33134460449219, -20.809326171875, -3.775146484375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000449.npy"}
|
|
{"epoch": 0.9403141361256544, "step": 450, "batch_size": 128, "mean": 37.319732666015625, "std": 59.8698844909668, "min": -119.85946655273438, "p10": -33.86149444580077, "median": 30.35187339782715, "p90": 118.9798225402832, "max": 159.91744995117188, "pos_frac": 0.71875, "sample": [0.0, -66.76025390625, 88.70964050292969, -15.453651428222656, 70.049560546875, 5.961280822753906, 107.91218566894531, -7.4185791015625, 129.8106689453125, 43.79779052734375, 84.87017822265625, 139.69888305664062, -0.7272796630859375, 32.78955078125, 107.312744140625, 9.433883666992188, -6.453125, 83.74554443359375, 11.72601318359375, 21.79897117614746, -37.44633483886719, 43.32295227050781, 143.79559326171875, 93.45997619628906, 84.56583404541016, 60.182373046875, -13.248252868652344, 19.47491455078125, 65.6861343383789, 23.764007568359375, -3.8088512420654297, -7.6530914306640625, 27.601409912109375, 8.093399047851562, 46.78265380859375, 12.263427734375, 159.91744995117188, -96.18972778320312, 47.46600341796875, -45.583099365234375, -2.780813217163086, 99.67251586914062, 29.98128890991211, 89.2254638671875, 19.328659057617188, 21.911033630371094, -25.349214553833008, 69.97518920898438, 140.81378173828125, 63.485382080078125, 6.6839752197265625, 115.92024230957031, 15.273548126220703, 8.629737854003906, 44.964996337890625, -14.837173461914062, 33.37199401855469, 138.87120056152344, 12.508758544921875, -81.73187255859375, 45.50616455078125, -119.85946655273438, 102.19439697265625, 37.177398681640625, 9.48095703125, 135.50106811523438, 110.82527160644531, 3.1953277587890625, 80.1212158203125, 74.13241577148438, -4.11151123046875, 85.33258056640625, -2.013824462890625, 119.72003173828125, 1.9233245849609375, 28.47174835205078, 111.92803955078125, 116.22613525390625, -74.43077087402344, 14.1746826171875, 79.84359741210938, 39.60310363769531, -45.74183654785156, 12.2615966796875, 138.6163330078125, 31.21923828125, -1.8369560241699219, 16.8548583984375, 26.357986450195312, -11.024269104003906, 31.89892578125, 62.30805587768555, 124.411376953125, 153.43582153320312, 21.79962158203125, 104.31442260742188, -27.105987548828125, -73.29299926757812, -32.32513427734375, -41.69970703125, 13.73532485961914, 73.114990234375, 113.33615112304688, 106.67936706542969, 141.24795532226562, 38.276397705078125, -76.30845642089844, 32.49781799316406, 8.0943603515625, 119.37567138671875, 111.33183288574219, 32.235595703125, 30.722457885742188, -13.577362060546875, 118.81017303466797, -70.52210998535156, 7.64019775390625, -6.90673828125, 36.169891357421875, 76.76266479492188, 0.0, -10.244010925292969, 73.54965209960938, 53.46417999267578, -8.94244384765625, 109.5765380859375, -18.75341796875, -59.00342559814453], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000450.npy"}
|
|
{"epoch": 0.9424083769633508, "step": 451, "batch_size": 128, "mean": 41.29623031616211, "std": 62.20350646972656, "min": -123.19468688964844, "p10": -27.096408081054687, "median": 30.696640014648438, "p90": 125.69543151855468, "max": 171.8900146484375, "pos_frac": 0.7734375, "sample": [72.83795166015625, 35.91753387451172, -18.670928955078125, 7.160457611083984, 148.5065155029297, 112.68258666992188, 35.770751953125, 2.047687530517578, -120.49032592773438, -23.56781005859375, 91.94839477539062, 3.179595947265625, 138.73214721679688, 22.34716796875, 31.6029052734375, 76.6094970703125, 91.89654541015625, 11.042694091796875, 136.87271118164062, -28.339080810546875, 103.79425048828125, 4.2115478515625, 10.360458374023438, 21.968643188476562, 153.9276123046875, 1.91650390625, 171.85977172851562, -18.986724853515625, -5.665191650390625, 29.467010498046875, -37.926513671875, 30.728492736816406, -123.19468688964844, 130.67117309570312, 25.861968994140625, 31.373374938964844, 16.638137817382812, 63.430267333984375, 34.879852294921875, 97.01692962646484, 25.116870880126953, 106.677490234375, 92.66534423828125, 108.03378295898438, 87.34488677978516, 121.38478088378906, 4.83984375, 13.927749633789062, 145.718994140625, -4.53619384765625, 4.34490966796875, 145.972412109375, 10.595611572265625, 157.53109741210938, 88.13531494140625, 108.91732788085938, -21.410781860351562, 52.66204833984375, -92.96478271484375, 92.73551940917969, 69.41844177246094, 94.34548950195312, 13.588638305664062, 123.91891479492188, 2.0043487548828125, 30.66478729248047, -81.32295989990234, 0.9691009521484375, 50.03973388671875, 77.48307800292969, 18.742874145507812, 6.5531005859375, -26.788360595703125, 114.63577270507812, 11.82183837890625, -43.8170166015625, 33.59611511230469, 7.952789306640625, 104.94308471679688, 45.16778564453125, 2.16790771484375, -3.556793212890625, -19.77520751953125, 68.88230895996094, 117.08248901367188, -27.815185546875, 14.199111938476562, 73.7862548828125, 19.867294311523438, 12.772422790527344, 87.44851684570312, 83.37317657470703, -14.95941162109375, 125.60064697265625, 18.051589965820312, -71.47967529296875, 33.36811065673828, -29.135772705078125, -15.265663146972656, 128.20187377929688, 59.443634033203125, 116.85836791992188, 57.417633056640625, 125.91659545898438, -19.3780517578125, -54.30291748046875, -67.8587646484375, -24.184524536132812, 74.49298095703125, 48.6119384765625, 93.16445922851562, 16.040573120117188, 134.00164794921875, 112.1433334350586, -4.460289001464844, 15.732172012329102, 86.60565185546875, 3.88525390625, 47.2413330078125, 171.8900146484375, 29.129741668701172, -2.7159881591796875, -22.913436889648438, 5.69952392578125, 97.39337158203125, 60.663299560546875, -40.563812255859375, 91.12007141113281], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000451.npy"}
|
|
{"epoch": 0.9445026178010472, "step": 452, "batch_size": 128, "mean": 40.284423828125, "std": 60.382511138916016, "min": -113.98211669921875, "p10": -23.92687530517578, "median": 27.0947265625, "p90": 127.68614807128907, "max": 170.35012817382812, "pos_frac": 0.75, "sample": [-53.30902099609375, 106.3890380859375, 35.755615234375, 155.8487548828125, -39.185638427734375, 1.4757614135742188, 10.314682006835938, 132.2163543701172, -11.695816040039062, -26.5623779296875, 127.96896362304688, 3.2955322265625, 8.79644775390625, 8.054008483886719, -1.8555755615234375, 7.42401123046875, 59.130767822265625, 5.044525146484375, -0.888427734375, 12.2625732421875, 0.11591339111328125, -77.52864074707031, 79.45075988769531, 104.18904113769531, 87.86358642578125, 54.913482666015625, 34.931884765625, 20.756881713867188, -0.24542999267578125, -55.20306396484375, -66.10406494140625, 140.68692016601562, -7.40179443359375, 148.0595703125, 70.91815185546875, -17.767242431640625, 131.339599609375, 126.33163452148438, -19.07196044921875, 101.65037536621094, 60.846580505371094, 14.78656005859375, 9.224878311157227, 17.99609375, 100.13958740234375, 10.484346389770508, 34.05506134033203, 88.8216552734375, 65.38278198242188, 15.668205261230469, 137.6346435546875, 33.963623046875, 82.45034790039062, -6.208831787109375, 12.18410873413086, -37.21868896484375, 11.733230590820312, 16.19659423828125, 16.724838256835938, 17.673828125, 10.046974182128906, 37.50054931640625, -9.964996337890625, 74.50115966796875, -50.827423095703125, 108.49330139160156, 82.49887084960938, 120.36505126953125, 51.3941650390625, 12.368560791015625, 9.318313598632812, 127.56494140625, 35.969886779785156, 3.06768798828125, 74.47109985351562, 170.35012817382812, 108.70045471191406, -16.082626342773438, 24.038818359375, 82.60768127441406, -103.66442108154297, 79.43026733398438, 134.88885498046875, -0.7867393493652344, 129.6650848388672, 107.19136047363281, -4.649436950683594, 105.57118225097656, 118.80166625976562, 30.150634765625, 50.93544006347656, 39.6011962890625, -0.846038818359375, -0.5992584228515625, 23.743141174316406, -39.54695129394531, 109.79020690917969, 32.93049621582031, 52.055389404296875, -10.0989990234375, 161.86221313476562, 45.888824462890625, 17.056365966796875, 102.27178955078125, 48.62474060058594, 4.56171989440918, -23.72271728515625, 13.902000427246094, 120.856201171875, 117.88423156738281, 166.3636474609375, 132.03448486328125, -8.362625122070312, 32.82806396484375, 119.73104858398438, 0.0, 49.14164733886719, -2.1196517944335938, -74.20947265625, -24.403244018554688, 30.68536376953125, 55.089447021484375, -113.98211669921875, 34.2109375, 4.298004150390625, 124.97621154785156, 9.02294921875, 0.06502723693847656], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000452.npy"}
|
|
