Files
ModelHub XC e8efbe1ca9 初始化项目,由ModelHub XC社区提供模型
Model: W-61/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725
Source: Original Platform
2026-05-20 08:27:04 +08:00

478 lines
1.3 MiB

{"epoch": 0.0, "step": 1, "batch_size": 128, "mean": -0.02793477475643158, "std": 0.6237562894821167, "min": -2.3379974365234375, "p10": -0.7331054687499999, "median": -0.01795196533203125, "p90": 0.5892669677734375, "max": 1.66827392578125, "pos_frac": 0.4921875, "sample": [0.1674652099609375, 0.0925140380859375, 0.15960693359375, -0.04441070556640625, 1.4956512451171875, 0.2601776123046875, -0.199493408203125, 0.20355224609375, -0.645050048828125, 0.5536956787109375, 1.0389938354492188, -0.570037841796875, 0.210723876953125, 0.2867774963378906, 0.47210693359375, -0.19388961791992188, 0.446380615234375, -1.2149658203125, 0.22918701171875, 1.66827392578125, -0.710540771484375, -0.70892333984375, 0.5078010559082031, 0.137725830078125, -1.50103759765625, -0.093780517578125, -0.19947052001953125, 0.131744384765625, 0.47847747802734375, 0.55963134765625, -0.65667724609375, 0.0279388427734375, -0.722320556640625, -0.1856689453125, -0.7859954833984375, -0.54620361328125, -0.29620361328125, 0.31866455078125, -0.099853515625, -0.62738037109375, 0.5283050537109375, 0.406494140625, -0.2256622314453125, 0.79986572265625, 0.8321533203125, -0.26869964599609375, -0.19608306884765625, -0.14408111572265625, -0.6990966796875, -1.551727294921875, -0.18653106689453125, 0.19101333618164062, -0.385894775390625, -0.2215576171875, 0.587554931640625, 0.01473236083984375, 0.3006134033203125, 0.037418365478515625, -0.1775360107421875, -0.0821533203125, 0.2209625244140625, -0.175567626953125, -1.319793701171875, -0.29390716552734375, -0.7856903076171875, -1.133575439453125, -2.3379974365234375, -0.582763671875, -0.407562255859375, 0.3812713623046875, -0.08927154541015625, 0.34177398681640625, 0.3060302734375, 0.201385498046875, 1.053680419921875, 0.5845870971679688, -1.07904052734375, 0.0, -0.0983428955078125, 0.143341064453125, -0.0359039306640625, 1.14263916015625, -0.6820068359375, -0.6137542724609375, 0.59326171875, 0.4530029296875, 0.039764404296875, -0.214813232421875, 0.38768768310546875, 0.952484130859375, 0.9595947265625, 0.2894439697265625, 0.4634552001953125, 0.46490478515625, -0.1591033935546875, -0.5848388671875, -0.758270263671875, 0.43994140625, -0.9514617919921875, -0.097900390625, -0.6257400512695312, -0.23140716552734375, 0.3023681640625, 1.04931640625, -0.2509117126464844, 0.964508056640625, -0.245361328125, 0.3499603271484375, 0.4720001220703125, 0.113616943359375, 0.24009323120117188, -0.6871795654296875, -0.339508056640625, -0.3977317810058594, -1.02557373046875, 0.550262451171875, -0.042972564697265625, 0.3683624267578125, -0.143951416015625, 0.9142074584960938, -0.1939697265625, 0.131011962890625, -1.03314208984375, 0.1519756317138672, -0.14737701416015625, 0.253753662109375, 0.50762939453125, -0.6018524169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000001.npy"}
{"epoch": 0.0020942408376963353, "step": 2, "batch_size": 128, "mean": 0.014312252402305603, "std": 0.6668511629104614, "min": -1.97247314453125, "p10": -0.8530120849609375, "median": 0.042736053466796875, "p90": 0.8162986755371093, "max": 2.1260986328125, "pos_frac": 0.515625, "sample": [-0.946807861328125, 0.58673095703125, -0.07851600646972656, 0.3317985534667969, -0.2125396728515625, -0.79388427734375, -0.061969757080078125, 0.03466033935546875, -1.2871551513671875, 0.349517822265625, -0.5836944580078125, -0.7088851928710938, 0.17462158203125, -0.19950103759765625, -0.4486236572265625, 0.1865081787109375, 0.31939697265625, 0.51416015625, 0.603607177734375, -0.423065185546875, 0.29950523376464844, 0.19829559326171875, -0.7109375, 0.547821044921875, -1.3863983154296875, -0.48331451416015625, -0.283416748046875, 0.917938232421875, 0.440765380859375, -0.3415374755859375, -0.322021484375, -0.1216583251953125, -0.14400482177734375, -1.139923095703125, -0.21550750732421875, -0.4138031005859375, 1.39569091796875, -0.739013671875, -1.128082275390625, 0.9674072265625, 0.99932861328125, 0.067962646484375, 0.8251266479492188, 0.1144866943359375, -0.85760498046875, -0.9853515625, 0.0034122467041015625, -0.28753662109375, -0.39898681640625, 0.06365966796875, 1.42681884765625, -1.2214508056640625, 2.1260986328125, 0.19042205810546875, 0.0, -0.1408672332763672, 0.07848358154296875, -0.04154014587402344, -1.97247314453125, -0.28961181640625, 0.4712066650390625, 0.6232757568359375, -0.58184814453125, 0.06168365478515625, 0.069488525390625, 0.7487945556640625, -1.0627593994140625, 0.75299072265625, -0.244537353515625, 0.9249114990234375, -0.5446319580078125, -0.7351150512695312, 0.09808349609375, 0.936187744140625, -0.0081024169921875, 0.050811767578125, 0.619873046875, 0.8125152587890625, -0.04132080078125, 0.397979736328125, 0.41588592529296875, -0.06989288330078125, -0.0491485595703125, 0.23805618286132812, 0.6832046508789062, 0.138214111328125, 1.1375579833984375, 0.3428459167480469, 0.33001708984375, 0.06795883178710938, -0.851043701171875, 0.13287353515625, -0.1810455322265625, 1.65087890625, -0.05999755859375, 0.15135574340820312, -0.0111846923828125, 0.565032958984375, -0.2957305908203125, -0.1913604736328125, 0.564849853515625, -0.92486572265625, 0.498077392578125, -0.666595458984375, 0.482147216796875, 0.06561279296875, 0.35321044921875, 0.1884613037109375, -0.12225723266601562, 1.531219482421875, 0.1341400146484375, 0.465179443359375, 0.7451171875, 1.089202880859375, 0.103302001953125, -0.5128173828125, 0.10821533203125, -0.3601531982421875, -0.32855224609375, -0.611328125, 0.48419189453125, -1.094207763671875, -0.7486114501953125, -0.090087890625, -0.925384521484375, -0.3560943603515625, -0.7297210693359375, 0.6012115478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000002.npy"}
{"epoch": 0.004188481675392671, "step": 3, "batch_size": 128, "mean": 0.05597783625125885, "std": 0.7559072971343994, "min": -1.98895263671875, "p10": -0.891766357421875, "median": 0.060558319091796875, "p90": 1.0278747558593748, "max": 2.13446044921875, "pos_frac": 0.546875, "sample": [-1.06011962890625, 0.1492481231689453, 0.06862258911132812, -0.2413330078125, 0.18970108032226562, 0.3020477294921875, 0.65966796875, 2.13446044921875, 0.0, 0.025054931640625, -0.3876953125, -0.9686279296875, 1.020843505859375, -1.52471923828125, -0.19903564453125, -0.6427841186523438, 0.1014404296875, 1.3555908203125, 1.4489898681640625, -0.03424072265625, 0.196563720703125, -0.48394775390625, 0.27008056640625, -0.25455474853515625, -0.300018310546875, 0.63909912109375, -0.82647705078125, 0.1082763671875, 0.359619140625, 0.383209228515625, 0.11820602416992188, 0.060337066650390625, 0.076751708984375, 0.464569091796875, 1.805908203125, 0.5835647583007812, 0.1717376708984375, -1.0531005859375, -0.013458251953125, 0.18603515625, 0.513916015625, 0.315643310546875, 0.127777099609375, -0.1472930908203125, 0.820404052734375, 0.388336181640625, 0.4271392822265625, 1.3128662109375, 0.715850830078125, 1.521270751953125, 0.09063720703125, -0.08895111083984375, -0.274627685546875, 0.03917503356933594, -1.45440673828125, 0.39288330078125, -1.2495040893554688, 0.87548828125, 0.035037994384765625, 0.3464851379394531, 0.0, 2.079071044921875, 2.07489013671875, -0.824737548828125, -0.2358245849609375, -0.244964599609375, -0.646514892578125, -0.466278076171875, -0.130126953125, 0.1496601104736328, 0.8148193359375, -0.29961395263671875, -0.88922119140625, -0.591278076171875, 0.090576171875, -0.20556640625, -0.1995086669921875, -1.98895263671875, -0.804656982421875, 1.43359375, -1.234161376953125, 0.2048797607421875, 1.138519287109375, -0.2908935546875, 0.160400390625, -1.014739990234375, -1.12445068359375, 0.4654998779296875, 0.43939208984375, -0.008880615234375, 0.008378982543945312, -0.101226806640625, -1.04229736328125, 0.5679931640625, 0.1778411865234375, -0.222076416015625, -0.42154693603515625, 0.26709747314453125, -0.18804931640625, 0.048397064208984375, 0.21762847900390625, -0.7423553466796875, 0.5053291320800781, 1.044281005859375, -0.638916015625, 0.244659423828125, -0.335540771484375, 0.2552337646484375, -0.56353759765625, 0.22592544555664062, -1.7333984375, -0.0456085205078125, 0.7318572998046875, 1.652313232421875, -0.405242919921875, -0.2100067138671875, 0.3005218505859375, -0.404266357421875, 0.0981292724609375, -0.2973175048828125, -0.897705078125, 0.060779571533203125, 0.13820648193359375, -0.1299724578857422, 0.23322296142578125, 1.889251708984375, -0.10874557495117188, -0.462646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000003.npy"}
{"epoch": 0.0062827225130890054, "step": 4, "batch_size": 128, "mean": -0.0333935022354126, "std": 0.7856199145317078, "min": -3.5147705078125, "p10": -0.9661102294921875, "median": -0.0025806427001953125, "p90": 0.8790130615234375, "max": 2.19281005859375, "pos_frac": 0.5, "sample": [0.113433837890625, 0.244110107421875, 0.1883831024169922, 0.3968505859375, -0.853057861328125, -0.16425323486328125, -0.263397216796875, -0.71649169921875, -0.194854736328125, -0.89520263671875, 0.694976806640625, -0.0308837890625, 0.3065643310546875, 0.8671875, 0.31182098388671875, -0.6666107177734375, -0.478546142578125, 1.29522705078125, 0.89892578125, 1.03387451171875, -1.318145751953125, -1.3935546875, -0.8017578125, -0.287200927734375, 0.388397216796875, -0.017486572265625, 0.830413818359375, 0.09268569946289062, 1.87152099609375, -0.40810394287109375, 0.250579833984375, 0.1681804656982422, -0.423095703125, -0.800384521484375, 0.27667236328125, 0.6103515625, 0.36773681640625, 1.063446044921875, -0.0147857666015625, -0.710052490234375, 0.31386756896972656, -0.0977325439453125, -0.10095024108886719, -0.2679443359375, -0.737396240234375, 0.18243408203125, 0.708526611328125, -0.013423919677734375, -1.3714599609375, -0.17279052734375, -0.0748291015625, 0.6247406005859375, -1.731903076171875, 0.20635986328125, 0.3399810791015625, -0.24371337890625, 0.288116455078125, 0.28204345703125, -1.84808349609375, -0.7744140625, -0.0589599609375, -0.4230194091796875, -0.1342010498046875, -0.096588134765625, 1.2069091796875, -0.4677734375, -0.2729339599609375, 0.890655517578125, 0.4720458984375, -0.15863037109375, 0.2811126708984375, 0.50567626953125, -0.047119140625, -0.6298828125, -3.5147705078125, 0.362548828125, -0.99444580078125, -0.457794189453125, 0.396484375, 0.1920166015625, -0.4300079345703125, 0.02799224853515625, 1.2530517578125, -1.047149658203125, 1.5955963134765625, 0.399871826171875, 0.308563232421875, -0.389923095703125, 0.456512451171875, -0.438751220703125, -0.591796875, -0.98052978515625, 0.48248291015625, -0.597320556640625, 1.56597900390625, -0.3818359375, 0.31219482421875, -0.21600341796875, 0.00826263427734375, 0.2032470703125, 0.596282958984375, -0.4344635009765625, -0.285919189453125, 0.8919677734375, 0.6463241577148438, 0.140594482421875, 0.14499664306640625, 0.643402099609375, -0.02044677734375, -1.17156982421875, 0.163177490234375, -0.7862548828125, -0.8353958129882812, -1.02276611328125, 0.654754638671875, -0.959930419921875, 2.19281005859375, 0.07171630859375, 0.8740234375, -0.0463104248046875, 0.0095977783203125, -0.26715087890625, -1.8798828125, 0.2457122802734375, -1.339385986328125, 0.369354248046875, 0.26165771484375, 0.93206787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000004.npy"}
{"epoch": 0.008376963350785341, "step": 5, "batch_size": 128, "mean": -0.0680682510137558, "std": 0.8271145820617676, "min": -2.817138671875, "p10": -1.0277130126953125, "median": -0.11671829223632812, "p90": 0.9896049499511713, "max": 2.519439697265625, "pos_frac": 0.421875, "sample": [0.6903762817382812, 0.1658935546875, -0.1139678955078125, 1.75274658203125, 0.01617431640625, 0.799407958984375, -0.97772216796875, -0.349517822265625, 0.449981689453125, -1.184783935546875, -0.98577880859375, 0.17388916015625, 1.274932861328125, 0.3591461181640625, -0.12604522705078125, -2.817138671875, 0.23439788818359375, 0.936492919921875, -0.4024658203125, 0.0704345703125, -0.7808837890625, 0.27166748046875, 0.0421295166015625, -0.796630859375, 2.519439697265625, 1.801177978515625, -0.518829345703125, -0.08544158935546875, -0.7707595825195312, -0.11946868896484375, -1.1806640625, -0.22127532958984375, -0.42022705078125, -1.20965576171875, -0.1119537353515625, 0.385009765625, 0.632354736328125, -0.0547637939453125, -0.371551513671875, -0.3709716796875, 0.559814453125, -0.2083740234375, 0.202911376953125, 0.110992431640625, -0.1663818359375, -0.03230476379394531, 1.398223876953125, 0.7232818603515625, 0.3746528625488281, -0.76947021484375, 1.2403564453125, 0.2452392578125, 0.042724609375, 0.05299949645996094, 0.23838043212890625, 2.3448486328125, 0.03131103515625, -0.497528076171875, -0.7991485595703125, 0.1285247802734375, 0.55804443359375, 0.880126953125, -0.053253173828125, -0.17737579345703125, -0.6240615844726562, -0.18889617919921875, -0.382293701171875, 0.805419921875, -0.49072265625, 0.3236083984375, 0.3489990234375, -1.6115646362304688, -0.29717254638671875, -0.9738845825195312, -0.10188674926757812, 0.62139892578125, 0.2257080078125, -0.37872314453125, 0.031524658203125, -0.5006256103515625, 1.695098876953125, -1.16552734375, -0.7911376953125, 0.166778564453125, 0.19927978515625, -1.3916015625, 0.53643798828125, -1.507232666015625, 1.406524658203125, -0.5008392333984375, -1.178802490234375, 0.06818199157714844, 0.298828125, -0.23297119140625, -0.623321533203125, -1.138671875, 2.11968994140625, -0.251007080078125, -1.08917236328125, -0.348785400390625, -0.1474609375, 0.12286376953125, -0.39196014404296875, -0.748931884765625, -0.271484375, -0.639190673828125, -0.6842041015625, -0.333709716796875, 1.1135330200195312, -1.21270751953125, -1.001373291015625, -0.061553955078125, -0.3245849609375, 1.587127685546875, 0.407257080078125, -0.020626068115234375, -0.2458648681640625, -0.67822265625, -1.29803466796875, -0.23931884765625, -0.847503662109375, -0.13219451904296875, 1.11767578125, -0.3921699523925781, 0.0540771484375, -0.508636474609375, -0.0469207763671875, -1.000946044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000005.npy"}
{"epoch": 0.010471204188481676, "step": 6, "batch_size": 128, "mean": 0.08753781020641327, "std": 0.7662262916564941, "min": -2.0205230712890625, "p10": -0.6774795532226563, "median": 0.02655029296875, "p90": 1.017974853515625, "max": 3.165008544921875, "pos_frac": 0.5234375, "sample": [-2.0205230712890625, -0.551361083984375, -0.2641143798828125, 0.57061767578125, 0.760345458984375, 0.36810302734375, 1.49188232421875, 0.3556365966796875, -0.5350494384765625, -0.6757659912109375, 0.0805816650390625, -0.31982421875, 1.25787353515625, 0.32281494140625, -0.1230621337890625, 0.3756256103515625, 0.663482666015625, -0.16178131103515625, 0.3018169403076172, -0.470550537109375, -0.383270263671875, -0.2151947021484375, -0.486053466796875, 0.506256103515625, 0.40142822265625, -0.166778564453125, -0.92132568359375, 0.402740478515625, 0.235809326171875, 0.0028667449951171875, -0.07318115234375, 0.1678314208984375, -0.407928466796875, -0.594451904296875, -0.3803863525390625, -0.8621826171875, -0.3088836669921875, 0.474395751953125, 0.7080841064453125, 0.1287841796875, 0.1363067626953125, 0.214324951171875, -0.69915771484375, 0.641571044921875, 0.3548297882080078, 0.845916748046875, 0.11696243286132812, 0.09332275390625, 0.2791748046875, 1.79547119140625, 0.6236457824707031, -0.2766876220703125, -0.346588134765625, 1.8275909423828125, -0.79156494140625, 0.4741249084472656, -0.206787109375, -0.3002166748046875, -1.49908447265625, 1.3804931640625, -0.20393943786621094, 0.113006591796875, 0.4490966796875, 0.5628738403320312, 1.69732666015625, -0.5312957763671875, -0.02739715576171875, 0.2215576171875, 0.47900390625, -0.0554962158203125, -0.71441650390625, 0.02154541015625, -0.486175537109375, -1.148040771484375, -0.35040283203125, -0.3629913330078125, -0.139373779296875, 0.8178329467773438, 0.0, -0.506195068359375, 0.2334136962890625, -0.509796142578125, 0.14278411865234375, -0.40921783447265625, 0.03155517578125, -0.6088104248046875, 0.005859375, -0.9503021240234375, -0.5886001586914062, -0.34833526611328125, 1.178924560546875, -0.6768798828125, -0.1007537841796875, -1.532501220703125, 1.0455322265625, -0.14031982421875, 0.1622467041015625, 0.28333282470703125, 0.2061309814453125, 1.48309326171875, -0.150909423828125, 3.165008544921875, 1.661163330078125, 0.72802734375, 0.6417236328125, 0.04583740234375, 0.80987548828125, 0.96728515625, -0.33664703369140625, -0.05992889404296875, -0.59393310546875, -0.06083106994628906, 1.565521240234375, 0.2592926025390625, 0.36962890625, 1.12939453125, 0.53704833984375, -1.909698486328125, -0.025848388671875, 0.7850799560546875, -0.08108711242675781, -1.59674072265625, -0.6788787841796875, -0.553375244140625, 0.630645751953125, 1.00616455078125, -0.3458423614501953, 0.23403167724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000006.npy"}
{"epoch": 0.012565445026178011, "step": 7, "batch_size": 128, "mean": -0.09795580804347992, "std": 0.677897572517395, "min": -2.435302734375, "p10": -0.9408065795898437, "median": -0.06395530700683594, "p90": 0.745269775390625, "max": 2.177459716796875, "pos_frac": 0.453125, "sample": [-1.203765869140625, -0.066925048828125, -1.23419189453125, -0.777008056640625, 0.9637908935546875, -1.3431243896484375, -0.0045013427734375, -0.585113525390625, -0.2456512451171875, 0.15567779541015625, -1.303497314453125, 0.8104248046875, 0.12640380859375, 0.02497100830078125, -0.9607696533203125, -0.556365966796875, -0.5747833251953125, 0.5303192138671875, 0.0, 0.05609130859375, -0.3421630859375, -0.1358184814453125, 0.502349853515625, 0.613800048828125, 1.286865234375, -0.809906005859375, -0.7210845947265625, -0.061336517333984375, -0.8671875, 0.7642974853515625, -0.7974853515625, -0.68939208984375, 0.074462890625, 0.011016845703125, 0.482879638671875, -0.1292724609375, -0.0665740966796875, 0.7557373046875, -0.9730224609375, 1.1668777465820312, -0.2947998046875, 0.0549774169921875, -1.12432861328125, -0.86065673828125, 0.120269775390625, 0.41815185546875, -0.633758544921875, -0.07204437255859375, -0.12318801879882812, -0.639984130859375, -1.1536102294921875, -0.472015380859375, -2.435302734375, 0.3537445068359375, -0.0460205078125, 0.30139923095703125, 0.6127166748046875, -1.073150634765625, 0.2947845458984375, -0.2253570556640625, 0.027652740478515625, -0.9322509765625, -0.07927322387695312, 0.240234375, 0.4590606689453125, -0.470489501953125, 0.81085205078125, 0.19359588623046875, 0.74078369140625, -0.124664306640625, 0.5439300537109375, 2.177459716796875, 0.06640243530273438, -0.635009765625, -0.3129730224609375, 0.695098876953125, 0.019687652587890625, 0.807891845703125, 0.29620361328125, 0.484893798828125, -1.717376708984375, -0.18360519409179688, -0.4583282470703125, 0.117767333984375, -0.364837646484375, -0.09139251708984375, 0.06182861328125, 0.8553466796875, -0.618011474609375, 0.000949859619140625, -0.86737060546875, -0.016693115234375, 0.32277679443359375, -0.07026290893554688, 0.59576416015625, -0.0765838623046875, -0.2909049987792969, 0.40642356872558594, -0.2129058837890625, -0.118682861328125, -0.7846221923828125, -0.55657958984375, -0.41424560546875, 0.703369140625, -0.473907470703125, 0.00201416015625, -0.42864990234375, -0.9210205078125, -0.035430908203125, 0.6397705078125, -0.4168243408203125, 0.3855018615722656, 1.2122802734375, 0.092681884765625, 0.3706817626953125, 0.298309326171875, 0.87994384765625, -0.1301727294921875, -0.572479248046875, -1.35028076171875, 0.05401611328125, -1.241729736328125, 0.0011138916015625, 0.2519798278808594, 0.3169746398925781, 1.18017578125, -0.086456298828125, -0.6765975952148438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000007.npy"}
{"epoch": 0.014659685863874346, "step": 8, "batch_size": 128, "mean": 0.054484546184539795, "std": 0.7802243232727051, "min": -1.833831787109375, "p10": -0.835101318359375, "median": 0.00921630859375, "p90": 1.0271453857421875, "max": 2.908935546875, "pos_frac": 0.5078125, "sample": [-0.214569091796875, 0.0716400146484375, 0.11273193359375, -0.283599853515625, -0.00042724609375, 0.098388671875, -0.6026611328125, -0.4194450378417969, -0.3620109558105469, -0.942626953125, -0.030498504638671875, -0.683685302734375, -1.0078125, 0.170745849609375, 0.4341888427734375, 0.7650222778320312, 0.334625244140625, -0.45001220703125, -0.16582298278808594, 1.57305908203125, 0.51568603515625, -0.72381591796875, -0.822906494140625, 0.4649658203125, 0.3017425537109375, 1.20928955078125, -0.087066650390625, 0.029205322265625, 1.973419189453125, 1.36785888671875, 0.428375244140625, 1.4127197265625, -0.089599609375, -0.14251708984375, -0.1577606201171875, 0.020050048828125, 0.31328582763671875, 0.190826416015625, -0.783416748046875, 0.249053955078125, -1.63348388671875, -0.50482177734375, -0.415740966796875, -0.3927764892578125, 0.6248321533203125, 0.8487548828125, -0.16235733032226562, 0.6401443481445312, 2.908935546875, -0.3875732421875, -0.881439208984375, -0.9117584228515625, 1.43646240234375, -0.33919525146484375, 0.0206451416015625, 0.11627197265625, 0.3502349853515625, -0.31427001953125, 0.007781982421875, 0.699737548828125, -1.55328369140625, 0.02362060546875, -0.76287841796875, 0.6824951171875, -0.7545013427734375, 1.1705322265625, 1.15142822265625, -0.9088211059570312, 0.46697998046875, -0.126953125, -0.107696533203125, 0.67779541015625, -0.45440673828125, -0.044586181640625, -0.93963623046875, -0.863555908203125, -1.833831787109375, -0.22621726989746094, -0.3879547119140625, 1.019317626953125, 0.5433006286621094, -0.78289794921875, 0.457855224609375, -0.17243003845214844, 0.4476470947265625, -0.04390716552734375, 0.509490966796875, 0.68096923828125, 0.010650634765625, 0.98577880859375, 2.565032958984375, 0.69720458984375, -0.1033935546875, 0.40790557861328125, 0.4047279357910156, -0.12392425537109375, 1.514129638671875, 0.11798095703125, -0.034801483154296875, 1.04541015625, -0.06737136840820312, -0.03521728515625, 0.32818603515625, -1.818511962890625, 0.1413116455078125, 0.012777328491210938, 0.64117431640625, -0.0021209716796875, 0.4384765625, -1.414215087890625, 0.168243408203125, -1.443603515625, 0.32122802734375, -0.5496826171875, 1.553985595703125, 0.1914520263671875, -0.483306884765625, 0.252532958984375, -0.6395492553710938, -0.310302734375, -0.8030853271484375, -0.135467529296875, 0.35943603515625, -0.30584716796875, -0.24861907958984375, -0.4155120849609375, 0.08148193359375, 0.024562835693359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000008.npy"}
{"epoch": 0.016753926701570682, "step": 9, "batch_size": 128, "mean": 0.04120419919490814, "std": 0.749523401260376, "min": -2.317596435546875, "p10": -0.7442489624023436, "median": 0.0, "p90": 0.9752136230468749, "max": 2.0499267578125, "pos_frac": 0.484375, "sample": [0.8119430541992188, 0.49053955078125, -1.0464324951171875, 0.018402099609375, 0.681304931640625, 0.493408203125, -0.3972930908203125, -0.04698944091796875, 0.595916748046875, 0.19599151611328125, -0.2434234619140625, 1.3193359375, 1.62091064453125, -0.5660552978515625, 1.382904052734375, 0.66162109375, 1.204559326171875, -0.67120361328125, -0.37525177001953125, -0.10843658447265625, -0.14045333862304688, 2.0499267578125, -0.008159637451171875, 0.447998046875, -0.665435791015625, -0.4892425537109375, 0.02801513671875, 1.0442047119140625, -0.141021728515625, 0.14447021484375, -0.0278472900390625, -0.69854736328125, 0.7614669799804688, 0.10034942626953125, -0.9493560791015625, 0.16097259521484375, 0.776702880859375, -0.00588226318359375, -0.29305267333984375, 0.63427734375, -0.3221893310546875, -0.2345733642578125, 1.38507080078125, 0.487762451171875, -0.6533203125, 1.49346923828125, 0.126678466796875, -1.289520263671875, -0.9246826171875, -0.33739471435546875, 0.221771240234375, 0.264556884765625, 0.35845947265625, 0.4195709228515625, -1.8472900390625, -0.3271636962890625, 0.6636962890625, -1.7088623046875, -1.140167236328125, 0.91351318359375, 0.577178955078125, -0.24205780029296875, -0.42041015625, 0.40708160400390625, -1.18475341796875, -0.576171875, 1.2119293212890625, -0.577728271484375, 1.06689453125, -0.12029266357421875, 0.46514892578125, -0.01397705078125, 0.28387451171875, 0.42223358154296875, -0.230926513671875, 0.415985107421875, 0.45458984375, -0.174835205078125, -0.16304779052734375, -0.227447509765625, -0.496917724609375, -1.279571533203125, -0.327545166015625, -0.2658538818359375, -0.08245849609375, 0.119354248046875, 0.0, -0.6068878173828125, 0.434906005859375, -0.4364013671875, 1.0086669921875, -0.1808185577392578, -0.6083526611328125, 0.3520660400390625, -0.8495330810546875, 0.16121292114257812, 0.271026611328125, -0.699127197265625, -0.425506591796875, -0.114715576171875, 0.6731605529785156, 0.0, 0.910003662109375, 1.685791015625, -0.2362060546875, -0.19666671752929688, 0.0, -0.42486572265625, 0.28186798095703125, 0.29697418212890625, 0.96087646484375, -0.2569732666015625, -0.308563232421875, -2.317596435546875, -0.404937744140625, 0.7257957458496094, 0.79730224609375, 0.13873291015625, 0.517059326171875, -0.10894775390625, -0.26041412353515625, -2.04949951171875, 0.0126495361328125, -1.140869140625, 0.04636383056640625, 0.00872802734375, 1.44818115234375, 0.828857421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000009.npy"}
{"epoch": 0.018848167539267015, "step": 10, "batch_size": 128, "mean": 0.05231724679470062, "std": 0.6874677538871765, "min": -2.920867919921875, "p10": -0.6731887817382812, "median": 0.04116058349609375, "p90": 0.917737579345703, "max": 1.856597900390625, "pos_frac": 0.5234375, "sample": [-0.108154296875, -0.09771728515625, -0.779541015625, 0.9883575439453125, 0.026153564453125, -1.090087890625, -0.32010650634765625, 0.872344970703125, 0.7037811279296875, 0.5043869018554688, -0.319091796875, -0.34671592712402344, -0.4853973388671875, 0.508209228515625, 1.488494873046875, 0.18020248413085938, 0.17594146728515625, -0.4818572998046875, 1.4372406005859375, -0.03851318359375, 0.39276123046875, 0.7154159545898438, 0.64239501953125, 0.646148681640625, 1.1132049560546875, -0.09635734558105469, 0.75341796875, 0.018505096435546875, 1.20086669921875, -0.387939453125, -0.4633026123046875, -0.1408538818359375, -0.4484386444091797, 0.2081756591796875, -0.29953765869140625, -0.0980987548828125, -0.0201416015625, -0.0714111328125, 1.1004638671875, 1.020660400390625, 1.856597900390625, 0.21511459350585938, -0.12865638732910156, -0.7423095703125, 0.844146728515625, -0.1419219970703125, 0.13257980346679688, -0.28426361083984375, -0.236083984375, -0.6974334716796875, -0.021617889404296875, 0.06868743896484375, -0.09576416015625, 0.523956298828125, -0.4559173583984375, 0.5394134521484375, 0.5468292236328125, -0.660980224609375, 0.24780654907226562, -0.449127197265625, -0.5991668701171875, 0.8034896850585938, -0.841094970703125, 0.6155319213867188, -0.1454925537109375, 1.02630615234375, -0.10919189453125, 0.078033447265625, 0.1683807373046875, 0.7000732421875, -0.6262741088867188, 0.504791259765625, -2.1327667236328125, 0.1444854736328125, 0.01548004150390625, 0.108367919921875, -0.2723865509033203, 0.06109619140625, 0.61236572265625, -0.26361083984375, 0.181793212890625, -0.752288818359375, -0.110992431640625, 0.8275146484375, 0.19776153564453125, 0.3114776611328125, 0.127197265625, 0.1407318115234375, 0.1552734375, -1.242279052734375, 0.5706329345703125, -0.24048614501953125, 0.199066162109375, -0.5103759765625, -0.466461181640625, 0.279937744140625, 0.9063644409179688, 0.345306396484375, -0.6894989013671875, -1.707427978515625, -0.5266571044921875, 0.43202972412109375, 0.0561676025390625, -0.66619873046875, 1.2628173828125, 0.1353759765625, 0.24810791015625, -0.22802734375, 0.11524200439453125, -1.065765380859375, -0.01220703125, -0.03914451599121094, 0.4148712158203125, 0.0, -0.08380126953125, 1.44171142578125, -0.17804718017578125, -0.3641357421875, 0.20249366760253906, -2.920867919921875, -1.0101165771484375, -0.05789947509765625, -0.646484375, 0.4366455078125, 0.94976806640625, -0.3687286376953125, 0.1885986328125, 0.94427490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000010.npy"}
{"epoch": 0.020942408376963352, "step": 11, "batch_size": 128, "mean": -0.060673072934150696, "std": 0.766075611114502, "min": -2.0888671875, "p10": -0.9184402465820313, "median": -0.03846549987792969, "p90": 0.7583465576171875, "max": 2.9488525390625, "pos_frac": 0.484375, "sample": [0.5234375, -1.850616455078125, 0.15386962890625, -0.1058349609375, 0.6153564453125, -0.4139404296875, 0.56243896484375, 0.469207763671875, 2.9488525390625, 0.0594940185546875, 0.6890869140625, -0.3862457275390625, 0.226654052734375, -0.286041259765625, 0.2755889892578125, -0.58087158203125, 0.058563232421875, 0.050212860107421875, -0.59375, -0.0720977783203125, -0.1659221649169922, 1.0016937255859375, -0.2315673828125, 0.7763671875, -0.650787353515625, 0.749603271484375, 0.328369140625, 0.0666961669921875, -0.220855712890625, -0.5697860717773438, 0.612762451171875, 0.331298828125, -0.167938232421875, -2.0888671875, -1.2770233154296875, 0.86480712890625, 0.95562744140625, -0.22524261474609375, -0.2380523681640625, 0.0171966552734375, -0.599334716796875, -0.8494873046875, 0.5316162109375, -1.725189208984375, -0.890411376953125, -0.3309326171875, 0.37439727783203125, -0.78631591796875, 1.259033203125, 0.7655029296875, -0.05194091796875, 1.2041015625, -1.0973052978515625, 0.740478515625, 0.393951416015625, -0.4530029296875, -0.308319091796875, -0.717529296875, -0.42718505859375, -0.433319091796875, -0.3992919921875, 1.2164306640625, 0.0311279296875, -0.921356201171875, 0.71099853515625, -0.3242454528808594, -0.6825714111328125, -0.180877685546875, 0.0653533935546875, -0.09734344482421875, -1.7694091796875, -0.293792724609375, 0.194580078125, -0.2084808349609375, 0.06737136840820312, 0.755279541015625, 0.2415924072265625, 0.44571685791015625, 0.174896240234375, 2.007568359375, 0.16819000244140625, -0.05047607421875, -0.026454925537109375, 0.4823760986328125, -0.9171905517578125, 0.3395805358886719, 0.309600830078125, 0.449371337890625, -1.59136962890625, -0.238922119140625, -0.946258544921875, 0.015300750732421875, -0.8389892578125, 0.07045745849609375, -0.7627410888671875, -1.9901123046875, -0.6077880859375, 0.57733154296875, -0.07909393310546875, -0.213165283203125, -0.8568878173828125, -0.412750244140625, 0.061614990234375, -1.0181732177734375, 0.5220947265625, -1.018585205078125, 0.3006591796875, -0.903564453125, -0.635589599609375, -0.0052490234375, 0.05194091796875, -0.2954864501953125, 0.34832763671875, 0.33393096923828125, -0.5178375244140625, -0.89398193359375, 0.1942901611328125, -0.6243896484375, 0.618927001953125, 1.2133026123046875, 0.58160400390625, 0.7234344482421875, -0.5165557861328125, 1.070465087890625, -0.5670623779296875, 0.2778587341308594, -1.0447235107421875, 1.2524871826171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000011.npy"}
{"epoch": 0.023036649214659685, "step": 12, "batch_size": 128, "mean": 0.026523590087890625, "std": 0.7754247784614563, "min": -2.055145263671875, "p10": -0.9881500244140624, "median": -0.0184326171875, "p90": 1.0617874145507813, "max": 2.22052001953125, "pos_frac": 0.4921875, "sample": [0.92144775390625, 0.161468505859375, -0.23687744140625, -0.90478515625, 0.1877613067626953, 0.5687255859375, -0.003936767578125, -0.8633270263671875, 2.22052001953125, 0.47707366943359375, -0.3097381591796875, -0.4949798583984375, -1.44989013671875, 0.14946746826171875, 0.562652587890625, 0.32073974609375, -0.523406982421875, -0.6803741455078125, -0.800537109375, 0.0024566650390625, 0.664306640625, 0.91375732421875, 1.504119873046875, 0.152130126953125, 0.094024658203125, -0.3693695068359375, 0.12166595458984375, -0.5086135864257812, 0.2484130859375, -0.520355224609375, -1.518157958984375, 0.6844100952148438, -1.046661376953125, -0.826080322265625, 2.05975341796875, -0.223175048828125, -0.1409454345703125, -0.238006591796875, 0.767822265625, 0.16031646728515625, -0.12444305419921875, 0.81365966796875, -0.271209716796875, -0.032928466796875, 0.341644287109375, 0.08953094482421875, 1.190216064453125, 0.6066436767578125, 0.0908203125, -0.5858306884765625, 1.264739990234375, -0.0743865966796875, 0.2531890869140625, 0.688751220703125, 0.7702980041503906, 0.20395278930664062, -0.05670166015625, 1.22064208984375, 0.3144378662109375, 0.29998779296875, -0.383880615234375, -1.1251220703125, 0.237091064453125, 0.2560272216796875, -0.0589599609375, 1.1378021240234375, 0.4329986572265625, 0.208251953125, 1.537109375, 1.3240966796875, -0.25977325439453125, 0.1866302490234375, 0.27545166015625, -2.055145263671875, -0.186187744140625, 0.4420166015625, 0.17921829223632812, -0.3249359130859375, -0.417877197265625, -0.3118133544921875, -0.3365325927734375, -0.1276397705078125, 1.52313232421875, 0.46472930908203125, -0.21429061889648438, -0.5198974609375, 0.35388755798339844, 0.3231964111328125, 1.79541015625, 0.99383544921875, 1.024688720703125, 0.3994598388671875, -0.20943450927734375, 0.341156005859375, -1.0997314453125, 0.510986328125, -0.8109130859375, -0.092041015625, 0.504058837890625, 0.6986083984375, -1.079559326171875, -0.17395782470703125, -1.145965576171875, -0.092041015625, -1.0911865234375, -0.73974609375, -1.11163330078125, -0.354034423828125, -0.84649658203125, -1.6357421875, -1.376312255859375, -1.287841796875, -0.48438262939453125, -0.582305908203125, 1.434234619140625, -0.96307373046875, -0.2881927490234375, 0.0400390625, -0.670196533203125, -0.1350250244140625, -0.713226318359375, -0.09246826171875, 1.05596923828125, 0.1681060791015625, -0.2442474365234375, -0.0564117431640625, -0.1171417236328125, 1.0753631591796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000012.npy"}
{"epoch": 0.025130890052356022, "step": 13, "batch_size": 128, "mean": 0.1632617861032486, "std": 0.6334633231163025, "min": -1.7305908203125, "p10": -0.559869384765625, "median": 0.11257553100585938, "p90": 0.9768768310546875, "max": 2.11236572265625, "pos_frac": 0.65625, "sample": [-0.755706787109375, 0.5462131500244141, -0.23361587524414062, 0.99249267578125, 0.4315643310546875, -1.43841552734375, 2.11236572265625, -0.83074951171875, 0.7137222290039062, 0.12425994873046875, 0.8909149169921875, -0.9791259765625, 0.0931549072265625, 0.12080001831054688, 0.08224296569824219, -0.07458114624023438, 0.154571533203125, -0.1968994140625, 1.10919189453125, -1.0173263549804688, 0.4751548767089844, 0.0919189453125, 0.24334716796875, 0.374542236328125, 0.481536865234375, -0.08663368225097656, 0.38608360290527344, 0.970184326171875, -0.26869964599609375, 0.139251708984375, -0.05389404296875, -0.48360443115234375, 0.995330810546875, -0.8441162109375, 0.957000732421875, 0.10803985595703125, 0.03369331359863281, 0.073516845703125, 0.6001663208007812, 1.029510498046875, 0.09717369079589844, 1.040985107421875, -0.45068359375, 1.29852294921875, 0.263946533203125, -0.261444091796875, -0.147308349609375, -1.46270751953125, 0.18719482421875, 1.52587890625, -0.488555908203125, 0.9608154296875, 0.266265869140625, -0.54296875, -0.791015625, 0.0, 0.40283203125, 0.289154052734375, 0.06822967529296875, 0.9418106079101562, 0.31934356689453125, -0.26092529296875, -0.09110260009765625, 0.05640983581542969, 0.11492919921875, 0.559814453125, 0.582916259765625, -0.155731201171875, 0.4286346435546875, 0.1735820770263672, 0.26861572265625, -0.409271240234375, 0.016971588134765625, -0.39013671875, 1.30792236328125, -0.59930419921875, 0.08442306518554688, 0.93450927734375, 0.5396957397460938, -0.38014984130859375, 0.10535621643066406, 0.5727920532226562, 0.195068359375, 0.056182861328125, -0.30770111083984375, -0.938232421875, 0.4518775939941406, 0.05260467529296875, -0.406890869140625, 1.0971755981445312, 1.177764892578125, 0.502288818359375, 0.02349090576171875, 0.01909637451171875, 0.437164306640625, 0.11022186279296875, -0.2844581604003906, 0.0397796630859375, 0.701629638671875, -0.43896484375, -0.352813720703125, 0.49637603759765625, 0.3759918212890625, -0.11638641357421875, 1.67755126953125, -0.26025390625, -1.7305908203125, -0.04541015625, 0.255523681640625, -0.171783447265625, 0.24654388427734375, 0.703094482421875, 0.3780975341796875, -0.081298828125, 1.5103759765625, -1.04632568359375, 0.1769256591796875, 0.2528076171875, 0.614898681640625, 0.683929443359375, 0.045166015625, 0.704254150390625, 0.8113250732421875, 0.01081085205078125, 0.192901611328125, -0.03656768798828125, -0.762420654296875, -0.1681365966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000013.npy"}
{"epoch": 0.027225130890052355, "step": 14, "batch_size": 128, "mean": 0.04503123462200165, "std": 0.7652790546417236, "min": -2.7520751953125, "p10": -0.7850028991699218, "median": 0.02538299560546875, "p90": 0.8944869995117187, "max": 2.3323974609375, "pos_frac": 0.53125, "sample": [1.0322418212890625, 0.31156158447265625, 0.3787841796875, 0.75653076171875, -0.410980224609375, -0.2389373779296875, -0.3778533935546875, 1.9678955078125, 0.24110794067382812, 0.2318115234375, 0.940277099609375, 0.2867584228515625, -1.61737060546875, -1.243194580078125, 2.3323974609375, 0.4681396484375, 0.96759033203125, 1.15478515625, 0.2541465759277344, 0.5521087646484375, 0.04772186279296875, -0.33274269104003906, 0.38214111328125, 0.33960723876953125, -0.495391845703125, -1.07000732421875, 0.01947021484375, -0.25299835205078125, -2.7520751953125, -1.45599365234375, 0.020599365234375, 0.71893310546875, -0.7723388671875, -0.54296875, 0.20267486572265625, -1.149200439453125, -0.073394775390625, -0.20161056518554688, 0.36065673828125, 0.462677001953125, -0.04543113708496094, 0.01513671875, 0.9053497314453125, 0.8818359375, 0.0, 0.344329833984375, 0.25042724609375, -2.1927490234375, -0.23404693603515625, -0.1100311279296875, 0.782806396484375, 0.026214599609375, -0.0050144195556640625, 0.88983154296875, 0.7729415893554688, 0.150909423828125, -0.029937744140625, -0.12274169921875, -0.4664306640625, -0.0064697265625, 0.7611083984375, -0.41082763671875, 0.487701416015625, -0.338623046875, -1.40228271484375, 0.14776611328125, 0.494354248046875, -0.13482666015625, -0.632659912109375, -0.5247802734375, 0.447998046875, 0.38031005859375, -0.12945556640625, 1.563934326171875, 0.5460205078125, -0.09418487548828125, -0.3961181640625, -1.2447509765625, 0.742431640625, 0.38379669189453125, -0.09515762329101562, 0.1992950439453125, -0.3840179443359375, -0.11989593505859375, -0.372802734375, 0.16786956787109375, -0.3421363830566406, 0.0549774169921875, -0.7268829345703125, 0.11225700378417969, -0.2845916748046875, 0.326873779296875, -0.218902587890625, -0.193115234375, -0.050048828125, -1.31805419921875, -0.47545623779296875, -0.124847412109375, -1.046142578125, -0.661376953125, -0.054046630859375, 1.7126312255859375, -0.3199615478515625, 0.2220458984375, -0.3489837646484375, 0.797515869140625, 0.13959503173828125, 0.4605560302734375, 0.0245513916015625, 1.62774658203125, -0.1506500244140625, 0.7850799560546875, 0.40618896484375, 0.29010009765625, -0.3538837432861328, 1.40118408203125, 0.4271392822265625, 0.914215087890625, -0.3629913330078125, -0.8145523071289062, -2.0841064453125, 0.96240234375, 0.11773681640625, 0.313995361328125, 0.5482559204101562, -0.015350341796875, 0.60528564453125, 0.16705322265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000014.npy"}
{"epoch": 0.02931937172774869, "step": 15, "batch_size": 128, "mean": 0.025799661874771118, "std": 0.755837082862854, "min": -1.712799072265625, "p10": -0.89849853515625, "median": 0.024690628051757812, "p90": 0.9151580810546873, "max": 3.16754150390625, "pos_frac": 0.515625, "sample": [0.1033782958984375, 0.5335845947265625, -0.0003662109375, 3.16754150390625, 0.08538818359375, 0.6427841186523438, -1.3860321044921875, 0.59686279296875, -0.3810577392578125, 0.10929107666015625, 0.105712890625, -0.303070068359375, -0.6114654541015625, 1.0412139892578125, -0.11008453369140625, -0.84857177734375, -1.4776153564453125, 0.04377555847167969, 0.1642608642578125, 0.201141357421875, -1.10491943359375, -0.32232666015625, -0.001953125, -0.498504638671875, 0.5331954956054688, -0.375396728515625, -0.8734130859375, -0.92584228515625, 1.229278564453125, 0.21984100341796875, -0.1224365234375, -0.78369140625, -0.5859375, 0.27264404296875, -0.40252685546875, -0.040653228759765625, -0.42401123046875, -0.33734130859375, 1.3646240234375, 0.5591354370117188, 0.5398483276367188, -1.019439697265625, 0.07354736328125, -0.5777206420898438, 0.4091339111328125, 0.009490966796875, 0.019222259521484375, 0.0955352783203125, -0.4568023681640625, -0.38043212890625, 0.7771644592285156, -0.33956146240234375, -1.712799072265625, 0.52154541015625, -0.88677978515625, 0.352874755859375, -0.797943115234375, 0.41448974609375, -0.25610923767089844, 0.656982421875, -0.71826171875, -0.13060569763183594, 1.288818359375, 0.22406578063964844, 0.5344390869140625, 0.8495712280273438, 1.984771728515625, -0.3031005859375, -0.834625244140625, 0.89801025390625, 0.5266189575195312, 0.034637451171875, -0.812408447265625, 0.039764404296875, 0.423492431640625, -0.4965057373046875, -0.40673828125, 0.16486358642578125, -0.34649658203125, 1.095703125, 0.955169677734375, 0.05769157409667969, 1.3411865234375, -0.835174560546875, -0.2484130859375, 1.063873291015625, 0.454803466796875, 0.72650146484375, -0.07814407348632812, -1.6728515625, -0.14089584350585938, -0.06201171875, -1.037689208984375, 0.5769195556640625, -0.10748291015625, -0.270751953125, -0.15747642517089844, 1.77301025390625, -0.59912109375, 0.20458984375, -1.16290283203125, -1.086700439453125, 1.012451171875, -1.0775299072265625, 0.851593017578125, 0.438568115234375, 0.75091552734375, -0.4854393005371094, 0.136688232421875, -0.06413650512695312, -0.09076690673828125, 0.5257492065429688, 0.03015899658203125, 0.2097015380859375, -1.08209228515625, 0.3349876403808594, 1.4898681640625, -0.17633819580078125, 0.22737884521484375, 0.83465576171875, -1.089080810546875, 0.14789962768554688, 0.03515625, 0.637847900390625, -0.316375732421875, 0.53369140625, -0.19959259033203125, -0.522430419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000015.npy"}
{"epoch": 0.031413612565445025, "step": 16, "batch_size": 128, "mean": 0.019194379448890686, "std": 0.8540195226669312, "min": -3.5052490234375, "p10": -0.946685791015625, "median": 0.06932258605957031, "p90": 1.037422180175781, "max": 2.1101531982421875, "pos_frac": 0.5390625, "sample": [0.672332763671875, 0.0436553955078125, 0.6255264282226562, 0.38250732421875, -1.262115478515625, 0.1055908203125, 0.258514404296875, 0.505523681640625, -2.03839111328125, -1.53607177734375, -0.3883857727050781, 0.1760406494140625, 0.35144805908203125, -0.131317138671875, -0.9014892578125, 0.4662628173828125, -2.337646484375, -0.59649658203125, -1.041290283203125, -0.4908599853515625, -0.3144989013671875, 0.7628326416015625, 0.40985107421875, 0.32232666015625, 0.16815185546875, -0.4786224365234375, -1.700836181640625, 1.675201416015625, -0.0947265625, 0.86395263671875, -0.3194580078125, -0.644287109375, -0.910125732421875, 0.3702392578125, 0.490081787109375, -0.06494140625, -0.9468994140625, -0.19304656982421875, -0.49905967712402344, 0.497589111328125, 0.5280609130859375, -2.0952911376953125, 1.177398681640625, 0.10150146484375, 0.5445404052734375, -0.10417938232421875, -0.51519775390625, 1.0071868896484375, 0.88812255859375, 0.038311004638671875, 0.07267379760742188, 0.85125732421875, -0.5982589721679688, -0.7491455078125, 0.10723876953125, 1.178436279296875, -0.078125, -0.5873565673828125, 0.1417083740234375, -1.2490234375, 1.1961669921875, 0.06597137451171875, -0.307373046875, -1.715850830078125, 0.4456329345703125, 0.6517333984375, -0.09696578979492188, 1.10797119140625, 0.8992919921875, -0.199371337890625, -0.14746856689453125, 0.798065185546875, -0.5997772216796875, -0.4002685546875, 0.28253173828125, 0.6004180908203125, -0.482147216796875, -0.0322418212890625, 0.570587158203125, 0.819610595703125, 0.3248291015625, 0.18828201293945312, 0.641815185546875, 2.1101531982421875, -0.04132080078125, -0.330322265625, 0.0, 0.102142333984375, 1.404266357421875, 1.997314453125, -0.01158905029296875, 0.9954833984375, -0.391998291015625, -0.366455078125, 0.23675537109375, 1.278076171875, 0.007568359375, -0.4772796630859375, -0.12066650390625, 0.3347587585449219, -0.0040283203125, 1.17034912109375, -0.94659423828125, -0.6545867919921875, 0.371734619140625, -0.0451507568359375, 0.014251708984375, 0.3996124267578125, -0.9951171875, -0.566619873046875, 0.377593994140625, 0.1068115234375, 1.5218963623046875, 0.437225341796875, -1.7347412109375, 0.407318115234375, -0.265289306640625, 0.3819732666015625, -0.195037841796875, 0.2162933349609375, -0.628265380859375, 0.353851318359375, 0.4188232421875, -0.768524169921875, -3.5052490234375, -0.018798828125, 1.147979736328125, 1.20391845703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000016.npy"}
{"epoch": 0.033507853403141365, "step": 17, "batch_size": 128, "mean": -0.06319691240787506, "std": 0.723469614982605, "min": -2.378173828125, "p10": -0.8288223266601562, "median": -0.049427032470703125, "p90": 0.752978515625, "max": 2.2659912109375, "pos_frac": 0.4296875, "sample": [-0.8130416870117188, -0.701171875, 1.6234130859375, -0.8252105712890625, 0.0325469970703125, -0.0304718017578125, 1.657470703125, 0.49078369140625, 0.509002685546875, 0.069000244140625, -0.16231155395507812, -1.11456298828125, -0.689483642578125, -1.297698974609375, -1.838134765625, -0.36922454833984375, -0.684295654296875, -1.452392578125, -0.458709716796875, -0.05279541015625, -0.027587890625, 0.0753021240234375, -0.635498046875, -0.2306365966796875, 0.028383255004882812, -0.091552734375, 0.760498046875, -0.342376708984375, 1.4886474609375, -0.1461181640625, -0.20733642578125, -2.378173828125, -0.25304412841796875, 0.92022705078125, -0.420135498046875, -0.41295623779296875, -0.4637451171875, -0.666351318359375, 0.1102142333984375, 0.32921600341796875, 0.16851806640625, -1.418731689453125, 0.30556488037109375, -0.6377372741699219, -0.017303466796875, -0.696685791015625, -0.04605865478515625, -0.13428115844726562, -1.152130126953125, 0.6371078491210938, -0.241180419921875, -0.3483734130859375, 0.4375457763671875, 1.764312744140625, -0.22075653076171875, -0.8993988037109375, 0.4762115478515625, 0.274200439453125, 0.39289093017578125, -0.200592041015625, -0.4251251220703125, 0.32611083984375, -0.3550872802734375, 0.043792724609375, -1.583465576171875, -0.0760955810546875, 0.4808349609375, 0.5827217102050781, 0.00249481201171875, 0.13153076171875, 0.246002197265625, 0.1013641357421875, 1.39373779296875, 0.8858795166015625, 0.38604736328125, -0.328369140625, -0.488037109375, -0.0058746337890625, -0.2221832275390625, -0.1945343017578125, 1.6060791015625, -0.0048065185546875, -0.9152679443359375, -0.494659423828125, 0.26384735107421875, 0.32373046875, -0.557861328125, -0.340576171875, 0.749755859375, 0.476593017578125, -0.15797042846679688, 0.7438125610351562, -0.2808837890625, 1.148529052734375, -0.837249755859375, 0.084320068359375, -0.28191375732421875, 0.04364013671875, -0.09776687622070312, 1.039398193359375, -0.353271484375, -0.60595703125, -0.02227783203125, -0.62786865234375, 0.11362838745117188, 0.15423583984375, -0.026828765869140625, -0.5092620849609375, -0.1436767578125, 0.32159423828125, -0.22666549682617188, -0.119873046875, -1.6365966796875, 0.11773681640625, 0.949615478515625, -0.038848876953125, -0.4608154296875, -0.77001953125, 0.23809814453125, 0.08294677734375, 0.26891326904296875, 0.00637054443359375, -1.336212158203125, -0.07318878173828125, 2.2659912109375, 0.610107421875, 0.5261001586914062, 0.021514892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000017.npy"}
{"epoch": 0.0356020942408377, "step": 18, "batch_size": 128, "mean": 0.14981183409690857, "std": 0.7894093990325928, "min": -3.361297607421875, "p10": -0.591162109375, "median": 0.1528778076171875, "p90": 1.0868743896484367, "max": 2.2703704833984375, "pos_frac": 0.6328125, "sample": [-0.3795013427734375, 0.08953857421875, 0.2612762451171875, -0.0168914794921875, -0.0807647705078125, 0.02423095703125, 0.057521820068359375, -0.2552680969238281, -0.007610321044921875, 0.1881103515625, 1.261810302734375, 0.799713134765625, 0.377532958984375, 1.01190185546875, 0.1075897216796875, 0.2316741943359375, -2.09423828125, 1.339599609375, -0.103912353515625, 1.6218719482421875, -0.555999755859375, 2.0650787353515625, 0.906158447265625, -1.3742523193359375, 1.71551513671875, 0.11182212829589844, 2.2703704833984375, 0.3958282470703125, -0.13897705078125, -0.5582351684570312, -1.040283203125, -0.725006103515625, 0.19126129150390625, 0.308197021484375, 0.0472564697265625, 0.32714080810546875, -0.15338134765625, 0.1868152618408203, -0.1331634521484375, 0.74127197265625, -0.00634002685546875, 1.8941650390625, 1.517578125, -3.361297607421875, 0.16750335693359375, 0.40343475341796875, -0.222930908203125, 0.0461883544921875, 0.7628936767578125, -0.06256103515625, 0.1353607177734375, 0.224212646484375, 0.2457733154296875, 0.292327880859375, 0.17266845703125, 0.75677490234375, 0.638427734375, 0.1737060546875, -0.714752197265625, 0.029144287109375, -0.597015380859375, 0.00588226318359375, -0.15948486328125, 0.218963623046875, -0.20489501953125, 0.503662109375, 0.0360107421875, 0.406219482421875, 1.77264404296875, 1.5, 0.2862281799316406, -0.25247955322265625, -0.11639404296875, 0.150390625, 0.019256591796875, 0.033252716064453125, -0.438140869140625, 0.155364990234375, 0.0711212158203125, -1.486968994140625, 0.43780517578125, 0.08260345458984375, -0.588653564453125, 0.3668670654296875, 0.51629638671875, -0.5045242309570312, 0.24355316162109375, -1.058837890625, 0.082550048828125, -0.12051773071289062, 1.61651611328125, 0.5852508544921875, -0.3192901611328125, 0.6603012084960938, -0.28662109375, 0.93603515625, -0.2606391906738281, -0.08945465087890625, 0.3962860107421875, 0.2252960205078125, -0.1065826416015625, 0.48577880859375, -0.15887451171875, 1.39862060546875, 0.44451141357421875, 0.343414306640625, -0.2196044921875, 0.706207275390625, -0.22353363037109375, -0.697509765625, 0.2154541015625, 0.6882781982421875, 0.570068359375, 0.2684326171875, -0.39337158203125, 0.658843994140625, 0.2140655517578125, 0.331207275390625, 0.67364501953125, -1.6361083984375, 0.29254150390625, -0.3961029052734375, -0.2639923095703125, -1.495086669921875, 1.5513916015625, -0.21612548828125, 0.62689208984375, -1.42486572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000018.npy"}
{"epoch": 0.03769633507853403, "step": 19, "batch_size": 128, "mean": 0.013604193925857544, "std": 0.5980666875839233, "min": -1.97393798828125, "p10": -0.7484100341796874, "median": 0.054012298583984375, "p90": 0.6295509338378905, "max": 1.4740447998046875, "pos_frac": 0.546875, "sample": [0.5860824584960938, 0.913421630859375, -0.9029693603515625, -0.6839599609375, 0.266937255859375, 1.2789306640625, 0.469940185546875, 0.278778076171875, -0.0932769775390625, -1.81854248046875, -0.3596763610839844, -0.609619140625, 0.5537261962890625, -1.97393798828125, 0.4253501892089844, -0.035888671875, 0.3377685546875, 0.52899169921875, -0.4570770263671875, 0.1742095947265625, 0.29280853271484375, -0.1680908203125, 0.08758926391601562, -0.70904541015625, 0.4868316650390625, 0.5665130615234375, 0.11161231994628906, 0.592437744140625, 0.26605224609375, 0.020503997802734375, 0.09423828125, 0.063690185546875, 0.02154541015625, -0.3851318359375, 0.4075775146484375, 0.5610618591308594, -0.12176513671875, -0.239013671875, 0.090057373046875, 0.3326911926269531, -0.740692138671875, 0.3789634704589844, -0.6339111328125, -0.5111541748046875, -1.1273193359375, 0.26416015625, 0.23352432250976562, -0.0271453857421875, 0.12506103515625, 0.6213912963867188, 0.1830291748046875, -0.76641845703125, 0.02044677734375, 0.896209716796875, 0.4372100830078125, -0.12933349609375, -0.361572265625, 0.30078125, 0.13494110107421875, -0.16613006591796875, -0.04210662841796875, 0.0074462890625, 0.092193603515625, 0.5499000549316406, -0.28985595703125, 0.78009033203125, -0.1974945068359375, -0.0382080078125, 1.2522430419921875, 0.2704925537109375, 0.343292236328125, 1.28887939453125, 0.173919677734375, -0.09136962890625, -0.453704833984375, -1.004241943359375, 0.0540771484375, 0.17559814453125, -0.4331512451171875, -0.0349578857421875, 0.368865966796875, -0.52545166015625, 0.2500591278076172, -0.137054443359375, 0.50982666015625, -0.06281280517578125, 0.1477069854736328, 1.46539306640625, 0.05394744873046875, 0.648590087890625, -0.86114501953125, 0.4517822265625, 0.2262115478515625, -1.258697509765625, 0.25628662109375, -0.436309814453125, -1.048736572265625, -0.35973358154296875, -0.36103057861328125, 0.39769554138183594, -0.1504974365234375, -0.388427734375, 1.4740447998046875, -0.9129638671875, -0.36688232421875, 0.46087646484375, 0.90069580078125, -0.202880859375, 0.29364013671875, -0.4093780517578125, -0.4987335205078125, 0.0194091796875, -0.881683349609375, -0.0501556396484375, -0.34283447265625, 0.2552642822265625, -0.498626708984375, -0.52349853515625, -0.853424072265625, -0.3644866943359375, -0.4138946533203125, 1.289825439453125, 0.9560546875, 0.38739013671875, -0.03326416015625, -0.7968597412109375, 0.7599945068359375, 0.1208038330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000019.npy"}
{"epoch": 0.039790575916230364, "step": 20, "batch_size": 128, "mean": 0.10843226313591003, "std": 0.760266125202179, "min": -1.94183349609375, "p10": -0.8086296081542969, "median": 0.08954429626464844, "p90": 1.0796722412109372, "max": 2.446258544921875, "pos_frac": 0.5625, "sample": [0.888946533203125, 0.503326416015625, -0.20819091796875, -0.7413330078125, 0.707611083984375, 1.477630615234375, 0.1004180908203125, 0.1240234375, 0.41986083984375, -1.0234375, 0.498809814453125, 0.070068359375, 0.2366943359375, 2.099639892578125, 0.7174072265625, -0.1628265380859375, -0.354400634765625, 1.3369140625, 0.08817672729492188, -0.05765533447265625, -0.9569854736328125, -0.13653564453125, -0.812469482421875, 0.4881172180175781, -0.1700000762939453, 0.207427978515625, 0.9412841796875, -0.798187255859375, -0.3571624755859375, 0.22613525390625, 1.167572021484375, 0.10630989074707031, 0.16306686401367188, -1.74847412109375, 0.086151123046875, -0.366851806640625, -0.841583251953125, 0.062103271484375, 0.416534423828125, 1.9261016845703125, 0.6134033203125, 0.1297607421875, 1.30108642578125, 0.010288238525390625, 0.6332321166992188, 0.244476318359375, -0.0167694091796875, 0.24334716796875, -0.9519195556640625, 0.2031402587890625, 0.4670867919921875, -0.04077911376953125, 0.0, -0.346832275390625, 1.380859375, 0.17865753173828125, -0.10322952270507812, -0.79644775390625, -1.1716766357421875, -0.2030181884765625, -0.20463943481445312, 0.35821533203125, -0.492919921875, -1.94183349609375, 1.0579833984375, 1.130279541015625, 0.25177001953125, 0.29540252685546875, -0.2073211669921875, 0.67645263671875, -0.9647216796875, -0.33419227600097656, 0.004547119140625, -0.8069839477539062, -0.835845947265625, -0.3978271484375, 0.011026382446289062, 0.411529541015625, -0.5874099731445312, 0.43265533447265625, 0.774871826171875, 1.1309814453125, 0.0050506591796875, -0.10763168334960938, 0.554931640625, -0.5090408325195312, -0.0338134765625, 1.5609130859375, -0.03273773193359375, 0.090911865234375, 0.5042495727539062, 0.0980224609375, 0.68084716796875, -0.217742919921875, -0.2398834228515625, -0.206512451171875, -0.8590850830078125, 0.7022476196289062, 0.2423858642578125, -0.0045909881591796875, 0.8907470703125, -1.6669921875, -0.4812164306640625, 0.15900039672851562, 2.446258544921875, 0.108367919921875, 0.60833740234375, 0.3341064453125, -0.5045547485351562, 0.330322265625, -0.3299560546875, -0.1125335693359375, 1.4069747924804688, 0.83526611328125, -1.762298583984375, -0.24428558349609375, 2.367645263671875, 0.56793212890625, 0.411834716796875, 0.2861137390136719, 0.400421142578125, -0.25372314453125, -0.33441162109375, -0.55474853515625, -0.4092884063720703, 0.39363861083984375, -0.3295135498046875, -0.7735595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000020.npy"}
{"epoch": 0.041884816753926704, "step": 21, "batch_size": 128, "mean": 0.13965462148189545, "std": 0.7703147530555725, "min": -2.09124755859375, "p10": -0.8499786376953125, "median": 0.18199920654296875, "p90": 1.0477737426757812, "max": 2.06683349609375, "pos_frac": 0.6015625, "sample": [-0.76617431640625, -0.28542327880859375, -1.42022705078125, 0.544830322265625, 1.420562744140625, -0.386474609375, -0.1951904296875, -0.10833740234375, -0.248138427734375, 0.0551605224609375, 0.35223388671875, -1.968597412109375, -0.0495452880859375, 1.348846435546875, -0.89190673828125, 0.5119552612304688, -1.170684814453125, 0.015777587890625, -0.2098388671875, -0.40260887145996094, 1.2249755859375, -0.10752487182617188, -2.09124755859375, -1.109527587890625, 0.912811279296875, -0.807861328125, -0.205841064453125, -0.89385986328125, -0.5865631103515625, 0.287872314453125, -0.1514434814453125, -0.30731201171875, 0.817779541015625, 0.9395828247070312, 0.061489105224609375, 0.30104827880859375, 0.6802177429199219, 0.280853271484375, -0.480194091796875, 1.0833740234375, 0.8361434936523438, -0.412445068359375, -0.429901123046875, 0.8617362976074219, 0.079803466796875, 1.0731201171875, 0.21604156494140625, 0.009552001953125, 0.7472457885742188, 0.8899993896484375, -0.8468017578125, -0.30255126953125, 0.0, 0.2748870849609375, 0.292388916015625, -0.08234786987304688, -1.0833740234375, 0.977752685546875, -0.3232421875, 0.16381072998046875, 0.507568359375, -0.857391357421875, -0.14211654663085938, 1.48944091796875, 0.49755859375, 0.87176513671875, -1.5986328125, -0.5194854736328125, -0.824005126953125, 0.707489013671875, 1.10498046875, 0.6256256103515625, 0.697021484375, 2.06683349609375, -0.9577789306640625, 0.1173095703125, 1.464996337890625, -0.162841796875, 0.62725830078125, 0.4346923828125, 0.34226226806640625, 0.19403457641601562, 1.7403564453125, -0.5925140380859375, 0.593505859375, 0.722747802734375, -0.12580490112304688, 0.1836090087890625, 0.680328369140625, 0.3349418640136719, -0.82403564453125, 0.5576553344726562, 0.51348876953125, 0.3325347900390625, 0.141021728515625, 0.1699676513671875, 0.32733917236328125, -0.0684814453125, 0.6319580078125, 1.437896728515625, -0.7066650390625, 0.9259185791015625, 0.88824462890625, 0.510406494140625, 0.84375, -1.19378662109375, 0.05617523193359375, 0.148834228515625, 0.6741867065429688, 1.0369110107421875, -0.0962982177734375, -0.8465805053710938, 1.8343505859375, 1.5157470703125, 0.40839385986328125, -0.826202392578125, 0.544830322265625, 0.180389404296875, 0.642425537109375, 0.3275146484375, -0.9608154296875, -0.564666748046875, 0.35364532470703125, 0.003570556640625, -0.322540283203125, 0.2547454833984375, 0.43378448486328125, -0.5702438354492188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000021.npy"}
{"epoch": 0.04397905759162304, "step": 22, "batch_size": 128, "mean": 0.052607640624046326, "std": 0.7412273287773132, "min": -2.17352294921875, "p10": -0.7310211181640625, "median": -0.015287399291992188, "p90": 0.9285797119140625, "max": 2.85357666015625, "pos_frac": 0.46875, "sample": [-0.378204345703125, -0.21165084838867188, -0.5393753051757812, 0.5003662109375, 1.309967041015625, 0.06803131103515625, 0.1131439208984375, -0.3914375305175781, -0.22137451171875, -0.106292724609375, 0.927764892578125, -0.05323028564453125, -0.317779541015625, 0.26915740966796875, -0.4716339111328125, 2.85357666015625, -1.69830322265625, 0.03790283203125, 0.455169677734375, -0.1773834228515625, 0.3097991943359375, -0.233978271484375, 0.113372802734375, 0.07756805419921875, 0.17486572265625, -0.0682373046875, -2.17352294921875, -0.2098388671875, 0.7187042236328125, 0.4939117431640625, -0.810791015625, 0.079254150390625, 0.0, 0.68359375, 0.29576873779296875, -1.5845947265625, -0.742279052734375, 1.47967529296875, -0.29099273681640625, 1.4186859130859375, 0.22509765625, -0.006317138671875, -0.430816650390625, 0.93048095703125, 0.46836090087890625, -0.4351654052734375, 0.34112548828125, 0.1291656494140625, -0.73077392578125, -0.536773681640625, -0.1333942413330078, -0.5697708129882812, 0.546356201171875, 0.0927886962890625, 0.12209320068359375, 0.5941238403320312, 0.3504981994628906, 1.1640625, 0.91363525390625, 1.4010009765625, 2.12786865234375, -0.884185791015625, 0.69830322265625, 0.7695693969726562, -0.15823745727539062, -0.2597618103027344, -0.05621337890625, -0.48388671875, 0.3047027587890625, -0.022735595703125, -0.1592693328857422, -0.10205078125, 0.21445465087890625, -0.004364013671875, -0.731597900390625, 0.7398834228515625, 0.319610595703125, -0.17908477783203125, -0.57489013671875, -0.18994140625, 0.5962371826171875, -0.11267852783203125, -0.1361236572265625, 0.26922035217285156, 0.8815765380859375, -0.012603759765625, -0.3759002685546875, -0.0181427001953125, 0.9759521484375, -0.284881591796875, 1.0980987548828125, 0.1290283203125, -0.7216949462890625, -0.05844879150390625, 0.7958984375, -0.42955780029296875, -1.986785888671875, 0.0804901123046875, -0.07403564453125, -0.034149169921875, -0.8458251953125, -0.1558971405029297, -0.225341796875, 0.917236328125, -0.60113525390625, 0.0982208251953125, 0.77288818359375, 1.072418212890625, -1.916473388671875, -0.8539276123046875, 0.24893569946289062, 0.455078125, 0.1365966796875, 0.22737884521484375, 0.4281597137451172, -0.216552734375, -0.3212127685546875, -1.121826171875, 1.046722412109375, 1.706512451171875, -0.1444854736328125, -0.59832763671875, -0.08539581298828125, -0.5096282958984375, 0.415008544921875, -0.246368408203125, -0.017971038818359375, -1.015838623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000022.npy"}
{"epoch": 0.04607329842931937, "step": 23, "batch_size": 128, "mean": 0.11437398195266724, "std": 0.7655512094497681, "min": -2.972320556640625, "p10": -0.6565223693847655, "median": 0.136138916015625, "p90": 1.0035339355468749, "max": 2.300048828125, "pos_frac": 0.5859375, "sample": [0.01883697509765625, 1.0784149169921875, 0.134521484375, 0.6824493408203125, -0.3079833984375, 2.300048828125, 0.09259033203125, 0.6963348388671875, -0.4849395751953125, -2.62469482421875, 1.1846923828125, -1.247222900390625, 0.36296844482421875, 1.160797119140625, 2.0238037109375, 0.144012451171875, 0.98846435546875, -0.913543701171875, -0.16114234924316406, 0.5157318115234375, 0.3236236572265625, -1.2171630859375, 0.238525390625, -0.1744384765625, 1.3658447265625, -0.1255950927734375, 0.68768310546875, -0.14117431640625, -0.38610076904296875, -0.28183746337890625, 0.0, -0.8927841186523438, 0.3674468994140625, -0.7438888549804688, 0.14300537109375, 0.3890838623046875, -0.33037567138671875, 0.58123779296875, 0.675750732421875, -0.61907958984375, 0.211212158203125, 0.729339599609375, 0.71246337890625, 0.18995285034179688, -0.3670654296875, 1.3975982666015625, 0.0882415771484375, -0.31790924072265625, 0.73028564453125, -0.122772216796875, -0.16096115112304688, -0.017208099365234375, -0.128570556640625, -0.393280029296875, 0.807708740234375, -1.514678955078125, 0.4202880859375, -0.0654296875, -0.2091217041015625, 0.616851806640625, 0.5906982421875, 0.8480873107910156, -0.51947021484375, -0.24996185302734375, -0.4069671630859375, -0.236663818359375, -0.3349609375, 0.2349853515625, 0.0087738037109375, -0.957794189453125, 1.081268310546875, 0.300750732421875, 0.06884765625, -0.41009521484375, 1.153167724609375, -0.0037384033203125, 2.2850341796875, 0.2105541229248047, 1.18634033203125, -1.6887969970703125, 0.463409423828125, 0.15320968627929688, -0.48055267333984375, 0.401458740234375, 0.276275634765625, 0.108489990234375, 0.56805419921875, 0.140777587890625, 0.458984375, 0.13775634765625, -0.43720245361328125, -0.23776626586914062, -0.7525634765625, -0.05120086669921875, -0.6126556396484375, 0.8455810546875, 0.0889892578125, 1.24761962890625, 0.25139617919921875, 1.0386962890625, -0.509674072265625, 0.6556015014648438, -0.05803680419921875, -0.34789276123046875, 0.007965087890625, 0.574462890625, 0.20513916015625, 0.1721343994140625, 0.1419525146484375, -0.752777099609375, 0.5798377990722656, -0.175323486328125, -0.023876190185546875, 0.287841796875, 0.0664825439453125, 0.385589599609375, 0.70513916015625, -0.1821746826171875, 0.07305908203125, -1.407470703125, 0.536224365234375, -0.42043304443359375, 0.5449371337890625, 0.663177490234375, -2.972320556640625, 0.48687744140625, 0.6597137451171875, -0.13595199584960938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000023.npy"}
{"epoch": 0.048167539267015703, "step": 24, "batch_size": 128, "mean": 0.23063026368618011, "std": 0.8571506142616272, "min": -2.36419677734375, "p10": -0.7006317138671875, "median": 0.19688129425048828, "p90": 1.2109191894531248, "max": 3.35443115234375, "pos_frac": 0.6171875, "sample": [0.3768577575683594, 0.8895263671875, 0.677459716796875, -0.5174407958984375, 0.89422607421875, 0.3233642578125, -0.47216796875, -0.248565673828125, 0.3213539123535156, 1.165496826171875, -0.0592193603515625, 0.3393096923828125, 0.42919921875, 0.4557228088378906, 0.05852508544921875, -0.0501708984375, -0.6732635498046875, 3.35443115234375, -0.025787353515625, 0.062297821044921875, 0.21649169921875, 0.309906005859375, 0.376007080078125, 0.0, -0.12076568603515625, 0.017406463623046875, 1.517181396484375, 0.0002288818359375, -0.5600738525390625, 0.94012451171875, -1.01837158203125, 0.21424102783203125, 1.3250732421875, 0.3057861328125, 0.42205810546875, -0.00421142578125, 1.371826171875, 2.06732177734375, 0.1342315673828125, 1.2822265625, -0.20172119140625, 0.504608154296875, -0.09651947021484375, -2.36419677734375, -0.9056396484375, 0.7018966674804688, 0.0, 0.356170654296875, 0.887939453125, 0.6883544921875, 0.306243896484375, 0.28429412841796875, 0.9348602294921875, -0.0828704833984375, 0.69354248046875, 0.565887451171875, 0.029937744140625, -0.20550155639648438, -0.95709228515625, 0.061431884765625, -1.0892333984375, -0.129241943359375, 0.8189697265625, 0.8029632568359375, -1.2928466796875, -0.24810791015625, -0.3629608154296875, -0.31330108642578125, -0.208099365234375, -0.509521484375, -0.4268798828125, 0.224273681640625, -0.97003173828125, 0.56231689453125, 0.07916259765625, -0.726287841796875, -0.606689453125, 2.949951171875, 0.65313720703125, 0.7795867919921875, -0.68963623046875, -0.37841796875, 0.7611083984375, 0.183990478515625, 0.5677032470703125, -0.641265869140625, 0.39874267578125, -1.231781005859375, 1.455322265625, 0.2635040283203125, 0.771820068359375, 0.3465576171875, 1.43597412109375, -0.8047027587890625, 0.062835693359375, 0.1946582794189453, -0.684295654296875, 0.0370941162109375, 0.930084228515625, 1.07281494140625, -1.36376953125, -0.01348876953125, -0.004486083984375, 0.859161376953125, -1.103759765625, 1.08905029296875, 0.4341583251953125, 2.56243896484375, 0.5410614013671875, 1.489288330078125, 1.18035888671875, -0.6609649658203125, -0.1935272216796875, 1.85821533203125, 0.19910430908203125, -0.28631591796875, -0.5277099609375, 0.39740753173828125, 0.1414794921875, 0.3060150146484375, 2.05889892578125, 0.598876953125, 0.18501663208007812, -0.022613525390625, -2.0584716796875, 0.11653900146484375, 0.55596923828125, -0.15599822998046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000024.npy"}
{"epoch": 0.050261780104712044, "step": 25, "batch_size": 128, "mean": 0.15958915650844574, "std": 0.788824737071991, "min": -2.7294921875, "p10": -0.6211746215820313, "median": 0.12298202514648438, "p90": 0.8641777038574219, "max": 2.34600830078125, "pos_frac": 0.609375, "sample": [-0.1775970458984375, 0.700531005859375, 0.716094970703125, 0.4403228759765625, -0.676788330078125, 2.23577880859375, 0.732574462890625, -0.064605712890625, 2.34600830078125, -0.86431884765625, 0.7430877685546875, -0.744598388671875, -0.3628692626953125, -0.406768798828125, 0.0403289794921875, -0.625885009765625, -0.5122451782226562, 0.00284576416015625, 1.630126953125, 0.124542236328125, -0.22222900390625, -0.011411666870117188, 0.6269912719726562, 0.648345947265625, 0.8637542724609375, 0.0189208984375, -0.5588836669921875, -0.6027450561523438, -0.2966117858886719, 0.668212890625, 0.0802764892578125, 0.8651657104492188, 0.0, -0.344207763671875, -0.065704345703125, 0.6678581237792969, 0.6116943359375, 0.2326202392578125, -2.08807373046875, 1.0755615234375, -1.55108642578125, 0.0, 1.0728759765625, -0.4935302734375, -0.5706939697265625, 0.30084228515625, -0.8906326293945312, 0.31103515625, -0.21414947509765625, 0.4855499267578125, -0.0818328857421875, 0.2675323486328125, 0.5486602783203125, -0.20248794555664062, -0.543853759765625, -0.0876617431640625, -0.6304855346679688, 0.580902099609375, 0.019439697265625, -0.3179168701171875, 0.9422225952148438, 0.14203643798828125, 0.5096435546875, 2.1734619140625, 0.4246978759765625, -0.02812957763671875, -0.4534912109375, 0.238555908203125, 0.2764129638671875, 0.214599609375, 1.2985687255859375, 1.4625091552734375, 0.7761459350585938, -2.6424560546875, 0.0771484375, 0.0205078125, 0.06195068359375, 0.05695533752441406, -0.00347900390625, 0.478485107421875, -0.8948822021484375, 0.6417999267578125, -0.1034393310546875, -0.166717529296875, 0.1369171142578125, 2.076263427734375, 0.07148361206054688, 0.2036571502685547, -0.188751220703125, 0.632659912109375, 0.095001220703125, 0.5947341918945312, 0.2761077880859375, 0.839447021484375, 0.95281982421875, -0.34930419921875, -0.021877288818359375, 0.80487060546875, -0.6191558837890625, 0.7018890380859375, -0.29973602294921875, 0.123870849609375, 0.2481842041015625, -0.3104228973388672, -0.0284423828125, 2.06646728515625, -0.36822509765625, 0.637908935546875, 0.8333740234375, -1.560791015625, 0.4865264892578125, 0.845001220703125, 0.2283935546875, 0.6608695983886719, -0.633392333984375, 0.33599853515625, 0.76629638671875, 0.21550750732421875, 0.748992919921875, 0.12209320068359375, -0.33587646484375, 0.0947265625, -2.7294921875, -0.3100242614746094, 0.6356201171875, 0.565521240234375, 0.124725341796875, 0.10526084899902344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000025.npy"}