{"epoch": 0.9465968586387434, "step": 453, "batch_size": 128, "mean": 43.77484130859375, "std": 60.057586669921875, "min": -106.36248779296875, "p10": -35.768969726562496, "median": 35.0722541809082, "p90": 120.97967834472657, "max": 190.9234619140625, "pos_frac": 0.7890625, "sample": [23.90918731689453, 101.64456939697266, -49.61334228515625, 74.61502075195312, 122.559326171875, 4.6777191162109375, 19.3450927734375, 4.865333557128906, 107.63232421875, 68.2213134765625, 37.38323211669922, 64.24922180175781, 13.782440185546875, 62.10382080078125, -46.92071533203125, -35.164794921875, 42.58880615234375, -94.57562255859375, 25.5406494140625, 121.4820556640625, -21.541748046875, 115.44741821289062, 115.19869995117188, 103.39970397949219, 46.1419677734375, 9.21099853515625, -37.1787109375, 116.9989013671875, 47.31239318847656, 115.92855834960938, 16.124740600585938, 95.99969482421875, 35.64338684082031, 98.30766296386719, 24.92474365234375, 33.6396484375, 34.26995849609375, -5.4074859619140625, 125.30455017089844, 133.2144775390625, 7.2218017578125, 127.48382568359375, 29.268798828125, 38.980377197265625, 89.91301727294922, -71.78826904296875, 13.917236328125, 27.911972045898438, 8.8876953125, -41.4068603515625, 11.058441162109375, 75.3194580078125, 5.6033477783203125, 4.9299163818359375, -106.36248779296875, 34.002227783203125, -0.868408203125, -41.025909423828125, 40.48545837402344, 23.2783203125, 88.51211547851562, -8.684043884277344, 0.0, 89.58633422851562, 63.10282897949219, 24.669036865234375, 84.70851135253906, 9.609283447265625, 98.74458312988281, 32.055816650390625, 64.84451293945312, 51.6656494140625, 43.333404541015625, -99.20449829101562, 13.070281982421875, 117.19947052001953, 142.49261474609375, 119.67877197265625, -2.4340972900390625, 107.57562255859375, -12.970550537109375, 138.77886962890625, 99.02606201171875, -6.526214599609375, 13.599884033203125, 19.059814453125, 142.75665283203125, -56.2740478515625, -31.242401123046875, 137.4205322265625, 40.70904541015625, 101.34304809570312, -43.014617919921875, -31.1744384765625, 120.76437377929688, 14.2027587890625, -79.83465576171875, 114.91487121582031, -6.6524200439453125, 190.9234619140625, 54.46952819824219, 47.8157958984375, 33.269630432128906, 60.060821533203125, 12.930511474609375, 50.87841796875, 0.876708984375, 2.4287338256835938, 120.11563110351562, 34.501121520996094, 115.03514099121094, 95.09297180175781, 24.940185546875, 95.77224731445312, 40.384063720703125, 88.82615661621094, 128.47274780273438, 136.79449462890625, -7.5770111083984375, 7.040061950683594, 66.31982421875, 8.433486938476562, 89.24728393554688, -9.048675537109375, 99.28947448730469, -40.09906005859375, 29.626625061035156, 162.87149047851562], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000453.npy"}
|
|
{"epoch": 0.9486910994764398, "step": 454, "batch_size": 128, "mean": 40.46157455444336, "std": 61.08564758300781, "min": -159.33694458007812, "p10": -20.415739440917964, "median": 22.70474338531494, "p90": 128.52785034179686, "max": 180.08737182617188, "pos_frac": 0.8046875, "sample": [45.65582275390625, 113.87294006347656, -100.64071655273438, -12.018905639648438, -4.531463623046875, 9.89068603515625, 134.57693481445312, 99.907958984375, 6.96636962890625, 29.324172973632812, 29.969093322753906, 45.964271545410156, -14.560457229614258, 90.34014129638672, 36.496864318847656, 82.7715835571289, -81.18396759033203, 5.806705474853516, 18.107315063476562, -2.4032421112060547, 35.55377197265625, 12.28564453125, 14.2960205078125, 3.169342041015625, 167.49603271484375, 180.08737182617188, -8.25994873046875, 137.74514770507812, -159.33694458007812, 4.554779052734375, 18.45061492919922, 106.81248474121094, 46.10381317138672, 93.5464096069336, 112.81227111816406, 19.19115447998047, 126.37600708007812, -19.443069458007812, 14.72247314453125, 14.205886840820312, 68.75634765625, 11.585311889648438, 130.77276611328125, 91.56234741210938, 21.46868133544922, 109.1505126953125, 94.40652465820312, 1.93017578125, 136.00570678710938, -10.71234130859375, 23.467803955078125, 35.96124267578125, 24.960662841796875, 2.155120849609375, 78.44683837890625, -16.461029052734375, 71.88449096679688, 7.7017669677734375, 119.52816772460938, 140.4056396484375, 89.42904663085938, 106.54609680175781, 133.0523681640625, 65.59439086914062, 130.36163330078125, 122.97674560546875, 115.48207092285156, 14.2266845703125, 26.431549072265625, 4.1109619140625, 128.15310668945312, 106.09292602539062, 7.33563232421875, -45.77176284790039, 9.555450439453125, 25.737167358398438, -25.473976135253906, 57.601531982421875, 16.04681396484375, -93.098388671875, -39.47315979003906, 1.6446151733398438, 131.07772827148438, -38.14227294921875, -18.552703857421875, 36.67193603515625, 110.67561340332031, 50.72625732421875, -22.685302734375, 120.48529052734375, 12.623397827148438, 10.313545227050781, 17.41741943359375, 132.5377197265625, 152.14907836914062, -1.1019287109375, 17.59416961669922, 8.978866577148438, 124.99888610839844, 18.809646606445312, -30.147689819335938, 12.15399169921875, 129.40225219726562, 21.941682815551758, 33.75048828125, 19.110748291015625, 105.91409301757812, 38.564849853515625, -62.218109130859375, 98.11152648925781, 4.3553009033203125, 38.208465576171875, 62.23187255859375, -1.1902313232421875, 10.599395751953125, 103.31288146972656, 49.513671875, 1.7252655029296875, -57.18511962890625, 36.55458068847656, 14.9637451171875, -7.1011962890625, 17.403411865234375, 120.19815063476562, 32.7618408203125, 3.630767822265625, 15.242095947265625, -51.52413558959961], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000454.npy"}
|
|
{"epoch": 0.9507853403141361, "step": 455, "batch_size": 128, "mean": 33.728511810302734, "std": 60.37581253051758, "min": -125.51254272460938, "p10": -29.52757415771484, "median": 26.97484588623047, "p90": 117.35160217285156, "max": 173.43634033203125, "pos_frac": 0.703125, "sample": [99.13705444335938, 55.932395935058594, 24.679733276367188, -17.795318603515625, 80.980224609375, 24.462005615234375, 89.37997436523438, 124.98362731933594, 26.798263549804688, -30.988807678222656, 4.568511962890625, 80.20559692382812, -4.057098388671875, 66.77240753173828, 42.54956817626953, 117.45063781738281, 126.68775939941406, 88.0108642578125, 162.37771606445312, -49.53277587890625, 7.49517822265625, 35.3233642578125, 18.083480834960938, 117.7630615234375, 173.43634033203125, 114.61920166015625, 113.13622283935547, 17.49273681640625, 30.473812103271484, 1.8380889892578125, -12.870635986328125, -125.51254272460938, -0.8538742065429688, 18.5623779296875, 17.791458129882812, 11.920318603515625, -12.744575500488281, -30.739425659179688, -67.1236801147461, 23.13140869140625, 23.94342041015625, -7.884765625, 88.2070083618164, 158.43569946289062, 13.1922607421875, 75.63385009765625, 95.90777587890625, -7.3270721435546875, 120.19232177734375, 25.860595703125, 7.810569763183594, 8.046630859375, 60.23529052734375, -6.352241516113281, -8.961898803710938, -5.92572021484375, 11.322784423828125, -75.34542846679688, -51.79197692871094, -0.6875514984130859, 18.42528533935547, -21.392196655273438, 7.199043273925781, -29.008209228515625, -28.440582275390625, -25.230417251586914, 76.33920288085938, 51.62626647949219, 96.58372497558594, 51.81999206542969, -115.4249267578125, 82.15869903564453, 86.52783203125, 113.74130249023438, 42.2852783203125, -10.722824096679688, -11.614013671875, 121.08648681640625, 112.68673706054688, 57.40034484863281, 27.15142822265625, -13.306709289550781, 7.8019256591796875, -66.74160766601562, 117.30915832519531, -22.809921264648438, -104.45664978027344, 46.35753631591797, -5.51129150390625, 16.871395111083984, -14.117767333984375, 63.89617919921875, 20.26263427734375, 29.933013916015625, 33.3323974609375, 49.71148681640625, 152.19244384765625, 158.6129150390625, 40.759613037109375, -27.104278564453125, -11.959808349609375, 5.834098815917969, 40.427978515625, 44.8565673828125, 7.8925933837890625, -1.6922245025634766, 6.548027038574219, 62.404754638671875, 32.789306640625, 56.646202087402344, -53.33729553222656, 81.2666015625, 51.84814453125, 59.9752197265625, 103.10101318359375, -49.61262893676758, -123.12371826171875, 48.93731689453125, -20.404266357421875, 105.44448852539062, 100.16122436523438, 51.404388427734375, 134.26693725585938, 45.92620849609375, 56.84075927734375, 51.844390869140625, 124.3302001953125, 30.105865478515625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000455.npy"}
|
|
{"epoch": 0.9528795811518325, "step": 456, "batch_size": 128, "mean": 48.97715759277344, "std": 68.83734893798828, "min": -123.1712646484375, "p10": -27.99955978393554, "median": 33.412567138671875, "p90": 136.02459564208985, "max": 227.14276123046875, "pos_frac": 0.7421875, "sample": [29.557220458984375, -33.81846237182617, -45.887786865234375, 99.69969177246094, 39.170509338378906, -17.67333221435547, 113.54310607910156, -57.736846923828125, 142.02392578125, 81.94137573242188, 128.67791748046875, -123.1712646484375, 17.635391235351562, -1.9866943359375, -22.29840087890625, 139.63037109375, 91.06379699707031, -36.06721496582031, 3.179443359375, 14.063232421875, 127.62826538085938, 65.6378173828125, -7.08453369140625, 121.04544067382812, 0.8786811828613281, -18.719970703125, 57.3948974609375, 18.616302490234375, 112.13606262207031, -32.455474853515625, 104.36318969726562, -86.30902099609375, 102.92611694335938, -26.089881896972656, 41.52093505859375, 132.86712646484375, 74.30732727050781, 105.7054214477539, 78.5758056640625, 86.45587158203125, 168.62103271484375, -20.473236083984375, 133.1619873046875, 21.340988159179688, 137.56939697265625, 100.32931518554688, 148.11358642578125, -3.805908203125, 9.06671142578125, 116.1607666015625, 34.457061767578125, 67.42074584960938, -3.066925048828125, 30.871326446533203, -44.8851318359375, 19.579666137695312, 6.804656982421875, 20.093482971191406, 11.9420166015625, 159.1246337890625, 1.51434326171875, -6.605140686035156, 16.72205352783203, -81.34540557861328, 2.08837890625, 115.49600219726562, 109.3428955078125, -103.19227600097656, 28.53515625, 18.655792236328125, 11.715574264526367, 82.1297607421875, -3.2962913513183594, 96.5382080078125, 32.368072509765625, 90.9737548828125, 159.72817993164062, 93.84869384765625, -12.391815185546875, 50.202789306640625, 104.67517852783203, 74.9112548828125, -59.67047119140625, 135.72808837890625, -1.1426849365234375, -16.88226318359375, 23.646408081054688, 130.57005310058594, 118.77877807617188, 143.00572204589844, 99.10342407226562, 149.87127685546875, -5.865943908691406, 127.25253295898438, 0.0, 3.401744842529297, 89.86707305908203, 130.49636840820312, 92.5835952758789, 90.20370483398438, 16.55828857421875, 13.204620361328125, 46.19354248046875, 1.897125244140625, 49.15660858154297, -42.0093994140625, 101.89398193359375, 2.659759521484375, 8.63885498046875, -13.26837158203125, 5.684478759765625, 60.830322265625, 45.281341552734375, 91.05108642578125, 136.71644592285156, 0.03668975830078125, 207.7568359375, 15.848342895507812, 227.14276123046875, 155.134765625, 119.7076416015625, 31.067108154296875, -20.18231201171875, -8.16845703125, 121.82267761230469, -22.323989868164062, -85.49856567382812, 135.30889892578125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000456.npy"}