{"epoch": 0.05235602094240838, "step": 26, "batch_size": 128, "mean": 0.17617014050483704, "std": 0.8798357248306274, "min": -2.275665283203125, "p10": -0.8424209594726562, "median": 0.000194549560546875, "p90": 1.4249359130859376, "max": 3.112060546875, "pos_frac": 0.5, "sample": [3.112060546875, -0.05770683288574219, 2.3040771484375, -0.5952301025390625, 1.424652099609375, 0.22850799560546875, -0.06513214111328125, -0.5117874145507812, -1.3436279296875, -0.06173515319824219, -0.10038375854492188, 1.166534423828125, 0.0, -0.07155609130859375, -1.542877197265625, -0.4949798583984375, 0.1273193359375, -0.08582305908203125, -1.470794677734375, 0.2939910888671875, 0.356170654296875, 0.263824462890625, 0.00504302978515625, -0.444610595703125, 0.358612060546875, 0.6627655029296875, 0.7499465942382812, -0.1598052978515625, 0.844024658203125, -0.30741119384765625, -0.1880950927734375, -0.5162353515625, 0.28319549560546875, -0.5130615234375, -1.3353271484375, 1.69427490234375, 0.297821044921875, -0.90142822265625, 0.4018402099609375, -0.339019775390625, 0.5362548828125, 0.33063507080078125, 0.41664886474609375, -0.114776611328125, -0.993072509765625, -0.428955078125, 0.5402984619140625, 1.42559814453125, 0.99456787109375, -0.197479248046875, 1.36822509765625, -0.259124755859375, 0.02744293212890625, 0.70849609375, 0.667205810546875, 0.00038909912109375, -0.233642578125, 0.451507568359375, -0.320068359375, 1.286376953125, 0.1907176971435547, 2.13934326171875, 1.12884521484375, -0.202972412109375, 0.038177490234375, -0.274627685546875, -1.07672119140625, -1.72760009765625, 2.1964111328125, 0.3965568542480469, -0.17919158935546875, -0.2225780487060547, 0.9394683837890625, 1.17742919921875, -0.10250091552734375, -0.3517341613769531, -0.61077880859375, -0.5396022796630859, 1.76885986328125, -0.58050537109375, 0.880340576171875, -0.41046142578125, 0.00751495361328125, -0.173065185546875, 1.153167724609375, -0.3652229309082031, 0.986175537109375, 1.151641845703125, 0.4815635681152344, -0.48183441162109375, 1.467620849609375, -0.5396347045898438, 0.7143325805664062, 0.7490234375, 1.80609130859375, 1.4525299072265625, -0.0468292236328125, -0.13369178771972656, 0.22682952880859375, -0.0009613037109375, -0.8217620849609375, -0.01947784423828125, 1.746063232421875, 0.4583587646484375, -0.41766357421875, -0.77105712890625, -0.01214599609375, 0.0, -0.17584228515625, -0.24298477172851562, 0.7393798828125, 0.17116546630859375, -1.4858245849609375, 0.4997100830078125, -0.2242279052734375, -0.0542144775390625, -0.890625, 1.545166015625, 0.104248046875, -0.9335784912109375, 1.692962646484375, 0.75299072265625, 0.4304046630859375, 0.700347900390625, -0.3524627685546875, -0.9089508056640625, -2.275665283203125, 0.588775634765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000026.npy"}
{"epoch": 0.05445026178010471, "step": 27, "batch_size": 128, "mean": 0.1770738810300827, "std": 0.9671897888183594, "min": -2.4495849609375, "p10": -0.9879547119140625, "median": 0.13506698608398438, "p90": 1.5356750488281246, "max": 3.108154296875, "pos_frac": 0.5859375, "sample": [0.807525634765625, 0.366363525390625, -0.3032035827636719, -0.18802261352539062, 0.1112060546875, -0.4580078125, -0.374481201171875, -0.125091552734375, 0.0472564697265625, -0.1498870849609375, 0.23809814453125, -1.501922607421875, -0.17156982421875, -0.07326507568359375, -2.4495849609375, 0.244384765625, -0.9948577880859375, 0.01861572265625, 0.4482879638671875, 0.71466064453125, -0.2970733642578125, 0.341949462890625, 1.497589111328125, -0.488250732421875, -0.99053955078125, 0.71112060546875, -0.39886474609375, 0.094390869140625, -0.4328727722167969, 0.4011077880859375, 1.86834716796875, 0.09930419921875, 0.601104736328125, 0.42291259765625, 0.6725311279296875, 0.0535888671875, -0.4178619384765625, 0.077484130859375, 0.95098876953125, 0.2322998046875, -0.30096435546875, 1.624542236328125, 1.853973388671875, 0.629150390625, -0.1280517578125, 0.824676513671875, 0.2168426513671875, 0.4853973388671875, 1.01763916015625, 0.6093902587890625, 1.70648193359375, 0.1841583251953125, 1.2941055297851562, 0.29117584228515625, 2.2227783203125, 0.8077392578125, 0.5953445434570312, 1.268951416015625, -0.27872467041015625, 3.108154296875, -0.3431396484375, -1.0584716796875, -0.3216094970703125, 0.429443359375, 0.8844985961914062, -0.8479461669921875, 0.7246551513671875, 0.002277374267578125, -1.1773681640625, -0.17755126953125, -1.7762451171875, -1.70611572265625, -2.3486328125, -0.7486572265625, -0.29839324951171875, -0.018890380859375, 0.960540771484375, 1.681060791015625, -1.809600830078125, -0.224212646484375, 0.11483001708984375, 2.0732421875, -0.2657470703125, 0.315216064453125, 1.06134033203125, 1.434234619140625, -0.363372802734375, -0.3611164093017578, -0.3460540771484375, 0.9979095458984375, 1.8823699951171875, 0.6326141357421875, -2.151123046875, 0.4755096435546875, 0.748504638671875, -1.291229248046875, -0.774017333984375, 0.10502052307128906, -0.970306396484375, -1.2174072265625, 0.155303955078125, -0.8544521331787109, -0.986846923828125, 1.8016357421875, -0.14044761657714844, 1.83807373046875, 0.932403564453125, 0.212432861328125, -0.92559814453125, 0.009387969970703125, 2.283905029296875, 0.1838207244873047, -0.018304824829101562, 0.4007568359375, 0.4148712158203125, -0.1606292724609375, 0.3177490234375, 0.5864524841308594, -0.44818115234375, 0.37005615234375, 2.49468994140625, 0.46666717529296875, 0.588531494140625, -0.23248291015625, -0.34320068359375, -0.012176513671875, 0.382080078125, 0.18437767028808594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000027.npy"}
{"epoch": 0.05654450261780105, "step": 28, "batch_size": 128, "mean": 0.1266840100288391, "std": 0.8673487901687622, "min": -3.871337890625, "p10": -0.7564918518066406, "median": 0.057033538818359375, "p90": 1.0705230712890623, "max": 2.9512939453125, "pos_frac": 0.53125, "sample": [0.191131591796875, -3.871337890625, 0.14591217041015625, 0.35528564453125, 0.2796478271484375, -0.15401649475097656, -0.33084869384765625, -0.7395248413085938, 0.5127468109130859, 0.68408203125, 0.4894390106201172, 0.9872283935546875, 0.972198486328125, -1.01702880859375, 0.829559326171875, 0.9208831787109375, -1.08172607421875, -0.10906219482421875, 0.5862579345703125, 0.6041259765625, -0.13041305541992188, -0.5836429595947266, 0.7169570922851562, 1.141937255859375, 0.603515625, -0.2198486328125, -1.1529541015625, -0.1263904571533203, -0.481964111328125, -0.83984375, 0.8389892578125, 0.36981201171875, -0.61572265625, 0.0646209716796875, -0.39275360107421875, -0.1396331787109375, 1.0318756103515625, 0.9305419921875, -0.636077880859375, -0.30352783203125, 0.498077392578125, -0.4754638671875, 0.5031585693359375, -0.73681640625, -0.5341796875, -0.32501220703125, -1.377227783203125, -0.837799072265625, 1.184478759765625, 1.6444168090820312, 0.238067626953125, 0.6552734375, -0.42694091796875, 1.842529296875, -1.56109619140625, 0.750701904296875, -0.2725830078125, 0.557891845703125, -0.508758544921875, 0.64727783203125, 0.936004638671875, -0.4629058837890625, 1.410736083984375, -0.08821487426757812, -0.8817138671875, 1.4021453857421875, -0.53717041015625, -1.864654541015625, 0.81256103515625, -0.341705322265625, -0.2213897705078125, -0.258697509765625, 0.18030548095703125, 0.02825927734375, -0.56634521484375, -0.054962158203125, -0.673095703125, 0.06813812255859375, -0.10692596435546875, 2.9512939453125, -0.46945953369140625, 0.43487548828125, -0.15093994140625, -0.0183868408203125, -0.4796905517578125, -0.37896728515625, -0.0145721435546875, -0.220916748046875, -0.375762939453125, 0.24363136291503906, 0.38555908203125, 2.7799072265625, 1.1470184326171875, 1.0399169921875, 0.22408294677734375, -0.06586456298828125, 2.291473388671875, -0.4792327880859375, 0.01312255859375, 0.382781982421875, 0.436187744140625, 0.378570556640625, -0.944580078125, 0.1507415771484375, 1.38897705078125, -0.79608154296875, -0.47928619384765625, -0.17230987548828125, 0.708526611328125, 0.04944610595703125, 0.09844970703125, 0.1504058837890625, 1.02734375, 0.08681488037109375, 0.842926025390625, 0.5098419189453125, -0.05712890625, 0.17657852172851562, 1.32958984375, 0.0419158935546875, -0.804901123046875, 1.50360107421875, -0.21649169921875, 0.69683837890625, -0.675750732421875, 0.3692512512207031, 0.0, 0.5994110107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000028.npy"}
{"epoch": 0.05863874345549738, "step": 29, "batch_size": 128, "mean": 0.15327222645282745, "std": 0.8613457083702087, "min": -2.43450927734375, "p10": -0.87786865234375, "median": 0.07238197326660156, "p90": 1.46195068359375, "max": 2.551910400390625, "pos_frac": 0.5546875, "sample": [-0.00714874267578125, -1.365020751953125, 2.12200927734375, 0.130889892578125, 1.2375946044921875, -0.9224853515625, -0.46136474609375, 1.67919921875, -0.7293701171875, -0.460693359375, 1.733306884765625, -0.971160888671875, 2.406036376953125, 1.560699462890625, -0.4780616760253906, 0.0, 0.809844970703125, 0.25732421875, 0.026824951171875, 0.07745742797851562, -0.2802734375, 1.4534912109375, 0.454010009765625, 0.58013916015625, 0.078582763671875, 0.0960693359375, -0.740447998046875, 0.048980712890625, -0.22606658935546875, 0.9406356811523438, 1.1717987060546875, 0.764678955078125, 0.047149658203125, -0.40594482421875, -0.9943008422851562, -0.70050048828125, 0.00341796875, -0.790924072265625, 1.79620361328125, -0.7296066284179688, -0.525115966796875, -0.1748027801513672, 1.578765869140625, -0.639892578125, -0.92034912109375, -0.6731529235839844, 0.213104248046875, 0.510223388671875, 0.2930755615234375, 1.0554046630859375, -0.89495849609375, -0.4144287109375, -0.268707275390625, 0.264923095703125, -0.174163818359375, -1.024078369140625, -2.43450927734375, 0.6613006591796875, 0.1268310546875, 0.44663238525390625, -0.09893798828125, -0.27996826171875, 1.2615966796875, 0.38214111328125, 0.05574798583984375, 0.708709716796875, 0.6748046875, -0.3923187255859375, 0.0, 0.6093292236328125, 2.11639404296875, 0.595916748046875, 0.3230438232421875, -1.0767822265625, 1.1741256713867188, 0.521331787109375, -0.383331298828125, -0.02242279052734375, 1.538360595703125, 0.662261962890625, 1.300628662109375, -1.2113418579101562, 0.41851806640625, -0.23023223876953125, -0.87054443359375, 0.0673065185546875, 0.28359222412109375, 0.46441650390625, 0.4273529052734375, -1.636688232421875, -0.5670318603515625, -0.5611572265625, -0.6682281494140625, 0.564178466796875, -0.293243408203125, 0.2705841064453125, -1.383392333984375, 0.1044464111328125, -0.42957305908203125, 0.827056884765625, 0.86041259765625, -0.4272804260253906, -0.3817291259765625, 0.2596282958984375, -0.206756591796875, -0.26526641845703125, -0.1402130126953125, -0.215362548828125, 0.486846923828125, 1.481689453125, -0.039276123046875, 0.2606048583984375, -0.66552734375, 0.1211090087890625, -0.071990966796875, 2.551910400390625, -0.32343292236328125, 1.365081787109375, 0.330810546875, 0.3668975830078125, -0.0811614990234375, 0.46772003173828125, 1.498321533203125, 0.00382232666015625, 0.242431640625, -1.11810302734375, 1.6787109375, 0.11322021484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000029.npy"}
{"epoch": 0.060732984293193716, "step": 30, "batch_size": 128, "mean": 0.26843076944351196, "std": 0.997687816619873, "min": -3.457916259765625, "p10": -0.8089767456054687, "median": 0.19603824615478516, "p90": 1.5072021484375, "max": 4.030487060546875, "pos_frac": 0.640625, "sample": [0.00750732421875, 0.1324462890625, 1.04510498046875, 1.603424072265625, 0.2990989685058594, 0.662841796875, -0.8303375244140625, -1.451934814453125, 0.93951416015625, 0.3830413818359375, -0.25867462158203125, 0.2073974609375, 1.229034423828125, -0.5317535400390625, 0.8043365478515625, -0.716766357421875, 0.296875, -0.802764892578125, -0.08490753173828125, 0.4592742919921875, 0.38043212890625, 0.94671630859375, 0.114288330078125, -0.46090126037597656, 0.19970703125, 0.858795166015625, 1.50537109375, 0.19489479064941406, 0.4150657653808594, -0.320465087890625, -0.00347900390625, 0.07547378540039062, -0.4427490234375, -1.525634765625, 0.1709136962890625, 1.534515380859375, -0.4979705810546875, -0.00786590576171875, -0.8234710693359375, 1.9788818359375, 0.145416259765625, -0.6081161499023438, 1.452850341796875, 0.1312255859375, 1.05084228515625, -2.1739501953125, -0.3566551208496094, 0.78387451171875, -0.442962646484375, -0.88983154296875, 0.3937721252441406, -0.015045166015625, 1.29412841796875, 1.0592041015625, 1.642578125, 0.43024444580078125, -1.254913330078125, 0.23350906372070312, -0.8443603515625, 1.4904327392578125, 0.41510009765625, 0.443572998046875, -0.1837615966796875, -0.5067138671875, -0.7996444702148438, 0.045623779296875, 0.695892333984375, 0.1365966796875, 0.7020645141601562, 1.71148681640625, -0.21480941772460938, 3.080474853515625, 0.7085418701171875, -0.5016288757324219, -0.67852783203125, 3.4222412109375, -0.8833694458007812, 0.1248016357421875, 0.01537322998046875, -0.040557861328125, 0.26335906982421875, 1.755706787109375, 2.522430419921875, 0.11647796630859375, 0.19718170166015625, 0.6395263671875, 4.030487060546875, 0.3899993896484375, 0.59423828125, -0.0719146728515625, -0.31519317626953125, 0.321014404296875, 0.2967681884765625, 0.3651123046875, -1.011383056640625, 1.0465087890625, 0.15655517578125, 0.004180908203125, -1.213134765625, 1.4737548828125, 1.511474609375, -0.2353363037109375, 0.1900177001953125, 0.5484848022460938, 0.597869873046875, 0.30289459228515625, 0.187347412109375, 0.504852294921875, -0.110504150390625, -0.623321533203125, 1.556976318359375, -0.4373626708984375, 2.46929931640625, 0.29766845703125, 0.2778663635253906, -0.7641525268554688, 0.47658538818359375, 0.9518661499023438, 0.2816295623779297, 0.16314697265625, -0.95361328125, -0.357757568359375, -0.306610107421875, -3.457916259765625, -0.00589752197265625, 0.2571258544921875, -0.192169189453125, 0.7667236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000030.npy"}
{"epoch": 0.06282722513089005, "step": 31, "batch_size": 128, "mean": 0.23408250510692596, "std": 0.9485920667648315, "min": -2.3157806396484375, "p10": -0.7738288879394531, "median": 0.16403579711914062, "p90": 1.2856338500976563, "max": 4.05242919921875, "pos_frac": 0.59375, "sample": [1.3523941040039062, 0.481292724609375, -0.442626953125, -0.149383544921875, 0.0399017333984375, 1.2955780029296875, -0.62652587890625, 0.2783203125, -0.19635772705078125, 0.014129638671875, 1.67816162109375, -0.7794418334960938, 0.10906982421875, -1.21014404296875, 0.6143798828125, 1.02362060546875, 0.5004959106445312, -0.01430511474609375, 1.8710556030273438, 0.139190673828125, 0.80517578125, -1.322540283203125, 0.028045654296875, -0.77142333984375, 0.17760467529296875, -1.450531005859375, 0.6009521484375, 0.1049041748046875, 0.22405242919921875, 0.45893096923828125, 0.5987358093261719, -0.5457611083984375, 0.4667243957519531, -0.00240325927734375, 0.380035400390625, -0.978118896484375, 0.2878570556640625, 0.419189453125, 0.2578887939453125, 0.0, 0.2807159423828125, 0.313323974609375, 0.0, -0.1656494140625, 0.3838844299316406, 0.4653472900390625, 0.6053466796875, 0.2411346435546875, 0.7679901123046875, -0.4168853759765625, -2.277740478515625, -2.3157806396484375, 0.0487060546875, 0.3546295166015625, -0.022426605224609375, 0.23691558837890625, 2.87066650390625, 1.153411865234375, -0.120574951171875, 0.3647193908691406, 0.0, 0.9180908203125, 2.164276123046875, 0.50274658203125, 0.8609161376953125, 0.7933502197265625, 0.7449798583984375, -0.664825439453125, 1.1209716796875, -0.40887451171875, 0.674774169921875, 1.98431396484375, -0.40415191650390625, 0.1975250244140625, 0.8113250732421875, 1.13037109375, -0.4535675048828125, -0.1636371612548828, 2.71099853515625, 0.543487548828125, 0.036693572998046875, 0.16742706298828125, 2.2213134765625, -0.1579761505126953, -0.208343505859375, -0.53021240234375, -1.5157470703125, 2.31005859375, 0.5781784057617188, 1.2813720703125, -0.7086181640625, 0.5960979461669922, 2.775848388671875, -0.042572021484375, -0.1797332763671875, -0.8284378051757812, -0.662628173828125, -0.456207275390625, 0.794525146484375, 0.0, -0.84527587890625, 4.05242919921875, -0.2896270751953125, -0.4040565490722656, -0.43769073486328125, 0.005645751953125, -1.386260986328125, -0.24123764038085938, 1.73809814453125, 0.02130126953125, 0.22076416015625, 0.24892425537109375, -0.08242034912109375, -0.79345703125, 0.6934814453125, 0.219818115234375, 0.011621475219726562, -0.1408843994140625, -0.10650634765625, 1.11639404296875, 0.16064453125, 0.18813323974609375, -0.015750885009765625, -1.11956787109375, -0.38904762268066406, 0.2903900146484375, -0.65570068359375, 0.8824310302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000031.npy"}
{"epoch": 0.06492146596858639, "step": 32, "batch_size": 128, "mean": 0.27091488242149353, "std": 1.0241347551345825, "min": -2.4141845703125, "p10": -1.110845947265625, "median": 0.3087615966796875, "p90": 1.3992233276367188, "max": 2.77911376953125, "pos_frac": 0.6015625, "sample": [0.799957275390625, -0.3563575744628906, -1.473114013671875, 0.2452392578125, -0.580078125, 0.07840728759765625, 0.2606697082519531, -0.6782989501953125, 0.184478759765625, 0.3973388671875, -0.17327880859375, 0.7746963500976562, -0.606353759765625, 0.2899932861328125, 1.810028076171875, -0.20179176330566406, 2.77911376953125, 0.79888916015625, 0.661163330078125, 0.419586181640625, -0.6473007202148438, 1.287109375, -0.064056396484375, 1.2098846435546875, 1.4123077392578125, 0.951080322265625, 1.149322509765625, -0.08130455017089844, 1.4400634765625, 0.8520736694335938, 0.292633056640625, 0.10369873046875, -0.68804931640625, -1.2281494140625, 0.5987548828125, 0.568206787109375, -0.7981719970703125, 2.437835693359375, 0.644256591796875, 0.9891815185546875, 0.7082977294921875, 0.285552978515625, 0.548828125, -0.9769668579101562, 0.9103546142578125, 0.313201904296875, 1.2740020751953125, -0.2233734130859375, -1.196563720703125, -0.3882598876953125, 0.6903076171875, -1.23089599609375, 1.1220703125, -1.18634033203125, 1.32861328125, -0.19074249267578125, 0.895904541015625, 2.3192138671875, 0.928314208984375, -1.0784912109375, 0.3624267578125, -0.7328338623046875, 0.9008636474609375, 1.226318359375, -0.2100830078125, -1.3689422607421875, -0.5469970703125, -1.3516845703125, -0.47381591796875, 0.2879180908203125, 0.513763427734375, 0.3626251220703125, 0.026020050048828125, -0.53460693359375, 0.8842010498046875, -0.46600341796875, -0.3872528076171875, -2.4141845703125, 0.9406204223632812, 2.568115234375, 0.558868408203125, 1.755462646484375, -0.2476024627685547, -0.1788806915283203, -2.16107177734375, 1.2823486328125, 1.44866943359375, 0.6204681396484375, -0.235382080078125, -0.54376220703125, -0.5268325805664062, -0.23920822143554688, 1.376678466796875, -0.64263916015625, -1.862701416015625, 0.3043212890625, 0.1577606201171875, 0.075225830078125, 0.6914749145507812, -0.975250244140625, 1.0516815185546875, 1.67987060546875, 1.3370819091796875, 0.6704902648925781, -2.10577392578125, 0.5853729248046875, -0.33017730712890625, -0.9151153564453125, -0.9052276611328125, 0.9153289794921875, 0.34613037109375, 0.7127838134765625, -0.03997802734375, 1.8170013427734375, 1.03656005859375, 1.319091796875, 1.220703125, -0.22540283203125, 1.235107421875, -1.524169921875, 1.6935272216796875, 0.5328140258789062, -0.08843994140625, 1.39361572265625, -0.5013198852539062, -1.205718994140625, 1.3353271484375, 2.6788330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000032.npy"}
{"epoch": 0.06701570680628273, "step": 33, "batch_size": 128, "mean": 0.4421673119068146, "std": 1.0764532089233398, "min": -2.820220947265625, "p10": -0.6772735595703123, "median": 0.3492393493652344, "p90": 1.763777160644531, "max": 4.06207275390625, "pos_frac": 0.6953125, "sample": [0.6934738159179688, 1.98651123046875, 0.9822998046875, -1.6827392578125, 0.175537109375, 0.565277099609375, -0.8493194580078125, 2.0128173828125, -0.31048583984375, 0.34627532958984375, 0.172943115234375, 0.25304412841796875, 4.06207275390625, 0.66693115234375, 0.0, 0.5847244262695312, 0.1781158447265625, 0.715667724609375, 0.5445556640625, -2.820220947265625, 0.05572509765625, 0.85614013671875, 0.5132675170898438, 0.567413330078125, -1.0115509033203125, -0.2028656005859375, -0.17484283447265625, 0.310516357421875, 2.02801513671875, -0.526611328125, -0.030731201171875, 0.296112060546875, -0.469940185546875, -1.505279541015625, -0.1527080535888672, 1.27703857421875, 0.827301025390625, 0.89312744140625, -2.500213623046875, 1.195709228515625, 0.3610076904296875, -0.21352005004882812, 0.3830986022949219, 0.3321533203125, -1.97406005859375, -0.642913818359375, 2.256988525390625, -0.6110763549804688, 0.206756591796875, 0.035797119140625, -0.0302734375, -0.8796234130859375, 1.61865234375, 0.0, 0.78350830078125, 0.21611785888671875, -0.0011157989501953125, 0.131317138671875, 1.057586669921875, 0.76812744140625, 2.900238037109375, -0.116058349609375, 0.26493072509765625, 0.7187652587890625, 0.17557334899902344, -1.47467041015625, -0.04656028747558594, 0.1984100341796875, -1.2127685546875, 1.691436767578125, 0.927337646484375, -0.6051483154296875, -0.7574462890625, 0.352203369140625, 0.27496337890625, 0.084747314453125, -1.0645751953125, 0.712066650390625, 0.2536773681640625, 1.8468017578125, 1.290435791015625, -0.5197906494140625, 1.5101318359375, 0.229949951171875, 0.99951171875, 1.496734619140625, 1.1496429443359375, 0.24066162109375, 2.3141632080078125, 1.689544677734375, 0.7757110595703125, 0.2864990234375, 1.1966018676757812, -0.22384262084960938, 0.5273246765136719, 1.5030517578125, 2.7087249755859375, 2.211212158203125, -0.451934814453125, 0.47900390625, -0.35198974609375, 0.11648941040039062, 1.306396484375, 1.310577392578125, -0.5535202026367188, 0.4973907470703125, 0.24825286865234375, 1.7281951904296875, 0.5745086669921875, 0.47747802734375, -0.224151611328125, 0.3728485107421875, 2.322265625, 0.46783447265625, 0.7440719604492188, 0.4732513427734375, -1.118499755859375, 1.2259902954101562, 0.4215087890625, 0.0406494140625, 1.6015625, 0.43353271484375, 2.037109375, -0.49053955078125, 0.73162841796875, -0.13372039794921875, -0.3956298828125, 3.875030517578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000033.npy"}
{"epoch": 0.06910994764397906, "step": 34, "batch_size": 128, "mean": 0.6636799573898315, "std": 1.2328981161117554, "min": -2.2054443359375, "p10": -0.5166366577148437, "median": 0.4062080383300781, "p90": 2.420849609375, "max": 4.24554443359375, "pos_frac": 0.6796875, "sample": [0.38838958740234375, -0.12652587890625, -0.06207275390625, 0.40290069580078125, 0.2806243896484375, 1.5647735595703125, -0.815673828125, 0.5772552490234375, -0.10968017578125, 0.77215576171875, 0.26153564453125, -0.390533447265625, 4.24554443359375, 0.973388671875, 1.038848876953125, 0.715087890625, 1.3675537109375, -0.359588623046875, 0.1390533447265625, 0.46730804443359375, -0.15737152099609375, 1.124755859375, 1.64935302734375, 0.84832763671875, -0.9432373046875, 2.315338134765625, 2.9881591796875, -0.7138671875, -0.2050647735595703, -2.2054443359375, 0.4559783935546875, 0.36473846435546875, 0.34136962890625, -0.2253570556640625, -0.02337646484375, 0.7194976806640625, -0.1126708984375, 0.31182861328125, -0.193145751953125, -0.2646942138671875, -0.566009521484375, -0.9875640869140625, 0.61920166015625, 2.40740966796875, 1.9365234375, -0.03546142578125, -1.6829833984375, 0.9521484375, -0.177337646484375, -0.793609619140625, 0.5143585205078125, 0.8610687255859375, 0.2525501251220703, 1.8236083984375, 1.7147216796875, -0.4616241455078125, 0.19744110107421875, 0.4549560546875, 3.556396484375, 2.679290771484375, 1.688629150390625, 0.13992691040039062, 0.653167724609375, -0.3390960693359375, 3.7554931640625, 2.84124755859375, 0.548065185546875, 0.12626075744628906, 1.2174606323242188, -0.987152099609375, 0.19141387939453125, 0.252105712890625, 0.81317138671875, 0.97357177734375, 3.106781005859375, -0.33984375, 0.5413818359375, 0.282257080078125, 0.3527870178222656, -0.2474079132080078, 1.850555419921875, 2.45220947265625, 2.83648681640625, -0.007781982421875, 1.3634033203125, 3.042694091796875, 0.952392578125, 2.225616455078125, -0.1340179443359375, 2.2716064453125, 1.1791000366210938, -2.09722900390625, 4.05889892578125, 1.169830322265625, 0.488433837890625, 0.60882568359375, 0.8429718017578125, 0.0943603515625, 0.99273681640625, 0.10054779052734375, -0.2283935546875, 0.731109619140625, 0.409515380859375, 0.559234619140625, 3.99932861328125, -0.4370441436767578, -0.3156890869140625, 0.39279937744140625, 0.16216278076171875, 2.054534912109375, 1.2274322509765625, -0.49822998046875, -1.1800537109375, -0.1751708984375, -0.375701904296875, 0.55352783203125, -0.14337158203125, 1.1782379150390625, -1.0569000244140625, 1.5576019287109375, -0.5595855712890625, 0.2479400634765625, 1.2295188903808594, 0.246307373046875, 3.74591064453125, -0.2046661376953125, 0.1732177734375, 1.1250457763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000034.npy"}
{"epoch": 0.0712041884816754, "step": 35, "batch_size": 128, "mean": 0.30603715777397156, "std": 1.4546723365783691, "min": -4.946044921875, "p10": -1.1990905761718749, "median": 0.23419952392578125, "p90": 1.915088653564453, "max": 4.495697021484375, "pos_frac": 0.625, "sample": [-0.17652130126953125, 0.5501708984375, -0.7235870361328125, -0.947601318359375, -1.4839019775390625, 4.16717529296875, 0.09295654296875, 1.0824127197265625, -3.977691650390625, -0.97528076171875, 1.24420166015625, -0.16522979736328125, 0.11761474609375, -2.61273193359375, 0.44708251953125, 0.37932586669921875, 0.188079833984375, 0.03277587890625, -0.360321044921875, -0.2881050109863281, -0.0157623291015625, 0.24456787109375, 0.579925537109375, -0.20086669921875, 0.6945114135742188, -0.218536376953125, 0.5133056640625, -1.1862640380859375, -0.0146636962890625, -1.18365478515625, -0.39933013916015625, -0.22443771362304688, -0.731781005859375, 0.1147308349609375, 1.515380859375, 2.68017578125, 0.61395263671875, -2.16705322265625, 1.1233139038085938, 0.0, -1.254058837890625, 0.748779296875, 0.2238311767578125, 0.953948974609375, -0.21430206298828125, 0.114105224609375, 2.12530517578125, -1.693084716796875, 0.5398712158203125, -1.089874267578125, 1.015533447265625, -0.38910675048828125, 3.2581787109375, 1.89898681640625, 2.816986083984375, 0.46984100341796875, 2.39935302734375, -0.426910400390625, 0.937774658203125, 0.17108917236328125, 2.50360107421875, 0.51556396484375, 2.201751708984375, 1.1067047119140625, -1.0196533203125, 1.226104736328125, 0.0607452392578125, 0.19015121459960938, -0.9976348876953125, 1.19024658203125, 0.6909942626953125, -3.79156494140625, -0.776153564453125, 0.1537933349609375, 0.21056747436523438, -0.0736083984375, 0.12264251708984375, 1.585693359375, 4.495697021484375, 0.79632568359375, 0.9189109802246094, 0.19334793090820312, 1.063507080078125, 0.9840850830078125, 0.8921279907226562, 1.9526596069335938, 0.08083343505859375, 1.0274810791015625, -4.403411865234375, 0.32117462158203125, 1.129180908203125, 0.8007888793945312, 0.1880474090576172, 0.6623382568359375, 0.31758689880371094, 1.087554931640625, -0.6605224609375, -4.946044921875, 1.62939453125, 0.2679901123046875, -0.522216796875, 2.251739501953125, 0.0, -1.2820587158203125, 2.9228515625, 1.073211669921875, -1.6194610595703125, 1.036651611328125, -0.8580322265625, 0.81829833984375, 1.6585693359375, -0.050323486328125, -0.330413818359375, -0.7048187255859375, 1.3766708374023438, 1.63330078125, 1.0887565612792969, 0.4349365234375, -1.4437713623046875, 0.8993644714355469, -0.5341033935546875, -0.38409423828125, 1.8345947265625, -0.126739501953125, 1.3747711181640625, 3.4539794921875, -1.2290191650390625, 1.5665283203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000035.npy"}
{"epoch": 0.07329842931937172, "step": 36, "batch_size": 128, "mean": 0.7109129428863525, "std": 1.4718480110168457, "min": -2.6549072265625, "p10": -1.1006805419921872, "median": 0.6314201354980469, "p90": 2.6609230041503906, "max": 4.50018310546875, "pos_frac": 0.6640625, "sample": [1.30340576171875, 2.597625732421875, 0.5409393310546875, 1.0505752563476562, 0.116424560546875, -0.64984130859375, 0.4107666015625, 0.6403350830078125, 1.682281494140625, -0.8096122741699219, 0.683441162109375, 0.071075439453125, 2.0050048828125, 0.35693359375, -1.35552978515625, 0.6285934448242188, -1.6090927124023438, 1.322052001953125, 0.634246826171875, 2.6496047973632812, 0.8723373413085938, 2.036865234375, -1.3242721557617188, 0.0, 1.3421478271484375, 0.337554931640625, -0.629302978515625, 1.84173583984375, -0.2939910888671875, -0.13858413696289062, 4.433837890625, 2.3304290771484375, 0.965057373046875, 1.22601318359375, -2.116790771484375, 2.505279541015625, 0.9580764770507812, 0.873138427734375, -0.020528793334960938, 2.6873321533203125, 4.50018310546875, -0.1882781982421875, -0.3153076171875, 0.7228546142578125, 1.0348663330078125, 1.60247802734375, 0.0933380126953125, 1.309295654296875, 1.3170928955078125, 1.00775146484375, -1.03167724609375, 1.7417449951171875, 1.57196044921875, 0.489715576171875, -0.9370574951171875, -0.5524063110351562, -1.4094085693359375, 0.947998046875, 2.926422119140625, 3.9442138671875, 0.0, -0.384857177734375, 2.69989013671875, 1.3870086669921875, 3.592193603515625, 2.024322509765625, 0.4947357177734375, -0.07306671142578125, 0.8272705078125, 1.067047119140625, 1.880584716796875, 1.02264404296875, -1.931610107421875, -0.9513397216796875, -1.4520416259765625, 0.5720939636230469, 0.7196044921875, 3.530792236328125, -0.729034423828125, 0.882568359375, -2.6549072265625, -0.34375, -0.2178802490234375, -0.307403564453125, -0.7501220703125, 0.04000663757324219, -2.45452880859375, -0.71728515625, 1.2886810302734375, -0.0814361572265625, 1.25384521484375, 2.18310546875, 1.7463226318359375, -0.14073944091796875, -1.261688232421875, 1.8245849609375, -0.2486095428466797, 0.348907470703125, -0.3520965576171875, 0.5457382202148438, -2.32904052734375, 3.462890625, 0.2159423828125, 1.26153564453125, 0.17547607421875, 1.6502227783203125, 1.64697265625, 2.933929443359375, -1.960723876953125, 0.8886260986328125, 0.08508110046386719, 0.5279083251953125, -0.0385284423828125, 3.20953369140625, -0.15398406982421875, -0.3329048156738281, 2.43597412109375, 4.027557373046875, 0.415802001953125, 0.378753662109375, 2.1549072265625, 4.31219482421875, 1.0410614013671875, 0.26749229431152344, -1.344818115234375, -0.07152175903320312, 0.9582061767578125, 1.3674163818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000036.npy"}
{"epoch": 0.07539267015706806, "step": 37, "batch_size": 128, "mean": 0.5112783312797546, "std": 1.5212631225585938, "min": -3.5174560546875, "p10": -1.3530029296875, "median": 0.44659423828125, "p90": 2.2638977050781244, "max": 5.104217529296875, "pos_frac": 0.640625, "sample": [1.3466796875, 1.60748291015625, 0.6230010986328125, -0.33349609375, 1.766265869140625, -0.2228107452392578, 5.104217529296875, 0.3284873962402344, 0.2740478515625, -1.020843505859375, 4.12811279296875, 0.31288719177246094, 0.61083984375, 1.3439788818359375, 2.9744873046875, 2.37066650390625, 1.397705078125, -0.4109992980957031, 0.538604736328125, -0.0888214111328125, 0.6773757934570312, 1.6064453125, -0.47113990783691406, 0.05718994140625, -2.2484130859375, -1.689453125, 0.7113037109375, 0.996826171875, 0.468841552734375, -1.29254150390625, 1.2117156982421875, -0.3249664306640625, 0.7748947143554688, -2.445556640625, 1.7059402465820312, 0.42974853515625, -0.694305419921875, -1.396514892578125, 3.185028076171875, 1.8424530029296875, 0.38648223876953125, -0.723388671875, -0.36896514892578125, 2.741485595703125, 1.8016204833984375, 0.3575439453125, 0.1436138153076172, -0.3662567138671875, -0.9647064208984375, -2.402313232421875, 1.3699951171875, -0.476715087890625, 1.54296875, 2.7034912109375, -0.23919677734375, 0.50396728515625, 0.203369140625, 1.367156982421875, 1.1221923828125, -2.292755126953125, 1.372161865234375, -0.707763671875, 1.6401519775390625, 0.894378662109375, -3.5174560546875, 0.2559356689453125, 3.910400390625, -0.55047607421875, 2.203125, -2.005615234375, 2.2181396484375, 0.2677764892578125, 2.17437744140625, 1.914306640625, 0.282806396484375, -0.771942138671875, 0.497314453125, 3.585418701171875, 1.14678955078125, -1.06768798828125, -0.29193878173828125, 4.527862548828125, 0.1778106689453125, 0.837921142578125, 1.8675537109375, -0.037994384765625, 0.078643798828125, -0.3234710693359375, 0.67120361328125, 3.36859130859375, -0.559722900390625, 0.305023193359375, -0.752593994140625, 0.538177490234375, 0.1398468017578125, 0.580474853515625, -0.528656005859375, 0.8073654174804688, 0.44500732421875, -0.4801025390625, 0.976318359375, -0.12860107421875, -1.366973876953125, -1.0041427612304688, -0.01767730712890625, 1.62982177734375, 0.44818115234375, 1.9085540771484375, 1.1804962158203125, 0.232513427734375, 1.5868988037109375, -1.6943359375, 2.906982421875, -2.68402099609375, 1.171234130859375, 0.73114013671875, 2.6707763671875, -2.843841552734375, -0.1554412841796875, -1.347015380859375, -0.934539794921875, -1.58843994140625, 1.497161865234375, 1.298614501953125, 0.981048583984375, 1.229888916015625, -0.3692626953125, 1.850189208984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000037.npy"}
{"epoch": 0.0774869109947644, "step": 38, "batch_size": 128, "mean": 0.6297144889831543, "std": 1.6352430582046509, "min": -5.1422271728515625, "p10": -1.3539016723632813, "median": 0.5485687255859375, "p90": 2.6667953491210934, "max": 4.6800537109375, "pos_frac": 0.65625, "sample": [0.85467529296875, 2.037689208984375, -0.2468719482421875, -0.8997039794921875, -1.101318359375, 1.4854888916015625, -0.8662033081054688, 4.658477783203125, 3.08935546875, 0.900482177734375, 0.862091064453125, -0.145233154296875, 0.07080078125, 2.485748291015625, 0.41107177734375, -1.13275146484375, 1.830963134765625, -0.1739044189453125, 2.181884765625, -1.35028076171875, 3.063873291015625, -1.391204833984375, -0.0969390869140625, 0.37664794921875, 0.39209747314453125, -0.687347412109375, -0.28778076171875, -1.858184814453125, 0.89569091796875, -0.3226337432861328, 0.566009521484375, 1.3041534423828125, 1.2847900390625, 2.3055419921875, 1.1540985107421875, 2.6253814697265625, 3.257904052734375, -0.14375686645507812, 0.20086669921875, 0.504150390625, 0.8785247802734375, -0.05411720275878906, 3.75042724609375, -1.1629180908203125, -2.1795654296875, 0.1187286376953125, 3.065673828125, 1.223876953125, 0.637969970703125, 0.898895263671875, -0.832977294921875, 1.1164093017578125, -0.75189208984375, 0.17840576171875, -0.17855453491210938, -1.8886566162109375, -2.070098876953125, 0.1123504638671875, 1.0526123046875, 0.6464996337890625, 0.1815948486328125, 0.4842529296875, 4.168182373046875, 1.125213623046875, 1.899078369140625, 0.31864356994628906, 0.9323577880859375, -5.1422271728515625, 1.24560546875, -2.14178466796875, 3.2353515625, 1.365966796875, 0.5311279296875, -0.57501220703125, -2.558380126953125, 0.1265869140625, 4.22052001953125, -0.02858734130859375, 1.81683349609375, -2.40203857421875, -0.351287841796875, -0.5989913940429688, 1.52117919921875, 1.6046142578125, 0.6731033325195312, 2.773834228515625, 4.6800537109375, 2.1424102783203125, 1.741943359375, -0.23473739624023438, 0.4586639404296875, -0.134552001953125, 2.568023681640625, 0.7314910888671875, 0.9396514892578125, 1.3262290954589844, 1.7804946899414062, -1.3623504638671875, 1.34637451171875, -1.746368408203125, -1.2337493896484375, 0.0, -0.28729248046875, -3.086669921875, 0.5221099853515625, 2.4530792236328125, 4.340179443359375, 0.43310546875, 2.195526123046875, 2.763427734375, 1.91339111328125, -1.6759796142578125, 1.84588623046875, -0.2269287109375, -0.7520751953125, 1.199066162109375, 0.84832763671875, 0.28759002685546875, 0.3730316162109375, 2.14208984375, 0.154541015625, 1.269317626953125, -0.391387939453125, 0.8756217956542969, 2.49566650390625, 1.0041427612304688, -0.9118537902832031, 0.7328109741210938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000038.npy"}
{"epoch": 0.07958115183246073, "step": 39, "batch_size": 128, "mean": 1.247907042503357, "std": 1.8530126810073853, "min": -2.80853271484375, "p10": -0.8768684387207031, "median": 0.9792060852050781, "p90": 3.531591796874999, "max": 7.2615966796875, "pos_frac": 0.734375, "sample": [1.074432373046875, -0.42303466796875, 0.648529052734375, 1.736175537109375, 2.21710205078125, 0.825714111328125, 0.6983642578125, -0.01905059814453125, -1.287322998046875, 1.7271728515625, 0.47344970703125, 4.11505126953125, 5.33074951171875, 2.121826171875, 3.905120849609375, 0.2708740234375, -0.2710723876953125, 3.396636962890625, 1.263031005859375, 2.161224365234375, -0.43804931640625, 1.08721923828125, -0.9887351989746094, 2.207183837890625, 4.594970703125, 2.2193832397460938, 0.84661865234375, -2.4482269287109375, 1.911407470703125, 2.53082275390625, 1.586883544921875, -0.3759765625, -0.07647705078125, -0.5702056884765625, 3.925872802734375, -0.01605224609375, 0.11387252807617188, 1.1959381103515625, 1.503265380859375, -1.3220443725585938, 0.20172119140625, 2.0874195098876953, 2.447479248046875, 1.0837268829345703, 0.9942398071289062, 6.162933349609375, 4.064727783203125, -0.07142257690429688, 1.7470550537109375, 1.71929931640625, 2.1121368408203125, -0.8708419799804688, 3.3771514892578125, 2.0264892578125, 0.8926544189453125, -0.21767425537109375, -1.10125732421875, 1.4287872314453125, 0.96417236328125, 1.078033447265625, 0.34743499755859375, -0.51263427734375, -2.746368408203125, 2.2130126953125, 0.67340087890625, 2.432708740234375, 0.1990814208984375, -0.1805572509765625, 0.5167236328125, -0.6838836669921875, 0.90118408203125, 0.429046630859375, 0.44506263732910156, 3.09619140625, 0.334442138671875, -0.03466796875, 2.4451904296875, -2.80853271484375, 2.0519256591796875, 4.41302490234375, 2.9271240234375, -0.27838134765625, 2.514495849609375, 2.5216217041015625, 0.08721923828125, 1.5917510986328125, 5.520660400390625, 5.751708984375, 2.291107177734375, 1.0885162353515625, 0.7476234436035156, -0.89093017578125, -0.08555221557617188, -1.1225166320800781, 0.0, 0.3052825927734375, 0.8274192810058594, -1.84332275390625, 2.605926513671875, 2.12469482421875, 1.115234375, 0.9607162475585938, 7.2615966796875, -1.8125, 3.44586181640625, 2.2993621826171875, 0.620330810546875, 0.7855682373046875, 6.34246826171875, -1.875701904296875, -0.4388275146484375, 0.8371963500976562, 0.4302520751953125, 1.8714065551757812, 2.502777099609375, 0.14165115356445312, 2.534210205078125, 3.063079833984375, 3.04833984375, -0.28913116455078125, 2.5147781372070312, 1.7173919677734375, -0.211212158203125, 3.73162841796875, -1.144439697265625, 0.0714874267578125, 0.1632232666015625, 2.2496414184570312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000039.npy"}
{"epoch": 0.08167539267015707, "step": 40, "batch_size": 128, "mean": 0.715647280216217, "std": 1.8487064838409424, "min": -3.83428955078125, "p10": -1.1728713989257813, "median": 0.3820610046386719, "p90": 3.0483200073242185, "max": 9.5880126953125, "pos_frac": 0.625, "sample": [-0.3851947784423828, -1.005523681640625, 0.18524169921875, 0.922607421875, -0.8721580505371094, 1.4232025146484375, -1.16595458984375, 0.706298828125, 0.5343093872070312, 1.63385009765625, 1.849609375, -1.0874404907226562, 2.67633056640625, 3.51873779296875, 0.03759765625, 2.6311264038085938, -0.26354217529296875, 1.502960205078125, -0.20684242248535156, 1.0408935546875, -0.42401123046875, 3.0101318359375, -0.09092330932617188, 0.3110504150390625, 2.0284194946289062, 4.12890625, -0.831390380859375, -0.9980545043945312, 0.144195556640625, 3.0354156494140625, -0.08014297485351562, 3.135894775390625, -0.22998809814453125, 2.176116943359375, 0.719970703125, -0.24847412109375, 0.924896240234375, -1.1890106201171875, 4.2716064453125, -1.8115234375, 9.5880126953125, 1.322509765625, -0.2476806640625, 4.3931884765625, -2.0569839477539062, -0.47667694091796875, -1.0794525146484375, 2.2480010986328125, 1.2935791015625, 1.5342559814453125, -1.405120849609375, -2.820709228515625, 1.731170654296875, 1.32110595703125, 0.95526123046875, 3.4241943359375, 1.6368255615234375, 1.5175399780273438, -0.201904296875, 0.16485595703125, -0.1617431640625, 0.2838287353515625, 0.5327911376953125, -3.83428955078125, 1.51324462890625, -0.08819580078125, 0.520050048828125, 2.0376434326171875, 2.38067626953125, 0.596160888671875, -0.9656295776367188, 0.24776458740234375, 0.7379074096679688, 2.16058349609375, 0.242706298828125, 1.8501014709472656, 0.009857177734375, 0.22880172729492188, -1.322540283203125, 4.89678955078125, -0.652191162109375, 0.996307373046875, -0.34836578369140625, -0.18094635009765625, -1.315185546875, 0.6050033569335938, -1.1369972229003906, 4.183441162109375, -0.33501434326171875, 0.3020172119140625, 3.07843017578125, 0.3594970703125, 2.7449951171875, 3.3621826171875, 1.51641845703125, 1.76995849609375, 4.095855712890625, -1.5720596313476562, 1.996551513671875, 0.9100265502929688, 0.9061279296875, 0.0825042724609375, -0.653900146484375, -2.001148223876953, -0.2350006103515625, -0.6788330078125, -0.068572998046875, -0.7987327575683594, -1.135101318359375, 0.75537109375, -1.8115234375, 0.2117919921875, 1.0048980712890625, 0.40462493896484375, -1.240631103515625, -3.52435302734375, -0.8657379150390625, 1.1756439208984375, 2.46685791015625, 5.559783935546875, 1.7147674560546875, 0.031887054443359375, 1.4937286376953125, -0.047393798828125, 0.7352371215820312, 0.1949462890625, -0.43280029296875, 1.610809326171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000040.npy"}
{"epoch": 0.08376963350785341, "step": 41, "batch_size": 128, "mean": 0.7740228176116943, "std": 1.7908726930618286, "min": -5.74273681640625, "p10": -1.2662971496582032, "median": 0.8138008117675781, "p90": 2.8523193359374996, "max": 7.026702880859375, "pos_frac": 0.65625, "sample": [1.850341796875, -0.1757049560546875, 3.81439208984375, -0.6218643188476562, 1.875732421875, -0.57757568359375, 1.92181396484375, 0.6250457763671875, 0.2675323486328125, -0.694915771484375, 0.8615531921386719, 2.37115478515625, 0.05736541748046875, -1.091064453125, -0.9902801513671875, 2.483184814453125, 0.759246826171875, -1.46759033203125, 4.655303955078125, 3.63385009765625, -0.2035064697265625, 3.447906494140625, 1.6512451171875, -0.314605712890625, -2.671173095703125, 1.4595489501953125, 1.39129638671875, 2.23638916015625, -0.6469879150390625, 1.4826335906982422, 1.7113723754882812, -5.74273681640625, 0.8507499694824219, 7.026702880859375, -1.34039306640625, -0.596710205078125, 3.90570068359375, 4.597869873046875, 0.1098480224609375, 1.6159210205078125, 2.71063232421875, 0.0586090087890625, 1.772735595703125, -0.199951171875, 0.248321533203125, 1.74517822265625, 3.155303955078125, -2.776458740234375, -0.292633056640625, 2.131378173828125, 1.6313018798828125, -0.722412109375, 0.75836181640625, -1.3169021606445312, 4.29266357421875, -2.1380615234375, -1.2513351440429688, 3.5001220703125, 1.2928466796875, 0.2275848388671875, 1.199798583984375, -1.2181243896484375, -0.523529052734375, -0.492919921875, -0.22723388671875, -1.1688995361328125, -0.5715789794921875, 1.044189453125, 0.0, 1.484527587890625, 0.152313232421875, 0.81085205078125, 0.7249908447265625, 0.080474853515625, -0.264556884765625, -3.692626953125, 2.2254409790039062, -1.4911727905273438, 0.743865966796875, 2.3907470703125, 0.8167495727539062, 0.09872817993164062, 0.06325531005859375, 1.3484344482421875, 1.252197265625, -0.188201904296875, 3.0467681884765625, 1.43072509765625, 1.246612548828125, 0.8958587646484375, -0.87353515625, 2.3983612060546875, 1.44964599609375, -0.60711669921875, 1.2039794921875, 2.39324951171875, -1.321533203125, 2.4624481201171875, -0.1363067626953125, 0.6650543212890625, 4.9200439453125, -1.64117431640625, 2.34857177734375, 0.9166717529296875, 0.8599090576171875, 0.8536300659179688, 0.123504638671875, -1.2320022583007812, -0.771240234375, 1.0858306884765625, 2.0584182739257812, 2.182373046875, -0.08770751953125, -1.30120849609375, 2.963958740234375, -0.49591064453125, 2.804473876953125, 0.901031494140625, 0.5026168823242188, 0.3773345947265625, -1.6343994140625, 1.876434326171875, 2.30291748046875, 1.2347412109375, 0.9199676513671875, 1.54852294921875, 2.2664031982421875, -0.0166015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000041.npy"}
{"epoch": 0.08586387434554973, "step": 42, "batch_size": 128, "mean": 0.9031499624252319, "std": 2.4427549839019775, "min": -8.63287353515625, "p10": -1.430033874511718, "median": 0.7522964477539062, "p90": 4.1253112792968745, "max": 7.83453369140625, "pos_frac": 0.703125, "sample": [0.4939098358154297, 1.699981689453125, 5.860015869140625, -0.77679443359375, -0.26666259765625, 5.70013427734375, 2.5887908935546875, 1.54022216796875, 2.236785888671875, 0.11743927001953125, 0.550384521484375, 3.44049072265625, -2.99884033203125, 1.6610488891601562, -2.524383544921875, 0.0076389312744140625, 1.8380126953125, 1.545654296875, -1.9243087768554688, 5.51708984375, 0.05306243896484375, -0.29132843017578125, -0.52777099609375, 2.3499298095703125, 4.664947509765625, -0.4318695068359375, -0.251861572265625, 0.08417320251464844, 0.90850830078125, -0.346588134765625, -0.63037109375, 0.9083633422851562, 2.508697509765625, 0.79986572265625, 4.84027099609375, 1.7401962280273438, 0.7001190185546875, 1.7775115966796875, 0.35341644287109375, 1.07196044921875, -0.80633544921875, -0.6133060455322266, 1.894195556640625, -1.218597412109375, 3.8729248046875, 1.2943115234375, 0.32076263427734375, 0.1286468505859375, 2.39453125, 0.5223941802978516, 1.8690185546875, 1.7226905822753906, 1.7041473388671875, 0.9222412109375, 0.4993324279785156, 0.07906723022460938, 0.8280487060546875, -0.9217529296875, -1.9233856201171875, 1.9569168090820312, -5.240234375, 0.64141845703125, -0.715545654296875, -0.450103759765625, -0.09283447265625, 5.25341796875, -0.5613174438476562, 0.25311279296875, -1.0431137084960938, 1.3133525848388672, 1.761474609375, 0.40462493896484375, 5.320404052734375, 5.5506591796875, -3.6868896484375, -0.3601245880126953, -6.82415771484375, 0.6262435913085938, 5.72900390625, 0.05884552001953125, 0.5743789672851562, 0.9547119140625, 0.706024169921875, 1.8594207763671875, 4.168975830078125, 0.7985687255859375, 4.4205322265625, 0.887908935546875, -0.28139686584472656, -2.3582000732421875, -2.30303955078125, 1.07647705078125, 6.24212646484375, -0.24028396606445312, 1.94366455078125, 0.2818145751953125, -2.835784912109375, 7.83453369140625, -0.9147872924804688, 0.3672637939453125, 2.877471923828125, 1.35003662109375, 0.0, 2.39154052734375, 1.3404541015625, 1.4468536376953125, 4.106597900390625, -0.8689002990722656, 0.3457374572753906, 1.3084869384765625, 2.6212158203125, 2.447998046875, 0.2151336669921875, -0.5172939300537109, 0.98175048828125, 3.7000732421875, 1.885284423828125, -0.007965087890625, 1.3944015502929688, 0.518524169921875, 1.4678802490234375, 3.7145233154296875, -4.81048583984375, 3.4058837890625, -8.63287353515625, 2.9403533935546875, -2.3711700439453125, 0.1168365478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000042.npy"}
{"epoch": 0.08795811518324607, "step": 43, "batch_size": 128, "mean": 1.1817758083343506, "std": 2.3028178215026855, "min": -8.728668212890625, "p10": -1.0475830078125, "median": 0.8715858459472656, "p90": 4.634921264648438, "max": 6.89971923828125, "pos_frac": 0.734375, "sample": [1.37371826171875, 4.19207763671875, 1.722625732421875, -0.956268310546875, 2.0084228515625, 3.163665771484375, 3.91558837890625, 0.5996627807617188, -0.10534477233886719, 0.09159088134765625, -0.4389610290527344, 2.02813720703125, -0.1776885986328125, 5.297576904296875, 1.0511932373046875, 1.730682373046875, 0.190216064453125, 1.2669219970703125, 2.184967041015625, -5.3226318359375, 3.82049560546875, 0.7031326293945312, 3.94732666015625, 5.110687255859375, -0.7849884033203125, -0.316619873046875, -1.1117095947265625, 1.979400634765625, 0.11279678344726562, 5.744781494140625, 0.74078369140625, 2.0668716430664062, -2.4710693359375, -8.728668212890625, 4.9471435546875, 1.2445220947265625, 0.7667694091796875, 1.8405532836914062, 0.8018760681152344, -0.6121063232421875, -1.6765594482421875, 1.983245849609375, -0.6420879364013672, 4.788055419921875, 0.5512008666992188, 0.09060287475585938, -0.70550537109375, -1.2896194458007812, -3.710174560546875, 1.7179412841796875, -1.9708404541015625, 0.662017822265625, 0.015016555786132812, -1.0529556274414062, 1.41046142578125, -0.484375, 3.43536376953125, 3.6415252685546875, 0.5177154541015625, 4.68408203125, -0.089874267578125, 1.5314483642578125, 1.560211181640625, -1.87841796875, 4.492919921875, 2.686798095703125, 6.89971923828125, 0.1363525390625, 0.9460906982421875, 3.321014404296875, 4.672882080078125, 0.28875732421875, 0.6061553955078125, 0.44482421875, 0.7778778076171875, -0.10695648193359375, -0.126373291015625, -2.969207763671875, 0.1861419677734375, 0.7205810546875, 1.0195770263671875, 2.2161865234375, 1.372314453125, 1.0169830322265625, 2.010223388671875, -0.0438232421875, 1.6023101806640625, -2.6309890747070312, 0.619720458984375, -0.34029388427734375, 2.0723419189453125, 0.80926513671875, 0.8416748046875, 1.467864990234375, 4.971435546875, -0.16138839721679688, 1.2288360595703125, 1.8089599609375, 0.34307861328125, 0.36203765869140625, 4.3428955078125, 0.9014968872070312, 1.70220947265625, -1.0056838989257812, 4.959320068359375, 6.065887451171875, 4.38262939453125, 0.3468017578125, 2.93212890625, 2.815093994140625, 5.3256683349609375, 4.818145751953125, 0.5158004760742188, -1.0452804565429688, -0.621917724609375, 2.1357421875, 2.817901611328125, 1.49505615234375, 4.61865234375, 2.60064697265625, 0.01190185546875, -1.927337646484375, 0.33951759338378906, 1.80462646484375, 0.328521728515625, -0.28672027587890625, 1.429901123046875, -0.8067951202392578], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000043.npy"}
{"epoch": 0.09005235602094241, "step": 44, "batch_size": 128, "mean": 1.5971367359161377, "std": 3.101762294769287, "min": -9.436859130859375, "p10": -0.9993593215942382, "median": 1.6522064208984375, "p90": 4.8422088623046875, "max": 11.71185302734375, "pos_frac": 0.765625, "sample": [3.641845703125, 1.1330413818359375, 9.8995361328125, 1.150726318359375, 1.0692520141601562, 1.4906005859375, 0.3609962463378906, 1.165374755859375, 2.812255859375, -0.9783725738525391, 3.648529052734375, 2.208404541015625, 0.832794189453125, -0.5329322814941406, 5.7904052734375, 0.7302474975585938, 2.954254150390625, 0.62701416015625, -3.709320068359375, 1.539215087890625, 2.524169921875, 1.7359619140625, 0.10595703125, 0.5260696411132812, 0.8630638122558594, 1.9151535034179688, 2.2479248046875, 0.3316497802734375, -8.103668212890625, -0.8560409545898438, 5.1132659912109375, 2.3818511962890625, -0.8260498046875, -0.8624267578125, 1.0315380096435547, 4.2135009765625, 1.634796142578125, -1.0483283996582031, 1.9749755859375, 2.6342010498046875, 11.71185302734375, -0.162628173828125, 0.733428955078125, 4.918365478515625, 4.242279052734375, 0.92071533203125, 4.944732666015625, 3.24041748046875, 2.818206787109375, -0.3328399658203125, 0.7352695465087891, 0.219512939453125, -0.8462982177734375, 4.165191650390625, 0.23540496826171875, 6.628936767578125, -4.7655029296875, 1.7935791015625, 0.055660247802734375, 4.8095703125, 4.0446929931640625, 3.189056396484375, 3.0467147827148438, 4.379150390625, 3.972930908203125, 0.7008209228515625, 2.302703857421875, -9.436859130859375, 6.61016845703125, 3.180267333984375, -8.653045654296875, 3.372936248779297, -0.579833984375, 6.59063720703125, -0.713043212890625, 4.2629241943359375, 2.8814697265625, -4.862335205078125, -0.193450927734375, -3.942535400390625, 2.596954345703125, 0.5764617919921875, 3.875244140625, 0.6288471221923828, 4.47857666015625, -1.2961807250976562, 0.688873291015625, 4.0379638671875, 6.285400390625, 1.8402328491210938, 3.21990966796875, 4.290496826171875, 2.538299560546875, 3.675445556640625, -1.075347900390625, 3.951812744140625, -0.65240478515625, 0.653472900390625, -0.9484710693359375, -2.068359375, 4.513916015625, -0.18141746520996094, -0.8639984130859375, 3.1484375, 0.8473320007324219, 2.72711181640625, 2.75299072265625, 1.62872314453125, 0.20833587646484375, 1.460296630859375, 5.92974853515625, 1.708709716796875, 0.10098648071289062, 6.32537841796875, 2.251708984375, 4.272674560546875, 2.34130859375, 1.66961669921875, -3.72479248046875, -0.01274871826171875, -5.723419189453125, 1.9665069580078125, 0.8028564453125, 1.4163665771484375, 5.61199951171875, -0.7916717529296875, 2.9489822387695312, 4.209686279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000044.npy"}
{"epoch": 0.09214659685863874, "step": 45, "batch_size": 128, "mean": 1.3486623764038086, "std": 2.882704973220825, "min": -5.891754150390625, "p10": -1.851513671875, "median": 0.8795623779296875, "p90": 5.1312744140625, "max": 12.47979736328125, "pos_frac": 0.6875, "sample": [2.171295166015625, -2.323822021484375, 5.05645751953125, -0.8891372680664062, 0.1553363800048828, 0.75335693359375, -0.009204864501953125, 0.484100341796875, 1.0701332092285156, 0.75799560546875, 4.622772216796875, 5.08001708984375, 0.2904052734375, -0.0207977294921875, -1.672332763671875, -1.700958251953125, -0.0006427764892578125, 2.463623046875, 0.0, -2.2811279296875, 0.477142333984375, 2.1978912353515625, 0.114288330078125, -0.016704559326171875, -0.8953857421875, 6.925323486328125, 1.1100234985351562, 1.27447509765625, -0.00970458984375, 2.7927627563476562, 0.74798583984375, 0.5034637451171875, 1.7515869140625, 3.2675018310546875, -0.3054084777832031, 4.053863525390625, 1.743408203125, 0.07488632202148438, 5.450225830078125, 5.65264892578125, 5.033782958984375, 3.241455078125, 1.8195877075195312, -1.78985595703125, 12.47979736328125, 0.8626823425292969, 4.9684295654296875, -2.1563796997070312, 5.55828857421875, 2.5841445922851562, -1.0325927734375, 2.0780715942382812, -1.3251953125, -2.7916259765625, 0.9350757598876953, 1.115478515625, 0.9901657104492188, 2.0412750244140625, -0.3164939880371094, -5.891754150390625, -4.59930419921875, 2.567169189453125, -0.233734130859375, 2.2861175537109375, -2.410888671875, 2.2050323486328125, -1.128204345703125, 1.3701019287109375, 3.4213409423828125, 1.4796371459960938, 0.278228759765625, 2.44122314453125, -0.57122802734375, 0.7188835144042969, 7.828094482421875, 1.1663055419921875, 5.1746826171875, 0.179168701171875, 7.187042236328125, 1.4857177734375, 1.6751785278320312, 6.71990966796875, 0.536468505859375, 9.069976806640625, 0.9939193725585938, -0.071624755859375, -5.78094482421875, 0.6545867919921875, 2.0914535522460938, -1.8687744140625, 3.77764892578125, 4.74798583984375, 4.46783447265625, -1.3820037841796875, 5.1126708984375, 0.3289337158203125, 1.76849365234375, 4.53582763671875, 1.18011474609375, -1.0567626953125, -1.884521484375, -0.258819580078125, 2.7535171508789062, -0.2708282470703125, -0.7323760986328125, -2.72723388671875, -0.1883544921875, 0.5630035400390625, 0.20888519287109375, 0.8964424133300781, 0.4839057922363281, -1.1212615966796875, 1.725341796875, 2.765380859375, 1.9175262451171875, -1.8441162109375, 3.211669921875, -3.1336669921875, 7.666168212890625, 2.3807373046875, 1.028778076171875, 5.499603271484375, 0.2909698486328125, 0.5038299560546875, 6.9874267578125, 0.51715087890625, -3.057861328125, 0.7791213989257812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000045.npy"}
{"epoch": 0.09424083769633508, "step": 46, "batch_size": 128, "mean": 1.482360601425171, "std": 2.6986091136932373, "min": -7.51177978515625, "p10": -1.1688690185546873, "median": 0.9503307342529297, "p90": 5.122030639648438, "max": 11.7021484375, "pos_frac": 0.7109375, "sample": [3.17840576171875, 3.7799072265625, 1.1351318359375, -0.07781982421875, 1.251739501953125, 0.5899658203125, 5.943115234375, 5.43707275390625, 0.8432464599609375, -0.5915756225585938, -2.115570068359375, 0.6238555908203125, -1.836181640625, 0.29656982421875, 2.722076416015625, -0.0943603515625, 1.49432373046875, -0.039752960205078125, 1.7187652587890625, -1.10565185546875, 0.013397216796875, 1.5807647705078125, 0.972900390625, -1.9275665283203125, 3.1819305419921875, 0.992340087890625, -2.657257080078125, 0.5638427734375, 6.23614501953125, -0.7334136962890625, -0.31770896911621094, 0.9277610778808594, 4.616546630859375, 5.6935577392578125, 6.2236328125, 3.3165283203125, 5.09814453125, 1.1766586303710938, -0.68310546875, 2.582550048828125, 5.105621337890625, 2.487518310546875, 0.6214218139648438, 5.6201324462890625, 0.583740234375, 0.69317626953125, 2.667236328125, -0.2669219970703125, -0.7500114440917969, 1.5979080200195312, 5.800018310546875, 3.7963714599609375, 2.4940338134765625, 1.8034820556640625, 0.28631591796875, 4.3771820068359375, 1.80108642578125, -0.398193359375, 1.5823974609375, 0.24090576171875, 0.5525283813476562, 1.4816322326660156, 0.35301971435546875, -0.97955322265625, -0.77001953125, 0.6023330688476562, 0.2674674987792969, -3.7080078125, -1.316375732421875, 3.267181396484375, -4.478363037109375, -0.3314208984375, 5.725433349609375, -1.837188720703125, 0.737762451171875, 3.9623565673828125, -1.9410057067871094, 0.15357208251953125, 1.5392875671386719, 2.960540771484375, 0.8638076782226562, 2.72296142578125, 6.4312744140625, 2.5762939453125, 0.097320556640625, 1.5379791259765625, -7.51177978515625, -1.0792617797851562, 1.5279693603515625, 5.117279052734375, -0.5794525146484375, 4.922637939453125, 5.0323486328125, 0.5141525268554688, 0.8956756591796875, 4.9510498046875, 0.8019828796386719, -0.1971282958984375, -2.214324951171875, 2.659423828125, -0.6456813812255859, 3.347930908203125, -2.3014068603515625, -0.0081634521484375, 11.7021484375, 3.1703948974609375, -0.652679443359375, 0.64410400390625, 5.13311767578125, 4.343963623046875, 2.797210693359375, -4.277313232421875, -0.13122940063476562, -0.22141075134277344, 0.19376182556152344, 1.727813720703125, 7.462982177734375, 1.0154190063476562, 0.316650390625, 2.760162353515625, 0.0, 2.63720703125, 2.6094970703125, -0.689727783203125, 0.19512176513671875, 6.67279052734375, 2.45904541015625, 4.014739990234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000046.npy"}
{"epoch": 0.09633507853403141, "step": 47, "batch_size": 128, "mean": 1.0569851398468018, "std": 3.506481170654297, "min": -11.587127685546875, "p10": -2.538148498535156, "median": 1.1863555908203125, "p90": 5.16317138671875, "max": 10.87261962890625, "pos_frac": 0.6640625, "sample": [-2.01019287109375, 1.2351207733154297, 1.7277374267578125, -0.5006866455078125, 6.61431884765625, -5.5587158203125, -5.2314453125, 2.570587158203125, 5.371307373046875, -2.123199462890625, 3.221221923828125, 1.2230682373046875, -0.5008010864257812, -3.229694366455078, -7.23919677734375, -0.585845947265625, 1.1496429443359375, -1.4143829345703125, 8.1005859375, -6.81707763671875, 3.43634033203125, -6.5843505859375, 0.7507858276367188, -0.85980224609375, 3.583282470703125, 2.222564697265625, -0.2998046875, -1.7533721923828125, -7.249755859375, -4.9571075439453125, 5.215850830078125, 0.0, 2.304290771484375, -3.3712234497070312, -1.263397216796875, 3.2928924560546875, 3.2989501953125, 2.4149169921875, 0.844390869140625, 1.6690292358398438, 0.84857177734375, -0.47485923767089844, 10.87261962890625, 5.140594482421875, 2.726104736328125, 0.71832275390625, 2.127197265625, 7.6768798828125, 1.3142852783203125, 0.4129905700683594, 0.1256866455078125, -0.6717567443847656, 6.942352294921875, 1.7049102783203125, 0.15509033203125, 4.868865966796875, -7.312286376953125, 6.42596435546875, 3.146148681640625, 2.2479171752929688, -0.774078369140625, 3.00634765625, -0.7358551025390625, 4.1759033203125, 2.5298538208007812, 0.33858489990234375, -2.471282958984375, 0.978424072265625, 0.3170623779296875, 6.2395172119140625, 2.4601287841796875, 0.08242034912109375, -0.779541015625, 0.64715576171875, 0.25262451171875, 3.9872055053710938, 0.0, 7.26959228515625, 7.29327392578125, 8.6220703125, 2.90948486328125, -1.9134521484375, 0.09763336181640625, -2.6941680908203125, 2.1923828125, -1.585357666015625, 0.30890655517578125, -1.644500732421875, 1.3803863525390625, 5.4269256591796875, 4.156894683837891, 2.023101806640625, -0.7488632202148438, 4.812591552734375, 1.2717437744140625, 0.968658447265625, 3.829254150390625, 1.731292724609375, 5.0350341796875, 1.3038482666015625, 0.00823974609375, 1.712005615234375, 0.271331787109375, 2.11309814453125, 4.84332275390625, -2.01251220703125, -0.7237548828125, 1.825408935546875, 1.0294342041015625, -0.7199516296386719, 2.846923828125, -0.6656112670898438, 0.0311279296875, 1.2398929595947266, -0.297821044921875, 4.104949951171875, -1.8513031005859375, 2.922760009765625, -0.7777938842773438, 1.781402587890625, -0.2688140869140625, 1.91925048828125, 2.0693817138671875, 4.67401123046875, 2.036651611328125, -11.587127685546875, -5.672607421875, 4.4485626220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000047.npy"}
{"epoch": 0.09842931937172775, "step": 48, "batch_size": 128, "mean": 1.5294491052627563, "std": 3.239652395248413, "min": -6.513519287109375, "p10": -1.8501495361328122, "median": 1.1332359313964844, "p90": 5.433256530761717, "max": 17.295654296875, "pos_frac": 0.6875, "sample": [2.1112213134765625, 1.59820556640625, 1.7194442749023438, 6.3686676025390625, 4.151763916015625, 2.082000732421875, 2.4749755859375, 8.373992919921875, 3.688934326171875, 0.1861572265625, -1.2707138061523438, -0.6952362060546875, 0.7667655944824219, -6.513519287109375, -2.6744651794433594, 10.34442138671875, -3.18560791015625, -4.9843597412109375, 1.689788818359375, -0.12537765502929688, 3.054718017578125, 0.8007354736328125, 4.478302001953125, 2.073577880859375, 2.3193893432617188, 0.6973495483398438, 0.7149887084960938, 0.5596923828125, -0.656494140625, -2.004364013671875, 2.0656471252441406, 7.440948486328125, 10.0914306640625, 2.68658447265625, 4.330810546875, 2.194263458251953, 0.10885810852050781, 2.67718505859375, 1.4153213500976562, 3.40087890625, 0.215301513671875, 2.068206787109375, 0.607421875, -0.011981964111328125, 1.9219894409179688, 0.41965484619140625, -3.500274658203125, 1.1854705810546875, -0.63848876953125, -1.2403182983398438, 8.29913330078125, 1.651815414428711, -0.8885650634765625, 1.0810012817382812, 1.4757080078125, 4.358551025390625, 0.5107040405273438, -0.29741668701171875, -0.0077571868896484375, 2.329071044921875, -1.404449462890625, 3.741668701171875, 0.6835479736328125, 0.3414459228515625, 1.4002685546875, -4.372955322265625, 2.709503173828125, -1.582305908203125, 3.7474822998046875, 2.8465576171875, -2.8251495361328125, 1.0709228515625, 0.0, 1.51715087890625, 3.24517822265625, 1.7130317687988281, 4.130859375, -0.2534523010253906, 0.7120819091796875, 1.846954345703125, 1.2428436279296875, 5.2731170654296875, -0.8538055419921875, 8.81292724609375, -0.82489013671875, 5.955078125, 2.667449951171875, 0.6236038208007812, -0.34644126892089844, 2.305389404296875, -1.1392364501953125, -0.294219970703125, -0.23992919921875, 0.9979324340820312, 4.0584716796875, 7.888671875, 2.612640380859375, 1.2813873291015625, 0.2067718505859375, 1.4162368774414062, -1.11749267578125, 4.694305419921875, 0.0, 6.847869873046875, -1.1326904296875, 4.3987884521484375, 17.295654296875, 0.9071578979492188, -3.441864013671875, 2.8029708862304688, -0.6444854736328125, -1.443389892578125, -1.7840576171875, 1.95648193359375, 0.9421043395996094, -2.0222015380859375, -2.0305938720703125, -2.9043750762939453, 0.38729095458984375, 1.9630661010742188, 1.4166259765625, 0.1536865234375, -2.0554656982421875, -0.0172119140625, 0.18243408203125, 2.693115234375, 6.906402587890625, 5.806915283203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000048.npy"}
{"epoch": 0.10052356020942409, "step": 49, "batch_size": 128, "mean": 1.6482040882110596, "std": 3.48956298828125, "min": -9.236358642578125, "p10": -2.3019973754882814, "median": 1.1149969100952148, "p90": 6.070468139648437, "max": 10.25299072265625, "pos_frac": 0.640625, "sample": [-0.28936767578125, 4.29296875, 2.0563430786132812, 1.1293258666992188, -2.8624114990234375, 3.79681396484375, 4.3974609375, 6.5111083984375, 5.786163330078125, 0.8784809112548828, -1.1979141235351562, 1.410675048828125, -5.67340087890625, 2.585906982421875, -0.729888916015625, 10.25299072265625, 0.7083892822265625, 2.2992401123046875, -1.145050048828125, 5.0361328125, -1.469268798828125, -0.5590248107910156, 4.19500732421875, 4.07586669921875, 6.75738525390625, -0.217681884765625, -2.3301849365234375, 9.69427490234375, 0.6010055541992188, 5.959625244140625, 1.1919021606445312, 4.8297271728515625, 5.0006103515625, 2.3427352905273438, 0.845703125, 1.191640853881836, -0.16060638427734375, 2.3749542236328125, 0.0295562744140625, -0.657379150390625, 1.2458267211914062, 7.2154541015625, -2.7254791259765625, 0.7897567749023438, 4.099937438964844, 2.5047760009765625, 1.8156280517578125, 5.649169921875, -0.02362060546875, -1.216552734375, -0.228240966796875, -6.7816619873046875, 0.6720085144042969, 5.6431121826171875, -2.7301807403564453, 1.42333984375, -0.5714263916015625, 0.43668365478515625, -0.412139892578125, -0.9676742553710938, -0.3299217224121094, 6.70770263671875, 4.191650390625, 2.39752197265625, -0.7532958984375, 2.246734619140625, -0.16851806640625, -3.036224365234375, 5.44696044921875, 7.16143798828125, 3.72308349609375, 5.50994873046875, -0.8034133911132812, -1.4934196472167969, 1.100667953491211, 5.657135009765625, -9.236358642578125, 5.103973388671875, 7.1985015869140625, -1.325592041015625, 0.6759719848632812, 4.55975341796875, 6.3291015625, 0.08878326416015625, 0.6695404052734375, 3.840850830078125, -1.4661865234375, 5.11199951171875, 7.71160888671875, 5.04827880859375, 2.009857177734375, -0.17958450317382812, 4.178489685058594, 9.019866943359375, 3.0577163696289062, -1.87054443359375, -0.2397308349609375, -0.338043212890625, 1.1595516204833984, 0.9749755859375, -1.44378662109375, 0.768035888671875, -4.37384033203125, -2.470611572265625, 8.264251708984375, -0.672515869140625, -2.631927490234375, 5.93804931640625, 4.8398284912109375, 0.5553855895996094, 4.380012512207031, -0.3316192626953125, 2.6527748107910156, 1.2287635803222656, 3.25457763671875, 2.946380615234375, -0.19914627075195312, 7.6768646240234375, 2.2803878784179688, 0.8951416015625, -8.65655517578125, 0.6586208343505859, 0.796112060546875, -2.2899169921875, 3.0007781982421875, -2.628570556640625, -1.3087158203125, -0.5779876708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000049.npy"}
{"epoch": 0.10261780104712041, "step": 50, "batch_size": 128, "mean": 1.1223524808883667, "std": 3.9583282470703125, "min": -11.33135986328125, "p10": -3.3684906005859374, "median": 0.6664142608642578, "p90": 5.947853088378905, "max": 12.301361083984375, "pos_frac": 0.609375, "sample": [2.415069580078125, -3.2003250122070312, 1.87255859375, -0.87274169921875, -2.11517333984375, 1.095062255859375, -0.513275146484375, 0.24776458740234375, 1.903411865234375, 12.301361083984375, -1.447021484375, 0.12722396850585938, 2.3428955078125, 4.201934814453125, 9.58355712890625, 1.6943511962890625, 3.2354278564453125, -8.1387939453125, 4.860260009765625, 0.566925048828125, 0.22064208984375, -2.7191162109375, -5.350738525390625, 4.247802734375, 2.56304931640625, 3.9205322265625, 3.100231170654297, -0.25998687744140625, 4.34149169921875, 1.980133056640625, -0.038818359375, 10.26953125, 4.4929351806640625, 0.665191650390625, 2.346435546875, 1.7889251708984375, -0.06758880615234375, 2.744476318359375, -1.143798828125, 1.14630126953125, -1.6006546020507812, -4.9255218505859375, 3.8854827880859375, 6.43841552734375, 0.22927093505859375, 4.58648681640625, 1.91400146484375, 4.929046630859375, -0.8104228973388672, -0.4009857177734375, 8.3890380859375, -0.3014068603515625, 3.4733200073242188, 0.6552085876464844, 1.598419189453125, -1.50860595703125, 6.99169921875, -2.188385009765625, 2.512359619140625, 4.4925537109375, 0.3126678466796875, 0.4544525146484375, -5.321136474609375, 0.8575592041015625, -2.244354248046875, -3.49664306640625, -1.07427978515625, -1.028778076171875, 2.7124099731445312, -0.7610816955566406, -0.597869873046875, -1.4337940216064453, -0.6405544281005859, 1.0840301513671875, 0.2003173828125, -0.673583984375, 4.8521728515625, -0.505279541015625, 8.6077880859375, -0.12591171264648438, 1.3446044921875, -0.2784881591796875, -11.33135986328125, -3.313568115234375, -0.1612262725830078, 6.23565673828125, -3.562255859375, -4.053497314453125, 3.58148193359375, -4.5933837890625, 6.449981689453125, -3.998046875, 11.2530517578125, 0.517974853515625, 0.5863800048828125, -2.044464111328125, -0.8204269409179688, -0.0500640869140625, 0.3544464111328125, 2.62646484375, 3.77685546875, 4.29388427734375, 2.0310516357421875, 5.8245086669921875, -0.812103271484375, 0.382049560546875, -0.23607635498046875, 4.957489013671875, 0.7611083984375, -10.669921875, -7.251220703125, 2.3794326782226562, 11.06787109375, 3.2242431640625, -0.5523681640625, -5.90716552734375, 5.4246826171875, 3.724273681640625, -2.512054443359375, 0.6676368713378906, 1.8441162109375, 1.1587696075439453, 7.908782958984375, 1.581787109375, 1.13336181640625, 1.050384521484375, -1.356689453125, 7.077606201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000050.npy"}
{"epoch": 0.10471204188481675, "step": 51, "batch_size": 128, "mean": 1.8258390426635742, "std": 4.971269607543945, "min": -13.04144287109375, "p10": -3.165008544921875, "median": 1.3413734436035156, "p90": 8.291357421874999, "max": 18.0653076171875, "pos_frac": 0.6328125, "sample": [3.46978759765625, -0.39594268798828125, 4.151613235473633, 1.98822021484375, 1.2206611633300781, -1.5137100219726562, 4.2674713134765625, 7.5028076171875, -4.149017333984375, -0.4736785888671875, 5.763031005859375, 6.161102294921875, 6.25701904296875, 3.19451904296875, -1.2168540954589844, -0.991424560546875, 8.96234130859375, -0.08428955078125, 2.033355712890625, 5.5689544677734375, -13.04144287109375, -2.9068756103515625, 1.112152099609375, -2.1815948486328125, 5.52593994140625, 18.0653076171875, 0.0950469970703125, 1.098388671875, 10.0296630859375, 0.6298828125, -0.44854736328125, 9.530120849609375, 9.8070068359375, 5.57965087890625, 1.8235397338867188, 9.1158447265625, 10.6646728515625, 4.360588073730469, 2.5838623046875, 0.046764373779296875, 2.9275588989257812, 5.269317626953125, 8.488616943359375, 3.9864501953125, -0.2980461120605469, 7.268035888671875, 5.346832275390625, 1.2872238159179688, -4.663360595703125, -7.8338623046875, 7.491729736328125, -2.203094482421875, -13.014739990234375, 3.38165283203125, -1.75262451171875, 5.9781646728515625, 2.0730667114257812, 1.8595962524414062, -8.3138427734375, 5.8345947265625, -0.3673095703125, 1.3680419921875, -1.9807586669921875, 1.2112274169921875, -11.406951904296875, -3.431549072265625, -0.09928703308105469, -1.7261199951171875, 3.756622314453125, 7.438385009765625, 1.8584136962890625, 1.16241455078125, 0.2591400146484375, 5.300048828125, -3.2374267578125, 0.0, 9.79150390625, -5.0548858642578125, 1.3466262817382812, 7.139007568359375, 2.108673095703125, -0.5604782104492188, -0.0991973876953125, 2.2952117919921875, -10.056182861328125, -1.21630859375, -0.624481201171875, 1.9107894897460938, -3.60174560546875, 12.89581298828125, -0.3052558898925781, -3.13397216796875, 0.3256072998046875, -0.0743560791015625, 0.7815055847167969, 2.6483306884765625, -2.268585205078125, 2.314788818359375, 8.491485595703125, -1.798858642578125, -2.44970703125, -2.2575836181640625, -0.7904205322265625, 7.142059326171875, 12.9527587890625, 10.08734130859375, 6.860809326171875, 8.206817626953125, 1.33612060546875, 0.76116943359375, 0.393310546875, 1.0454654693603516, 3.1880950927734375, 2.6854782104492188, -0.6390838623046875, -0.887420654296875, -3.0709228515625, -4.643707275390625, 1.03497314453125, 3.085540771484375, 7.8592987060546875, 1.65155029296875, 2.4599533081054688, -0.2973480224609375, 2.357452392578125, 4.871063232421875, -0.63623046875, 1.721435546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000051.npy"}
{"epoch": 0.1068062827225131, "step": 52, "batch_size": 128, "mean": 3.051837921142578, "std": 5.002547264099121, "min": -7.58831787109375, "p10": -1.9246002197265621, "median": 2.091064453125, "p90": 9.98106689453125, "max": 19.0938720703125, "pos_frac": 0.7890625, "sample": [4.9380340576171875, 2.100677490234375, -7.04156494140625, 1.5802001953125, 0.532257080078125, 1.62109375, 10.35638427734375, 1.9640884399414062, 1.1109886169433594, 9.962646484375, 1.63720703125, 2.224395751953125, -2.2836456298828125, -7.58831787109375, 2.931854248046875, 12.0811767578125, 4.3982086181640625, 3.9836044311523438, 0.3031158447265625, -0.01573944091796875, 6.75286865234375, 1.7325897216796875, 2.0019149780273438, 0.7936019897460938, 4.1555328369140625, -0.9908905029296875, 0.811737060546875, 7.367706298828125, 5.543426513671875, -2.2022247314453125, -7.226806640625, -0.37633514404296875, 1.2044906616210938, 2.081451416015625, 3.6573486328125, 0.33101654052734375, 0.6482162475585938, -3.32232666015625, 2.504058837890625, 1.7960205078125, 14.41424560546875, 0.4027290344238281, -1.8056182861328125, 0.6600341796875, 9.42584228515625, 1.7737045288085938, 6.8902587890625, 4.175048828125, 4.84771728515625, 8.4642333984375, 2.5692214965820312, 2.4801483154296875, 2.811187744140625, 2.757171630859375, -0.026594161987304688, 4.821990966796875, -0.5199813842773438, 2.87933349609375, 1.4429702758789062, 5.092803955078125, 1.6855621337890625, 1.6661224365234375, 6.931365966796875, -4.30572509765625, 9.79058837890625, 1.915008544921875, -3.50775146484375, -1.1673851013183594, 0.2110595703125, -0.27995872497558594, 3.32098388671875, 2.7538833618164062, 9.626556396484375, 4.1247711181640625, -4.5540771484375, -5.9449462890625, 1.6202392578125, 8.558349609375, 6.062591552734375, 12.46624755859375, 2.162872314453125, 11.815673828125, 0.655059814453125, 2.8236083984375, 11.023193359375, -4.804595947265625, 0.8623924255371094, 5.500701904296875, 19.0938720703125, 3.198883056640625, 2.9886474609375, 0.8529510498046875, 4.859405517578125, 14.541213989257812, 4.79791259765625, 0.7048072814941406, 0.09930419921875, 12.217864990234375, 1.1964187622070312, 0.490570068359375, 0.261474609375, 11.0106201171875, 3.8173828125, 2.27325439453125, 17.30322265625, 6.497249603271484, -7.33551025390625, 2.4374847412109375, 15.3875732421875, 9.860595703125, -0.83551025390625, 10.0240478515625, -0.8290786743164062, -1.6858596801757812, 7.725921630859375, 0.08738517761230469, 0.751861572265625, 2.062358856201172, 1.8987960815429688, 6.4976806640625, 3.52349853515625, 3.9693145751953125, -6.61602783203125, -1.1049652099609375, 3.21453857421875, -0.24660491943359375, 9.8695068359375, -0.855743408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000052.npy"}
{"epoch": 0.10890052356020942, "step": 53, "batch_size": 128, "mean": 2.216883659362793, "std": 5.618834972381592, "min": -14.34771728515625, "p10": -3.4587722778320313, "median": 1.4286117553710938, "p90": 10.097047424316406, "max": 15.923095703125, "pos_frac": 0.640625, "sample": [6.52899169921875, -3.4411163330078125, -8.415664672851562, -4.9101104736328125, 3.2645416259765625, -1.15911865234375, 10.492156982421875, -7.798065185546875, -0.04840087890625, -2.3670196533203125, -1.51788330078125, -2.34906005859375, -5.4965362548828125, -9.4322509765625, 2.1206207275390625, 7.369873046875, 5.3647308349609375, -1.573089599609375, 5.31878662109375, 1.064178466796875, -1.672637939453125, -1.1753692626953125, -1.83209228515625, 9.3736572265625, 10.7962646484375, 12.1356201171875, 4.7869720458984375, 9.14166259765625, 6.27154541015625, 9.778778076171875, 0.65533447265625, 9.709930419921875, 0.171661376953125, 2.281463623046875, 2.8613128662109375, -0.728485107421875, 8.113311767578125, -14.34771728515625, 0.18688011169433594, 3.7964096069335938, -0.491424560546875, 1.278839111328125, -0.7313804626464844, -2.0335693359375, 0.8766326904296875, 3.3607177734375, -3.027679443359375, -2.0645523071289062, 6.13238525390625, -5.4127197265625, 5.698455810546875, 12.161895751953125, 1.5743789672851562, -9.5806884765625, 0.0084381103515625, -1.4268798828125, 7.76641845703125, 1.6413421630859375, -0.463592529296875, 9.064437866210938, 10.119552612304688, 2.9437713623046875, -2.9247665405273438, -0.529083251953125, 5.5565338134765625, -2.610881805419922, 10.08740234375, -0.0067596435546875, 4.047264099121094, -1.7046566009521484, 1.3798370361328125, 5.370361328125, 15.071319580078125, 6.4281463623046875, 12.139434814453125, 11.17962646484375, -3.4342498779296875, 2.158905029296875, 5.74505615234375, 0.53271484375, 3.6448974609375, 1.477386474609375, 5.833709716796875, 1.540374755859375, -1.2734222412109375, 9.749755859375, -11.3175048828125, 10.93365478515625, 1.7874031066894531, -3.499969482421875, 2.8697128295898438, 1.2944183349609375, 11.322052001953125, 0.7752304077148438, 0.83612060546875, 15.923095703125, -0.7538490295410156, -6.880828857421875, 2.2367172241210938, 0.2588653564453125, -0.4252605438232422, 7.78125, 1.266448974609375, -6.42218017578125, -0.27124786376953125, -1.2775344848632812, 0.023040771484375, 5.8759918212890625, 0.23681640625, -9.75250244140625, -0.5529556274414062, 0.35054779052734375, 6.4984588623046875, 5.987274169921875, 2.4295196533203125, 0.99560546875, 11.465301513671875, 0.0, 9.99725341796875, -0.321502685546875, 8.021392822265625, -2.694091796875, 1.7606086730957031, 4.0067138671875, 2.76153564453125, 2.7733612060546875, 7.158966064453125, 10.129425048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000053.npy"}
{"epoch": 0.11099476439790576, "step": 54, "batch_size": 128, "mean": 1.856428861618042, "std": 5.586865425109863, "min": -11.36065673828125, "p10": -5.2593948364257805, "median": 1.2746047973632812, "p90": 8.98740844726562, "max": 16.9866943359375, "pos_frac": 0.625, "sample": [8.46380615234375, 6.554534912109375, 8.0447998046875, 4.283905029296875, 3.32708740234375, 1.6343841552734375, -2.76776123046875, 12.40838623046875, 0.18218231201171875, -3.10101318359375, 1.242706298828125, -5.1160736083984375, 0.06918144226074219, -1.8392486572265625, 11.01202392578125, -0.28127288818359375, 11.542083740234375, 4.466163635253906, -6.457550048828125, -2.6857757568359375, 2.68975830078125, 8.535888671875, 6.2488250732421875, 0.48846435546875, -4.7452850341796875, 3.089569091796875, 8.39312744140625, 1.1799774169921875, 3.09912109375, -0.4153594970703125, 1.1854629516601562, 7.1798095703125, -3.2854766845703125, 2.3591079711914062, -0.6064453125, 8.001983642578125, 13.7760009765625, -6.56817626953125, 6.6114501953125, 2.436603546142578, 5.109798431396484, 2.169586181640625, 14.3006591796875, -2.96331787109375, 14.10064697265625, 1.33221435546875, -6.507293701171875, -3.55279541015625, -0.87835693359375, -7.664093017578125, 6.709075927734375, 11.544158935546875, 0.7713623046875, 1.3065032958984375, -0.3969402313232422, 2.882915496826172, -0.5359954833984375, -0.01580810546875, 4.73443603515625, -1.1877975463867188, 0.9824581146240234, 3.6829681396484375, -4.4161376953125, -5.84515380859375, 2.0467987060546875, 3.8155136108398438, -3.88385009765625, 3.41571044921875, 8.024261474609375, 1.329376220703125, -1.3324012756347656, 2.2448463439941406, 5.050323486328125, 2.704864501953125, -2.8079833984375, 2.4984588623046875, 3.401947021484375, -3.149566650390625, 3.2966156005859375, 0.34060096740722656, -4.544578552246094, -5.59381103515625, 0.33100128173828125, -2.6590347290039062, 7.2757568359375, 3.6092529296875, -3.448028564453125, 4.1393280029296875, 3.7757644653320312, 4.1760101318359375, -0.962432861328125, 10.04095458984375, -6.193939208984375, 5.684600830078125, 4.7764892578125, 0.08807373046875, -6.0598297119140625, 16.9866943359375, 16.583740234375, 1.576416015625, 8.37127685546875, 5.002777099609375, 8.486495971679688, 0.78961181640625, 11.474945068359375, -4.044013977050781, -7.2073974609375, -0.749725341796875, -11.36065673828125, 12.188034057617188, 0.836029052734375, -5.6654052734375, 0.0411376953125, -3.300506591796875, 11.110626220703125, -0.31103515625, -2.72686767578125, -0.713165283203125, -0.43438720703125, 1.170684814453125, -3.483978271484375, -0.2518310546875, 3.3232078552246094, -9.76971435546875, -7.21112060546875, 1.0977020263671875, 3.607959747314453, 4.474212646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000054.npy"}
{"epoch": 0.1130890052356021, "step": 55, "batch_size": 128, "mean": 2.2327780723571777, "std": 4.963666915893555, "min": -15.0863037109375, "p10": -3.042767333984375, "median": 1.5206451416015625, "p90": 9.491178894042966, "max": 20.190643310546875, "pos_frac": 0.640625, "sample": [-3.7255401611328125, -1.892120361328125, 0.52850341796875, -2.668792724609375, 1.541900634765625, -3.94219970703125, 5.169303894042969, 1.18408203125, 5.3798828125, -1.1219024658203125, 4.314239501953125, 4.401039123535156, 5.368408203125, 4.530609130859375, 0.801544189453125, 0.8862571716308594, 2.2523231506347656, -3.352294921875, 2.6102333068847656, 2.794025421142578, 4.886604309082031, 2.048095703125, 2.664093017578125, 10.294418334960938, 10.3594970703125, -0.5495758056640625, 0.0526275634765625, 10.881591796875, 1.5200042724609375, -4.46563720703125, 0.183990478515625, -0.22574234008789062, -1.2346038818359375, 2.924407958984375, 2.661407470703125, 2.783721923828125, -1.1007080078125, 5.56396484375, -0.0899658203125, 12.65728759765625, 0.92425537109375, -1.30291748046875, -0.0823211669921875, 1.387603759765625, 2.42388916015625, 3.9326934814453125, 1.571197509765625, 0.0284576416015625, 3.430419921875, 10.454345703125, 1.2433700561523438, -2.7588043212890625, 3.34271240234375, 2.0556793212890625, -1.543975830078125, 2.1936874389648438, -7.972564697265625, 7.903350830078125, -1.0934333801269531, 1.3040771484375, -0.9348602294921875, 2.007781982421875, 14.01190185546875, 1.0437507629394531, 14.128265380859375, -0.56890869140625, -3.0644073486328125, 1.30450439453125, 4.86322021484375, -2.2361278533935547, -3.7541656494140625, 2.8172874450683594, -3.0334930419921875, -0.5205078125, -0.20423126220703125, -0.1420116424560547, -1.6551513671875, 3.89923095703125, 3.0950279235839844, 4.722434997558594, 1.446563720703125, 12.85247802734375, 4.45111083984375, -0.3717002868652344, -0.0119476318359375, -3.104156494140625, -1.9803428649902344, 10.113174438476562, 8.379608154296875, 1.5212860107421875, -0.685516357421875, 5.7754669189453125, 4.700714111328125, 2.516357421875, 13.232177734375, -1.4916305541992188, 8.042877197265625, -1.8820457458496094, 1.9754371643066406, 6.3035125732421875, -3.85955810546875, 5.234466552734375, 2.920574188232422, 10.21917724609375, 1.0074920654296875, -4.86566162109375, 20.190643310546875, 2.43328857421875, 5.0135955810546875, -1.7865142822265625, -15.0863037109375, 1.1105613708496094, 7.64068603515625, -1.1906661987304688, -0.871612548828125, -0.3330841064453125, 3.353057861328125, -4.36785888671875, 6.040546417236328, -2.106903076171875, 0.19037628173828125, 3.1993865966796875, -5.056976318359375, 9.224609375, -1.795928955078125, 14.498260498046875, 7.6574859619140625, 5.302768707275391], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000055.npy"}
{"epoch": 0.11518324607329843, "step": 56, "batch_size": 128, "mean": 3.1832268238067627, "std": 6.551545143127441, "min": -14.521453857421875, "p10": -2.9638244628906247, "median": 2.149364471435547, "p90": 10.8928466796875, "max": 26.287750244140625, "pos_frac": 0.671875, "sample": [3.96209716796875, -10.500411987304688, 6.605377197265625, 12.72076416015625, 3.532989501953125, 0.4665374755859375, -0.51611328125, 3.804046630859375, -1.4692230224609375, 9.84759521484375, 9.8575439453125, 7.692535400390625, -0.6483154296875, -1.8966751098632812, 0.13970947265625, 0.563812255859375, -2.2639122009277344, -2.1605911254882812, 11.02008056640625, -4.66796875, 2.44140625, 0.881439208984375, 8.843505859375, -5.621734619140625, -0.6869964599609375, 26.287750244140625, 9.310943603515625, 10.217529296875, 8.590011596679688, -1.05078125, 12.2568359375, 6.268524169921875, 3.2385101318359375, -1.45703125, 14.63787841796875, 3.958984375, -6.599365234375, 3.9314117431640625, 5.9001312255859375, 0.7016677856445312, 17.627944946289062, 1.9017333984375, -0.3146820068359375, -0.13068199157714844, 1.0659027099609375, 21.2666015625, 5.464752197265625, -1.1105327606201172, 1.0189361572265625, 1.397705078125, -1.076904296875, -2.1539382934570312, 8.772750854492188, -8.973876953125, 6.975311279296875, -5.38372802734375, 0.21506690979003906, -1.8205299377441406, 6.264556884765625, 0.8349761962890625, 10.693244934082031, 14.66888427734375, 10.83831787109375, 8.452178955078125, -3.25311279296875, -2.493743896484375, 5.381439208984375, 3.596099853515625, 3.302398681640625, 2.7195663452148438, -0.5696258544921875, -1.3258552551269531, -12.3775634765625, 1.1568012237548828, 0.017889022827148438, 4.77911376953125, -5.49200439453125, 3.9625244140625, -0.6035614013671875, 1.07220458984375, -14.521453857421875, 2.844146728515625, 6.5528717041015625, -3.565093994140625, 4.605339050292969, 17.3011474609375, 1.47930908203125, 9.804595947265625, -4.460479736328125, 7.48492431640625, 0.3818359375, 15.442230224609375, 4.70159912109375, 8.137931823730469, 0.46466064453125, 20.44500732421875, 3.7994842529296875, -0.9130630493164062, -2.0973052978515625, -0.8236618041992188, 4.6363677978515625, -5.44146728515625, 2.3969955444335938, -2.1630096435546875, -2.83984375, 12.67120361328125, 0.2012176513671875, 9.6121826171875, -1.6096649169921875, -2.04571533203125, 7.2769622802734375, 2.749298095703125, 4.387664794921875, -2.322998046875, 0.6896553039550781, 4.770538330078125, 0.2145843505859375, 1.4112968444824219, 2.492523193359375, 0.217041015625, -2.7298583984375, -1.4902420043945312, 6.315826416015625, 10.440338134765625, 6.88812255859375, 10.314056396484375, 13.876312255859375, 4.9625701904296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000056.npy"}
{"epoch": 0.11727748691099477, "step": 57, "batch_size": 128, "mean": 4.168484687805176, "std": 5.92910099029541, "min": -8.85882568359375, "p10": -2.6502983093261716, "median": 3.282075881958008, "p90": 12.41477355957031, "max": 18.7393798828125, "pos_frac": 0.75, "sample": [-2.7219314575195312, 8.082427978515625, -6.524627685546875, 1.947845458984375, -2.619598388671875, -3.94073486328125, 13.3499755859375, 1.5411224365234375, 3.8428955078125, 2.503631591796875, 1.51348876953125, -1.758697509765625, 12.216644287109375, 1.7489242553710938, 9.088035583496094, 2.1284122467041016, 5.40869140625, 8.07720947265625, 4.43951416015625, 5.30938720703125, 5.895231246948242, 13.2388916015625, -0.5316390991210938, 2.0422897338867188, 0.4637451171875, 10.652099609375, -8.749267578125, 4.71197509765625, -0.8099746704101562, -2.7867355346679688, 5.0318603515625, 5.8082275390625, -0.1401519775390625, 1.493377685546875, 6.21575927734375, 18.7393798828125, 6.1970977783203125, 2.79547119140625, -7.314544677734375, 7.746246337890625, 10.302169799804688, -0.2996940612792969, 6.4136962890625, 18.510787963867188, -1.7366867065429688, 3.793548583984375, -0.16471099853515625, 1.584197998046875, 2.220703125, 8.497283935546875, -1.659881591796875, 2.6830406188964844, 3.930450439453125, 15.83282470703125, 4.218536376953125, 4.875732421875, -0.419281005859375, 2.0112762451171875, 4.6168212890625, 11.0628662109375, 2.19268798828125, 0.11279296875, 0.9476375579833984, 6.449127197265625, 5.4357757568359375, 10.96527099609375, 16.51287841796875, 13.36090087890625, -8.814178466796875, 4.3054962158203125, 16.233505249023438, 5.414154052734375, -1.6374969482421875, -1.4799747467041016, 15.339141845703125, 1.90869140625, 4.239166259765625, 8.907470703125, 3.61767578125, 2.0941162109375, 1.80218505859375, -0.7735214233398438, 17.545013427734375, 8.4442138671875, -8.85882568359375, -2.4050140380859375, -0.3640594482421875, 11.6971435546875, -3.728118896484375, -3.6733551025390625, 7.6475830078125, 10.012451171875, 0.29494667053222656, 3.0540199279785156, 2.095458984375, -4.573024749755859, 9.2960205078125, 2.4580078125, 4.29339599609375, -3.5528564453125, 2.509033203125, -0.316162109375, 11.36798095703125, 3.3550186157226562, 1.5384063720703125, 16.18463134765625, -0.01413726806640625, 12.8770751953125, 7.4344024658203125, 3.2091331481933594, -0.913970947265625, 1.56707763671875, 8.156280517578125, 0.8797683715820312, 6.837535858154297, 10.657073974609375, 8.437667846679688, -0.5958938598632812, 1.7007083892822266, 0.3135948181152344, 9.944793701171875, -4.0731201171875, 15.93133544921875, 3.4371337890625, 8.006942749023438, 0.1569061279296875, 9.9794921875, 7.5751953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000057.npy"}
{"epoch": 0.1193717277486911, "step": 58, "batch_size": 128, "mean": 2.439037799835205, "std": 6.015893459320068, "min": -20.708038330078125, "p10": -4.586059188842771, "median": 2.490753173828125, "p90": 9.650296020507813, "max": 20.9398193359375, "pos_frac": 0.7109375, "sample": [6.397369384765625, 11.3272705078125, 3.8345184326171875, 4.155517578125, 2.6715469360351562, -1.1600723266601562, 2.892772674560547, -15.15716552734375, 2.98248291015625, 7.40350341796875, 3.453632354736328, 9.780914306640625, 3.9745864868164062, 7.9581298828125, 1.43963623046875, -4.085914611816406, 1.4298477172851562, 11.08880615234375, 9.383384704589844, -1.30157470703125, 10.120849609375, 11.504913330078125, 7.0972900390625, -7.7431182861328125, -7.523681640625, 1.4385223388671875, 5.2968597412109375, 5.802490234375, 10.493118286132812, -0.681884765625, 0.9165725708007812, 6.0153961181640625, -2.694427490234375, -0.7774143218994141, 6.69500732421875, -6.119354248046875, -2.27679443359375, 7.255218505859375, 0.20726394653320312, 4.84039306640625, -12.09539794921875, 0.36163330078125, 0.22037506103515625, 1.6929168701171875, 5.14642333984375, -6.989936828613281, 1.2090778350830078, 14.2547607421875, 0.2013225555419922, -20.708038330078125, -0.07755279541015625, 14.738433837890625, -1.7928924560546875, -2.1577224731445312, 0.29974365234375, 0.2964324951171875, -3.0470428466796875, 5.216743469238281, 1.390869140625, 6.7813720703125, 6.214012145996094, -1.10699462890625, 7.153106689453125, -0.7156219482421875, 5.650543212890625, 4.539398193359375, -6.80517578125, 9.659332275390625, 2.31439208984375, 1.56817626953125, 2.06024169921875, -5.753063201904297, -1.3684158325195312, 3.174591064453125, 5.830402374267578, 5.107635498046875, 17.08099365234375, 2.8096351623535156, -2.590423583984375, 3.3720779418945312, 5.3690032958984375, -1.8331298828125, 20.9398193359375, -7.82720947265625, 4.308319091796875, 4.8622589111328125, -0.043979644775390625, 0.9426803588867188, 3.8659820556640625, 10.307861328125, 0.5041694641113281, -0.139801025390625, 2.885009765625, 2.7301177978515625, 8.204345703125, 4.01904296875, 2.307708740234375, -7.646087646484375, 0.73297119140625, 3.92608642578125, 0.8879852294921875, 0.5550460815429688, -2.6087188720703125, -2.021240234375, 8.795196533203125, 3.2838783264160156, 0.3296852111816406, 9.120468139648438, 4.350189208984375, 3.21368408203125, -8.002548217773438, 2.6671142578125, -2.8443450927734375, 9.64642333984375, 1.892608642578125, 2.215484619140625, -1.468719482421875, 10.87615966796875, 8.190879821777344, 9.624359130859375, 6.3949432373046875, 1.991729736328125, -7.5794677734375, 9.092742919921875, -1.2368087768554688, -3.33935546875, 6.38311767578125, 1.898406982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000058.npy"}
{"epoch": 0.12146596858638743, "step": 59, "batch_size": 128, "mean": 1.5757670402526855, "std": 7.5829758644104, "min": -24.124359130859375, "p10": -5.497962951660156, "median": 0.8295135498046875, "p90": 10.831759643554685, "max": 23.411651611328125, "pos_frac": 0.6015625, "sample": [1.2286911010742188, 9.70538330078125, 3.65545654296875, -20.423583984375, 0.872406005859375, 7.2740936279296875, 2.375244140625, -4.071014404296875, 3.83721923828125, 3.753997802734375, 1.7277374267578125, 4.774383544921875, -5.6844024658203125, 8.310714721679688, -9.787200927734375, 1.1342315673828125, 3.7332382202148438, 1.2490005493164062, -6.0785064697265625, 4.04144287109375, 0.102325439453125, 0.24921035766601562, 0.21619033813476562, -0.2365570068359375, -5.704925537109375, 0.7127227783203125, 1.24139404296875, 12.6099853515625, -13.327545166015625, -1.24737548828125, -4.3248291015625, 2.6557388305664062, -3.7384033203125, 11.394805908203125, 0.0, 10.5904541015625, -0.6265792846679688, 5.654693603515625, 0.0777130126953125, -8.33868408203125, 0.788177490234375, 17.3975830078125, 7.227142333984375, 16.4920654296875, -0.059844970703125, -14.490142822265625, 2.8819198608398438, 22.53253173828125, 5.9818115234375, 8.4453125, 7.502479553222656, 0.563507080078125, 14.240966796875, 20.17547607421875, -1.7137680053710938, 3.1913299560546875, -1.0755386352539062, -4.6563720703125, -3.441009521484375, 0.9104080200195312, -4.97760009765625, 4.4496917724609375, 3.45458984375, -2.370697021484375, -5.213409423828125, 16.443084716796875, 19.457427978515625, -2.4830322265625, 7.721893310546875, 4.5282440185546875, -3.36480712890625, 11.559219360351562, -4.75103759765625, 23.411651611328125, -1.6814079284667969, -15.24139404296875, 1.6818389892578125, -24.124359130859375, -3.8692779541015625, 0.31906700134277344, 6.0693359375, -0.6954345703125, 2.4956817626953125, 1.609283447265625, 3.0055770874023438, 6.7761688232421875, 9.6446533203125, -0.8433380126953125, -5.418060302734375, -9.390548706054688, 7.099609375, 4.8779754638671875, 0.4448108673095703, -0.9962863922119141, -4.9695892333984375, 9.706283569335938, -1.29180908203125, -0.84100341796875, -4.8980255126953125, 4.917236328125, -0.899505615234375, 2.408203125, 3.0583038330078125, 4.098052978515625, -0.34915924072265625, -9.679168701171875, -5.17572021484375, -4.879486083984375, -0.9160003662109375, 3.72760009765625, 0.870849609375, 6.72467041015625, 18.54156494140625, 0.49652099609375, -0.54156494140625, -3.4456405639648438, 11.52593994140625, 5.7606658935546875, 3.389892578125, -1.700286865234375, -8.4853515625, 3.52801513671875, 3.373626708984375, 0.10662841796875, 0.7022914886474609, 0.42978668212890625, -3.880889892578125, -1.8267593383789062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000059.npy"}
{"epoch": 0.12356020942408377, "step": 60, "batch_size": 128, "mean": 3.46097993850708, "std": 7.463657855987549, "min": -21.21142578125, "p10": -4.583740997314453, "median": 2.6651554107666016, "p90": 12.453407287597656, "max": 23.376480102539062, "pos_frac": 0.7265625, "sample": [-17.028045654296875, 11.198455810546875, 3.8620243072509766, 0.1964111328125, 5.32135009765625, 9.03448486328125, 0.66534423828125, 5.2628021240234375, 5.3634033203125, -13.402099609375, 1.2810821533203125, 0.04353141784667969, 0.3075523376464844, 5.824195861816406, -0.3378448486328125, 6.138916015625, 1.9882659912109375, 0.57061767578125, 6.13824462890625, 18.08416748046875, 6.516227722167969, -6.352020263671875, 16.407012939453125, -4.930572509765625, -3.8386402130126953, 2.39117431640625, 2.68707275390625, 0.0, -3.435731887817383, 16.98486328125, -3.2533721923828125, 10.7359619140625, 2.307952880859375, 8.292236328125, 2.6199951171875, -5.4795379638671875, 7.796051025390625, -10.228363037109375, 3.1584396362304688, -0.4465808868408203, 0.0527801513671875, 5.813262939453125, 3.479736328125, 4.39495849609375, -0.34609222412109375, -5.9758148193359375, -4.362493515014648, -1.53570556640625, 10.779678344726562, 1.1927299499511719, -2.9317359924316406, 2.19256591796875, 2.816864013671875, 11.84588623046875, 0.3313751220703125, -4.5377197265625, 7.964378356933594, -8.1768798828125, 0.6207046508789062, 7.0883941650390625, 2.3795089721679688, 10.32952880859375, 13.092300415039062, 0.000408172607421875, -21.21142578125, 12.170867919921875, -3.0443115234375, -4.691123962402344, 1.7325668334960938, -2.446868896484375, 2.347747802734375, 10.874404907226562, 8.444099426269531, 12.85260009765625, 11.6572265625, 4.957672119140625, 1.6281814575195312, 4.4876251220703125, 2.0196990966796875, 5.9207763671875, 16.1016845703125, 3.9665908813476562, -1.9253082275390625, 4.97467041015625, 23.376480102539062, 0.019500732421875, 9.988677978515625, -17.409881591796875, 4.854101181030273, 9.107666015625, 4.389892578125, 2.384014129638672, -2.0477294921875, -0.11670112609863281, 8.7769775390625, 12.38714599609375, 1.043853759765625, 14.640625, -0.10420989990234375, 5.0335693359375, 4.653472900390625, 17.28155517578125, -2.4428863525390625, -6.1829833984375, 17.2918701171875, 9.8984375, 1.492431640625, 7.1731109619140625, 12.06341552734375, 0.0, 22.60260009765625, -9.983154296875, 6.519256591796875, -1.9246063232421875, 0.81036376953125, 0.82635498046875, 4.174591064453125, 2.643238067626953, 8.1490478515625, -4.4112548828125, 6.390892028808594, 12.608016967773438, -0.7766075134277344, 16.06304931640625, 6.930572509765625, 1.6573944091796875, 2.7117919921875, 10.690475463867188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000060.npy"}
{"epoch": 0.1256544502617801, "step": 61, "batch_size": 128, "mean": 3.210594415664673, "std": 8.144835472106934, "min": -21.88897705078125, "p10": -6.2286437988281245, "median": 2.1660690307617188, "p90": 14.453407287597656, "max": 23.056793212890625, "pos_frac": 0.6640625, "sample": [-6.93780517578125, -10.6446533203125, -0.37597084045410156, -3.4443817138671875, 1.74920654296875, 6.1015625, -1.3052520751953125, 3.765106201171875, 1.8794307708740234, 5.247886657714844, 9.6241455078125, -16.0340576171875, -3.523681640625, 0.7045745849609375, -2.952178955078125, 3.4659481048583984, -21.88897705078125, -1.607666015625, -6.015838623046875, 1.54705810546875, 2.0409622192382812, 3.0252685546875, 13.74945068359375, 11.066558837890625, 11.8701171875, -2.808563232421875, 9.400360107421875, -7.4393768310546875, 18.657135009765625, 15.776123046875, 1.393167495727539, 0.2041778564453125, 0.38056373596191406, -0.09370803833007812, 18.237091064453125, 2.2911758422851562, -0.8167877197265625, 14.530685424804688, 8.27154541015625, 5.739080429077148, -2.6212844848632812, -0.2845458984375, 11.781219482421875, -4.479148864746094, -0.1927490234375, 1.445556640625, 11.407989501953125, 4.6427001953125, 1.217132568359375, 3.08392333984375, -1.9814453125, 14.798599243164062, 16.951416015625, 18.807373046875, 20.92669677734375, 13.406585693359375, 6.281158447265625, 5.36561393737793, 15.832382202148438, 1.5736236572265625, 8.282028198242188, 4.994544982910156, 6.19354248046875, 0.6444854736328125, 6.679878234863281, 13.22088623046875, 14.4202880859375, 21.42327880859375, 12.012786865234375, 7.63470458984375, 1.079498291015625, 3.6544647216796875, 7.159915924072266, -6.767578125, -4.660400390625, 4.93695068359375, 0.026056289672851562, -5.0233306884765625, 4.875831604003906, -7.1270751953125, 12.488006591796875, 1.4563751220703125, 4.725677490234375, -0.74249267578125, 1.2326202392578125, 5.698921203613281, 0.4175872802734375, -3.0869827270507812, -2.6835403442382812, 5.425140380859375, -1.4303092956542969, -6.725189208984375, -8.56591796875, 7.1485443115234375, 0.3320465087890625, 3.0270614624023438, 1.8190650939941406, 19.228790283203125, -2.6807708740234375, 13.1943359375, 14.57489013671875, 23.056793212890625, 6.9259033203125, 0.0, 1.3769378662109375, 4.857391357421875, 4.398002624511719, 3.2742462158203125, 2.8097000122070312, -12.8892822265625, 9.15606689453125, -2.31634521484375, -1.345184326171875, 1.990203857421875, -15.09649658203125, 8.30865478515625, -17.98870849609375, -2.0541610717773438, -4.221954345703125, 7.964447021484375, -2.53692626953125, -3.8174057006835938, -2.0221328735351562, 5.421150207519531, 4.848258972167969, 6.4869232177734375, 12.081069946289062, -9.017959594726562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000061.npy"}
{"epoch": 0.12774869109947645, "step": 62, "batch_size": 128, "mean": 3.3294644355773926, "std": 7.579824924468994, "min": -15.125762939453125, "p10": -4.907225036621093, "median": 2.2018814086914062, "p90": 15.326805114746092, "max": 30.98779296875, "pos_frac": 0.671875, "sample": [16.748291015625, 3.4324493408203125, 16.283447265625, 9.451133728027344, 7.734100341796875, 3.16107177734375, 8.61822509765625, 3.392578125, -11.92022705078125, -3.232208251953125, -1.470977783203125, -3.2048377990722656, -4.8795623779296875, 10.13720703125, 0.373443603515625, 0.7691650390625, 1.204559326171875, 6.8722686767578125, -0.5541458129882812, -3.9271240234375, 4.483329772949219, 3.893413543701172, 6.7653961181640625, 1.4144096374511719, -4.339508056640625, 18.9464111328125, -7.1477813720703125, 1.2047119140625, 3.1835403442382812, 2.15472412109375, 0.599456787109375, -2.043212890625, 7.8247222900390625, 6.5429840087890625, 1.8119621276855469, 0.6005592346191406, 2.72528076171875, -4.8316802978515625, 6.002155303955078, -2.272796630859375, 2.2371978759765625, 10.45355224609375, 7.851371765136719, -1.5012969970703125, -5.8702545166015625, -1.509521484375, -0.37834930419921875, -0.694091796875, 6.690338134765625, 18.47998046875, -14.4912109375, 15.740447998046875, -4.971771240234375, 14.60052490234375, 4.73150634765625, -7.6914520263671875, 16.40570068359375, 0.0, 7.3867645263671875, -3.1743698120117188, -15.125762939453125, -0.5865097045898438, 8.81463623046875, -0.4459075927734375, 3.3235931396484375, 0.5915069580078125, -1.6698760986328125, 2.56488037109375, 1.782867431640625, 7.509368896484375, 19.692291259765625, 3.2557220458984375, 1.7296791076660156, -5.1229705810546875, 7.963958740234375, 16.439300537109375, 5.912628173828125, -4.2272186279296875, 1.7824325561523438, 4.232866287231445, 3.2194747924804688, -4.3014984130859375, 4.639892578125, 4.04595947265625, -0.90850830078125, 12.99346923828125, 18.6287841796875, 0.76239013671875, -0.594512939453125, 9.0029296875, 0.9524803161621094, -2.923126220703125, 25.625, 17.010467529296875, 9.649871826171875, 11.33843994140625, -7.1213836669921875, 30.98779296875, -7.002044677734375, -6.81951904296875, 9.511993408203125, 1.2469635009765625, -4.325584411621094, 2.16656494140625, 15.219528198242188, 7.0566558837890625, 2.9469146728515625, 8.185440063476562, 1.6411781311035156, 15.577117919921875, 2.990447998046875, -4.130668640136719, 4.435455322265625, -0.58258056640625, 0.272369384765625, 0.32729339599609375, -6.18792724609375, 0.4254150390625, 5.024147033691406, 1.3303813934326172, 3.8961334228515625, 2.655609130859375, 7.5203857421875, 5.9532928466796875, 4.051239013671875, -1.628997802734375, -7.356109619140625, -0.45703887939453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000062.npy"}
{"epoch": 0.12984293193717278, "step": 63, "batch_size": 128, "mean": 3.5506744384765625, "std": 7.564035415649414, "min": -15.8685302734375, "p10": -4.129953002929687, "median": 2.581073760986328, "p90": 13.538134002685544, "max": 23.024658203125, "pos_frac": 0.6796875, "sample": [-0.395965576171875, 2.271484375, 0.3336906433105469, -13.6522216796875, -9.4967041015625, -0.5368785858154297, 5.387847900390625, 0.18108367919921875, 0.63934326171875, 8.912353515625, -10.325592041015625, -2.996063232421875, -5.820098876953125, -0.8449859619140625, 9.058639526367188, 1.2692337036132812, 9.875946044921875, 12.19580078125, 16.835693359375, 16.831146240234375, -3.933837890625, -1.436614990234375, 16.659912109375, -3.42779541015625, 0.099853515625, 12.509735107421875, 9.545501708984375, -4.630271911621094, 2.7278289794921875, 5.89013671875, 0.2754974365234375, 4.437225341796875, -4.050811767578125, 1.4294815063476562, 2.3621997833251953, 21.44268798828125, -3.66534423828125, 6.1349639892578125, -7.6803741455078125, 3.549224853515625, -4.191070556640625, 1.5149307250976562, 3.538787841796875, -4.103759765625, 3.156402587890625, 8.805923461914062, 17.25189208984375, -15.73785400390625, -12.3353271484375, 6.7407073974609375, -0.6974430084228516, 13.11083984375, 7.7447357177734375, 0.0, -1.5735244750976562, 11.046661376953125, 4.10028076171875, -3.3502655029296875, 3.899211883544922, 4.515483856201172, 2.8513565063476562, 1.8060302734375, 7.403106689453125, 23.024658203125, -1.41412353515625, -0.7685546875, 4.5755462646484375, 14.34130859375, -1.2743377685546875, 7.92315673828125, 20.469482421875, 2.593597412109375, 9.244354248046875, 7.04638671875, 15.44677734375, -3.177581787109375, 13.193916320800781, 4.181114196777344, 3.8066558837890625, -3.37158203125, -15.8685302734375, 16.075439453125, 5.9777069091796875, 2.10247802734375, 8.761283874511719, 4.799446105957031, 8.26283073425293, 0.9243640899658203, 2.0042572021484375, 2.5945205688476562, 10.245407104492188, -0.4232177734375, -10.955368041992188, 10.0281982421875, 1.6233940124511719, 11.912506103515625, -1.87030029296875, 15.589569091796875, 10.335685729980469, 12.26922607421875, 0.4949951171875, -1.236297607421875, 0.0489501953125, -1.454498291015625, 2.3596649169921875, -0.1242523193359375, 2.315338134765625, -1.1366462707519531, 15.552398681640625, 4.5367431640625, 6.57720947265625, -0.951904296875, -0.773651123046875, 5.0215301513671875, 2.1415576934814453, 20.4798583984375, 2.6470184326171875, 3.3919677734375, -2.1103515625, 2.5685501098632812, -6.757904052734375, 0.234222412109375, 6.183624267578125, 11.095428466796875, -5.746246337890625, 1.4644088745117188, 12.95465087890625, 11.020233154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000063.npy"}
{"epoch": 0.1319371727748691, "step": 64, "batch_size": 128, "mean": 4.244952201843262, "std": 8.640124320983887, "min": -23.924560546875, "p10": -4.584197998046874, "median": 4.161899566650391, "p90": 14.590255737304688, "max": 29.3658447265625, "pos_frac": 0.6953125, "sample": [8.7283935546875, 1.6044921875, -19.61956787109375, 19.62042236328125, -2.79949951171875, 11.11041259765625, 4.4013824462890625, 5.829681396484375, 23.82806396484375, 7.764656066894531, 7.858642578125, 6.55987548828125, 16.18231201171875, 7.31427001953125, 5.0861358642578125, -11.012008666992188, 27.72552490234375, 6.0480499267578125, 1.0344696044921875, -7.20928955078125, 9.56640625, 5.69964599609375, -0.6207733154296875, -2.542449951171875, 2.6175193786621094, -15.512832641601562, 2.5073013305664062, 2.8986892700195312, 9.91998291015625, 14.627960205078125, 5.31903076171875, -3.017547607421875, 10.428436279296875, 3.5291824340820312, 6.64483642578125, 0.7437667846679688, -2.224517822265625, 6.661842346191406, 5.0733489990234375, 1.1004791259765625, 10.45233154296875, -2.026569366455078, -3.48480224609375, 8.122535705566406, 7.40765380859375, 23.2763671875, 18.0281982421875, -2.7441177368164062, 4.55828857421875, 5.058868408203125, 6.70745849609375, 0.534820556640625, -2.4313106536865234, 4.296905517578125, 2.142129898071289, -2.3067626953125, -9.32537841796875, 6.203620910644531, 11.512657165527344, 6.2110443115234375, 14.91644287109375, 9.156978607177734, -5.4439697265625, 3.227508544921875, 0.103790283203125, -2.64154052734375, 2.3373184204101562, -5.1624755859375, -4.33636474609375, 9.1370849609375, -2.56817626953125, 6.677001953125, 14.50091552734375, 10.313674926757812, -0.3348197937011719, 13.37701416015625, 6.475250244140625, 0.420562744140625, -1.4272994995117188, -1.7800445556640625, -2.8774490356445312, 14.5740966796875, -0.1922454833984375, -0.9377708435058594, 4.108856201171875, 17.974090576171875, 3.155548095703125, 1.7404708862304688, 3.7140350341796875, -3.286376953125, 11.276885986328125, 29.3658447265625, 8.4339599609375, 17.8824462890625, 3.2652511596679688, -5.36334228515625, -18.8875732421875, 5.782501220703125, -3.75225830078125, -5.90283203125, 3.33819580078125, 13.813232421875, 4.214942932128906, 0.0, 0.1342334747314453, 0.4232349395751953, 5.927642822265625, 11.853591918945312, 10.23101806640625, -5.35009765625, 10.60101318359375, 0.9211349487304688, 0.043975830078125, 17.275558471679688, 4.5683135986328125, -2.919597625732422, -2.934906005859375, 21.97467041015625, -5.1746826171875, 2.4708480834960938, 0.0, 13.156707763671875, 13.7896728515625, 8.482070922851562, 11.931167602539062, 7.6785430908203125, -23.924560546875, -1.86572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000064.npy"}
{"epoch": 0.13403141361256546, "step": 65, "batch_size": 128, "mean": 3.3691043853759766, "std": 7.65395450592041, "min": -16.85382080078125, "p10": -4.896014404296874, "median": 2.9160871505737305, "p90": 14.273463439941406, "max": 21.480072021484375, "pos_frac": 0.6171875, "sample": [-1.8004875183105469, -2.77667236328125, 5.66644287109375, 1.1274070739746094, 14.245223999023438, 4.955217361450195, 1.3819313049316406, 10.364105224609375, 11.30975341796875, -11.316009521484375, -0.8094482421875, -12.938262939453125, 4.41900634765625, 9.30145263671875, -4.5094451904296875, 15.27490234375, 17.2020263671875, -2.821533203125, 3.0074119567871094, 0.87152099609375, -1.831390380859375, -0.9436378479003906, 5.88153076171875, 13.485382080078125, 2.9838314056396484, 2.8483428955078125, 10.31817626953125, 2.2692184448242188, -2.714874267578125, -0.7632675170898438, 7.364959716796875, -1.5024871826171875, -2.07391357421875, 9.344329833984375, 20.19805908203125, -2.150054931640625, 0.33962059020996094, 5.1590576171875, -0.5643692016601562, 1.2677001953125, 3.644134521484375, -4.5690155029296875, -5.22247314453125, 7.2870941162109375, 19.31549072265625, -0.2961578369140625, 12.75830078125, 5.073486328125, 12.690338134765625, 4.032661437988281, 2.5185470581054688, 11.949554443359375, -13.5399169921875, 6.3236083984375, 3.4417953491210938, -5.9666290283203125, 11.08074951171875, 3.6338958740234375, 1.1350841522216797, 6.728080749511719, -1.95965576171875, -1.5359268188476562, 0.20555877685546875, -1.5503616333007812, 5.092926025390625, -0.5169677734375, 0.7621994018554688, -8.436614990234375, -6.891448974609375, 0.989898681640625, 9.165496826171875, 10.8330078125, -4.756103515625, 3.197021484375, 9.122116088867188, 7.427921295166016, -0.2028656005859375, 1.1136302947998047, -1.23138427734375, -1.069732666015625, -12.13677978515625, -1.8978271484375, 0.0, -3.7006683349609375, 19.012969970703125, -0.21894073486328125, -6.1508636474609375, 3.524078369140625, 15.85443115234375, 5.9411468505859375, 9.04681396484375, 21.480072021484375, 15.2635498046875, 10.630142211914062, 19.544235229492188, -3.6474609375, 5.251708984375, 6.827667236328125, -0.8175392150878906, 12.103851318359375, 9.4873046875, 10.81060791015625, -15.5816650390625, 3.458648681640625, 3.7955474853515625, 7.351287841796875, -0.5557098388671875, 18.69769287109375, 15.4813232421875, 14.33935546875, 3.073822021484375, -0.0372161865234375, 6.890472412109375, 2.1848907470703125, -6.85174560546875, -3.2334747314453125, 7.8552398681640625, -16.85382080078125, -0.0589141845703125, 5.075958251953125, -5.448089599609375, -4.360908508300781, 14.654449462890625, 4.467803955078125, 7.972320556640625, -0.3191337585449219, 1.9496536254882812, -1.759033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000065.npy"}
{"epoch": 0.13612565445026178, "step": 66, "batch_size": 128, "mean": 4.190981388092041, "std": 10.356176376342773, "min": -19.836639404296875, "p10": -7.380337524414061, "median": 3.3365402221679688, "p90": 19.158451843261716, "max": 30.777740478515625, "pos_frac": 0.65625, "sample": [6.204353332519531, 1.301919937133789, 24.73382568359375, -7.0120849609375, -2.4794387817382812, 15.56488037109375, -5.6089324951171875, -0.7813301086425781, 7.147216796875, 0.8871688842773438, -1.3279037475585938, -2.803943634033203, 3.9741134643554688, 9.503341674804688, 5.295928955078125, -1.03656005859375, -1.3231773376464844, 2.0944976806640625, 26.4691162109375, 7.918087005615234, -1.13299560546875, -14.483001708984375, 8.212570190429688, 11.23516845703125, -4.938140869140625, 9.025985717773438, 3.0877838134765625, 1.662841796875, -3.015472412109375, 8.511276245117188, 15.546112060546875, -0.8706588745117188, 4.2960205078125, -15.45477294921875, 5.1280364990234375, -0.6611251831054688, 6.48046875, 27.45257568359375, 3.585296630859375, -2.0502471923828125, -13.576751708984375, 11.003072738647461, 20.93353271484375, 4.0087890625, 3.6963119506835938, 6.315803527832031, -14.19891357421875, 29.424072265625, 25.2386474609375, 1.5907745361328125, -11.24853515625, 0.5794296264648438, 2.80242919921875, 11.130615234375, 20.88958740234375, 1.3545379638671875, 9.585418701171875, 10.557159423828125, -19.836639404296875, 3.0425567626953125, 4.60406494140625, 9.1717529296875, 20.210601806640625, 8.166351318359375, 18.738922119140625, -17.995986938476562, 5.9338531494140625, -6.696746826171875, 9.326263427734375, -0.6977462768554688, -5.574798583984375, -0.41217041015625, 30.777740478515625, -6.854400634765625, 13.91802978515625, 4.2303009033203125, 8.943191528320312, -2.3097286224365234, 18.91766357421875, -16.211700439453125, 5.9844970703125, 6.410606384277344, -2.097442626953125, 13.033416748046875, -3.07830810546875, 9.112030029296875, 4.684883117675781, 2.5908889770507812, 9.8677978515625, 13.561920166015625, 3.7196044921875, 4.35882568359375, 1.373291015625, -18.04119873046875, 1.0887603759765625, -4.9284210205078125, -6.5883636474609375, -8.704086303710938, 7.755401611328125, 1.944793701171875, 21.03143310546875, 2.7927398681640625, -0.5, 0.0, 5.9599151611328125, -1.5031890869140625, -0.98028564453125, 19.720291137695312, 6.921638488769531, -6.76904296875, 8.9183349609375, 1.481353759765625, 1.8037567138671875, -15.07647705078125, 0.06048583984375, 14.41973876953125, 1.1953887939453125, 24.04986572265625, 10.220703125, 5.65576171875, -8.239593505859375, -0.8363037109375, 15.38140869140625, -8.688911437988281, 3.027557373046875, -0.5114059448242188, 28.0216064453125, 17.023773193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000066.npy"}
{"epoch": 0.1382198952879581, "step": 67, "batch_size": 128, "mean": 3.692615509033203, "std": 8.990950584411621, "min": -18.858306884765625, "p10": -7.2746330261230465, "median": 3.1976184844970703, "p90": 14.93597412109375, "max": 27.350494384765625, "pos_frac": 0.6640625, "sample": [1.8862838745117188, 16.619842529296875, 6.063323974609375, 27.350494384765625, -1.2114677429199219, -11.23004150390625, 2.0917129516601562, 10.914093017578125, 10.137786865234375, -7.072395324707031, 4.106624603271484, 23.6220703125, 3.83673095703125, 10.67840576171875, 18.54779052734375, 9.71368408203125, 5.7201080322265625, 16.9898681640625, 3.908905029296875, -5.471923828125, 4.3597259521484375, 0.098876953125, 14.87518310546875, -1.7521209716796875, 4.617805480957031, 10.95556640625, -13.40325927734375, 4.577056884765625, 7.215385437011719, 1.333831787109375, -14.39276123046875, 1.339029312133789, -3.3692626953125, 7.296272277832031, 15.07781982421875, -4.96246337890625, 0.0, 2.003337860107422, 12.300384521484375, 3.480966567993164, -2.8270263671875, 9.285308837890625, 4.1590576171875, 13.574310302734375, 3.1967201232910156, -1.0530643463134766, 8.183837890625, -5.865814208984375, -6.7115478515625, -0.32032012939453125, -14.566879272460938, 21.057830810546875, 2.6092987060546875, 11.3292236328125, 3.2471923828125, 3.0037994384765625, 18.795196533203125, -15.07708740234375, 0.2112274169921875, 5.867572784423828, 7.48150634765625, 9.490947723388672, 14.802001953125, 4.5731201171875, 9.354454040527344, -0.2059154510498047, 3.0305023193359375, -9.69757080078125, -0.185546875, 3.198516845703125, 9.808319091796875, -5.308258056640625, -3.957632064819336, 16.1654052734375, -18.858306884765625, 26.742156982421875, 17.19482421875, 10.823699951171875, 2.4790802001953125, -1.646087646484375, 9.72357177734375, -5.899196624755859, 13.2413330078125, -2.945281982421875, -10.356231689453125, -0.14151573181152344, 14.024589538574219, -6.389739990234375, 1.9967117309570312, 14.593734741210938, 14.762939453125, -14.74346923828125, 9.98114013671875, 15.639892578125, -10.72784423828125, 10.915603637695312, -0.94732666015625, 3.7987518310546875, 7.0838470458984375, 10.878448486328125, 13.422882080078125, -0.2927703857421875, -0.25911712646484375, -2.39141845703125, 1.8121337890625, -1.175750732421875, -3.8942604064941406, -18.5771484375, -8.521110534667969, 2.7834625244140625, 8.279632568359375, 9.595458984375, 7.56536865234375, -1.8103179931640625, 0.051265716552734375, 16.208251953125, 2.25164794921875, 3.1010055541992188, 4.69061279296875, -2.6885528564453125, 4.920928955078125, -0.19628524780273438, 1.37646484375, 4.115943908691406, 2.1695098876953125, -7.74652099609375, 8.8670654296875, 2.2691383361816406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000067.npy"}
{"epoch": 0.14031413612565444, "step": 68, "batch_size": 128, "mean": 4.679655075073242, "std": 11.033845901489258, "min": -23.73919677734375, "p10": -7.461688995361328, "median": 3.391063690185547, "p90": 17.004547119140618, "max": 39.5206298828125, "pos_frac": 0.6796875, "sample": [10.124740600585938, 1.6872062683105469, 1.501617431640625, -4.034332275390625, -4.397411346435547, 0.24486541748046875, -0.94378662109375, 15.311660766601562, -4.640266418457031, 4.324531555175781, 5.939216613769531, -2.2030181884765625, -2.320281982421875, 18.609375, 16.24468994140625, -7.40008544921875, 0.77020263671875, 3.3632736206054688, 4.66363525390625, 9.718292236328125, 14.428314208984375, 2.6777572631835938, -6.484291076660156, -9.946380615234375, 38.206298828125, 5.117591857910156, 4.138820648193359, 12.90911865234375, -0.233978271484375, -17.72357177734375, 8.267288208007812, 19.701416015625, 36.266265869140625, 14.376953125, 0.33703041076660156, 20.690765380859375, -1.8108978271484375, -3.3532257080078125, -8.618545532226562, -14.34130859375, 3.418853759765625, -1.2050399780273438, 10.177631378173828, 1.1644287109375, 3.6992454528808594, -23.73919677734375, -0.7169036865234375, 5.3210296630859375, 16.3729248046875, 11.90069580078125, 3.19512939453125, -6.359619140625, -1.08056640625, 11.69940185546875, 31.729766845703125, -6.667999267578125, 11.080047607421875, 20.561767578125, -1.5134086608886719, 4.583396911621094, 13.02602767944336, 6.6385650634765625, 1.4911079406738281, 11.08953857421875, 5.129825592041016, -0.9109535217285156, -18.2623291015625, 18.750640869140625, -2.948883056640625, 39.5206298828125, 0.7189826965332031, -10.804290771484375, 9.39361572265625, 1.910430908203125, -10.64984130859375, 4.476165771484375, -3.00909423828125, 3.8013916015625, 37.80657958984375, -2.4767379760742188, -6.0663604736328125, 7.2012176513671875, 1.5418968200683594, 3.1244659423828125, 3.515869140625, 5.09625244140625, -7.2109375, 15.410110473632812, -8.95404052734375, 3.8907318115234375, 7.071533203125, 2.19635009765625, 6.247398376464844, -0.46820068359375, 2.15277099609375, 18.47833251953125, -5.1220703125, -9.937103271484375, 8.2801513671875, 11.582778930664062, 14.286544799804688, -0.7433185577392578, 2.2249526977539062, 13.518096923828125, 13.195037841796875, 6.160980224609375, 14.62249755859375, 18.5491943359375, 1.95855712890625, 8.823333740234375, 10.367156982421875, -8.893943786621094, 12.746841430664062, 1.7265148162841797, -7.605430603027344, 3.509857177734375, 11.80291748046875, 1.3356094360351562, 0.06939697265625, 8.022758483886719, 3.31707763671875, -4.4878387451171875, 3.587066650390625, 36.75151062011719, 8.0582275390625, -9.604522705078125, -2.4702491760253906, 0.6533355712890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000068.npy"}
{"epoch": 0.1424083769633508, "step": 69, "batch_size": 128, "mean": 6.056118011474609, "std": 10.002968788146973, "min": -22.64996337890625, "p10": -5.432426452636719, "median": 4.365547180175781, "p90": 18.160862731933587, "max": 33.115478515625, "pos_frac": 0.75, "sample": [9.699966430664062, 3.90576171875, 7.102954864501953, 12.188201904296875, 5.28656005859375, -12.64495849609375, 2.1866607666015625, 7.943950653076172, 8.59368896484375, 2.2787094116210938, 12.532272338867188, 4.0570220947265625, 27.654205322265625, 1.321685791015625, 19.759201049804688, -0.5313491821289062, 23.729278564453125, 1.59796142578125, 12.32635498046875, -1.6184844970703125, -3.5625457763671875, 1.0718097686767578, -5.747295379638672, -5.027862548828125, 9.0841064453125, 29.886260986328125, 10.952301025390625, 16.32061767578125, -0.06700515747070312, 16.275177001953125, -22.64996337890625, 12.666046142578125, 3.0277023315429688, 1.7099761962890625, 4.674072265625, 0.7813568115234375, 0.9198493957519531, -3.8268585205078125, 21.498260498046875, -5.165313720703125, 33.115478515625, -15.42706298828125, 2.165346145629883, 3.9634246826171875, 1.1345367431640625, 6.469842910766602, -5.651824951171875, 11.480133056640625, 7.9447021484375, -2.2118072509765625, 17.093994140625, -1.77215576171875, 10.312652587890625, 22.1981201171875, 9.81719970703125, -9.490020751953125, -14.6707763671875, 2.9143524169921875, 12.52972412109375, 1.7075557708740234, -5.481292724609375, 12.1114501953125, 2.9018630981445312, 14.88690185546875, -2.7427825927734375, 9.201080322265625, 3.9753875732421875, 12.748489379882812, 2.037078857421875, 11.0582275390625, 20.572509765625, 8.683853149414062, 0.0006103515625, -0.727996826171875, 0.19464111328125, -4.426666259765625, 17.43121337890625, 3.555835723876953, 13.69366455078125, 9.439865112304688, -4.225553512573242, -8.554901123046875, 8.214258193969727, 2.0403900146484375, -2.1091766357421875, 1.0223236083984375, 10.78082275390625, 8.176963806152344, 12.449813842773438, 22.768798828125, 14.36529541015625, 28.942611694335938, 7.617618560791016, 11.2962646484375, -4.1231842041015625, -8.088348388671875, -0.06842803955078125, 2.0771331787109375, 26.294189453125, 1.6417579650878906, 11.714248657226562, 9.917144775390625, 30.6156005859375, 9.42291259765625, 9.62841796875, 16.849380493164062, -6.8846435546875, 11.560638427734375, 0.91302490234375, 4.980888366699219, -5.4114837646484375, -4.757904052734375, 3.4918060302734375, 27.9244384765625, 17.475860595703125, -5.766693115234375, -4.937530517578125, -7.532196044921875, 13.519554138183594, 7.797454833984375, 12.878814697265625, 2.7928466796875, 2.283782958984375, 0.6558170318603516, 2.7334671020507812, 9.230560302734375, 12.9927978515625, 7.647773742675781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000069.npy"}
{"epoch": 0.14450261780104712, "step": 70, "batch_size": 128, "mean": 5.823676109313965, "std": 11.366315841674805, "min": -31.62249755859375, "p10": -5.132281494140623, "median": 4.014892578125, "p90": 19.54637145996093, "max": 41.13934326171875, "pos_frac": 0.7109375, "sample": [-17.213531494140625, 12.734893798828125, 4.50439453125, -4.183685302734375, 1.564910888671875, 2.29998779296875, 7.373291015625, 7.04302978515625, -14.02081298828125, 30.57098388671875, -6.4216461181640625, 0.6684761047363281, -3.4146881103515625, 14.310302734375, 24.113525390625, 10.2860107421875, 12.65692138671875, 5.088592529296875, 0.2659339904785156, 11.63311767578125, 1.6243057250976562, 18.182891845703125, 3.050994873046875, 2.920450210571289, -13.6788330078125, 15.224769592285156, 3.0391387939453125, 12.738433837890625, 6.1782989501953125, 1.2579345703125, 22.15655517578125, 9.779403686523438, 3.79803466796875, -11.08905029296875, -0.65972900390625, 3.5766448974609375, 16.602142333984375, -9.180305480957031, -12.841094970703125, -2.04412841796875, 1.5557670593261719, -1.0234756469726562, 28.24786376953125, 5.293235778808594, 8.593624114990234, -0.05389404296875, 5.663383483886719, 14.160736083984375, -3.6873397827148438, -12.0478515625, 2.0294189453125, -3.7790679931640625, -10.4609375, -0.6740493774414062, -3.2546539306640625, 21.18011474609375, 12.089935302734375, -0.917724609375, 10.544708251953125, 1.3207855224609375, 9.517005920410156, 14.19451904296875, -0.4459075927734375, 9.390289306640625, 5.393218994140625, 13.5394287109375, 7.228248596191406, 5.62750244140625, 3.3159046173095703, -2.794281005859375, -8.81964111328125, 1.5587692260742188, -1.2857494354248047, 8.767730712890625, 2.776123046875, 16.6085205078125, 8.90966796875, 11.714618682861328, 25.58544921875, 15.95928955078125, 27.03546142578125, 11.132232666015625, 4.4243927001953125, 13.880126953125, 11.395736694335938, -0.733062744140625, 1.2848358154296875, 3.875, -2.4780960083007812, 30.088623046875, 20.902450561523438, 18.39654541015625, 0.0, 1.0124702453613281, 17.14581298828125, 18.71673583984375, 18.965194702148438, 14.9603271484375, 8.777511596679688, 33.8868408203125, 5.4962158203125, 31.49365234375, 1.5257568359375, 1.4134597778320312, 8.202125549316406, -3.4864654541015625, 22.9140625, 2.6026611328125, 9.016265869140625, -7.304962158203125, -3.2158660888671875, -4.017467498779297, 2.3780288696289062, 11.881538391113281, -25.75933837890625, -31.62249755859375, -4.5796966552734375, -1.0076370239257812, 4.15478515625, 12.645050048828125, 15.753082275390625, 10.022056579589844, 2.0870819091796875, -2.830638885498047, 41.13934326171875, 0.0, 0.7236785888671875, 1.1189498901367188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000070.npy"}
{"epoch": 0.14659685863874344, "step": 71, "batch_size": 128, "mean": 5.189933776855469, "std": 12.832728385925293, "min": -40.413848876953125, "p10": -10.037261962890625, "median": 5.57966423034668, "p90": 23.44851379394531, "max": 34.55218505859375, "pos_frac": 0.703125, "sample": [-4.052879333496094, 31.462860107421875, 0.069061279296875, -13.2152099609375, 6.364776611328125, -24.39117431640625, -4.739891052246094, 25.72467041015625, -19.778549194335938, 9.65203857421875, -0.09951019287109375, 7.010978698730469, -7.126556396484375, 2.840087890625, 5.283084869384766, 14.76080322265625, 2.355621337890625, 7.516082763671875, 11.41448974609375, 17.72467041015625, 6.258567810058594, -10.482711791992188, 6.91552734375, 33.913055419921875, 7.3747406005859375, -1.3329696655273438, 5.185943603515625, 29.52783203125, 18.235694885253906, 13.518402099609375, -11.736190795898438, 3.7308807373046875, 6.737739562988281, 28.99066162109375, -20.21923828125, 9.177764892578125, 18.07305908203125, -0.03377532958984375, 24.97613525390625, -16.111785888671875, 0.1376953125, -2.287994384765625, -10.39764404296875, 5.311180114746094, 2.355560302734375, 0.151519775390625, 5.729736328125, -40.413848876953125, -8.7427978515625, -1.8310356140136719, 6.0394287109375, 22.56378173828125, -1.4815673828125, 7.567298889160156, 4.326595306396484, 13.899627685546875, 24.83343505859375, 3.270416259765625, -1.6121711730957031, -8.6378173828125, 9.65423583984375, -6.03204345703125, 24.013824462890625, 34.55218505859375, 1.0879669189453125, 6.1988525390625, 10.419384002685547, -13.5640869140625, 11.81805419921875, 2.2230186462402344, 0.22344589233398438, 25.665130615234375, 12.232177734375, -9.8828125, -5.803474426269531, 17.20794677734375, 25.81829833984375, -8.165679931640625, -9.10845947265625, 3.308816909790039, -12.116546630859375, -5.944671630859375, 0.461181640625, 8.006324768066406, 9.8128662109375, 20.975799560546875, -0.554718017578125, 8.25250244140625, 13.22320556640625, 9.512908935546875, 6.457145690917969, 8.125030517578125, -1.419281005859375, -0.39790821075439453, 2.1155548095703125, 25.189697265625, -0.178619384765625, 18.096328735351562, 13.225273132324219, 11.42462158203125, 10.084442138671875, -15.422622680664062, 0.11738014221191406, 15.147735595703125, 15.728424072265625, -2.9013442993164062, 4.357330322265625, 4.794027328491211, -8.974609375, 26.55084228515625, 9.559089660644531, 4.4218902587890625, 1.14276123046875, 12.588752746582031, 8.077667236328125, 7.085485458374023, 10.174812316894531, 23.20623779296875, 1.5573196411132812, 9.4425048828125, 5.03515625, 7.2383880615234375, 9.285614013671875, 5.429592132568359, -0.9013175964355469, 21.3956298828125, -35.8997802734375, 7.6004486083984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000071.npy"}
{"epoch": 0.1486910994764398, "step": 72, "batch_size": 128, "mean": 5.0998945236206055, "std": 11.27873706817627, "min": -35.15765380859375, "p10": -6.749811172485351, "median": 4.43699836730957, "p90": 20.386669921874997, "max": 31.516510009765625, "pos_frac": 0.6953125, "sample": [13.690673828125, 6.3408203125, -2.5474319458007812, -7.159942626953125, 27.0045166015625, -1.298980712890625, 7.345771789550781, 16.831634521484375, 3.3713531494140625, 1.5658836364746094, 2.527099609375, -1.66937255859375, 1.26593017578125, 5.4319610595703125, 16.3489990234375, 15.940032958984375, -8.642292022705078, 3.965576171875, -14.805450439453125, 4.53302001953125, 11.872261047363281, 21.228515625, 3.86993408203125, 21.044586181640625, 8.277969360351562, 3.2013015747070312, -5.434600830078125, 10.005867004394531, -7.813728332519531, -0.8225479125976562, 8.525726318359375, -31.33428955078125, -5.5911865234375, 0.5884246826171875, 7.9012298583984375, 6.32806396484375, 0.7505035400390625, -4.137182235717773, 15.343536376953125, 5.9045867919921875, 5.1094970703125, -6.277801513671875, 21.16607666015625, 3.43927001953125, 12.531784057617188, 12.226287841796875, 4.8875274658203125, -3.1944656372070312, -21.375457763671875, -9.02301025390625, -2.2115707397460938, 8.09381103515625, -16.8280029296875, 0.0, 21.94244384765625, 16.632843017578125, 22.727890014648438, -2.0797576904296875, 30.236785888671875, 17.93243408203125, -2.636383056640625, -7.1221160888671875, 4.521595001220703, 16.68145751953125, 7.527915954589844, 13.564743041992188, 5.0494232177734375, 4.55377197265625, 0.7252464294433594, 2.873382568359375, -6.2935638427734375, 5.872486114501953, 6.322265625, 10.358970642089844, 20.104705810546875, 3.7413482666015625, 2.977771759033203, 9.174774169921875, -9.03729248046875, 17.250164031982422, 4.3524017333984375, 3.574125289916992, -6.590251922607422, -1.439239501953125, 0.8373565673828125, -5.184112548828125, 9.872421264648438, 3.75054931640625, 0.6414470672607422, 13.814422607421875, -14.37200927734375, 5.99493408203125, -0.469635009765625, 6.396575927734375, 4.186187744140625, 14.46511459350586, 18.244461059570312, 2.530609130859375, 9.82061767578125, 13.695465087890625, 12.146087646484375, 25.371856689453125, 0.0, 19.447921752929688, -1.0920066833496094, -1.1874351501464844, 28.71478271484375, 31.516510009765625, 11.797103881835938, 8.172569274902344, 2.0456008911132812, -0.7922515869140625, -4.307525634765625, -35.15765380859375, 23.27081298828125, 0.4293670654296875, -2.8881988525390625, 0.0, 4.1806182861328125, 13.903030395507812, 0.5667572021484375, 9.282569885253906, -19.37396240234375, 21.94439697265625, 8.368377685546875, -6.4411773681640625, 14.63970947265625, 22.211181640625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000072.npy"}
{"epoch": 0.15078534031413612, "step": 73, "batch_size": 128, "mean": 4.807818412780762, "std": 12.773103713989258, "min": -52.339019775390625, "p10": -10.816107177734375, "median": 4.442623138427734, "p90": 19.99943389892578, "max": 35.26995849609375, "pos_frac": 0.6875, "sample": [6.339927673339844, 8.616363525390625, 15.05303955078125, 14.104949951171875, -8.329193115234375, 5.484596252441406, -10.673797607421875, 1.13189697265625, 1.389801025390625, 15.479866027832031, 6.2406005859375, 12.757598876953125, 3.3896484375, 23.41937255859375, -0.8149833679199219, -1.0210037231445312, -26.3359375, -16.004791259765625, -1.7480316162109375, 1.5168304443359375, 5.311492919921875, 1.53094482421875, 0.7866439819335938, 9.784912109375, 10.064102172851562, 1.7186012268066406, 15.990570068359375, 3.276702880859375, 9.14971923828125, 13.21734619140625, -17.693603515625, -19.56572723388672, 16.63092041015625, 30.835113525390625, -0.28839111328125, 6.325523376464844, -11.407218933105469, 32.42364501953125, 14.636161804199219, 7.0178070068359375, 13.221481323242188, 4.5403594970703125, -0.585479736328125, -1.4951171875, 32.353515625, 0.3166351318359375, -5.8276519775390625, -6.38775634765625, 7.9394683837890625, 32.369415283203125, -11.148162841796875, -4.2138671875, 21.070556640625, -52.339019775390625, 23.460906982421875, 3.9653587341308594, 3.041351318359375, -18.466415405273438, 20.420944213867188, -11.836479187011719, 0.983612060546875, 7.5009765625, 4.344886779785156, 5.0174560546875, -9.26165771484375, 15.676300048828125, 12.077861785888672, 0.493408203125, 23.001129150390625, -0.24313926696777344, -1.9167289733886719, -14.7437744140625, -0.454376220703125, 6.982147216796875, -2.1292724609375, 0.7936458587646484, -14.911895751953125, -14.76190185546875, -3.4790573120117188, 4.31256103515625, 7.7320556640625, 19.81878662109375, 12.04119873046875, 1.876495361328125, -4.995025634765625, 13.686965942382812, -2.0419158935546875, 3.806018829345703, 1.9554672241210938, 12.053985595703125, 5.269304275512695, 9.032211303710938, 20.985076904296875, 13.493621826171875, 12.710845947265625, 16.6304931640625, 3.434385299682617, 7.07025146484375, -18.349960327148438, 15.74102783203125, -3.67718505859375, 17.484619140625, 9.700881958007812, 13.10711669921875, 12.0997314453125, 12.252059936523438, -0.8570785522460938, 2.588623046875, 33.39892578125, 16.488616943359375, 11.74560546875, -6.4051513671875, 5.892784118652344, -2.302776336669922, 3.8822174072265625, -2.7789306640625, 3.8563594818115234, 35.26995849609375, -9.0068359375, -3.08941650390625, 26.791168212890625, 3.3092269897460938, 12.14654541015625, 6.176338195800781, 6.666042327880859, 8.308517456054688, 10.139892578125, -9.16265869140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000073.npy"}
{"epoch": 0.15287958115183245, "step": 74, "batch_size": 128, "mean": 4.41343879699707, "std": 12.234708786010742, "min": -25.543487548828125, "p10": -11.89435119628906, "median": 3.3111209869384766, "p90": 19.556332397460938, "max": 39.5848388671875, "pos_frac": 0.65625, "sample": [3.2196044921875, 19.3052978515625, -4.681182861328125, 0.7799072265625, 18.004592895507812, -0.97509765625, -15.165451049804688, 2.761962890625, 11.588966369628906, 12.52587890625, -7.66943359375, 22.745086669921875, 0.0, 12.8614501953125, -3.3341064453125, 24.6466064453125, 17.1131591796875, 27.853485107421875, -13.830734252929688, -4.77880859375, 1.24725341796875, 10.242263793945312, 2.3494873046875, 8.645439147949219, 23.308563232421875, -21.424163818359375, -14.51995849609375, 0.19964599609375, -20.04095458984375, -11.370697021484375, -8.5902099609375, -14.69921875, -7.636962890625, -4.432685852050781, 19.607666015625, 18.85400390625, -3.8033523559570312, 9.272071838378906, 0.702178955078125, 13.536209106445312, -0.8870143890380859, -6.785125732421875, -11.3214111328125, -17.16400146484375, -0.2723541259765625, 19.534332275390625, 1.46734619140625, -9.306243896484375, -2.47637939453125, 4.174079895019531, -13.16607666015625, -20.52392578125, 7.543975830078125, 10.203033447265625, -2.659820556640625, -3.9799423217773438, 3.1444473266601562, 2.0489349365234375, -24.07720947265625, 19.130645751953125, 6.0344085693359375, 7.1411895751953125, -2.70623779296875, 20.5318603515625, 9.51629638671875, 1.196075439453125, 18.29864501953125, 5.8795013427734375, 5.02313232421875, 39.5848388671875, 11.88690185546875, 2.560302734375, -0.103118896484375, 0.891357421875, 7.584075927734375, 3.2498931884765625, 7.3276824951171875, 3.0986328125, -7.56292724609375, 21.01324462890625, 8.582107543945312, -0.1280364990234375, 6.737945556640625, -25.543487548828125, 16.80047607421875, 3.3723487854003906, 4.53485107421875, 7.313873291015625, 2.2149410247802734, -3.1044960021972656, -0.5775146484375, -5.2958221435546875, 3.02734375, 13.649803161621094, 12.77117919921875, 8.7811279296875, 12.059844970703125, -17.3287353515625, -7.960296630859375, -2.97698974609375, 13.45709228515625, -4.7277374267578125, 8.229537963867188, 20.9622802734375, 12.47772216796875, 12.685928344726562, 14.31402587890625, -3.28680419921875, 12.72454833984375, -13.1162109375, 2.975078582763672, 2.6614913940429688, 9.44488525390625, 4.5361480712890625, 32.2330322265625, 3.916015625, 5.253883361816406, 11.91192626953125, 13.102386474609375, -2.883575439453125, 15.359161376953125, 6.963104248046875, 5.9034881591796875, 22.531097412109375, 32.80096435546875, 16.028106689453125, 0.815338134765625, 31.225982666015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000074.npy"}
{"epoch": 0.1549738219895288, "step": 75, "batch_size": 128, "mean": 6.536618232727051, "std": 11.554920196533203, "min": -28.661964416503906, "p10": -7.121087455749512, "median": 5.650794982910156, "p90": 21.44186096191406, "max": 37.1639404296875, "pos_frac": 0.7265625, "sample": [-4.234657287597656, 7.105686187744141, 2.2396163940429688, 4.2243499755859375, -12.8939208984375, 12.792724609375, 17.480850219726562, -8.895416259765625, -3.4681243896484375, 4.796943664550781, 7.257988929748535, -15.83197021484375, 4.911693572998047, 1.0978584289550781, 21.7318115234375, -3.375276565551758, 31.138671875, 0.36669921875, 22.354217529296875, 10.447052001953125, 15.610107421875, 16.58226203918457, 17.355560302734375, -18.803955078125, -3.7328720092773438, 17.29486083984375, -7.163330078125, 14.814727783203125, 5.180023193359375, 0.33685302734375, 9.8668212890625, 24.915878295898438, 15.314208984375, 5.7410430908203125, -0.8292007446289062, -7.102983474731445, 14.503799438476562, 7.62762451171875, 14.086372375488281, -0.8846435546875, 21.317596435546875, 11.438156127929688, 3.339111328125, -16.73236083984375, 9.357864379882812, 4.7236328125, -1.0293502807617188, 3.3893890380859375, 8.88507080078125, -1.3504867553710938, 33.420379638671875, 15.251434326171875, 12.666961669921875, 11.054779052734375, 7.7947998046875, 13.408218383789062, -12.21466064453125, 20.702880859375, -5.6134490966796875, 10.74920654296875, 13.43621826171875, 5.054658889770508, -0.19051361083984375, 10.616012573242188, 2.544647216796875, 6.49072265625, 21.790924072265625, -2.6267242431640625, 10.541084289550781, 16.530181884765625, 8.06439208984375, 5.560546875, -0.5362968444824219, 37.1639404296875, 26.45648193359375, -0.6461944580078125, -18.9110107421875, -28.661964416503906, 3.288055419921875, 11.73348617553711, 27.172821044921875, 19.3099365234375, 12.419403076171875, -13.666168212890625, -2.421875, 23.17828369140625, 1.0205612182617188, -11.122268676757812, 5.118682861328125, 8.799942016601562, 6.280517578125, -3.102142333984375, 12.25152587890625, 12.871337890625, 17.014678955078125, 22.23822021484375, 0.3579864501953125, 14.44940185546875, 1.823333740234375, 11.675403594970703, -15.865386962890625, 4.11053466796875, 25.3011474609375, 10.08807373046875, 19.326385498046875, 22.014892578125, 1.2193984985351562, 19.217666625976562, 0.899993896484375, -6.54083251953125, 16.51129150390625, 0.4597797393798828, 5.29193115234375, 10.310245513916016, 4.255035400390625, 18.8504638671875, -0.883148193359375, -0.192626953125, 14.308502197265625, -1.276275634765625, 4.02020263671875, -4.803680419921875, 4.971004486083984, 16.3687744140625, -2.0306549072265625, 1.803436279296875, -16.945419311523438, 2.009063720703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000075.npy"}
{"epoch": 0.15706806282722513, "step": 76, "batch_size": 128, "mean": 7.157843589782715, "std": 13.303543090820312, "min": -29.0667724609375, "p10": -5.235491943359374, "median": 4.766223907470703, "p90": 25.500369262695312, "max": 46.228729248046875, "pos_frac": 0.7109375, "sample": [17.72344970703125, -3.90631103515625, 9.397418975830078, 4.338287353515625, 2.4894790649414062, 11.38232421875, 8.194061279296875, 5.373067855834961, 6.700798034667969, 3.407958984375, 37.6072998046875, -0.21195220947265625, 10.00250244140625, -2.339385986328125, 15.998046875, 15.69427490234375, -1.7772979736328125, -11.42047119140625, 7.137664794921875, 1.875946044921875, 39.45343017578125, 1.9925155639648438, 0.32293701171875, -1.341827392578125, 4.3055877685546875, 28.793212890625, -2.5511322021484375, 10.236793518066406, 4.328189849853516, -0.88299560546875, 10.1041259765625, -1.964956283569336, 8.965713500976562, 19.91021728515625, -27.337432861328125, 38.308624267578125, 9.804359436035156, 15.43621826171875, 10.13580322265625, 9.0433349609375, 25.605621337890625, 22.021697998046875, 15.734954833984375, 16.676559448242188, -3.477386474609375, 3.441669464111328, 9.541488647460938, 5.26751708984375, -5.080474853515625, 18.679473876953125, 0.0, 2.04779052734375, 5.635986328125, 1.3875732421875, 27.598861694335938, -15.12890625, 46.228729248046875, -29.0667724609375, -5.597198486328125, -15.79736328125, 4.04510498046875, 42.33807373046875, -8.6036376953125, 25.45526123046875, 10.839344024658203, 0.9854660034179688, -3.5377197265625, 1.8846893310546875, 13.962028503417969, -7.245262145996094, 1.3682098388671875, 21.87890625, -3.0272216796875, -1.1899871826171875, -12.614913940429688, 0.0, 21.332550048828125, 10.374717712402344, 13.180068969726562, 33.30865478515625, 13.125, -0.6860122680664062, 3.88433837890625, 4.442955017089844, 6.096153259277344, 7.58203125, 12.329635620117188, 24.28570556640625, 7.0450897216796875, 8.41143798828125, 10.7601318359375, 1.7884674072265625, 1.5194473266601562, 11.1229248046875, -2.0313243865966797, 1.9687576293945312, -10.365364074707031, -5.066558837890625, 5.4194793701171875, -0.8347091674804688, -10.8402099609375, 3.5730056762695312, 22.099273681640625, 29.802825927734375, 21.9866943359375, 26.851898193359375, 37.53511047363281, 5.32415771484375, 0.60894775390625, -11.273651123046875, -0.7095184326171875, -4.6654205322265625, 30.778182983398438, -3.6844482421875, -20.597076416015625, 2.0401611328125, 20.5389404296875, 8.904769897460938, 22.7303466796875, 2.6875152587890625, 2.7520065307617188, 9.64453125, 5.0894927978515625, 8.772567749023438, -4.586673736572266, 2.1912384033203125, -0.8507843017578125, 3.5184783935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000076.npy"}
{"epoch": 0.15916230366492146, "step": 77, "batch_size": 128, "mean": 5.799104690551758, "std": 14.499119758605957, "min": -35.3572998046875, "p10": -10.219247436523437, "median": 4.422325134277344, "p90": 25.56023254394531, "max": 49.813079833984375, "pos_frac": 0.640625, "sample": [9.458314895629883, 4.1151275634765625, 6.615875244140625, 13.000946044921875, 4.24212646484375, -0.7085342407226562, 19.864501953125, 16.542022705078125, -3.728506088256836, 17.234085083007812, -5.84515380859375, -0.5333423614501953, 13.5570068359375, 16.5157470703125, 26.613677978515625, 9.67115592956543, 2.22265625, 14.3050537109375, -33.300628662109375, 14.206893920898438, -2.7615394592285156, -4.80670166015625, -8.8739013671875, -11.0821533203125, 31.569244384765625, 20.656585693359375, 14.333389282226562, -0.2623138427734375, 15.1920166015625, 15.203125, -4.0556640625, 6.83599853515625, 4.6025238037109375, 20.897064208984375, -0.62615966796875, -1.5449295043945312, 8.502426147460938, -3.86553955078125, 11.7635498046875, 0.9311676025390625, 0.7542266845703125, -3.386199951171875, 26.166046142578125, 10.166412353515625, 5.6779632568359375, 10.0888671875, 3.233478546142578, 36.88897705078125, -8.26806640625, 7.855560302734375, 4.19964599609375, -0.16107177734375, 27.990966796875, 49.813079833984375, 15.48040771484375, 10.539314270019531, -22.635040283203125, -0.9871826171875, 4.9760589599609375, 0.8136692047119141, -6.7747802734375, -9.59503173828125, 28.27142333984375, 44.2403564453125, -7.840545654296875, 15.45159912109375, 6.436256408691406, 4.9791259765625, 25.411102294921875, 6.725559234619141, 14.042938232421875, 9.050483703613281, -35.3572998046875, -15.549636840820312, -9.987342834472656, -2.28143310546875, 10.827423095703125, 3.2412643432617188, -1.396331787109375, 3.0151824951171875, -6.99737548828125, 20.012985229492188, -11.720619201660156, -16.74237060546875, 2.9154891967773438, 12.433135986328125, 30.778762817382812, -16.6055908203125, -14.175582885742188, 15.918617248535156, 11.39178466796875, 4.145851135253906, -5.319370269775391, 5.6644744873046875, 12.846389770507812, -6.7708740234375, -9.215507507324219, -15.04364013671875, 17.172088623046875, 3.7881927490234375, -13.47698974609375, 5.4878082275390625, -0.7200336456298828, 10.205352783203125, 16.832550048828125, 3.5894622802734375, -1.4402313232421875, 27.83837890625, 34.98516845703125, -1.9727630615234375, 6.310279846191406, -21.830596923828125, -10.563201904296875, 45.505615234375, 1.3548927307128906, 1.769561767578125, -10.07183837890625, 10.979568481445312, -4.596488952636719, 3.2403564453125, 5.43695068359375, -0.196044921875, 24.292236328125, 23.2030029296875, 14.577362060546875, 25.908203125, -8.494110107421875, 0.8794670104980469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000077.npy"}