|
|
{"epoch": 0.9549738219895288, "step": 457, "batch_size": 128, "mean": 46.16623306274414, "std": 65.70210266113281, "min": -144.36175537109375, "p10": -34.55280761718749, "median": 38.50090026855469, "p90": 129.10523376464843, "max": 156.72044372558594, "pos_frac": 0.75, "sample": [136.61448669433594, 15.4136962890625, 2.6167068481445312, -3.715240478515625, 102.84552001953125, 26.700927734375, 117.27215576171875, 123.48004150390625, 144.92059326171875, 13.778472900390625, 118.1839370727539, -0.4821434020996094, 140.775146484375, 82.80984497070312, 4.907470703125, 148.99114990234375, 5.467376708984375, 25.0772705078125, -83.42382049560547, 135.05245971679688, 104.56867980957031, 113.45791625976562, 83.14042663574219, 156.72044372558594, 101.00762939453125, 55.056671142578125, 127.57321166992188, 6.1342926025390625, 124.42314910888672, 13.157180786132812, 83.09698486328125, -10.58221435546875, -4.325927734375, -25.397659301757812, 107.66830444335938, 119.08067321777344, -7.840675354003906, 37.253875732421875, -73.60536193847656, 82.91268920898438, 3.1715087890625, 116.2275161743164, -19.676010131835938, 10.94537353515625, 4.262050628662109, -144.36175537109375, 139.7015380859375, 28.32818603515625, 141.14422607421875, -11.137176513671875, -32.267303466796875, 57.5975341796875, 77.36289978027344, -54.194580078125, 54.40998840332031, 18.384672164916992, -75.2545166015625, 89.63775634765625, 40.380828857421875, 24.1318359375, -25.3697509765625, -13.707969665527344, 15.259490966796875, -24.004600524902344, 39.7479248046875, 92.11170959472656, 95.7688980102539, 20.118186950683594, 146.0293426513672, -6.102203369140625, -0.9518280029296875, 109.03759765625, 103.31744384765625, 34.92608642578125, 20.824417114257812, -97.77073669433594, -39.885650634765625, 33.13079833984375, 25.74493408203125, 75.06028747558594, 19.9527587890625, 122.67892456054688, 87.63493347167969, 72.50537109375, 56.471160888671875, -15.03851318359375, 72.7491226196289, 1.7792816162109375, 91.88626098632812, 147.91107177734375, 7.30402946472168, -43.668243408203125, 125.39697265625, -55.89044189453125, -40.96624755859375, 119.72281646728516, -49.7762451171875, 94.05003356933594, -17.045318603515625, 127.47503662109375, 101.78224182128906, 43.64630126953125, 7.7416534423828125, 45.72064208984375, -70.29736328125, 135.69989013671875, 29.636383056640625, 86.92485046386719, 8.35418701171875, 89.8778076171875, 131.07186889648438, 80.59339141845703, 12.57525634765625, 123.58558654785156, 109.56649780273438, 102.69068145751953, 4.535869598388672, 1.8437519073486328, -91.57501220703125, 91.18803405761719, 143.04608154296875, 0.8751049041748047, 0.0, 128.26239013671875, -9.9122314453125, 105.99301147460938, 48.20953369140625, -0.35718536376953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000457.npy"}
|
|
{"epoch": 0.9570680628272251, "step": 458, "batch_size": 128, "mean": 40.97951126098633, "std": 63.69767761230469, "min": -126.11370849609375, "p10": -26.27118988037109, "median": 21.191543579101562, "p90": 134.03260192871093, "max": 174.4048614501953, "pos_frac": 0.7578125, "sample": [-1.9996261596679688, -42.798095703125, 12.089523315429688, -115.43028259277344, 148.0847930908203, 14.258777618408203, -35.6627197265625, 133.4561767578125, -2.466827392578125, -29.08880615234375, 15.032958984375, -47.99320602416992, -41.4415283203125, 0.7343101501464844, 1.6854219436645508, 74.24897766113281, 33.27166748046875, 38.11737060546875, -21.18072509765625, 55.523834228515625, 135.84579467773438, -1.2768115997314453, 89.8720703125, 0.087860107421875, 23.208236694335938, 4.81181526184082, 114.6568603515625, 135.37759399414062, -12.0621337890625, 102.13763427734375, 27.4569091796875, -22.89885711669922, 95.25471496582031, 162.03988647460938, 55.4713134765625, 15.6708984375, 21.9971923828125, 31.485565185546875, 132.43844604492188, -25.525680541992188, 108.72787475585938, -2.731660842895508, -67.64657592773438, 28.201171875, 7.79620361328125, 5.7932281494140625, 19.395751953125, 12.286762237548828, 12.69913101196289, 13.969902038574219, 4.225322723388672, -20.217086791992188, -11.55364990234375, 123.41055297851562, 107.707275390625, -30.707244873046875, -6.18896484375, -2.3605270385742188, 90.11766052246094, 158.21127319335938, 20.54412841796875, 2.1388654708862305, 142.9940185546875, 74.61868286132812, 78.69660949707031, 2.448394775390625, 8.637420654296875, -72.01982879638672, -28.010711669921875, 145.10513305664062, 112.10986328125, 65.95147705078125, 9.279754638671875, 23.98883056640625, -29.52789306640625, 112.40642547607422, 75.67550659179688, -15.126129150390625, 6.596595764160156, 94.65599060058594, 106.17635345458984, -20.84710693359375, -5.444099426269531, 27.878021240234375, 155.04461669921875, 20.83697509765625, 174.4048614501953, 7.26593017578125, 42.6055908203125, 112.6805419921875, 21.546112060546875, 147.40536499023438, 104.70330810546875, 97.27363586425781, 159.76968383789062, -2.5548171997070312, 100.70819091796875, -4.327880859375, 7.148134231567383, 88.58441162109375, 42.10882568359375, 12.00592041015625, 7.610565185546875, 119.90339660644531, 81.798583984375, 43.20465850830078, 7.449394226074219, 16.686477661132812, 118.15068054199219, 108.10093688964844, 41.7705078125, 148.15040588378906, 114.39268493652344, 63.14105224609375, 89.1009521484375, 17.984375, 94.26007080078125, 2.5694427490234375, -126.11370849609375, -118.86727905273438, 98.24520874023438, 140.73828125, 9.946208953857422, 3.1466064453125, 94.64636993408203, 30.183502197265625, 5.814537048339844, -18.399978637695312], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000458.npy"}
|
|
{"epoch": 0.9591623036649215, "step": 459, "batch_size": 128, "mean": 31.661806106567383, "std": 66.34986114501953, "min": -129.1346435546875, "p10": -42.78375854492187, "median": 18.084976196289062, "p90": 123.63932189941406, "max": 172.993896484375, "pos_frac": 0.671875, "sample": [8.066818237304688, -30.295944213867188, 5.109952926635742, 105.38778686523438, 6.97320556640625, 95.98928833007812, -101.65290832519531, 36.35467529296875, 60.33509826660156, 28.456588745117188, 72.306884765625, 77.6455078125, 28.997020721435547, -14.78472900390625, 7.098213195800781, 7.050495147705078, 136.80592346191406, 96.83145141601562, -14.879308700561523, 32.7994384765625, 59.96923828125, 13.58209228515625, -6.409576416015625, 27.0869140625, -22.928009033203125, -8.8636474609375, 98.82827758789062, -2.8927078247070312, 107.2016830444336, -17.84449005126953, -16.61308479309082, 0.0, 16.52783203125, -51.41046142578125, 108.25747680664062, -75.76910400390625, 47.88134002685547, -37.548343658447266, -11.07830810546875, 52.531219482421875, 111.51513671875, -39.35296630859375, 16.45166015625, 62.104827880859375, -107.25810241699219, 74.14611053466797, -17.096710205078125, 21.3819580078125, -16.879718780517578, 138.81866455078125, 5.847381591796875, 6.964239120483398, -2.7204647064208984, 163.13113403320312, 34.393531799316406, 13.351646423339844, 68.89665222167969, -103.43179321289062, 121.10054016113281, 172.993896484375, 78.75991821289062, -25.226028442382812, 125.23712158203125, 91.70356750488281, 110.0860824584961, 139.56301879882812, 68.55386352539062, 104.83992004394531, -67.4644775390625, -80.22659301757812, 8.9276123046875, 41.5152587890625, 102.36299133300781, 116.36166381835938, -12.252182006835938, -123.13973999023438, 141.15582275390625, 56.95631408691406, 10.624481201171875, -50.7889404296875, -129.1346435546875, -9.1607666015625, 4.9154052734375, -107.86685180664062, 124.763427734375, 83.30224609375, -12.9259033203125, 104.3779296875, 0.0, 13.990524291992188, 108.87886047363281, 2.5355224609375, 19.642120361328125, 154.68670654296875, -68.82373046875, 126.19033813476562, -5.422576904296875, 102.40711212158203, 16.084091186523438, 53.49115753173828, -55.156646728515625, 69.06729125976562, 16.1065673828125, -28.645339965820312, -9.838623046875, 33.716835021972656, 87.7723388671875, 20.686676025390625, 10.736541748046875, -12.826324462890625, -37.28460693359375, -8.992568969726562, 20.499908447265625, -34.558868408203125, 116.69523620605469, 142.174072265625, 2.74395751953125, 14.7078857421875, 123.15756225585938, 27.6767578125, -5.032783508300781, 143.71322631835938, 149.7679443359375, 26.48462677001953, 31.0185546875, 8.00946044921875, 42.428985595703125, 88.94052124023438], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000459.npy"}
|
|