{"epoch": 0.1612565445026178, "step": 78, "batch_size": 128, "mean": 8.663091659545898, "std": 14.819053649902344, "min": -24.371566772460938, "p10": -8.625130462646483, "median": 6.768598556518555, "p90": 29.53636779785156, "max": 61.7733154296875, "pos_frac": 0.7265625, "sample": [-3.8975677490234375, 10.84442138671875, -16.450775146484375, 17.519027709960938, 33.987091064453125, 0.46990203857421875, 11.191734313964844, 22.69403076171875, -11.630470275878906, 15.382865905761719, -1.2652130126953125, 2.8387680053710938, 14.022541046142578, -7.4896087646484375, -1.86236572265625, 11.584625244140625, 5.050910949707031, 15.993362426757812, 29.380096435546875, -0.1998291015625, 0.0, 25.521148681640625, -17.834762573242188, 10.448051452636719, 5.7239227294921875, 5.534152984619141, 20.69268798828125, 1.9582595825195312, -11.454254150390625, 11.79730224609375, -7.174491882324219, 21.65472412109375, 26.266571044921875, 0.430908203125, 21.326080322265625, -13.383026123046875, 16.217880249023438, 34.228790283203125, -11.490264892578125, -24.371566772460938, 9.340156555175781, 17.1802978515625, 4.7577667236328125, 61.7733154296875, 35.47454833984375, -4.893218994140625, 11.197799682617188, 8.10498046875, 20.336517333984375, 0.3900604248046875, 18.613037109375, -14.34173583984375, 4.125404357910156, 6.987209320068359, 15.18353271484375, -2.3478641510009766, 4.200531005859375, 10.60150146484375, 11.461654663085938, -3.0503463745117188, 37.66180419921875, -3.05279541015625, 8.206718444824219, 18.855010986328125, 7.662441253662109, -15.083969116210938, 14.14276123046875, 12.96087646484375, 11.87603759765625, -2.5977783203125, 40.19854736328125, 14.366348266601562, 5.584014892578125, 6.4067840576171875, 37.00274658203125, 17.57379150390625, 5.325069427490234, 8.81155776977539, -1.184783935546875, 1.6962509155273438, -17.277679443359375, -6.198173522949219, 29.9010009765625, 30.30963134765625, -4.983917236328125, 6.473884582519531, 0.6615066528320312, 2.203460693359375, -9.257171630859375, 24.82958984375, 2.995767593383789, -8.354255676269531, 20.473800659179688, -5.631496429443359, -16.353240966796875, 0.227508544921875, 4.199554443359375, -16.808319091796875, 19.2265625, 14.290176391601562, -2.1802940368652344, 2.2835235595703125, -8.105621337890625, 38.7781982421875, 9.4251708984375, 10.130462646484375, 14.269012451171875, -5.007354736328125, 5.3070068359375, 1.2980728149414062, 34.2303466796875, 8.6376953125, -0.1627197265625, 25.619308471679688, 20.98968505859375, 4.600341796875, 26.8935546875, 31.00128173828125, 23.377700805664062, -7.59161376953125, 0.741912841796875, 0.8484725952148438, 40.32940673828125, 1.4753990173339844, 15.310928344726562, 8.148551940917969, 6.54998779296875, 24.986846923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000078.npy"}
{"epoch": 0.16335078534031414, "step": 79, "batch_size": 128, "mean": 8.30783462524414, "std": 15.502094268798828, "min": -47.507080078125, "p10": -7.883104133605956, "median": 6.136613845825195, "p90": 30.596826171874998, "max": 58.45536804199219, "pos_frac": 0.71875, "sample": [-3.1640625, 6.1374664306640625, 58.45536804199219, -11.06268310546875, 8.680877685546875, -3.216949462890625, 1.64520263671875, -0.388763427734375, 2.7486724853515625, 2.1666107177734375, 36.2764892578125, -1.45672607421875, 0.0, 4.963623046875, -8.287176132202148, -6.444002151489258, -3.348419189453125, -6.1593475341796875, -17.89263153076172, 0.855438232421875, 13.125518798828125, -22.41375732421875, -2.3099899291992188, 0.7252578735351562, -9.901611328125, 6.506103515625, 9.082061767578125, 0.0, 26.409317016601562, 7.79844856262207, -21.552215576171875, 13.40264892578125, 18.10113525390625, 10.577133178710938, 11.908042907714844, 1.486083984375, 2.321533203125, 11.011962890625, 7.3250885009765625, -0.22119140625, 19.323410034179688, -20.1104736328125, 15.476776123046875, -1.9579105377197266, 1.264434814453125, 20.425369262695312, 6.168407440185547, 33.7630615234375, 0.17095184326171875, 4.871185302734375, 11.5194091796875, -8.6627197265625, -47.507080078125, -15.27337646484375, 11.602935791015625, 22.793426513671875, 6.234527587890625, 23.550323486328125, 21.513885498046875, 48.810516357421875, -3.127391815185547, 2.899993896484375, 15.215934753417969, 28.092575073242188, 30.282012939453125, -4.490264892578125, 20.526214599609375, 5.864559173583984, 6.9020233154296875, 16.42498779296875, -9.2738037109375, 13.014640808105469, 6.135761260986328, 4.05572509765625, 2.12335205078125, -7.273162841796875, 26.315338134765625, 6.100433349609375, 29.1376953125, 4.5174713134765625, -3.7628097534179688, 37.61285400390625, 33.069091796875, 19.711334228515625, 6.899566650390625, 16.26336669921875, 8.282127380371094, 1.3534278869628906, -0.5765266418457031, 3.358348846435547, 1.4074535369873047, 11.183116912841797, 5.189208984375, 7.08575439453125, 5.214427947998047, 37.3585205078125, 2.957456588745117, 36.18132019042969, -3.6367034912109375, 32.2366943359375, 25.86090087890625, 29.454437255859375, 10.793853759765625, 9.479400634765625, 2.455230712890625, 4.402181625366211, 15.427154541015625, 7.1105804443359375, 31.331390380859375, 1.163330078125, -3.568450927734375, 36.627899169921875, -1.77423095703125, 0.17979812622070312, 11.212873458862305, 17.272537231445312, 16.61468505859375, 25.44775390625, -7.1581573486328125, 14.709060668945312, -1.0871124267578125, 18.052459716796875, -14.37176513671875, 33.983154296875, 8.892768859863281, -10.322906494140625, -7.709930419921875, 40.224273681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000079.npy"}
{"epoch": 0.16544502617801046, "step": 80, "batch_size": 128, "mean": 6.596972465515137, "std": 14.170262336730957, "min": -27.111572265625, "p10": -8.818209838867187, "median": 4.923023223876953, "p90": 25.779098510742188, "max": 58.8798828125, "pos_frac": 0.6328125, "sample": [4.749549865722656, 14.277259826660156, -2.683624267578125, 13.339218139648438, 5.181060791015625, -6.17706298828125, -3.7085113525390625, -26.56683349609375, 24.70550537109375, -4.078857421875, 1.360809326171875, -0.20166015625, 1.154144287109375, 10.07037353515625, 4.546913146972656, -6.375205993652344, -0.0308837890625, 23.109222412109375, 12.299835205078125, 8.60693359375, 36.259307861328125, -3.0100250244140625, 6.354705810546875, -0.1132049560546875, 8.82366943359375, 18.99945068359375, 12.39105224609375, -7.81866455078125, 10.047876358032227, 27.284515380859375, 7.478515625, 2.5185604095458984, 3.384033203125, -3.4060611724853516, -1.218017578125, 2.51385498046875, -27.111572265625, 25.923828125, 0.0, 29.482086181640625, -4.40875244140625, 20.43438720703125, -0.5870208740234375, 3.2669677734375, 9.665283203125, -18.339111328125, -4.363250732421875, 12.212646484375, 2.3739013671875, -6.724311828613281, -3.993427276611328, -2.9822998046875, 9.666213989257812, 20.149581909179688, 14.827606201171875, -1.5186386108398438, -2.8748855590820312, 0.8115921020507812, -2.10955810546875, 30.334823608398438, 5.283111572265625, 8.074104309082031, 6.059749603271484, -9.919952392578125, -3.0474853515625, 4.60455322265625, 18.905441284179688, 10.775650024414062, 4.341377258300781, 8.80877685546875, -8.497711181640625, 8.26153564453125, 9.746429443359375, -9.259330749511719, 58.8798828125, 2.784393310546875, -2.913238525390625, 1.93084716796875, 15.1240234375, -18.380035400390625, 16.026214599609375, 7.648307800292969, -9.766632080078125, 32.921478271484375, 18.111572265625, 22.857807159423828, -14.564239501953125, 16.39141845703125, -1.7823257446289062, 27.34161376953125, 25.41083526611328, 9.060653686523438, 27.803024291992188, 7.24346923828125, 6.080322265625, 25.717071533203125, -1.98944091796875, 5.627098083496094, -17.706451416015625, -2.0681724548339844, -17.081298828125, 36.737335205078125, -15.448005676269531, 21.98406982421875, 22.48651123046875, 19.69757080078125, 3.140289306640625, -7.254341125488281, -2.8142547607421875, 27.975860595703125, -9.33599853515625, 9.599258422851562, -0.03155517578125, -7.350730895996094, 5.09649658203125, -8.629158020019531, 10.190019607543945, -1.4298591613769531, 6.182155609130859, -9.385986328125, 24.11651611328125, 37.07342529296875, 14.43560791015625, 12.59674072265625, 44.05291748046875, 1.3218631744384766, 2.9567947387695312, 13.400634765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000080.npy"}
{"epoch": 0.16753926701570682, "step": 81, "batch_size": 128, "mean": 7.144075393676758, "std": 17.26014518737793, "min": -43.38214111328125, "p10": -12.854739379882812, "median": 6.122616767883301, "p90": 30.977801513671874, "max": 55.2291259765625, "pos_frac": 0.6875, "sample": [0.7553558349609375, -1.3203086853027344, 19.81861114501953, -20.64508056640625, 3.63848876953125, 0.3546295166015625, -17.509002685546875, 7.840095520019531, 42.75634765625, 31.196380615234375, 19.134414672851562, 22.01171875, 14.621597290039062, -7.319770812988281, 17.367172241210938, 0.7436676025390625, 11.7119140625, 29.344772338867188, 10.388580322265625, 11.49151611328125, 8.427318572998047, 38.202239990234375, 8.234703063964844, -6.189178466796875, 6.423614501953125, 1.4694862365722656, -1.743438720703125, 18.82635498046875, 23.4857177734375, 33.695953369140625, 28.68426513671875, -8.284698486328125, 19.379844665527344, 2.96099853515625, -13.144134521484375, 5.9550933837890625, -14.979736328125, 43.46929931640625, 29.165802001953125, 9.829803466796875, -10.597610473632812, -10.863983154296875, 24.036041259765625, -2.1214065551757812, -26.819503784179688, -0.4359283447265625, -43.38214111328125, -10.44769287109375, 19.431243896484375, 55.2291259765625, 6.823089599609375, 8.293075561523438, -8.94256591796875, 23.511016845703125, -15.341815948486328, -4.983940124511719, 10.00408935546875, 27.05499267578125, 3.5941162109375, -35.797698974609375, -10.5670166015625, -0.09716796875, 2.386444091796875, 5.8295135498046875, -3.4686145782470703, 9.981925964355469, -5.9351806640625, -0.4705848693847656, 7.363037109375, 6.270416259765625, -24.6019287109375, 9.105091094970703, -19.89642333984375, 30.884124755859375, 30.824737548828125, 23.73480224609375, -2.186920166015625, 14.4442138671875, 3.9776172637939453, 14.32061767578125, -23.75189208984375, 9.45281982421875, 34.615631103515625, -10.571823120117188, 9.2667236328125, 9.499614715576172, 5.1702880859375, -12.730712890625, 0.73663330078125, 5.974817276000977, -18.57501220703125, -8.513839721679688, -24.098175048828125, -6.869140625, 2.947479248046875, 2.2399673461914062, 6.4815216064453125, 1.5634441375732422, 32.955322265625, 35.46881103515625, 35.419891357421875, 22.50433349609375, 12.4539794921875, -4.62847900390625, 27.88421630859375, 6.934131622314453, 0.6068115234375, 2.1562042236328125, 7.7902727127075195, 16.406280517578125, -11.630393981933594, 16.18787384033203, 2.221698760986328, -3.0014190673828125, 3.17962646484375, 24.20728302001953, 26.601318359375, -2.4170761108398438, 33.44598388671875, 8.357330322265625, 7.256195068359375, 37.10968017578125, 38.68115234375, 20.992050170898438, 5.43377685546875, 2.5217056274414062, 2.5431556701660156, -6.4300689697265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000081.npy"}
{"epoch": 0.16963350785340314, "step": 82, "batch_size": 128, "mean": 7.9389190673828125, "std": 16.92216682434082, "min": -34.385589599609375, "p10": -10.834771728515625, "median": 6.942386627197266, "p90": 25.45235900878906, "max": 59.448699951171875, "pos_frac": 0.7109375, "sample": [5.2554931640625, 25.227752685546875, -0.28348541259765625, -11.55096435546875, 4.941474914550781, 10.2347412109375, 11.01300048828125, 57.753753662109375, 16.4080810546875, -6.576301574707031, -3.861358642578125, 8.8140869140625, -7.6083221435546875, -22.335601806640625, -1.04083251953125, 21.143798828125, 27.108184814453125, 11.937881469726562, 16.53106689453125, 17.732345581054688, 4.1298828125, -15.892608642578125, 15.291656494140625, -5.060760498046875, 2.7809410095214844, 26.96490478515625, -25.386444091796875, 11.172065734863281, -2.9756927490234375, 3.7041168212890625, 55.834228515625, -4.4522857666015625, 22.314376831054688, 13.36578369140625, 28.784637451171875, 16.1497802734375, 0.24138641357421875, 8.079185485839844, 14.196319580078125, -34.385589599609375, -16.624343872070312, -2.1924514770507812, 6.0529022216796875, 10.670074462890625, -14.398735046386719, 52.4462890625, -15.156524658203125, 46.03399658203125, 44.5989990234375, -8.159576416015625, -4.304901123046875, 59.448699951171875, 15.133316040039062, 10.147392272949219, 15.3004150390625, 0.1497173309326172, 6.66748046875, -5.324859619140625, 4.1107025146484375, 2.8139801025390625, 21.826522827148438, -18.545944213867188, -29.303558349609375, 4.216510772705078, -13.7235107421875, 4.5508575439453125, -6.8020477294921875, -6.140098571777344, 27.525360107421875, 12.415512084960938, -1.29449462890625, 22.900665283203125, 8.655487060546875, 8.369621276855469, -10.28857421875, 8.438568115234375, 6.860748291015625, 5.127859115600586, 9.041202545166016, 15.654510498046875, 6.550971984863281, 55.71942138671875, -9.141571044921875, 24.28350830078125, 7.568023681640625, 13.021339416503906, 18.96087646484375, 14.669418334960938, 7.024024963378906, 10.244712829589844, 6.543724060058594, 1.91595458984375, 9.549980163574219, 22.1893310546875, 4.044361114501953, 8.583984375, -10.52783203125, 18.289825439453125, 20.695953369140625, 21.5247802734375, 11.842254638671875, 1.0302982330322266, 5.47509765625, 12.4378662109375, -3.595062255859375, 6.367116928100586, 2.553985595703125, 0.619476318359375, -6.86114501953125, 5.776031494140625, -4.619819641113281, 15.77880859375, 17.681087493896484, 18.574256896972656, -16.183746337890625, 9.3543701171875, 40.93084716796875, -1.8155899047851562, 18.042739868164062, 16.081253051757812, -8.552032470703125, 24.874755859375, 25.9764404296875, 7.320281982421875, 2.3199501037597656, 2.6709442138671875, -8.961669921875, -23.246437072753906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000082.npy"}
{"epoch": 0.17172774869109947, "step": 83, "batch_size": 128, "mean": 8.669137001037598, "std": 19.22634506225586, "min": -57.71478271484375, "p10": -10.47749557495117, "median": 7.230964660644531, "p90": 35.31246643066406, "max": 53.88385009765625, "pos_frac": 0.734375, "sample": [12.20599365234375, 0.0, 33.867401123046875, -0.965087890625, 14.389328002929688, 1.4141845703125, -10.258979797363281, 28.799346923828125, 10.90484619140625, -6.5070343017578125, -5.966522216796875, -57.71478271484375, 5.795616149902344, -42.798614501953125, 1.649627685546875, 3.078094482421875, 19.230224609375, 49.14959716796875, 10.665321350097656, 11.617084503173828, 17.622100830078125, 19.785144805908203, 2.467864990234375, 19.761734008789062, 27.152587890625, -10.98736572265625, 6.8196258544921875, -0.32558631896972656, -6.536041259765625, 41.749176025390625, 38.8060302734375, -14.3271484375, 22.26812744140625, 4.149383544921875, -11.247909545898438, 31.0936279296875, 1.1008529663085938, 5.807037353515625, 43.566375732421875, 7.642303466796875, 9.896675109863281, -2.5951614379882812, 5.99464225769043, 2.570343017578125, -3.9429931640625, 35.10955810546875, 3.7485809326171875, 3.001312255859375, 12.056365966796875, -4.7451934814453125, 13.852920532226562, 20.10174560546875, 46.44482421875, 16.0677490234375, -50.128082275390625, 0.34521484375, 18.140533447265625, 11.91900634765625, 35.785919189453125, -5.10113525390625, -3.6771087646484375, 20.194549560546875, 13.254180908203125, 28.362518310546875, 3.740509033203125, 4.7142333984375, -5.22821044921875, 5.328742980957031, -23.13726806640625, 13.158828735351562, -20.284957885742188, 14.266433715820312, 9.477546691894531, -0.473602294921875, -14.6080322265625, 8.926040649414062, 53.88385009765625, -20.9169921875, 12.435546875, 26.725311279296875, -9.891090393066406, 31.8157958984375, 13.731658935546875, 1.021575927734375, -46.21923828125, -18.658294677734375, 16.518630981445312, 44.427886962890625, -28.327606201171875, 24.89898681640625, 0.12412071228027344, -5.848182678222656, 4.5287933349609375, 1.888153076171875, 9.78598403930664, 43.16552734375, 37.1500244140625, 9.89385986328125, 12.099838256835938, 6.577156066894531, 30.606353759765625, 1.027801513671875, 17.16351318359375, 27.6043701171875, 1.8487548828125, -3.5162429809570312, 9.965606689453125, 16.33269500732422, 32.05464172363281, -5.764404296875, 16.736083984375, -4.782707214355469, 21.5323486328125, 38.026458740234375, 0.9268798828125, 4.47760009765625, 9.5538330078125, 36.89898681640625, 42.502105712890625, -2.710174560546875, 2.7972640991210938, 0.3972625732421875, 13.026138305664062, -0.70745849609375, 6.1102752685546875, 13.383056640625, 2.3375091552734375, 13.548927307128906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000083.npy"}
{"epoch": 0.17382198952879582, "step": 84, "batch_size": 128, "mean": 7.290060997009277, "std": 17.69985580444336, "min": -50.165863037109375, "p10": -14.233480834960936, "median": 6.763496398925781, "p90": 30.026144409179686, "max": 58.373687744140625, "pos_frac": 0.6875, "sample": [1.6512775421142578, 15.59405517578125, 44.476043701171875, -3.717071533203125, -1.661834716796875, -18.794448852539062, 12.27520751953125, 15.155105590820312, 10.299102783203125, -9.55157470703125, 17.02796173095703, 18.38818359375, 4.618461608886719, -0.6718292236328125, 9.3079833984375, 10.702499389648438, 0.7798385620117188, -5.237640380859375, 4.8731231689453125, 23.887222290039062, -5.159828186035156, 20.193603515625, 9.6375732421875, 14.99359130859375, 20.88531494140625, 12.14398193359375, -0.4956779479980469, 14.20892333984375, 18.22369384765625, 12.904510498046875, -3.16339111328125, -22.7568359375, -50.165863037109375, 28.9888916015625, 12.333221435546875, 3.21649169921875, 14.5491943359375, -2.4709320068359375, 30.65899658203125, 3.0503768920898438, -1.025421142578125, -0.3343505859375, 15.8687744140625, -6.7598114013671875, 22.71893310546875, -24.894058227539062, 9.518058776855469, -0.4605712890625, 20.77618408203125, 13.98175048828125, 58.373687744140625, -28.292724609375, 33.522857666015625, 13.602294921875, 11.318603515625, -19.102264404296875, 15.634696960449219, -9.610809326171875, 14.039886474609375, 5.644012451171875, 10.349365234375, -8.947525024414062, 3.8822479248046875, 8.344482421875, 30.23333740234375, -1.9318695068359375, 8.76544189453125, -2.6800155639648438, 2.380462646484375, 5.687744140625, -13.965286254882812, 2.316192626953125, 4.91851806640625, 1.8005142211914062, 0.5976181030273438, 29.937347412109375, 14.33980941772461, 37.35392761230469, -31.57952880859375, 14.12396240234375, -0.5446014404296875, 3.9107818603515625, -7.188297271728516, 4.03709602355957, -35.36767578125, -1.1500930786132812, -7.80169677734375, 15.316329956054688, 46.168701171875, 3.23138427734375, 28.341705322265625, 5.7033843994140625, -20.88470458984375, 10.366912841796875, -6.9605712890625, 7.4791412353515625, 16.818099975585938, 3.4605674743652344, 14.939437866210938, 10.125091552734375, 6.0478515625, -2.9268722534179688, 5.82843017578125, -6.66180419921875, 17.78363800048828, 15.580787658691406, -11.5760498046875, 34.99163818359375, -32.12017822265625, 26.00228500366211, 34.1195068359375, 5.3608551025390625, 3.443805694580078, 42.026519775390625, 43.20819091796875, 30.6578369140625, -16.50469970703125, -14.859268188476562, 27.968917846679688, 12.946136474609375, 3.106952667236328, 18.133193969726562, 51.36370849609375, -13.114883422851562, 13.74560546875, 15.72064208984375, 12.5697021484375, -21.33966064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000084.npy"}
{"epoch": 0.17591623036649215, "step": 85, "batch_size": 128, "mean": 5.7236199378967285, "std": 16.49332046508789, "min": -33.77081298828125, "p10": -14.226966857910154, "median": 6.057580947875977, "p90": 27.890478515625, "max": 55.211273193359375, "pos_frac": 0.640625, "sample": [5.11846923828125, -8.412216186523438, -0.875732421875, 34.05712890625, 7.019176483154297, 22.07085418701172, -6.165496826171875, -9.92633056640625, 34.154327392578125, 19.481109619140625, 4.038196563720703, -2.1639633178710938, 7.330650329589844, 21.473770141601562, 20.28778076171875, -5.04180908203125, 24.9273681640625, 22.403926849365234, 27.73828125, 9.883079528808594, 5.777191162109375, 2.544219970703125, 38.94941711425781, 33.837432861328125, -3.246368408203125, -13.55206298828125, 15.109146118164062, -6.895576477050781, -10.70648193359375, 35.249542236328125, 12.703094482421875, 1.1717376708984375, -7.910919189453125, 11.6661376953125, 18.613128662109375, 29.784088134765625, -11.799758911132812, 23.257904052734375, -0.32755279541015625, -1.902435302734375, 30.46026611328125, 18.34857177734375, -13.61871337890625, 17.986427307128906, 23.649017333984375, -33.77081298828125, 6.4676971435546875, 7.396141052246094, 4.137399673461914, 33.4168701171875, 0.316558837890625, -18.866668701171875, 7.6551971435546875, 11.680145263671875, 6.066974639892578, -3.8762359619140625, 16.729141235351562, 3.818927764892578, 11.041885375976562, 11.144012451171875, -17.198883056640625, 10.962211608886719, 9.796005249023438, -4.1493377685546875, 23.906051635742188, 19.633331298828125, 37.01922607421875, 19.91241455078125, -10.469375610351562, -23.338531494140625, -7.52850341796875, 17.869213104248047, -28.31884765625, -8.572113037109375, -11.32275390625, -5.99566650390625, 24.49176025390625, 11.87115478515625, 2.3610706329345703, 15.692245483398438, 6.048187255859375, 9.832275390625, 5.85955810546875, 16.6658935546875, 36.200653076171875, -30.531768798828125, -15.72882080078125, 21.415985107421875, 11.362388610839844, -4.680206298828125, -18.625274658203125, 10.041816711425781, 1.915252685546875, 3.2785778045654297, 28.24560546875, -2.7616348266601562, -11.643501281738281, 6.980804443359375, 10.271896362304688, 6.07275390625, -9.656997680664062, 14.64471435546875, 1.3096027374267578, -4.15789794921875, -13.545257568359375, -19.802459716796875, -27.562423706054688, 3.81793212890625, 32.042999267578125, 3.9664840698242188, -15.646224975585938, -25.3974609375, -2.213348388671875, 17.889633178710938, -19.207656860351562, 1.857086181640625, 17.988914489746094, 8.937644958496094, 55.211273193359375, -11.33258056640625, -5.0349884033203125, 10.0552978515625, -7.958732604980469, 3.70819091796875, 6.07183837890625, -10.916727066040039, 11.03936767578125, 9.770830154418945], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000085.npy"}
{"epoch": 0.17801047120418848, "step": 86, "batch_size": 128, "mean": 8.965091705322266, "std": 17.821704864501953, "min": -45.921234130859375, "p10": -9.76460266113281, "median": 10.387138366699219, "p90": 30.184376525878907, "max": 59.46923828125, "pos_frac": 0.6640625, "sample": [17.162521362304688, -45.921234130859375, 17.667755126953125, -16.661285400390625, 20.87006378173828, -4.1119537353515625, 20.6451416015625, 7.816478729248047, 21.082626342773438, 10.225067138671875, 59.46923828125, -0.703765869140625, 6.552022933959961, 10.551239013671875, 14.495384216308594, -5.688663482666016, -13.76019287109375, -5.405181884765625, -1.8167858123779297, -2.1132278442382812, 5.1120758056640625, 11.304367065429688, -1.1427440643310547, -1.85052490234375, 38.927947998046875, -3.2183074951171875, -29.039306640625, 7.6798095703125, 37.7303466796875, 18.7742919921875, -10.78411865234375, 23.09136962890625, -3.853759765625, 29.1531982421875, 48.77252197265625, 22.95733642578125, 20.40185546875, -40.97686767578125, 19.8480224609375, 34.55999755859375, -1.2031936645507812, -12.85638427734375, 17.906272888183594, -7.2919464111328125, 28.883544921875, 0.688232421875, 23.834686279296875, 26.787384033203125, 30.0975341796875, 4.792747497558594, -24.995880126953125, 8.36962890625, 16.7001953125, 11.660110473632812, 19.004730224609375, -4.8791351318359375, 32.171234130859375, -25.06011962890625, 12.143394470214844, 16.08001708984375, 32.38726806640625, 30.387008666992188, 6.105865478515625, 24.161788940429688, 20.133544921875, 6.84515380859375, 0.0, -1.685302734375, -0.4490966796875, 17.41748046875, 33.065338134765625, 24.84649658203125, -26.789642333984375, 24.43225860595703, 2.693462371826172, -2.8211135864257812, -5.396030426025391, -9.277496337890625, 39.97593688964844, 27.40545654296875, 1.9466400146484375, 12.624664306640625, 22.93023681640625, -1.2675933837890625, 5.9796295166015625, -19.2318115234375, -38.4510498046875, 14.3961181640625, -3.387969970703125, 16.953369140625, 1.5961036682128906, 12.486923217773438, 4.3678741455078125, 2.5265884399414062, 14.68304443359375, -21.9349365234375, 16.403717041015625, 17.510986328125, 15.031158447265625, 12.611274719238281, -4.868194580078125, -9.327667236328125, 2.112579345703125, 27.6064453125, 37.0733642578125, 15.717376708984375, -3.087749481201172, 49.106689453125, -0.45520782470703125, 16.9215087890625, -0.7037429809570312, 11.862716674804688, 30.851943969726562, 2.646209716796875, 27.6802978515625, -7.569549560546875, -1.248138427734375, 23.780181884765625, -0.5311164855957031, 8.586883544921875, 0.824066162109375, 15.887798309326172, 13.7711181640625, 14.908721923828125, -1.69073486328125, 8.65838623046875, 16.61920166015625, 10.549209594726562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000086.npy"}
{"epoch": 0.18010471204188483, "step": 87, "batch_size": 128, "mean": 10.276609420776367, "std": 19.75945281982422, "min": -50.2193603515625, "p10": -8.916351318359375, "median": 5.3208160400390625, "p90": 37.03846435546874, "max": 67.1849365234375, "pos_frac": 0.6875, "sample": [12.302032470703125, -1.2925796508789062, -5.9947509765625, 25.218109130859375, 25.73663330078125, 0.0, 9.783443450927734, -7.9701995849609375, 18.92169189453125, 11.062164306640625, 0.32830810546875, 24.164840698242188, -18.79364013671875, 19.365066528320312, 54.654754638671875, -2.3429832458496094, 0.32149505615234375, 8.614990234375, 41.392486572265625, 7.920372009277344, 3.57818603515625, -19.72540283203125, -1.07037353515625, 23.268035888671875, 1.05572509765625, -9.401702880859375, -7.343292236328125, 16.1773681640625, 46.632537841796875, 6.351814270019531, 27.6451416015625, -26.2984619140625, 18.8094482421875, 36.08203125, 22.994903564453125, 13.525726318359375, 19.059608459472656, 0.76904296875, 24.158660888671875, 12.87750244140625, -2.093952178955078, 2.532073974609375, 23.762939453125, -2.0599365234375, -0.40105438232421875, -3.336578369140625, 51.470245361328125, -18.190216064453125, 50.55975341796875, -0.31707763671875, 13.301254272460938, 0.23615264892578125, 34.30645751953125, 27.66754150390625, 61.84686279296875, 58.53253173828125, 17.730316162109375, 20.95367431640625, 67.1849365234375, 39.2701416015625, -17.26922607421875, -1.900482177734375, 3.567537307739258, 19.894607543945312, 15.233383178710938, 49.75982666015625, 13.043716430664062, 0.0, 4.0169219970703125, 16.68865966796875, 4.057197570800781, -0.99853515625, 42.73406982421875, 24.347686767578125, 25.2274169921875, 5.643096923828125, 8.850753784179688, 0.7952423095703125, 25.541671752929688, -1.4378890991210938, -3.40081787109375, -6.7691650390625, -8.453941345214844, -11.7169189453125, -8.723564147949219, 25.754638671875, 2.808074951171875, 14.692176818847656, 10.854110717773438, 26.11529541015625, 10.133316040039062, 2.4743881225585938, 9.996391296386719, -10.21868896484375, 4.44024658203125, 4.99853515625, 19.550140380859375, 7.730485916137695, -21.47747802734375, 25.3740234375, 2.1630401611328125, 26.621627807617188, -9.366188049316406, -7.658966064453125, 26.137954711914062, -50.2193603515625, -0.09832763671875, -46.16551208496094, 2.602630615234375, 1.2735824584960938, 30.997772216796875, 41.658355712890625, 2.0377960205078125, -6.5337066650390625, -7.168670654296875, -0.2762603759765625, 41.3785400390625, 4.219371795654297, 0.5047607421875, 0.23160552978515625, 21.60870361328125, -3.592041015625, -12.363639831542969, 2.4234142303466797, 13.613571166992188, 0.4158782958984375, 34.53936767578125, -1.0309028625488281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000087.npy"}
{"epoch": 0.18219895287958116, "step": 88, "batch_size": 128, "mean": 8.950811386108398, "std": 19.623144149780273, "min": -66.58090209960938, "p10": -11.246018981933593, "median": 4.81805419921875, "p90": 38.26380310058593, "max": 64.38055419921875, "pos_frac": 0.6875, "sample": [2.0788040161132812, 31.049407958984375, 34.238525390625, -3.6572265625, 42.1910400390625, 3.1075820922851562, -24.584381103515625, 0.45063018798828125, 28.450363159179688, 10.865753173828125, 9.804817199707031, -10.425003051757812, 3.71966552734375, 64.38055419921875, -10.035636901855469, 7.32470703125, -1.2085151672363281, 7.40234375, 4.40771484375, -15.05230712890625, 52.41291809082031, -11.799774169921875, 39.3438720703125, -7.39349365234375, -4.525177001953125, 18.537948608398438, -11.231109619140625, -17.378746032714844, 52.016845703125, -20.559234619140625, -19.3226318359375, 39.122161865234375, -6.4297332763671875, 22.14422607421875, -4.395866394042969, 4.182075500488281, 37.89593505859375, 0.0, 0.9442253112792969, 9.39910888671875, 12.41790771484375, 3.1463623046875, 44.18882751464844, -6.32476806640625, 54.2039794921875, 40.479400634765625, 10.80908203125, -14.4359130859375, 3.941661834716797, 7.7154998779296875, 0.0, 12.393524169921875, 12.007232666015625, 21.3626708984375, 25.697845458984375, -11.280807495117188, 0.557891845703125, -0.68011474609375, 12.09326171875, 19.8519287109375, 29.524185180664062, 1.76190185546875, -3.1966476440429688, -23.535606384277344, 24.849044799804688, 0.56842041015625, 1.1655807495117188, 24.978607177734375, 2.753326416015625, 26.71160888671875, 12.857006072998047, -10.281219482421875, 13.98785400390625, 0.9644947052001953, 5.2283935546875, -32.311981201171875, 15.961456298828125, 4.021265029907227, -3.98748779296875, -11.729095458984375, 18.139938354492188, 27.795578002929688, 15.145751953125, -1.9607391357421875, 6.70672607421875, 1.338134765625, -3.4605331420898438, 4.057861328125, 36.193115234375, 21.132171630859375, -4.135015487670898, 11.698394775390625, 0.6395187377929688, 15.3836669921875, 21.422027587890625, 47.59886169433594, -0.88140869140625, -8.354736328125, 5.303558349609375, 41.05914306640625, 7.333740234375, 8.040939331054688, -1.164407730102539, 24.0211181640625, -8.583831787109375, 41.71820068359375, 31.49517822265625, 2.386402130126953, 17.2030029296875, -12.69915771484375, 26.638519287109375, 1.116058349609375, 5.2691650390625, 6.80908203125, 53.3682861328125, 7.967626571655273, -8.270227432250977, 4.2630157470703125, 0.5821075439453125, -9.6063232421875, -66.58090209960938, 2.879669189453125, -1.7532958984375, 20.016006469726562, 10.663360595703125, 13.4283447265625, 34.28498840332031, -5.853858947753906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000088.npy"}
{"epoch": 0.18429319371727748, "step": 89, "batch_size": 128, "mean": 9.292278289794922, "std": 19.352781295776367, "min": -48.96661376953125, "p10": -12.017070007324216, "median": 7.258411407470703, "p90": 35.75755920410155, "max": 61.787078857421875, "pos_frac": 0.765625, "sample": [1.178497314453125, -29.7474365234375, 27.277923583984375, 9.533050537109375, 5.7392578125, 40.320892333984375, 0.8026218414306641, 13.630401611328125, 6.2066802978515625, 20.79095458984375, 1.63092041015625, 9.679107666015625, 4.0273590087890625, 25.830413818359375, 6.9053955078125, 21.503402709960938, 4.1573333740234375, -22.865081787109375, 13.711585998535156, 8.07830810546875, 14.777252197265625, 22.946685791015625, 16.8074951171875, 2.3690147399902344, 22.300682067871094, -1.7678680419921875, 13.525726318359375, 1.546905517578125, 3.2912139892578125, 7.218231201171875, 0.5735015869140625, 3.733245849609375, 1.36474609375, 34.33270263671875, 34.418426513671875, 61.787078857421875, 51.333251953125, -32.181121826171875, -2.837005615234375, 11.659820556640625, -48.96661376953125, -14.500244140625, 1.984619140625, -37.7354736328125, 6.892578125, 24.47747802734375, 40.991546630859375, -25.618804931640625, -2.29583740234375, -4.288848876953125, 0.6317367553710938, 18.711334228515625, 7.1946868896484375, -3.7336273193359375, 22.333053588867188, 26.627593994140625, 29.988418579101562, -3.8069496154785156, 20.018630981445312, 31.20037841796875, 7.298591613769531, 26.109283447265625, 16.1473388671875, 15.047210693359375, 13.38665771484375, 12.078521728515625, -14.1893310546875, -7.915679931640625, 0.2382659912109375, -3.9679107666015625, 4.479286193847656, 3.596527099609375, 2.0112380981445312, 25.726409912109375, -32.86781311035156, -0.75848388671875, 10.105735778808594, 29.76605224609375, 48.477783203125, -22.43658447265625, 6.494842529296875, 14.99896240234375, 17.215736389160156, 38.8822021484375, 12.635223388671875, 51.147613525390625, 0.2122344970703125, 45.3494873046875, -5.88714599609375, 41.888946533203125, 17.920143127441406, 6.05718994140625, 45.05671691894531, 1.9942131042480469, 4.3061676025390625, 19.053497314453125, 40.61181640625, 2.7070236206054688, 7.519996643066406, 7.329170227050781, 26.1939697265625, 3.0495986938476562, -2.9532012939453125, -6.963958740234375, 4.106956481933594, 8.661491394042969, -15.88578987121582, -4.888824462890625, 11.9366455078125, 9.083740234375, 13.599945068359375, 7.376766204833984, 25.71131134033203, 39.885345458984375, -13.905166625976562, 13.816375732421875, 8.595191955566406, -11.2078857421875, 40.92622375488281, -44.370086669921875, -2.8298892974853516, 33.727325439453125, 7.0858917236328125, 0.3500823974609375, 2.171079635620117, 11.515106201171875, -4.7595367431640625, -4.14154052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000089.npy"}
{"epoch": 0.18638743455497384, "step": 90, "batch_size": 128, "mean": 8.221304893493652, "std": 18.490219116210938, "min": -46.05755615234375, "p10": -10.341761398315429, "median": 7.4660186767578125, "p90": 29.890585327148436, "max": 55.7684326171875, "pos_frac": 0.625, "sample": [14.799270629882812, 1.7705078125, -3.2652740478515625, 50.73388671875, -10.164993286132812, -2.061767578125, -8.530487060546875, 7.2836151123046875, -7.193267822265625, 23.86761474609375, 14.732551574707031, 13.652450561523438, -32.35124206542969, -3.738433837890625, -7.063812255859375, 14.45684814453125, 36.24613952636719, 40.091888427734375, 16.41357421875, 11.096343994140625, 16.824020385742188, -6.571380615234375, 8.777862548828125, -6.721805572509766, -38.846435546875, -8.79302978515625, 16.720008850097656, 22.65106201171875, 18.07098388671875, 0.254302978515625, -11.037734985351562, 12.209640502929688, 40.610565185546875, 12.063323974609375, 21.073974609375, 27.110992431640625, -8.4862060546875, -2.904052734375, -6.663421630859375, 12.413299560546875, -10.225284576416016, 5.1639251708984375, -0.7869338989257812, -41.1168212890625, 18.87506103515625, 55.7684326171875, 25.148147583007812, 10.811279296875, 26.290611267089844, -1.455657958984375, 27.042953491210938, -7.0340728759765625, -2.893310546875, -0.7956447601318359, 3.67266845703125, 13.132553100585938, -18.335357666015625, -3.8925094604492188, 6.131233215332031, 13.945449829101562, -0.634735107421875, 0.0, 25.416473388671875, 33.47837829589844, 23.087833404541016, -8.343643188476562, 16.045318603515625, 7.6484222412109375, -3.316162109375, 27.39495849609375, -3.1615676879882812, 13.292449951171875, 5.4083251953125, 44.79052734375, 18.348907470703125, 13.25262451171875, -0.854278564453125, 4.4150543212890625, 7.8345184326171875, 27.03753662109375, 18.740234375, -46.05755615234375, 24.13153076171875, -1.503997802734375, -0.6513748168945312, 3.8159637451171875, 26.764442443847656, 2.68212890625, 10.629659652709961, -7.216766357421875, 23.07745361328125, -13.6302490234375, 24.2939453125, 37.5819091796875, -8.441131591796875, 13.943084716796875, -10.613540649414062, 6.563751220703125, 15.782318115234375, 26.29730224609375, -4.4638671875, 25.41773223876953, 21.198150634765625, 10.723159790039062, -12.915969848632812, 54.739105224609375, 4.337860107421875, 23.3856201171875, -14.484786987304688, -34.20965576171875, 14.313949584960938, -5.4074249267578125, -8.06085205078125, 32.623779296875, -0.66680908203125, 2.378124237060547, 29.96746826171875, 4.630615234375, 30.2540283203125, 1.1267547607421875, -13.424293518066406, -9.374113082885742, 42.306610107421875, 22.970306396484375, 2.1538429260253906, 16.7110595703125, -20.069122314453125, 29.857635498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000090.npy"}
{"epoch": 0.18848167539267016, "step": 91, "batch_size": 128, "mean": 9.504834175109863, "std": 20.621761322021484, "min": -34.16131591796875, "p10": -13.64654388427734, "median": 6.839154243469238, "p90": 38.48640747070312, "max": 59.911865234375, "pos_frac": 0.6171875, "sample": [5.0460205078125, -7.813270568847656, -4.395606994628906, 52.470550537109375, 43.896392822265625, 54.714263916015625, 2.4777755737304688, 18.372222900390625, 14.723419189453125, 11.158195495605469, 15.991897583007812, 17.462387084960938, 28.4813232421875, 18.68627166748047, 13.653972625732422, 25.160430908203125, 16.31829071044922, -26.1475830078125, 6.929573059082031, -0.9828033447265625, 16.56000518798828, 22.9744873046875, 8.231117248535156, 12.427703857421875, -1.4275951385498047, 10.710372924804688, 8.90673828125, -4.2098388671875, -1.422994613647461, -24.010757446289062, 25.26446533203125, -3.2468948364257812, -5.7424468994140625, -31.8016357421875, -34.16131591796875, 25.5628662109375, -9.297531127929688, 17.516143798828125, 27.170738220214844, 21.09332275390625, -19.82086181640625, 1.3475914001464844, -8.587387084960938, 6.7593994140625, 17.96112060546875, 56.381591796875, -12.975357055664062, -15.626373291015625, 0.0, -26.911788940429688, -2.4150848388671875, -11.600570678710938, -25.34161376953125, 6.162017822265625, 4.9755096435546875, 7.829193115234375, -5.21728515625, -5.979454040527344, 16.273834228515625, 34.6884765625, 23.119537353515625, -31.142257690429688, 14.37872314453125, 6.714717864990234, -1.710317611694336, -9.286468505859375, 59.911865234375, -3.72735595703125, -1.1843719482421875, 31.448516845703125, 6.8310546875, -11.209442138671875, 55.352386474609375, 18.102859497070312, 44.2154541015625, 6.847253799438477, 30.092681884765625, -0.1945781707763672, 16.52129364013672, 46.5528564453125, -29.51519775390625, -7.076873779296875, 45.57049560546875, 30.89575958251953, 22.65948486328125, 0.0, 22.942283630371094, -4.694610595703125, 18.722412109375, 15.95916748046875, 2.1949691772460938, 33.661041259765625, -0.993804931640625, -22.866943359375, -7.6246337890625, 23.2216796875, -23.906532287597656, -1.8984928131103516, 5.15924072265625, 37.9775390625, 32.954681396484375, 8.559173583984375, 37.161590576171875, 4.52911376953125, 13.3660888671875, -8.540679931640625, -10.874038696289062, 22.93927764892578, -1.805206298828125, 47.678466796875, -5.04205322265625, 55.61474609375, 39.96209716796875, -1.9796218872070312, 5.225799560546875, 15.817123413085938, 5.026695251464844, -15.212646484375, 11.263702392578125, -3.6574020385742188, 35.56591796875, -6.514739990234375, 6.3937835693359375, 5.51453971862793, 9.1767578125, 7.592987060546875, 39.67376708984375, -0.99615478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000091.npy"}
{"epoch": 0.1905759162303665, "step": 92, "batch_size": 128, "mean": 10.168839454650879, "std": 21.967079162597656, "min": -53.49725341796875, "p10": -17.948135375976562, "median": 8.234336853027344, "p90": 40.29635925292968, "max": 64.56591796875, "pos_frac": 0.71875, "sample": [-21.182342529296875, 4.284965515136719, -28.358123779296875, 14.106735229492188, 15.873703002929688, 18.168991088867188, 21.672500610351562, 9.529815673828125, 33.578399658203125, 4.5623321533203125, 0.6178970336914062, 7.678001403808594, 49.5108642578125, 0.507293701171875, 48.246795654296875, 32.5567626953125, 43.661590576171875, -9.5159912109375, 20.527297973632812, 0.12828445434570312, -22.529640197753906, 26.1715087890625, 38.26768493652344, 30.750518798828125, 5.29248046875, 8.222808837890625, 14.07147216796875, 25.616851806640625, 12.87884521484375, -7.501678466796875, 3.06402587890625, 26.36407470703125, 31.445068359375, 8.970016479492188, 3.168792724609375, -15.386383056640625, 5.6527099609375, 12.626571655273438, 28.79034423828125, 12.3599853515625, -1.7834510803222656, -9.85211181640625, 0.33855628967285156, -53.49725341796875, 16.969375610351562, 7.623382568359375, 13.10943603515625, 0.70599365234375, 9.3536376953125, -6.1334075927734375, -19.1317138671875, -17.440887451171875, -40.272491455078125, 6.67828369140625, 43.843536376953125, -10.73150634765625, -24.396255493164062, -9.844829559326172, -26.5703125, 19.1915283203125, 8.245864868164062, 24.511856079101562, 11.231842041015625, 16.91876220703125, 33.72515869140625, 11.444305419921875, 29.008934020996094, 0.0, -21.329681396484375, 46.2410888671875, -22.827911376953125, 1.572296142578125, -1.417877197265625, -5.886692047119141, 9.593589782714844, 21.2200927734375, -2.4316787719726562, 20.32105255126953, 33.889373779296875, -1.9534883499145508, -16.37359619140625, 46.7069091796875, 29.042633056640625, -50.9344482421875, -9.55206298828125, 3.75823974609375, 30.74761962890625, 40.518035888671875, 47.88433837890625, 21.366546630859375, 16.534549713134766, 20.439193725585938, -6.657524108886719, -9.457046508789062, 2.260141372680664, 56.940093994140625, -7.6005859375, 1.870819091796875, 40.20135498046875, 8.007781982421875, 11.914031982421875, 33.1341552734375, -33.46661376953125, 25.122100830078125, 21.33697509765625, -6.4781646728515625, 64.56591796875, 17.749862670898438, 12.315475463867188, 1.9599609375, 50.39459228515625, 6.548728942871094, 21.18933868408203, -23.367401123046875, 27.465286254882812, 29.18231201171875, 3.6070804595947266, 4.958648681640625, -5.214447021484375, -2.081878662109375, -5.325286865234375, 7.932500839233398, 6.343803405761719, 6.904264450073242, 51.90472412109375, 42.276153564453125, 39.59504699707031, 0.752960205078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000092.npy"}
{"epoch": 0.19267015706806281, "step": 93, "batch_size": 128, "mean": 9.561741828918457, "std": 20.356063842773438, "min": -40.70062255859375, "p10": -14.959495544433594, "median": 8.231685638427734, "p90": 35.215728759765625, "max": 74.51809692382812, "pos_frac": 0.6796875, "sample": [-1.3405723571777344, 25.35015869140625, 2.6600341796875, 59.31842041015625, -15.388076782226562, 19.85424041748047, -0.7835693359375, 11.533676147460938, 11.21917724609375, 11.861982345581055, 11.34906005859375, -19.125442504882812, 14.955078125, 2.252899169921875, 29.43475341796875, 33.128173828125, 6.660736083984375, 35.796234130859375, -9.068382263183594, 12.805479049682617, 16.174514770507812, 21.752105712890625, 8.16457748413086, 14.9412841796875, 36.64512634277344, 12.403717041015625, -2.965911865234375, 14.15771484375, 25.647750854492188, 19.882217407226562, 18.487762451171875, -29.714569091796875, 6.818695068359375, -6.762786865234375, 7.960662841796875, 26.48748779296875, 13.10552978515625, 46.766632080078125, 4.739471435546875, 36.3291015625, -12.189315795898438, 26.63939666748047, 9.2564697265625, -4.839080810546875, -5.8687286376953125, 6.155574798583984, 1.92578125, 63.7037353515625, 3.169586181640625, 8.29879379272461, 15.429862976074219, 20.122528076171875, 27.411102294921875, -10.09661865234375, 18.7227783203125, 14.784423828125, -1.648040771484375, -10.441978454589844, 40.6510009765625, 2.948711395263672, 19.98046875, -25.5274658203125, 35.3646240234375, -2.4029693603515625, 8.682296752929688, -28.82232666015625, 35.15191650390625, 27.57635498046875, -20.034439086914062, 8.102798461914062, 13.704141616821289, 4.458854675292969, 1.897216796875, 6.5946044921875, 18.85382080078125, -21.765777587890625, 34.893280029296875, -8.521240234375, 33.08197021484375, -40.70062255859375, 50.77899169921875, 18.18682861328125, 24.267486572265625, -21.748046875, 12.748321533203125, 7.389457702636719, 0.0, -12.104522705078125, -0.8515205383300781, 20.646278381347656, 41.72784423828125, 31.02313232421875, -13.559326171875, -14.77581787109375, 22.529624938964844, -34.21697998046875, 11.9251708984375, -1.6867847442626953, -6.12506103515625, -6.484134674072266, 17.862106323242188, 19.091827392578125, -10.51416015625, 0.642303466796875, 17.673004150390625, 0.0, 3.4693832397460938, -7.5680389404296875, -22.252182006835938, -9.29681396484375, -2.76007080078125, 1.5633125305175781, 21.240692138671875, 6.515327453613281, -30.586334228515625, 74.51809692382812, 49.496337890625, 4.9313507080078125, -16.255882263183594, -9.62567138671875, 10.724365234375, 21.57958984375, 4.8824462890625, 55.31732177734375, -0.5123100280761719, 21.046966552734375, 4.4907684326171875, 20.359619140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000093.npy"}
{"epoch": 0.19476439790575917, "step": 94, "batch_size": 128, "mean": 9.423165321350098, "std": 22.725751876831055, "min": -73.22357177734375, "p10": -22.238592529296874, "median": 8.39956283569336, "p90": 40.04353332519531, "max": 61.494873046875, "pos_frac": 0.6953125, "sample": [-6.254913330078125, 33.95185852050781, -21.81622314453125, 35.740478515625, 36.199798583984375, 45.3531494140625, 52.28773498535156, 5.964776992797852, 9.178619384765625, 21.163040161132812, -3.14471435546875, -17.566848754882812, -40.093292236328125, -2.3918685913085938, 34.766265869140625, -8.612548828125, 39.79840087890625, 7.676490783691406, 9.122634887695312, 16.951065063476562, 31.725372314453125, 15.992019653320312, 7.645599365234375, -5.99285888671875, 54.93959045410156, 24.048828125, 2.8970108032226562, -1.2676849365234375, -11.567138671875, 3.90545654296875, 44.006195068359375, 41.136962890625, -34.16070556640625, 49.57464599609375, 19.61156463623047, 10.955230712890625, 27.1112060546875, 7.516244888305664, -23.4703369140625, 58.1201171875, 13.071327209472656, 44.7255859375, 19.442138671875, -10.402763366699219, 42.62481689453125, 14.512557983398438, 21.431793212890625, -31.783447265625, 15.071243286132812, -6.060192108154297, -3.565216064453125, 10.625228881835938, 6.188920974731445, 19.759918212890625, 12.030410766601562, -7.1469879150390625, 32.09742736816406, -24.4754638671875, 10.89471435546875, -11.40570068359375, -6.31781005859375, 3.836517333984375, 0.40177154541015625, -8.450668334960938, 4.68109130859375, 14.30767822265625, 20.83636474609375, 21.505752563476562, -1.9083480834960938, 29.001327514648438, 1.91693115234375, 2.711963653564453, 2.94989013671875, 20.425140380859375, 38.0863037109375, -31.043701171875, 9.45263671875, -4.855316162109375, 13.848876953125, 1.054443359375, 6.58966064453125, 27.109451293945312, -10.782478332519531, -3.837066650390625, -26.079238891601562, 40.615509033203125, 41.762237548828125, -24.8056640625, 26.733062744140625, -27.3536376953125, -73.22357177734375, 32.43791198730469, 5.300975799560547, 21.899389266967773, 16.787094116210938, 20.745193481445312, 27.43951416015625, 10.56756591796875, 50.662078857421875, -38.59654235839844, -0.9332389831542969, -18.80322265625, 19.528594970703125, 5.579669952392578, 12.145736694335938, 23.899627685546875, 5.303260803222656, 16.795387268066406, 32.56915283203125, 7.129119873046875, 1.9599456787109375, 4.41357421875, 11.218536376953125, -26.028564453125, -2.1180419921875, 2.16986083984375, 21.266693115234375, 4.0560760498046875, -7.649810791015625, -17.688323974609375, 34.181640625, 9.4056396484375, 61.494873046875, 30.730995178222656, -23.22412109375, 4.2564697265625, -1.3019981384277344, 0.7578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000094.npy"}
{"epoch": 0.1968586387434555, "step": 95, "batch_size": 128, "mean": 10.187210083007812, "std": 24.099838256835938, "min": -71.49609375, "p10": -13.18660736083984, "median": 8.577560424804688, "p90": 40.82756042480468, "max": 74.32818603515625, "pos_frac": 0.6953125, "sample": [-7.572961807250977, -3.661468505859375, -1.70526123046875, 69.55520629882812, 14.07275390625, -1.38995361328125, 14.734909057617188, 8.1634521484375, 43.026702880859375, 19.62896728515625, 3.923583984375, 1.1566925048828125, 3.8459014892578125, 6.549003601074219, -30.167022705078125, 17.603240966796875, 9.360321044921875, 9.244804382324219, 20.218048095703125, 8.545013427734375, 27.277023315429688, -0.2871246337890625, -33.54736328125, 14.537353515625, 5.29705810546875, 8.63580322265625, 69.47723388671875, -15.2322998046875, 48.9268798828125, -56.232391357421875, 51.59466552734375, -12.1602783203125, 20.710067749023438, -35.67054748535156, 21.420654296875, 30.45050048828125, -4.431365966796875, 2.048370361328125, 32.97886657714844, 14.9320068359375, 20.090621948242188, 17.82232666015625, -1.818359375, 9.082117080688477, 2.3342132568359375, 17.789291381835938, 14.142322540283203, 28.486160278320312, 43.77490234375, 8.072891235351562, 9.132492065429688, 31.526756286621094, 23.324264526367188, 0.8256359100341797, 20.572509765625, -11.82177734375, 12.233787536621094, 23.164955139160156, 30.46112060546875, 19.06494140625, 22.782958984375, -0.4961700439453125, -1.154052734375, -3.829559326171875, 17.368186950683594, 71.20706176757812, -8.376296997070312, 24.553009033203125, 8.610107421875, 7.336997985839844, -11.277359008789062, 27.37481689453125, 19.785675048828125, 43.72760009765625, -9.264801025390625, -4.506103515625, 2.583688735961914, 39.88507080078125, 1.6477508544921875, 14.410079956054688, -4.93017578125, 24.18679428100586, 16.7481689453125, 6.128684997558594, 22.496063232421875, 49.328857421875, 37.31645965576172, 6.0032806396484375, -8.30352783203125, -27.17584228515625, 15.896537780761719, 22.489425659179688, -7.8729095458984375, 19.760009765625, -40.752685546875, 5.65911865234375, 26.44342041015625, 19.71966552734375, -30.07373046875, 17.037811279296875, -7.229217529296875, -2.7806644439697266, -5.553924560546875, 3.217395782470703, 4.531578063964844, -25.0167236328125, 1.0610198974609375, -9.014289855957031, 70.3687744140625, -23.8074951171875, -15.013137817382812, 3.03729248046875, -3.8462066650390625, 22.81683349609375, 5.611724853515625, 31.1273193359375, -3.4143524169921875, 20.217681884765625, 14.01654052734375, -34.23455810546875, -12.40380859375, 0.6204109191894531, 7.432403564453125, 74.32818603515625, 4.732147216796875, 57.49322509765625, -71.49609375, 50.568603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000095.npy"}
{"epoch": 0.19895287958115182, "step": 96, "batch_size": 128, "mean": 10.200647354125977, "std": 24.265975952148438, "min": -72.108642578125, "p10": -21.389726257324217, "median": 8.665267944335938, "p90": 40.20163269042968, "max": 80.72119140625, "pos_frac": 0.671875, "sample": [-8.85894775390625, -39.737548828125, -2.0977783203125, 19.888431549072266, 41.8330078125, 26.222549438476562, -23.776676177978516, -4.44171142578125, 26.30828857421875, 5.33648681640625, 10.808456420898438, 0.6584320068359375, -13.09002685546875, 4.10113525390625, 18.62866973876953, 7.496501922607422, 22.99713134765625, -22.035369873046875, 19.33135986328125, 8.738006591796875, 1.636627197265625, 5.295440673828125, 42.904815673828125, 28.493026733398438, -8.01080322265625, 13.604827880859375, 64.8636474609375, 23.958953857421875, 2.3621139526367188, 16.619720458984375, 80.72119140625, 20.840057373046875, 39.502471923828125, 30.709835052490234, 26.738189697265625, 7.370262145996094, -1.6247024536132812, 1.118072509765625, -13.680572509765625, -1.2552947998046875, 25.674957275390625, 21.148971557617188, -28.381439208984375, 47.34002685546875, -5.9801025390625, 8.869857788085938, -8.645416259765625, 5.05937385559082, 22.574729919433594, -4.244537353515625, 2.6650543212890625, -5.074554443359375, 25.487945556640625, -30.0810546875, 8.592529296875, -36.93922424316406, 15.876800537109375, 21.620758056640625, 6.918304443359375, 28.3089599609375, -11.176563262939453, -13.630615234375, 22.881988525390625, 20.255645751953125, 4.960723876953125, 55.21331787109375, 55.65753173828125, 14.732666015625, 3.4247817993164062, 45.8448486328125, 35.34783935546875, 16.999168395996094, 20.15174102783203, -25.082839965820312, 74.28964233398438, 9.533447265625, 19.088546752929688, -3.387359619140625, 65.087646484375, -10.00787353515625, 7.0550994873046875, -1.7126312255859375, 2.9230575561523438, -12.708282470703125, -16.401809692382812, -22.952865600585938, -2.62493896484375, 26.814834594726562, 38.355865478515625, -10.811080932617188, -1.8127117156982422, -3.2120513916015625, 6.045942306518555, 30.47479248046875, 48.34027099609375, -23.936668395996094, 15.102630615234375, 24.358749389648438, 10.344886779785156, 29.919677734375, -44.78948974609375, 32.060760498046875, 10.883392333984375, 1.9225883483886719, -7.6898193359375, 18.756134033203125, -11.2840576171875, 56.55291748046875, 10.491157531738281, 19.66668701171875, 49.553924560546875, 0.0, -21.113021850585938, 6.405330657958984, 4.142461776733398, 22.984577178955078, -29.894927978515625, 16.929443359375, 20.6121826171875, 34.8157958984375, -25.30799102783203, 28.313472747802734, -4.3760833740234375, 20.813232421875, -1.236328125, 17.262603759765625, 6.299224853515625, -72.108642578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000096.npy"}
{"epoch": 0.20104712041884817, "step": 97, "batch_size": 128, "mean": 12.005033493041992, "std": 23.192520141601562, "min": -50.442535400390625, "p10": -18.177217864990233, "median": 10.62900161743164, "p90": 42.912640380859365, "max": 75.4185791015625, "pos_frac": 0.734375, "sample": [13.442222595214844, 22.95556640625, 17.693038940429688, -13.42401123046875, 8.76742172241211, 8.730213165283203, -11.354637145996094, 8.70404052734375, 21.64593505859375, 34.11305236816406, -0.6533145904541016, 46.2491455078125, 54.549957275390625, 50.29280090332031, 1.707681655883789, 4.8018035888671875, 37.90655517578125, 2.76397705078125, -2.516357421875, 22.024276733398438, 13.13360595703125, 0.6422348022460938, -15.848907470703125, 2.056793212890625, 32.3336181640625, -30.765594482421875, -1.781005859375, -11.528205871582031, 5.386135101318359, 22.689414978027344, -28.478172302246094, 19.857620239257812, -10.361068725585938, -34.531494140625, 23.111984252929688, 58.74012756347656, 41.077110290527344, -5.78387451171875, -23.693984985351562, 8.724395751953125, 2.155548095703125, 9.6558837890625, 9.413284301757812, 31.14263916015625, -26.44921875, 26.74542236328125, 16.844757080078125, 32.7042236328125, 13.1689453125, 6.5256500244140625, 27.075088500976562, 3.59918212890625, 18.08599853515625, 5.1622314453125, 22.043487548828125, 45.347900390625, -13.426651000976562, 23.205718994140625, 9.31951904296875, -1.040283203125, 29.28929901123047, 17.267074584960938, -16.62701416015625, 18.049699783325195, 25.647628784179688, 0.0, 51.1363525390625, 25.645828247070312, 11.994293212890625, 41.03448486328125, -15.708866119384766, 54.154022216796875, 29.808303833007812, 7.0158538818359375, 0.3377399444580078, 53.16143798828125, 17.8551025390625, -42.056610107421875, 23.53580093383789, 35.59014892578125, 12.45361328125, 6.3493499755859375, 9.881591796875, 11.185844421386719, 46.575042724609375, 11.426006317138672, 9.926084518432617, 75.4185791015625, 26.462783813476562, -9.285980224609375, 18.005878448486328, -27.92498779296875, 17.92755126953125, 4.8331298828125, 19.488006591796875, -21.314254760742188, 0.0, 12.19049072265625, 18.252838134765625, 0.2968902587890625, 8.657928466796875, 66.72097778320312, 10.072158813476562, 4.5308685302734375, 2.3048553466796875, 51.80586242675781, -19.795150756835938, -7.844573974609375, 64.2796630859375, 19.228271484375, -4.31158447265625, 27.44549560546875, -17.48381805419922, -50.442535400390625, 14.385009765625, 23.65142822265625, 16.65985107421875, 41.86895751953125, -7.436637878417969, -25.036819458007812, 7.778656005859375, -23.99957275390625, 2.35223388671875, 26.0704345703125, 33.1849365234375, -24.501220703125, -0.92669677734375, 33.486846923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000097.npy"}
{"epoch": 0.2031413612565445, "step": 98, "batch_size": 128, "mean": 12.31692886352539, "std": 25.665485382080078, "min": -66.30841064453125, "p10": -14.019522094726561, "median": 9.069562911987305, "p90": 49.29403076171875, "max": 95.302734375, "pos_frac": 0.6875, "sample": [30.900188446044922, 58.73199462890625, 49.7996826171875, 13.385965347290039, 74.15695190429688, 33.526611328125, 37.2288818359375, -34.34649658203125, 13.71722412109375, -5.8899688720703125, 59.62237548828125, -34.12437438964844, -5.575653076171875, 0.344390869140625, 28.65472412109375, 24.13079833984375, 9.690017700195312, -13.12786865234375, -17.412857055664062, -3.522552490234375, 5.58148193359375, -6.040313720703125, -10.56463623046875, 15.94805908203125, -1.715118408203125, -8.35597038269043, 9.704025268554688, -8.173141479492188, -13.5816650390625, -44.97509765625, 3.578794479370117, -66.30841064453125, -19.4891357421875, -6.40185546875, 7.1834869384765625, -11.427627563476562, -11.735528945922852, 58.512481689453125, 30.643798828125, 10.804862976074219, 6.2303619384765625, 11.030609130859375, -19.1695556640625, 17.13653564453125, -3.9465484619140625, 49.767333984375, 9.549728393554688, 28.789794921875, 1.761138916015625, 7.736766815185547, 49.0911865234375, 95.302734375, -7.743011474609375, -1.42095947265625, 1.9292488098144531, -4.87371826171875, 63.62066650390625, 21.9967041015625, -14.3875732421875, 49.80438232421875, -42.84757995605469, -6.33917236328125, -14.6026611328125, -6.524139404296875, 58.22454833984375, 0.05023193359375, 37.34967041015625, 3.9530029296875, 22.432741165161133, -4.7990570068359375, 32.99897766113281, 33.131103515625, -17.086639404296875, -4.8306732177734375, 44.326385498046875, 10.9935302734375, 22.957489013671875, 16.19036865234375, 11.78204345703125, 33.962310791015625, 8.530052185058594, 31.644989013671875, -13.861785888671875, 8.63564682006836, 35.614898681640625, 13.840675354003906, 0.12099456787109375, 3.1868553161621094, 29.6065673828125, 21.99853515625, 51.483795166015625, 33.521209716796875, 25.572654724121094, -7.7631988525390625, 27.599266052246094, 2.89166259765625, 6.723072052001953, 25.13550567626953, 0.480438232421875, 39.138427734375, 9.50347900390625, 31.213760375976562, 68.77716064453125, 18.048919677734375, 8.499725341796875, -12.614814758300781, -16.716064453125, 21.140182495117188, 10.239410400390625, 24.025543212890625, 3.462005615234375, -2.280181884765625, 7.7794036865234375, 6.989513397216797, 0.059173583984375, 22.339447021484375, 3.0599365234375, -5.598114013671875, 19.581035614013672, 14.496063232421875, 28.89227294921875, -3.6290740966796875, 77.382080078125, 34.29339599609375, 27.9716796875, 16.449356079101562, -32.33148193359375, 4.82598876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000098.npy"}
{"epoch": 0.20523560209424083, "step": 99, "batch_size": 128, "mean": 12.683259963989258, "std": 28.451452255249023, "min": -55.9681396484375, "p10": -22.98889465332031, "median": 11.648849487304688, "p90": 50.0251220703125, "max": 84.51177978515625, "pos_frac": 0.6875, "sample": [-50.733734130859375, -2.9843521118164062, -39.7763671875, 56.805938720703125, -2.2430419921875, 8.400238037109375, 17.77062225341797, 66.291015625, -1.518310546875, -6.362152099609375, 34.31390380859375, -16.058746337890625, -17.11541748046875, 16.699447631835938, -8.190460205078125, 20.75677490234375, 55.77593994140625, 39.24507141113281, 46.291473388671875, 11.332199096679688, 23.551036834716797, 14.397918701171875, 23.202880859375, -55.9681396484375, 67.82574462890625, 37.63714599609375, 8.396249771118164, -11.98974609375, 13.634017944335938, 29.01715087890625, 45.87005615234375, -3.59814453125, 61.2369384765625, -1.1856346130371094, -11.742462158203125, 7.358642578125, -4.52960205078125, -5.81854248046875, 32.009857177734375, 4.885272979736328, 40.95570373535156, 45.06103515625, 26.547752380371094, 0.96331787109375, 57.154937744140625, -22.121551513671875, 21.630054473876953, -0.1905956268310547, 10.004898071289062, 17.6787109375, -31.98126983642578, -51.451904296875, -51.312591552734375, 28.237701416015625, 14.251678466796875, -6.78955078125, 28.654735565185547, -18.076095581054688, 39.88421630859375, 22.63604736328125, 15.798797607421875, -12.536201477050781, 49.562957763671875, 0.7817306518554688, -14.840972900390625, 24.32928466796875, -43.631591796875, 16.7486572265625, 1.2961807250976562, 49.979949951171875, 3.1563720703125, 29.305511474609375, 14.748886108398438, -3.801675796508789, 6.25360107421875, 37.79632568359375, 26.0863037109375, 66.30667114257812, 18.498260498046875, -8.416522979736328, 22.23785400390625, 15.766555786132812, -10.74951171875, 4.760734558105469, 32.85939025878906, 7.2540435791015625, 50.130523681640625, -19.86395263671875, 28.230911254882812, 11.96392822265625, -39.183685302734375, 6.057258605957031, -25.0126953125, 18.436843872070312, 36.154937744140625, 11.333770751953125, 19.0732421875, -26.11114501953125, 4.433469772338867, -26.27490234375, 44.324859619140625, 2.9202804565429688, 0.0, 5.59967041015625, -7.89105224609375, 46.82765197753906, 1.5959014892578125, -28.000755310058594, 58.1451416015625, 43.70147705078125, 65.00947570800781, 20.724929809570312, -41.476898193359375, 43.24969482421875, 8.470197677612305, 13.4320068359375, 15.22381591796875, 84.51177978515625, -11.039459228515625, 1.4758453369140625, 9.32891845703125, -20.12701416015625, 54.764617919921875, 10.677467346191406, 38.74078369140625, 21.582382202148438, 64.19395446777344, 3.943511962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000099.npy"}
{"epoch": 0.20732984293193718, "step": 100, "batch_size": 128, "mean": 12.510345458984375, "std": 25.535541534423828, "min": -43.67205810546875, "p10": -19.24569854736328, "median": 10.48331069946289, "p90": 47.98862915039061, "max": 79.95306396484375, "pos_frac": 0.6875, "sample": [15.284393310546875, 16.840118408203125, 30.650360107421875, -0.050750732421875, 35.16015625, 3.70001220703125, -36.829315185546875, 45.639678955078125, -5.31304931640625, 12.899322509765625, -4.724853515625, 10.40869140625, 2.17108154296875, 33.2508544921875, 23.831878662109375, 67.037353515625, -5.640928268432617, -10.536834716796875, -26.326385498046875, 57.478607177734375, 10.81982421875, -2.016204833984375, 11.55474853515625, 3.7035751342773438, -8.681503295898438, 6.95025634765625, 22.065505981445312, 12.005241394042969, -12.939231872558594, 58.174591064453125, -25.12689208984375, -11.89569091796875, -1.8357696533203125, 17.56365203857422, -9.49542236328125, 11.706527709960938, 14.169464111328125, 46.3033447265625, 2.5336227416992188, 39.123992919921875, 0.91455078125, 15.486740112304688, 7.13494873046875, 23.16009521484375, -2.216276168823242, 20.8609619140625, 29.499507904052734, 55.26983642578125, 51.92095947265625, 30.948471069335938, -20.73657989501953, -13.0081787109375, -18.28571319580078, -5.031301498413086, 10.314743041992188, 38.25511169433594, -27.760498046875, -2.53363037109375, -43.67205810546875, -2.3179931640625, 3.5746994018554688, 56.415008544921875, -13.69464111328125, 27.27503204345703, 38.386810302734375, 5.0380706787109375, -15.44281005859375, -27.04449462890625, 7.981201171875, 16.39483642578125, 25.5528564453125, 16.1177978515625, 10.557929992675781, 6.114530563354492, 4.475311279296875, -16.97265625, -25.392608642578125, -1.5728759765625, 28.402923583984375, 16.628231048583984, -35.26860046386719, 22.349746704101562, 52.770782470703125, -31.377593994140625, 17.79827880859375, 9.088973999023438, 5.892814636230469, -3.59381103515625, 39.259490966796875, 79.95306396484375, 25.31793212890625, 3.1382598876953125, -12.775421142578125, 7.302001953125, 68.145751953125, 16.890968322753906, -3.74798583984375, 14.043815612792969, 2.144224166870117, 23.922515869140625, 72.14358520507812, 3.397308349609375, 63.15411376953125, 39.935791015625, 33.49192810058594, -20.712295532226562, 38.67304992675781, 66.23260498046875, 8.495637893676758, 21.185028076171875, 62.927581787109375, 6.95489501953125, 17.264068603515625, -18.617156982421875, 14.937469482421875, 15.335235595703125, 0.3203449249267578, 16.711822509765625, 38.812530517578125, -26.1185302734375, 2.575450897216797, 40.000946044921875, 34.7674560546875, -15.864822387695312, 29.793472290039062, -4.51153564453125, 31.2113037109375, -33.04112243652344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000100.npy"}
{"epoch": 0.2094240837696335, "step": 101, "batch_size": 128, "mean": 14.883186340332031, "std": 27.56473731994629, "min": -53.44940185546875, "p10": -22.53305130004883, "median": 11.597613334655762, "p90": 50.14897766113281, "max": 99.1500244140625, "pos_frac": 0.734375, "sample": [40.828765869140625, -53.44940185546875, 9.162506103515625, 25.15478515625, 28.027328491210938, -25.326751708984375, 32.21169662475586, 2.3756103515625, -13.4638671875, 29.750030517578125, 43.156707763671875, 41.487548828125, 13.54522705078125, 57.55328369140625, 99.1500244140625, 61.763397216796875, 6.182262420654297, 8.285179138183594, -2.190305709838867, 36.995574951171875, 54.222442626953125, 29.933441162109375, 58.586639404296875, 29.85260009765625, 4.868646621704102, 38.8306884765625, 16.972335815429688, 17.7855224609375, -0.901123046875, 63.90380859375, -2.2379837036132812, 14.533447265625, 7.635009765625, 2.3575592041015625, 35.82489013671875, 12.622650146484375, 61.78401184082031, 27.11724853515625, 34.748329162597656, 1.537445068359375, -17.5477294921875, -1.945892333984375, 29.4215087890625, 47.57318115234375, 41.658050537109375, 35.36775207519531, 6.824676513671875, 35.2303466796875, 1.2883682250976562, -23.417083740234375, 0.3495025634765625, 37.926239013671875, -36.22430419921875, 24.360931396484375, 57.486724853515625, 43.77178955078125, 63.891998291015625, -40.760101318359375, -4.754987716674805, 9.525627136230469, -23.071739196777344, 4.01483154296875, 12.2747802734375, 32.079811096191406, -10.243438720703125, -8.836191177368164, 12.94287109375, 40.07667541503906, 5.342109680175781, 55.97785949707031, 11.036712646484375, 12.158514022827148, 8.79339599609375, 6.9512176513671875, 9.304880142211914, 3.355684280395508, 50.007354736328125, -1.6502532958984375, 3.0940933227539062, -46.4632568359375, -45.811187744140625, 31.8062744140625, -18.029451370239258, 32.553466796875, -2.3327178955078125, 0.4010772705078125, 27.264312744140625, 31.440338134765625, 5.724639892578125, 6.5079803466796875, 37.711936950683594, -2.2085838317871094, 8.81298828125, 31.80914306640625, 14.227691650390625, 15.789031982421875, 5.916656494140625, 58.9869384765625, 32.569915771484375, 22.983901977539062, -31.9970703125, -43.10333251953125, 25.50384521484375, 7.6236419677734375, 30.053741455078125, -11.11798095703125, 7.0447998046875, 20.271942138671875, 95.994384765625, 1.891448974609375, 23.805755615234375, -2.6651077270507812, 50.47943115234375, -2.87969970703125, 3.8852005004882812, 46.277244567871094, 8.457305908203125, -14.954330444335938, -23.94140625, 19.042266845703125, -27.447998046875, -1.6654052734375, -22.30218505859375, -0.8507328033447266, -26.870079040527344, -1.7658557891845703, 23.785995483398438, 18.018020629882812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000101.npy"}
{"epoch": 0.21151832460732983, "step": 102, "batch_size": 128, "mean": 14.828296661376953, "std": 26.35175323486328, "min": -79.08340454101562, "p10": -17.394499206542967, "median": 14.872581481933594, "p90": 44.806062316894526, "max": 66.70709228515625, "pos_frac": 0.71875, "sample": [-0.3097076416015625, -16.24407958984375, 43.221588134765625, 6.083770751953125, 17.317367553710938, 55.24700927734375, 43.032745361328125, -12.705575942993164, 12.966888427734375, 50.657470703125, 18.067642211914062, 14.185760498046875, 43.877227783203125, 8.43988037109375, 1.390106201171875, 35.9981689453125, 37.059722900390625, -16.38311767578125, 26.405288696289062, 13.688766479492188, -17.48126220703125, 65.00309753417969, 28.246063232421875, 55.894927978515625, 35.62969970703125, -7.5234375, 6.284912109375, -22.281494140625, 7.55902099609375, -4.2604827880859375, -20.80548095703125, 46.41749572753906, 10.577075958251953, 42.3486328125, -13.391754150390625, 43.3221435546875, 19.7073974609375, -3.1693954467773438, 8.939353942871094, -3.335296630859375, -27.05517578125, -3.6552505493164062, 3.009899139404297, 13.217445373535156, 9.344024658203125, 12.595695495605469, 22.693458557128906, -23.979156494140625, -38.9149169921875, -64.31166076660156, 35.47705078125, -30.701507568359375, 20.154495239257812, 8.488677978515625, 26.174407958984375, 25.8084716796875, 27.40667724609375, 63.04461669921875, 19.080657958984375, 32.506439208984375, 14.645126342773438, -15.108856201171875, 35.818084716796875, 39.44549560546875, 15.10003662109375, -17.357315063476562, -27.230926513671875, 7.1329345703125, -14.098499298095703, 0.2962379455566406, 38.52751922607422, 31.68170166015625, 32.952293395996094, 42.93695068359375, 16.713905334472656, 42.595184326171875, 19.72614288330078, 32.40419006347656, 60.15447998046875, -35.60688781738281, -5.14171028137207, 38.869384765625, 0.0, 4.3811798095703125, -2.4911651611328125, 53.177398681640625, 0.0, 23.913833618164062, 16.2235107421875, 1.125223159790039, 20.744674682617188, 25.254608154296875, -3.543975830078125, 0.8544921875, 61.25634765625, -2.476104736328125, 16.11541748046875, 2.5937576293945312, 13.743621826171875, 51.98008728027344, 38.853546142578125, 14.030632019042969, 22.81268310546875, 55.03271484375, 20.302947998046875, 27.397369384765625, -0.7095832824707031, 33.17927169799805, -4.2088470458984375, 29.217559814453125, -79.08340454101562, 40.42645263671875, 13.428924560546875, 5.874536514282227, 29.117233276367188, -26.20501708984375, 2.703857421875, 1.3176727294921875, 66.70709228515625, -1.0707855224609375, 52.851806640625, 44.115447998046875, 26.917327880859375, 39.7852783203125, 16.047317504882812, 43.85931396484375, -6.41314697265625, -35.63713073730469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000102.npy"}
{"epoch": 0.2136125654450262, "step": 103, "batch_size": 128, "mean": 13.64555835723877, "std": 26.974367141723633, "min": -53.036865234375, "p10": -13.237801361083983, "median": 7.760765075683594, "p90": 53.57145690917969, "max": 90.227783203125, "pos_frac": 0.7265625, "sample": [-34.80877685546875, -4.0471649169921875, 74.21142578125, 30.49609375, 56.27105712890625, 11.214601516723633, -12.322128295898438, 2.1182479858398438, 86.7186279296875, 7.24652099609375, -1.9136962890625, 40.93743896484375, 5.3004913330078125, 6.966400146484375, 0.0, 12.308380126953125, -20.669647216796875, -15.435317993164062, -14.133880615234375, 14.96087646484375, 7.415933609008789, -0.7998199462890625, 9.516860961914062, -40.85430908203125, 4.291023254394531, 16.37823486328125, 12.853073120117188, 14.197479248046875, 19.726516723632812, 15.064788818359375, 6.284576416015625, 7.765655517578125, 3.143678665161133, -11.530731201171875, 9.1292724609375, 0.5224170684814453, -0.39989471435546875, 55.721649169921875, 45.703216552734375, 25.846416473388672, 0.540069580078125, 62.78831481933594, -18.39862060546875, 81.52569580078125, 1.9897327423095703, 63.78706359863281, 20.026107788085938, 22.561187744140625, 34.44097900390625, 74.61495971679688, 54.20452880859375, 16.95098876953125, 31.8602294921875, -29.320068359375, 9.765594482421875, 4.76239013671875, -0.5774459838867188, 25.2452392578125, 30.1607666015625, 28.86669921875, 90.227783203125, 9.59112548828125, 8.720512390136719, 57.1246337890625, 5.502067565917969, 53.192047119140625, 13.3604736328125, 51.24920654296875, 22.205276489257812, 22.78546142578125, 7.086662292480469, 9.254867553710938, 1.59185791015625, -10.90234375, -8.56304931640625, -1.0194931030273438, -17.132102966308594, -53.036865234375, 4.2721405029296875, -9.474441528320312, 38.1031494140625, 41.566802978515625, 2.246631622314453, 39.40727233886719, -2.1084251403808594, 7.367584228515625, 14.554573059082031, -6.508460998535156, 53.300140380859375, -10.90130615234375, 7.7558746337890625, 9.17266845703125, 26.4189453125, 4.6632232666015625, 1.0421867370605469, 1.4768142700195312, 88.31710815429688, 12.354034423828125, 0.486083984375, -33.8646240234375, 0.0, 7.9109954833984375, 7.016460418701172, 38.713226318359375, 2.945648193359375, 58.3226318359375, -19.44598388671875, 5.724700927734375, -12.853767395019531, 33.241539001464844, 35.25547790527344, 13.439895629882812, 15.088493347167969, 40.733245849609375, -6.4167022705078125, 1.3036212921142578, 27.842544555664062, 18.24896240234375, 5.490234375, 38.779754638671875, -0.578857421875, 11.83203125, -3.854583740234375, -10.81378173828125, -9.646835327148438, -49.68519592285156, -20.199005126953125, 6.1606292724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000103.npy"}
{"epoch": 0.2157068062827225, "step": 104, "batch_size": 128, "mean": 10.099353790283203, "std": 27.13100242614746, "min": -58.484619140625, "p10": -22.322125244140626, "median": 8.266579627990723, "p90": 46.15570373535156, "max": 94.04391479492188, "pos_frac": 0.65625, "sample": [8.204071044921875, -4.68951416015625, 11.569564819335938, 52.0513916015625, -54.555694580078125, 13.202781677246094, 13.025360107421875, 9.741256713867188, 47.68524169921875, -1.523651123046875, 55.09149169921875, -15.316780090332031, 13.115386962890625, 1.24212646484375, 35.466217041015625, 20.519363403320312, 74.18243408203125, 51.981475830078125, 9.26104736328125, -10.607368469238281, -1.4811248779296875, 1.2197265625, 13.116073608398438, -49.789093017578125, 18.346466064453125, -20.780288696289062, 4.05682373046875, -1.9566383361816406, 11.236700057983398, 46.5316162109375, 15.497093200683594, 26.619354248046875, -22.998504638671875, 1.9787368774414062, 16.123687744140625, -39.3486328125, 5.87677001953125, -2.3451461791992188, 56.398193359375, 5.277496337890625, 31.3997802734375, 13.873809814453125, -11.399833679199219, -11.0936279296875, -18.75689697265625, 22.769805908203125, -13.48516845703125, 32.279876708984375, -0.1047210693359375, -10.399200439453125, -33.832977294921875, 5.857391357421875, 27.471221923828125, -13.07098388671875, 57.44775390625, 12.686084747314453, 37.55047607421875, -12.71417236328125, -13.008819580078125, 0.20817947387695312, -18.308578491210938, 31.728408813476562, 3.9661865234375, 9.116012573242188, 3.625762939453125, 21.97235107421875, -3.1741943359375, 25.832305908203125, 14.782958984375, -37.509979248046875, 34.148223876953125, 31.2529296875, 16.749183654785156, -58.484619140625, 76.12030029296875, -3.9198760986328125, 40.80584716796875, 43.257530212402344, 40.49092102050781, 50.44232177734375, 10.68304443359375, -45.118438720703125, -13.914093017578125, 31.860214233398438, 33.60260009765625, -11.229423522949219, 3.8475379943847656, 32.486785888671875, -4.61175537109375, 35.772308349609375, 94.04391479492188, -3.577606201171875, 22.04559326171875, 20.78369140625, -2.8778610229492188, 6.192451477050781, -45.15234375, 2.8990707397460938, 42.49957275390625, 24.752288818359375, 45.994598388671875, 47.347137451171875, -11.070793151855469, 37.13531494140625, -22.0390625, -9.47747802734375, -28.16424560546875, 3.3052902221679688, -7.6491546630859375, -29.6568603515625, 0.0, 27.879730224609375, 8.32908821105957, 18.411117553710938, 6.566009521484375, 16.584075927734375, 28.638671875, -29.384979248046875, 0.0005950927734375, 6.5687255859375, -22.98260498046875, 33.23004150390625, 1.0788135528564453, -2.41265869140625, 57.12774658203125, 6.78515625, 21.748886108398438, 14.039222717285156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000104.npy"}
{"epoch": 0.21780104712041884, "step": 105, "batch_size": 128, "mean": 20.59231948852539, "std": 28.82282066345215, "min": -74.6837158203125, "p10": -12.809596252441402, "median": 19.004554748535156, "p90": 52.94543457031249, "max": 105.17974853515625, "pos_frac": 0.78125, "sample": [12.13189697265625, 18.914276123046875, 49.96775817871094, 8.433551788330078, -5.99609375, 22.238525390625, 46.436553955078125, 41.550994873046875, 26.278671264648438, -5.87567138671875, 32.540679931640625, 9.111083984375, 50.15399169921875, 47.37811279296875, -0.096466064453125, 64.82815551757812, 20.458694458007812, -11.841796875, 45.3712158203125, 51.8885498046875, 1.8961181640625, 7.466064453125, 14.0455322265625, 12.749870300292969, 55.938873291015625, 39.4788818359375, -49.063629150390625, 4.556526184082031, 2.7955322265625, 63.00341796875, 38.18955993652344, 12.90045166015625, 1.3847503662109375, 57.3306884765625, 7.7322235107421875, 38.451751708984375, 47.5565185546875, -7.633056640625, 48.2945556640625, -28.21734619140625, 9.502700805664062, 58.339599609375, 40.3590087890625, 50.78204345703125, 4.648162841796875, 30.364044189453125, 37.1370849609375, -17.7474365234375, 42.7491455078125, 74.96914672851562, -2.230499267578125, 40.873382568359375, 20.8814697265625, 1.56011962890625, -6.9021759033203125, 21.661163330078125, 16.448699951171875, 42.48834228515625, 13.742118835449219, -26.09552001953125, -4.03759765625, -17.28558349609375, -5.418060302734375, 34.250579833984375, 48.91754150390625, 16.81890869140625, 49.138519287109375, 3.471435546875, 15.504348754882812, -7.675323486328125, 25.54644775390625, -62.259185791015625, 22.511669158935547, 48.19427490234375, 19.55645751953125, -11.836029052734375, 45.5208740234375, 45.220245361328125, 105.17974853515625, -3.6188201904296875, 14.36749267578125, 18.84912109375, 36.41551208496094, 10.153556823730469, 15.567962646484375, 31.4744873046875, -35.73931884765625, 54.56854248046875, 25.173805236816406, -42.452911376953125, 14.268056869506836, 52.24981689453125, 58.131439208984375, -25.614501953125, 7.72479248046875, -0.75775146484375, 19.094833374023438, -15.067794799804688, 9.1412353515625, 0.0, 7.419597625732422, 16.61334228515625, -5.166656494140625, 15.954612731933594, 39.96331787109375, 5.6276092529296875, 69.174072265625, 55.163909912109375, 32.188568115234375, 16.656723022460938, 30.954498291015625, 79.32972717285156, 0.8166656494140625, 33.592018127441406, 47.075897216796875, 24.724151611328125, -15.112777709960938, 40.8348388671875, 12.554962158203125, -74.6837158203125, 34.48187255859375, 65.208740234375, 20.762649536132812, 8.3533935546875, -21.44073486328125, 50.32318115234375, 43.00788879394531, 39.929290771484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000105.npy"}
{"epoch": 0.2198952879581152, "step": 106, "batch_size": 128, "mean": 12.57497787475586, "std": 30.50245475769043, "min": -70.78054809570312, "p10": -24.685305786132812, "median": 11.671539306640625, "p90": 49.05346069335936, "max": 83.23626708984375, "pos_frac": 0.6484375, "sample": [3.62371826171875, -5.6336669921875, 21.407135009765625, 16.958343505859375, 11.65313720703125, 47.78759765625, -11.6558837890625, 19.506179809570312, -10.062164306640625, 31.060264587402344, -35.763336181640625, -4.171852111816406, 0.1576385498046875, 55.2813720703125, 61.44189453125, 56.667877197265625, 34.253936767578125, -18.1478271484375, -18.627243041992188, -8.825759887695312, 22.46031951904297, -6.0282745361328125, 32.27484130859375, 42.085205078125, 26.046112060546875, 11.68994140625, 27.963327407836914, -13.7049560546875, 29.65441131591797, 52.00714111328125, 2.64190673828125, 21.955467224121094, 12.88641357421875, 10.586471557617188, 2.3335494995117188, -10.035911560058594, 43.59033203125, -38.01606750488281, 18.07537841796875, 38.308509826660156, 43.49432373046875, -22.596084594726562, 29.14301300048828, -32.76995849609375, -14.792221069335938, 27.75457763671875, -24.501861572265625, -0.62884521484375, 38.55815124511719, 30.964378356933594, 1.9664306640625, 3.4090118408203125, -15.187311172485352, 63.364837646484375, -69.30314636230469, -60.51826477050781, 37.396148681640625, 72.78097534179688, 75.09715270996094, 41.58062744140625, 0.02044677734375, -4.935455322265625, -44.864227294921875, -70.78054809570312, 14.317596435546875, -11.410293579101562, -3.583404541015625, -2.117584228515625, 54.03997802734375, -25.11334228515625, 43.53692626953125, 30.87554931640625, -15.326515197753906, 20.66351318359375, -32.6876220703125, 2.437288284301758, -15.814277648925781, -17.260147094726562, 26.47198486328125, 13.697662353515625, 9.662948608398438, 83.23626708984375, 19.312698364257812, 8.00482177734375, -7.8431396484375, 18.5904541015625, 69.04656982421875, 2.0680084228515625, 43.00225830078125, -36.798858642578125, 34.0181884765625, -8.686996459960938, 30.646148681640625, 2.95831298828125, -0.74603271484375, 23.65692138671875, 74.19131469726562, 29.97113037109375, 7.3565826416015625, 46.40809631347656, 71.27777099609375, -32.359832763671875, -1.93310546875, -12.506240844726562, 20.82073974609375, 10.739704132080078, 45.55534362792969, -26.081649780273438, 9.20703125, 10.89373779296875, -32.073883056640625, -13.6708984375, 70.97137451171875, -1.2699718475341797, -8.904937744140625, 25.826278686523438, 12.832183837890625, -16.462936401367188, 47.30403137207031, 14.71185302734375, 44.82861328125, 27.62567138671875, 1.20733642578125, 29.873321533203125, 42.491546630859375, 18.944839477539062, 41.28147888183594, -0.6528167724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000106.npy"}
{"epoch": 0.22198952879581152, "step": 107, "batch_size": 128, "mean": 19.688522338867188, "std": 28.87677764892578, "min": -104.24774169921875, "p10": -11.535554885864258, "median": 15.888893127441406, "p90": 57.675140380859375, "max": 91.37237548828125, "pos_frac": 0.75, "sample": [15.693710327148438, -6.38787841796875, 9.720794677734375, 16.350486755371094, 1.5260009765625, -8.3526611328125, 31.302490234375, 58.670379638671875, 13.231597900390625, 24.697509765625, -27.627151489257812, 8.102188110351562, 24.68645477294922, 50.34556579589844, 2.4242401123046875, -5.01556396484375, 30.8970947265625, -19.568878173828125, 31.205581665039062, 68.7894287109375, -3.574810028076172, 49.79954528808594, 36.56275177001953, 11.438507080078125, -14.56329345703125, 45.102630615234375, 24.43280029296875, -26.17864990234375, 50.44677734375, -15.752090454101562, 69.28524780273438, 4.2039947509765625, 10.92572021484375, 18.6182861328125, 52.68798828125, 50.445770263671875, 28.025360107421875, 16.11089324951172, 91.37237548828125, 5.71728515625, 42.731842041015625, 0.38916015625, 5.367645263671875, 39.36387634277344, 19.49468994140625, 7.74151611328125, -14.0782470703125, 16.41119384765625, 31.198707580566406, 11.62445068359375, 11.626934051513672, -2.0695571899414062, 21.503280639648438, 35.64982604980469, 37.11492919921875, -11.80294418334961, 2.172332763671875, 28.175384521484375, 39.30377197265625, 36.52473449707031, 4.1478271484375, 7.1588134765625, -1.3931884765625, 36.1729736328125, -0.459136962890625, -17.442108154296875, 11.191574096679688, 57.944091796875, 46.76458740234375, 66.42535400390625, -18.9254150390625, 60.5289306640625, 2.01837158203125, 53.8104248046875, 76.77447509765625, -3.288726806640625, 0.877838134765625, 54.3660888671875, 40.53401184082031, 33.28860092163086, 43.2139892578125, 25.845108032226562, 6.39532470703125, -10.102310180664062, 15.633148193359375, 18.550247192382812, 32.81982421875, -15.545135498046875, 88.8055419921875, 0.6082763671875, 57.137603759765625, 54.37989807128906, -1.0123748779296875, 44.855712890625, -6.1620941162109375, 2.4561805725097656, 28.69482421875, 1.12310791015625, 10.26007080078125, 7.094266891479492, -104.24774169921875, -22.541748046875, 22.3182373046875, 59.440185546875, 24.03765106201172, -2.5100479125976562, 90.38339233398438, 6.306304931640625, 39.31951904296875, -0.5368919372558594, 62.53150939941406, -17.514007568359375, 16.084075927734375, 6.1995849609375, -3.55706787109375, 38.2744140625, 28.565322875976562, -2.7703857421875, -9.072311401367188, 2.402587890625, -11.42095947265625, 26.697998046875, -10.813812255859375, 39.56800842285156, -1.8400020599365234, 77.10610961914062, 10.372394561767578, 57.55987548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000107.npy"}
{"epoch": 0.22408376963350785, "step": 108, "batch_size": 128, "mean": 18.23785972595215, "std": 32.64109802246094, "min": -72.27932739257812, "p10": -22.938506317138668, "median": 15.210098266601562, "p90": 60.37649383544921, "max": 116.7845458984375, "pos_frac": 0.765625, "sample": [53.109375, -22.240020751953125, -7.312187194824219, 23.727935791015625, 49.07312774658203, 11.3253173828125, 13.993091583251953, 16.027320861816406, 0.3156089782714844, 58.16969299316406, 0.097747802734375, -12.11962890625, -26.247840881347656, 35.833953857421875, 2.027587890625, -34.87273406982422, 68.4866943359375, -6.3366241455078125, -4.002521514892578, 29.078521728515625, 6.31890869140625, 21.125045776367188, 27.7269287109375, -5.6031494140625, 38.379852294921875, 18.972854614257812, 68.630126953125, -22.176239013671875, 0.0, -43.620147705078125, 41.50480270385742, 0.9541015625, 35.59796142578125, 16.4764404296875, 90.05789184570312, 1.7105712890625, 37.142578125, 22.59283447265625, 2.522979736328125, 0.7616863250732422, 1.9386558532714844, -30.071456909179688, 6.56976318359375, -25.595489501953125, 18.49951171875, 59.28004455566406, 34.432861328125, -43.03558349609375, -32.867095947265625, 85.7100830078125, 36.64020538330078, 17.484413146972656, 34.171905517578125, 111.22802734375, 9.952316284179688, 13.548126220703125, 0.0, 41.097412109375, 2.370737075805664, 1.516897201538086, 0.8737888336181641, 68.21612548828125, 32.97889709472656, 11.19110107421875, -71.22203063964844, -36.47900390625, 42.07086181640625, 19.00385856628418, 15.550552368164062, 26.16265869140625, 42.8826904296875, 24.114212036132812, -1.3968505859375, -3.420623779296875, -10.895675659179688, 41.84796142578125, 58.094818115234375, -5.640739440917969, 116.7845458984375, 54.054473876953125, 21.77099609375, -27.249004364013672, 77.8369140625, 35.33355712890625, 21.285964965820312, 1.217376708984375, 40.67181396484375, 3.1671371459960938, 4.408935546875, 43.862762451171875, 3.527252197265625, 41.94500732421875, -5.043426513671875, 17.937149047851562, 11.362817764282227, 11.219390869140625, 73.6077880859375, 50.05902099609375, 81.92681884765625, 62.93487548828125, 10.509117126464844, 6.3451690673828125, 8.902420043945312, 0.0, -1.451751708984375, 13.459983825683594, 2.6020736694335938, 66.71078491210938, 15.672855377197266, 76.5748291015625, 36.578887939453125, -72.27932739257812, 34.233184814453125, 12.116498947143555, 17.894912719726562, 52.50616455078125, 57.9193115234375, 14.869644165039062, 6.6302032470703125, 16.712135314941406, -28.062744140625, -14.400047302246094, -24.56830596923828, 5.956695556640625, 20.7635498046875, 6.55316162109375, 24.608245849609375, 20.451812744140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000108.npy"}
{"epoch": 0.2261780104712042, "step": 109, "batch_size": 128, "mean": 20.344404220581055, "std": 32.708526611328125, "min": -49.5335693359375, "p10": -17.720138549804688, "median": 15.519851684570312, "p90": 66.69007568359375, "max": 130.12103271484375, "pos_frac": 0.765625, "sample": [-4.15008544921875, -8.73828125, -27.152931213378906, 45.13862609863281, 61.470123291015625, 17.099365234375, 28.3267822265625, 6.124847412109375, -43.754058837890625, 15.768707275390625, 23.29315185546875, 116.98095703125, 1.8409194946289062, -14.839279174804688, -11.133255004882812, 4.345703125, -13.147109985351562, 7.970245361328125, 67.89535522460938, 0.5539207458496094, 10.94150161743164, 14.395309448242188, -21.224365234375, -1.531524658203125, -15.392337799072266, 14.650421142578125, 1.996124267578125, 11.078399658203125, -0.10830307006835938, 36.455535888671875, 15.794387817382812, 1.8437957763671875, 49.16876220703125, 74.7132568359375, 33.627838134765625, 55.70951461791992, 39.46534729003906, 69.81085205078125, 0.9264297485351562, 27.4193115234375, 41.6619873046875, 40.879638671875, 7.905235290527344, -2.0677013397216797, 21.83380126953125, 75.04391479492188, 0.7637176513671875, 68.83673095703125, 24.974685668945312, 9.6865234375, 66.85879516601562, 5.134979248046875, 29.3614501953125, 21.7734375, -46.27410888671875, 12.98089599609375, 30.1182861328125, 2.926891326904297, 35.725730895996094, 46.124603271484375, 66.61776733398438, 14.582870483398438, -1.472442626953125, 33.707244873046875, 17.35791015625, 9.347579956054688, -39.41960144042969, 20.559600830078125, 19.772071838378906, 24.99811553955078, 23.008758544921875, -20.966461181640625, -31.5738525390625, 21.76171875, 32.518463134765625, 42.70416259765625, 55.311126708984375, -29.80859375, 0.7717399597167969, 67.904296875, 5.2581787109375, 38.83717346191406, 54.548858642578125, 22.35504913330078, 6.6246795654296875, 8.330062866210938, 37.77764892578125, 0.6765823364257812, -21.362106323242188, 88.981689453125, 46.36895751953125, 45.6934814453125, 8.569381713867188, 103.13958740234375, 6.226291656494141, 15.27099609375, 1.2206535339355469, 130.12103271484375, -12.767807006835938, 29.77973175048828, -17.593475341796875, 76.32728576660156, 0.0, 18.152374267578125, 67.87844848632812, 13.0146484375, 62.81756591796875, 30.955078125, -0.7316436767578125, 44.554473876953125, 6.074310302734375, 11.29296875, -18.01568603515625, -8.580436706542969, 47.092369079589844, -2.0214691162109375, 39.62457275390625, 55.35906982421875, -46.35247802734375, -49.5335693359375, -33.04374694824219, 36.786376953125, 22.843124389648438, 63.01728820800781, 14.093658447265625, -14.727203369140625, 10.117889404296875, 43.46580505371094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000109.npy"}
{"epoch": 0.22827225130890053, "step": 110, "batch_size": 128, "mean": 19.17294692993164, "std": 34.54230880737305, "min": -74.665771484375, "p10": -20.086402893066406, "median": 16.05860137939453, "p90": 66.42215270996093, "max": 110.88427734375, "pos_frac": 0.7421875, "sample": [14.046234130859375, -74.665771484375, -31.68804168701172, 62.48773193359375, 6.380317687988281, 11.38846206665039, -16.528167724609375, 12.09259033203125, -17.55950927734375, 100.27157592773438, 28.25421142578125, 3.4426116943359375, -1.845571517944336, 44.434814453125, -2.3169021606445312, -7.228584289550781, 20.2998046875, 19.147560119628906, 47.73780822753906, 2.040924072265625, -20.850799560546875, 27.277252197265625, -1.4952659606933594, 44.85040283203125, 24.395004272460938, 9.296924591064453, -17.944561004638672, 40.78730010986328, 18.42169189453125, -6.9394378662109375, 87.9752197265625, 62.357879638671875, 83.6500244140625, 4.221015930175781, 71.86083984375, 17.454010009765625, 57.45512390136719, 52.32511901855469, 63.99456787109375, -3.23016357421875, 15.893875122070312, 2.9566650390625, -19.758804321289062, 35.76910400390625, -18.515625, 10.18087387084961, 26.007431030273438, -30.195037841796875, 13.564651489257812, 4.2081298828125, 49.817352294921875, -51.18328857421875, 20.995849609375, 13.98028564453125, 44.19036865234375, 4.308319091796875, 7.891937255859375, -55.846954345703125, 35.87152099609375, 110.88427734375, 1.512176513671875, 16.074142456054688, 34.06581115722656, -1.254913330078125, -0.836700439453125, 43.440155029296875, -34.776153564453125, 57.859771728515625, 52.468475341796875, -23.70892333984375, 0.2639904022216797, 28.4913330078125, 18.010719299316406, 32.160621643066406, 18.91968536376953, 20.442428588867188, -3.9133453369140625, -27.95111083984375, -47.94221496582031, 42.74053955078125, 0.0, -15.82855224609375, 41.838134765625, 16.043060302734375, -34.203792572021484, 40.711212158203125, -0.5098686218261719, -46.06172180175781, 64.66903686523438, 10.012725830078125, 32.362884521484375, 77.19711303710938, 19.875457763671875, 22.435791015625, 21.820335388183594, 31.83074951171875, 43.0185546875, 3.6212310791015625, 5.0754241943359375, 49.95623779296875, 74.75360107421875, 5.0137939453125, 9.801219940185547, 74.32403564453125, -63.54693603515625, 36.185699462890625, 10.033462524414062, -10.500228881835938, 70.51275634765625, 54.135589599609375, 0.9241485595703125, 81.81707763671875, 0.18393707275390625, 5.985931396484375, 27.012771606445312, 76.82086181640625, 81.35479736328125, 4.779754638671875, 42.22174072265625, 33.04705810546875, -7.6027984619140625, 80.07608032226562, 13.362091064453125, 18.66457748413086, 2.726409912109375, 26.781280517578125, -17.932220458984375, 56.195220947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000110.npy"}
{"epoch": 0.23036649214659685, "step": 111, "batch_size": 128, "mean": 20.33997344970703, "std": 36.08433151245117, "min": -82.20513916015625, "p10": -22.9857177734375, "median": 17.947214126586914, "p90": 60.81106910705564, "max": 118.2772216796875, "pos_frac": 0.7578125, "sample": [2.5936355590820312, -18.820892333984375, 50.96514892578125, -49.8165283203125, 117.1414794921875, -1.9535179138183594, 51.771148681640625, 83.86572265625, -2.2993202209472656, 6.6556854248046875, 71.899658203125, 2.566070556640625, 18.348602294921875, 20.143524169921875, -9.198089599609375, 56.10384750366211, 19.276634216308594, 72.03158569335938, -36.78857421875, -18.5048828125, 20.605987548828125, 29.267072677612305, 27.159423828125, 28.1982421875, 38.00567626953125, 6.682891845703125, 10.733589172363281, 54.18719482421875, -7.998046875, 39.245384216308594, 31.114952087402344, 43.427001953125, -8.94708251953125, 101.69259643554688, -9.22540283203125, 4.366889953613281, -13.18310546875, 42.319183349609375, 0.95379638671875, 69.15257263183594, 18.48534393310547, 27.549705505371094, -5.636177062988281, 52.0682373046875, 7.4893798828125, 2.530029296875, 110.87185668945312, -8.461822509765625, 67.60311889648438, 49.237449645996094, 3.137908935546875, 118.2772216796875, -54.1815185546875, 2.4109954833984375, 5.575927734375, 20.0340576171875, 8.782958984375, 56.29896545410156, -10.100662231445312, 94.67364501953125, 19.35528564453125, 9.341598510742188, 11.829803466796875, 23.097747802734375, 6.7722015380859375, -38.92390441894531, 3.986236572265625, 49.944366455078125, 40.401123046875, 6.99853515625, 23.65533447265625, 4.224403381347656, 14.07598876953125, 14.716384887695312, 57.72723388671875, 5.665754318237305, -3.256103515625, 43.8795166015625, 4.343353271484375, -26.89044189453125, 7.831428527832031, 43.01019287109375, -28.29632568359375, 1.8695068359375, 18.297698974609375, -25.161880493164062, -14.278465270996094, 1.177734375, 58.242794036865234, 49.160125732421875, 57.39862060546875, 37.64190673828125, 50.53394317626953, -1.3668136596679688, 49.97953796386719, 31.285125732421875, 10.452447891235352, 1.510162353515625, 90.60421752929688, 24.128273010253906, 32.11054229736328, 7.8600616455078125, -51.849822998046875, -82.20513916015625, -41.903076171875, 53.81500244140625, -19.561065673828125, 38.233551025390625, 10.877288818359375, 30.624862670898438, 81.44621276855469, -3.74261474609375, 57.79998779296875, 57.804046630859375, 66.8037109375, 35.975799560546875, 55.04315185546875, -46.09088134765625, 42.067413330078125, 17.596729278564453, 26.77227020263672, -34.74334716796875, 27.157623291015625, 13.174789428710938, 25.0701904296875, 4.6197967529296875, -22.520538330078125, -24.071136474609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000111.npy"}
{"epoch": 0.2324607329842932, "step": 112, "batch_size": 128, "mean": 13.129972457885742, "std": 39.163936614990234, "min": -84.00848388671875, "p10": -35.734076690673824, "median": 9.024760246276855, "p90": 64.02428131103514, "max": 115.64617919921875, "pos_frac": 0.6171875, "sample": [50.769378662109375, 8.153251647949219, -9.43768310546875, 16.553466796875, -4.5093536376953125, 25.966506958007812, -6.171684265136719, -77.91461181640625, -17.42945098876953, 1.42022705078125, 9.853399276733398, -2.070648193359375, -56.85577392578125, 69.09677124023438, -5.02679443359375, 12.946189880371094, 51.67828369140625, 57.498138427734375, 0.33502960205078125, 46.73291015625, -37.21199035644531, 35.102996826171875, 21.51275634765625, 54.439910888671875, -84.00848388671875, -7.2950439453125, -13.348480224609375, 27.442222595214844, 62.92271423339844, 34.622291564941406, -35.100685119628906, 41.58623504638672, -47.89642333984375, 17.703887939453125, 19.157264709472656, -50.22808837890625, -3.99920654296875, 24.274372100830078, 8.069015502929688, -9.069923400878906, 42.501556396484375, 6.04083251953125, 55.4736328125, 68.41763305664062, 26.58190155029297, -14.631011962890625, 50.71044921875, 8.196121215820312, 80.60601806640625, -28.686065673828125, 16.416473388671875, -3.2186431884765625, -3.4032669067382812, 2.710205078125, -11.28314208984375, -0.102386474609375, -46.307037353515625, -10.470901489257812, -66.24948120117188, -6.549701690673828, -8.135103225708008, -12.168235778808594, 34.509124755859375, 111.117919921875, 13.8828125, -27.760833740234375, 56.814239501953125, 68.76298522949219, 7.4479522705078125, 38.10613250732422, 23.264095306396484, -1.8480720520019531, 40.641021728515625, 36.15032958984375, 12.260650634765625, 115.64617919921875, -3.4448089599609375, 6.740142822265625, 75.60809326171875, 16.31235122680664, -12.03839111328125, 46.217864990234375, 15.050178527832031, 0.3283195495605469, 21.1558837890625, 40.499969482421875, 24.646240234375, 5.4459228515625, 10.909904479980469, -18.128662109375, 113.78216552734375, -19.568145751953125, 85.73727416992188, 1.327484130859375, -58.392547607421875, 23.778091430664062, -30.727703094482422, -1.4949970245361328, -59.915191650390625, -9.614669799804688, -45.10929870605469, 44.15826416015625, -13.952713012695312, -26.0518798828125, 36.54856872558594, 68.9002685546875, -14.581329345703125, 72.8431396484375, 21.6212158203125, 55.38006591796875, 66.5946044921875, -2.524444580078125, 2.8538818359375, 45.833770751953125, 30.1043701171875, 4.3015289306640625, -42.97747802734375, 13.205047607421875, 20.838638305664062, -21.86199951171875, 59.67633056640625, -26.740631103515625, 56.892547607421875, 57.959686279296875, 71.72073364257812, -74.76654052734375, 0.5191116333007812, 39.328887939453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000112.npy"}
{"epoch": 0.23455497382198953, "step": 113, "batch_size": 128, "mean": 24.571178436279297, "std": 40.96280288696289, "min": -93.47064208984375, "p10": -19.08989620208739, "median": 25.12017822265625, "p90": 74.61513366699218, "max": 168.66094970703125, "pos_frac": 0.734375, "sample": [-6.5423583984375, 39.893577575683594, 55.9561767578125, -3.2383041381835938, 32.170684814453125, -3.656627655029297, -10.338775634765625, 71.85986328125, 34.6549072265625, 21.198944091796875, 24.729888916015625, -11.598159790039062, 68.72552490234375, -12.032903671264648, -16.51123046875, 2.6708240509033203, 1.8861885070800781, 0.375518798828125, -3.2762451171875, 3.6766586303710938, 120.51495361328125, 32.423828125, -0.5711288452148438, 89.58551025390625, 27.70684814453125, -34.154998779296875, 35.64697265625, 41.845703125, 14.011810302734375, -26.979839324951172, 15.903289794921875, -14.78173828125, 168.66094970703125, 40.13720703125, 47.2381591796875, 15.438812255859375, 33.81227111816406, -35.75526428222656, 54.822479248046875, -67.75961303710938, 17.312713623046875, 39.98138427734375, 5.6037139892578125, 60.803466796875, 96.22299194335938, 30.5303955078125, -15.5390625, 1.9125137329101562, -0.48001861572265625, 30.4854736328125, 20.251007080078125, 38.642921447753906, -48.369873046875, 48.98603820800781, 38.2906494140625, 40.54359436035156, 33.26069259643555, 26.511749267578125, 31.1737060546875, -14.830184936523438, 7.9365386962890625, 27.896438598632812, -25.106782913208008, 90.59283447265625, -5.71356201171875, 38.486541748046875, 6.02520751953125, 22.72149658203125, -28.34027099609375, 28.100982666015625, 9.48931884765625, 164.87762451171875, -2.3353614807128906, 48.72660827636719, -37.82246398925781, -0.8643264770507812, 47.16790771484375, 59.210296630859375, 74.4365234375, 2.3594512939453125, 47.234336853027344, -28.43768310546875, 13.32940673828125, 61.856048583984375, 91.34149169921875, 43.86412048339844, 11.459785461425781, -6.50885009765625, 2.5796356201171875, 8.254974365234375, 75.03189086914062, 76.17998504638672, 35.11067199707031, 25.510467529296875, -29.133697509765625, 63.73162841796875, -0.362518310546875, 49.79425048828125, 101.860107421875, 46.937744140625, 23.88288116455078, -14.81146240234375, 36.453216552734375, 39.64414978027344, 95.43914794921875, 51.6097412109375, 94.70027160644531, 101.685791015625, -14.2177734375, 7.31884765625, 38.56626892089844, 51.73374938964844, 2.524871826171875, 64.50238037109375, -12.778289794921875, -93.47064208984375, 1.320892333984375, 5.4414825439453125, 37.21931457519531, 34.75126647949219, 2.41748046875, 49.85652160644531, -35.33251953125, 38.2176513671875, 19.523902893066406, 3.8885498046875, 26.4144287109375, -58.51441192626953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000113.npy"}
{"epoch": 0.23664921465968586, "step": 114, "batch_size": 128, "mean": 13.87610149383545, "std": 37.18177795410156, "min": -87.621337890625, "p10": -32.357908630371085, "median": 13.782920837402344, "p90": 60.51480255126953, "max": 98.453125, "pos_frac": 0.671875, "sample": [11.29461669921875, 28.9140625, 28.223052978515625, 46.607852935791016, -9.577262878417969, 4.70166015625, 2.3282413482666016, 14.82718276977539, 21.351829528808594, -15.033676147460938, -11.725051879882812, 64.36424255371094, 28.42047119140625, 61.90545654296875, 4.5684814453125, 56.662078857421875, -24.383831024169922, 6.826873779296875, -51.43353271484375, 6.534210205078125, 2.799957275390625, -80.746337890625, 14.756950378417969, 59.714080810546875, 37.97881317138672, 29.4010009765625, 15.98870849609375, 44.640655517578125, 54.230743408203125, 32.654296875, 22.713653564453125, 47.59648132324219, 12.492607116699219, 16.848922729492188, 28.8802490234375, -85.72760009765625, 64.32307434082031, -54.71954345703125, 60.15087890625, -3.375946044921875, 26.9757080078125, -8.266845703125, 5.144815444946289, -17.93585205078125, 9.181327819824219, 7.0555572509765625, 29.978103637695312, 33.449462890625, 23.19586181640625, -14.34991455078125, -8.086387634277344, -8.18277359008789, 43.33210754394531, -46.791900634765625, -42.39190673828125, 3.3708953857421875, -3.845703125, 60.97320556640625, 60.865966796875, -42.813812255859375, 68.14869689941406, -74.71878051757812, 72.21331787109375, -39.518463134765625, -11.941619873046875, 97.94638061523438, 39.20935821533203, -28.88409423828125, 21.109970092773438, 94.4820556640625, -87.621337890625, 49.07769775390625, -20.997955322265625, 48.886505126953125, -6.541748046875, 51.953857421875, -0.9404830932617188, -12.182456970214844, 16.635726928710938, -59.416656494140625, 14.270736694335938, 15.215423583984375, 18.72802734375, -7.90301513671875, -2.1558055877685547, -9.212806701660156, -36.698333740234375, 37.37004089355469, 0.0, 71.93618774414062, 13.29510498046875, -16.105712890625, 52.64347839355469, 0.208953857421875, 3.491729736328125, 57.580230712890625, -2.975341796875, 32.93519592285156, 2.522613525390625, 12.115249633789062, 50.32994079589844, 49.813385009765625, 19.401004791259766, -0.2963104248046875, 98.453125, 55.562591552734375, 20.643341064453125, 4.455451965332031, -30.273727416992188, 0.0, 89.87881469726562, -25.012481689453125, -30.497726440429688, 35.08013916015625, -27.4732666015625, 60.36430358886719, 14.867401123046875, -37.53466796875, 8.743526458740234, 11.309036254882812, 22.342315673828125, 61.632843017578125, 24.21217918395996, 6.05755615234375, 19.103179931640625, 25.149139404296875, 10.366216659545898, 58.465179443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000114.npy"}
{"epoch": 0.2387434554973822, "step": 115, "batch_size": 128, "mean": 13.929231643676758, "std": 40.63798141479492, "min": -84.53817749023438, "p10": -28.703385925292967, "median": 8.929817199707031, "p90": 68.26329345703125, "max": 152.3170166015625, "pos_frac": 0.6171875, "sample": [112.32363891601562, -29.109664916992188, 44.398529052734375, -13.163753509521484, -6.7432861328125, -31.12249755859375, -15.20025634765625, 21.51495361328125, -16.017959594726562, 14.656778335571289, -13.487571716308594, 29.68804931640625, -12.72833251953125, 0.19930267333984375, 35.903526306152344, -26.237579345703125, -4.4339752197265625, -32.44938659667969, 47.61415100097656, 23.376937866210938, -39.65557098388672, 15.77947998046875, 7.882743835449219, 8.03216552734375, -6.459949493408203, 0.3339691162109375, 24.2711181640625, 26.209190368652344, 23.610870361328125, 17.04266357421875, -3.191986083984375, 23.67803955078125, -7.8935546875, -3.2826080322265625, 3.194063186645508, 13.65118408203125, -7.192962646484375, -38.360198974609375, -19.40265655517578, 120.27227783203125, 12.074676513671875, 53.76458740234375, 53.6043701171875, 30.92046356201172, -44.096038818359375, -22.314910888671875, 49.094268798828125, -16.535125732421875, 46.47662353515625, 83.44879150390625, 21.21484375, 140.30670166015625, -36.383026123046875, 13.15545654296875, -13.59661865234375, -5.712158203125, 67.86813354492188, 44.9737548828125, 70.42744445800781, -73.18887329101562, -72.72525024414062, -9.27984619140625, 60.105072021484375, 29.29400634765625, -27.3486328125, 30.436859130859375, 4.758766174316406, 19.2789306640625, 51.632415771484375, 84.2705078125, -0.1417675018310547, 9.22674560546875, 87.80374908447266, 29.016860961914062, 26.68204689025879, 16.751272201538086, 101.38330078125, 13.158050537109375, -16.320404052734375, 43.936920166015625, -0.683837890625, -6.354766845703125, -13.21331787109375, 91.59405517578125, 43.76658630371094, -8.7640380859375, -84.53817749023438, 10.651473999023438, 10.956096649169922, 69.18533325195312, -29.94826889038086, 19.480422973632812, -17.37603759765625, 87.86215209960938, 20.373992919921875, 3.168975830078125, -8.862274169921875, -47.49310302734375, -12.698837280273438, 6.2086334228515625, 47.62353515625, -7.099937438964844, 9.862922668457031, 11.248453140258789, 27.6971435546875, 72.47286987304688, 152.3170166015625, 5.501068115234375, -27.413482666015625, -28.529266357421875, 8.197586059570312, 3.6254405975341797, 28.36029052734375, -15.995925903320312, 7.2861480712890625, -9.44451904296875, -16.689651489257812, 33.526611328125, 8.275520324707031, 61.45599365234375, -78.38751220703125, 1.500762939453125, 40.898895263671875, 19.185508728027344, 8.632888793945312, 13.850616455078125, -21.491744995117188, 48.23663330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000115.npy"}
{"epoch": 0.24083769633507854, "step": 116, "batch_size": 128, "mean": 18.022811889648438, "std": 36.29035186767578, "min": -121.35394287109375, "p10": -19.736117935180665, "median": 13.441768646240234, "p90": 65.88084564208984, "max": 114.74163818359375, "pos_frac": 0.6796875, "sample": [-0.017425537109375, -19.652400970458984, 29.840621948242188, -2.4085159301757812, 15.57708740234375, 28.15045166015625, 37.40724182128906, 48.65447998046875, 85.01980590820312, 31.83673095703125, 31.73095703125, 10.7861328125, 48.54972839355469, 18.24468231201172, -9.08526611328125, 0.24432373046875, 50.20848846435547, 50.55694580078125, -16.11501693725586, 14.623123168945312, -66.22268676757812, -6.429874420166016, 64.40119171142578, 40.81787109375, 2.9387283325195312, -33.06396484375, 49.28028869628906, -4.585445404052734, 31.468292236328125, -34.751739501953125, 4.674522399902344, 38.8504638671875, -9.62774658203125, 79.23092651367188, 12.130218505859375, 16.1285400390625, 23.191909790039062, 19.7686767578125, -18.04791259765625, 0.1273956298828125, -5.576755523681641, 31.718994140625, 22.025726318359375, -20.924449920654297, 12.8876953125, 19.04073715209961, -7.249267578125, 47.25262451171875, -39.050140380859375, 40.81011962890625, -121.35394287109375, 15.785263061523438, 72.63249206542969, -17.94659423828125, -9.603103637695312, 59.39453125, 38.03778076171875, 70.33238220214844, 2.9075927734375, 7.982246398925781, 13.755752563476562, 56.470481872558594, 12.205963134765625, -22.125579833984375, -0.0317535400390625, 0.0, -0.4656791687011719, 90.36715698242188, -5.158638000488281, 7.1201171875, 114.74163818359375, 66.71682739257812, 11.13323974609375, -18.526336669921875, -3.7888259887695312, 19.45874786376953, -16.77320098876953, 12.729522705078125, 10.927635192871094, 51.186737060546875, 13.127784729003906, 57.40399169921875, 36.673675537109375, 0.0, -2.584787368774414, 19.70541000366211, -6.699554443359375, 16.1451416015625, 92.72467041015625, 6.6263580322265625, 11.661788940429688, 104.0550537109375, 67.67312622070312, -39.27960205078125, 9.8519287109375, 59.977447509765625, -6.133201599121094, 26.220226287841797, 10.496665954589844, 3.493896484375, -0.054050445556640625, 4.170989990234375, -37.12052917480469, 65.52256774902344, 45.9112548828125, -41.9111328125, 21.39959716796875, 59.8516845703125, -4.7330322265625, -8.627708435058594, 19.250534057617188, 24.707778930664062, 73.72308349609375, -8.747596740722656, 47.929046630859375, 50.02928161621094, 23.053558349609375, 36.1141357421875, 91.16900634765625, 39.28243637084961, -75.10816955566406, 40.649452209472656, 7.46209716796875, 79.62966918945312, -28.354766845703125, 7.343482971191406, -19.93145751953125, 29.689208984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000116.npy"}
{"epoch": 0.24293193717277486, "step": 117, "batch_size": 128, "mean": 20.927654266357422, "std": 36.125511169433594, "min": -68.81460571289062, "p10": -24.727099609374996, "median": 19.894412994384766, "p90": 68.49182815551757, "max": 107.10191345214844, "pos_frac": 0.7109375, "sample": [8.128047943115234, 30.61907958984375, 20.760303497314453, 37.46589660644531, 17.471176147460938, 31.097686767578125, 7.829071044921875, -1.334310531616211, 0.56292724609375, -27.213943481445312, 46.86692810058594, -32.790679931640625, 14.89263916015625, 46.959197998046875, 29.84500503540039, 37.086517333984375, 21.535354614257812, 13.849716186523438, 82.50653076171875, -9.89129638671875, 55.463287353515625, -31.748855590820312, -0.37984466552734375, 0.7554721832275391, 14.63800048828125, 38.725128173828125, -1.8370361328125, -47.194122314453125, 41.851806640625, 76.36666870117188, 40.664825439453125, 72.74397277832031, 86.3714599609375, 28.22077751159668, 14.694269180297852, -7.8079986572265625, 35.702239990234375, -0.0222930908203125, 2.969696044921875, -14.4537353515625, 91.71319580078125, -1.9057998657226562, 38.71044921875, 49.94989013671875, 0.0, 61.192535400390625, -10.886474609375, -1.38348388671875, 78.6136474609375, 103.44625854492188, 0.423004150390625, 94.58114624023438, -46.75042724609375, 92.71426391601562, 47.185882568359375, 25.743606567382812, -11.7708740234375, 35.404518127441406, 19.143211364746094, 28.261123657226562, 53.11328125, 49.0533447265625, 27.941314697265625, 53.86548614501953, 7.14227294921875, 47.202728271484375, 41.318603515625, 2.530670166015625, -9.007797241210938, -11.867462158203125, 49.11376953125, -51.87738037109375, 0.4574623107910156, 12.525009155273438, 107.10191345214844, -68.81460571289062, 87.88900756835938, 33.15185546875, 65.068603515625, 32.76264953613281, 14.967838287353516, 24.91943359375, -14.697494506835938, 90.40835571289062, 19.09375, -50.669189453125, -48.31712341308594, -14.8245849609375, 27.770965576171875, -3.247243881225586, 32.812469482421875, 24.63592529296875, 67.3128662109375, 70.9596176147461, 6.089191436767578, 36.262359619140625, -43.921417236328125, 23.007064819335938, -17.087440490722656, 45.86376953125, -46.917938232421875, -5.025390625, 2.559293746948242, -26.540054321289062, -39.05387878417969, 4.1063995361328125, 57.791107177734375, 41.205177307128906, 52.13226318359375, -5.504985809326172, 4.663299560546875, 29.456512451171875, 67.4342041015625, 16.362060546875, 15.599540710449219, -15.537689208984375, 11.746088027954102, 6.154560089111328, 20.645614624023438, -23.950119018554688, 9.156448364257812, 56.83673095703125, -8.82464599609375, 32.32831573486328, -3.6136322021484375, 44.08086395263672, 57.210662841796875, 27.87384033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000117.npy"}
{"epoch": 0.2450261780104712, "step": 118, "batch_size": 128, "mean": 23.154701232910156, "std": 44.704429626464844, "min": -74.38992309570312, "p10": -34.018923950195315, "median": 15.556488037109375, "p90": 81.82196655273437, "max": 141.55557250976562, "pos_frac": 0.6796875, "sample": [59.96534729003906, -12.101730346679688, 45.26325988769531, -20.3922119140625, 52.40620422363281, 50.77398681640625, -74.38992309570312, 1.0111312866210938, 67.49015808105469, 39.29833984375, 18.95489501953125, 10.84149169921875, 39.539093017578125, 3.7359466552734375, -40.4326171875, 112.0546875, 56.564239501953125, 85.1272201538086, 94.59119415283203, -20.8515625, 18.890457153320312, 71.76092529296875, 97.8693618774414, 3.40985107421875, -46.05938720703125, -34.0645751953125, 84.85482788085938, -12.54803466796875, -55.920562744140625, 93.09869384765625, 42.056182861328125, 75.44229125976562, 82.6494140625, 47.4459228515625, 40.01095199584961, 57.11643981933594, -12.729400634765625, 13.825027465820312, 51.37895965576172, 30.255691528320312, -0.06793212890625, 58.962799072265625, 67.60302734375, -0.04524993896484375, 15.18255615234375, -16.365020751953125, 71.96047973632812, 59.645751953125, 31.50421142578125, 5.488719940185547, 36.012542724609375, -4.51409912109375, 68.14810180664062, 61.79486083984375, 59.54443359375, 117.56878662109375, -37.85707092285156, 4.468658447265625, 13.927446365356445, 6.0947265625, -5.289447784423828, 0.0, -21.817626953125, 20.18901824951172, -15.674591064453125, -18.031301498413086, -53.75018310546875, 13.534683227539062, 26.153854370117188, -33.999359130859375, -16.833045959472656, 71.69073486328125, -4.74969482421875, 38.87135314941406, -25.337890625, 72.016357421875, 80.75881958007812, 0.900604248046875, -37.444732666015625, 125.83926391601562, -8.210182189941406, 24.480606079101562, -30.561187744140625, 7.1079864501953125, 92.82119750976562, 34.189727783203125, -69.87545776367188, -26.82421875, -17.432586669921875, -41.19696044921875, 115.05252075195312, -6.525909423828125, 71.76455688476562, 23.66748046875, 0.7909393310546875, 12.474349975585938, 34.71913528442383, -10.051002502441406, 46.9560546875, 4.588287353515625, 21.73916244506836, 81.46734619140625, 8.17218017578125, 46.513427734375, 14.669418334960938, 19.623443603515625, 1.6253204345703125, -40.819671630859375, 47.465354919433594, 65.15609741210938, -21.363502502441406, -42.855987548828125, 58.76708221435547, 75.29985046386719, 8.68505859375, -25.05255126953125, -53.532867431640625, 15.930419921875, 29.54827880859375, 95.61099243164062, 14.153451919555664, 79.24078369140625, 1.9779415130615234, -17.4261474609375, 0.0, 7.95477294921875, 17.484527587890625, 141.55557250976562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000118.npy"}
{"epoch": 0.24712041884816754, "step": 119, "batch_size": 128, "mean": 16.9993896484375, "std": 44.39560317993164, "min": -85.66033935546875, "p10": -39.44447784423828, "median": 13.551979064941406, "p90": 77.68600158691406, "max": 138.95208740234375, "pos_frac": 0.6328125, "sample": [-72.71905517578125, -51.87287902832031, 49.0059814453125, 48.112518310546875, 6.14801025390625, 7.3703765869140625, -57.93260192871094, -28.863128662109375, -3.1070556640625, -20.089630126953125, 52.23321533203125, -29.0616455078125, 41.49102783203125, 66.2696533203125, -21.980712890625, 24.9171142578125, -12.868438720703125, 37.906005859375, 27.43292236328125, -14.102142333984375, -7.87213134765625, 47.81719970703125, 14.207345962524414, -31.976036071777344, 19.8333740234375, 13.208511352539062, 26.79906463623047, 12.17413330078125, 79.10879516601562, 17.311904907226562, 31.506216049194336, -47.04057312011719, -9.766860961914062, 50.24530029296875, 41.27972412109375, 11.527450561523438, 38.14808654785156, 19.728302001953125, 5.832916259765625, -1.2899131774902344, 48.440704345703125, -50.36566925048828, -1.1875, -6.117523193359375, 90.0306396484375, 1.6774730682373047, -12.277145385742188, -54.46087646484375, -4.586883544921875, -30.895751953125, 125.65548706054688, 30.07537841796875, -39.09648132324219, 77.07623291015625, -23.125335693359375, -10.157760620117188, 29.6502685546875, -2.9680519104003906, 25.388702392578125, 54.67735290527344, 13.1090087890625, 4.1875, 69.42330932617188, 116.39727783203125, 98.10430908203125, 24.122833251953125, 132.54049682617188, 20.958541870117188, 60.08892822265625, -10.19744873046875, 11.057695388793945, 13.89544677734375, 23.236618041992188, 0.0, 79.23683166503906, -22.72808837890625, 20.01385498046875, -85.33978271484375, 37.067138671875, -47.54957580566406, 31.19659423828125, 64.10874938964844, 8.14532470703125, -41.85675048828125, 30.034912109375, 17.059799194335938, 49.7279052734375, -7.1561126708984375, 36.15965270996094, 8.296585083007812, 41.453155517578125, -43.26031494140625, -14.518386840820312, 138.95208740234375, -2.426055908203125, 102.59457397460938, 43.6739501953125, -18.3558349609375, 26.51092529296875, -75.56298828125, 104.54452514648438, 63.673370361328125, 38.69146728515625, 28.845687866210938, -20.719085693359375, 53.2266845703125, -28.167274475097656, 8.456085205078125, 1.969818115234375, -40.2564697265625, 10.071075439453125, -85.66033935546875, 112.57012939453125, 46.78443908691406, 83.73507690429688, -8.014739990234375, -11.181320190429688, 41.5440673828125, 30.762298583984375, 28.96636962890625, 35.103515625, 9.07952880859375, 34.8370361328125, -11.781963348388672, 87.64341735839844, 12.817245483398438, -5.626678466796875, -24.9005126953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000119.npy"}
{"epoch": 0.24921465968586387, "step": 120, "batch_size": 128, "mean": 19.44042205810547, "std": 44.264198303222656, "min": -91.66619873046875, "p10": -28.82939529418945, "median": 15.595802307128906, "p90": 72.86211853027343, "max": 147.76544189453125, "pos_frac": 0.640625, "sample": [-23.728729248046875, 58.5391845703125, -5.890830993652344, 40.25376892089844, -14.492515563964844, 67.78656005859375, 61.6851806640625, -67.40834045410156, 14.498382568359375, 20.834381103515625, -7.582340240478516, -30.68842315673828, 23.97667694091797, -9.941375732421875, -69.75338745117188, 0.04815673828125, 7.2847900390625, -10.332002639770508, 14.105438232421875, 5.513164520263672, -26.861175537109375, 102.80691528320312, 147.76544189453125, 45.12384033203125, 62.52105712890625, 40.46623992919922, -7.258941650390625, -3.742462158203125, -18.377178192138672, 46.74957275390625, 109.2086181640625, 41.432220458984375, 113.85626220703125, 1.251007080078125, -79.09576416015625, 21.45574951171875, -1.1024017333984375, 38.24493408203125, 44.654296875, 18.269134521484375, 7.661964416503906, 19.04435920715332, 17.81475830078125, -21.1671142578125, 0.0, -16.199981689453125, -38.650360107421875, -21.142181396484375, -0.8351516723632812, 12.730667114257812, -16.481964111328125, -11.048583984375, 103.1644287109375, 44.9283447265625, 45.38446044921875, 45.63575744628906, 6.840217590332031, 24.50128173828125, 6.277374267578125, 19.931655883789062, -6.148963928222656, 0.0, 2.932830810546875, 50.77073669433594, 66.22756958007812, 32.86488342285156, 61.44921875, 91.91305541992188, -15.09405517578125, 77.51751708984375, 72.40460205078125, 16.693222045898438, -7.334724426269531, 33.22662353515625, 22.956756591796875, -24.601150512695312, 8.26849365234375, 39.386444091796875, 38.61932373046875, 32.98687744140625, 58.637664794921875, -64.35479736328125, -39.46223449707031, -28.032669067382812, -26.705581665039062, 82.67694091796875, -13.035530090332031, 58.88043212890625, 85.60125732421875, 63.615272521972656, 9.494697570800781, 51.414825439453125, -61.761749267578125, 56.05401611328125, 38.88182067871094, 8.790313720703125, -22.75347900390625, 20.4241943359375, 51.491233825683594, 62.62146759033203, -46.53586196899414, 8.375234603881836, -10.830848693847656, 24.602401733398438, 47.9716796875, -0.11759185791015625, 41.165618896484375, -8.067451477050781, 120.10317993164062, -91.66619873046875, 83.358642578125, 43.194000244140625, 61.85328674316406, -4.3084716796875, -52.88592529296875, 30.260986328125, 40.42259216308594, 73.92965698242188, 4.2130279541015625, -43.37886047363281, -16.651264190673828, 43.86613464355469, -34.85679626464844, 143.31658935546875, 2.6022377014160156, 12.514404296875, -8.281295776367188, 36.222442626953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000120.npy"}
{"epoch": 0.2513089005235602, "step": 121, "batch_size": 128, "mean": 13.999553680419922, "std": 41.786399841308594, "min": -94.014404296875, "p10": -30.15413360595703, "median": 6.724071502685547, "p90": 65.60823059082031, "max": 166.42422485351562, "pos_frac": 0.609375, "sample": [-6.5165557861328125, 1.580078125, -1.3888416290283203, 1.8239822387695312, 34.60108947753906, -16.620803833007812, -11.1295166015625, 47.88628387451172, 18.014236450195312, 126.24142456054688, 15.776676177978516, -8.029251098632812, 64.87590026855469, 12.789518356323242, 47.29248046875, 13.2462158203125, -0.4239501953125, -11.307403564453125, -7.951658248901367, 99.47837829589844, 59.82237243652344, -8.596084594726562, 65.450439453125, 27.10400390625, -94.014404296875, 0.0, 34.02782440185547, -6.8107147216796875, -24.73918914794922, 39.11865234375, 12.965694427490234, 37.50886154174805, -70.2860107421875, 57.3446044921875, 39.83035659790039, -54.424407958984375, -42.04247283935547, 39.18755340576172, 48.8375244140625, 20.6202392578125, -20.10235595703125, 1.78338623046875, 51.6715087890625, -0.07058143615722656, -13.354393005371094, 6.993370056152344, 80.45706176757812, -17.858001708984375, 1.0592041015625, 0.0, 28.464725494384766, 7.616672515869141, 0.526092529296875, 36.823577880859375, 11.8817138671875, -17.20270538330078, -2.688232421875, 38.002532958984375, 63.9259033203125, -12.811126708984375, -39.028289794921875, -49.62493896484375, -14.509376525878906, 120.35479736328125, 28.919517517089844, 40.727272033691406, 68.18577575683594, 27.9581298828125, 3.5710296630859375, 12.68756103515625, -38.34162902832031, 13.47796630859375, 6.45477294921875, 50.51725387573242, -2.196208953857422, 6.436767578125, 0.2590007781982422, 9.144058227539062, -8.742324829101562, 42.4609375, 30.4501953125, -9.95050048828125, -7.8597869873046875, 66.5414047241211, 101.9227294921875, 1.2474288940429688, -63.772186279296875, 1.670074462890625, -79.30728149414062, 79.29898071289062, 37.657806396484375, 78.09925842285156, 14.716148376464844, 4.6957244873046875, 92.49690246582031, -15.268630981445312, -33.102996826171875, 27.299209594726562, 7.256683349609375, -45.444122314453125, 8.913782119750977, -24.685943603515625, 16.66333770751953, -9.626129150390625, 0.019683837890625, 30.479248046875, -0.37158203125, 47.563568115234375, -51.03150939941406, -17.346107482910156, 21.93597412109375, 166.42422485351562, 41.71064758300781, -16.18359375, 42.082672119140625, -0.775299072265625, 65.97640991210938, -21.941246032714844, 39.139739990234375, 5.81689453125, 13.297252655029297, -28.315399169921875, 116.97309875488281, -11.922149658203125, -0.5508670806884766, -28.890335083007812, -57.17987060546875, 30.145736694335938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000121.npy"}
{"epoch": 0.2534031413612565, "step": 122, "batch_size": 128, "mean": 19.668943405151367, "std": 47.95038604736328, "min": -105.37001037597656, "p10": -42.700742340087885, "median": 15.863372802734375, "p90": 83.8408447265625, "max": 147.70004272460938, "pos_frac": 0.671875, "sample": [10.144805908203125, -28.782623291015625, 51.246917724609375, -34.005126953125, -24.32647705078125, 53.22039794921875, 26.947418212890625, 23.202301025390625, -57.232940673828125, 43.7806396484375, 26.520309448242188, 0.0, 23.849029541015625, -70.43508911132812, 63.35228729248047, -105.37001037597656, 24.313125610351562, 69.91256713867188, -37.44123840332031, 20.163925170898438, -13.16192626953125, 64.96630859375, -28.462387084960938, -88.47760009765625, 16.8521728515625, 113.6204833984375, -14.201141357421875, 48.597930908203125, 147.70004272460938, -41.773826599121094, 9.710357666015625, -53.07670593261719, 21.630081176757812, -5.0021514892578125, 55.711456298828125, -21.702354431152344, 51.167510986328125, -61.98992919921875, 21.200042724609375, 44.0616455078125, -63.25877380371094, 28.013336181640625, 14.76959228515625, 66.46826171875, -5.696321487426758, -87.03083801269531, 21.08464813232422, 12.122112274169922, 78.94332885742188, -40.578857421875, -1.282379150390625, 8.981132507324219, 34.03131103515625, -51.231658935546875, 15.70843505859375, 6.204383850097656, -0.5369033813476562, -2.2110443115234375, -78.38729858398438, 29.01666259765625, 16.018310546875, 88.55252075195312, 60.61091613769531, -8.9857177734375, -0.17688751220703125, 64.88668823242188, 63.60387420654297, 68.1944580078125, 12.67718505859375, 83.8359375, 92.56753540039062, 39.853981018066406, -6.325653076171875, 96.59063720703125, 105.76043701171875, 38.92826843261719, 40.99079895019531, 0.540802001953125, 89.26734924316406, 18.4832763671875, 22.46691131591797, -13.802642822265625, 47.210845947265625, -50.00177001953125, -3.0849609375, 0.65435791015625, 8.754623413085938, 47.14591979980469, 51.90667724609375, 12.37371826171875, 24.767303466796875, -0.6984100341796875, -2.456106185913086, -41.94220733642578, 8.207588195800781, -48.353363037109375, 23.273086547851562, 90.1868667602539, 17.8079833984375, 10.585052490234375, 7.796661376953125, -4.832244873046875, 45.387725830078125, 50.685638427734375, 2.652721405029297, 74.69744873046875, 50.952728271484375, 94.8519287109375, 2.9646453857421875, -21.653545379638672, -10.201301574707031, 6.187980651855469, 8.621719360351562, 17.44891357421875, 142.281494140625, -0.907440185546875, 59.190765380859375, -44.47065734863281, 77.16006469726562, 132.74102783203125, 9.18292236328125, 76.3065185546875, 1.1411762237548828, 83.852294921875, -4.580028533935547, 55.86480712890625, 93.76884460449219, 8.09649658203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000122.npy"}
{"epoch": 0.2554973821989529, "step": 123, "batch_size": 128, "mean": 25.55533218383789, "std": 44.51418685913086, "min": -144.80517578125, "p10": -20.034252929687497, "median": 21.71009063720703, "p90": 87.37093658447264, "max": 150.01791381835938, "pos_frac": 0.7265625, "sample": [55.87977600097656, -14.636016845703125, 102.54008483886719, -19.62347412109375, 35.84242248535156, 3.8571090698242188, 21.132461547851562, -3.1045684814453125, 25.760498046875, 22.175811767578125, -18.197528839111328, 3.11712646484375, 5.832489013671875, 12.411163330078125, 9.966156005859375, -11.2030029296875, -14.9349365234375, 16.43096923828125, 115.562744140625, 86.52549743652344, -50.980224609375, 5.684051513671875, 35.83882141113281, 16.076635360717773, 90.40048217773438, -18.812332153320312, 55.8818359375, 52.0408935546875, 68.70697021484375, -5.821998596191406, 40.222900390625, 44.69499206542969, 22.42828369140625, 13.7386474609375, 40.839599609375, 67.262451171875, 31.965789794921875, -144.80517578125, -2.885650634765625, 82.02243041992188, 23.54888916015625, -4.409393310546875, 1.7602996826171875, 44.58416748046875, 49.64007568359375, 99.283447265625, -10.867107391357422, 10.040214538574219, 36.875274658203125, 105.99993896484375, -8.706920623779297, 31.675018310546875, 68.1640625, 101.13433837890625, 98.06155395507812, 14.146804809570312, -7.19329833984375, 16.595947265625, -10.81390380859375, 85.87939453125, 45.0391845703125, 14.517623901367188, -114.51702880859375, 41.271240234375, 35.65806579589844, 10.3388671875, 65.20733642578125, 4.090545654296875, 89.3436279296875, 36.8319091796875, 55.7296142578125, -18.1107177734375, 12.273208618164062, -27.58867645263672, 36.55079650878906, 57.85008239746094, 21.58856201171875, -26.04913330078125, -37.923248291015625, 55.417724609375, -8.875656127929688, 38.07368469238281, 24.893321990966797, 7.2025146484375, 21.831619262695312, 56.775634765625, -37.42999267578125, 22.77947998046875, 59.355255126953125, 90.97796630859375, 0.0, -30.31855010986328, 150.01791381835938, 25.796981811523438, -28.164703369140625, 0.0, 26.01422119140625, 19.684799194335938, 3.159414291381836, 82.02537536621094, -8.086250305175781, -14.430145263671875, 19.144668579101562, 135.41412353515625, 22.069137573242188, 7.318550109863281, 41.767333984375, 74.33197021484375, 70.30494689941406, -25.637527465820312, -2.99725341796875, -20.99273681640625, 104.45147705078125, 31.345481872558594, -26.335601806640625, 83.92056274414062, 38.75262451171875, 5.154083251953125, 36.499755859375, 26.129928588867188, -11.996711730957031, 0.191925048828125, 18.02362060546875, 0.512969970703125, 111.5504150390625, -35.96710205078125, 15.51080322265625, 68.57955932617188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000123.npy"}
{"epoch": 0.25759162303664923, "step": 124, "batch_size": 128, "mean": 21.752338409423828, "std": 42.03032684326172, "min": -114.83758544921875, "p10": -25.89643783569336, "median": 19.98969268798828, "p90": 73.4637664794922, "max": 127.82470703125, "pos_frac": 0.7109375, "sample": [-1.52850341796875, 31.313735961914062, -97.8946533203125, 9.80999755859375, 12.727167129516602, 4.32861328125, 59.794158935546875, 61.667877197265625, 56.066680908203125, 106.82293701171875, 46.32305908203125, -30.46630859375, 22.614852905273438, -17.103973388671875, 27.09820556640625, 28.381141662597656, -1.9040374755859375, 56.33294677734375, -19.86967658996582, 39.09747314453125, 36.5, -25.318641662597656, 6.2166748046875, 67.32563781738281, 21.1142578125, -27.24462890625, -3.519561767578125, -8.611778259277344, 15.644622802734375, 18.865127563476562, 65.82563781738281, -54.540252685546875, -19.52605438232422, 54.717864990234375, 97.86566162109375, 85.1209716796875, 53.4650764465332, 88.1302490234375, 46.1396484375, 72.0670166015625, 60.051361083984375, 4.957496643066406, 98.66665649414062, 8.288604736328125, 28.09481430053711, 3.593250274658203, -21.410140991210938, 7.3467559814453125, 42.280364990234375, -33.653682708740234, -2.400909423828125, -51.65442657470703, 45.24811553955078, 0.0, -5.569520950317383, 2.2584915161132812, 31.548492431640625, 10.06378173828125, 32.628204345703125, -37.553985595703125, 57.329925537109375, -55.8681640625, -58.0694580078125, 56.8956298828125, 9.0257568359375, 39.521026611328125, 15.0225830078125, 102.46932983398438, 39.32456970214844, 26.09356689453125, 30.92181396484375, 43.206207275390625, -114.83758544921875, -56.6851806640625, 41.514244079589844, 18.60790252685547, 9.0213623046875, 34.407127380371094, 46.92010498046875, 41.21678161621094, 25.523284912109375, 78.58834838867188, -10.552200317382812, -19.47265625, -14.30303955078125, 13.422847747802734, -7.5854644775390625, 73.403076171875, 127.82470703125, 55.805545806884766, 21.744186401367188, 44.81096649169922, -8.675888061523438, 56.05293273925781, 9.025543212890625, 15.663566589355469, 12.0059814453125, 91.66134643554688, 29.539960861206055, 64.19300842285156, 5.054145812988281, -11.0611572265625, 73.318115234375, -1.21771240234375, 12.4954833984375, 104.74090576171875, -24.852035522460938, -36.99821472167969, 8.872520446777344, 7.118896484375, 22.061050415039062, -17.780136108398438, -13.42333984375, 6.75567626953125, 10.726547241210938, 54.656288146972656, 22.146087646484375, 103.55769348144531, -8.380195617675781, -38.130210876464844, 52.1075439453125, 66.04052734375, 85.01492309570312, 73.60537719726562, 29.406494140625, 24.302001953125, 3.9715499877929688, -21.15619659423828], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000124.npy"}
{"epoch": 0.25968586387434556, "step": 125, "batch_size": 128, "mean": 19.8980712890625, "std": 52.14939498901367, "min": -106.88485717773438, "p10": -42.29658050537109, "median": 14.948760986328125, "p90": 88.03409805297852, "max": 159.553466796875, "pos_frac": 0.6328125, "sample": [29.01922607421875, -24.822994232177734, -101.0405044555664, 59.84710693359375, 25.46941566467285, 49.94642639160156, 44.49275207519531, -16.207489013671875, -1.2596759796142578, -64.80087280273438, -21.586318969726562, 32.438575744628906, 66.26092529296875, 38.2384033203125, -1.105438232421875, 92.294677734375, -79.62030029296875, 31.801681518554688, 74.05624389648438, 10.569156646728516, -15.91973876953125, 13.297874450683594, 34.891357421875, -25.05133819580078, 4.2520751953125, 29.006576538085938, 27.649017333984375, -11.056671142578125, -22.507080078125, 159.553466796875, 11.00799560546875, -53.65721130371094, 41.9036865234375, -5.29847526550293, -8.901195526123047, 88.0277328491211, 31.02166748046875, 80.31497192382812, 20.425567626953125, -8.140144348144531, -49.1944694519043, -43.9222412109375, -23.970184326171875, -19.399566650390625, -35.30455017089844, 53.22918701171875, -17.35076904296875, -1.673492431640625, -64.50375366210938, 25.767730712890625, 48.2601318359375, -79.20062255859375, 63.96929931640625, 157.84521484375, 35.75823211669922, 13.028812408447266, 6.3205413818359375, 46.91316223144531, 63.206634521484375, 70.67947387695312, -18.560791015625, 80.9630126953125, 2.23785400390625, -1.7231388092041016, -64.2310791015625, 29.5911865234375, 0.44415283203125, 12.449913024902344, 1.5902099609375, 14.5755615234375, 126.9755859375, 39.555809020996094, 112.20809936523438, 119.85687255859375, -3.7687454223632812, -106.88485717773438, 37.987762451171875, 37.52116394042969, 80.10379028320312, 76.78805541992188, 34.411895751953125, 117.43402099609375, -14.020187377929688, 80.41433715820312, 5.3616943359375, 76.55300903320312, -30.557754516601562, -12.99322509765625, 46.14881896972656, 53.330108642578125, 65.70421600341797, 0.086181640625, -41.59986877441406, 0.16192626953125, -46.334075927734375, 18.199600219726562, -12.579345703125, 59.2210693359375, 25.211334228515625, -31.832008361816406, 2.9849700927734375, 0.0, 2.5615806579589844, -61.00811767578125, 88.0489501953125, -31.388458251953125, 20.395736694335938, -2.5931167602539062, 15.32196044921875, 105.865234375, -16.452980041503906, -30.7513427734375, 46.89715576171875, 132.89346313476562, 7.921714782714844, 52.913818359375, 90.4395751953125, 26.3638916015625, 34.82159423828125, 33.502197265625, 83.0831298828125, -93.18203735351562, -23.645957946777344, 107.1263427734375, 27.587387084960938, 102.53173828125, -14.906105041503906, -15.651153564453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000125.npy"}
{"epoch": 0.2617801047120419, "step": 126, "batch_size": 128, "mean": 22.455135345458984, "std": 46.01070022583008, "min": -121.5299072265625, "p10": -40.686466979980466, "median": 20.678482055664062, "p90": 85.5032257080078, "max": 145.76141357421875, "pos_frac": 0.7109375, "sample": [43.748138427734375, 69.39248657226562, 37.670753479003906, 31.70086669921875, 43.583251953125, 68.91143798828125, 13.298439025878906, 24.10052490234375, 7.146371841430664, 9.892364501953125, -65.0946044921875, 32.194366455078125, -3.537996292114258, 31.476043701171875, 53.16627883911133, 44.326690673828125, 31.365203857421875, 49.497772216796875, -42.319244384765625, -20.6044921875, 14.367424011230469, 133.53256225585938, -9.31134033203125, 0.26675987243652344, 20.97491455078125, -40.142181396484375, -4.875823974609375, -43.93036651611328, -7.205352783203125, 7.632080078125, -0.081756591796875, 28.867660522460938, 99.59213256835938, 0.854705810546875, 49.054779052734375, -48.294952392578125, -41.95646667480469, 23.93927001953125, -44.039825439453125, 1.9724884033203125, 51.252349853515625, 46.27679443359375, 41.08636474609375, -8.406967163085938, 33.144287109375, 86.25750732421875, 3.9072189331054688, 11.542510986328125, -23.94122314453125, -32.723785400390625, 68.58395385742188, -13.793701171875, 49.3245849609375, 145.76141357421875, 23.263736724853516, 20.382049560546875, 56.19837951660156, 15.673477172851562, -52.944000244140625, 16.693607330322266, 18.03424835205078, 50.70843505859375, 16.123779296875, 44.85174560546875, -43.76678466796875, 58.319305419921875, -2.7871170043945312, 14.891799926757812, 41.48545837402344, 105.84716796875, 59.34584045410156, 12.077423095703125, 30.925384521484375, 68.20513916015625, 6.615001678466797, 42.723541259765625, 16.83740234375, -1.280548095703125, -1.851327896118164, -16.3184814453125, 67.29582214355469, 83.73556518554688, -3.4096412658691406, 80.2340087890625, 101.55029296875, 32.61187744140625, -57.149749755859375, 18.71893310546875, 42.75732421875, -62.30572509765625, 116.61953735351562, -39.192779541015625, 63.2098388671875, 33.41246032714844, -94.61370849609375, 5.3783721923828125, 8.157466888427734, 30.951171875, -4.176239013671875, 3.130889892578125, 89.7828369140625, 34.327728271484375, 46.81629943847656, 88.22308349609375, -5.508270263671875, 94.59619140625, -15.514129638671875, 123.5885009765625, 46.060760498046875, 87.07443237304688, -121.5299072265625, 4.99737548828125, 72.9341049194336, 110.73348999023438, 26.74081802368164, 18.443450927734375, -7.3195953369140625, 0.86358642578125, 21.12493896484375, -48.981597900390625, 40.6436767578125, -29.291748046875, -5.17877197265625, 24.937652587890625, 38.20733642578125, 85.17996215820312, 1.4778366088867188, -35.7457275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000126.npy"}
{"epoch": 0.2638743455497382, "step": 127, "batch_size": 128, "mean": 26.494121551513672, "std": 49.177188873291016, "min": -129.1256103515625, "p10": -23.174453735351562, "median": 20.604084014892578, "p90": 97.27836990356445, "max": 178.3515625, "pos_frac": 0.6953125, "sample": [-9.504629135131836, 4.4021453857421875, -18.89276123046875, 28.47259521484375, 22.03460693359375, 14.708221435546875, 59.001708984375, 12.644668579101562, -29.591102600097656, -129.1256103515625, 30.799774169921875, -21.54088592529297, 68.59982299804688, 51.66279220581055, -5.7464141845703125, -63.02130126953125, 22.712310791015625, 133.23387145996094, -13.618606567382812, 20.68927001953125, 27.9151611328125, 11.525238037109375, 49.016510009765625, 16.898345947265625, 52.316497802734375, 95.91082763671875, 60.413818359375, 51.426788330078125, -6.01953125, 39.565032958984375, 51.792724609375, 1.6843986511230469, 142.26055908203125, 8.31903076171875, 9.46856689453125, 108.9619140625, -28.224708557128906, -28.89947509765625, -2.871795654296875, 19.350341796875, 103.68527221679688, -0.2851715087890625, -28.80072021484375, -8.8741455078125, -99.62162780761719, -34.36145782470703, 37.922943115234375, 66.33688354492188, 48.522247314453125, -22.74468994140625, 30.068099975585938, 30.23651123046875, 12.6429443359375, -19.81244659423828, 42.71246337890625, 57.73443603515625, 70.2425537109375, 35.74317932128906, -3.02471923828125, 20.518898010253906, -0.160125732421875, 7.493011474609375, 95.79139709472656, 21.46588134765625, 30.260162353515625, 85.0604248046875, -41.7481689453125, 10.775497436523438, 102.25604248046875, 85.3729248046875, 29.959075927734375, -2.8426666259765625, -7.878379821777344, -3.4960403442382812, 21.665985107421875, -117.45448303222656, 0.94189453125, -15.591758728027344, 25.157791137695312, -11.885871887207031, 18.64764404296875, 30.170013427734375, 70.83905029296875, 13.381534576416016, 96.63174438476562, 71.2874755859375, 11.21002197265625, 178.3515625, 103.541748046875, 71.04791259765625, 100.53375244140625, -23.293121337890625, 55.82110595703125, -15.378173828125, 27.092330932617188, -7.485071182250977, -17.5895938873291, 21.419830322265625, 16.267345428466797, 16.746612548828125, 24.779190063476562, -2.3046875, -13.30474853515625, 20.044296264648438, 98.78716278076172, 10.557106018066406, 150.96142578125, 110.99911499023438, -2.611572265625, -23.12359619140625, 11.492431640625, 23.6444091796875, 51.3516845703125, 101.57647705078125, -31.305519104003906, 16.184188842773438, 66.60089111328125, 57.01202392578125, -20.012306213378906, 5.089324951171875, 69.08651733398438, 40.93756103515625, 16.80975341796875, 64.06689453125, 38.197357177734375, -29.718505859375, 77.85282897949219, 105.63934326171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000127.npy"}
{"epoch": 0.26596858638743454, "step": 128, "batch_size": 128, "mean": 30.295085906982422, "std": 50.65873336791992, "min": -98.83001708984375, "p10": -29.559535980224606, "median": 30.423072814941406, "p90": 95.42489013671874, "max": 155.29339599609375, "pos_frac": 0.7421875, "sample": [53.673187255859375, 76.83908081054688, 87.10496520996094, 32.6583251953125, -94.49200439453125, -65.87361145019531, -28.1275634765625, 45.05973815917969, 89.01629638671875, 44.46954345703125, 87.96151733398438, -14.40313720703125, 92.44807434082031, 11.655181884765625, 149.03018188476562, 7.464822769165039, 9.355873107910156, 61.6822509765625, 58.91896057128906, 71.88154602050781, -19.874786376953125, -11.820472717285156, 93.46524047851562, 113.92422485351562, 21.15045166015625, 30.709808349609375, -17.587005615234375, 25.421295166015625, 49.78668212890625, 40.172760009765625, 24.33110809326172, 13.997148513793945, -52.239105224609375, 101.43336486816406, -12.188880920410156, 99.47303771972656, -37.876434326171875, 14.215888977050781, -16.5802001953125, 31.142608642578125, 48.9775390625, -61.511016845703125, -24.490646362304688, 24.808685302734375, 109.05132293701172, -40.70643615722656, 13.331840515136719, 36.48247528076172, 56.4178466796875, 77.86691284179688, 34.48614501953125, 68.42037963867188, 11.008399963378906, 153.52572631835938, -7.58612060546875, -17.458969116210938, 8.630859375, 35.73435592651367, 62.54644775390625, 76.50927734375, 58.94525909423828, 30.136337280273438, 3.5097732543945312, 34.78016662597656, 3.4202423095703125, 19.377960205078125, 86.569091796875, 62.034576416015625, 35.30670166015625, 9.497695922851562, 17.936813354492188, 71.54627990722656, 63.3487548828125, 42.2081298828125, 85.95681762695312, 61.50408935546875, 35.115562438964844, -36.39894104003906, 1.977294921875, 17.872024536132812, 97.9549560546875, 39.32696533203125, -25.219512939453125, 38.411773681640625, 72.22076416015625, -98.83001708984375, 53.154998779296875, -50.41322326660156, 51.19291687011719, 94.340576171875, -28.789337158203125, -46.09844970703125, 40.44830322265625, 22.75360107421875, -18.88861083984375, 0.6007003784179688, -21.97967529296875, -9.191436767578125, 22.64056396484375, 12.680618286132812, 114.24850463867188, -31.356666564941406, 11.6805419921875, 106.8919677734375, 68.56623077392578, 84.92889404296875, -18.998435974121094, 39.408538818359375, 109.19976806640625, 5.984954833984375, 21.28411865234375, 0.0, -84.87127685546875, -17.998916625976562, 138.62741088867188, -18.366744995117188, -51.908111572265625, 26.138427734375, 39.07926940917969, 52.287933349609375, 4.67230224609375, 26.80670166015625, 57.88127136230469, 103.33367919921875, 155.29339599609375, 37.07456970214844, 17.91961669921875, -4.4232025146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000128.npy"}