{"epoch": 0.9612565445026178, "step": 460, "batch_size": 128, "mean": 32.531761169433594, "std": 68.05662536621094, "min": -144.517822265625, "p10": -57.92813262939452, "median": 21.825942993164062, "p90": 119.15548248291016, "max": 178.48699951171875, "pos_frac": 0.7265625, "sample": [15.854705810546875, 56.525543212890625, -89.36541748046875, 107.12802124023438, 43.62274169921875, 116.2623291015625, -84.76513671875, 98.40594482421875, -42.40557861328125, 133.04180908203125, -89.1314697265625, 12.990814208984375, 19.27294921875, 9.295364379882812, 40.39146423339844, -4.515777587890625, -22.032257080078125, 10.3642578125, 118.49916076660156, 131.70635986328125, 19.71331787109375, 23.537841796875, 24.0413818359375, 66.34994506835938, 47.289886474609375, 16.17345428466797, 85.2296142578125, -4.1905059814453125, 154.07257080078125, -19.48114013671875, -10.495849609375, 61.00001525878906, 36.1263427734375, 178.48699951171875, 76.75601196289062, -43.15907287597656, 92.98811340332031, -96.88592529296875, 163.84942626953125, -15.783477783203125, -132.4623260498047, 130.78408813476562, -69.71585083007812, 17.155414581298828, 58.376800537109375, 111.25814819335938, -34.60345458984375, 101.92402648925781, 13.200958251953125, -52.06709289550781, 68.80654907226562, 0.4669609069824219, 2.0999984741210938, 2.3827896118164062, -33.97423553466797, 11.542037963867188, 6.368064880371094, -5.9241943359375, -6.7137908935546875, 9.598907470703125, 114.11376953125, 57.892791748046875, 97.73809051513672, 20.114044189453125, 30.051513671875, -74.55059814453125, 3.541318893432617, 36.88056945800781, 109.60311889648438, -0.12979507446289062, -6.7897491455078125, 33.9613037109375, 18.0313720703125, 13.20855712890625, 97.593505859375, 84.09736633300781, 111.42050170898438, 105.99444580078125, 32.663970947265625, 25.93880844116211, 137.40399169921875, 10.025161743164062, -49.637786865234375, 172.4423828125, 41.45306396484375, 163.54794311523438, 119.26666259765625, 116.669921875, -73.27407836914062, 127.98616027832031, 76.28955078125, 67.25985717773438, -69.54931640625, 7.572288513183594, -33.6446533203125, 16.512176513671875, 114.60678100585938, -144.517822265625, 85.44769287109375, -41.4510498046875, 42.761268615722656, -92.90509033203125, 113.7247314453125, 47.98271179199219, 15.222724914550781, -65.68646240234375, -13.716552734375, -54.60313415527344, 119.10783386230469, 33.885162353515625, 26.35693359375, 100.81381225585938, 31.98388671875, 120.54429626464844, -82.7943115234375, 1.253499984741211, 5.149799346923828, 100.1014404296875, 31.378173828125, 143.2422637939453, 102.3236083984375, 13.116706848144531, 32.400238037109375, -26.069095611572266, -9.993499755859375, 3.5158023834228516, 19.67938232421875, 14.236923217773438], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000460.npy"}
|
|
{"epoch": 0.9633507853403142, "step": 461, "batch_size": 128, "mean": 39.557273864746094, "std": 66.738525390625, "min": -124.658203125, "p10": -33.55711212158203, "median": 28.331634521484375, "p90": 131.47277221679687, "max": 162.40057373046875, "pos_frac": 0.7109375, "sample": [17.890342712402344, -98.779052734375, 139.86279296875, 16.6195068359375, 27.76397705078125, 65.85736083984375, 112.46443939208984, 41.07110595703125, -18.391551971435547, 141.74850463867188, 69.2237548828125, 0.0, 10.890228271484375, 162.40057373046875, 67.89405822753906, 95.2605972290039, 96.3816909790039, 42.280487060546875, -7.14495849609375, 9.710853576660156, 44.49082946777344, 132.382080078125, 92.94416046142578, -56.95941162109375, 76.13299560546875, -2.5997772216796875, 13.2362060546875, 143.93707275390625, -31.039901733398438, 111.7255630493164, -3.1864662170410156, 153.16372680664062, 50.1180419921875, 109.71875, -19.796737670898438, 114.69294738769531, 143.50482177734375, 15.0570068359375, -93.22894287109375, -30.908294677734375, 134.84353637695312, -102.36001586914062, 4.3730926513671875, 18.120742797851562, 58.420318603515625, 3.4268951416015625, 86.17939758300781, 87.68521118164062, 136.7715301513672, 3.0111083984375, 93.08680725097656, -33.43986511230469, 24.25341796875, -2.1775970458984375, 10.64080810546875, -14.970657348632812, 104.0966796875, -124.658203125, -50.262176513671875, 48.129730224609375, 89.66770935058594, 3.215667724609375, -11.79156494140625, 76.55473327636719, 131.08306884765625, 20.848785400390625, -121.66172790527344, 9.91583251953125, 129.02857971191406, 5.0512237548828125, 21.307937622070312, 119.23832702636719, -8.067214965820312, 123.7032470703125, -10.004547119140625, 146.09344482421875, 74.74932861328125, 92.99554443359375, -48.563140869140625, -29.92694091796875, 119.01402282714844, 3.63818359375, 97.55551147460938, 72.35745239257812, 115.75070190429688, 12.593971252441406, -17.153289794921875, -71.42100524902344, -8.111551284790039, 52.492034912109375, 63.916168212890625, 83.01820373535156, 10.57562255859375, 100.8675537109375, 3.78631591796875, 141.0453643798828, -2.55035400390625, 29.207305908203125, 113.2196044921875, -81.18305969238281, 67.62102508544922, 1.8332481384277344, 138.5843048095703, 17.760528564453125, -33.8306884765625, -11.722930908203125, 54.456939697265625, 112.64176940917969, 116.41261291503906, -6.124153137207031, -31.74219512939453, 67.82880401611328, -4.958797454833984, 80.91814422607422, 12.17987060546875, 1.6881484985351562, 71.00747680664062, 73.43467712402344, -103.6635513305664, -18.049461364746094, 33.936737060546875, 13.866798400878906, -58.399200439453125, 138.69879150390625, -12.52419662475586, 28.8992919921875, 88.32650756835938, 130.63314819335938], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000461.npy"}
|
|
{"epoch": 0.9654450261780104, "step": 462, "batch_size": 128, "mean": 50.583683013916016, "std": 68.07189178466797, "min": -122.9022216796875, "p10": -28.545180511474605, "median": 56.87080383300781, "p90": 130.300341796875, "max": 180.268310546875, "pos_frac": 0.765625, "sample": [100.84178161621094, 59.54734802246094, 4.820770263671875, 77.9176025390625, 127.62564086914062, 37.473663330078125, 97.91996765136719, 88.48834228515625, 49.97064971923828, 126.42410278320312, -102.80490112304688, 97.08953857421875, 144.498046875, 62.83064270019531, 123.83792114257812, 11.928665161132812, 61.031532287597656, 96.69674682617188, 59.130767822265625, 5.84906005859375, 35.863250732421875, 98.7313232421875, 101.43815612792969, 41.91754150390625, 61.24750900268555, 2.44903564453125, 90.39021301269531, -1.2300052642822266, 115.107666015625, -71.02931213378906, -95.38925170898438, 74.85516357421875, -7.7191619873046875, 48.6822509765625, -10.701019287109375, 119.10267639160156, 129.62255859375, -55.61775207519531, -13.2286376953125, 11.68194580078125, 66.83798217773438, 119.07884979248047, 107.28219604492188, 5.6473236083984375, 172.697265625, 72.74225616455078, 58.68611145019531, 127.54013061523438, 103.71380615234375, 2.751556396484375, 135.77783203125, 5.793067932128906, 89.89990234375, -79.90739440917969, 13.026603698730469, -9.829513549804688, 138.470458984375, 41.396759033203125, 12.030509948730469, 15.55792236328125, 110.31884765625, 162.32611083984375, -2.063934326171875, 126.0460205078125, -12.832611083984375, 71.4773941040039, -76.86341094970703, 54.0396728515625, 172.35418701171875, -11.214828491210938, 115.25762939453125, 127.03055572509766, 107.044677734375, 107.13125610351562, 74.2259521484375, 57.274261474609375, -25.3016357421875, 19.200225830078125, 4.428333282470703, -51.0120849609375, 13.281681060791016, -23.13134765625, -3.625335693359375, 10.732025146484375, -107.984619140625, 47.69537353515625, 83.07491302490234, 172.15396118164062, -67.64268493652344, 83.23222351074219, -55.677146911621094, 149.6551513671875, 52.11854553222656, 37.69049835205078, 41.20820617675781, 180.268310546875, -32.10901641845703, 70.71343994140625, 95.20318603515625, 121.15536499023438, 73.06160736083984, -0.99627685546875, 177.30816650390625, 25.594757080078125, -16.12384033203125, 14.713043212890625, 108.27916717529297, 40.276824951171875, 44.72479248046875, 11.28204345703125, 88.82861328125, 7.204345703125, 56.46734619140625, 131.8818359375, -92.1341552734375, 160.55416870117188, -27.017822265625, 83.33246612548828, 90.84640502929688, -18.071060180664062, 141.7896728515625, 38.10411071777344, -4.28082275390625, -19.70043182373047, 79.34454345703125, 98.07830810546875, 100.9007568359375, -122.9022216796875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000462.npy"}
|
|
{"epoch": 0.9675392670157068, "step": 463, "batch_size": 128, "mean": 37.481258392333984, "std": 61.26836395263672, "min": -117.69635009765625, "p10": -27.37388458251953, "median": 25.153152465820312, "p90": 121.7119659423828, "max": 164.56134033203125, "pos_frac": 0.734375, "sample": [94.40518188476562, 24.838333129882812, 16.48431396484375, 7.411643981933594, 11.27069091796875, -0.6928157806396484, 145.2283935546875, 7.2801513671875, -117.69635009765625, 58.96875, 116.96086883544922, 5.304229736328125, -100.67381286621094, -36.06071472167969, 7.7158355712890625, 123.19320678710938, 57.99632263183594, 54.12200927734375, 15.00372314453125, 4.923482894897461, 17.356788635253906, -79.99005126953125, -18.351287841796875, 6.710655212402344, 68.00111389160156, -2.0611724853515625, -56.05926513671875, 27.659423828125, 119.65982055664062, 131.6263427734375, -109.90684509277344, 78.25904083251953, 13.116363525390625, -30.2281494140625, -10.10986328125, 138.1805419921875, 121.0771484375, 112.64112854003906, 164.56134033203125, -34.72065734863281, 90.28024291992188, -24.7933349609375, 69.54890441894531, 152.3861083984375, 32.68696594238281, 7.90618896484375, 146.182861328125, -9.72775650024414, 23.02679443359375, 1.5306625366210938, 18.016197204589844, 25.467971801757812, 131.20822143554688, 43.96478271484375, 11.333984375, 19.93902587890625, 129.11587524414062, -10.10162353515625, 110.03067016601562, 99.40843963623047, 59.11151123046875, 113.13739013671875, 6.433258056640625, -40.143341064453125, -22.096282958984375, 14.817703247070312, 127.38397216796875, 3.625213623046875, 123.49893188476562, 37.175872802734375, 115.08636474609375, -27.2796630859375, -2.44195556640625, 50.450592041015625, 58.149444580078125, 3.17449951171875, 86.06953430175781, -15.743865966796875, -14.690696716308594, 39.816436767578125, 108.56201934814453, 15.4229736328125, 73.41476440429688, 88.49478149414062, 75.431640625, 5.806295394897461, 36.70875549316406, 104.21968841552734, 46.01487731933594, -11.477436065673828, -17.533599853515625, 92.909423828125, -27.593734741210938, 0.0, -5.49951171875, 4.536762237548828, 7.707386016845703, 144.0372314453125, 31.4102783203125, -13.795677185058594, 61.17352294921875, 82.27064514160156, -11.648513793945312, 0.7104930877685547, 35.98754119873047, -58.175506591796875, 35.94378662109375, 18.206520080566406, 77.17117309570312, -0.5552616119384766, 2.8846893310546875, -52.93878173828125, -4.565643310546875, 62.99518585205078, 3.212158203125, 111.88121032714844, 120.55259704589844, 46.77926254272461, 114.51788330078125, 79.85943603515625, 146.9337158203125, 30.860137939453125, 77.29339599609375, 58.026702880859375, 91.06063842773438, 114.81927490234375, -116.71978759765625, -26.0643310546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000463.npy"}