{"epoch": 0.2680628272251309, "step": 129, "batch_size": 128, "mean": 27.243267059326172, "std": 50.70091247558594, "min": -109.93714904785156, "p10": -25.28871459960937, "median": 23.542755126953125, "p90": 89.46893310546875, "max": 237.53662109375, "pos_frac": 0.7265625, "sample": [-20.8929443359375, 88.1241455078125, -44.535491943359375, 21.87237548828125, -2.0496597290039062, 90.0316162109375, 63.072296142578125, 100.28581237792969, 100.28982543945312, 26.176849365234375, 60.137969970703125, 8.868766784667969, 28.65234375, 11.0384521484375, -4.8116607666015625, -1.5391082763671875, 54.79231262207031, 91.64030456542969, 9.593582153320312, -38.955116271972656, 40.273406982421875, 68.13178253173828, -11.048152923583984, 23.78948974609375, 67.32333374023438, -109.93714904785156, 78.58992004394531, 1.785980224609375, 237.53662109375, 89.227783203125, 52.83251953125, 85.56069946289062, 132.8582000732422, 29.26611328125, 31.880096435546875, -99.15829467773438, 69.60525512695312, 38.4752197265625, 155.13983154296875, 33.194053649902344, 17.288421630859375, 19.852996826171875, 49.87164306640625, 54.09490966796875, 0.4583892822265625, 94.60629272460938, -20.259429931640625, 10.905540466308594, 2.3061485290527344, 23.876731872558594, 71.405517578125, 34.17185974121094, 37.00355529785156, 85.73355102539062, -6.79736328125, 0.10249519348144531, -1.4404754638671875, 52.94915771484375, 19.395034790039062, 48.56526184082031, 8.24490737915039, 16.815658569335938, 55.35284423828125, 74.07396697998047, 115.97918701171875, -1.97955322265625, 21.183700561523438, 41.76568603515625, 42.5303955078125, -67.83926391601562, 6.737060546875, 24.22380828857422, 5.2729949951171875, -5.5150909423828125, 19.92840576171875, -68.37753295898438, 70.0510482788086, 38.464881896972656, -1.9070281982421875, 107.53070068359375, 64.33323669433594, 11.4774169921875, 5.0643310546875, 35.89928436279297, 6.884283065795898, -92.8919677734375, 82.08970642089844, -28.9532470703125, 29.051513671875, 134.36795043945312, -31.938079833984375, -7.31451416015625, -12.745864868164062, -0.402313232421875, -14.576156616210938, 78.833984375, -4.600921630859375, -53.9652099609375, -43.58268737792969, 23.2960205078125, 100.58584594726562, 8.54998779296875, -9.336669921875, -11.167289733886719, -83.3573226928711, -45.330291748046875, 10.277511596679688, -7.05877685546875, 27.951324462890625, 35.001708984375, 49.48130798339844, 16.029407501220703, 37.1861572265625, 26.951171875, -2.0799598693847656, 92.41021728515625, 7.4862060546875, 78.42805480957031, -23.71820068359375, 29.06561279296875, 28.254791259765625, 10.379156112670898, 14.212760925292969, 50.005584716796875, 2.6043853759765625, -2.4979019165039062, 57.89239501953125, 54.8579216003418], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000129.npy"}
{"epoch": 0.27015706806282724, "step": 130, "batch_size": 128, "mean": 27.08936309814453, "std": 49.64395523071289, "min": -94.82351684570312, "p10": -33.177206420898436, "median": 23.025299072265625, "p90": 91.93963623046875, "max": 167.011474609375, "pos_frac": 0.6953125, "sample": [49.626556396484375, 8.29400634765625, 43.77098083496094, -21.33978271484375, 29.140838623046875, 32.46440124511719, 43.070465087890625, 70.67752075195312, 78.39025115966797, -40.621856689453125, -7.144245147705078, 112.0511474609375, 56.511871337890625, 60.181793212890625, -14.97113037109375, 20.110595703125, 2.00189208984375, -40.46022033691406, 11.305030822753906, 14.69036865234375, 19.310531616210938, 69.81954956054688, -25.939586639404297, 62.07829284667969, -4.707004547119141, 51.77363586425781, -0.6875, 89.49700927734375, 123.31683349609375, 78.00439453125, -69.8719482421875, 13.516448974609375, 0.0, -16.118545532226562, 50.088043212890625, 31.0748291015625, 60.36077880859375, -66.75161743164062, -21.52698516845703, 91.44451904296875, 77.90176391601562, 16.617645263671875, 87.460693359375, -7.4864349365234375, -75.33502197265625, -94.82351684570312, -6.3491668701171875, 102.78829956054688, -6.751068115234375, -3.06829833984375, 74.76799011230469, -40.743072509765625, 17.988929748535156, 65.32711791992188, -61.48040771484375, -25.385284423828125, 49.04839324951172, 3.850067138671875, 96.517578125, -42.0887451171875, -29.922561645507812, 29.728466033935547, 51.72906494140625, 67.92686462402344, 44.549713134765625, 130.2853546142578, 40.23883056640625, -19.3638916015625, -34.54693603515625, 9.758209228515625, 93.89727783203125, 9.662223815917969, 108.99053955078125, 8.99615478515625, 1.1912994384765625, 58.221588134765625, -5.251701354980469, 41.19073486328125, 13.605239868164062, 33.936920166015625, 86.63371276855469, 40.86492919921875, 127.8731689453125, 28.5579833984375, 29.042755126953125, 1.470102310180664, 22.19207763671875, -0.587646484375, 106.464599609375, -6.783317565917969, 6.08367919921875, 41.33953857421875, 4.894683837890625, 27.411460876464844, 52.87469482421875, -66.71697998046875, -29.431228637695312, 13.74090576171875, 3.72235107421875, 89.80609130859375, 167.011474609375, 83.2086181640625, 138.3634796142578, 16.638839721679688, -76.14044189453125, 61.0098876953125, 23.10443878173828, 47.866455078125, 39.93670654296875, -6.7340850830078125, 55.060211181640625, -2.7621212005615234, -42.733001708984375, -12.227249145507812, 33.54780578613281, 51.3153076171875, 130.2862548828125, 46.69207763671875, -2.7150630950927734, 46.752044677734375, 22.94615936279297, -24.07220458984375, 10.281768798828125, 38.270469665527344, 40.5972900390625, -32.590179443359375, 15.962116241455078, 93.09490966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000130.npy"}
{"epoch": 0.27225130890052357, "step": 131, "batch_size": 128, "mean": 21.811826705932617, "std": 53.98527145385742, "min": -120.05615234375, "p10": -47.74461364746094, "median": 23.076183319091797, "p90": 96.07093505859375, "max": 147.65719604492188, "pos_frac": 0.6796875, "sample": [37.78334045410156, 10.491424560546875, 66.40826416015625, -39.42034912109375, 36.49067687988281, 64.16706085205078, 18.04729652404785, 39.79292297363281, -70.676513671875, 145.77642822265625, 39.43013000488281, 70.0576171875, 127.6568603515625, 42.01348876953125, -80.10931396484375, -21.081344604492188, 5.6646728515625, -104.5413818359375, 32.22706604003906, 110.33612060546875, 29.33917236328125, 6.08587646484375, -14.43646240234375, -3.360952377319336, 64.3235092163086, 1.146148681640625, -17.92266845703125, 67.77740478515625, 39.106292724609375, 64.65615844726562, 46.69066619873047, 102.572509765625, 32.06520080566406, 0.9732818603515625, -3.32501220703125, 2.9426345825195312, -66.4140625, 76.866455078125, 0.01171875, 134.271728515625, 6.2144775390625, 30.86223602294922, 18.253459930419922, -34.64715576171875, 63.1416015625, 10.103660583496094, 138.87057495117188, -48.84039306640625, 43.69044494628906, 28.76214599609375, -1.3398818969726562, 5.207000732421875, -32.23028564453125, 36.111083984375, 102.62648010253906, 98.34896850585938, -60.9066162109375, -30.39764404296875, 60.847198486328125, 118.60629272460938, -19.684310913085938, -33.320343017578125, -20.42901611328125, 52.931854248046875, 36.1317138671875, -11.37435531616211, 11.66748046875, 89.1568603515625, 118.74169921875, 32.386627197265625, -120.05615234375, -59.71296691894531, 61.0787353515625, 4.062103271484375, 1.3852310180664062, 24.351455688476562, -98.43386840820312, 147.65719604492188, -26.738479614257812, 56.41413879394531, 45.35198974609375, -51.565582275390625, 0.18184661865234375, 38.638336181640625, 54.11273193359375, 70.67092895507812, -37.6763916015625, 65.94345092773438, 95.09463500976562, -30.016586303710938, 23.41301727294922, -13.84674072265625, 22.739349365234375, -3.0194835662841797, 46.206878662109375, 24.48993682861328, -3.6562576293945312, 27.20807647705078, 5.5435943603515625, 37.219886779785156, 18.72979736328125, -46.168548583984375, 69.6002197265625, 10.465019226074219, 112.05583190917969, 69.3289794921875, -1.397674560546875, -63.03619384765625, 105.832275390625, 13.661529541015625, 69.7362060546875, 85.35308837890625, 36.011993408203125, 37.6976318359375, -84.5546875, -3.96051025390625, 51.2373046875, -40.675689697265625, 54.32887268066406, 12.70843505859375, 50.37213134765625, 43.684547424316406, 19.333251953125, -12.700576782226562, -16.182342529296875, -55.40641784667969, -47.274993896484375, -7.282508850097656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000131.npy"}
{"epoch": 0.2743455497382199, "step": 132, "batch_size": 128, "mean": 30.582592010498047, "std": 56.06938171386719, "min": -128.81668090820312, "p10": -37.30238494873047, "median": 23.43506622314453, "p90": 96.0305908203125, "max": 177.88665771484375, "pos_frac": 0.7109375, "sample": [-6.107421875, 27.27294921875, 70.67434692382812, 79.25911712646484, 9.467269897460938, -37.568634033203125, 64.66158294677734, 13.801162719726562, -26.314697265625, 106.75860595703125, 37.63182067871094, -56.28998565673828, -42.2232666015625, 67.46527099609375, 60.43000793457031, -29.4136962890625, -1.4932861328125, 56.697509765625, 71.37393188476562, 174.35659790039062, 86.34915161132812, 79.50399780273438, 81.4273681640625, 96.6929931640625, -0.05224609375, 6.311981201171875, 3.8662490844726562, -84.09658813476562, 40.711273193359375, 69.32827758789062, 0.862823486328125, 155.58419799804688, 36.26544189453125, 14.92901611328125, -45.825927734375, -33.43341064453125, 28.150501251220703, 42.054595947265625, -8.6572265625, 10.31610107421875, 161.8519287109375, 74.32833862304688, -37.18827819824219, -77.80667114257812, -16.330829620361328, 1.64752197265625, 37.861602783203125, -0.06256103515625, 6.829620361328125, 22.62909698486328, 67.58158874511719, -4.672981262207031, 50.550445556640625, 91.5367431640625, 25.994827270507812, 4.675018310546875, -9.056549072265625, 66.78515625, 155.760009765625, -3.5682144165039062, 68.850830078125, 36.210174560546875, 39.12266540527344, -42.7369384765625, 76.50743103027344, 14.874755859375, 3.1444664001464844, 22.9686279296875, -22.180145263671875, 32.378692626953125, 21.099822998046875, 10.738914489746094, 22.367294311523438, 124.36279296875, -27.28961181640625, 21.180130004882812, 0.0, 95.7467041015625, 23.901504516601562, 7.029296875, -0.3967437744140625, 76.56005859375, 89.19486236572266, -21.755325317382812, -128.81668090820312, 102.51092529296875, 17.943618774414062, 111.7432861328125, -16.292724609375, 29.779083251953125, 53.49493408203125, 22.93377685546875, 66.59768676757812, 40.00238037109375, 82.16365051269531, 177.88665771484375, -21.7015380859375, 88.187744140625, -2.70111083984375, 7.682830810546875, 61.005157470703125, 70.79312133789062, 9.87481689453125, -1.2024612426757812, 59.03253173828125, 123.43008422851562, 18.15447998046875, 31.564727783203125, 0.0, -41.7774658203125, 74.9024658203125, 96.71755981445312, -30.207000732421875, 29.657623291015625, -86.91972351074219, 40.9920654296875, 1.25775146484375, 85.89215087890625, -53.19425964355469, 46.519195556640625, 63.18439483642578, 145.27713012695312, -97.25617980957031, 3.475311279296875, -53.479644775390625, 75.30787658691406, 10.509857177734375, 87.69204711914062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000132.npy"}
{"epoch": 0.2764397905759162, "step": 133, "batch_size": 128, "mean": 28.45734405517578, "std": 51.36540985107422, "min": -114.6754150390625, "p10": -25.182954216003417, "median": 23.810951232910156, "p90": 96.25709838867186, "max": 147.8929443359375, "pos_frac": 0.6953125, "sample": [41.6119384765625, -25.47715187072754, -58.39385986328125, 0.0, 36.752838134765625, 39.199951171875, -0.1635894775390625, 15.00185775756836, -7.781269073486328, 26.470550537109375, 57.3538818359375, -17.968780517578125, -11.53314208984375, 57.388702392578125, 68.83242797851562, 147.8929443359375, 0.4646759033203125, 79.8140869140625, 11.753517150878906, -23.881500244140625, 60.750885009765625, 117.1324462890625, 122.5240478515625, -15.028060913085938, 44.350341796875, 82.75045776367188, 42.599708557128906, 104.88818359375, 0.0, 14.40264892578125, 98.93408203125, 34.96551513671875, -12.857751846313477, 3.086944580078125, 77.54031372070312, 125.57455444335938, 23.52105712890625, 45.07745361328125, -59.715606689453125, 43.05029296875, 79.32704162597656, 23.63800048828125, -21.472442626953125, -23.979217529296875, 13.401397705078125, -20.643524169921875, -96.52995300292969, 20.335830688476562, -7.770538330078125, 25.7137451171875, 54.614410400390625, 66.02325439453125, -5.0716552734375, 2.2872467041015625, 115.27743530273438, -10.71453857421875, 23.983901977539062, 146.12841796875, 8.08270263671875, 87.84231567382812, -34.368133544921875, 35.34172439575195, 42.7486572265625, -11.52783203125, 56.78733825683594, 1.3147354125976562, 58.00679016113281, 97.4871826171875, 85.75164794921875, 78.56338500976562, -48.25567626953125, -21.443572998046875, 46.677703857421875, 4.219970703125, 135.03424072265625, 4.745386123657227, -7.099357604980469, -50.069732666015625, 10.510116577148438, 131.47100830078125, 8.066673278808594, -25.056869506835938, 44.34123229980469, -60.78515625, 89.54132080078125, 10.607513427734375, 2.576690673828125, 59.45372009277344, 11.22332763671875, 47.557769775390625, -15.340576171875, 95.72991943359375, 28.338043212890625, 39.825653076171875, -5.966011047363281, -23.6258544921875, 19.170013427734375, 49.481475830078125, 88.42636108398438, 93.23490905761719, 8.132125854492188, 117.44734191894531, 38.22991943359375, 71.59053802490234, -32.8072509765625, 11.637752532958984, -3.764129638671875, 67.49722290039062, -49.063880920410156, 117.29107666015625, 27.79107666015625, 35.62969970703125, 15.16973876953125, -19.604248046875, -74.36347961425781, 81.29107666015625, -114.6754150390625, 26.512714385986328, -12.724021911621094, 43.673126220703125, 4.61724853515625, 51.204864501953125, 68.78634643554688, -45.16545104980469, -8.29339599609375, 4.285926818847656, 81.89202880859375, 86.26627349853516], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000133.npy"}
{"epoch": 0.27853403141361255, "step": 134, "batch_size": 128, "mean": 21.43670654296875, "std": 46.86186218261719, "min": -124.76348876953125, "p10": -33.02749557495117, "median": 15.921783447265625, "p90": 76.85419464111328, "max": 169.94277954101562, "pos_frac": 0.7109375, "sample": [62.50688171386719, 62.70215606689453, -62.85313415527344, -15.89593505859375, -62.099334716796875, 6.03289794921875, 40.98870849609375, 14.53924560546875, -14.498260498046875, 35.01202392578125, -124.76348876953125, -21.478439331054688, 13.41290283203125, 48.92950439453125, 78.26063537597656, 1.868661880493164, 9.189624786376953, 69.0238037109375, 89.12884521484375, 5.101404190063477, 51.40350341796875, 72.39779663085938, 0.0, 107.263916015625, 70.12158203125, 61.651702880859375, 42.48388671875, -9.26531982421875, 43.85302734375, 100.15556335449219, -45.77031707763672, -7.123992919921875, 42.601409912109375, 12.253864288330078, 44.401344299316406, -24.57318115234375, 47.977294921875, 45.255332946777344, 76.25143432617188, 58.194793701171875, -57.73744201660156, 41.82969665527344, 34.1119384765625, 43.60467529296875, 169.94277954101562, 29.508209228515625, 5.912200927734375, -25.785888671875, 125.52276611328125, 13.181121826171875, 18.42718505859375, 72.64000701904297, 57.108062744140625, -34.31208038330078, 17.84430694580078, -19.681800842285156, 50.143218994140625, 78.95735168457031, 96.71565246582031, 13.340072631835938, -49.96482849121094, -9.184356689453125, 11.42791748046875, -35.1282958984375, -75.0845947265625, 142.38998413085938, -12.52777099609375, -0.248931884765625, 29.136993408203125, 50.63292694091797, 1.2290973663330078, 12.030975341796875, -32.02009582519531, -4.962032318115234, 13.834014892578125, 100.45210266113281, 0.554229736328125, 1.3095703125, 9.901464462280273, -73.63772583007812, 74.127685546875, 23.083969116210938, 32.253684997558594, 2.64788818359375, 5.215087890625, 4.86553955078125, 43.311676025390625, 53.52659606933594, 79.49603271484375, 5.907060623168945, -9.978645324707031, 19.82867431640625, 26.073348999023438, 36.28277587890625, -44.99591064453125, 49.12522888183594, 43.79254150390625, 35.8919677734375, 12.772468566894531, -70.14399719238281, 38.62639617919922, 71.38031005859375, -32.476959228515625, -99.10662841796875, -8.852981567382812, 33.53045654296875, -13.20440673828125, -0.18536376953125, 43.55218505859375, 2.050506591796875, 1.3902645111083984, 62.57594299316406, 26.91132354736328, 81.1788330078125, 55.0123291015625, 87.06318664550781, 46.26606750488281, -14.704727172851562, 6.5904541015625, -16.2744140625, 17.3043212890625, 13.40374755859375, -28.54931640625, -11.34320068359375, 57.3343505859375, 59.00733947753906, 6.279815673828125, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000134.npy"}
{"epoch": 0.2806282722513089, "step": 135, "batch_size": 128, "mean": 25.045719146728516, "std": 52.447715759277344, "min": -182.2353057861328, "p10": -29.17824516296387, "median": 19.456443786621094, "p90": 103.13248291015626, "max": 137.14309692382812, "pos_frac": 0.7109375, "sample": [-4.428279876708984, -12.673690795898438, 4.467998504638672, 67.75758361816406, -22.529449462890625, 137.14309692382812, 2.499948501586914, 4.685546875, 63.95664978027344, 16.142980575561523, 111.55670166015625, 35.5419921875, -41.95880126953125, -24.841033935546875, 46.79681396484375, 28.62445068359375, 21.51757049560547, 7.3316650390625, 22.271942138671875, 15.032630920410156, 35.670135498046875, 25.969058990478516, -4.708168029785156, 13.019210815429688, -42.310089111328125, -0.8664016723632812, 47.547454833984375, 97.8643798828125, 45.178863525390625, 84.60696411132812, 23.87481689453125, 4.237480163574219, 65.17605590820312, 5.8990478515625, 3.659637451171875, 104.46157836914062, 0.6346282958984375, -7.336090087890625, 38.858917236328125, 3.26715087890625, 12.025680541992188, 32.481597900390625, 37.30769348144531, 74.02090454101562, 126.72979736328125, 27.421722412109375, 113.15557861328125, 49.5841064453125, -0.78387451171875, -39.4931640625, 12.411163330078125, 51.624298095703125, 7.924171447753906, -11.3614501953125, -9.542800903320312, -29.37884521484375, 104.29080200195312, 17.189117431640625, 58.225311279296875, 22.4403076171875, 94.3793716430664, 20.37933349609375, 60.397071838378906, 12.932708740234375, 11.151901245117188, 40.224517822265625, 24.872802734375, 46.531463623046875, 70.13885498046875, 14.864402770996094, 56.13200378417969, 32.049598693847656, 103.10186767578125, 12.87054443359375, 49.68666076660156, 58.93196105957031, 42.235321044921875, -11.734649658203125, -6.674896240234375, 74.65878295898438, 11.871109008789062, -38.18171691894531, -3.950031280517578, 2.263427734375, -49.33465576171875, 62.909027099609375, -12.040252685546875, 118.03253173828125, 39.979583740234375, 130.92184448242188, 34.323760986328125, 103.20391845703125, -26.471588134765625, -150.13528442382812, 134.58612060546875, -21.912826538085938, 39.715423583984375, 54.10289001464844, -34.85595703125, 100.45205688476562, 19.296066284179688, 19.6168212890625, 4.3253936767578125, 132.72601318359375, -6.2074737548828125, -182.2353057861328, -29.092273712158203, -0.9545860290527344, 107.0396728515625, 65.63760375976562, 33.92730712890625, -19.187606811523438, 4.6058349609375, -7.215911865234375, 0.0, 28.633224487304688, -42.934722900390625, -113.06443786621094, -4.330986022949219, 82.62644958496094, 6.259307861328125, 103.35592651367188, 10.175323486328125, 83.87521362304688, 41.308349609375, -77.58384704589844, -34.69525146484375, -2.535858154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000135.npy"}
{"epoch": 0.28272251308900526, "step": 136, "batch_size": 128, "mean": 28.557849884033203, "std": 56.8984375, "min": -117.755615234375, "p10": -27.05443572998047, "median": 25.25253677368164, "p90": 92.17042236328125, "max": 223.331298828125, "pos_frac": 0.7109375, "sample": [49.6427001953125, 19.10858917236328, 25.912734985351562, 39.09485626220703, 108.61699676513672, 87.38418579101562, 92.158447265625, 56.60804748535156, 31.115402221679688, 71.98809814453125, 4.551513671875, 4.57550048828125, 75.35610961914062, 14.214553833007812, 63.31370162963867, 79.09423828125, 72.30599975585938, 47.0780029296875, 6.348388671875, -25.227867126464844, -103.78866577148438, 14.981941223144531, -23.25933837890625, 5.1800994873046875, -43.907867431640625, -4.701332092285156, -27.3914794921875, -86.89239501953125, 30.777633666992188, 21.259490966796875, 8.930614471435547, -10.02001953125, 9.498199462890625, 2.3409423828125, 63.3795166015625, -3.950927734375, -25.07012939453125, 28.452301025390625, 36.81304931640625, -6.260223388671875, 16.059051513671875, -26.453292846679688, 92.1983642578125, -109.29473876953125, 53.422149658203125, -15.019157409667969, 12.41851806640625, -3.8201751708984375, 49.464111328125, 17.019332885742188, 67.1922607421875, -98.3583984375, 139.44610595703125, 65.8045654296875, 129.8077392578125, 185.59780883789062, -25.487091064453125, 21.508102416992188, 1.54327392578125, 79.78498840332031, 155.60211181640625, -1.6517333984375, 92.5909423828125, 25.31256866455078, 26.93359375, -28.543655395507812, -49.52177429199219, 28.34429931640625, 90.51889038085938, -24.468231201171875, 31.512405395507812, -6.47503662109375, 14.41495132446289, 59.45672607421875, 79.83760070800781, 158.42181396484375, -4.895668029785156, 38.672515869140625, 65.3923110961914, -57.79754638671875, 54.79559326171875, -55.1983642578125, 43.493682861328125, 133.76943969726562, 72.92971801757812, 61.5382080078125, 19.967575073242188, -9.391447067260742, 8.087182998657227, 87.6063232421875, -4.011432647705078, 22.00244140625, -26.909988403320312, 39.20672607421875, 8.497222900390625, -24.04875946044922, 120.58712768554688, 3.5318374633789062, 40.493255615234375, 67.32553100585938, -1.1198959350585938, 39.65596008300781, 25.1925048828125, 10.144832611083984, 111.790771484375, -117.755615234375, -18.388870239257812, 56.46223449707031, 97.6895751953125, 65.3480224609375, 86.97076416015625, 34.16480255126953, -49.843231201171875, -8.595367431640625, 12.757888793945312, 32.44146728515625, 223.331298828125, -2.585613250732422, -22.4197998046875, 49.66337585449219, 39.10980224609375, 66.44059753417969, 72.90010070800781, 0.3217315673828125, 12.626663208007812, -93.8450927734375, 52.044677734375, 60.52735900878906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000136.npy"}
{"epoch": 0.2848167539267016, "step": 137, "batch_size": 128, "mean": 24.885669708251953, "std": 53.65616226196289, "min": -94.9351806640625, "p10": -38.24596557617187, "median": 16.360008239746094, "p90": 111.7396484375, "max": 165.2734375, "pos_frac": 0.734375, "sample": [-83.666015625, 7.413848876953125, 47.81982421875, 8.355148315429688, 92.33128356933594, 31.647361755371094, 165.2734375, -37.57080078125, 16.048477172851562, 6.40911865234375, 20.60572052001953, 23.720748901367188, 32.78857421875, 142.05328369140625, 10.144332885742188, 1.926788330078125, -13.836318969726562, 103.841064453125, -12.51702880859375, -39.82135009765625, 59.702484130859375, 30.261795043945312, 20.38775634765625, -6.75531005859375, -94.9351806640625, 12.794807434082031, -5.592864990234375, -51.42362976074219, 81.19181823730469, -5.550422668457031, 131.7935791015625, 19.572265625, 57.89967346191406, 134.50405883789062, -32.2957763671875, -22.9334716796875, 45.647125244140625, 22.498008728027344, 11.614501953125, -33.82749938964844, 13.191436767578125, 2.1683616638183594, 15.790840148925781, -10.36053466796875, -52.53558349609375, 21.450469970703125, 34.73844909667969, -22.86383056640625, -70.017578125, 10.391185760498047, -20.99932861328125, 111.5179443359375, 9.531417846679688, 17.24188232421875, 28.599212646484375, 72.15481567382812, 9.255340576171875, 87.3197021484375, -28.47528076171875, 10.151718139648438, -11.042816162109375, 0.36721038818359375, 4.8978271484375, 60.603363037109375, 1.1546630859375, 113.68572998046875, -60.735107421875, 1.3946819305419922, 4.255859375, -55.433685302734375, 4.8876953125, 29.32110595703125, 108.79412841796875, -35.41206359863281, 118.96233367919922, 45.889404296875, -3.7303123474121094, 112.2569580078125, 77.36351013183594, 138.374755859375, 16.671539306640625, 124.805419921875, 17.248123168945312, 122.54034423828125, 13.255706787109375, 49.505889892578125, 86.32208251953125, 37.031219482421875, 18.553085327148438, 95.16659545898438, 6.800140380859375, -30.216079711914062, -8.61724853515625, 16.892181396484375, 2.891876220703125, 148.46157836914062, 113.485595703125, 11.432804107666016, 40.904876708984375, 77.51536560058594, -63.9892578125, -57.23321533203125, 66.35946655273438, -90.97508239746094, 0.065582275390625, 113.06709289550781, -28.08224868774414, 43.773406982421875, 57.205596923828125, 14.714622497558594, 30.150054931640625, 51.92401123046875, -42.278900146484375, 48.082855224609375, 55.788330078125, 74.93406677246094, -24.404159545898438, 9.585617065429688, -67.36038208007812, 13.460365295410156, -6.784217834472656, 20.28969955444336, 28.414947509765625, 44.78794860839844, 61.475677490234375, 42.717041015625, 30.242095947265625, 7.156303405761719], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000137.npy"}
{"epoch": 0.2869109947643979, "step": 138, "batch_size": 128, "mean": 28.920318603515625, "std": 61.38934326171875, "min": -105.65042114257812, "p10": -47.06645126342774, "median": 28.661720275878906, "p90": 106.38112716674803, "max": 178.66107177734375, "pos_frac": 0.6484375, "sample": [178.66107177734375, 47.43731689453125, 70.13796997070312, -30.49280548095703, -36.20623779296875, 32.64227294921875, 31.59490966796875, 135.3136749267578, -7.93524169921875, 116.57467651367188, 29.633407592773438, 4.06640625, 100.5596923828125, 78.53524780273438, 12.232522964477539, -23.38335418701172, -64.44586181640625, 33.336669921875, -56.698760986328125, -47.52906799316406, -2.699310302734375, 53.20635986328125, -68.77122497558594, 18.28338623046875, -90.42752075195312, 51.134735107421875, 28.5093994140625, 66.65917205810547, -2.2367477416992188, 141.75131225585938, -41.87481689453125, -6.146026611328125, 16.65301513671875, 25.67486572265625, 6.666749954223633, 41.06977844238281, -42.652374267578125, 92.94281005859375, 66.96417236328125, -22.470321655273438, 57.198516845703125, -4.045146942138672, 104.81354522705078, 119.14886474609375, 29.840179443359375, 100.92222595214844, -3.43487548828125, 18.793933868408203, 39.131561279296875, -15.77398681640625, 37.69459533691406, 41.23223876953125, 89.45919799804688, -34.41578674316406, 125.27944946289062, 33.69105529785156, 98.8294677734375, 92.96516418457031, 59.16904067993164, -40.2603759765625, -63.24632263183594, 15.207763671875, 0.404022216796875, -14.966880798339844, 102.92303466796875, -80.07369995117188, 96.54876708984375, 78.54182434082031, 64.69863891601562, -9.098876953125, -58.27716064453125, 41.845458984375, -24.591659545898438, 113.196533203125, 35.89259338378906, 16.218826293945312, -51.82286071777344, 80.34503173828125, 134.4566192626953, 0.4224853515625, 60.00233459472656, 76.98309326171875, 83.68376159667969, 62.551849365234375, -102.21453857421875, -34.02314758300781, 9.647216796875, -14.819313049316406, -7.268177032470703, -39.623046875, 63.480865478515625, 101.57293701171875, 172.29696655273438, 70.7984619140625, -12.07623291015625, -15.9212646484375, 33.75749206542969, 51.115997314453125, -25.700469970703125, -21.023025512695312, 110.038818359375, 3.462249755859375, 175.89495849609375, -46.868186950683594, 17.816429138183594, 10.844970703125, -24.932144165039062, -63.483001708984375, -42.17076110839844, 90.47615051269531, -7.108795166015625, 77.3001708984375, 6.4063720703125, 123.8189697265625, 7.309906005859375, 46.43588638305664, 151.15219116210938, -59.609130859375, 93.17913818359375, -105.65042114257812, 20.625385284423828, 69.92100524902344, 66.3531494140625, -3.21966552734375, 84.636474609375, 30.724014282226562, 28.814041137695312, -38.72453308105469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000138.npy"}
{"epoch": 0.28900523560209423, "step": 139, "batch_size": 128, "mean": 20.569684982299805, "std": 53.47947692871094, "min": -90.87643432617188, "p10": -48.04597473144531, "median": 16.522438049316406, "p90": 87.0692657470703, "max": 161.82171630859375, "pos_frac": 0.625, "sample": [-74.60870361328125, 27.785110473632812, -54.395111083984375, 32.393951416015625, -40.738067626953125, 53.4132080078125, 39.367286682128906, 45.14312744140625, 15.892410278320312, -16.763568878173828, 72.470458984375, 10.332778930664062, 93.68798828125, 33.272216796875, -47.651519775390625, 85.490234375, 106.63082885742188, 13.568679809570312, 3.4078903198242188, 4.853054046630859, 6.1510009765625, 74.98835754394531, 14.259109497070312, 56.20057678222656, -12.98486328125, 18.133697509765625, -2.66241455078125, 65.03521728515625, -6.107021331787109, -80.78448486328125, 6.040693283081055, -5.788352966308594, 86.5286865234375, 87.94338989257812, 49.871002197265625, -1.4691486358642578, 37.454010009765625, -4.30535888671875, -1.182769775390625, 98.89393615722656, -40.898155212402344, -15.325607299804688, 2.405975341796875, -16.9813232421875, 29.222270965576172, 20.47039794921875, -90.822509765625, -48.96636962890625, -15.506629943847656, 83.4652099609375, 7.2010498046875, -7.670989990234375, -55.157196044921875, -55.989959716796875, 86.34783172607422, 94.44122314453125, -31.282073974609375, 17.1524658203125, -64.8946533203125, -31.93426513671875, 86.69464111328125, -12.50706672668457, 53.7181396484375, 27.768173217773438, 86.53289794921875, -9.63751220703125, 7.225988388061523, -3.379791259765625, 30.12261962890625, 82.06968688964844, 4.851676940917969, 56.09662628173828, 27.79681396484375, 137.37753295898438, 32.701820373535156, -71.04531860351562, 100.61944580078125, 134.52215576171875, 48.29840087890625, 76.74105834960938, 58.8917236328125, 17.77728271484375, 56.51832580566406, 66.16427612304688, 55.15753173828125, -27.795326232910156, 126.30078125, -69.37835693359375, 44.01646423339844, -8.250585556030273, -6.6513671875, 116.14541625976562, 30.5418701171875, 40.9110107421875, 73.09564208984375, 50.517425537109375, 14.154159545898438, -55.34584045410156, -38.101043701171875, 3.28155517578125, 86.67294311523438, 25.009033203125, -17.016632080078125, 22.39264488220215, -30.343353271484375, -41.05859375, 57.54893493652344, 147.11907958984375, 15.249122619628906, 84.90817260742188, 161.82171630859375, 44.003692626953125, -24.85247802734375, -42.4488525390625, -70.97821044921875, -90.87643432617188, 0.0, 17.70556640625, -12.037155151367188, 107.29045867919922, -37.86688995361328, -45.19903564453125, 12.755203247070312, -18.283782958984375, 25.636566162109375, -4.89703369140625, 37.5029296875, 23.59893035888672], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000139.npy"}
{"epoch": 0.29109947643979056, "step": 140, "batch_size": 128, "mean": 31.077434539794922, "std": 50.27806854248047, "min": -106.7275390625, "p10": -29.556124114990233, "median": 26.975509643554688, "p90": 96.85001678466797, "max": 160.8956298828125, "pos_frac": 0.7265625, "sample": [21.40289306640625, 86.34515380859375, 131.8575897216797, -44.17485046386719, -17.97527313232422, 54.03460693359375, -25.878219604492188, 39.04786682128906, 1.285308837890625, -7.624839782714844, -25.361289978027344, -23.899658203125, 44.17962646484375, 51.52256774902344, 15.509284973144531, 70.45690155029297, 1.857147216796875, 91.22821044921875, 104.28945922851562, -8.39984130859375, 29.86468505859375, -40.04759216308594, 59.23622131347656, -58.6702880859375, -14.7747802734375, 53.61224365234375, -106.7275390625, -2.0605316162109375, 67.28297424316406, 87.73526763916016, 118.28485107421875, 75.88177490234375, 108.4345703125, 97.01866149902344, -65.75857543945312, 62.611419677734375, 48.73114013671875, -1.481170654296875, 101.51216125488281, 97.15603637695312, -9.231666564941406, -62.480743408203125, 44.79472351074219, 4.704891204833984, 28.598419189453125, 37.78851318359375, 96.77774047851562, 31.51544189453125, -39.869171142578125, 160.8956298828125, 21.359214782714844, 61.113853454589844, 106.39765930175781, 2.787994384765625, 0.19239044189453125, 14.03106689453125, 34.957305908203125, -3.3629074096679688, 12.380111694335938, 17.880569458007812, 70.77899169921875, -3.9619140625, -30.402931213378906, 61.75201416015625, -12.1517333984375, 74.08489990234375, 20.299823760986328, -15.6190185546875, 0.0, -85.69233703613281, 5.12158203125, -56.151763916015625, 130.14596557617188, 33.275604248046875, 32.152069091796875, 69.10244750976562, 66.49618530273438, 91.62542724609375, 8.230369567871094, 81.84481811523438, 27.1402587890625, 10.927093505859375, -1.118356704711914, 87.77267456054688, 83.82467651367188, 57.30338668823242, 70.67633056640625, 12.829486846923828, 64.66937255859375, 62.230621337890625, 18.529380798339844, 0.9427337646484375, 79.52259826660156, 12.894271850585938, 1.8910751342773438, 89.84461975097656, 0.0, -2.647653579711914, 50.525146484375, -32.213523864746094, 1.430816650390625, 103.05889892578125, -29.193206787109375, 52.53192138671875, 15.604400634765625, -24.958290100097656, 65.452880859375, 10.495330810546875, 22.295639038085938, 23.313674926757812, 89.67486572265625, 4.2215576171875, 93.11773681640625, 13.311195373535156, 68.65382385253906, 17.830245971679688, 37.7550048828125, -8.664764404296875, 27.360000610351562, -70.05290985107422, 46.04412841796875, 146.79592895507812, 26.810760498046875, -4.34735107421875, 36.21368408203125, -32.12261962890625, 114.45794677734375, 55.570648193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000140.npy"}
{"epoch": 0.2931937172774869, "step": 141, "batch_size": 128, "mean": 32.050987243652344, "std": 58.07513427734375, "min": -166.86642456054688, "p10": -38.251984405517575, "median": 30.778900146484375, "p90": 103.48312988281249, "max": 151.9569854736328, "pos_frac": 0.7421875, "sample": [117.01206970214844, 25.4749755859375, -48.760032653808594, -23.855499267578125, 57.6112060546875, 58.04393768310547, 15.883270263671875, 74.40963745117188, -77.15875244140625, 143.52569580078125, 5.89398193359375, 41.84442138671875, 93.37391662597656, 145.43280029296875, 50.971649169921875, 85.59170532226562, 31.801345825195312, 81.37521362304688, 56.31708526611328, 91.61679077148438, 105.322998046875, 21.767333984375, 34.67955780029297, 108.80288696289062, -18.56242561340332, -34.41733169555664, 88.54732513427734, 21.119247436523438, 16.792037963867188, 149.6397705078125, 15.38262939453125, -32.3985595703125, -37.423004150390625, 104.49664306640625, 30.37109375, 90.5726318359375, -38.65203094482422, -81.01958465576172, 11.668548583984375, -38.080535888671875, 124.68756103515625, 54.2882080078125, 58.796043395996094, -13.843597412109375, -56.27170181274414, 52.15850830078125, 62.190093994140625, -17.160606384277344, 77.8349609375, 18.911575317382812, 30.093795776367188, 61.46051025390625, 16.84063720703125, 14.958724975585938, 88.98681640625, 38.755615234375, -17.488338470458984, 12.869918823242188, 51.21040344238281, -21.80755615234375, 83.43707275390625, 88.52786254882812, -38.82208251953125, -55.38677978515625, -11.87286376953125, 96.141845703125, 21.592559814453125, 33.15180969238281, -14.842170715332031, 86.60054016113281, 7.160308837890625, 37.98365783691406, 27.147871017456055, 73.00225067138672, 45.252655029296875, 74.35919189453125, 124.30224609375, -19.2630615234375, 0.0, 1.7645301818847656, 85.09014892578125, -15.528289794921875, 4.62005615234375, 103.04876708984375, -7.97210693359375, 151.9569854736328, -166.86642456054688, 21.05078125, 2.842132568359375, 31.699371337890625, 31.18670654296875, 17.43700408935547, 12.836509704589844, -152.3105010986328, -57.573974609375, -9.54217529296875, 16.157997131347656, 136.5417022705078, 32.97807312011719, 49.480255126953125, -72.91737365722656, 56.14654541015625, 58.61235046386719, 89.06597900390625, 42.48345947265625, 10.527421951293945, 76.98052978515625, 99.16400146484375, 70.18869018554688, -48.61322021484375, -0.9985790252685547, 9.15753173828125, 122.28826904296875, -70.76347351074219, 31.781387329101562, 16.023239135742188, 55.004180908203125, 25.029739379882812, 87.42584228515625, 22.433998107910156, -30.04498291015625, 70.0863037109375, 135.390380859375, -24.468048095703125, 43.874664306640625, 66.94015502929688, 5.9523773193359375, 25.918319702148438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000141.npy"}
{"epoch": 0.29528795811518327, "step": 142, "batch_size": 128, "mean": 26.73249626159668, "std": 59.700496673583984, "min": -160.85523986816406, "p10": -44.669258117675774, "median": 17.8192138671875, "p90": 105.69256591796872, "max": 167.09548950195312, "pos_frac": 0.6953125, "sample": [20.64276123046875, -6.158348083496094, 31.541580200195312, 32.073577880859375, 17.891860961914062, -21.70721435546875, 4.776153564453125, -10.330108642578125, 113.18902587890625, 80.04017639160156, 101.56515502929688, -41.78871154785156, -36.778900146484375, 63.91717529296875, 41.28190612792969, 62.00048828125, -37.5712890625, 1.7632598876953125, 150.18487548828125, 44.68878173828125, 62.848114013671875, 22.1566162109375, 0.0, 96.03340148925781, 37.46331787109375, 50.379364013671875, 24.006195068359375, 0.922393798828125, -0.3694725036621094, -0.2227783203125, -4.933502197265625, 37.77197265625, -6.757652282714844, -51.8302001953125, 6.067047119140625, 3.187793731689453, 21.395782470703125, 48.389366149902344, -52.816253662109375, 55.235809326171875, 18.710693359375, 113.30453491210938, 124.22283935546875, 33.872650146484375, 40.60504150390625, -81.03675842285156, 10.359479904174805, 113.4112548828125, 43.50782012939453, 102.030029296875, 10.32440185546875, -14.343194961547852, 32.008209228515625, 32.48020935058594, -18.531784057617188, 6.7772216796875, -1.69091796875, 86.44381713867188, -26.43133544921875, -27.128860473632812, 9.846649169921875, 77.22479248046875, -28.182022094726562, -73.65966796875, 143.51992797851562, 80.61676025390625, 72.8935546875, -33.337249755859375, 88.39002990722656, 4.947479248046875, 6.38348388671875, -27.423431396484375, 13.634063720703125, 9.955968856811523, -56.56866455078125, 55.27507019042969, -53.854888916015625, 59.26386260986328, 76.19088745117188, -51.390533447265625, -110.52230834960938, 102.47979736328125, 115.677490234375, 44.979827880859375, -14.465042114257812, 33.353248596191406, 16.33074951171875, 135.90240478515625, 65.63253021240234, 15.258323669433594, -14.99479866027832, 6.8813934326171875, 89.6671142578125, 87.55189514160156, 124.8951416015625, 6.426490783691406, 101.16525268554688, 7.9691009521484375, 86.49908447265625, 167.09548950195312, -79.54769897460938, 10.21337890625, -71.46847534179688, 1.6688308715820312, 12.038848876953125, -71.01751708984375, 119.30039978027344, 17.408470153808594, 63.99871826171875, -26.683242797851562, 147.43743896484375, -121.78408813476562, 17.746566772460938, 50.465057373046875, 64.16726684570312, 5.520782470703125, 47.3721923828125, -2.9382705688476562, 30.969635009765625, -26.61456298828125, -160.85523986816406, 100.53363037109375, -0.5005550384521484, -5.9372100830078125, 9.637939453125, 55.06285095214844, 161.33828735351562, 73.6720962524414], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000142.npy"}
{"epoch": 0.2973821989528796, "step": 143, "batch_size": 128, "mean": 32.3115348815918, "std": 53.434757232666016, "min": -115.05072021484375, "p10": -26.3854736328125, "median": 28.37578582763672, "p90": 112.79147949218749, "max": 208.02587890625, "pos_frac": 0.71875, "sample": [47.78400421142578, 125.05426025390625, 120.80169677734375, 53.78778076171875, 119.06764221191406, 17.52886962890625, 53.53901672363281, 14.98822021484375, -22.925369262695312, -3.1976566314697266, 63.918182373046875, -12.250335693359375, 59.27496337890625, 20.97357177734375, -6.851081848144531, 11.396514892578125, 105.0865478515625, 42.34228515625, -30.948143005371094, 120.68026733398438, -36.78094482421875, 2.153228759765625, 46.292724609375, 97.73399353027344, 38.33270263671875, 49.39344787597656, 49.793609619140625, 126.739013671875, 208.02587890625, -40.9080810546875, 0.10888671875, 65.04043579101562, 138.0851593017578, 82.201904296875, 38.367279052734375, -25.8956298828125, -96.45849609375, -5.725181579589844, 71.01315307617188, -1.45867919921875, 14.798797607421875, -36.42413330078125, 12.600364685058594, 89.97222900390625, 30.820842742919922, -11.404296875, 17.17609977722168, -12.688980102539062, 23.00244903564453, -115.05072021484375, 46.321044921875, 57.581787109375, 35.07233428955078, 78.22956848144531, 42.13043212890625, 49.68328857421875, -6.40081787109375, 31.84754753112793, 1.6846771240234375, 2.1211624145507812, -7.652984619140625, 4.287586212158203, 164.12210083007812, 60.9005126953125, 7.058784484863281, 4.3191375732421875, 80.0994873046875, 52.37164306640625, 89.36724853515625, -10.1912841796875, 62.122100830078125, 87.14381408691406, 23.735923767089844, 4.388648986816406, 6.264286041259766, -4.6954345703125, -56.669677734375, 42.84794616699219, 13.986953735351562, 63.67279815673828, 12.851959228515625, -1.7413330078125, 8.816818237304688, 62.75187683105469, 18.008392333984375, -16.28765869140625, -17.183639526367188, -7.981517791748047, 6.295745849609375, 29.125839233398438, 63.54853439331055, -0.066741943359375, 18.85040283203125, 51.6015625, 68.59869384765625, 42.13787841796875, -0.21620941162109375, 125.7769775390625, -17.939058303833008, 50.8076171875, 120.11654663085938, 59.90777587890625, -27.53326416015625, 34.6842041015625, -27.5284423828125, 150.42657470703125, 128.55372619628906, -3.8933467864990234, 2.324626922607422, 15.265579223632812, -30.4422607421875, 62.88214111328125, -11.992263793945312, 113.54608154296875, 43.822662353515625, 112.46807861328125, 53.347991943359375, 90.87277221679688, -83.75173950195312, 36.94532012939453, -61.626617431640625, 57.3629150390625, -70.6131591796875, -5.315521240234375, 69.90078735351562, 27.625732421875, 31.94635009765625, 4.158056259155273], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000143.npy"}
{"epoch": 0.2994764397905759, "step": 144, "batch_size": 128, "mean": 26.48564910888672, "std": 59.540321350097656, "min": -148.69537353515625, "p10": -42.851571655273425, "median": 24.602035522460938, "p90": 103.17300109863281, "max": 267.55535888671875, "pos_frac": 0.6953125, "sample": [0.3702678680419922, 66.12996673583984, 65.44015502929688, 42.82250213623047, 9.096969604492188, 44.99102783203125, -12.798736572265625, -82.74252319335938, 0.30716705322265625, 155.18740844726562, -2.9679718017578125, 7.0668792724609375, 24.769866943359375, 116.23858642578125, 12.258865356445312, 64.68832397460938, 22.123008728027344, 77.53445434570312, 25.173851013183594, 3.570465087890625, -24.2762451171875, 60.5325927734375, 61.598175048828125, -0.3669891357421875, 9.05908203125, -9.453857421875, 28.381629943847656, -8.57938003540039, 39.85736083984375, -15.443975448608398, -76.475830078125, 10.827009201049805, 15.9178466796875, 187.894775390625, 150.03143310546875, 85.80287170410156, 65.94644165039062, 145.243408203125, -7.651588439941406, 75.1719970703125, -8.81220817565918, -78.51313781738281, 38.98747253417969, 10.58673095703125, -5.372186660766602, -65.11185455322266, -23.342071533203125, 38.153411865234375, -54.74810791015625, -51.346038818359375, 21.125808715820312, 99.39093017578125, 59.12694549560547, -148.69537353515625, -108.50836181640625, -68.258544921875, 32.835693359375, -21.13372802734375, -28.600555419921875, 31.65325927734375, 43.759185791015625, 44.43994140625, -35.84185791015625, -4.125030517578125, 54.79734802246094, 13.3294677734375, 69.478515625, 134.02093505859375, 55.66349792480469, 80.40850830078125, 14.805938720703125, 32.40532684326172, 145.601318359375, 26.90631103515625, 32.44158935546875, 72.20819091796875, 0.74609375, 54.068397521972656, 50.835601806640625, 267.55535888671875, 36.460235595703125, 41.33168411254883, -23.28485107421875, -53.08221435546875, 110.43171691894531, 71.74981689453125, 1.09881591796875, 46.679931640625, 0.2635040283203125, 120.54812622070312, 89.41680908203125, 62.899505615234375, 38.58589172363281, 2.0619125366210938, 2.9477901458740234, 108.572509765625, 15.805267333984375, -4.63134765625, -15.42352294921875, 58.4923095703125, -23.591064453125, -2.772062301635742, 0.0, -28.755264282226562, -53.52070617675781, -51.06134033203125, 0.41900634765625, -87.37275695800781, 41.18975830078125, 39.087589263916016, 58.281768798828125, 15.040435791015625, 102.86544799804688, -23.325592041015625, 49.85595703125, 2.165679931640625, 38.347900390625, 55.631385803222656, 35.5341796875, -12.193511962890625, 37.199798583984375, 40.72967529296875, 4.5666046142578125, 103.890625, -39.333099365234375, -10.523712158203125, 106.2481689453125, 24.4342041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000144.npy"}
{"epoch": 0.30157068062827225, "step": 145, "batch_size": 128, "mean": 34.678043365478516, "std": 56.68185043334961, "min": -115.56735229492188, "p10": -25.652963256835935, "median": 36.224151611328125, "p90": 104.81713104248047, "max": 189.607177734375, "pos_frac": 0.71875, "sample": [-1.524078369140625, 57.46199035644531, -115.56735229492188, 30.236663818359375, 0.9091873168945312, -34.7294921875, 74.28303527832031, 49.07159423828125, 117.70834350585938, -72.1964111328125, 1.3330078125, 124.53330993652344, 34.481300354003906, 101.93635559082031, 16.0472412109375, 55.71868896484375, 28.9891357421875, 49.997772216796875, -11.29595947265625, 9.41448974609375, -23.317222595214844, -17.179779052734375, 60.9552001953125, 21.06182289123535, -0.03079986572265625, 96.86654663085938, -55.217529296875, 125.78402709960938, -8.41583251953125, 44.26953125, 4.10675048828125, 73.383056640625, 43.494354248046875, 79.05415344238281, 183.1571044921875, 6.796295166015625, 129.45278930664062, -78.90192413330078, 15.60406494140625, 2.115081787109375, 44.325653076171875, 24.53485107421875, 6.6198883056640625, 174.61260986328125, 141.68988037109375, 35.47705078125, 105.57736206054688, 85.99310302734375, 55.624168395996094, 42.853515625, 65.45891571044922, -30.887557983398438, 24.444656372070312, 104.49131774902344, 66.77813720703125, 91.71142578125, -23.9119873046875, -9.593708038330078, -75.87554931640625, 13.21406364440918, 38.1561279296875, 77.12826538085938, -44.1014404296875, 27.042999267578125, 116.58599853515625, -15.782279968261719, 69.44561767578125, -23.698822021484375, 43.0428466796875, 59.59545135498047, 69.58013916015625, -27.453399658203125, 43.28209686279297, 66.831298828125, 76.8707275390625, 93.3876953125, 14.069744110107422, 51.31523132324219, -23.477745056152344, 29.716766357421875, -84.42108917236328, 61.94923400878906, 46.40346908569336, 36.97125244140625, 178.82403564453125, 26.45888900756836, 40.514244079589844, 89.70701599121094, -52.88587951660156, -15.085166931152344, 1.5468978881835938, 0.0, 64.83956909179688, 38.37493896484375, 35.04558563232422, 10.545265197753906, 93.39595031738281, 127.91836547851562, 60.631744384765625, 55.521202087402344, -51.961734771728516, -93.072509765625, 13.759979248046875, 38.873634338378906, 33.750396728515625, -23.7855224609375, 63.73650360107422, 41.54791259765625, -1.58526611328125, -3.138559341430664, -24.88134765625, 130.946044921875, 68.05331420898438, 47.82159423828125, -15.172607421875, -7.09368896484375, 47.51979064941406, 47.15547180175781, 74.44705200195312, 72.90667724609375, 29.304473876953125, 55.95137023925781, -7.960210800170898, -5.530311584472656, 189.607177734375, -10.1826171875, 15.622589111328125, -8.627227783203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000145.npy"}
{"epoch": 0.3036649214659686, "step": 146, "batch_size": 128, "mean": 34.039329528808594, "std": 57.4796142578125, "min": -129.63189697265625, "p10": -34.34421997070312, "median": 25.110153198242188, "p90": 103.38761291503906, "max": 207.80984497070312, "pos_frac": 0.7421875, "sample": [58.86931610107422, 7.620677947998047, 7.815093994140625, -24.464202880859375, 16.5877685546875, 58.646026611328125, -18.198776245117188, 24.349655151367188, 88.21636962890625, -78.8583984375, 144.89157104492188, 101.01974487304688, 38.3692626953125, -38.325714111328125, 51.372039794921875, 14.51641845703125, 186.88949584960938, 34.363006591796875, 19.476757049560547, 8.811065673828125, 60.965118408203125, 47.07819366455078, 66.97399139404297, -47.627227783203125, -45.309417724609375, -1.257598876953125, 19.34579849243164, 57.185028076171875, 41.323036193847656, 27.6597900390625, 13.365028381347656, 9.164249420166016, 77.5009765625, 26.588165283203125, -9.0362548828125, 75.29507446289062, 80.73857116699219, 56.06404113769531, 77.91268920898438, 18.216773986816406, 4.546722412109375, 94.69009399414062, 2.7665023803710938, 207.80984497070312, 0.0, 103.04345703125, -3.0133056640625, -8.134405136108398, 66.24037170410156, 71.5238037109375, 8.767990112304688, -37.004852294921875, 55.342769622802734, 24.321853637695312, 124.361572265625, 126.16171264648438, -129.63189697265625, 75.62657165527344, 79.66595458984375, 111.64563751220703, 22.39288330078125, 61.798370361328125, 17.891578674316406, 74.1435546875, 38.680999755859375, -33.203948974609375, -31.7471923828125, 28.462448120117188, 10.77532958984375, -74.41453552246094, 84.72561645507812, 5.031951904296875, -7.01080322265625, 183.3086395263672, -32.78656005859375, 148.12948608398438, 75.96429443359375, 4.665679931640625, -6.446136474609375, 68.24105834960938, 3.550323486328125, 23.07659912109375, 5.6843109130859375, -21.98046875, 54.9530029296875, 9.733980178833008, -8.161285400390625, 25.662628173828125, 0.0, 5.369682312011719, -43.94236755371094, -3.6924896240234375, 22.323165893554688, -58.868408203125, 23.886962890625, 110.95596313476562, -39.502716064453125, 47.08013916015625, 19.13378143310547, 36.9407958984375, -30.649078369140625, -1.490386962890625, 39.622337341308594, 29.6561279296875, 94.40017700195312, -8.693817138671875, 98.21514892578125, 45.174346923828125, 76.48439025878906, 153.2476348876953, 104.19064331054688, -102.3582763671875, 44.448638916015625, 145.76495361328125, 28.7314453125, 85.595703125, -40.828216552734375, 122.03109741210938, 13.168167114257812, 74.38473510742188, -12.060791015625, 19.657943725585938, 55.51068115234375, 95.4813232421875, 36.40264892578125, -71.91759490966797, 24.55767822265625, 54.656982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000146.npy"}
{"epoch": 0.3057591623036649, "step": 147, "batch_size": 128, "mean": 26.7027530670166, "std": 60.65266799926758, "min": -162.38095092773438, "p10": -35.39232635498046, "median": 23.45578384399414, "p90": 105.88370971679684, "max": 189.26025390625, "pos_frac": 0.671875, "sample": [-24.080596923828125, -2.88232421875, -10.901809692382812, 26.815277099609375, -16.30506134033203, -26.93157958984375, 148.6024627685547, -31.237258911132812, 0.0, 130.13507080078125, -15.817794799804688, 23.31591033935547, 24.962738037109375, 38.1904296875, 8.605169296264648, 19.117755889892578, -12.84771728515625, 75.18792724609375, 64.6468505859375, -17.61871337890625, 11.1370849609375, 40.689056396484375, 72.5246810913086, 71.56121826171875, -8.189476013183594, 165.4058380126953, 54.06635284423828, 98.47511291503906, 43.05487060546875, 6.922882080078125, 119.453369140625, 21.759521484375, 15.837532043457031, 138.13394165039062, 24.92559814453125, -53.736968994140625, -38.15068054199219, -24.9775390625, 11.2659912109375, 53.568641662597656, 28.086669921875, 79.11749267578125, 33.45147705078125, 54.792213439941406, -33.368160247802734, 56.074951171875, -14.35784912109375, -1.8394088745117188, 7.986242294311523, 45.43012237548828, -27.598236083984375, 148.55352783203125, 23.595657348632812, 57.521453857421875, 102.95394897460938, 12.028839111328125, 189.26025390625, 34.78166198730469, 74.46609497070312, 27.406631469726562, 47.595947265625, 79.310791015625, 117.48234558105469, 48.73590087890625, 131.87322998046875, -21.57659912109375, 24.44598388671875, 139.37014770507812, 75.48110961914062, 9.877876281738281, -93.15072631835938, 9.87799072265625, -19.6834716796875, -50.18994140625, -75.048095703125, -18.520706176757812, 26.763952255249023, 54.03278350830078, 0.1490345001220703, 0.8826446533203125, 62.34953308105469, 0.18070220947265625, 12.881195068359375, 80.02527618408203, 59.974937438964844, 60.430389404296875, 6.713653564453125, 58.8427734375, -59.462799072265625, 123.6520004272461, 55.401336669921875, 68.6595458984375, -64.27810668945312, -98.26028442382812, -159.09375, -2.3499603271484375, 76.00784301757812, 125.1494140625, -0.575225830078125, -162.38095092773438, -33.25128936767578, 13.018795013427734, 112.71981811523438, 15.36328125, -0.660491943359375, -20.430801391601562, 30.631622314453125, -11.446014404296875, -47.55812072753906, -24.649917602539062, -81.24173736572266, 74.16104125976562, -68.00505065917969, 55.00888442993164, 48.84056091308594, 14.793365478515625, 35.51014709472656, 77.817138671875, 59.930511474609375, -34.210174560546875, -6.143890380859375, 81.7923583984375, 99.34963989257812, 94.3751220703125, 86.11480712890625, 4.6238555908203125, -19.781192779541016, 6.674957275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000147.npy"}
{"epoch": 0.3078534031413613, "step": 148, "batch_size": 128, "mean": 28.292701721191406, "std": 65.34690856933594, "min": -102.00747680664062, "p10": -47.362895202636714, "median": 15.738476753234863, "p90": 108.79282836914061, "max": 256.74737548828125, "pos_frac": 0.625, "sample": [59.34088134765625, 38.82463073730469, 41.96406555175781, 4.41558837890625, -13.159774780273438, 81.92317199707031, 0.9409217834472656, 63.16326904296875, 106.591552734375, 54.05912780761719, 7.587522506713867, 49.95182800292969, -70.30648803710938, 146.45413208007812, -7.711639404296875, 103.75328063964844, 89.522216796875, 8.84991455078125, 122.31137084960938, 12.546707153320312, -80.153564453125, 132.21014404296875, -92.59297943115234, -1.0991134643554688, -41.44671630859375, -75.36322021484375, 40.50489807128906, -26.36121368408203, 30.250244140625, -37.76976013183594, 15.128921508789062, 31.8533935546875, -50.319618225097656, -102.00747680664062, -46.79541015625, 24.670928955078125, 205.06793212890625, -24.391571044921875, -99.40414428710938, 71.489501953125, 256.74737548828125, 1.2907371520996094, -52.446807861328125, -1.10345458984375, -1.186553955078125, -4.832456588745117, -62.19458770751953, -41.415618896484375, 54.57098388671875, 71.53897094726562, 13.775333404541016, 29.43273162841797, 15.628541946411133, 86.66461181640625, -40.180938720703125, 180.24569702148438, 46.66151428222656, 63.20256042480469, 83.27420043945312, 196.5107421875, 69.69052124023438, 88.54247283935547, -0.8206405639648438, -3.694732666015625, 0.0, 49.29082489013672, 127.32208251953125, 30.401824951171875, 22.186317443847656, 51.40985107421875, 8.51254653930664, -25.387588500976562, -28.670196533203125, -28.276260375976562, 101.05567932128906, -0.20557403564453125, -15.948562622070312, 92.51754760742188, -15.738067626953125, 99.85386657714844, 3.0735397338867188, 3.6828765869140625, 21.8389892578125, 36.2386474609375, 156.39666748046875, 54.120792388916016, 3.1059112548828125, 27.1728515625, -32.52345275878906, -78.33273315429688, 68.04866027832031, 15.848411560058594, 8.064229965209961, -0.1634063720703125, 100.10873413085938, 38.89741516113281, -48.68702697753906, -7.008888244628906, -44.85779571533203, 95.18362426757812, -26.876205444335938, 72.32133483886719, 126.84881591796875, 39.44970703125, -7.7746734619140625, 2.304088592529297, 30.772464752197266, -29.908050537109375, 91.568359375, 59.3753662109375, 77.5594253540039, 35.10108184814453, 49.844390869140625, 105.2401123046875, -66.43157958984375, -17.60478973388672, -87.310302734375, -5.12548828125, 119.05044555664062, 0.0, 1.54730224609375, 60.3560791015625, 78.41421508789062, -12.535453796386719, -16.0858154296875, 113.92913818359375, 125.4019775390625, -10.893333435058594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000148.npy"}
{"epoch": 0.3099476439790576, "step": 149, "batch_size": 128, "mean": 38.4803352355957, "std": 54.289161682128906, "min": -88.70999145507812, "p10": -26.486218261718747, "median": 34.353233337402344, "p90": 110.22623977661132, "max": 181.86767578125, "pos_frac": 0.734375, "sample": [91.76116943359375, -19.682891845703125, 42.59405517578125, 67.70650482177734, 137.17672729492188, 83.858642578125, 30.53265380859375, 3.8645172119140625, -0.6461181640625, 16.855361938476562, 82.98370361328125, -17.2152099609375, 107.44003295898438, 79.72754669189453, 38.938720703125, -1.5703277587890625, 58.2025146484375, 3.726776123046875, -41.2598876953125, -36.741844177246094, 80.20899963378906, 35.334808349609375, 90.04391479492188, 35.555755615234375, 112.8641586303711, -16.399993896484375, -4.835710525512695, 21.793487548828125, 88.94308471679688, -0.27175140380859375, 47.70780944824219, 53.51141357421875, 31.193389892578125, 14.190475463867188, 0.4646015167236328, -5.3165740966796875, 113.78262329101562, -33.80644989013672, 7.390869140625, 3.7558364868164062, 76.35455322265625, 160.12042236328125, 8.868682861328125, -21.723838806152344, 64.49465942382812, 93.1788330078125, 42.798187255859375, -13.407455444335938, 69.13897705078125, 38.953094482421875, 28.392913818359375, -8.45989990234375, -5.178680419921875, 50.88798522949219, -38.38587951660156, 107.102783203125, -41.7108154296875, 31.77423095703125, 83.17478942871094, -3.56732177734375, 16.37991714477539, 95.71717834472656, -62.055450439453125, 140.53439331054688, -14.50677490234375, 30.15167236328125, 8.01190185546875, -21.0975341796875, 33.41815185546875, 22.634357452392578, 52.14826202392578, 94.78628540039062, 22.580387115478516, 34.313995361328125, 61.5771484375, 46.18830871582031, 91.64181518554688, 37.633941650390625, 164.5450439453125, -18.500534057617188, 69.29278564453125, 26.449317932128906, 115.38204956054688, 108.1937255859375, 78.9476318359375, 47.5279541015625, 127.42169189453125, 10.094390869140625, -88.70999145507812, 19.076492309570312, 16.221878051757812, 89.93595886230469, -8.275344848632812, -30.61004638671875, 36.55030822753906, -22.72735595703125, 109.095703125, 16.92218017578125, 107.55232238769531, 96.33027648925781, -14.59576416015625, -35.3038330078125, 62.491363525390625, -4.5103759765625, 181.86767578125, 53.238861083984375, -69.5289306640625, 18.794265747070312, 34.39247131347656, 123.1072998046875, 41.2130126953125, -25.97418212890625, 53.724609375, 43.17042541503906, 0.723052978515625, 36.7264404296875, 118.31217956542969, 151.02601623535156, -36.92497634887695, -76.8782958984375, 31.194580078125, 62.87445068359375, 12.7794189453125, 149.2864990234375, 74.2462158203125, 27.373580932617188, -27.68096923828125, 52.393951416015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000149.npy"}
{"epoch": 0.31204188481675393, "step": 150, "batch_size": 128, "mean": 29.768699645996094, "std": 57.03544235229492, "min": -109.4830322265625, "p10": -42.45194244384765, "median": 25.814241409301758, "p90": 104.29613342285155, "max": 186.05386352539062, "pos_frac": 0.703125, "sample": [18.93096923828125, 59.30615234375, -7.420967102050781, 119.97879028320312, 61.698097229003906, -76.16653442382812, 15.733848571777344, 2.06829833984375, -47.07635498046875, 129.56576538085938, 48.61553192138672, 72.65261840820312, 76.8685302734375, 1.75830078125, -45.256439208984375, 11.78125, 28.508743286132812, 65.25279998779297, 7.8616943359375, 103.71524047851562, 67.35202026367188, -18.070297241210938, 7.7801513671875, 12.269012451171875, 7.386884689331055, 1.399658203125, -49.95220947265625, -39.082733154296875, 93.36648559570312, 43.79472351074219, 93.6318359375, -13.776191711425781, 33.977294921875, 38.67951965332031, 45.145538330078125, 106.76608276367188, 93.54632568359375, -84.36422729492188, 79.8232421875, 63.41461181640625, 35.399444580078125, 66.61198425292969, 11.924468994140625, 97.13705444335938, -7.8452911376953125, -62.240440368652344, -38.62810516357422, 13.11297607421875, 78.25091552734375, 118.49786376953125, 49.93651580810547, 170.63003540039062, 34.888580322265625, 56.308990478515625, 112.63067626953125, 46.899688720703125, 64.89202880859375, 23.918832778930664, -62.533782958984375, -17.616960525512695, -16.882598876953125, -11.804241180419922, -104.44195556640625, 128.82403564453125, 116.37939453125, 47.28349304199219, -13.312332153320312, 7.54241943359375, 71.60020446777344, 49.05133056640625, 58.733428955078125, 91.4302978515625, 18.735626220703125, -8.395721435546875, 52.33588409423828, -41.25001525878906, -38.01568603515625, 89.91876220703125, 24.743595123291016, -9.055343627929688, 78.25860595703125, -76.17068481445312, -19.996490478515625, -16.25726318359375, -1.6723098754882812, 76.32818603515625, -109.4830322265625, 186.05386352539062, 44.6712646484375, 119.077880859375, 7.085296630859375, -28.3314208984375, 13.640792846679688, 50.34038543701172, -74.58108520507812, -9.426460266113281, 51.158416748046875, -7.5726318359375, -65.46707153320312, -3.232837677001953, 12.713348388671875, 141.01947021484375, 5.71038818359375, 49.03388977050781, 11.305444717407227, 8.428924560546875, 26.8848876953125, 105.65155029296875, 90.6168212890625, 77.99996948242188, 6.199676513671875, 74.83767700195312, 15.739484786987305, 69.82232666015625, 42.341827392578125, 12.239700317382812, 47.19715881347656, -32.14341735839844, 63.955322265625, 53.051666259765625, 0.0, 64.22406005859375, 88.6142578125, -18.959320068359375, -3.336627960205078, 13.572174072265625, 112.11878204345703, -79.92930603027344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000150.npy"}
{"epoch": 0.31413612565445026, "step": 151, "batch_size": 128, "mean": 43.797603607177734, "std": 62.388099670410156, "min": -133.02975463867188, "p10": -35.11344833374023, "median": 45.68891906738281, "p90": 122.1595504760742, "max": 200.44281005859375, "pos_frac": 0.734375, "sample": [47.270263671875, -22.816650390625, 11.877864837646484, -84.12205505371094, 1.3816146850585938, 131.81341552734375, 105.06114196777344, -26.338058471679688, 52.21258544921875, 43.935333251953125, -9.33697509765625, 93.91336822509766, 37.938018798828125, 49.97784423828125, -45.82206726074219, -46.69783020019531, 21.31915283203125, 87.04771423339844, 92.61947631835938, 29.91991424560547, -71.8133544921875, 43.01307678222656, 25.565963745117188, 36.92356872558594, 33.6085205078125, 79.0023193359375, 48.302398681640625, -12.0037841796875, 120.19377136230469, 126.74636840820312, 67.31686401367188, 18.747894287109375, -30.010498046875, 44.674713134765625, 74.13555145263672, -3.1769561767578125, 47.76153564453125, 84.12214660644531, -58.64643859863281, 200.44281005859375, 68.38381958007812, 85.00969696044922, 35.10679626464844, -8.45458984375, 142.02841186523438, -23.060653686523438, 96.02969360351562, -43.12348937988281, 63.659576416015625, 165.07980346679688, 53.252655029296875, -36.8792724609375, 5.134092330932617, -34.356666564941406, 51.26556396484375, 79.31436157226562, 39.20556640625, 20.0206298828125, 24.169036865234375, 58.00657653808594, -10.981109619140625, 163.526123046875, 35.13897705078125, -57.511077880859375, 92.534912109375, 77.62875366210938, -3.92010498046875, 25.29510498046875, 90.41204833984375, 63.4232177734375, -133.02975463867188, -31.88299560546875, 87.05401611328125, 70.77930450439453, 40.22630310058594, 100.49520874023438, 109.05099487304688, 83.76776123046875, 157.8082275390625, -41.3641357421875, -62.511474609375, -19.001434326171875, 67.01751708984375, 54.592987060546875, 160.31549072265625, 21.410892486572266, 82.18019104003906, -2.552297592163086, 0.0, 6.39361572265625, 79.274658203125, 52.99078369140625, 112.27313995361328, -9.410430908203125, 115.50552368164062, 68.96919250488281, 63.92413330078125, 108.39456176757812, -32.72981262207031, 136.65789794921875, -60.4959716796875, 47.09834289550781, 39.741851806640625, 80.80595397949219, 153.87918090820312, 67.78372192382812, 23.143753051757812, 15.173454284667969, 78.68087768554688, 46.703125, 36.9498291015625, 101.36663818359375, -28.126007080078125, 3.955272674560547, 162.98272705078125, 3.64453125, 160.32431030273438, 52.08843994140625, 192.4483642578125, 88.6324462890625, 94.19305419921875, -23.987548828125, -5.91998291015625, 18.22845458984375, 83.41136932373047, -9.18206787109375, 36.08612060546875, -61.515838623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000151.npy"}
{"epoch": 0.3162303664921466, "step": 152, "batch_size": 128, "mean": 38.130916595458984, "std": 67.77798461914062, "min": -100.41999816894531, "p10": -39.830398559570305, "median": 31.560211181640625, "p90": 119.58782348632812, "max": 241.68841552734375, "pos_frac": 0.671875, "sample": [161.44174194335938, 193.1693115234375, 102.85873413085938, 48.59918212890625, 34.827972412109375, 47.04888916015625, 24.141326904296875, 41.140594482421875, 131.81674194335938, 18.925613403320312, 84.49357604980469, 142.82022094726562, 128.07345581054688, -22.52484130859375, 75.8321533203125, -14.647567749023438, 32.0775146484375, 113.1395263671875, 17.37689208984375, -22.234405517578125, -58.7298583984375, 46.514556884765625, -79.60885620117188, -24.2158203125, 5.010398864746094, 93.01669311523438, 20.164596557617188, 9.574516296386719, 236.3076171875, 1.098724365234375, -3.2964935302734375, 70.40858459472656, 97.39761352539062, 92.6719970703125, 106.9088134765625, 33.57159423828125, 80.02621459960938, 2.719757080078125, 115.73406982421875, -2.60931396484375, -22.818077087402344, 208.8800048828125, -93.1705322265625, -19.0828857421875, 59.83985900878906, 3.2346534729003906, -63.045654296875, 84.357421875, -15.449575424194336, -0.0390777587890625, 1.5563812255859375, -11.854476928710938, 49.82276916503906, -48.07203674316406, -7.525392532348633, -7.5877685546875, 11.688079833984375, -58.593109130859375, -1.494873046875, -21.043289184570312, 241.68841552734375, -45.99662780761719, -86.09086608886719, 51.960296630859375, 160.76507568359375, -16.87957763671875, 35.206939697265625, -1.644134521484375, 47.52716827392578, 16.95623779296875, 8.98672866821289, -37.18772888183594, -12.284381866455078, 28.54345703125, 6.682525634765625, 90.78564453125, 90.3136215209961, 93.14312744140625, 155.52670288085938, 69.54159545898438, 120.89254760742188, -72.1626205444336, -2.9373855590820312, -16.063323974609375, -17.295013427734375, -5.071624755859375, 85.500732421875, 66.0435791015625, 132.88150024414062, -2.0104446411132812, 36.05824279785156, -93.39520263671875, -14.733016967773438, 119.02865600585938, 131.47036743164062, 78.2922592163086, 31.04290771484375, 15.8900146484375, 10.263717651367188, -22.788482666015625, 69.73370361328125, 118.428466796875, 113.40072631835938, 70.5245361328125, -47.6978759765625, 109.2017822265625, 56.16832733154297, 10.150289535522461, -12.582427978515625, -13.210700988769531, 76.42346954345703, 48.887481689453125, -85.90374755859375, 47.7391357421875, -27.370811462402344, -100.41999816894531, 80.10981750488281, 53.91712188720703, 67.25486755371094, 110.87686157226562, 87.64480590820312, 35.0107421875, 106.36763000488281, 66.72607421875, 18.504562377929688, 23.124839782714844, 0.854736328125, 87.79672241210938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000152.npy"}
{"epoch": 0.3183246073298429, "step": 153, "batch_size": 128, "mean": 53.922462463378906, "std": 63.72970199584961, "min": -97.54222106933594, "p10": -12.147044372558593, "median": 43.316078186035156, "p90": 135.7642822265625, "max": 253.196044921875, "pos_frac": 0.796875, "sample": [85.0306396484375, 5.187644958496094, 127.434326171875, 14.36419677734375, 98.2379150390625, 132.80433654785156, 135.55398559570312, 4.82672119140625, 114.56402587890625, 12.15850830078125, 88.09823608398438, 0.407379150390625, 10.748138427734375, 13.217857360839844, 79.05287170410156, 22.64098358154297, -0.0908355712890625, -12.647445678710938, 18.268768310546875, 139.27349853515625, 139.486083984375, 81.91085815429688, 88.69644165039062, 5.908653259277344, 55.32745361328125, -9.0567626953125, 80.29714965820312, -50.95000457763672, 0.5151901245117188, 67.85986328125, -43.196929931640625, 50.822837829589844, 112.128662109375, -6.6534576416015625, 31.6982421875, 18.97991943359375, 29.519989013671875, 100.16233825683594, 4.342222213745117, 0.0, 141.45916748046875, 115.09541320800781, 172.31265258789062, 59.90150451660156, 98.12164306640625, 6.208892822265625, 81.27706909179688, -0.9714202880859375, 8.979705810546875, 38.821815490722656, 118.677734375, 68.08685302734375, -90.32965087890625, -14.053153991699219, -33.522674560546875, 222.7896728515625, 76.70578002929688, 90.64773559570312, -21.329147338867188, 94.01809692382812, 31.55303955078125, 8.174697875976562, -16.880355834960938, 177.7200927734375, 43.45396423339844, 59.58892822265625, -21.358642578125, -6.22967529296875, 68.66618347167969, 112.59573364257812, -9.3790283203125, 72.046875, 21.769760131835938, 57.060577392578125, 16.4649658203125, 17.093738555908203, 3.4364013671875, -11.932586669921875, 81.68365478515625, 114.47088623046875, 87.92112731933594, -2.8488006591796875, 121.94107055664062, 136.25497436523438, 72.41546630859375, 44.067596435546875, 4.296394348144531, -0.1828460693359375, 90.75374603271484, 2.796142578125, 21.416961669921875, 70.27447509765625, 131.4569549560547, 129.97006225585938, 193.45849609375, 45.20136260986328, 3.0862903594970703, 1.4575347900390625, 22.96405029296875, 0.21038818359375, 43.178192138671875, 92.93975830078125, 157.35801696777344, 18.392372131347656, -12.704727172851562, -7.2208251953125, 75.63665771484375, 100.61691284179688, 35.849876403808594, 233.6123046875, 151.9241943359375, 40.05845642089844, 20.20989990234375, 52.715728759765625, -4.744562149047852, 141.47314453125, 36.78229522705078, 253.196044921875, 131.60882568359375, -5.806427001953125, -31.34161376953125, 73.20046997070312, 120.42514038085938, 101.48880004882812, -15.685546875, -97.54222106933594, 76.5895767211914, 43.125396728515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000153.npy"}
{"epoch": 0.3204188481675393, "step": 154, "batch_size": 128, "mean": 41.93115997314453, "std": 75.73348999023438, "min": -170.70391845703125, "p10": -55.92290496826172, "median": 42.302520751953125, "p90": 129.37157592773437, "max": 360.7144775390625, "pos_frac": 0.7578125, "sample": [156.27577209472656, 110.64671325683594, 15.23223876953125, 240.832763671875, 53.157562255859375, 15.234085083007812, 9.414573669433594, 29.33135986328125, 15.529380798339844, 54.948974609375, -69.82760620117188, 110.69390869140625, 36.84654998779297, 66.73831176757812, 114.44589233398438, -126.99411010742188, 68.71957397460938, -115.00900268554688, 360.7144775390625, -12.846588134765625, 82.38047790527344, 30.043701171875, 1.58416748046875, -5.8246002197265625, 19.641937255859375, 3.800994873046875, 30.2569580078125, -5.5881195068359375, 55.1231689453125, 173.44674682617188, 71.22628784179688, -12.644207000732422, 95.75823974609375, 162.13558959960938, 24.596511840820312, 56.740455627441406, 68.1727294921875, 127.5029296875, 74.77890014648438, 112.596923828125, 111.92672729492188, 122.78515625, 118.06973266601562, 53.75640869140625, -58.178924560546875, 186.755615234375, 135.573974609375, 22.01097869873047, 41.39642333984375, 27.792236328125, -5.6681671142578125, -0.6525344848632812, 25.9326171875, -28.501739501953125, 45.31671142578125, 55.377532958984375, 147.49710083007812, -54.95603942871094, -7.673126220703125, -77.13955688476562, -13.35940933227539, 31.864837646484375, 44.50947570800781, 107.22076416015625, 51.99775695800781, -170.70391845703125, -6.488441467285156, -2.1662521362304688, -19.951385498046875, 21.9267578125, 49.469947814941406, -66.26953125, 136.18447875976562, 106.36663818359375, 72.15176391601562, -101.8651351928711, 26.990947723388672, 1.429534912109375, 56.298370361328125, 39.04132080078125, -49.91459655761719, 7.890964508056641, -22.15105438232422, 8.723175048828125, 43.2086181640625, -88.75885009765625, 61.062286376953125, 25.8487548828125, 38.849151611328125, 4.5655059814453125, 58.987823486328125, -67.54908752441406, 116.07977294921875, 68.28840637207031, 47.5867919921875, 51.673370361328125, 44.322418212890625, -37.409454345703125, 114.601318359375, 9.401031494140625, 133.73175048828125, 169.91897583007812, -26.3233642578125, -141.1632080078125, 105.52088928222656, 60.17486572265625, 33.55104064941406, 45.287811279296875, 67.8385009765625, 66.99566650390625, 165.72146606445312, 76.47753143310547, 62.011199951171875, 62.165435791015625, 23.257965087890625, 27.64044189453125, 165.08895874023438, 92.43496704101562, 10.532234191894531, 55.93768310546875, -91.57525634765625, -66.49359893798828, 105.94796752929688, -28.540481567382812, 32.03147506713867, 24.150146484375, 73.66085815429688, 100.01632690429688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000154.npy"}
{"epoch": 0.3225130890052356, "step": 155, "batch_size": 128, "mean": 36.712886810302734, "std": 70.29187774658203, "min": -172.21231079101562, "p10": -37.42165985107422, "median": 29.20954132080078, "p90": 109.76050720214842, "max": 316.2041015625, "pos_frac": 0.703125, "sample": [105.3819580078125, -14.812149047851562, 11.0662841796875, 7.80035400390625, 78.68368530273438, 5.648281097412109, 38.79380798339844, 17.52069091796875, 41.455902099609375, 92.27752685546875, 126.27920532226562, 7.1713409423828125, 51.96124267578125, 32.98396301269531, 29.39544677734375, 108.32000732421875, -8.0123291015625, 133.09039306640625, 103.4669189453125, -9.8775634765625, 11.4403076171875, 172.32115173339844, 91.42002868652344, 154.94293212890625, -2.680206298828125, 108.30889892578125, -39.34162902832031, 0.0, -36.3712158203125, -27.292724609375, -6.6126708984375, 245.64996337890625, 93.50570678710938, 84.30796813964844, 19.10784912109375, -172.21231079101562, -90.09332275390625, 2.7038421630859375, -47.8465576171875, -2.0241851806640625, -16.8070068359375, 62.99737548828125, 16.885498046875, 99.23190307617188, 157.8123779296875, 74.621337890625, 67.732421875, 65.82232666015625, 12.910469055175781, 21.910785675048828, -12.603919982910156, -19.463623046875, 5.496036529541016, -15.59747314453125, 12.21270751953125, -36.59881591796875, 29.3931884765625, 38.410797119140625, 182.78115844726562, 185.2893829345703, 3.4823036193847656, -59.60731506347656, 95.4852294921875, 4.3414306640625, -30.093399047851562, -7.156991958618164, 105.07334899902344, 162.05734252929688, 37.32774353027344, 17.19611358642578, 14.423324584960938, -40.61252212524414, 86.15145874023438, 4.77655029296875, -30.53564453125, 77.18844604492188, 20.812896728515625, 5.07684326171875, 316.2041015625, -69.84317016601562, 64.736572265625, 117.66258239746094, -59.78179931640625, -6.68963623046875, 17.80950927734375, 68.58352661132812, 29.819976806640625, -22.45391845703125, -5.3436279296875, 62.78385925292969, 40.477081298828125, 18.686004638671875, 42.1453857421875, 33.251190185546875, -3.39288330078125, 41.15419006347656, 74.08787536621094, 113.12167358398438, 22.833831787109375, 29.025894165039062, -23.05633544921875, 31.448654174804688, -51.625335693359375, 35.32856750488281, 37.74237060546875, 99.83514404296875, 15.830574035644531, -21.06262969970703, -97.270751953125, 0.5946121215820312, -72.79547119140625, 75.67021179199219, 69.76605987548828, -7.2464599609375, 73.0220947265625, -18.334228515625, 81.8325424194336, 67.9617919921875, 30.588287353515625, 270.520263671875, -48.655364990234375, 37.611968994140625, -40.464744567871094, 82.81124877929688, 29.805755615234375, 59.645111083984375, 56.76295471191406, 82.45147705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000155.npy"}
{"epoch": 0.32460732984293195, "step": 156, "batch_size": 128, "mean": 50.82361602783203, "std": 81.22702026367188, "min": -170.4060821533203, "p10": -33.696408081054685, "median": 40.817832946777344, "p90": 155.9842987060547, "max": 276.892822265625, "pos_frac": 0.7578125, "sample": [126.4027328491211, 122.22869873046875, 12.318229675292969, 25.66961669921875, 6.688323974609375, -14.24237060546875, 97.5111083984375, 61.338714599609375, 109.12411499023438, 52.815147399902344, 96.59532928466797, 41.072235107421875, 71.67388916015625, 68.08718872070312, 31.846824645996094, -170.4060821533203, -110.94918823242188, 48.117431640625, 49.437522888183594, 10.939811706542969, -11.294891357421875, 74.00601196289062, -7.7213592529296875, 16.43579864501953, 22.93072509765625, 1.921051025390625, -1.1143035888671875, 39.57164001464844, -23.132034301757812, 129.2930908203125, 144.729248046875, -7.0570068359375, 97.08966064453125, 7.4254150390625, -3.0142345428466797, 6.555755615234375, -44.49078369140625, -42.234458923339844, 89.435546875, 3.1062469482421875, 16.655784606933594, 82.12222290039062, 66.8858642578125, 21.4525146484375, -95.33946228027344, 71.6561279296875, 106.1894302368164, -33.6318359375, 89.04733276367188, 68.72232055664062, 10.50650405883789, 95.71576690673828, 45.656402587890625, 203.25918579101562, 185.05197143554688, -11.16314697265625, -61.352874755859375, 214.09609985351562, -49.740386962890625, 17.740577697753906, -33.847076416015625, 225.2821044921875, 28.064468383789062, 108.28860473632812, 7.3631591796875, 123.94088745117188, 126.1407470703125, 121.08416748046875, 125.60015869140625, 4.1882171630859375, 29.85003662109375, 40.56343078613281, -18.98004913330078, 76.985595703125, 127.36563110351562, 58.94000244140625, 214.92535400390625, 27.53668212890625, 1.4086036682128906, -11.044017791748047, 155.00601196289062, 149.6260986328125, 148.4002685546875, -15.864059448242188, -3.3130416870117188, 103.6952896118164, 8.771841049194336, 103.34689331054688, 50.35023498535156, 48.99494934082031, 43.960205078125, 135.94223022460938, 158.2669677734375, 188.2152099609375, 184.31854248046875, 32.97882080078125, -168.10556030273438, 26.76903533935547, 71.04910278320312, 15.31719970703125, 69.55810546875, 22.462249755859375, -21.71295166015625, -20.644775390625, -150.4580078125, 192.03961181640625, 12.83192253112793, 177.900390625, -5.5239105224609375, -87.0616455078125, 23.53887939453125, 76.39126586914062, 168.0963134765625, 88.7977294921875, -115.37936401367188, -27.697898864746094, 8.434234619140625, 110.53294372558594, 96.240234375, 170.975830078125, 27.728408813476562, 103.58451843261719, 124.4954833984375, 146.98492431640625, -8.694107055664062, 276.892822265625, -60.887123107910156, 12.3731689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000156.npy"}
{"epoch": 0.3267015706806283, "step": 157, "batch_size": 128, "mean": 40.32099533081055, "std": 70.26083374023438, "min": -161.3160400390625, "p10": -43.58781433105467, "median": 31.43640899658203, "p90": 131.19791564941406, "max": 227.8387451171875, "pos_frac": 0.7109375, "sample": [130.77667236328125, 14.370979309082031, 148.55020141601562, 130.57955932617188, 63.083587646484375, -29.24925994873047, 25.02008056640625, 112.39498901367188, 90.72280883789062, 61.435821533203125, -9.50103759765625, -27.669418334960938, -72.8109130859375, 26.008148193359375, 13.582710266113281, 35.87176513671875, -9.260154724121094, -34.21476745605469, 12.324237823486328, -101.55856323242188, -30.203781127929688, 151.8544921875, 41.338714599609375, -52.098876953125, 146.83250427246094, -24.26407241821289, -73.11813354492188, 31.774734497070312, 33.45867919921875, 130.43905639648438, -5.90618896484375, -53.80133056640625, 13.110198974609375, -6.31536865234375, 47.11564636230469, 216.47052001953125, 104.99844360351562, 17.726943969726562, 15.278976440429688, -70.27490234375, 227.8387451171875, 71.07972717285156, 132.18081665039062, 1.11700439453125, 151.20388793945312, 27.558639526367188, 45.11512756347656, 26.92290496826172, -16.17828369140625, 0.0, 37.31605529785156, 46.449493408203125, 12.72344970703125, -52.581207275390625, 149.90289306640625, 24.873748779296875, 97.95921325683594, -5.087005615234375, 13.401275634765625, -10.761405944824219, 69.6993408203125, 36.401123046875, 132.8333282470703, 41.989837646484375, 49.22816467285156, -13.16965103149414, 114.511474609375, -4.62109375, -87.1014404296875, 68.71041870117188, 31.09808349609375, 34.662384033203125, -39.940216064453125, -130.97586059570312, -12.808197021484375, 10.44879150390625, -22.79998779296875, 34.376708984375, 95.6241455078125, 100.34210205078125, 71.86077880859375, 100.28787231445312, -8.2945556640625, -18.59716796875, 4.1607818603515625, 134.36732482910156, -8.820404052734375, 98.140869140625, 110.45059204101562, 0.0, -54.717132568359375, 175.92578125, 120.03408813476562, 102.02304077148438, 41.324249267578125, 19.48651123046875, 77.07247161865234, 86.23843383789062, 168.59149169921875, 78.31411743164062, 61.607940673828125, -84.64579772949219, 63.643157958984375, 27.60554313659668, 26.2279052734375, 25.942611694335938, 122.8962173461914, 63.33378219604492, 79.05850219726562, -14.26544189453125, -27.70123291015625, 69.63800048828125, 13.480392456054688, 118.47564697265625, 114.42976379394531, 42.183624267578125, 13.864105224609375, 12.183225631713867, -161.3160400390625, 129.41506958007812, -66.29769897460938, 22.675987243652344, 2.9256343841552734, 86.96537780761719, 24.626571655273438, 115.54277038574219, 181.91802978515625, 94.40640258789062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000157.npy"}
{"epoch": 0.3287958115183246, "step": 158, "batch_size": 128, "mean": 45.4599494934082, "std": 72.13594055175781, "min": -247.57861328125, "p10": -30.439942932128904, "median": 41.45944595336914, "p90": 131.54903106689449, "max": 266.94683837890625, "pos_frac": 0.7421875, "sample": [20.97905158996582, 73.09355163574219, 40.26806640625, 60.75921630859375, -29.59027099609375, 33.21339416503906, 2.2643280029296875, -28.36737060546875, -109.42355346679688, 94.9146728515625, -9.4595947265625, 24.24737548828125, -46.1475830078125, -88.20077514648438, 161.20294189453125, 29.835235595703125, 76.85111999511719, 78.74576568603516, 93.98294067382812, 127.908203125, 25.50678253173828, 120.21798706054688, 145.14111328125, 4.35308837890625, 66.09661865234375, 162.05166625976562, 116.38023376464844, -247.57861328125, 75.0833740234375, 39.51213073730469, -31.250396728515625, 106.77259826660156, 48.73236083984375, 91.71923828125, 31.66363525390625, 108.84935760498047, -1.613311767578125, -64.16857147216797, 69.84491729736328, -71.57974243164062, 84.42794799804688, 38.93939208984375, 50.268760681152344, -18.544429779052734, -24.661376953125, -8.523674011230469, -38.504241943359375, -6.552581787109375, 10.998870849609375, 24.049758911132812, 66.27252197265625, -4.89836311340332, -11.730024337768555, 12.32680892944336, 17.10936737060547, 161.3034210205078, -74.13343048095703, -16.93841552734375, 113.78619384765625, 54.00736999511719, 188.721435546875, 31.184432983398438, -6.88629150390625, 63.817138671875, 125.93280029296875, 16.814056396484375, -16.346481323242188, 0.0, 41.524131774902344, 42.114593505859375, 68.52163696289062, 7.33502197265625, 4.92742919921875, 55.782958984375, 117.18505859375, 82.037353515625, -30.092605590820312, 35.08570861816406, -93.90158081054688, -18.424081802368164, 205.5333251953125, 3.622894287109375, -89.00640869140625, 41.39476013183594, -0.3210601806640625, 266.94683837890625, 112.10140991210938, 3.0210494995117188, 84.73126220703125, 199.31256103515625, 64.97219848632812, 141.40264892578125, 74.914306640625, 29.094329833984375, 176.583251953125, 55.74468994140625, 125.00686645507812, 48.0800666809082, 9.45233154296875, 23.745742797851562, 147.8610382080078, 86.44406127929688, 88.50906372070312, 121.57330322265625, 140.04429626464844, 69.48736572265625, 51.91815185546875, 114.57568359375, 71.1341552734375, -26.97237777709961, 33.81414794921875, 39.4281005859375, 94.29817199707031, 33.87646484375, 80.01095581054688, 48.36622619628906, 104.44096374511719, -0.4959144592285156, -28.454452514648438, -44.5338134765625, 102.173583984375, 123.4625244140625, -48.91497802734375, 59.577362060546875, 142.148681640625, 43.429229736328125, 37.7580680847168, 10.43271255493164], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000158.npy"}
{"epoch": 0.3308900523560209, "step": 159, "batch_size": 128, "mean": 26.61713409423828, "std": 78.7261734008789, "min": -171.04910278320312, "p10": -65.13068466186523, "median": 22.139999389648438, "p90": 117.77017822265623, "max": 376.33477783203125, "pos_frac": 0.6484375, "sample": [-12.611663818359375, 30.374053955078125, 59.16925048828125, 18.27880859375, 4.470130920410156, -70.41342163085938, -60.6417236328125, 115.292724609375, 37.05767822265625, 98.65890502929688, -87.75611877441406, 25.790733337402344, -158.58270263671875, -50.10503387451172, 16.387832641601562, 376.33477783203125, 37.15412139892578, -2.1044769287109375, -48.724609375, 114.6443099975586, 22.429061889648438, 2.644989013671875, -163.89663696289062, -70.5609359741211, 44.91583251953125, 148.76510620117188, -2.599822998046875, 75.82762145996094, 42.241363525390625, -4.1003875732421875, -1.2095527648925781, 22.22430419921875, 46.55584716796875, -8.343521118164062, 35.58628845214844, -73.79537963867188, 81.73965454101562, -20.88092041015625, -4.734432220458984, -62.136505126953125, 35.0125732421875, 6.6484832763671875, 44.13825988769531, 87.29495239257812, -14.52471923828125, 61.2430419921875, -17.000686645507812, -8.943395614624023, -116.49285888671875, -34.72357177734375, 22.055694580078125, -63.01673889160156, 76.30450439453125, -23.806625366210938, 79.15484619140625, 104.43231201171875, 3.9804725646972656, 42.12214660644531, 0.9007110595703125, 8.9429931640625, 148.14097595214844, 71.8201904296875, 21.45477294921875, 21.378158569335938, 82.34918212890625, 88.66635131835938, 24.967315673828125, 26.05242919921875, 42.4547119140625, 16.790267944335938, 10.745620727539062, -17.04461669921875, -33.528099060058594, 135.00970458984375, 23.38775634765625, -83.72787475585938, -63.91752624511719, 18.291656494140625, -78.29385375976562, 16.4224853515625, 98.21002197265625, -63.85076904296875, 82.68231201171875, 140.06900024414062, 164.4755096435547, 15.554813385009766, -19.473098754882812, -34.83540344238281, 55.32368469238281, 152.2222900390625, 69.76141357421875, 50.09895324707031, -57.369354248046875, 53.56217956542969, 54.42747497558594, 32.182037353515625, 52.09942626953125, 83.4278564453125, -19.062877655029297, 9.800167083740234, 31.250526428222656, -91.53943634033203, -171.04910278320312, 50.28436279296875, 233.4150390625, 207.19097900390625, 83.13613891601562, -67.96138763427734, 140.39303588867188, 11.388618469238281, -73.05575561523438, -50.5263671875, 123.5509033203125, -58.432891845703125, 153.5166015625, -54.676910400390625, 68.53253173828125, 103.80047607421875, 105.28057861328125, 142.397216796875, -1.5908203125, 106.31147766113281, -36.53627014160156, 95.67633056640625, 62.47313690185547, 4.310302734375, -11.5115966796875, 63.1722412109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000159.npy"}
{"epoch": 0.33298429319371725, "step": 160, "batch_size": 128, "mean": 51.03730392456055, "std": 62.293087005615234, "min": -76.75898742675781, "p10": -20.3068302154541, "median": 51.277610778808594, "p90": 122.34122924804687, "max": 238.11453247070312, "pos_frac": 0.765625, "sample": [26.16565704345703, 62.17179870605469, 0.0, 124.99826049804688, 14.299224853515625, 87.11080932617188, 64.964111328125, 71.23779296875, 111.94219970703125, 49.866455078125, 78.17112731933594, 103.373046875, 46.85148620605469, 77.39068603515625, -14.452850341796875, -4.993377685546875, 155.99420166015625, 51.61436462402344, 41.24022674560547, 19.929229736328125, 55.391204833984375, 47.966552734375, 92.04209899902344, 96.23777770996094, 71.15559387207031, -3.374114990234375, -5.4286651611328125, 93.98553466796875, -24.919015884399414, -63.1317138671875, 97.15972900390625, 211.0535888671875, -76.75898742675781, 209.8209228515625, 66.0157470703125, 3.544879913330078, 104.22283172607422, -18.410717010498047, 108.72615051269531, -9.095970153808594, 107.50028228759766, 78.25845336914062, 33.9122314453125, 0.0, 53.475341796875, -9.535736083984375, 82.80817413330078, -17.762054443359375, -56.466888427734375, 60.88232421875, 59.17523193359375, 48.505950927734375, 237.60931396484375, 108.93936157226562, -25.93115234375, 24.066757202148438, 71.78671264648438, -24.097686767578125, 7.6358489990234375, 17.380340576171875, 133.5899658203125, 114.90161895751953, 49.41064453125, -57.3580322265625, 4.08795166015625, 106.39523315429688, 86.0928955078125, 125.21653747558594, 15.327436447143555, -45.91766357421875, 92.54641723632812, 3.7497329711914062, 14.683692932128906, 71.16998291015625, 112.46321105957031, 86.19705200195312, 75.57022094726562, 78.03374481201172, 141.29913330078125, 69.33282470703125, -10.10662841796875, -28.00054931640625, 106.84185791015625, -2.7068099975585938, 5.47979736328125, 116.30364990234375, 19.43103790283203, 35.68988037109375, 85.19766235351562, 64.50210571289062, 19.747207641601562, 79.12406921386719, 130.3363037109375, -46.95782470703125, 75.89044189453125, -45.07086181640625, 5.379722595214844, -16.057464599609375, 79.46490478515625, 27.437423706054688, 25.652328491210938, 63.55265808105469, -21.549312591552734, 28.264801025390625, 35.393096923828125, 1.3755340576171875, 14.7777099609375, -9.316070556640625, 50.94085693359375, 123.27685546875, 200.35845947265625, -19.216400146484375, -15.8411865234375, 88.14872741699219, 25.429718017578125, 77.7735366821289, 238.11453247070312, 8.572235107421875, 106.35682678222656, 145.67430114746094, 74.91757202148438, 37.960113525390625, -70.12445068359375, 121.94024658203125, 66.28083801269531, 87.370849609375, -19.774337768554688, 5.525785446166992], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000160.npy"}
{"epoch": 0.33507853403141363, "step": 161, "batch_size": 128, "mean": 45.19239044189453, "std": 80.91664123535156, "min": -152.0895538330078, "p10": -39.643165588378906, "median": 38.25489807128906, "p90": 127.34018096923828, "max": 438.41925048828125, "pos_frac": 0.7265625, "sample": [-7.1233367919921875, 36.944801330566406, -18.24884033203125, -94.24771118164062, 18.28354835510254, 6.39129638671875, 5.7255401611328125, 13.67431640625, -56.286376953125, 13.44525146484375, 5.245515823364258, 118.10867309570312, 119.97648620605469, -103.49990844726562, -1.0765380859375, 3.8130226135253906, 134.259033203125, -25.6368408203125, 57.368064880371094, -7.819122314453125, 41.04656982421875, -6.7100677490234375, 207.08868408203125, 95.5167236328125, -40.96202087402344, 109.26654052734375, 32.36785888671875, 151.70721435546875, 92.24923706054688, 60.17931365966797, 38.661376953125, 62.82893371582031, -10.699676513671875, -97.97575378417969, 34.05448532104492, 113.89834594726562, 44.606475830078125, 53.4124755859375, 8.215110778808594, 54.66108322143555, -20.51885986328125, -152.0895538330078, -34.22589111328125, -30.227828979492188, 9.596481323242188, 50.67791748046875, -69.37396240234375, 15.64129638671875, 0.0, 117.78018188476562, -9.05908203125, 97.4139404296875, 28.838401794433594, 7.624752044677734, 21.20068359375, 55.86761474609375, 127.67884826660156, 112.05615234375, 47.403472900390625, 69.59904479980469, 27.74908447265625, 126.88629150390625, -39.07794189453125, 29.058868408203125, -18.203216552734375, 85.0467529296875, 65.86053466796875, 171.08242797851562, -118.97421264648438, 45.704345703125, 227.80712890625, 42.772640228271484, 86.39639282226562, 265.4135437011719, -44.250762939453125, 118.97076416015625, 108.196533203125, -96.1656494140625, 15.3209228515625, 107.51757049560547, -97.2274398803711, 36.581138610839844, 438.41925048828125, 141.77499389648438, 63.17399597167969, 147.21371459960938, 71.09443664550781, 92.02191162109375, 91.53677368164062, -8.763956069946289, 114.54620361328125, 111.90733337402344, -30.245086669921875, 130.67544555664062, 22.059051513671875, 20.28778076171875, 37.848419189453125, 29.454246520996094, 60.956817626953125, -11.06500244140625, 22.249725341796875, 144.3023681640625, 59.371673583984375, -34.8223876953125, 109.25652313232422, 99.53713989257812, -33.95060729980469, 27.258544921875, 71.78278350830078, 124.92910766601562, 85.685546875, 122.1341552734375, -27.88237762451172, 41.06256103515625, 69.58122253417969, -125.781982421875, 122.77001953125, 146.70391845703125, 32.34953308105469, 0.0, 25.554718017578125, 109.10256958007812, 127.19503784179688, -86.38561248779297, 36.315399169921875, 90.51950073242188, -34.576568603515625, 82.40814208984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000161.npy"}
{"epoch": 0.33717277486910996, "step": 162, "batch_size": 128, "mean": 41.90522766113281, "std": 72.87638092041016, "min": -161.287353515625, "p10": -34.23286437988281, "median": 37.95991516113281, "p90": 129.99269104003903, "max": 336.3901672363281, "pos_frac": 0.71875, "sample": [-11.209213256835938, 11.519298553466797, 160.14913940429688, 70.33668518066406, -15.80609130859375, 30.10028076171875, -67.64372253417969, 11.78515625, 10.672775268554688, 43.213829040527344, 68.49153137207031, 143.45477294921875, -2.1559791564941406, 57.52613830566406, -15.105224609375, -45.112060546875, 6.52349853515625, 38.0362548828125, -11.943122863769531, -7.619945526123047, -84.7515869140625, 81.70399475097656, 35.15260314941406, 95.61131286621094, 68.94769287109375, -36.935264587402344, 116.84710693359375, 26.71148681640625, -25.12078857421875, 37.883575439453125, -13.899120330810547, 101.98193359375, 100.04299926757812, -42.72669982910156, 177.4569091796875, 50.48553466796875, 46.72900390625, -147.00393676757812, -1.2393302917480469, 159.6236572265625, 0.617401123046875, 125.76885986328125, -6.797006607055664, 44.2415771484375, 192.89959716796875, -157.5528564453125, 182.13186645507812, 40.308502197265625, 115.45303344726562, 139.84829711914062, 68.97154235839844, 80.2703857421875, 17.29669189453125, 20.13018798828125, 103.43479919433594, 2.6836891174316406, 12.97711181640625, 52.99591064453125, 77.48614501953125, 51.920745849609375, 336.3901672363281, 141.84371948242188, -17.49371337890625, -27.844741821289062, 117.7542724609375, 198.09239196777344, 152.56085205078125, 105.291015625, -22.34613037109375, 79.35867309570312, 3.9818763732910156, 32.558197021484375, -161.287353515625, 46.734222412109375, 116.20138549804688, -28.755443572998047, -35.636962890625, 9.674652099609375, 12.905120849609375, 121.00349426269531, -15.49978256225586, 102.265380859375, 100.34303283691406, 58.96156311035156, 97.83938598632812, 116.25942993164062, 59.940528869628906, 10.40625, 59.841522216796875, 75.13508605957031, -12.863510131835938, 94.91781616210938, 46.42390060424805, -107.3392333984375, 18.21478271484375, 11.170211791992188, 77.67435455322266, 53.27557373046875, 34.98016357421875, 86.21371459960938, 54.852149963378906, -35.421142578125, 102.2498779296875, -2.5249176025390625, 15.996627807617188, 146.40924072265625, -9.715179443359375, -50.50126647949219, -14.677078247070312, 90.69235229492188, -70.23588562011719, -4.1795501708984375, -18.246902465820312, 64.57904052734375, 26.720001220703125, 142.5517578125, 39.55217361450195, 8.655141830444336, 49.81951904296875, 35.1531982421875, 77.66266632080078, -33.723602294921875, 34.339134216308594, 13.401947021484375, -29.95915985107422, 113.18258666992188, 21.728057861328125, 58.561279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000162.npy"}
{"epoch": 0.3392670157068063, "step": 163, "batch_size": 128, "mean": 47.2469482421875, "std": 68.87916564941406, "min": -136.26226806640625, "p10": -33.777714538574216, "median": 46.006744384765625, "p90": 131.00533142089841, "max": 240.48977661132812, "pos_frac": 0.765625, "sample": [62.4300537109375, 5.974666595458984, 98.8647689819336, -28.540618896484375, 88.11654663085938, -15.087966918945312, 53.25132751464844, 37.64508056640625, 65.5361328125, -3.158477783203125, 9.57623291015625, 115.34855651855469, 186.8396759033203, 28.25048828125, 141.6875, -45.389556884765625, 75.49835205078125, 41.1015625, -7.512298583984375, 44.489410400390625, -35.645965576171875, 107.745361328125, 71.27264404296875, -18.318519592285156, 12.82574462890625, 62.838653564453125, -11.229804992675781, -14.928604125976562, 117.06976318359375, 2.2215576171875, 22.99609375, 174.90399169921875, -20.93707275390625, 74.8358154296875, 85.1766357421875, 23.1015625, 47.381317138671875, 55.1387939453125, -108.047119140625, -31.565231323242188, -3.83502197265625, 54.881805419921875, 29.316207885742188, 67.6931381225586, -20.0577392578125, 164.4280548095703, -120.06301879882812, 102.16845703125, 89.3031234741211, 45.42535400390625, 93.56536102294922, 0.9669189453125, 56.548492431640625, 164.20057678222656, 45.20213317871094, 208.1020050048828, 2.970367431640625, 46.588134765625, 120.29722595214844, 104.92022705078125, 23.03363037109375, -2.7220306396484375, 31.3175048828125, 0.1063995361328125, -19.238601684570312, -44.802520751953125, 53.70738983154297, 105.71349334716797, 57.589599609375, 85.77220153808594, -117.97390747070312, 115.56918334960938, 47.88890075683594, 139.99331665039062, 24.254379272460938, -64.55117797851562, 127.49542236328125, 4.542236328125, 134.9378662109375, 143.7530517578125, 3.99395751953125, 103.7840576171875, 44.22833251953125, 40.1319465637207, 90.10685729980469, 109.45603942871094, 91.40399169921875, 109.91212463378906, 240.48977661132812, 129.97354125976562, 8.86418342590332, 19.371917724609375, 105.32841491699219, 106.4527587890625, 42.46217346191406, 7.34088134765625, -95.73851776123047, 55.95528030395508, 136.98748779296875, 21.077774047851562, 88.36138916015625, 124.759521484375, -92.08963012695312, 112.705322265625, -42.969482421875, 33.841522216796875, 90.26171875, 1.0975799560546875, -8.008209228515625, -58.935028076171875, 42.7020263671875, 71.80975341796875, 94.69790649414062, 60.67896270751953, 82.3818130493164, -3.1260986328125, 62.59320068359375, 133.412841796875, 120.13580322265625, 26.789459228515625, -7.959873199462891, -32.97703552246094, -136.26226806640625, 86.74177551269531, 146.55032348632812, 25.066986083984375, 33.90709686279297, -50.90625762939453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000163.npy"}
{"epoch": 0.3413612565445026, "step": 164, "batch_size": 128, "mean": 42.64482116699219, "std": 78.91986846923828, "min": -166.26443481445312, "p10": -55.88275756835937, "median": 33.5828857421875, "p90": 140.514404296875, "max": 324.19049072265625, "pos_frac": 0.6953125, "sample": [26.222686767578125, 41.81565856933594, 13.396242141723633, 162.22323608398438, 4.2126312255859375, -4.7078094482421875, 48.57901382446289, -42.826019287109375, 91.28007507324219, 94.19189453125, 140.02468872070312, -5.707611083984375, -16.867340087890625, 71.91119384765625, 94.2742919921875, -166.26443481445312, 13.957305908203125, 182.73367309570312, 7.003896713256836, 118.65179443359375, 15.15142822265625, -81.97994995117188, 130.5933074951172, 9.54791259765625, 8.0950927734375, -12.915565490722656, 123.07484436035156, -13.11029052734375, 10.640625, 99.80398559570312, 45.19126892089844, 119.13711547851562, 52.812469482421875, 182.51329040527344, -67.88450622558594, 144.572265625, -29.587539672851562, 22.619369506835938, 33.77740478515625, -78.389404296875, -30.46722412109375, 102.08525085449219, 95.434814453125, 57.1187744140625, 324.19049072265625, 197.29254150390625, 54.414794921875, 13.655426025390625, 40.27484130859375, 16.34674072265625, -28.41657257080078, -4.576606750488281, -27.977294921875, 6.957675933837891, -127.90975952148438, 176.57485961914062, 53.67509460449219, 135.12020874023438, 58.57090759277344, 49.600738525390625, -45.60246276855469, 93.52558898925781, 73.322265625, 33.38836669921875, -52.591583251953125, 89.61778259277344, -1.29315185546875, -73.15728759765625, 41.8681640625, 3.547271728515625, 134.11175537109375, 136.93328857421875, 132.8240966796875, -20.418182373046875, -77.91228485107422, 18.91162109375, 28.253799438476562, -4.25468635559082, -67.33822631835938, -23.452468872070312, 41.7711181640625, 97.95864868164062, -73.62065124511719, 58.171119689941406, 25.536407470703125, 54.328277587890625, 214.31625366210938, 119.26339721679688, 96.589111328125, 83.60525512695312, 1.4984970092773438, 128.6195831298828, -6.218116760253906, 128.004150390625, -20.79168701171875, -3.22705078125, 155.45230102539062, 63.5987548828125, -40.64814758300781, 10.7178955078125, 23.328582763671875, 65.45686340332031, -28.222198486328125, -24.628997802734375, 1.29248046875, -82.59402465820312, 141.65707397460938, -64.48565673828125, 108.91649627685547, 71.85774230957031, -19.065277099609375, 47.1717529296875, -9.158939361572266, 39.859405517578125, -63.562164306640625, -79.84249877929688, 9.260093688964844, -7.6092529296875, 114.12066650390625, 11.841278076171875, 102.71632385253906, 58.013580322265625, 142.79806518554688, 127.286376953125, 52.54168701171875, 211.8487091064453, 196.45831298828125, 6.338293075561523], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000164.npy"}
{"epoch": 0.34345549738219894, "step": 165, "batch_size": 128, "mean": 31.38031578063965, "std": 66.61856079101562, "min": -167.69564819335938, "p10": -45.37162780761717, "median": 25.841697692871094, "p90": 121.30095748901365, "max": 197.70899963378906, "pos_frac": 0.6796875, "sample": [25.573455810546875, 103.69024658203125, 2.4046897888183594, 55.4017333984375, -14.42315673828125, -21.700210571289062, 13.5811767578125, 36.4793701171875, 140.215087890625, 28.445068359375, 78.14300537109375, 12.819145202636719, -167.69564819335938, 114.32657623291016, 89.17803955078125, -15.944915771484375, -120.85418701171875, 92.50567626953125, -14.854110717773438, 181.9212646484375, 75.2972412109375, -15.49761962890625, 197.70899963378906, 82.58486938476562, 53.19073486328125, -77.11045837402344, -16.0751953125, -28.337814331054688, 49.00604248046875, 70.75259399414062, -24.819091796875, -32.185028076171875, -98.52984619140625, 29.750457763671875, 7.558769226074219, 27.738235473632812, 20.55097198486328, -2.5144805908203125, -17.27931022644043, -73.90634155273438, 34.4268798828125, 52.03569030761719, 178.61495971679688, 34.5809326171875, 76.38107299804688, 146.3392333984375, 0.22617340087890625, -19.410423278808594, 42.65176010131836, 36.41725158691406, 135.52789306640625, 39.77598571777344, 26.109939575195312, -56.1556396484375, 70.262939453125, -7.365478515625, -73.9222412109375, 19.850149154663086, -12.047355651855469, 9.959686279296875, 30.14935302734375, 65.76812744140625, 6.4057769775390625, 94.14157104492188, 41.673919677734375, -40.749908447265625, 62.417205810546875, 23.283370971679688, -12.22998046875, -58.81794738769531, 28.61334228515625, -86.98751068115234, 22.72021484375, 64.271484375, 106.2696533203125, -4.667205810546875, 130.89576721191406, -3.4837646484375, 58.275482177734375, 18.7021484375, -39.99485778808594, 89.69049835205078, 107.29367065429688, -9.81829833984375, 12.961750030517578, 77.88372802734375, 23.334266662597656, 87.5299072265625, 92.7626953125, 123.7813720703125, 13.7545166015625, 1.1028594970703125, -35.816497802734375, 85.02973937988281, 150.83364868164062, 8.303085327148438, 25.110336303710938, -1.445037841796875, 135.26829528808594, 24.351638793945312, 109.6995849609375, -35.821754455566406, 120.23792266845703, 115.6189956665039, 158.85751342773438, -39.1082763671875, 27.14675521850586, 77.67704772949219, 74.11508178710938, -56.532470703125, 2.9403419494628906, -2.4598121643066406, 166.68650817871094, 40.265289306640625, 123.97683715820312, 59.98968505859375, 44.95477294921875, -37.91797637939453, 65.99139404296875, 70.60187530517578, 9.928394317626953, 97.46426391601562, -84.30326843261719, -69.13595581054688, -37.97265625, -34.98686218261719, -59.872650146484375, 8.716262817382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000165.npy"}
{"epoch": 0.34554973821989526, "step": 166, "batch_size": 128, "mean": 39.917572021484375, "std": 74.06755065917969, "min": -111.190673828125, "p10": -49.010369873046876, "median": 23.69194793701172, "p90": 136.01804962158204, "max": 271.47796630859375, "pos_frac": 0.6640625, "sample": [75.10214233398438, 3.745452880859375, 6.10430908203125, 44.538726806640625, -2.14892578125, 47.637481689453125, 235.51602172851562, 103.81985473632812, -6.527862548828125, 16.162334442138672, 40.955345153808594, 15.8297119140625, 151.02767944335938, -94.85122680664062, 26.41192626953125, -27.495208740234375, 77.32783508300781, 113.51919555664062, 16.74028778076172, -16.52355194091797, -14.129951477050781, -79.80722045898438, 80.77798461914062, 87.67646789550781, -50.281982421875, -2.42510986328125, 4.466411590576172, 58.48210144042969, -93.52418518066406, -8.423126220703125, -7.294464111328125, 100.66162109375, 137.67054748535156, 189.41845703125, 137.12860107421875, 105.0157470703125, 21.148574829101562, 6.9857177734375, -80.94438934326172, -83.66018676757812, 5.569311141967773, -12.816207885742188, -0.4932403564453125, -1.555419921875, -33.31915283203125, 129.2767333984375, 23.927276611328125, 87.72506713867188, -70.7542724609375, 60.585601806640625, 46.11817169189453, 170.90438842773438, -5.945220947265625, -38.51788330078125, 193.51922607421875, -111.190673828125, -87.29067993164062, 110.84744262695312, 85.80587768554688, 96.45791625976562, -38.93175506591797, -23.254196166992188, 80.89630126953125, -5.1136474609375, 112.09732055664062, -32.49853515625, 140.29046630859375, 27.449951171875, 87.97842407226562, -11.379425048828125, 88.43241882324219, 71.54254150390625, 176.61288452148438, -12.817352294921875, -32.515708923339844, 49.70945739746094, 109.91094207763672, 39.195770263671875, 88.87269592285156, 15.72259521484375, -71.9624252319336, 84.06298828125, 23.456619262695312, -10.712310791015625, 16.300811767578125, 108.80044555664062, 61.60687255859375, 17.300888061523438, 8.706146240234375, -34.29969787597656, -22.322174072265625, 6.228828430175781, -7.106761932373047, 144.35494995117188, 121.19471740722656, 88.64588928222656, 26.507904052734375, 133.22434997558594, 72.79182434082031, 46.208648681640625, 126.88595581054688, -5.27110481262207, -49.04093933105469, 53.499122619628906, -0.8038043975830078, 64.9761962890625, 71.57415771484375, 135.54209899902344, 6.768276214599609, 134.6964569091797, -18.097076416015625, -81.5682373046875, 4.017333984375, 47.99223327636719, 158.5498046875, -74.86686706542969, -48.99726867675781, -3.09381103515625, 71.6688232421875, 92.81777954101562, 202.90072631835938, 8.727149963378906, 271.47796630859375, 4.930500030517578, 19.689910888671875, 103.77859497070312, 79.93684387207031, 0.8812255859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000166.npy"}
{"epoch": 0.34764397905759165, "step": 167, "batch_size": 128, "mean": 44.565155029296875, "std": 76.3853988647461, "min": -160.69361877441406, "p10": -40.91251373291014, "median": 39.95157241821289, "p90": 133.27510070800778, "max": 269.75299072265625, "pos_frac": 0.7734375, "sample": [102.69657897949219, 10.256729125976562, 17.17193603515625, 59.85052490234375, 105.7910385131836, 77.18609619140625, 80.37066650390625, 269.75299072265625, 104.93132019042969, 152.4327392578125, 102.51025390625, -147.32623291015625, 127.28150939941406, 3.0279979705810547, 70.94636535644531, 73.11117553710938, 16.74658203125, 137.6634521484375, 37.54119873046875, 35.995208740234375, 95.17080688476562, 58.847381591796875, -133.22933959960938, -119.84246826171875, -2.4292678833007812, -15.063423156738281, -49.43217468261719, 62.714202880859375, 90.2744140625, 120.83670043945312, 140.33908081054688, 153.9062042236328, 72.8327407836914, 131.39437866210938, 65.35163879394531, 19.93927001953125, 70.71401977539062, 5.59796142578125, 57.00787353515625, -10.662567138671875, 167.06988525390625, 11.888076782226562, 21.735076904296875, 9.917190551757812, -9.58203125, -160.69361877441406, 0.0, 84.85052490234375, 71.69012451171875, 62.57953643798828, -26.144027709960938, 92.048828125, -107.41561889648438, 128.90093994140625, 120.94509887695312, 3.161285400390625, 126.8069076538086, -12.78021240234375, 41.2308349609375, -13.120956420898438, -71.0111083984375, -13.5223388671875, 147.76473999023438, -0.7232246398925781, 99.896484375, 30.147750854492188, 158.04733276367188, -37.26123046875, 108.453125, 11.148513793945312, 4.1604156494140625, -1.8666019439697266, 71.53890228271484, -36.486541748046875, 102.78948211669922, 53.526222229003906, 32.757843017578125, 0.261505126953125, 181.062255859375, 64.704833984375, 17.130332946777344, 123.85205078125, 117.934326171875, -29.09344482421875, 117.965576171875, -55.69348907470703, 48.33485412597656, 3.1888389587402344, 84.95700073242188, 101.62432861328125, 70.17987060546875, 38.67230987548828, 49.369972229003906, 49.31036376953125, 59.04707336425781, 142.406982421875, -32.14701843261719, -57.54689025878906, 6.928855895996094, 2.1114959716796875, 18.094390869140625, 14.693864822387695, -123.30548095703125, 149.67758178710938, 2.7207489013671875, 16.97125244140625, 167.36822509765625, 17.013153076171875, 71.85723876953125, -138.6732635498047, 7.4332275390625, 122.60546875, -49.68592834472656, 20.82427978515625, 36.10980224609375, 26.230194091796875, 27.69140625, -8.7813720703125, 64.561279296875, 124.87793731689453, 106.42233276367188, 77.54000854492188, 8.796798706054688, 27.502342224121094, -102.01824951171875, 120.8128662109375, 12.13629150390625, 255.64874267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000167.npy"}
{"epoch": 0.34973821989528797, "step": 168, "batch_size": 128, "mean": 48.118675231933594, "std": 76.88212585449219, "min": -179.8536376953125, "p10": -41.31014633178711, "median": 50.35304260253906, "p90": 150.45755920410156, "max": 242.38613891601562, "pos_frac": 0.7421875, "sample": [95.45697021484375, 24.27734375, 150.05966186523438, 146.12841796875, 176.405517578125, 65.66357421875, -179.8536376953125, 56.9990234375, 65.22653198242188, 24.52227783203125, 160.12298583984375, 1.935699462890625, -41.6190185546875, 54.17246627807617, 5.58209228515625, 7.0836639404296875, 5.819183349609375, 12.168701171875, -41.177772521972656, 106.37644958496094, 62.30443572998047, 84.2303466796875, 33.485870361328125, 55.364044189453125, 137.77633666992188, -2.065753936767578, 82.73135375976562, 73.9773941040039, -22.209213256835938, 5.097259521484375, 0.0, -9.752006530761719, 103.9708251953125, 31.62384033203125, 59.6070556640625, 45.8878173828125, 54.888946533203125, -55.29058837890625, 60.865478515625, 37.12493896484375, 103.890869140625, 37.46260070800781, 0.38580322265625, 112.75714111328125, -19.469802856445312, 205.45672607421875, 71.65663146972656, 116.92166137695312, 151.82373046875, -76.5694580078125, 184.6297607421875, -118.39791870117188, 4.940849304199219, 180.25238037109375, -98.6781005859375, 73.50914001464844, 78.9833984375, 101.41876220703125, 41.51873016357422, 84.135986328125, -127.35946655273438, -17.0494384765625, 9.2716064453125, 95.24462890625, 74.87384033203125, -0.4130592346191406, 242.38613891601562, 55.27313232421875, 187.42446899414062, 106.015380859375, 132.70941162109375, -10.049365997314453, -63.2056884765625, 95.63250732421875, 128.90206909179688, -3.493734359741211, 118.40223693847656, 52.6697998046875, 17.889427185058594, 13.19122314453125, -37.760101318359375, 14.224014282226562, 111.33258056640625, 53.13568115234375, 55.804412841796875, 213.89599609375, 88.59848022460938, 77.69851684570312, 146.45626831054688, 160.98838806152344, -135.28839111328125, -0.258880615234375, 127.86175537109375, -7.1871337890625, 25.959823608398438, 48.036285400390625, 3.655426025390625, 43.942138671875, 59.9371337890625, -121.42855834960938, 98.33512878417969, 30.328659057617188, 98.10092163085938, 151.385986328125, 161.3209228515625, 36.03607177734375, -22.114883422851562, -13.786239624023438, 32.06931686401367, 2.1375732421875, -12.462860107421875, 40.105926513671875, 43.8173828125, -78.82186889648438, 154.89659118652344, 83.85888671875, 35.197418212890625, 140.33358764648438, 66.2393798828125, 101.39476013183594, -15.94122314453125, 85.38613891601562, -44.10638427734375, -33.63616943359375, -11.905364990234375, -5.4944305419921875, 110.74371337890625, -83.745361328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000168.npy"}
{"epoch": 0.3518324607329843, "step": 169, "batch_size": 128, "mean": 58.925193786621094, "std": 78.2546615600586, "min": -146.55076599121094, "p10": -24.35380973815918, "median": 46.90281677246094, "p90": 158.55488281249998, "max": 289.8297119140625, "pos_frac": 0.7578125, "sample": [-29.11468505859375, 157.66427612304688, 149.466552734375, 104.7401123046875, -9.93280029296875, 11.510986328125, -14.24871826171875, 8.683853149414062, -15.435455322265625, 3.271270751953125, 78.96907043457031, 148.65243530273438, 128.3759765625, 139.55255126953125, -121.57208251953125, -7.5877685546875, 86.67388916015625, -31.54693603515625, 10.68998908996582, 137.68386840820312, -16.953353881835938, -40.37384796142578, 98.87921142578125, 162.01629638671875, 24.931915283203125, 149.73898315429688, 56.50164794921875, -40.926971435546875, -12.132843017578125, 147.144775390625, 135.63113403320312, 171.05178833007812, 138.22860717773438, 5.663970947265625, 1.2368011474609375, 37.435882568359375, -24.300580978393555, -146.55076599121094, 60.0120849609375, -4.94891357421875, 25.768203735351562, 18.621673583984375, 32.47972869873047, 14.372611999511719, 0.0, -49.221954345703125, -24.478010177612305, 75.39437866210938, 75.64044189453125, 6.7245941162109375, 18.8604736328125, 35.313507080078125, 200.89508056640625, 104.91909790039062, 60.43517303466797, 75.36482238769531, -50.37054443359375, 119.0384750366211, -4.155080795288086, -3.5303802490234375, 115.2845458984375, 97.40452575683594, 18.869964599609375, 9.016571044921875, 31.64837646484375, 171.90008544921875, 115.2564697265625, 217.35968017578125, 17.666259765625, 104.3642578125, 29.471786499023438, 160.63296508789062, 33.29829406738281, 54.82879638671875, 84.87020874023438, 5.7865753173828125, 23.726333618164062, 52.268218994140625, 69.26968383789062, 28.38720703125, 29.36968994140625, -13.738182067871094, 34.61872482299805, 121.32699584960938, 1.117767333984375, 77.17745971679688, 100.46612548828125, -46.887481689453125, 84.646484375, 97.3817138671875, -10.23321533203125, 202.2877197265625, 98.42735290527344, 146.05374145507812, 137.74423217773438, 34.37481689453125, 54.940277099609375, 67.60513305664062, 121.88388061523438, 5.953216552734375, 1.8563003540039062, 114.74703216552734, 289.8297119140625, 62.639495849609375, -132.36285400390625, -61.77216339111328, 129.46466064453125, -7.1305999755859375, 223.1898193359375, 55.57452392578125, 126.44326782226562, -1.9688758850097656, 166.39129638671875, 67.63589477539062, 41.53741455078125, 0.0, 12.122161865234375, -17.70709991455078, 226.314453125, 94.30252075195312, 38.02998352050781, -0.2771167755126953, 93.51101684570312, 97.48275756835938, 213.83126831054688, 227.97750854492188, 156.17581176757812, -32.094764709472656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000169.npy"}
{"epoch": 0.3539267015706806, "step": 170, "batch_size": 128, "mean": 47.885047912597656, "std": 90.65646362304688, "min": -281.3389892578125, "p10": -44.51415328979491, "median": 44.33940124511719, "p90": 158.06913757324216, "max": 432.9127197265625, "pos_frac": 0.6875, "sample": [-12.331817626953125, 80.23773193359375, 220.14317321777344, -71.77105712890625, 60.014923095703125, 59.3248291015625, -23.550079345703125, 200.00701904296875, -40.836463928222656, 43.00341796875, 97.18048095703125, 163.22811889648438, 100.41130065917969, 52.25294494628906, -281.3389892578125, 249.9805908203125, -55.796600341796875, 54.737884521484375, 19.263771057128906, 2.4428749084472656, -21.11676025390625, -1.0570068359375, 49.47601318359375, 109.76918029785156, 10.246002197265625, 3.327272415161133, -38.677223205566406, -12.773813247680664, 121.337158203125, -16.019683837890625, 123.65115356445312, 23.107690811157227, 57.39164733886719, 7.728496551513672, 3.7049407958984375, 13.265594482421875, -14.387908935546875, -15.016647338867188, 82.74609375, 134.93484497070312, 10.794815063476562, -74.54336547851562, -13.02593994140625, 53.55419921875, 38.47593688964844, 126.90328979492188, 111.56488037109375, -17.60833740234375, -27.018356323242188, 67.69343566894531, 204.00975036621094, 105.35797882080078, 197.1605224609375, -116.87580108642578, 78.75180053710938, 39.81414794921875, 248.39495849609375, -12.938385009765625, 228.17578125, 54.55351257324219, 61.4283447265625, 154.5887451171875, 104.90213012695312, -65.64002990722656, 109.78921508789062, 49.26002502441406, -12.04638671875, 41.90931701660156, 156.35659790039062, -4.773529052734375, -38.911170959472656, 233.40069580078125, 51.196205139160156, 42.7347412109375, 91.23527526855469, 84.33316040039062, 0.0, 147.13284301757812, 53.84222412109375, 105.86715698242188, -71.798828125, 77.6949462890625, 96.66558837890625, 106.46417236328125, 124.3807373046875, 110.296875, -80.81219482421875, 47.62347412109375, 95.2960205078125, -105.55587768554688, 103.63375854492188, 47.099517822265625, 13.5665283203125, 89.1639404296875, -25.70520782470703, -91.60859680175781, 85.28070068359375, 35.057647705078125, 38.50872802734375, 432.9127197265625, -2.9319305419921875, -28.201507568359375, 77.19647216796875, 13.94232177734375, -135.159423828125, 22.05474853515625, 34.4853515625, 23.74909210205078, 6.74053955078125, 72.84503173828125, -2.3956146240234375, -12.4775390625, 140.44818115234375, -53.095428466796875, -40.09358215332031, -18.096405029296875, 48.654144287109375, 95.3465576171875, 9.077434539794922, -2.3883819580078125, -84.76791381835938, 45.675384521484375, 184.49923706054688, 162.0650634765625, -1.8914413452148438, 55.12443542480469, 0.94183349609375, 181.73382568359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000170.npy"}
{"epoch": 0.35602094240837695, "step": 171, "batch_size": 128, "mean": 58.5494384765625, "std": 77.75472259521484, "min": -123.92391967773438, "p10": -41.169744873046874, "median": 52.5262451171875, "p90": 160.85452575683593, "max": 283.40753173828125, "pos_frac": 0.7578125, "sample": [170.86727905273438, 55.763671875, 78.4652099609375, 33.727294921875, 181.896484375, 99.00407409667969, 166.0741729736328, -114.27459716796875, -65.96701049804688, 17.801406860351562, -59.33892822265625, 18.275100708007812, 149.80967712402344, -34.99586486816406, 39.09210205078125, 65.51810455322266, -25.036134719848633, 92.3646240234375, 48.400390625, 252.20440673828125, 84.73712158203125, -55.830257415771484, 43.11138916015625, 160.14385986328125, 38.7806396484375, 51.458160400390625, -33.69256591796875, 63.543212890625, -41.12786865234375, 38.24395751953125, -62.226905822753906, 129.50839233398438, -13.58782958984375, 69.27699279785156, 96.45660400390625, 91.71450805664062, 98.48307800292969, 82.83334350585938, 20.0035400390625, -46.73902893066406, 52.684722900390625, -72.89154052734375, -49.33323669433594, 20.632705688476562, 34.33357238769531, 93.76995849609375, 147.51385498046875, -25.982093811035156, -41.2674560546875, 102.29837036132812, 185.28427124023438, 114.00604248046875, 39.47515869140625, 216.86972045898438, 72.8592529296875, 60.76069641113281, 78.78964233398438, 160.53787231445312, 283.40753173828125, -74.69619750976562, 115.66757202148438, 44.42420196533203, 60.367889404296875, 131.32699584960938, 66.12797546386719, 161.5933837890625, 106.66595458984375, 106.59315490722656, 172.78643798828125, 122.54412841796875, 21.689407348632812, 140.16220092773438, 55.115447998046875, 17.4635009765625, -18.58539581298828, 4.0568084716796875, 148.89236450195312, 5.548007965087891, 28.531585693359375, 50.75209045410156, 0.07583808898925781, 3.5001144409179688, -81.61578369140625, 57.656219482421875, 37.1461181640625, -123.92391967773438, 149.7874298095703, 63.706329345703125, -11.166351318359375, 140.01715087890625, 85.9481201171875, 177.94107055664062, -79.90545654296875, -3.79022216796875, 107.248046875, 181.02706909179688, 87.12603759765625, -7.646968841552734, 231.1741943359375, -18.801025390625, 52.367767333984375, -10.97998046875, -3.332061767578125, 27.040985107421875, 68.55097961425781, -30.57586669921875, 90.130126953125, 130.60250854492188, 46.40165710449219, -9.604206085205078, 48.4935302734375, 32.841949462890625, 117.66770935058594, 131.61383056640625, 38.10969543457031, 91.22003173828125, 51.44317626953125, 38.648277282714844, 143.6436004638672, 26.750015258789062, 117.93949890136719, 119.74739837646484, -5.764667510986328, 52.03178024291992, -34.85572814941406, 167.91204833984375, 99.59434509277344, -24.332763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000171.npy"}
{"epoch": 0.3581151832460733, "step": 172, "batch_size": 128, "mean": 31.991592407226562, "std": 75.55733489990234, "min": -163.64990234375, "p10": -62.68626251220703, "median": 27.620376586914062, "p90": 128.49864807128904, "max": 206.0445556640625, "pos_frac": 0.6875, "sample": [39.654327392578125, -64.0142822265625, -62.11711120605469, 27.30169677734375, 116.80397033691406, 0.480865478515625, -1.372711181640625, -46.91288757324219, 58.84539794921875, 11.066986083984375, 2.020294189453125, 36.77092742919922, -109.18475341796875, -3.5993289947509766, 205.18153381347656, 20.181100845336914, 97.6534423828125, 149.42507934570312, 167.3238067626953, -4.8653106689453125, -0.648529052734375, -40.91070556640625, 61.31396484375, 164.03427124023438, -10.6453857421875, 25.246551513671875, 83.07943725585938, 5.815399169921875, -31.033294677734375, 69.88031005859375, 16.4681396484375, 141.21176147460938, -117.447509765625, 4.77105712890625, 72.9608154296875, 57.583953857421875, 13.7349853515625, 80.442138671875, -91.7320327758789, 1.1739654541015625, 18.351791381835938, 106.05503845214844, -96.39321899414062, 1.366811752319336, 119.67198944091797, -163.64990234375, 29.8734130859375, 96.90452575683594, 135.8541259765625, 43.491424560546875, -2.516693115234375, -41.21392822265625, 11.304412841796875, -27.89678192138672, -39.1015625, 19.400177001953125, 109.23541259765625, 117.70378112792969, -155.02880859375, 20.075271606445312, -21.7445068359375, 38.1171875, 66.00863647460938, -4.1157989501953125, 206.0445556640625, 103.88153076171875, 24.2657470703125, 93.10067749023438, 93.08296203613281, 60.34461212158203, 161.50732421875, -73.64230346679688, -52.244140625, 60.821533203125, 154.048828125, 162.42922973632812, 76.13958740234375, -22.04796600341797, 62.3037109375, -92.1292724609375, 36.8265380859375, 46.93000030517578, -21.20992660522461, -131.45294189453125, -11.56060791015625, 31.75775146484375, -27.717788696289062, 124.4005126953125, 25.604644775390625, -52.09593200683594, 126.63839721679688, 1.8287162780761719, 86.82196044921875, -69.83795166015625, -118.05488586425781, 156.9257049560547, -21.719940185546875, 21.83514404296875, -58.271812438964844, 30.993202209472656, -20.69036865234375, 132.8392333984375, 27.939056396484375, 65.03411865234375, 135.0109405517578, 35.440032958984375, -1.6302490234375, 94.20193481445312, 123.99419403076172, 26.56378173828125, 52.34259033203125, 81.88520050048828, 41.142860412597656, 59.09033203125, 95.86785888671875, 113.73666381835938, -16.5572509765625, 39.02203369140625, 45.585731506347656, 113.31655883789062, -147.32928466796875, 106.09942626953125, 14.345428466796875, 16.573272705078125, 0.0, 72.46527099609375, 59.06500244140625, 5.32672119140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000172.npy"}
{"epoch": 0.36020942408376966, "step": 173, "batch_size": 128, "mean": 43.928924560546875, "std": 80.18257904052734, "min": -133.7584228515625, "p10": -55.50082092285156, "median": 31.459701538085938, "p90": 148.86680450439454, "max": 316.59820556640625, "pos_frac": 0.6796875, "sample": [-66.0352783203125, -133.7584228515625, -114.3117904663086, 77.8349609375, 88.050537109375, -12.585769653320312, 74.11370849609375, -104.03659057617188, 170.33319091796875, 316.59820556640625, 21.459869384765625, 41.28338623046875, 13.187358856201172, 56.08349609375, 78.29058837890625, 161.77975463867188, 7.177947998046875, 32.04515838623047, 144.08319091796875, 188.50115966796875, 93.73648071289062, -5.1368408203125, 7.69537353515625, 49.9556884765625, 29.6475830078125, 12.1541748046875, -65.07270812988281, 36.675506591796875, 83.22366333007812, 107.90855407714844, -68.80938720703125, 0.0137939453125, 7.1654205322265625, 8.172332763671875, -4.49603271484375, 54.00215148925781, -57.452117919921875, 55.96881103515625, 22.4284725189209, 96.9659423828125, 103.24708557128906, 87.7000732421875, 42.78265380859375, 75.30633544921875, 191.00442504882812, 61.316162109375, -20.095901489257812, -29.252243041992188, 26.31829071044922, 25.84490966796875, 31.071945190429688, -4.660697937011719, -99.32150268554688, 206.684326171875, 40.413848876953125, -56.15264892578125, -51.41161346435547, -55.221466064453125, 95.02550506591797, 148.86065673828125, 108.0002670288086, 122.97073364257812, -6.7257537841796875, 148.8811492919922, 132.29132080078125, 98.3173828125, -81.60018920898438, 23.656036376953125, 4.8759002685546875, 147.0392303466797, 13.66162109375, -11.600563049316406, 185.71444702148438, 109.76708221435547, 34.38531494140625, 94.8284912109375, 4.691490173339844, 252.46847534179688, -1.056243896484375, 48.1461181640625, 135.072998046875, -23.26318359375, 97.99116516113281, 23.76922607421875, 44.503814697265625, 19.854354858398438, 31.847457885742188, 122.25440979003906, 6.0745849609375, 35.930450439453125, 88.57684326171875, -9.2440185546875, 121.80781555175781, 66.80886840820312, 248.027099609375, 142.26638793945312, 149.16253662109375, 44.36700439453125, 159.25189208984375, 119.67935180664062, 93.98945617675781, -51.994537353515625, -1.092966079711914, -71.41366577148438, -2.2430171966552734, 120.56083679199219, -42.837890625, 168.717041015625, -15.189361572265625, 66.49560546875, -20.015274047851562, 2.656829833984375, -72.9622802734375, -36.096527099609375, -11.980096817016602, -19.956695556640625, -9.250152587890625, -66.66098022460938, 138.4990234375, -11.3424072265625, 34.179443359375, -1.665435791015625, 0.0, 72.41581726074219, 2.159637451171875, -16.346649169921875, -28.42822265625, 24.951766967773438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000173.npy"}
{"epoch": 0.362303664921466, "step": 174, "batch_size": 128, "mean": 53.3561897277832, "std": 70.12322235107422, "min": -172.98214721679688, "p10": -22.46402893066406, "median": 52.837646484375, "p90": 145.1211669921875, "max": 224.15353393554688, "pos_frac": 0.8125, "sample": [175.84544372558594, 1.0809326171875, 4.824947357177734, 4.457191467285156, 146.518310546875, 146.27325439453125, 129.95697021484375, -172.98214721679688, -12.111541748046875, -23.389450073242188, 110.93890380859375, 67.10208129882812, -5.060821533203125, 91.08309173583984, 49.82318115234375, 65.93939208984375, 122.23226928710938, 116.44183349609375, 50.240570068359375, 52.41288757324219, -61.05937194824219, 102.77545166015625, -23.284393310546875, 176.81875610351562, 145.4605712890625, -7.49993896484375, 87.67050170898438, 59.310333251953125, 92.72920227050781, 84.14642333984375, 21.901596069335938, 31.637298583984375, 157.67868041992188, 15.701171875, -51.512733459472656, 20.953643798828125, 110.48193359375, 110.050537109375, 149.11752319335938, 74.7109375, 11.14706039428711, 55.693519592285156, 22.84229278564453, 68.15383911132812, 1.3752174377441406, 206.46484375, 64.44589233398438, 143.000244140625, 20.8115234375, 167.72744750976562, 0.4798583984375, -1.566925048828125, 82.37869262695312, 83.8331298828125, -32.228912353515625, 74.65048217773438, -3.053417205810547, 2.1919326782226562, 55.577392578125, 30.2152099609375, 5.081695556640625, 3.9237899780273438, 151.80950927734375, 224.15353393554688, 28.143341064453125, -22.192657470703125, 53.26240539550781, 69.70315551757812, -8.03466796875, 101.7420654296875, 58.28220748901367, 127.87074279785156, 139.51953125, 128.05584716796875, 144.9757080078125, 76.25474548339844, 101.91862487792969, -23.09722900390625, 129.0028076171875, -166.66461181640625, -41.49470520019531, 46.521583557128906, 25.533035278320312, 124.435791015625, 44.80778503417969, 5.386999130249023, 66.30026245117188, 83.51215362548828, 31.4678955078125, 26.82909393310547, -10.294136047363281, 30.123046875, 58.414276123046875, 71.84925079345703, 110.13433837890625, 138.9273681640625, 3.78143310546875, 187.75778198242188, 58.99842071533203, -12.650283813476562, 104.16950988769531, -49.28074645996094, 115.06576538085938, 32.7801513671875, 50.799407958984375, 17.544448852539062, -1.5111160278320312, 17.376220703125, 10.026611328125, -64.18942260742188, 71.29673767089844, 41.7845458984375, 13.670177459716797, 112.4075698852539, 73.69873046875, -101.94781494140625, 114.71432495117188, -21.01922607421875, 41.2535400390625, -127.21269226074219, 9.992935180664062, 71.01324462890625, 165.735595703125, 66.73580932617188, 119.00387573242188, 40.22508239746094, 32.7210693359375, 25.131134033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000174.npy"}
{"epoch": 0.3643979057591623, "step": 175, "batch_size": 128, "mean": 48.05918884277344, "std": 91.83658599853516, "min": -177.57247924804688, "p10": -52.574484252929686, "median": 27.8167724609375, "p90": 159.38633422851564, "max": 477.9869384765625, "pos_frac": 0.7109375, "sample": [78.99777221679688, 166.68154907226562, 73.9569091796875, 2.6735076904296875, 12.452491760253906, -74.88699340820312, 40.42437744140625, 100.8922119140625, 115.50363159179688, -78.18447875976562, -27.423049926757812, 27.81610107421875, 58.97357177734375, 0.2555999755859375, 93.2328109741211, 477.9869384765625, 5.915193557739258, 138.69581604003906, 119.23292541503906, 165.1931610107422, -61.37750244140625, 127.90164184570312, 169.99459838867188, -2.408843994140625, 36.96051025390625, -9.164764404296875, -20.218406677246094, -82.6171875, -1.876922607421875, -22.69940185546875, 34.482460021972656, -18.89429473876953, 27.81744384765625, 83.36148071289062, -110.31878662109375, 159.97061157226562, 15.933563232421875, 58.004852294921875, 89.52874755859375, 109.57635498046875, 15.974609375, 9.7862548828125, 26.922149658203125, 43.79949951171875, 25.80096435546875, 148.6329345703125, 96.87884521484375, 156.9603271484375, -16.13037109375, 157.6273193359375, 13.80096435546875, 58.03978729248047, -7.62481689453125, 4.418304443359375, -65.38009643554688, -116.5382080078125, 37.838897705078125, 13.536468505859375, -25.886703491210938, 107.09783935546875, 1.2635383605957031, 159.5986328125, 105.23573303222656, 64.3984146118164, 93.65231323242188, 38.588836669921875, 14.53753662109375, 2.008237838745117, 9.776702880859375, 1.961212158203125, 48.8583984375, -8.128808975219727, 9.889404296875, 168.58526611328125, 140.00851440429688, -85.26773071289062, -7.541900634765625, 191.81529235839844, 0.0, -177.57247924804688, 155.4482421875, 60.32269287109375, 296.1474609375, 18.7225341796875, -54.8123779296875, -17.389801025390625, -13.017333984375, -36.4119873046875, -63.84222412109375, 69.85684204101562, 87.32870483398438, 6.125091552734375, 159.29534912109375, 100.16937255859375, 135.9638671875, 266.17596435546875, 23.618682861328125, 6.449728012084961, -51.615386962890625, 7.4632415771484375, 136.9338836669922, 65.0577621459961, 64.67219543457031, 35.628936767578125, -26.245269775390625, -80.70440673828125, 106.25997924804688, -16.969146728515625, 15.258697509765625, 160.37188720703125, 69.08049774169922, 145.94488525390625, 83.527099609375, -39.44622802734375, 163.07052612304688, -19.8717041015625, 18.158851623535156, 148.98370361328125, 65.8978500366211, -37.25958251953125, 39.640228271484375, 26.32183837890625, 112.35629272460938, -12.26947021484375, -159.165771484375, 0.0, 85.61808776855469, 305.1575622558594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000175.npy"}
{"epoch": 0.36649214659685864, "step": 176, "batch_size": 128, "mean": 37.14612579345703, "std": 90.58794403076172, "min": -194.3504638671875, "p10": -81.18084106445312, "median": 34.1590576171875, "p90": 153.54898681640626, "max": 373.8965759277344, "pos_frac": 0.703125, "sample": [-87.40478515625, 24.2877197265625, 76.92202758789062, -27.818222045898438, 66.79000854492188, 50.78295135498047, -138.609130859375, -194.3504638671875, 89.35062408447266, 59.3414306640625, -127.00811767578125, 95.07365417480469, 121.90579223632812, 109.426513671875, 18.677215576171875, 48.712860107421875, 65.66191101074219, 5.690439224243164, 1.17791748046875, -41.77465057373047, -59.565948486328125, 23.606689453125, 0.5433216094970703, -98.31658935546875, 75.60659790039062, 208.96578979492188, 79.13502502441406, 126.89703369140625, 37.7841796875, -30.900360107421875, 85.7608642578125, -94.1982421875, -4.507072448730469, 36.0155029296875, 123.91336822509766, 25.46673583984375, 28.286376953125, 154.42909240722656, 33.698883056640625, -2.2207183837890625, -92.64041137695312, -1.0860671997070312, 59.207672119140625, -27.190277099609375, 29.9710693359375, 198.79766845703125, 27.247650146484375, -6.072521209716797, 33.064849853515625, 186.53741455078125, -58.646270751953125, 116.9563217163086, 47.1400146484375, -76.22157287597656, 25.389129638671875, 107.55987548828125, -32.16326904296875, 41.84040069580078, 49.394683837890625, 12.866519927978516, 56.446136474609375, 124.44233703613281, -0.8868408203125, -31.367916107177734, 126.06478881835938, 155.70361328125, 160.33551025390625, -21.54913330078125, 60.508636474609375, 10.39678955078125, -44.880126953125, 210.0865020751953, 68.37590026855469, -43.240570068359375, 373.8965759277344, 32.187957763671875, 14.612884521484375, -69.68380737304688, 74.20748901367188, -0.690673828125, 75.69610595703125, 74.41526794433594, 14.094749450683594, 73.98574829101562, 80.70588684082031, 124.48131561279297, 15.798233032226562, 35.22772216796875, -81.5667724609375, 178.447265625, 49.46632385253906, 83.23614501953125, -71.99542236328125, 82.62393188476562, -52.47880554199219, -7.165580749511719, 115.9913330078125, -103.52703857421875, -81.01544189453125, 50.1197509765625, -120.26233673095703, 148.39968872070312, 0.15770721435546875, 9.716278076171875, 153.1717987060547, -143.92730712890625, 76.40335083007812, 78.17919921875, -115.59764099121094, 4.99603271484375, 102.89840698242188, 21.622642517089844, 274.94207763671875, 119.56997680664062, 206.69027709960938, 79.31804656982422, 16.96807861328125, 43.319976806640625, -45.15228271484375, -191.45565795898438, 175.17575073242188, 34.619232177734375, 21.572481155395508, 69.61363983154297, 70.171630859375, 158.7089385986328, 17.316993713378906, -33.12865447998047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000176.npy"}
{"epoch": 0.36858638743455496, "step": 177, "batch_size": 128, "mean": 38.615257263183594, "std": 89.39845275878906, "min": -197.2039794921875, "p10": -59.669613647460935, "median": 26.819570541381836, "p90": 153.84076232910155, "max": 264.346923828125, "pos_frac": 0.703125, "sample": [116.80233764648438, 8.689992904663086, -5.34814453125, 181.0883331298828, 20.630874633789062, 86.5816650390625, 90.561767578125, -19.006622314453125, 150.46505737304688, 2.8862991333007812, 96.8876953125, -30.324642181396484, -197.2039794921875, -31.638599395751953, -35.4068603515625, -54.32881164550781, 23.239791870117188, 82.20126342773438, 118.54557800292969, 43.38494873046875, 25.140533447265625, 112.54241943359375, 56.65020751953125, 91.92864990234375, -60.197113037109375, 52.245849609375, -24.056045532226562, 133.62890625, 67.22442626953125, 112.531982421875, 15.041641235351562, 48.07200622558594, -168.39773559570312, 261.7698669433594, -46.61799621582031, 147.05047607421875, 164.40155029296875, 28.76408576965332, -7.367883682250977, 19.945831298828125, 149.93894958496094, -21.031982421875, 9.278200149536133, -64.57579040527344, -62.46533203125, -90.16956329345703, 7.103689193725586, 13.561695098876953, 76.8824462890625, 52.01823425292969, -20.632568359375, 12.727462768554688, 264.346923828125, 61.85081481933594, 28.234848022460938, 110.36970520019531, 34.98828125, -15.37506103515625, 100.03994750976562, -5.0306396484375, -88.57754516601562, 260.2821044921875, 3.101165771484375, -1.3638916015625, 112.39474487304688, 50.81341552734375, -53.59031677246094, 0.60577392578125, 34.58201599121094, 123.1265869140625, -28.944976806640625, 162.308837890625, -2.7132568359375, 8.448507308959961, -2.113758087158203, 61.384796142578125, 109.39205932617188, -193.8014373779297, 215.0465087890625, 10.197845458984375, 11.961151123046875, -120.55877685546875, 12.768745422363281, 72.3822021484375, -103.3265380859375, 9.4332275390625, 65.62881469726562, 2.3702392578125, 45.0908203125, 25.85614776611328, 102.0281982421875, 96.21429443359375, 89.91934204101562, 27.538433074951172, 26.1007080078125, -16.5377197265625, 79.00210571289062, 12.9345703125, 169.70538330078125, 4.844715118408203, -10.184677124023438, 2.3482818603515625, 35.266693115234375, 168.68603515625, 9.67115592956543, -113.0797119140625, 2.814849853515625, -39.69578552246094, -175.02294921875, -168.74819946289062, -50.704193115234375, 151.78451538085938, 127.92005157470703, 131.44540405273438, 35.22655487060547, 158.638671875, 192.9407958984375, 85.89102935791016, 193.52142333984375, 45.02374267578125, 127.6177978515625, -1.1180801391601562, -59.44354248046875, 49.67319107055664, -1.0277099609375, 54.048919677734375, 67.08059692382812, 207.17282104492188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000177.npy"}
{"epoch": 0.3706806282722513, "step": 178, "batch_size": 128, "mean": 55.74102020263672, "std": 86.89921569824219, "min": -143.453125, "p10": -54.01815643310547, "median": 38.86302947998047, "p90": 170.2355026245117, "max": 378.758056640625, "pos_frac": 0.71875, "sample": [111.22349548339844, 4.774440765380859, -6.6953887939453125, 35.98284912109375, 0.3994598388671875, 4.766157150268555, 92.34808349609375, 174.01190185546875, -15.719547271728516, 76.20817565917969, 235.64810180664062, 86.87484741210938, 96.33270263671875, 96.0938720703125, -73.68753051757812, 187.98895263671875, 0.0, 191.46640014648438, 200.52227783203125, 142.5211639404297, 129.4842529296875, 204.01165771484375, 14.787622451782227, 157.53582763671875, 115.78997802734375, -107.96914672851562, -58.478302001953125, 183.18414306640625, -14.436325073242188, 136.95361328125, -5.980367660522461, -115.22408294677734, 15.653564453125, 131.92501831054688, -56.505615234375, 94.48773193359375, 168.8887939453125, 229.21829223632812, -51.32305908203125, 140.82366943359375, 78.63278198242188, 17.302291870117188, 81.65634155273438, 125.25595092773438, 188.82098388671875, 94.50299072265625, -41.99021911621094, -1.0478668212890625, -5.407989501953125, 118.79580688476562, 120.04948425292969, 160.5576171875, -41.246856689453125, -143.453125, -18.839828491210938, -24.3624267578125, 33.424072265625, 94.23199462890625, 39.96800231933594, 20.950958251953125, 2.3786277770996094, 159.01727294921875, 48.522918701171875, 88.5205078125, 37.758056640625, 63.456756591796875, 65.0111083984375, 94.7705078125, 92.31600952148438, 51.90843200683594, 150.4859619140625, 24.129501342773438, -7.661582946777344, 14.391815185546875, -8.764530181884766, 223.28414916992188, -57.70379638671875, 31.485366821289062, -66.00021362304688, 19.34814453125, 54.865234375, 24.155227661132812, 73.25462341308594, -84.7239990234375, -91.7943115234375, -1.619842529296875, 378.758056640625, 78.8458251953125, 3.8885421752929688, 76.22161865234375, 17.486480712890625, 69.10368347167969, 37.360595703125, 0.7438507080078125, 108.80081939697266, 136.82015991210938, -65.85447692871094, -12.648025512695312, 4.578300476074219, 146.47323608398438, 29.237083435058594, 2.7931289672851562, -13.369544982910156, 136.0794677734375, 77.57875061035156, -59.49049377441406, -11.884513854980469, 173.37782287597656, 140.75491333007812, -52.95210266113281, 1.1826324462890625, 192.2322998046875, -18.623016357421875, 16.573333740234375, 126.78384399414062, 4.58331298828125, 44.81918716430664, 150.029541015625, -6.795783996582031, -11.302978515625, 48.808837890625, 35.472412109375, 32.1241455078125, -45.15472412109375, -60.583770751953125, 86.43234252929688, 117.24549865722656, 165.8397674560547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000178.npy"}