|
|
{"epoch": 0.9696335078534032, "step": 464, "batch_size": 128, "mean": 39.165565490722656, "std": 62.22639083862305, "min": -156.732666015625, "p10": -30.53175048828125, "median": 29.484819412231445, "p90": 115.65923614501952, "max": 157.6684112548828, "pos_frac": 0.7578125, "sample": [0.0, 77.3515625, 16.66460609436035, 92.765869140625, -7.1644134521484375, 94.44153594970703, 67.3436279296875, -31.0296630859375, 106.74002075195312, -0.269744873046875, 52.768890380859375, 35.91864013671875, 109.40605163574219, 29.935806274414062, 90.77897644042969, 18.67391014099121, 110.9583740234375, 1.011138916015625, 113.01749420166016, 65.58206176757812, 44.54840087890625, 114.79960632324219, 10.555723190307617, 16.234619140625, 50.940032958984375, 85.47348022460938, 36.59691619873047, -113.08489227294922, 157.6684112548828, 20.18450927734375, 6.20166015625, -12.4993896484375, 19.105941772460938, 141.3881072998047, 127.5458984375, 127.60690307617188, 42.6588134765625, -3.8579940795898438, -10.044036865234375, 79.37641906738281, 12.561279296875, 25.4208984375, 111.7498779296875, 70.78892517089844, 85.51641845703125, -0.5961761474609375, -10.39410400390625, 17.793365478515625, 1.2232818603515625, 73.82439422607422, 12.851669311523438, 15.06549072265625, 5.97418212890625, 103.29570007324219, 16.566028594970703, 69.11593627929688, 101.67852783203125, 87.90290832519531, -109.67132568359375, 86.1568603515625, -2.8021240234375, 106.79403686523438, -7.617973327636719, 112.8731689453125, 138.66973876953125, -13.66571044921875, 55.23371124267578, 24.585617065429688, 27.5592041015625, 80.27679443359375, -69.46746826171875, -156.732666015625, -75.65260314941406, 1.8666610717773438, 23.827972412109375, 33.91499328613281, -17.651641845703125, -45.704345703125, 98.09201049804688, 21.183006286621094, 128.8937225341797, 141.64035034179688, 106.78948974609375, 129.99957275390625, -24.257781982421875, 117.6650390625, 1.5399093627929688, -31.2916259765625, 13.3243408203125, 3.5260181427001953, 31.476272583007812, -87.70184326171875, 123.09127807617188, -7.261810302734375, -51.274017333984375, 13.526687622070312, -1.4067764282226562, 102.78421020507812, 5.6971435546875, 8.862504959106445, 86.58443450927734, 131.10789489746094, 38.07658386230469, 126.52342987060547, 120.2515869140625, -30.318359375, 67.77072143554688, 9.347702026367188, 105.11042785644531, -77.57075500488281, 14.970489501953125, 9.536441802978516, -0.15972518920898438, 108.11862182617188, 110.61508178710938, 50.06084060668945, 82.83583068847656, 112.01885986328125, 102.07028198242188, -134.1910400390625, 22.72613525390625, 23.81298828125, -31.525753021240234, 57.68255615234375, 29.033832550048828, -19.22802734375, 42.7027587890625, 30.905853271484375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000464.npy"}
|
|
{"epoch": 0.9717277486910995, "step": 465, "batch_size": 128, "mean": 49.35356903076172, "std": 66.03955078125, "min": -129.25784301757812, "p10": -25.889061737060544, "median": 38.893951416015625, "p90": 138.1471710205078, "max": 166.67318725585938, "pos_frac": 0.7578125, "sample": [-1.50341796875, -7.070037841796875, 9.004714965820312, 15.650955200195312, 124.7227783203125, -12.999052047729492, -24.592132568359375, 143.92800903320312, 11.93280029296875, 23.505966186523438, 3.89276123046875, 17.093017578125, 0.4332427978515625, 11.844696044921875, -86.90243530273438, -31.91668701171875, 137.857177734375, 117.73876953125, 140.638671875, 130.59051513671875, 117.7900390625, -64.71875, 38.29133605957031, 116.05490112304688, 28.940887451171875, 2.6441383361816406, 3.56402587890625, 125.18118286132812, 134.67233276367188, 101.59370422363281, -8.70721435546875, 106.25196838378906, 27.214691162109375, 88.75421142578125, 6.227203369140625, -5.1568603515625, -34.4281005859375, 141.69183349609375, 99.19807434082031, 166.67318725585938, 9.717117309570312, 59.51287841796875, -0.019775390625, -29.29095458984375, -11.639076232910156, 38.46600341796875, -3.2948455810546875, 8.495237350463867, -20.3956298828125, 92.55891418457031, 99.93942260742188, 78.84591674804688, 73.02264404296875, 98.83824157714844, 70.56733703613281, -67.24971771240234, 143.42788696289062, 32.87774658203125, -10.942596435546875, 44.82106018066406, 58.27581787109375, 47.8192138671875, 41.862762451171875, 68.69319152832031, 130.26177978515625, 20.256942749023438, 139.23788452148438, 147.77694702148438, -5.44561767578125, 29.29024314880371, 50.9937744140625, 119.89729309082031, 101.68681335449219, 88.93803405761719, 68.41371154785156, -69.74761962890625, -8.55377197265625, 43.08258056640625, 23.432235717773438, 74.3956298828125, 31.62042236328125, 7.28753662109375, 2.904388427734375, 15.768524169921875, 4.51190185546875, 96.8358383178711, 146.86819458007812, -43.381744384765625, 41.17510223388672, -12.26263427734375, 132.8511962890625, 148.39205932617188, 163.78228759765625, 26.106727600097656, 80.72799682617188, 24.642181396484375, 147.80047607421875, 128.8056640625, 132.10833740234375, 3.0487899780273438, 112.68890380859375, 121.89447021484375, 147.53274536132812, 124.56129455566406, 8.743072509765625, 105.18085479736328, -14.8731689453125, 109.64633178710938, -79.40266418457031, 39.3218994140625, -58.62518310546875, 4.5812530517578125, 129.796142578125, 121.33819580078125, 119.04971313476562, 138.82382202148438, -82.40875244140625, 112.64273834228516, 9.926055908203125, 8.334243774414062, -28.91522979736328, 83.24505615234375, -9.958984375, -14.507415771484375, 119.2183837890625, -129.25784301757812, -6.6451416015625, 51.323974609375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000465.npy"}
|
|
{"epoch": 0.9738219895287958, "step": 466, "batch_size": 128, "mean": 52.361289978027344, "std": 57.8995475769043, "min": -102.98574829101562, "p10": -10.702129745483397, "median": 50.87360763549805, "p90": 128.37342071533203, "max": 173.28463745117188, "pos_frac": 0.8046875, "sample": [-5.882368087768555, 78.43533325195312, 95.58543395996094, 109.97897338867188, 82.728515625, 134.08041381835938, 9.675689697265625, 0.0, 70.87313842773438, 55.94879913330078, 111.59278869628906, -21.376178741455078, 87.17770385742188, 72.99246215820312, 40.6673583984375, 157.5224609375, 45.79841613769531, 1.5537452697753906, 98.95953369140625, 57.29655456542969, 89.76712036132812, 24.779403686523438, 64.69956970214844, -38.3857421875, 17.99688720703125, 43.107696533203125, 72.47695922851562, 23.017486572265625, 96.63042449951172, 71.43775939941406, -1.9865951538085938, 112.57998657226562, 3.1973648071289062, 95.54827880859375, 89.367431640625, 9.304744720458984, 90.11862182617188, -7.0034637451171875, 117.30584716796875, 63.49015808105469, 169.5264434814453, 2.431182861328125, 151.59890747070312, 45.481658935546875, 89.4169921875, 140.07028198242188, 37.664344787597656, -46.59619140625, -87.80659484863281, 45.18499755859375, -1.0577545166015625, 107.42849731445312, 4.121185302734375, 108.47408294677734, 89.61093139648438, 117.40697479248047, -15.057586669921875, -88.89605712890625, 22.660888671875, 113.86279296875, 96.41590881347656, 106.90342712402344, 79.72889709472656, 66.62527465820312, 14.485626220703125, 97.51483154296875, 111.50216674804688, 97.9476318359375, 137.66159057617188, 103.49234008789062, 0.0, 111.88510131835938, 152.16412353515625, -25.251678466796875, -7.336906433105469, 26.540748596191406, 85.81720733642578, -23.999130249023438, 57.9498291015625, 173.28463745117188, 5.174385070800781, 9.02386474609375, 159.272705078125, 42.985015869140625, 24.70343017578125, -83.13363647460938, 19.69378662109375, 40.52680969238281, 22.624267578125, 84.52875518798828, 114.51290893554688, 18.278701782226562, 16.74573516845703, -16.257614135742188, 153.25244140625, 1.9604806900024414, 56.8121337890625, 85.06298828125, -38.54737854003906, -102.98574829101562, 58.08735656738281, 23.81390380859375, -0.966064453125, 129.98985290527344, 20.87834930419922, 0.0, 127.6806640625, 16.59857177734375, -0.31270599365234375, 68.97415161132812, 130.83529663085938, 79.68374633789062, -1.3884544372558594, 106.674560546875, 62.61219787597656, 27.035064697265625, 26.326080322265625, 25.403884887695312, 23.533000946044922, 8.52099609375, 24.306976318359375, 25.183876037597656, 151.59503173828125, 89.32843017578125, -11.511138916015625, -10.355411529541016, 70.399169921875, 25.167678833007812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000466.npy"}
|
|
{"epoch": 0.9759162303664921, "step": 467, "batch_size": 128, "mean": 37.7950439453125, "std": 67.66366577148438, "min": -122.44903564453125, "p10": -48.76395874023437, "median": 22.094154357910156, "p90": 129.2753463745117, "max": 165.76708984375, "pos_frac": 0.6640625, "sample": [129.25567626953125, 96.06086730957031, 30.430038452148438, 74.47392272949219, 40.98932647705078, -5.316856384277344, 27.09515380859375, 64.47992706298828, 87.05572509765625, 4.659210205078125, -122.44903564453125, -112.9722900390625, -10.3822021484375, -9.613296508789062, 17.383880615234375, 101.4671630859375, -80.45867919921875, 140.56259155273438, 2.01934814453125, 75.10224914550781, -17.618682861328125, 93.97177124023438, -17.219970703125, 74.75244140625, 124.65650939941406, 49.82989501953125, 17.701339721679688, 100.98368835449219, 1.5215682983398438, 12.401611328125, 19.094818115234375, 116.28526306152344, 71.258544921875, 115.44842529296875, 135.35009765625, 55.979888916015625, 14.732711791992188, 139.57196044921875, 50.70628356933594, -50.52569580078125, 135.50457763671875, -2.17889404296875, 0.029863357543945312, 125.70596313476562, 45.966156005859375, 100.939453125, -5.804010391235352, -33.326629638671875, 129.77151489257812, -90.05211639404297, 10.506965637207031, -77.92282104492188, -29.064468383789062, 116.84927368164062, -38.861297607421875, 105.0040283203125, -21.385433197021484, -32.389129638671875, 113.67583465576172, 146.2191162109375, 7.713623046875, 15.669595718383789, -0.4528942108154297, 10.702728271484375, 95.65402221679688, 13.09271240234375, -1.0128250122070312, 50.37055969238281, 6.89898681640625, -50.16139602661133, 9.4541015625, -103.23153686523438, 108.07223510742188, 0.0, -1.977691650390625, -3.38482666015625, 63.100433349609375, 144.10899353027344, 3.984039306640625, -0.501556396484375, 0.0, 22.347366333007812, 110.9150390625, -8.343658447265625, 105.8941650390625, 69.42280578613281, 106.08602905273438, -34.78680419921875, 162.61276245117188, -80.7288818359375, -48.744384765625, 145.0899658203125, 89.99346923828125, 84.50680541992188, 127.3287353515625, -1.3041000366210938, -16.215545654296875, 21.8409423828125, -6.448211669921875, 62.350616455078125, -29.002410888671875, 113.31809997558594, 116.3447265625, 33.678863525390625, -57.16566467285156, 21.21551513671875, -45.00822448730469, -48.80963134765625, 10.584943771362305, 97.435791015625, 92.7265625, 101.04350280761719, -92.54264831542969, 103.3614501953125, 10.842597961425781, -0.559326171875, -13.119438171386719, 34.17338562011719, 165.76708984375, 29.076335906982422, 125.01394653320312, 63.30015563964844, 35.5010986328125, 129.3212432861328, 130.2860107421875, 140.86508178710938, -60.163421630859375, -7.5496826171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000467.npy"}
|
|
{"epoch": 0.9780104712041885, "step": 468, "batch_size": 128, "mean": 35.084388732910156, "std": 59.991729736328125, "min": -100.5745849609375, "p10": -30.58871307373046, "median": 23.445068359375, "p90": 117.86330032348633, "max": 185.66082763671875, "pos_frac": 0.7109375, "sample": [36.47540283203125, 145.64459228515625, 101.43559265136719, -35.76408386230469, 25.75439453125, 14.66851806640625, -12.92779541015625, 105.10516357421875, 44.06218719482422, 24.45965576171875, 28.47125244140625, -26.432449340820312, 22.43048095703125, 133.93423461914062, -43.81890869140625, -12.484321594238281, -8.057373046875, 126.81196594238281, -79.03901672363281, -16.110015869140625, 54.22819519042969, 26.600746154785156, -53.333473205566406, 54.7552490234375, 49.600921630859375, 159.13055419921875, 46.11236572265625, 58.43470764160156, 21.582855224609375, -28.19631576538086, 3.85455322265625, 14.406387329101562, 140.53128051757812, -27.627939224243164, 32.358428955078125, -23.127687454223633, -17.768051147460938, 119.16717529296875, -100.5745849609375, 15.034942626953125, 9.794471740722656, 19.616943359375, -17.267181396484375, 97.11666870117188, 6.649898529052734, 13.502227783203125, 58.7650146484375, -41.115142822265625, 45.99043273925781, 3.8662796020507812, -0.6580657958984375, 1.4734458923339844, 26.95794677734375, 0.1654052734375, 9.969467163085938, 35.85447692871094, 10.012115478515625, 91.37623596191406, 98.82514190673828, 108.95248413085938, 128.7373046875, -17.73675537109375, -100.33724975585938, 12.889556884765625, 185.66082763671875, 9.530364990234375, 19.221176147460938, 107.19717407226562, -2.51348876953125, 43.008453369140625, 42.736663818359375, 47.339080810546875, -25.449798583984375, 67.93731689453125, 108.90691375732422, 110.9298095703125, 5.2487945556640625, 15.77557373046875, 8.232437133789062, 81.48466491699219, -26.484111785888672, 145.8758544921875, 27.8074893951416, 129.837890625, 33.05780029296875, 84.31129455566406, 110.579345703125, -14.686101913452148, 102.84555053710938, 79.0417709350586, -11.20892333984375, 5.51689338684082, 2.686309814453125, 145.2677001953125, -0.446044921875, 30.919403076171875, 91.34352111816406, 2.8060989379882812, 35.48095703125, -2.6923370361328125, 115.06663513183594, 117.30449676513672, 77.50080871582031, 5.54608154296875, -46.49725341796875, -24.089691162109375, -48.16229248046875, 60.07025146484375, 47.555908203125, 112.73397827148438, -14.8623046875, -5.8408203125, 90.42095947265625, -68.07527160644531, -43.540496826171875, 10.76904296875, 110.48756408691406, 2.0121994018554688, -5.647727966308594, -72.76485443115234, 99.00770568847656, 111.445556640625, 55.385284423828125, 119.60101318359375, -40.76812744140625, 122.55477905273438, -28.370697021484375, 111.66180419921875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000468.npy"}
|
|
{"epoch": 0.9801047120418848, "step": 469, "batch_size": 128, "mean": 46.153438568115234, "std": 67.12257385253906, "min": -147.03927612304688, "p10": -28.816894531249996, "median": 41.595855712890625, "p90": 129.25328216552734, "max": 166.4637451171875, "pos_frac": 0.7578125, "sample": [14.7515869140625, 127.38064575195312, 0.0, 11.774177551269531, 28.605712890625, 49.34159851074219, -41.9371337890625, 27.53973388671875, -5.2569580078125, 70.00262451171875, 86.74684143066406, 33.485748291015625, 113.43499755859375, -147.03927612304688, 24.69036865234375, 14.46295166015625, 142.0289764404297, 107.68110656738281, 19.487762451171875, 97.69866943359375, -37.90550231933594, 67.06985473632812, 6.3215179443359375, -2.443380355834961, -108.53689575195312, 123.07411193847656, 8.73529052734375, 125.6806640625, 128.42971801757812, 28.523956298828125, 100.23094940185547, 75.1114501953125, 159.3482666015625, 109.85031127929688, 1.153472900390625, 68.51445007324219, 46.5172119140625, 71.18548583984375, 127.7021484375, 43.765106201171875, -1.989349365234375, -32.66778564453125, 24.4017333984375, 8.678466796875, 105.36671447753906, -4.753997802734375, 19.20555877685547, 25.118194580078125, 43.16314697265625, 32.69378662109375, 93.12590026855469, 130.20782470703125, 122.93638610839844, 132.69830322265625, -92.2276611328125, 83.73380279541016, 101.01971435546875, 153.00469970703125, 115.69754028320312, -47.678375244140625, 5.541328430175781, -6.68328857421875, 141.120361328125, -70.195068359375, -85.82621002197266, 58.003662109375, 22.434066772460938, -6.712890625, -22.244232177734375, 121.89852905273438, 25.386993408203125, 103.6210708618164, 70.1011962890625, 121.61750793457031, -122.31509399414062, 4.067605972290039, 45.909881591796875, 111.19912719726562, -6.9109344482421875, 155.69638061523438, 121.73272705078125, 56.868499755859375, 35.6451416015625, 0.301239013671875, 128.22137451171875, -8.84307861328125, -140.27813720703125, 4.928293228149414, 43.604461669921875, 19.799232482910156, 55.406982421875, 132.3690185546875, -3.6063690185546875, 124.36807250976562, -27.752593994140625, -19.443130493164062, 0.5907554626464844, 2.73138427734375, -12.28384017944336, 98.97444152832031, -12.327362060546875, 2.8250198364257812, -13.17572021484375, 166.4637451171875, 17.70665740966797, -37.405548095703125, 128.8441925048828, 72.74441528320312, 103.49533081054688, 139.73297119140625, 89.82991027832031, 8.880111694335938, 124.26919555664062, 40.028564453125, -10.825809478759766, 65.01751708984375, 23.70162582397461, 114.52163696289062, 96.57183837890625, 49.490196228027344, 135.95556640625, 131.0836181640625, 71.37986755371094, -15.830047607421875, 64.19229125976562, 25.143310546875, 148.63961791992188, -31.300262451171875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000469.npy"}
|
|
{"epoch": 0.9821989528795811, "step": 470, "batch_size": 128, "mean": 48.272979736328125, "std": 67.69413757324219, "min": -114.586181640625, "p10": -29.526858520507812, "median": 43.635108947753906, "p90": 135.0876724243164, "max": 169.0737762451172, "pos_frac": 0.765625, "sample": [-16.74713134765625, 20.488922119140625, 123.19488525390625, 65.76126098632812, -4.7762298583984375, 52.684844970703125, 6.1483001708984375, 113.2718734741211, 124.6046142578125, 19.073089599609375, -29.9715576171875, 3.428070068359375, -3.5253372192382812, 101.62744140625, 101.636474609375, 112.7783203125, 129.5029296875, 1.7040252685546875, -11.870079040527344, 42.44403076171875, 85.16742706298828, -67.10107421875, -81.44306945800781, -86.66012573242188, 159.55755615234375, 104.86904907226562, 52.443275451660156, 15.52679443359375, 117.72506713867188, 111.65316772460938, 150.9588623046875, 92.80630493164062, 2.3356246948242188, -1.40838623046875, 92.03408813476562, -79.68695068359375, 101.10845947265625, 90.28468322753906, 23.684356689453125, -76.88572692871094, 10.08709716796875, 147.6583251953125, 90.06074523925781, 20.6524658203125, 8.985626220703125, 110.60800170898438, 147.08242797851562, -27.315780639648438, 31.709598541259766, 84.70315551757812, 6.44732666015625, 104.75262451171875, 23.73779296875, 22.61371421813965, -16.585174560546875, 98.5562744140625, 68.01701354980469, 1.264678955078125, 110.41018676757812, 14.63470458984375, 51.91919708251953, 3.7690486907958984, 117.60548400878906, 15.403701782226562, 25.291397094726562, 20.484466552734375, -2.0163040161132812, -114.586181640625, 15.26263427734375, 74.43417358398438, -22.086837768554688, 112.84468841552734, 87.77151489257812, -23.451065063476562, -81.54730224609375, -38.22016906738281, -30.687255859375, 92.20560455322266, 133.0999298095703, 109.8505859375, 30.2318115234375, 37.59051513671875, 103.04205322265625, 158.78811645507812, 44.82618713378906, -12.748298645019531, 10.520645141601562, 141.689697265625, 147.44775390625, 139.72573852539062, -19.072898864746094, 106.42605590820312, 94.11033630371094, 109.88259887695312, -1.5027923583984375, 42.430267333984375, 16.01384735107422, 98.51847839355469, 94.861572265625, 3.614990234375, 126.22046661376953, 132.82666015625, 81.02281188964844, 169.0737762451172, -1.020050048828125, 72.74920654296875, 1.3162689208984375, 7.2752685546875, -9.76214599609375, -107.74340057373047, 13.91241455078125, -75.87562561035156, 155.27297973632812, -26.50482177734375, 79.87474822998047, 153.20233154296875, 1.6820068359375, 54.164520263671875, 67.1619873046875, 54.830078125, 150.861328125, -83.69573974609375, 57.066925048828125, 30.4749755859375, 151.57464599609375, 110.48609161376953, 131.57687377929688, -29.336273193359375], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000470.npy"}
|
|
{"epoch": 0.9842931937172775, "step": 471, "batch_size": 128, "mean": 34.91949462890625, "std": 60.38799285888672, "min": -122.38920593261719, "p10": -31.03776359558105, "median": 26.897933959960938, "p90": 120.96441955566405, "max": 178.42822265625, "pos_frac": 0.6796875, "sample": [-57.55171203613281, 102.44102478027344, 28.101715087890625, 47.66265869140625, 96.91094970703125, 158.6773681640625, 78.52015686035156, -80.73504638671875, -10.170589447021484, 99.82591247558594, -35.84013366699219, -19.129150390625, 4.584442138671875, 81.99248504638672, -0.5346603393554688, 37.34649658203125, 18.7918701171875, 30.089004516601562, -21.55547332763672, -23.168792724609375, 25.69415283203125, 23.746002197265625, 17.08331298828125, 32.421173095703125, -45.936126708984375, -5.856163024902344, 147.36834716796875, -4.12872314453125, -20.095062255859375, 28.10406494140625, 95.50491333007812, 73.93841552734375, 57.82818603515625, 11.05218505859375, -5.061676025390625, 54.05851745605469, 58.57792663574219, 80.75642395019531, 30.45550537109375, 15.141387939453125, 8.372238159179688, -4.31988525390625, 122.052978515625, 128.0577392578125, -20.538433074951172, 60.518798828125, 32.31187438964844, 146.6319580078125, 100.77474975585938, -4.398353576660156, 5.5977020263671875, 13.5595703125, 125.41049194335938, 43.20538330078125, -51.7960205078125, 98.6781005859375, 85.04515075683594, 3.1879425048828125, -3.7626800537109375, 24.544105529785156, 162.48388671875, 178.42822265625, 12.736719131469727, -3.7689666748046875, -30.065113067626953, -7.3807830810546875, -9.332317352294922, 3.2825469970703125, -122.38920593261719, 38.037109375, 65.93368530273438, 21.52471923828125, 107.30686950683594, 36.78675842285156, -100.90068054199219, 33.111083984375, -12.901458740234375, 47.83356475830078, 48.224937438964844, -34.88824462890625, 48.486114501953125, -16.29571533203125, -1.91229248046875, -8.293190002441406, 47.2996826171875, 91.58126068115234, 12.464752197265625, 7.904853820800781, 99.64483642578125, -4.43450927734375, 69.72634887695312, 114.10610961914062, -3.66119384765625, 48.568878173828125, 19.598373413085938, 117.86149597167969, 127.9407958984375, -29.297473907470703, -1.9586029052734375, -16.455184936523438, 30.916065216064453, 13.184211730957031, 18.7647705078125, 50.7373046875, 61.353302001953125, 120.49789428710938, 39.416015625, 165.09634399414062, 61.81382751464844, 8.890594482421875, 55.3216552734375, 95.51458740234375, 129.63775634765625, -77.83739471435547, 116.4638671875, 25.05975341796875, 93.4967041015625, 0.0, 40.49945068359375, -48.18988037109375, 141.38861083984375, 112.21218872070312, 140.16949462890625, -16.085052490234375, 11.3873291015625, -33.307281494140625, -74.46461486816406, -89.22610473632812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000471.npy"}
|
|
{"epoch": 0.9863874345549738, "step": 472, "batch_size": 128, "mean": 46.27994155883789, "std": 65.01326751708984, "min": -110.82937622070312, "p10": -22.69380187988281, "median": 32.65887641906738, "p90": 126.02694702148438, "max": 216.784912109375, "pos_frac": 0.7421875, "sample": [216.784912109375, -8.268585205078125, -0.9641971588134766, -47.858673095703125, 4.703432083129883, 6.685516357421875, 4.763799667358398, 106.8875732421875, 0.797882080078125, 25.49737548828125, 66.74659729003906, 43.3963623046875, 122.82177734375, 40.99445343017578, -65.49494171142578, 110.24835968017578, -14.136497497558594, 9.305343627929688, -29.1939697265625, 65.23811340332031, -6.080291748046875, 25.0927734375, 80.75137329101562, 37.056396484375, -20.567405700683594, 117.87277221679688, 93.19120025634766, -85.25643920898438, 4.178508758544922, 112.16075897216797, 42.3458251953125, -20.351150512695312, 17.363189697265625, 161.508544921875, -1.821258544921875, -24.471893310546875, 46.900634765625, 125.277099609375, -21.9317626953125, 4.256721496582031, 18.826019287109375, 87.75653076171875, 79.61257934570312, 66.95361328125, -17.270828247070312, 19.016754150390625, 21.090972900390625, 3.227752685546875, 53.635101318359375, 12.936065673828125, 104.60232543945312, 44.93437194824219, 125.62979125976562, -86.9542236328125, 91.08029174804688, 99.91120910644531, 122.50341796875, 97.82339477539062, 23.657821655273438, 26.842391967773438, -7.8540191650390625, 73.85626220703125, 8.166858673095703, -1.745147705078125, 99.185546875, 120.39553833007812, 76.22605895996094, 68.73033142089844, 83.31065368652344, -3.1941070556640625, 31.867313385009766, 26.31446075439453, 25.552093505859375, 118.58625793457031, 96.79397583007812, 157.39337158203125, 133.5538330078125, -8.21664810180664, 122.78233337402344, 20.4813232421875, -10.27203369140625, 182.50860595703125, 120.49751281738281, 150.25808715820312, 22.188323974609375, 162.5771484375, 152.96633911132812, 128.86386108398438, 109.298583984375, -110.82937622070312, -31.463668823242188, -89.7840576171875, 67.21870422363281, 46.3028564453125, 5.35089111328125, 94.88168334960938, 111.72195434570312, 11.968582153320312, -30.3056640625, 80.96932983398438, 119.73379516601562, -9.508171081542969, 125.17489624023438, 99.40536499023438, 23.88286590576172, 61.97437286376953, -4.3355712890625, 14.639190673828125, -2.5539703369140625, 5.1913604736328125, 135.12738037109375, 132.68011474609375, 4.56395149230957, 11.686309814453125, 33.450439453125, -92.14027404785156, -13.389175415039062, 181.8519287109375, -9.7078857421875, 30.62921142578125, 126.95364379882812, 71.08416748046875, -60.350982666015625, 75.932373046875, 69.9263916015625, 82.75418090820312, -34.98644256591797, -15.186477661132812], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000472.npy"}
|
|
{"epoch": 0.9884816753926702, "step": 473, "batch_size": 128, "mean": 47.45037841796875, "std": 64.2818832397461, "min": -132.62232971191406, "p10": -23.287434387207014, "median": 40.72900581359863, "p90": 133.02793579101564, "max": 165.3851318359375, "pos_frac": 0.7421875, "sample": [8.266761779785156, 133.15350341796875, 40.53912353515625, -1.7880268096923828, 79.10435485839844, -6.548370361328125, 18.790863037109375, 14.4664306640625, 37.06024169921875, -8.541534423828125, 56.73445129394531, 32.424644470214844, -69.163330078125, 18.01466178894043, -62.977500915527344, 54.74929428100586, 44.49859619140625, 12.663742065429688, 61.133636474609375, 95.81271362304688, 68.69723510742188, 107.16534423828125, 139.35107421875, 160.52508544921875, 60.68951416015625, -79.68499755859375, -11.499282836914062, 94.31085205078125, 69.70880126953125, 15.611351013183594, 140.21807861328125, 126.24105834960938, 116.21878051757812, 34.58697509765625, -12.184394836425781, -35.72703552246094, 29.308929443359375, -61.795616149902344, 148.01473999023438, -9.288841247558594, 62.987030029296875, 94.03619384765625, 33.621337890625, -13.599010467529297, -12.332595825195312, -46.913482666015625, -44.800933837890625, 132.97412109375, 49.7493896484375, 34.1182861328125, 41.828582763671875, 16.078140258789062, -0.18759727478027344, 144.68399047851562, -16.40704345703125, 135.243408203125, -17.9561767578125, 90.23457336425781, 148.0576171875, -107.943603515625, 14.550872802734375, 137.49591064453125, 126.49618530273438, 3.6862106323242188, 27.177764892578125, -39.978271484375, 163.053466796875, 120.09075164794922, 81.37142944335938, 149.76812744140625, 72.92254638671875, 93.23279571533203, 80.14218139648438, 16.114959716796875, 4.117958068847656, 109.6056137084961, 112.49215698242188, 9.053932189941406, -46.0745849609375, 111.06207275390625, -5.3299560546875, -15.423728942871094, 125.73223876953125, 84.74348449707031, 90.69403076171875, 124.12799072265625, -132.62232971191406, -80.25390625, 104.8277587890625, -6.62969970703125, 48.1849365234375, 107.84576416015625, 2.6839828491210938, 37.62721252441406, -5.342987060546875, -16.681228637695312, 40.918888092041016, 15.89508056640625, 25.57091522216797, -16.266677856445312, 135.64962768554688, 98.60514831542969, 102.8114242553711, 0.0, 99.217041015625, 65.6756591796875, -61.082855224609375, 106.76359558105469, 24.37408447265625, 91.92344665527344, 108.76446533203125, 13.430404663085938, 2.781269073486328, 129.46835327148438, 52.31207275390625, -6.09417724609375, 5.03802490234375, -17.420364379882812, 26.16510009765625, 44.67573547363281, 115.90863037109375, 114.36825561523438, 118.52407836914062, 37.289588928222656, 165.3851318359375, 108.68492126464844, 92.13017272949219, 39.28143310546875], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000473.npy"}
|
|
{"epoch": 0.9905759162303664, "step": 474, "batch_size": 128, "mean": 36.236385345458984, "std": 57.944068908691406, "min": -124.79293823242188, "p10": -29.480023193359376, "median": 33.41061782836914, "p90": 112.70268325805664, "max": 165.32492065429688, "pos_frac": 0.75, "sample": [-7.0495452880859375, 6.60491943359375, 164.5574951171875, -4.357208251953125, -6.275503158569336, -24.7734375, 74.9310302734375, 139.19967651367188, 51.4713134765625, 108.15483093261719, 8.22406005859375, 43.06085205078125, -14.65234375, -87.18701171875, 7.146503448486328, -78.69447326660156, -1.5526123046875, 0.47724151611328125, 47.89093780517578, 27.448654174804688, 50.7667236328125, 132.8511199951172, 6.511962890625, -47.107940673828125, 96.05686950683594, -35.25981521606445, -59.4752197265625, 41.929931640625, -17.27613067626953, 42.23358154296875, 95.9158935546875, 40.147216796875, 6.7459716796875, 99.5347900390625, 45.248748779296875, 84.09576416015625, 163.8914794921875, 17.2618350982666, 3.2116851806640625, 141.44937133789062, 86.99945068359375, 126.44375610351562, 142.3094482421875, 127.48118591308594, 55.426605224609375, 6.397586822509766, 0.0, 4.341888427734375, 16.496307373046875, 6.098762512207031, -0.7651786804199219, -9.73931884765625, -44.931365966796875, 41.79399108886719, 28.113636016845703, 82.19921875, 33.37348175048828, -29.58727264404297, 120.63426208496094, 103.27748107910156, 23.011474609375, -9.448150634765625, -4.276390075683594, 79.35440063476562, 93.23709106445312, 22.90435791015625, 114.52855682373047, 55.830108642578125, -33.24249267578125, 8.446075439453125, 96.46224975585938, 28.843677520751953, -86.61188507080078, 39.46674728393555, 44.108856201171875, -21.25543212890625, 132.55865478515625, 91.10755920410156, 88.53753662109375, -5.2593231201171875, -5.45697021484375, 20.339263916015625, 0.2929859161376953, 3.893054962158203, 2.6662940979003906, -116.2054443359375, -29.434059143066406, 77.39118957519531, -76.00584411621094, 108.36859130859375, 67.30027770996094, -124.79293823242188, -25.831695556640625, -6.39703369140625, 77.10906982421875, 7.554058074951172, 45.84761047363281, -10.52783203125, 29.400344848632812, 88.13723754882812, 64.7445068359375, 111.920166015625, 75.275634765625, 57.625244140625, 165.32492065429688, 83.80459594726562, 100.92596435546875, 19.035003662109375, 37.860870361328125, 23.833267211914062, 40.493408203125, 59.69659423828125, 63.93402099609375, 51.9678955078125, 33.44775390625, 46.487640380859375, 81.33039855957031, 68.97555541992188, 15.860977172851562, 46.16064453125, 44.82740020751953, 127.50303649902344, 22.6173095703125, -59.963165283203125, 24.399566650390625, 8.68255615234375, 28.20263671875, 41.6080322265625], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000474.npy"}
|
|
{"epoch": 0.9926701570680628, "step": 475, "batch_size": 128, "mean": 37.36973571777344, "std": 62.08230972290039, "min": -141.79714965820312, "p10": -38.43451080322265, "median": 29.278457641601562, "p90": 116.02587432861327, "max": 181.27053833007812, "pos_frac": 0.7109375, "sample": [142.09469604492188, 111.43070983886719, -42.28727722167969, 114.8341064453125, 101.5751953125, 16.304473876953125, -141.79714965820312, 151.64694213867188, -60.49903106689453, -43.527984619140625, -13.843032836914062, -31.08880615234375, -65.6253662109375, -36.7833251953125, 18.6800537109375, 66.55551147460938, 59.183109283447266, 110.22035217285156, 118.65240478515625, -8.984138488769531, 142.35733032226562, -15.697349548339844, 34.54772186279297, 27.648712158203125, 131.69686889648438, 115.30595397949219, 104.16436004638672, -9.462432861328125, 169.0548095703125, -0.9966602325439453, 32.64381408691406, 36.440399169921875, -60.8011474609375, 108.21577453613281, 37.76197814941406, 18.09082794189453, -61.23797607421875, 24.3052978515625, 0.08389854431152344, 117.7056884765625, 92.78515625, 28.785987854003906, -94.27435302734375, 51.96986389160156, 68.85438537597656, -14.865036010742188, 79.71959686279297, -10.71673583984375, 104.91740417480469, -24.9976806640625, 12.335350036621094, -36.504417419433594, 103.12249755859375, 111.97592163085938, -63.338043212890625, 110.90625, 136.25225830078125, -6.821786880493164, 18.246004104614258, 35.582916259765625, 103.5609130859375, 20.506622314453125, 7.402492523193359, -18.836090087890625, 29.45501708984375, 34.407432556152344, 22.122909545898438, 40.019012451171875, 12.998397827148438, 24.2484130859375, 12.578155517578125, 118.16400146484375, 52.33610534667969, -14.07574462890625, 92.43194580078125, -29.91943359375, 3.4145050048828125, 181.27053833007812, 36.1956787109375, -4.5861663818359375, 56.6070556640625, 142.5755615234375, 109.25970458984375, -32.846832275390625, 106.89766693115234, -24.158008575439453, 57.13600540161133, 9.161376953125, 47.73870849609375, 19.31536865234375, 1.2415199279785156, 36.20109176635742, 106.36782836914062, 94.97694396972656, -1.725433349609375, 114.82638549804688, 105.27250671386719, 118.1587142944336, 39.65550231933594, 67.30555725097656, 11.421958923339844, -44.761474609375, 26.635040283203125, -0.1538238525390625, -7.8542938232421875, -25.434478759765625, -65.428466796875, 37.06011962890625, 92.97288513183594, 60.68097686767578, 76.80149841308594, 71.31663513183594, 14.839736938476562, 14.572235107421875, 96.94454956054688, -31.90606689453125, 60.00083923339844, -46.01940155029297, -61.5240478515625, 29.387847900390625, 109.21853637695312, 29.1690673828125, 17.1435546875, 148.15505981445312, 39.67677688598633, 12.717987060546875, -4.447460174560547, 1.973876953125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000475.npy"}
|
|
{"epoch": 0.9947643979057592, "step": 476, "batch_size": 128, "mean": 43.710411071777344, "std": 63.328163146972656, "min": -115.51130676269531, "p10": -20.135858154296876, "median": 31.516525268554688, "p90": 126.91989288330078, "max": 178.29571533203125, "pos_frac": 0.7265625, "sample": [-34.305137634277344, -67.8389892578125, -8.599838256835938, -30.899749755859375, 7.28369140625, 99.08746337890625, 2.9250640869140625, -5.16578483581543, 18.271697998046875, 66.41847229003906, 92.899169921875, 35.152099609375, 123.45277404785156, 139.62164306640625, 79.0567398071289, 158.5316162109375, -20.106781005859375, 134.2963104248047, 107.7254638671875, -2.998676300048828, -5.534629821777344, 31.589080810546875, 13.94537353515625, 31.4439697265625, 92.21908569335938, -16.87145233154297, 122.0221176147461, -5.735851287841797, -88.49703979492188, 142.32662963867188, 10.325439453125, 31.241851806640625, -24.919143676757812, 137.34219360351562, 116.81842041015625, 14.1563720703125, 125.46905517578125, -63.833213806152344, 25.70050048828125, -13.263153076171875, -2.04864501953125, -107.53292846679688, 91.981689453125, 18.647254943847656, 55.457763671875, 31.348907470703125, 100.4729995727539, 148.20932006835938, 33.19377136230469, -61.435943603515625, 68.31277465820312, 125.655029296875, 12.511199951171875, 12.880317687988281, 1.0917072296142578, 148.073486328125, 9.786773681640625, -14.96453857421875, 116.32881164550781, 25.083953857421875, 100.01126098632812, 76.74698638916016, -23.08538818359375, 126.76266479492188, 21.67657470703125, 139.2888946533203, 130.4603271484375, 15.07666015625, 106.36756896972656, 58.432525634765625, 26.026611328125, -11.790252685546875, 153.19020080566406, 37.5992431640625, 37.620018005371094, 0.8042488098144531, 101.50627899169922, -89.14655303955078, 108.63421630859375, 46.431396484375, 110.95497131347656, 100.38446044921875, 99.09294128417969, 0.720001220703125, 8.306404113769531, 6.5045928955078125, -4.855766296386719, 98.67115783691406, -115.51130676269531, 78.82953643798828, -8.957290649414062, 39.0919189453125, 136.50070190429688, 2.628204345703125, 33.16273498535156, -8.989028930664062, -10.265899658203125, 54.26300811767578, 3.25482177734375, 102.30931091308594, -20.203704833984375, 32.21484375, -17.6907958984375, 178.29571533203125, 17.333045959472656, -18.196487426757812, -8.107391357421875, 27.727291107177734, 75.51840209960938, 82.58526611328125, 40.552734375, 108.60409545898438, 111.12796020507812, 18.0091552734375, 8.313644409179688, -16.182830810546875, 97.36993408203125, -9.017990112304688, -2.9517974853515625, 43.39068603515625, 126.71038818359375, 100.08599853515625, -58.99688720703125, 126.50007629394531, -18.6427001953125, 127.28675842285156, 99.26426696777344, 71.51729583740234], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000476.npy"}
|
|
{"epoch": 0.9968586387434555, "step": 477, "batch_size": 128, "mean": 46.6090087890625, "std": 68.06424713134766, "min": -144.14205932617188, "p10": -35.2608154296875, "median": 40.37795829772949, "p90": 129.88113861083986, "max": 176.54266357421875, "pos_frac": 0.71875, "sample": [-64.05230712890625, 42.96588134765625, 108.43727111816406, -68.151611328125, 116.8587646484375, 35.08660888671875, -61.30023193359375, 102.69268798828125, -24.267501831054688, 72.84529113769531, -7.907978057861328, 129.8204345703125, 6.2653045654296875, 92.13632202148438, 63.90985107421875, -13.032073974609375, 144.380615234375, 74.58102416992188, -144.14205932617188, -16.0938720703125, 15.05859375, -34.322052001953125, 64.56645202636719, 135.19046020507812, 145.40696716308594, 135.78988647460938, 86.41322326660156, -5.621465682983398, 12.6544189453125, 29.92193603515625, 107.99581909179688, 123.0845947265625, 15.50299072265625, 161.56906127929688, 36.568817138671875, 21.1712646484375, 144.8282470703125, 29.41436004638672, -7.870330810546875, 0.0, 14.6661376953125, 107.65776062011719, 23.673202514648438, -25.862869262695312, 74.15050506591797, -10.474456787109375, 17.726806640625, 110.3079833984375, 114.724365234375, -57.9080810546875, -24.814247131347656, 1.156158447265625, -23.215591430664062, 114.04150390625, 20.662628173828125, 20.020172119140625, 108.13320922851562, 25.681884765625, 109.8154296875, 73.63224792480469, 102.36529541015625, 130.0227813720703, 176.54266357421875, -3.60736083984375, 64.7020263671875, 90.83036041259766, 139.863525390625, -60.25469970703125, 57.038963317871094, -48.69622802734375, 65.74736785888672, 114.1192626953125, 78.30598449707031, 99.19073486328125, 113.3779296875, 129.76974487304688, -32.28199768066406, 8.234024047851562, 0.7196426391601562, 110.0482177734375, -16.535247802734375, 125.93275451660156, 54.35272216796875, -19.45647430419922, -34.0272216796875, -112.11275482177734, 115.79859924316406, 85.97315979003906, 37.790035247802734, 83.64678955078125, -3.112940788269043, 15.59588623046875, -8.958301544189453, 145.811767578125, -16.62506103515625, 105.02098083496094, -62.28510284423828, 90.17662048339844, 37.060882568359375, 118.83870697021484, -42.220367431640625, 114.91549682617188, -43.660430908203125, -118.30941772460938, 13.20904541015625, -11.832733154296875, -37.451263427734375, -5.683563232421875, 149.07867431640625, 60.438446044921875, 35.049560546875, 171.48577880859375, 112.4639892578125, 139.47769165039062, 5.8955078125, 117.25493621826172, 111.87278747558594, -33.266387939453125, 124.19309997558594, 16.008514404296875, 45.261810302734375, 98.02346801757812, 108.94676208496094, 26.48143768310547, 1.1879501342773438, 76.23719787597656, 43.37689208984375, 6.4635772705078125], "npy": "/root/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.5-20260430-194457/margin_logs/step_0000477.npy"}
|