{"epoch": 0.37277486910994767, "step": 179, "batch_size": 128, "mean": 48.41162109375, "std": 86.52859497070312, "min": -200.7635498046875, "p10": -53.83055419921873, "median": 42.70685005187988, "p90": 175.3787139892578, "max": 237.23831176757812, "pos_frac": 0.7578125, "sample": [71.59512329101562, 8.635007858276367, 111.19140625, 6.1678466796875, 63.500823974609375, 185.98312377929688, 40.9495849609375, 31.018287658691406, -76.995849609375, 17.03411865234375, 104.01992797851562, -29.11419677734375, 164.6917724609375, -20.73321533203125, 195.92367553710938, -46.982330322265625, 76.6849365234375, 77.00067138671875, 39.898193359375, -9.176952362060547, 51.95556640625, 76.00566101074219, 9.148307800292969, 112.46519470214844, 33.60589599609375, 6.449451446533203, -49.49835205078125, 3.035005569458008, 49.59320831298828, -108.98820495605469, 109.32811737060547, 1.9061660766601562, 3.0253982543945312, -13.045654296875, 77.43563842773438, 67.669921875, 113.1898193359375, 92.64208984375, 52.91314697265625, 12.085575103759766, -63.93902587890625, -72.76284790039062, 34.56141662597656, 117.03744506835938, -10.9991455078125, 81.57147216796875, 106.45735931396484, 1.9478626251220703, 147.41900634765625, -149.400634765625, -14.539794921875, -81.69583129882812, 81.77568054199219, 11.002815246582031, 19.913814544677734, 189.67803955078125, 174.21417236328125, 197.71873474121094, 120.361572265625, -33.571990966796875, -71.0648193359375, -12.121673583984375, 194.12640380859375, 48.76295471191406, -5.793224334716797, 54.952301025390625, 237.23831176757812, -143.658447265625, 24.263641357421875, 70.53338623046875, 43.401126861572266, 71.27285766601562, 116.610107421875, 69.52130126953125, 55.791015625, 9.814958572387695, 125.20004272460938, 21.615966796875, 44.23396301269531, 42.0125732421875, 6.155080795288086, 28.437408447265625, 127.432861328125, 29.42902374267578, 152.62042236328125, -23.76025390625, 230.26438903808594, 182.1163330078125, 26.955066680908203, -107.73562622070312, 155.6199951171875, 1.7881965637207031, 235.283447265625, 2.743927001953125, 62.781219482421875, -21.081298828125, 91.60708618164062, 108.1728515625, 50.270599365234375, 26.710601806640625, 213.352294921875, 58.377288818359375, 110.45309448242188, 55.36456298828125, -200.7635498046875, 205.09759521484375, -133.78518676757812, 10.501754760742188, 2.4481658935546875, 127.70628356933594, 20.20099639892578, 90.171630859375, -136.261474609375, 178.09597778320312, -76.3116455078125, -24.35810089111328, 86.68366241455078, 107.04721069335938, -17.73870849609375, 13.256927490234375, 216.58343505859375, 137.41033935546875, -10.148468017578125, 3.55804443359375, -2.27001953125, -25.230987548828125, 142.56553649902344, 113.19226837158203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000179.npy"}
{"epoch": 0.374869109947644, "step": 180, "batch_size": 128, "mean": 54.568145751953125, "std": 85.422119140625, "min": -122.80526733398438, "p10": -28.08738594055175, "median": 38.32444381713867, "p90": 156.7189819335937, "max": 504.6225891113281, "pos_frac": 0.7265625, "sample": [-1.4714088439941406, -4.703649520874023, -116.99314880371094, -5.894020080566406, 29.16156005859375, -3.3022613525390625, 103.63873291015625, 29.7855224609375, 27.419921875, 176.79026794433594, -1.787261962890625, 136.40960693359375, 143.69906616210938, -122.80526733398438, -22.833038330078125, 178.23023986816406, -26.6307373046875, 65.52716064453125, 77.76779174804688, -90.27163696289062, 68.17927551269531, 153.58529663085938, 61.58968734741211, 50.86328125, 48.30609130859375, 52.06524658203125, 22.18756103515625, 193.74151611328125, 138.16046142578125, -31.48623275756836, 13.642745971679688, 0.0695343017578125, 111.18353271484375, 95.37664794921875, 211.76199340820312, 22.304832458496094, 208.37710571289062, 26.981552124023438, -0.3840217590332031, 87.48388671875, 122.77865600585938, -14.694168090820312, 80.1219253540039, 9.809576034545898, 191.55377197265625, 89.48590087890625, -96.71304321289062, 91.51734924316406, -0.648681640625, 36.918243408203125, 38.50141906738281, -12.541244506835938, -8.592681884765625, 22.842529296875, 93.2154541015625, 24.9676513671875, 107.76590728759766, 80.52035522460938, 110.55596923828125, 88.63330841064453, 190.65008544921875, 30.1485595703125, -54.462982177734375, 146.19198608398438, 164.03091430664062, -80.9119873046875, -9.89215087890625, 35.44610595703125, -79.24755859375, 134.40872192382812, 103.8740234375, -73.59819030761719, 38.14746856689453, 184.9324951171875, 185.1322479248047, 21.402923583984375, -22.41861343383789, -3.137666702270508, 136.39639282226562, 31.309890747070312, 188.2745361328125, 100.14994812011719, -43.283843994140625, 65.5281982421875, 125.1051025390625, 89.51575469970703, 54.10424041748047, -94.4266357421875, 25.779251098632812, 119.27737426757812, -22.202056884765625, 84.65911865234375, -7.259929656982422, 60.11053466796875, 97.564208984375, 85.43394470214844, 58.80499267578125, 40.811065673828125, -7.734338760375977, 53.967254638671875, 1.7719192504882812, 137.75857543945312, 115.47885131835938, 94.91244506835938, 5.186433792114258, -7.876750946044922, 56.34613800048828, -10.551422119140625, 504.6225891113281, 22.965240478515625, -32.4619140625, 7.460796356201172, 9.030624389648438, -17.11962890625, 103.33474731445312, 27.27191162109375, 46.09429931640625, 73.85638427734375, -62.311492919921875, 108.90802001953125, 8.888830184936523, 102.6806640625, 24.335662841796875, 29.999008178710938, 36.40531921386719, 289.145751953125, 9.737457275390625, -15.45501708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000180.npy"}
{"epoch": 0.3769633507853403, "step": 181, "batch_size": 128, "mean": 44.40808868408203, "std": 88.05331420898438, "min": -186.52792358398438, "p10": -59.25728149414062, "median": 38.72007751464844, "p90": 156.69759826660155, "max": 313.6751708984375, "pos_frac": 0.6953125, "sample": [-1.2947311401367188, 0.19181060791015625, -60.812713623046875, -33.25933837890625, 0.0, 52.51268005371094, 193.19122314453125, 27.10003662109375, 54.34629821777344, 1.456756591796875, 60.28369140625, -15.279197692871094, 148.5985107421875, -9.478584289550781, 164.7281494140625, 24.787109375, 0.0, 125.63600158691406, 4.173900604248047, -58.503265380859375, 150.0227508544922, 42.199249267578125, 212.863525390625, 115.85281372070312, -134.3611602783203, 171.45054626464844, -91.31829833984375, 110.09132385253906, -6.358848571777344, 16.296478271484375, 145.037353515625, -20.557113647460938, 18.466665267944336, 164.4961700439453, 25.86553955078125, 18.964431762695312, 154.72039794921875, 30.13031005859375, 55.500396728515625, 114.98300170898438, 2.0924453735351562, 130.85488891601562, -126.162841796875, -7.533063888549805, 47.14642333984375, 133.03469848632812, 64.0788345336914, -103.44931030273438, 240.17120361328125, -55.081153869628906, 64.3416748046875, -9.96014404296875, 34.96046447753906, 109.4273681640625, -78.78646850585938, -121.17547607421875, -28.594268798828125, 24.281982421875, 12.08807373046875, 60.9935302734375, 31.286865234375, -1.5239639282226562, -39.045654296875, 114.74586486816406, 47.49763488769531, 4.661945343017578, -45.28363037109375, 167.40411376953125, 14.87689208984375, 161.31106567382812, 68.41921997070312, 31.418947219848633, 51.98851013183594, 137.5040283203125, 248.31756591796875, 247.9776611328125, -186.52792358398438, 96.57249450683594, -49.5010986328125, 50.8753662109375, -34.278839111328125, 60.26727294921875, -23.610815048217773, -58.590667724609375, -28.278091430664062, 165.721435546875, 77.69912719726562, -61.86592102050781, 54.78094482421875, 4.70806884765625, -21.152679443359375, 65.15814208984375, 149.56161499023438, -115.54122924804688, 33.23115539550781, 89.0582275390625, 121.31387329101562, 110.12576293945312, 66.00849914550781, 84.4459228515625, 313.6751708984375, 29.555648803710938, 43.658538818359375, 110.0240478515625, -36.34307861328125, -50.09300994873047, 12.018341064453125, 135.03570556640625, -89.3826904296875, -142.4775390625, 124.406494140625, 35.24090576171875, 62.696189880371094, -11.632080078125, 74.05636596679688, 168.97349548339844, -33.514427185058594, 23.918212890625, 46.693328857421875, 91.16064453125, 55.37054443359375, 97.2216796875, 95.44122314453125, -85.08676147460938, 23.519622802734375, 103.150390625, 110.90160369873047, 118.85612487792969], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000181.npy"}
{"epoch": 0.37905759162303665, "step": 182, "batch_size": 128, "mean": 63.763023376464844, "std": 86.21107482910156, "min": -156.71600341796875, "p10": -34.265618896484376, "median": 58.44715118408203, "p90": 172.04954528808588, "max": 304.8025817871094, "pos_frac": 0.765625, "sample": [55.064361572265625, 125.21989440917969, 70.38327026367188, -0.5136508941650391, 39.71171569824219, 39.94505310058594, 198.46340942382812, 107.34835815429688, 114.38226318359375, -32.909393310546875, 102.60220336914062, 96.8565673828125, 49.06591796875, 66.6749267578125, 97.16207885742188, 11.42095947265625, 75.8063735961914, 123.08453369140625, 117.0406494140625, 148.90927124023438, -6.862945556640625, 7.5063018798828125, 61.162811279296875, 271.7349853515625, 88.48095703125, 153.78497314453125, 12.655517578125, 103.69633483886719, 218.94818115234375, 136.87429809570312, 116.58633422851562, 21.986183166503906, 1.0910797119140625, 92.25582885742188, 22.379150390625, 9.414970397949219, -99.95709228515625, -35.398590087890625, 9.188560485839844, -44.760528564453125, -41.09954833984375, 124.10031127929688, -149.15707397460938, 41.79228210449219, 217.484619140625, 8.978523254394531, 137.3447265625, -72.81224060058594, 165.33767700195312, 187.7105712890625, 116.908203125, 246.7110137939453, 136.15087890625, 304.8025817871094, 142.68582153320312, -33.780059814453125, 50.98332214355469, 36.692848205566406, 30.71612548828125, 82.56341552734375, 71.92202758789062, 91.7318115234375, 92.20465087890625, 112.57670593261719, 196.01254272460938, -32.647926330566406, 114.27944946289062, 135.79141235351562, 194.86709594726562, 2.3725662231445312, -156.71600341796875, 7.374225616455078, 64.05303955078125, 60.56288146972656, 0.440521240234375, -18.66693115234375, 105.1446533203125, 111.52472686767578, 2.5611419677734375, -3.86566162109375, 26.817825317382812, 68.66877746582031, -4.4254150390625, -16.356414794921875, -47.531982421875, -5.900356292724609, 113.90324401855469, -21.114181518554688, 73.27578735351562, -31.216522216796875, 123.4114990234375, 103.38006591796875, 121.28890991210938, 164.90017700195312, -27.57781982421875, -12.604377746582031, 137.42715454101562, 108.11299133300781, 56.3314208984375, 91.44247436523438, -51.61775207519531, 6.76043701171875, -22.845054626464844, 134.26736450195312, 5.168365478515625, 111.18539428710938, 30.395172119140625, 50.951690673828125, 201.51431274414062, 15.0244140625, 3.650543212890625, 150.3206024169922, 232.944580078125, -92.52273559570312, -36.59656524658203, 6.7642822265625, 15.176193237304688, 156.60528564453125, 47.04978942871094, -32.36834716796875, -10.81549072265625, 141.34259033203125, 215.6302490234375, -49.76617431640625, 246.21304321289062, 46.9100341796875, 46.19659423828125, -60.22819519042969], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000182.npy"}
{"epoch": 0.381151832460733, "step": 183, "batch_size": 128, "mean": 55.20054626464844, "std": 92.3909683227539, "min": -147.912841796875, "p10": -66.11729888916015, "median": 40.202911376953125, "p90": 177.74935607910157, "max": 381.53167724609375, "pos_frac": 0.7421875, "sample": [-60.31463623046875, 121.29550170898438, -16.648849487304688, 69.83851623535156, -70.64366149902344, 201.6187744140625, -100.29812622070312, 117.8770751953125, 166.77120971679688, 5.611717224121094, -59.5126953125, 50.001869201660156, 229.4442138671875, 9.356170654296875, -17.95062255859375, -115.76226806640625, -16.68718719482422, 225.5840606689453, 25.36932373046875, -76.7784423828125, 29.009750366210938, 31.011688232421875, 107.81085205078125, 26.702239990234375, 23.812713623046875, -115.53292846679688, 122.6078109741211, 61.529632568359375, 33.597259521484375, 107.85818481445312, -14.818817138671875, -147.912841796875, 102.22959899902344, 50.395904541015625, 31.306930541992188, 140.806396484375, 99.25259399414062, 156.3720703125, 52.93829345703125, 46.15449523925781, 381.53167724609375, 17.18958282470703, 120.40486907958984, -86.00555419921875, 91.99893188476562, 110.5074234008789, 192.0499267578125, -38.34233093261719, 183.23272705078125, 95.70977020263672, -64.17742919921875, 121.459228515625, 102.20030212402344, -31.6925048828125, 0.0, 29.805543899536133, 26.568206787109375, -10.248123168945312, -83.17433166503906, 63.184600830078125, 212.29574584960938, 196.47462463378906, 1.1871623992919922, 152.37527465820312, 70.80050659179688, -30.887939453125, 126.27610778808594, 105.30801391601562, 157.69532775878906, -17.412384033203125, 53.08464050292969, 177.88314819335938, 177.6920166015625, 40.45494079589844, 10.813217163085938, 32.280521392822266, -78.96812438964844, 24.52362060546875, 38.90966796875, 37.09834289550781, 53.89581298828125, 64.70883178710938, 30.274261474609375, -4.202566146850586, 7.245105743408203, 78.70330810546875, 61.17913818359375, 212.76295471191406, 39.95088195800781, 60.27734375, 11.734298706054688, -7.720367431640625, 15.032291412353516, 3.6785125732421875, -63.2958984375, 62.03050231933594, 151.3717803955078, 98.31353759765625, 125.44049072265625, -11.107444763183594, -80.39945983886719, 13.984054565429688, 24.265411376953125, 150.1780548095703, 80.30719757080078, -37.5028076171875, 4.930442810058594, 197.13429260253906, -31.62114715576172, 318.9374084472656, 103.29594421386719, 153.13369750976562, 98.9730224609375, -11.33404541015625, 18.269134521484375, -107.95711517333984, 37.827606201171875, 4.78619384765625, -76.15325927734375, 203.15966796875, 16.095458984375, 87.44586181640625, 122.09295654296875, 104.50830078125, -105.95352172851562, 158.61033630371094, 121.65953063964844, 163.33331298828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000183.npy"}
{"epoch": 0.3832460732984293, "step": 184, "batch_size": 128, "mean": 77.85972595214844, "std": 95.65695190429688, "min": -186.10009765625, "p10": -27.31365432739257, "median": 67.16641235351562, "p90": 193.51325073242185, "max": 357.46942138671875, "pos_frac": 0.8125, "sample": [127.33943176269531, 273.73333740234375, 288.5451354980469, -6.022491455078125, 17.836624145507812, 48.44389343261719, 143.42474365234375, -8.403524398803711, 119.258056640625, 209.197509765625, 18.145858764648438, 32.7806396484375, 21.5087890625, 50.32659149169922, -55.27775573730469, 11.9599609375, 226.9532470703125, 160.59226989746094, 56.031410217285156, 133.3267364501953, 83.63397216796875, 323.1524658203125, -143.26181030273438, 99.81906127929688, -88.57051086425781, 180.10333251953125, 116.72012329101562, 15.7452392578125, -41.73081970214844, 150.52780151367188, 100.16236877441406, 154.8338623046875, 59.157073974609375, -25.02776336669922, 168.23724365234375, 357.46942138671875, 11.637786865234375, 144.8504180908203, 144.2757568359375, 130.62115478515625, 69.67816162109375, 162.74249267578125, -77.62393188476562, 53.61497497558594, 152.48223876953125, 203.334716796875, -87.90167236328125, 64.6546630859375, 19.175857543945312, 212.23529052734375, 55.09135437011719, -186.10009765625, 119.43521118164062, 27.45806884765625, -52.635223388671875, 23.162384033203125, 151.90386962890625, -2.5885162353515625, 60.9652099609375, 219.33071899414062, 53.83453369140625, 4.203441619873047, 159.4140625, -112.6959228515625, 107.01016235351562, 164.84344482421875, 155.50665283203125, 141.62045288085938, 90.95452880859375, 127.22964477539062, 9.23333740234375, 30.444442749023438, -5.211936950683594, 106.13642120361328, 203.7188720703125, 12.835861206054688, 37.365875244140625, 148.8072509765625, 149.85081481933594, 105.95150756835938, 201.4482421875, 44.461669921875, 57.636962890625, 119.88810729980469, -21.925338745117188, -150.3503875732422, 99.75112915039062, 54.66558837890625, 15.847198486328125, 97.75225830078125, 30.4739990234375, 200.1068115234375, 118.56570434570312, 37.365234375, 73.49221801757812, -3.9560012817382812, -23.0040283203125, 0.6448516845703125, 3.421234130859375, 132.9783935546875, 49.452125549316406, 139.22372436523438, 75.92959594726562, 190.68743896484375, 11.637222290039062, 7.674564361572266, 141.76803588867188, 77.71044921875, -22.67969512939453, 172.06698608398438, 139.98565673828125, -8.710205078125, 168.73721313476562, 48.022308349609375, 142.32528686523438, -36.210113525390625, 164.26268005371094, 26.0965576171875, -24.805618286132812, 103.47894287109375, 29.750885009765625, -42.56134033203125, 171.13330078125, 6.94769287109375, 272.5221862792969, 183.07086181640625, -32.64739990234375, 30.385242462158203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000184.npy"}
{"epoch": 0.38534031413612563, "step": 185, "batch_size": 128, "mean": 57.08734130859375, "std": 98.45223999023438, "min": -177.86187744140625, "p10": -61.90018615722656, "median": 52.56963348388672, "p90": 176.29580841064453, "max": 384.45709228515625, "pos_frac": 0.703125, "sample": [55.28053283691406, -111.96112060546875, 130.88372802734375, -115.1080322265625, 49.858734130859375, 38.125732421875, 100.03974151611328, 249.7421875, 201.326904296875, -13.835273742675781, 139.299072265625, 66.41348266601562, 74.300048828125, 130.3548126220703, 56.09053039550781, 384.45709228515625, 68.47198486328125, -59.80279541015625, 82.02462768554688, 144.07159423828125, 110.24473571777344, 61.51800537109375, 129.1064910888672, 110.4315185546875, 29.03131103515625, 56.9522705078125, 69.98269653320312, 29.91738510131836, 17.47572135925293, 143.79458618164062, -177.86187744140625, 128.3164825439453, 123.35517120361328, 3.03448486328125, 5.852630615234375, 101.13967895507812, 58.8836669921875, 138.46726989746094, 111.10125732421875, 257.1502380371094, 71.54241943359375, 111.76861572265625, -84.8759994506836, 156.37167358398438, -13.64276123046875, 130.13360595703125, 43.7542724609375, -21.853271484375, 69.64453125, 268.9619140625, -8.84197998046875, -27.082382202148438, -112.66061401367188, -24.20343017578125, 28.777084350585938, -4.751741409301758, 84.55642700195312, 0.0, 5.497043609619141, 6.944549560546875, -4.2935333251953125, -79.28852081298828, 35.454193115234375, -6.666942596435547, 47.09123229980469, 162.0511016845703, 31.146793365478516, 67.6614990234375, 170.25238037109375, -4.508901596069336, -60.83935546875, 145.2567138671875, 1.646209716796875, 143.84896850585938, -79.40275573730469, 46.073974609375, -64.37545776367188, 43.19903564453125, -8.738473892211914, 177.0909881591797, 177.12924194335938, -15.277053833007812, 63.370361328125, 0.0, 68.72622680664062, 72.27752685546875, 219.0933380126953, -54.7225341796875, 175.95501708984375, 69.96002197265625, 81.34476470947266, 22.921142578125, -146.829833984375, -16.889095306396484, 43.6556396484375, 273.5547790527344, -48.3262939453125, 39.412261962890625, 63.67144775390625, 114.3798828125, 12.306045532226562, 56.07435607910156, 141.07223510742188, -59.7655029296875, 272.02783203125, -4.6926116943359375, 27.29059600830078, 5.417533874511719, 115.35260009765625, 302.1259765625, 49.80206298828125, 78.65233612060547, 34.97175598144531, -120.93505859375, 207.81915283203125, -95.43695068359375, 105.6669921875, 213.70001220703125, 134.27670288085938, -145.01937866210938, 135.76773071289062, 42.90864562988281, -4.087738037109375, -70.76815795898438, -9.757820129394531, -59.86061096191406, -1.2034378051757812, 119.441162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000185.npy"}
{"epoch": 0.387434554973822, "step": 186, "batch_size": 128, "mean": 53.57258987426758, "std": 93.9493637084961, "min": -169.53091430664062, "p10": -69.592236328125, "median": 55.16560363769531, "p90": 175.12715148925778, "max": 334.4276123046875, "pos_frac": 0.7109375, "sample": [132.40658569335938, 99.18844604492188, 220.95223999023438, 77.21435546875, -47.602630615234375, 145.5386962890625, 100.47216796875, -17.035980224609375, 40.57958984375, 35.71983337402344, 47.19696044921875, 56.287994384765625, 21.671173095703125, 182.38482666015625, -113.41177368164062, 47.88031005859375, 131.1851348876953, -2.8307838439941406, 203.7322998046875, -9.67950439453125, 67.49777221679688, -69.90451049804688, -79.17242431640625, 37.726470947265625, 6.034832000732422, 56.09783935546875, 121.80621337890625, 85.03048706054688, 48.1607666015625, -46.254974365234375, 32.36603546142578, 76.0015869140625, 98.98483276367188, 82.008056640625, -42.73516845703125, -160.28466796875, 37.019012451171875, 161.29330444335938, 1.12432861328125, 76.6240234375, -24.374183654785156, 16.35809326171875, 88.59366607666016, 168.78416442871094, 132.82498168945312, 113.77886962890625, -83.3890380859375, 6.3408660888671875, 61.39002990722656, 61.2083740234375, 74.67103576660156, 172.48745727539062, 193.68081665039062, -34.2645263671875, -50.20390319824219, -37.603515625, 64.45169067382812, 102.26097106933594, 1.97320556640625, -3.727691650390625, -15.765495300292969, -47.26690673828125, 143.30364990234375, -69.45840454101562, 119.1592788696289, -30.15545654296875, 187.8238525390625, 58.984619140625, 104.2244873046875, 109.1063232421875, 66.06319427490234, 107.20748901367188, 138.3150634765625, 129.4437255859375, 78.929931640625, 53.873443603515625, 103.29972839355469, -114.18966674804688, 31.978586196899414, 268.2144775390625, 15.058456420898438, -38.159698486328125, 181.28643798828125, 278.8970947265625, 68.748291015625, -92.03875732421875, 52.69537353515625, -77.25944519042969, -94.61029052734375, -139.107666015625, -72.72311401367188, 4.702178955078125, 109.68251037597656, 54.233367919921875, 187.63699340820312, 220.08639526367188, -35.745338439941406, 136.09994506835938, -169.53091430664062, -24.97756004333496, 61.58274841308594, 164.09219360351562, -133.9358367919922, 6.6370086669921875, 207.92352294921875, 27.96923828125, 29.139862060546875, 49.11982727050781, 83.43858337402344, 106.98077392578125, 124.585205078125, -22.65460205078125, 11.72125244140625, 25.822826385498047, 213.55441284179688, 0.0, 334.4276123046875, 165.8721466064453, -49.42488098144531, 170.00048828125, -11.031333923339844, -28.65904998779297, 62.81669616699219, -9.753936767578125, 115.80349731445312, 26.803359985351562, 101.36306762695312, 68.51556396484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000186.npy"}
{"epoch": 0.38952879581151834, "step": 187, "batch_size": 128, "mean": 47.3968505859375, "std": 107.73493957519531, "min": -388.880615234375, "p10": -74.94834594726562, "median": 54.094276428222656, "p90": 177.0575210571289, "max": 315.1778869628906, "pos_frac": 0.6875, "sample": [73.7899169921875, 204.84254455566406, 105.5380859375, 118.81439208984375, -388.880615234375, 38.225311279296875, 208.65689086914062, 1.0814056396484375, -95.89324951171875, 207.56512451171875, 197.36849975585938, 8.549240112304688, -26.53143310546875, 34.648284912109375, 176.61546325683594, 110.0911865234375, 132.42926025390625, 130.28890991210938, 164.8829345703125, 80.56216430664062, 154.30502319335938, 19.972946166992188, 71.26324462890625, -73.9478759765625, -28.973114013671875, -70.77560424804688, 171.1008758544922, -94.18484497070312, -83.92477416992188, -100.55364990234375, -46.00993347167969, 178.0889892578125, 47.147705078125, -114.25872802734375, -25.865570068359375, -39.512901306152344, -13.7911376953125, 1.066009521484375, 110.48493957519531, 115.913818359375, 87.11764526367188, -77.28277587890625, 146.18911743164062, -57.93556213378906, -6.57159423828125, 158.20420837402344, 174.55975341796875, 12.54998779296875, -84.75042724609375, 47.6138916015625, 2.35546875, 62.6539306640625, 31.603477478027344, 190.064697265625, 10.48626708984375, -83.607177734375, 34.32171630859375, 128.5544891357422, 73.8067626953125, 122.1177978515625, 137.25253295898438, 6.055877685546875, 217.79196166992188, -150.62876892089844, 117.59904479980469, 174.77316284179688, -36.47804260253906, -297.5274658203125, 98.23306274414062, 77.001220703125, 114.44503784179688, -12.387630462646484, -16.7830867767334, -23.005035400390625, 80.676513671875, -171.98428344726562, 223.50608825683594, 121.04539489746094, 55.5902099609375, 53.60125732421875, 65.3011474609375, 54.58729553222656, -26.475059509277344, -17.091323852539062, 34.18196105957031, -36.084938049316406, 31.29534912109375, 27.343368530273438, -282.12841796875, 217.69906616210938, 127.34442138671875, 183.67147827148438, 185.95913696289062, 82.9885025024414, 39.26263427734375, 315.1778869628906, -9.463226318359375, 88.15811157226562, -64.92369079589844, 113.83118438720703, -22.38262939453125, 8.56561279296875, 150.674560546875, 122.47918701171875, -6.740682601928711, 149.85629272460938, 14.70159912109375, 1.66778564453125, 119.31381225585938, 68.35269165039062, 65.70196533203125, 98.845458984375, -26.192794799804688, 93.47949981689453, 5.360593795776367, 78.40826416015625, 180.67605590820312, -2.4861602783203125, 36.229103088378906, 104.04090881347656, -62.6343994140625, 119.32333374023438, 70.7087631225586, -40.12506866455078, 0.0, 157.95944213867188, -62.71221923828125, 112.07296752929688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000187.npy"}
{"epoch": 0.39162303664921466, "step": 188, "batch_size": 128, "mean": 53.92945861816406, "std": 113.08765411376953, "min": -209.49517822265625, "p10": -87.02411499023437, "median": 41.2294807434082, "p90": 188.31574707031248, "max": 431.797119140625, "pos_frac": 0.6875, "sample": [8.263900756835938, 122.10250854492188, 47.09423828125, 0.0, 68.79896545410156, -0.79290771484375, 19.921600341796875, 352.9853210449219, 9.950033187866211, 151.36178588867188, 102.15090942382812, 29.360031127929688, 142.84100341796875, -31.7406005859375, 17.98199462890625, 25.817604064941406, 200.96527099609375, 49.522613525390625, -10.383522033691406, 163.76992797851562, 258.92742919921875, 211.30099487304688, -147.32098388671875, 20.92013931274414, -80.57130432128906, -190.48065185546875, -7.50872802734375, -59.147918701171875, -94.2730712890625, 155.9461669921875, 146.9884033203125, -0.01519775390625, 115.78071594238281, -99.15615844726562, -6.09613037109375, 47.062164306640625, 141.77499389648438, 181.88656616210938, 105.91380310058594, -6.936981201171875, 15.212890625, 97.1915283203125, 2.0304946899414062, 158.60269165039062, 280.47808837890625, -95.0628662109375, 83.4500732421875, 112.5281982421875, 98.56216430664062, 24.4659423828125, 111.7928466796875, 125.47610473632812, -149.58363342285156, -46.8480224609375, 80.37850952148438, 263.96148681640625, 192.24444580078125, 129.0704345703125, 141.31527709960938, 431.797119140625, 272.7998046875, -6.8332672119140625, 27.625701904296875, 165.73780822753906, 35.29158020019531, 5.46136474609375, 69.29319763183594, 164.1423797607422, 146.06472778320312, 169.4865264892578, 15.305206298828125, 160.30484008789062, 186.63201904296875, -21.091014862060547, 13.308837890625, 38.652923583984375, 93.85736083984375, 203.01364135742188, -25.324371337890625, -73.00920104980469, -83.91741943359375, 65.5362548828125, 42.948463439941406, 180.05068969726562, 77.45692443847656, -0.9658546447753906, 34.1866455078125, 124.86529541015625, 116.46771240234375, 1.88592529296875, -190.63446044921875, 197.71224975585938, -177.2025909423828, -58.46514892578125, 166.08035278320312, -34.52532958984375, 12.012718200683594, -209.49517822265625, 78.05270385742188, 50.14292907714844, 53.4005126953125, 37.36619567871094, -54.670196533203125, -140.40975952148438, 196.2781982421875, -14.00836181640625, 128.34385681152344, 101.7542724609375, 255.18394470214844, -38.15625762939453, -106.8133544921875, 6.853322982788086, 69.4739990234375, 38.452606201171875, 158.67703247070312, -59.316253662109375, 186.1127166748047, -16.16799545288086, 14.611740112304688, -120.10263061523438, -48.44146728515625, -127.064697265625, 0.0, 40.050140380859375, 59.53416442871094, 65.11544799804688, 42.40882110595703, -80.43692016601562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000188.npy"}
{"epoch": 0.393717277486911, "step": 189, "batch_size": 128, "mean": 68.63457489013672, "std": 97.28904724121094, "min": -174.08766174316406, "p10": -30.028078460693358, "median": 51.162837982177734, "p90": 198.6401489257812, "max": 441.2745361328125, "pos_frac": 0.7890625, "sample": [89.89743041992188, 85.29241943359375, -129.55084228515625, 34.35020446777344, 20.254364013671875, -91.88015747070312, 3.2272491455078125, 118.49557495117188, 119.17203521728516, 96.94866943359375, 83.21321105957031, 111.0693359375, 172.45751953125, 45.990081787109375, 157.35116577148438, 216.0399169921875, 213.74095153808594, -21.720056533813477, 129.40478515625, 77.29400634765625, 166.14654541015625, 216.28826904296875, -31.448463439941406, 105.64492797851562, 51.566162109375, 12.52642822265625, 49.80792236328125, 44.6905517578125, 125.14722442626953, 26.71417999267578, -64.18499755859375, 77.11814880371094, 89.23597717285156, 150.92431640625, 27.557861328125, 160.4365234375, 44.69598388671875, 0.2689552307128906, 38.34886169433594, 146.88427734375, 4.126890182495117, 167.7093505859375, 153.7572784423828, 22.112945556640625, 303.75299072265625, 135.1002960205078, 5.21148681640625, -46.7120361328125, 222.02066040039062, -24.364212036132812, -68.71417236328125, -132.66632080078125, -81.59747314453125, 238.02099609375, 31.260009765625, 40.9935302734375, 317.31060791015625, 19.890060424804688, 9.840385437011719, 42.03376770019531, 137.39508056640625, 53.74461364746094, 145.27691650390625, -65.01536560058594, 7.874603271484375, 85.27339935302734, 150.42056274414062, 50.75951385498047, 13.016510009765625, 104.38191986083984, 59.120506286621094, -12.462928771972656, -14.511543273925781, 79.82916259765625, -2.372955322265625, 4.75970458984375, 40.577301025390625, 6.273357391357422, 153.6424560546875, 110.8011474609375, 0.057830810546875, 64.2598876953125, 41.95751953125, -35.12188720703125, 117.0546875, -9.490921020507812, 170.82029724121094, -174.08766174316406, -6.376579284667969, 3.089925765991211, -4.27972412109375, 257.6908874511719, 63.799774169921875, -126.78494262695312, 83.62313079833984, 0.0, 441.2745361328125, 185.08758544921875, 121.49186706542969, 262.36285400390625, 31.2401123046875, 146.40084838867188, 129.34439086914062, 143.08724975585938, 27.859962463378906, 0.0, 68.11590576171875, 225.6514892578125, 6.630859375, 193.05865478515625, 72.84910583496094, 10.288665771484375, 62.05010986328125, -3.8026084899902344, 22.5699462890625, -90.50634765625, 53.77851104736328, 80.00221252441406, 113.93573760986328, 27.173439025878906, 119.3685302734375, 228.10794067382812, -4.935874938964844, 211.66363525390625, 40.25732421875, 31.566970825195312, -29.419342041015625, -24.82927894592285], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000189.npy"}
{"epoch": 0.3958115183246073, "step": 190, "batch_size": 128, "mean": 59.70409393310547, "std": 90.87604522705078, "min": -172.0939178466797, "p10": -48.526879119873044, "median": 46.42024230957031, "p90": 182.2460784912109, "max": 300.4928894042969, "pos_frac": 0.765625, "sample": [-20.8101806640625, 44.148681640625, 152.89794921875, 37.1248779296875, 22.786163330078125, 12.981756210327148, -110.87136840820312, -15.10604476928711, -172.0939178466797, 4.67376708984375, 200.58949279785156, -50.69556427001953, -51.32914733886719, 113.6829833984375, 82.69808959960938, 49.17205047607422, 63.90496826171875, -30.716522216796875, 70.00306701660156, -47.597442626953125, 1.2496089935302734, -1.7994537353515625, -75.22855377197266, 218.07315063476562, 145.34011840820312, -60.714996337890625, 190.96780395507812, 3.06011962890625, 26.323387145996094, 178.3060302734375, 260.5052185058594, -73.18962097167969, -44.18260955810547, 42.645713806152344, 142.10391235351562, -77.27777099609375, 243.3280029296875, 31.174888610839844, 206.33741760253906, 114.19735717773438, 188.6614990234375, -135.16983032226562, -33.678192138671875, 275.07269287109375, 123.9298324584961, 44.73345947265625, 106.08946228027344, 153.7607421875, 155.31344604492188, 169.08401489257812, 80.30236053466797, -26.048133850097656, 88.88805389404297, -6.915641784667969, -33.07135009765625, 102.89591979980469, 5.508949279785156, 20.191314697265625, 5.507476806640625, -7.0890045166015625, -19.388427734375, 50.81391143798828, 61.72955322265625, 54.559661865234375, 69.71878051757812, 70.18142700195312, 138.84732055664062, 22.403533935546875, 126.354736328125, 112.07623291015625, 142.15948486328125, 160.17919921875, 49.226959228515625, 0.0, 59.17156982421875, 242.08189392089844, -23.157073974609375, 125.04154205322266, 248.82830810546875, 127.259521484375, 19.77880859375, 11.558517456054688, 83.464111328125, 115.07771301269531, 140.03338623046875, 36.052490234375, 0.5783538818359375, 29.128173828125, 48.107025146484375, 50.751953125, 1.3495941162109375, 300.4928894042969, 179.49661254882812, 171.88287353515625, -2.5380210876464844, -117.99118041992188, 119.69215393066406, -80.12542724609375, 198.93035888671875, -25.498062133789062, 114.941650390625, 33.409149169921875, 39.90374755859375, 160.31036376953125, 59.802642822265625, 131.66229248046875, 26.244140625, 0.103363037109375, 36.574310302734375, 54.6224365234375, 5.390689849853516, 202.60400390625, 19.634775161743164, 2.502685546875, 26.999980926513672, 42.37066650390625, 86.66244506835938, 110.8696060180664, -69.57421875, 43.38104248046875, -14.746612548828125, 4.877847671508789, 68.77490234375, 124.41294860839844, 5.6746368408203125, 140.8040771484375, -77.67007446289062, 58.63934326171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000190.npy"}
{"epoch": 0.39790575916230364, "step": 191, "batch_size": 128, "mean": 56.297969818115234, "std": 97.27239227294922, "min": -242.14321899414062, "p10": -41.284611511230466, "median": 40.137542724609375, "p90": 182.3019561767578, "max": 386.64453125, "pos_frac": 0.734375, "sample": [202.33892822265625, 182.03042602539062, 44.955718994140625, -4.176849365234375, 140.4582061767578, 162.41928100585938, 6.5858001708984375, 36.388999938964844, -42.722373962402344, -28.04083251953125, -35.343994140625, 120.34658813476562, 95.64419555664062, 18.810546875, 22.659271240234375, 12.920074462890625, 22.887432098388672, 114.99368286132812, 16.316526412963867, 80.37553405761719, 28.09661865234375, 182.57913208007812, 132.81500244140625, -0.16493988037109375, 2.5457725524902344, 28.08575439453125, 34.026947021484375, -130.4945068359375, 212.51495361328125, 6.682777404785156, 36.11590576171875, 81.75208282470703, -69.28535461425781, 127.9801025390625, 9.9136962890625, -134.96771240234375, 119.57789611816406, -41.36761474609375, -168.61810302734375, 39.24029541015625, 35.43810272216797, 154.96533203125, 114.48992919921875, 46.98566436767578, 13.552314758300781, 386.64453125, 62.72039794921875, -16.244232177734375, 46.38447570800781, 65.28298950195312, 49.7794189453125, 117.40859985351562, 198.581298828125, 108.85245513916016, 52.717742919921875, -75.6639404296875, 30.10419464111328, 96.49429321289062, -7.95611572265625, 85.49853515625, 11.575408935546875, 115.8746337890625, 14.927679061889648, 11.417869567871094, 154.1735076904297, 24.680419921875, 182.18316650390625, 41.0347900390625, 53.4825439453125, -3.1660537719726562, 24.897216796875, 11.188308715820312, 113.06541442871094, -242.14321899414062, -11.213489532470703, -26.865936279296875, 70.55673217773438, 13.81488037109375, 218.581298828125, -41.24903869628906, 28.59759521484375, 137.24758911132812, -1.007223129272461, 214.098388671875, -37.295257568359375, 163.9754638671875, 212.078125, -130.07723999023438, 169.86093139648438, 284.838623046875, 154.42312622070312, 167.12551879882812, 74.329833984375, -2.435771942138672, 15.705642700195312, -19.621688842773438, 52.091400146484375, 82.29388427734375, 81.59210205078125, 53.91609191894531, 267.571533203125, -13.779769897460938, 127.95575714111328, 60.96874237060547, 197.34130859375, 207.50344848632812, 44.592857360839844, 3.7191314697265625, -101.5409927368164, 41.7957763671875, -74.13043212890625, -172.58804321289062, 119.80267333984375, 124.59385681152344, 103.60064697265625, 171.832763671875, -12.597930908203125, -13.181930541992188, -38.18678283691406, 158.58123779296875, -26.281227111816406, 16.049652099609375, -60.12726593017578, 34.640380859375, 0.0, 173.35269165039062, -1.346822738647461, 194.53350830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000191.npy"}
{"epoch": 0.4, "step": 192, "batch_size": 128, "mean": 46.239566802978516, "std": 86.69127655029297, "min": -198.35025024414062, "p10": -54.39428634643554, "median": 43.922088623046875, "p90": 147.23963928222656, "max": 303.68890380859375, "pos_frac": 0.7109375, "sample": [106.20724487304688, 113.7257080078125, 73.49887084960938, -125.94586181640625, 95.03704833984375, 65.26513671875, -23.530609130859375, 45.063079833984375, 18.854087829589844, 263.07672119140625, 186.420166015625, 55.33740234375, 68.73978424072266, 39.131591796875, 113.74075317382812, 7.2467041015625, -17.84204864501953, 95.34909057617188, -198.35025024414062, 198.7681121826172, -83.91961669921875, 303.68890380859375, -64.35018920898438, -124.04171752929688, 185.38394165039062, 173.269775390625, 19.529983520507812, 77.78031921386719, -123.92518615722656, 159.0880126953125, -114.19015502929688, 70.15426635742188, -13.343643188476562, 139.34002685546875, 43.8907470703125, 13.811334609985352, 43.95343017578125, 132.45571899414062, 32.4666748046875, 150.29925537109375, 14.996063232421875, -116.6927490234375, -22.79608154296875, 42.6571044921875, 132.0172119140625, 4.8751220703125, 91.8804931640625, -108.6632080078125, 0.0, 0.0, 11.067802429199219, -23.74359130859375, -84.83598327636719, 66.36273193359375, -93.30975341796875, 1.6141281127929688, 9.897796630859375, 69.84634399414062, 31.01244354248047, 6.703338623046875, 129.14813232421875, 11.80344009399414, 30.354888916015625, 144.21197509765625, 165.98178100585938, 69.11251831054688, 123.85298156738281, 139.6326904296875, 6.282588958740234, 70.05110168457031, -1.6041259765625, 11.510528564453125, 27.12176513671875, 79.16390991210938, 44.76884460449219, 140.59857177734375, 61.1904296875, 122.52267456054688, 84.30377197265625, -27.707481384277344, -4.206684112548828, 30.802276611328125, -11.324005126953125, -14.316680908203125, -55.9923095703125, 114.35842895507812, 57.99231719970703, 70.32669067382812, 80.3284683227539, 76.17962646484375, 8.57098388671875, -21.54168701171875, -126.81263732910156, 301.549560546875, 130.7948455810547, 4.485877990722656, -23.78362274169922, -51.25417709350586, 182.284912109375, 173.09823608398438, 73.23501586914062, 93.44244384765625, -12.4298095703125, -3.695556640625, -26.904296875, -5.77001953125, 55.593414306640625, 25.00421142578125, 129.487060546875, 131.3536376953125, 62.72027587890625, -28.19805908203125, 68.91217803955078, 78.90826416015625, 14.510894775390625, -2.52191162109375, 124.59149169921875, 57.935733795166016, -19.214324951171875, 47.50836181640625, 15.12533950805664, 221.1531982421875, 38.217185974121094, -1.66632080078125, 55.318206787109375, 145.92837524414062, -53.70941925048828, 45.963287353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000192.npy"}
{"epoch": 0.40209424083769635, "step": 193, "batch_size": 128, "mean": 53.602569580078125, "std": 99.38701629638672, "min": -224.83493041992188, "p10": -64.48727416992188, "median": 44.89767837524414, "p90": 170.0946044921875, "max": 417.23016357421875, "pos_frac": 0.7734375, "sample": [5.337192535400391, 31.668441772460938, 1.54754638671875, -19.690231323242188, 9.61627197265625, 90.24942016601562, 37.39308166503906, 126.69775390625, 63.47709655761719, 51.19427490234375, 172.57687377929688, 8.718496322631836, 46.151702880859375, -22.7281494140625, 35.96244812011719, 148.44894409179688, -52.23968505859375, 30.069007873535156, 5.444602966308594, 10.179864883422852, 149.17227172851562, -63.598663330078125, 113.2442626953125, -13.053466796875, 70.37506866455078, 54.917938232421875, -83.41827392578125, -78.56108093261719, 158.44375610351562, -26.76214599609375, 31.523170471191406, 9.659042358398438, 46.92352294921875, 18.8656005859375, 98.06793975830078, 22.593658447265625, 83.6483154296875, -138.80087280273438, -68.11131286621094, 43.643653869628906, -124.33074951171875, 32.9556884765625, 72.0959243774414, 57.28485107421875, 33.06770324707031, 66.14157104492188, 97.14373779296875, 119.10977172851562, 169.31704711914062, 171.90890502929688, 91.53401184082031, -27.057861328125, 59.4217529296875, 4.2071075439453125, 97.99739074707031, 34.109954833984375, 105.75374603271484, 94.7752685546875, -3.46966552734375, 65.8563232421875, 51.711456298828125, 9.985130310058594, 110.71929168701172, -18.150238037109375, 166.48593139648438, 84.8482666015625, 132.38629150390625, 122.53729248046875, 52.14068603515625, -38.581298828125, -224.83493041992188, -170.2752685546875, 89.99740600585938, 265.643310546875, -60.72698211669922, 85.03619384765625, 247.89712524414062, 236.74476623535156, 28.6436767578125, 417.23016357421875, 216.78997802734375, 93.80889892578125, 272.28460693359375, -167.35108947753906, 167.64974975585938, 172.24713134765625, 41.44522476196289, 23.49443817138672, 34.00444030761719, 28.17047119140625, 19.04241943359375, 5.3981170654296875, 146.9637451171875, 207.0242462158203, 75.78022003173828, 34.8677978515625, 124.36969757080078, 67.55853271484375, 27.706985473632812, 216.88623046875, 65.0458984375, 146.71578979492188, 6.517818450927734, -73.82868957519531, -37.463958740234375, 1.89935302734375, 297.28912353515625, -56.804595947265625, -67.45924377441406, -25.99884033203125, -37.815452575683594, 3.74554443359375, -72.31768798828125, 148.33840942382812, 71.72828674316406, 12.224065780639648, 277.5729675292969, -171.14697265625, 62.0625, -66.56069946289062, -11.298126220703125, 110.04864501953125, 11.25189208984375, 114.99674987792969, 148.74087524414062, 50.2525634765625, 43.435028076171875, 85.70803833007812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000193.npy"}
{"epoch": 0.4041884816753927, "step": 194, "batch_size": 128, "mean": 44.330955505371094, "std": 92.03105163574219, "min": -248.3227996826172, "p10": -59.97816009521484, "median": 31.40224838256836, "p90": 159.3809814453125, "max": 484.853271484375, "pos_frac": 0.6796875, "sample": [106.77365112304688, -2.5477447509765625, 81.64485168457031, -69.33978271484375, 8.8548583984375, 188.939697265625, 10.808380126953125, -17.665252685546875, 70.96743774414062, -17.322452545166016, 97.14411163330078, -50.23258972167969, -16.631093978881836, 74.56301879882812, 63.418914794921875, 15.198171615600586, -69.87257385253906, 51.224945068359375, 151.22247314453125, -1.68707275390625, 67.14291381835938, 0.7331600189208984, 40.666770935058594, -106.0455322265625, 24.373992919921875, 38.851722717285156, -63.99821472167969, 104.695556640625, 5.6085205078125, -94.39898681640625, -48.88055419921875, 32.1868896484375, 13.186103820800781, 16.473388671875, 74.64356994628906, 5.116943359375, 234.50836181640625, 110.95720672607422, 128.12078857421875, 178.7997283935547, -33.27064514160156, 154.56109619140625, -84.750732421875, 76.59263610839844, -46.42608642578125, 134.50054931640625, 31.647056579589844, 175.03152465820312, 217.32969665527344, -35.69500732421875, 22.86908721923828, 79.52691650390625, 122.76824188232422, 33.425750732421875, 484.853271484375, 59.213958740234375, 52.58844757080078, 63.2630615234375, 48.274330139160156, 75.72463989257812, 82.02874755859375, -1.001129150390625, 4.099632263183594, 88.72700500488281, -11.4495849609375, 37.905242919921875, 118.46697998046875, -132.18511962890625, 31.157440185546875, 91.24493408203125, 127.49884033203125, -2.4486083984375, -39.468231201171875, 159.7552490234375, 3.063507080078125, 130.64154052734375, -7.014400482177734, -72.92439270019531, 97.83309936523438, 227.71449279785156, -107.39013671875, 102.23681640625, 129.3660888671875, 131.32818603515625, 17.869182586669922, 30.650665283203125, 42.61534118652344, 17.14672088623047, 226.4869384765625, 96.897216796875, 18.92767333984375, -75.14968872070312, 55.396240234375, -4.656547546386719, -8.466300964355469, -12.91033935546875, -5.296077728271484, -248.3227996826172, 218.881591796875, 65.57801055908203, -58.255279541015625, 159.2205810546875, 113.93934631347656, 25.513999938964844, -26.762542724609375, -5.626914978027344, -34.89581298828125, 99.177734375, -31.230499267578125, 10.8740234375, -0.4267578125, 166.48101806640625, 167.24510192871094, -22.611648559570312, 135.61529541015625, 181.73114013671875, -30.242767333984375, 17.68825912475586, -106.59766387939453, 65.68133544921875, -43.59661865234375, 7.309844970703125, 108.74957275390625, 17.501922607421875, 11.779720306396484, 79.83711242675781, -73.5328369140625, 44.72955322265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000194.npy"}
{"epoch": 0.406282722513089, "step": 195, "batch_size": 128, "mean": 43.69676208496094, "std": 90.89944458007812, "min": -151.50958251953125, "p10": -50.23114318847656, "median": 26.01453399658203, "p90": 160.9257110595702, "max": 437.21435546875, "pos_frac": 0.6640625, "sample": [-20.008209228515625, -0.3043060302734375, 3.12957763671875, 68.6566162109375, 195.33792114257812, 63.707977294921875, -46.69927978515625, 19.819656372070312, 47.98291015625, 48.6849365234375, 82.19732666015625, 31.067543029785156, 8.21889877319336, -11.279844284057617, -6.6372528076171875, 237.22348022460938, 10.79541015625, -13.339599609375, 89.96792602539062, 63.88922119140625, -40.12872314453125, 24.089996337890625, 74.48518371582031, -23.382247924804688, -33.298828125, -29.326217651367188, 16.141029357910156, 2.36700439453125, -52.600799560546875, 13.26690673828125, -55.01690673828125, 148.62533569335938, -17.31378173828125, -29.610671997070312, 2.922576904296875, -4.894920349121094, 253.78408813476562, 103.955078125, -49.215576171875, -109.89439392089844, 29.423492431640625, 13.994232177734375, -3.5698623657226562, 20.8961181640625, 259.3125915527344, 97.39384460449219, 193.91619873046875, -54.934112548828125, -20.702789306640625, 8.13128662109375, -29.294784545898438, 54.452789306640625, 107.48696899414062, 204.64251708984375, -13.602386474609375, 11.163330078125, 24.224700927734375, -0.9401016235351562, 30.368453979492188, 10.906646728515625, 127.36477661132812, -67.382568359375, -16.866241455078125, 125.14781188964844, -74.14411163330078, -123.0904541015625, -86.38261413574219, -6.999267578125, -94.05880737304688, 69.96600341796875, 134.87274169921875, 109.4580078125, 131.80545043945312, -39.805511474609375, 75.191650390625, 141.17535400390625, -112.504150390625, 122.449462890625, 54.735260009765625, 65.12553405761719, 85.73997497558594, 65.42120361328125, 99.90948486328125, 98.71420288085938, -67.68960571289062, -30.34021759033203, 35.24266815185547, 14.473846435546875, 9.192489624023438, 21.085235595703125, 42.43780517578125, 271.91839599609375, 76.159423828125, 44.47442626953125, 65.66597747802734, 74.29647827148438, 245.4228515625, 19.306182861328125, 106.94090270996094, 189.6265869140625, 29.356101989746094, 24.908218383789062, -27.995262145996094, 52.12705993652344, 19.843048095703125, 27.120849609375, -29.220672607421875, 196.13076782226562, 141.47882080078125, 138.91204833984375, 437.21435546875, 40.777984619140625, -101.9444580078125, -1.417694091796875, -32.71641540527344, 45.02415084838867, 122.827392578125, -151.50958251953125, -0.6604461669921875, 28.260154724121094, 195.38037109375, -4.149871826171875, 27.617660522460938, 73.44711303710938, 100.31859588623047, 35.95851135253906, 218.91909790039062, -27.513076782226562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000195.npy"}
{"epoch": 0.4083769633507853, "step": 196, "batch_size": 128, "mean": 49.14500427246094, "std": 93.23129272460938, "min": -315.48895263671875, "p10": -39.71286277770995, "median": 38.01304626464844, "p90": 166.3500762939453, "max": 338.0561828613281, "pos_frac": 0.7109375, "sample": [-8.807857513427734, 16.22760009765625, 96.42909240722656, 35.74275207519531, 39.41099548339844, 46.6527099609375, -10.42974853515625, 33.81482696533203, 77.68209838867188, -24.712982177734375, -4.3399200439453125, 229.32762145996094, 42.71356201171875, 71.78073120117188, 110.4466552734375, 12.661285400390625, 259.74359130859375, 45.41082763671875, 152.70989990234375, 39.9949951171875, 18.616744995117188, 11.650108337402344, 119.33273315429688, 156.19651794433594, 8.859611511230469, 39.0963134765625, -19.434368133544922, -32.68646240234375, -21.87120819091797, 77.4144287109375, 167.85848999023438, 75.8447265625, 103.4052734375, 46.673828125, 61.61163330078125, 3.595733642578125, -79.74058532714844, -84.9061279296875, 92.41990661621094, 16.942232131958008, -32.215667724609375, 121.74105834960938, 114.91110229492188, -62.00215148925781, 85.55239868164062, 5.989849090576172, 43.101409912109375, 165.70361328125, -66.635498046875, 128.51336669921875, -82.6058349609375, -5.770969390869141, -15.605499267578125, 21.57939910888672, 26.98230743408203, 192.9622802734375, 45.064544677734375, 5.800779342651367, 80.72412109375, 2.4180908203125, 58.1319580078125, 6.537872314453125, 26.649250030517578, -3.68310546875, -8.613525390625, 112.4993896484375, -11.348939895629883, 81.96759033203125, 3.49737548828125, -67.05831909179688, -1.297607421875, 245.44757080078125, -4.646381378173828, 35.95427703857422, 99.46258544921875, 67.65113830566406, -10.148193359375, -12.6439208984375, 55.79454803466797, 41.564117431640625, 64.92253875732422, -315.48895263671875, 2.946258544921875, 11.63470458984375, -45.056640625, 172.47320556640625, 74.18255615234375, 234.30174255371094, 69.1478271484375, 36.929779052734375, 249.24380493164062, -179.8917236328125, 176.81195068359375, 219.78787231445312, 35.685791015625, -24.92333221435547, -9.343147277832031, -5.075469970703125, 146.401611328125, 150.5323486328125, 75.54997253417969, 98.44137573242188, 10.35723876953125, 267.6096496582031, 289.291748046875, -10.322322845458984, -37.422672271728516, 103.31963348388672, 12.15350341796875, 105.12185668945312, 88.26284790039062, 338.0561828613281, 114.98521423339844, 157.18728637695312, -69.052978515625, -49.88349151611328, -22.525291442871094, 55.531005859375, 12.776283264160156, 153.15231323242188, -18.820526123046875, 0.5220546722412109, -158.607177734375, 71.69146728515625, 50.682464599609375, -47.22503662109375, 101.7222900390625, 17.51802635192871], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000196.npy"}
{"epoch": 0.41047120418848165, "step": 197, "batch_size": 128, "mean": 61.486061096191406, "std": 92.0542221069336, "min": -164.4261474609375, "p10": -53.529118347167966, "median": 55.255001068115234, "p90": 178.8598876953125, "max": 392.523193359375, "pos_frac": 0.75, "sample": [-80.57286071777344, 62.07305908203125, -164.4261474609375, 66.38336181640625, 206.87994384765625, -25.018722534179688, 142.95074462890625, 45.859588623046875, -76.31116485595703, 157.86590576171875, 120.09173583984375, 163.811279296875, 2.8697967529296875, 74.37353515625, 114.75103759765625, -3.363809585571289, 181.57965087890625, 172.30697631835938, 21.191574096679688, 149.533447265625, 11.191438674926758, 67.51780700683594, 85.98526763916016, 83.20834350585938, 6.603721618652344, -81.02444458007812, 237.28997802734375, 208.824951171875, 138.27093505859375, 118.67710876464844, 147.14999389648438, -58.557220458984375, 1.5552978515625, 129.3817596435547, 34.35888671875, 173.20614624023438, 27.716663360595703, -23.51770782470703, 31.55328369140625, 231.14495849609375, 28.0223388671875, 392.523193359375, -45.82647705078125, 177.69427490234375, -129.06903076171875, -73.66079711914062, 120.09231567382812, 17.5911865234375, -26.69293212890625, 40.643768310546875, 79.08441162109375, -38.59087371826172, 96.65725708007812, -42.605621337890625, 64.14398193359375, 22.286460876464844, 188.04010009765625, 63.414154052734375, 59.95904541015625, 195.7550048828125, 46.978267669677734, 253.51519775390625, 10.425994873046875, 14.1221923828125, -55.4471435546875, 14.666839599609375, 123.0546875, 92.328369140625, 7.213409423828125, 22.0638427734375, 130.42117309570312, 73.1121826171875, -11.786155700683594, 207.6309051513672, -7.0787353515625, 83.44985961914062, -21.7518310546875, 13.677825927734375, -64.0376968383789, 3.2010498046875, -162.43505859375, 1.31536865234375, 131.96865844726562, 127.44775390625, 10.25653076171875, 84.69204711914062, 32.1531982421875, 59.12444305419922, 67.3670654296875, 35.34071350097656, 118.66424560546875, -0.06402587890625, 96.66070556640625, 51.38555908203125, 143.3921661376953, -10.745361328125, -39.171966552734375, -57.497802734375, -4.389595031738281, -89.77243041992188, 18.80859375, 81.38552856445312, -21.170028686523438, 135.31158447265625, 222.56597900390625, 20.609237670898438, 98.35818481445312, 105.73258972167969, 166.27041625976562, 36.0394287109375, 114.87637329101562, 102.27229309082031, 30.63970947265625, -52.70710754394531, 64.06324768066406, 33.6944580078125, 153.42327880859375, -3.9632568359375, 199.0896759033203, -0.6277923583984375, 0.0, 236.89743041992188, 152.84954833984375, 73.59478759765625, 156.77566528320312, -63.35662841796875, 136.15213012695312, 48.35044860839844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000197.npy"}
{"epoch": 0.41256544502617803, "step": 198, "batch_size": 128, "mean": 66.17696380615234, "std": 98.33187103271484, "min": -177.71270751953125, "p10": -37.93630599975586, "median": 45.273162841796875, "p90": 191.2107696533203, "max": 317.26080322265625, "pos_frac": 0.71875, "sample": [173.99249267578125, 15.7159423828125, -13.2027587890625, -36.48350524902344, 145.54742431640625, 32.34197998046875, 232.77581787109375, 241.681884765625, 188.00662231445312, 65.69442749023438, -11.050064086914062, -34.11714553833008, 32.710052490234375, 175.767578125, 35.29566955566406, -4.6051483154296875, 176.24761962890625, 202.524658203125, 52.1573486328125, 147.66964721679688, 183.09115600585938, -177.71270751953125, -84.26483154296875, 140.82281494140625, 184.6068115234375, 182.2154541015625, 31.07696533203125, 4.270111083984375, 45.35260009765625, 32.01519775390625, 118.63427734375, 41.988800048828125, 196.02499389648438, 184.5145263671875, 15.86297607421875, 16.74814224243164, -54.05084228515625, 175.520263671875, -145.85195922851562, 279.78985595703125, -14.493072509765625, 2.97998046875, 294.64385986328125, 194.35113525390625, -30.83609390258789, -27.47471809387207, 224.120361328125, -33.023521423339844, 7.452150344848633, 15.194053649902344, 65.34487915039062, -70.51382446289062, -1.4275665283203125, 60.84991455078125, 6.3586273193359375, 102.63223266601562, 317.26080322265625, -31.505386352539062, 69.21087646484375, 82.6229248046875, 85.61417388916016, 102.49383544921875, -6.408166885375977, 116.96807861328125, 130.60296630859375, 105.35855102539062, 1.9798450469970703, 189.50054931640625, 34.46195983886719, 15.147216796875, 111.41400146484375, -114.24114990234375, -29.66876220703125, 115.96145629882812, -151.68435668945312, -17.195465087890625, 59.31886291503906, 132.62738037109375, -39.12482452392578, 45.1937255859375, 135.90927124023438, 234.16510009765625, 78.11785888671875, 96.54940795898438, 75.06177520751953, 27.12979507446289, 158.935302734375, 250.063720703125, -89.62344360351562, 86.019287109375, 0.0, 124.6409912109375, 36.978179931640625, -44.197509765625, -44.92848205566406, 20.811508178710938, -10.641849517822266, 58.175621032714844, 149.21209716796875, 38.1309814453125, 247.91375732421875, 25.87628173828125, 150.34364318847656, 156.80062866210938, 98.87901306152344, 110.8824462890625, 77.95388793945312, -3.013599395751953, -6.1253662109375, -105.97421264648438, 131.41268920898438, 11.660797119140625, -17.143342971801758, 3.4396514892578125, -27.637771606445312, -42.048919677734375, 189.86489868164062, 41.3233642578125, 227.42019653320312, 33.23468017578125, 119.22061157226562, -37.42694091796875, 75.44692993164062, -10.495214462280273, 34.917144775390625, 184.81591796875, 110.11016845703125, -2.902191162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000198.npy"}
{"epoch": 0.41465968586387436, "step": 199, "batch_size": 128, "mean": 53.729469299316406, "std": 101.13272857666016, "min": -181.74319458007812, "p10": -71.46212615966797, "median": 44.82685852050781, "p90": 200.54844665527344, "max": 289.6500244140625, "pos_frac": 0.703125, "sample": [43.2396240234375, 100.7353515625, 200.16653442382812, 71.53067016601562, 160.37619018554688, 153.97373962402344, 75.93983459472656, -21.3257999420166, 129.09405517578125, 136.20458984375, 268.8026428222656, 219.47293090820312, -17.3414306640625, 95.165283203125, 210.22998046875, -117.43731689453125, 77.83328247070312, 43.79847717285156, 45.85818099975586, -1.1304874420166016, 89.14160919189453, -71.42448425292969, 2.3077869415283203, 13.087940216064453, -71.54995727539062, 19.423789978027344, -8.136314392089844, 28.885971069335938, 107.50140380859375, -115.572509765625, 129.2250213623047, 130.95263671875, 157.38702392578125, 3.56964111328125, 21.87945556640625, 236.2052459716797, -4.46429443359375, -73.08294677734375, 68.83715057373047, 0.157745361328125, 41.67424011230469, 95.3785400390625, 30.32421875, 10.641647338867188, 3.6046295166015625, 192.9700927734375, 65.24560546875, 29.789325714111328, 0.03570556640625, -87.40576171875, 53.595062255859375, 279.3058166503906, -12.542022705078125, 55.443450927734375, -101.52345275878906, 70.31245422363281, 22.605377197265625, 44.3380126953125, 160.4420166015625, 80.130859375, -181.74319458007812, 113.282958984375, 117.20343017578125, 155.8043670654297, -28.604904174804688, 23.78528594970703, 53.74720001220703, -76.18923950195312, -8.092041015625, 74.9852294921875, 241.65054321289062, -98.22831726074219, 15.065999984741211, 259.8096923828125, -32.05279541015625, -38.80845642089844, 66.96830749511719, 97.29629516601562, 2.046344757080078, 1.9327850341796875, 69.64614868164062, 29.691848754882812, 233.02691650390625, -26.6221923828125, 177.59027099609375, 184.38198852539062, -31.670242309570312, -15.8486328125, 260.31451416015625, 51.633636474609375, -57.188812255859375, -67.6036376953125, 94.6950454711914, 201.4395751953125, 45.315704345703125, 37.04156494140625, -6.776470184326172, -45.69482421875, 25.035369873046875, 141.36026000976562, 124.78964233398438, -136.00559997558594, 80.50146484375, -1.667755126953125, 141.91287231445312, -175.46246337890625, -31.45777130126953, 0.2294178009033203, -8.201210021972656, 100.60330200195312, 76.02565002441406, 68.71511840820312, 198.1260986328125, -59.75300598144531, 60.17620849609375, 73.741943359375, 226.01893615722656, 87.02705383300781, -66.16706848144531, 133.62899780273438, -1.007354736328125, -84.20309448242188, 1.5045623779296875, -59.91222381591797, 289.6500244140625, 265.9808349609375, 71.46086120605469, -132.39480590820312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000199.npy"}
{"epoch": 0.4167539267015707, "step": 200, "batch_size": 128, "mean": 50.33221435546875, "std": 113.19395446777344, "min": -288.2904357910156, "p10": -95.89669952392578, "median": 43.36781311035156, "p90": 196.0126342773437, "max": 331.333740234375, "pos_frac": 0.6953125, "sample": [59.762969970703125, -8.564132690429688, 240.15621948242188, 67.69378662109375, -78.97357177734375, 61.55701446533203, -26.60226058959961, -40.550628662109375, -31.345733642578125, 110.7326431274414, -97.37525939941406, 214.64556884765625, -0.736602783203125, 10.484542846679688, -125.35549926757812, 169.74710083007812, 130.85763549804688, -41.213768005371094, 81.64620971679688, 191.53900146484375, -141.8983154296875, -54.39666748046875, 163.47311401367188, 110.52665710449219, 188.40435791015625, 118.63592529296875, -81.21919250488281, -23.271942138671875, 43.173858642578125, 3.53814697265625, -160.47213745117188, -288.2904357910156, 62.981658935546875, 170.68572998046875, 68.16201782226562, 18.35821533203125, -60.335235595703125, 259.94293212890625, 170.03823852539062, 43.561767578125, 0.0, 74.3756103515625, 206.45111083984375, 12.80206298828125, -54.1341552734375, -109.03897094726562, 111.2823486328125, 25.525390625, 272.5775146484375, -83.76349639892578, 12.027862548828125, 67.43096923828125, 82.8619384765625, 6.684600830078125, 33.96662902832031, -21.428298950195312, 152.94808959960938, -206.87301635742188, 102.59625244140625, -26.363754272460938, 70.66288757324219, 78.8541259765625, -3.6978187561035156, 124.89852905273438, 212.326904296875, 126.37792205810547, 3.9024276733398438, 210.37298583984375, -176.89404296875, 62.4083251953125, 324.5554504394531, 111.06573486328125, 25.89117431640625, -19.572021484375, 139.4527587890625, 76.827392578125, -67.356201171875, -25.400550842285156, 2.8992919921875, 187.35821533203125, -152.61074829101562, 233.8570556640625, 331.333740234375, 38.091156005859375, 12.009117126464844, 175.3690185546875, 8.368072509765625, 188.771728515625, -140.39089965820312, 167.34161376953125, -95.26303100585938, 224.87969970703125, -44.88438415527344, 167.34136962890625, 86.3839111328125, 101.369384765625, 30.149559020996094, 211.04620361328125, -107.50067138671875, -21.746170043945312, 41.13067626953125, 104.48538208007812, 97.1904296875, 4.9571533203125, -115.93692016601562, 122.592529296875, 22.07403564453125, -47.77622985839844, -170.92233276367188, 22.7568359375, 20.74280548095703, 165.17788696289062, 82.28057861328125, 274.5560607910156, 59.19146728515625, 121.78302001953125, -11.255355834960938, 9.237518310546875, 87.35174560546875, 31.96123504638672, 79.94540405273438, 70.06614685058594, 165.14010620117188, -16.715438842773438, 56.9447021484375, 14.276748657226562, 100.16552734375, 12.6402587890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000200.npy"}
{"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 59.25444030761719, "std": 108.9170913696289, "min": -259.123291015625, "p10": -68.68003997802734, "median": 58.431785583496094, "p90": 210.7865264892578, "max": 340.91668701171875, "pos_frac": 0.6953125, "sample": [-123.25180053710938, -86.88143920898438, 139.001953125, 121.05078125, 70.76513671875, 67.8121337890625, -66.25225830078125, -11.992889404296875, 66.02713012695312, 172.818359375, 63.79974365234375, 88.9114990234375, -28.941505432128906, 122.85992431640625, -77.52056884765625, -65.54852294921875, 213.33352661132812, 255.21978759765625, 99.75222778320312, 30.699432373046875, 17.88946533203125, 172.44973754882812, 27.273521423339844, 0.0, 58.07579040527344, -52.269622802734375, 85.62759399414062, 274.4532165527344, 52.3553466796875, 78.55209350585938, 296.2781982421875, 127.41751098632812, -35.47401428222656, -88.37437438964844, 0.0, -175.6810302734375, 20.267410278320312, -13.140792846679688, -31.501564025878906, -74.34486389160156, -39.92283630371094, 88.92762756347656, 248.6925048828125, 66.85759735107422, -30.68865966796875, 154.6028594970703, 81.07745361328125, 121.78081512451172, -4.72186279296875, 16.634321212768555, -19.736831665039062, 101.16728210449219, 238.4937744140625, 297.573486328125, -14.128204345703125, 86.7411117553711, 174.50665283203125, 212.6842041015625, 128.64254760742188, 113.31765747070312, 12.357688903808594, 209.97323608398438, 104.11276245117188, 30.400604248046875, 183.11679077148438, 162.56185913085938, 27.195831298828125, 0.0, 40.645164489746094, -133.52423095703125, 92.25, -34.280426025390625, -21.879302978515625, 208.6708984375, 117.34048461914062, 71.64903259277344, 17.09772491455078, 14.952140808105469, -12.4429931640625, 81.78878784179688, 96.65066528320312, -27.733413696289062, 6.80088996887207, 5.59613037109375, -43.199249267578125, 161.2758331298828, -80.92393493652344, 249.02487182617188, -95.66607666015625, 246.19317626953125, 340.91668701171875, 9.78802490234375, 149.14981079101562, 15.159637451171875, 58.78778076171875, 1.697479248046875, 49.79621887207031, 152.58016967773438, 95.978515625, 257.32843017578125, 257.146728515625, 120.86466217041016, -137.98098754882812, 112.87435913085938, 45.465614318847656, -26.527565002441406, -142.99676513671875, 98.279296875, 141.13027954101562, 60.760711669921875, 129.1390838623047, 2.105255126953125, -50.1539306640625, -45.514892578125, 128.05809020996094, 139.7802734375, 9.743927001953125, 53.03984069824219, 25.35501480102539, 149.03350830078125, 51.46465301513672, -164.40609741210938, 59.64634704589844, 148.35723876953125, -46.72356414794922, 133.7738037109375, -43.22889709472656, -259.123291015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000201.npy"}
{"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 55.29514694213867, "std": 108.12387084960938, "min": -211.1282501220703, "p10": -70.15922622680664, "median": 49.0234375, "p90": 182.62352600097654, "max": 349.4645080566406, "pos_frac": 0.6640625, "sample": [317.82147216796875, -211.1282501220703, -59.02287292480469, 62.20536804199219, 200.0093994140625, 93.3379898071289, -13.24530029296875, 246.09637451171875, -62.91619873046875, 16.37628173828125, 53.2081298828125, 103.95062255859375, 87.0131607055664, 242.0684814453125, 129.14273071289062, 64.31353759765625, -98.86212158203125, 173.60452270507812, 93.23274230957031, 111.24714660644531, 82.97931671142578, 174.66390991210938, -68.12310791015625, 76.75796508789062, -6.34295654296875, -73.22189331054688, 176.4766845703125, 6.605064392089844, -4.7205963134765625, 44.23670959472656, -140.4835662841797, 68.29827880859375, -17.445068359375, 220.07608032226562, 32.35289001464844, 189.48101806640625, 145.6588134765625, 170.02444458007812, 52.02350616455078, -50.54502868652344, 103.63662719726562, 289.4090576171875, 138.49057006835938, 162.96795654296875, -17.8963623046875, 93.2099609375, -29.340545654296875, 40.05516052246094, 188.21636962890625, -84.62800598144531, 8.111328125, 142.30856323242188, 158.7237091064453, -32.617523193359375, -126.65847778320312, 17.985031127929688, 173.02694702148438, -36.125701904296875, -25.770294189453125, -5.718536376953125, 105.12074279785156, 70.05178833007812, 174.49063110351562, -59.929595947265625, -53.55181884765625, 18.86053466796875, 137.42608642578125, 10.12835693359375, -89.64529418945312, -1.4327392578125, 35.699607849121094, 136.22689819335938, 110.6461181640625, 3.3980560302734375, 180.22659301757812, 48.911376953125, 152.08102416992188, 71.27667236328125, 349.4645080566406, -70.70587158203125, 0.0, 162.98583984375, 47.82186508178711, 113.41043090820312, -69.65261840820312, -91.78240203857422, 19.960384368896484, -4.8627166748046875, -77.26985168457031, 49.135498046875, 35.5821533203125, 332.3929748535156, -10.307266235351562, 60.88451385498047, 46.779571533203125, 74.6068115234375, 111.96672058105469, -111.0689697265625, -53.540283203125, -23.20751953125, -191.64703369140625, 107.1279296875, 69.10284423828125, 279.27545166015625, 75.6820068359375, -35.7978515625, 137.8288116455078, 100.11422729492188, -10.415550231933594, -31.875926971435547, 94.467529296875, -34.93890380859375, 33.51683044433594, 53.289093017578125, 135.169677734375, -59.781951904296875, 95.61697387695312, 129.57476806640625, -190.79486083984375, 21.696762084960938, -12.086149215698242, 266.8348388671875, 38.117340087890625, 17.797035217285156, 17.168907165527344, 105.15362548828125, -69.9249496459961, 210.33706665039062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000202.npy"}
{"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 42.51974105834961, "std": 95.71865844726562, "min": -192.9566650390625, "p10": -88.6443878173828, "median": 46.66511535644531, "p90": 159.23348999023438, "max": 319.853271484375, "pos_frac": 0.703125, "sample": [26.596220016479492, 2.1697006225585938, 109.25054931640625, 73.84528350830078, 93.7498779296875, 129.3834686279297, 113.35040283203125, 71.859130859375, -6.24932861328125, 35.932525634765625, 12.87213134765625, 53.49249267578125, -61.00232696533203, 103.357177734375, -105.21040344238281, 163.2291259765625, -111.8812255859375, -79.98701477050781, -44.82292938232422, 92.58114624023438, 27.377899169921875, 0.0, -110.32608032226562, 73.33064270019531, -138.744384765625, 39.752662658691406, 46.778289794921875, 80.73065185546875, 74.13037109375, 12.59844970703125, 148.71536254882812, 134.16775512695312, 90.51321411132812, 53.99481201171875, 39.07695007324219, -20.685649871826172, 32.75494384765625, 319.853271484375, 150.83453369140625, -2.7039794921875, 82.50028991699219, 185.6856689453125, 157.953125, 95.15773010253906, 5.722484588623047, -55.68048095703125, 65.39720916748047, 162.22100830078125, -65.11137390136719, 0.4669647216796875, 8.303131103515625, 88.33045959472656, 125.31716918945312, 24.444503784179688, 237.9805145263672, -62.0989990234375, 53.6253662109375, 13.477998733520508, 217.4341278076172, 187.29098510742188, -41.51521301269531, 113.58649444580078, 76.076416015625, 53.18396759033203, -46.80829620361328, 36.68236541748047, -108.30276489257812, 154.3250732421875, 69.19561004638672, 3.540802001953125, 49.42376708984375, -87.98171997070312, -131.19216918945312, -90.19061279296875, -18.885589599609375, 2.9397106170654297, -191.48800659179688, 8.7584228515625, -100.90478515625, -23.638504028320312, 84.83929443359375, 25.789276123046875, -153.4107666015625, 209.99864196777344, 137.8811492919922, -58.45658874511719, 114.5810546875, 57.29534912109375, 249.30715942382812, 62.2818603515625, 44.430992126464844, 143.9134521484375, -46.548370361328125, 204.07064819335938, 55.06035232543945, 86.1053466796875, 198.97723388671875, 33.7080078125, 152.42544555664062, 156.8565673828125, 65.96630859375, -92.9095458984375, -3.13818359375, 30.030517578125, 85.50664520263672, -77.9443359375, -36.884185791015625, -71.43052673339844, -40.92913818359375, 213.881103515625, 0.0, 25.2791748046875, -44.40461730957031, 60.21031188964844, 107.12061309814453, 19.716705322265625, 46.55194091796875, 86.2999267578125, 28.324920654296875, 58.19317626953125, 132.30172729492188, -118.76617431640625, 112.29647827148438, 195.33761596679688, 69.93507385253906, -192.9566650390625, -29.00341033935547, 74.94683074951172], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000203.npy"}
{"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 60.053924560546875, "std": 102.3615493774414, "min": -178.56512451171875, "p10": -79.68594360351562, "median": 59.495643615722656, "p90": 185.43572387695312, "max": 324.6540222167969, "pos_frac": 0.734375, "sample": [164.27227783203125, 97.32818603515625, 17.654541015625, 143.0560302734375, 63.89888000488281, 165.4934844970703, 4.828853607177734, 153.49191284179688, 92.38462829589844, 183.760986328125, 64.89453125, 52.277557373046875, 35.06048583984375, -46.0621337890625, -178.56512451171875, -79.3626708984375, 43.2957763671875, -68.54071044921875, 43.54100799560547, 47.436798095703125, 113.38070678710938, 25.572677612304688, 44.3033447265625, 71.8514175415039, 63.649658203125, 191.4869384765625, -30.030807495117188, -93.63388061523438, 67.18803405761719, 60.971282958984375, 56.747283935546875, 230.749755859375, 7.960906982421875, 83.36589050292969, 124.96875, -97.33050537109375, 17.879657745361328, 324.6540222167969, 44.00025939941406, 136.70901489257812, 221.42556762695312, -119.3184814453125, 220.381103515625, 31.98351287841797, 69.53948974609375, 168.20346069335938, -147.78746032714844, 256.63751220703125, 153.33450317382812, 204.55642700195312, 184.01278686523438, -80.44024658203125, -127.443603515625, 153.3843994140625, 42.12432861328125, 7.704673767089844, 188.74154663085938, -11.763317108154297, -27.517738342285156, -126.07064819335938, 215.77859497070312, 78.79489135742188, 174.31781005859375, -48.640472412109375, -43.3448486328125, 186.43560791015625, 170.413818359375, 120.5262451171875, -77.457275390625, 44.103668212890625, -23.59292221069336, 120.12625122070312, -6.8541717529296875, -55.35059356689453, -7.175128936767578, -17.568687438964844, 183.3359375, 235.857177734375, -1.9580211639404297, 27.218994140625, 29.149154663085938, -117.92666625976562, -127.42169952392578, 139.08416748046875, 69.63027954101562, -39.33935546875, 167.27349853515625, 36.33531188964844, 174.04998779296875, 172.7377166748047, 65.65887451171875, 4.622314453125, -157.57427978515625, 155.1439208984375, -69.22634887695312, -81.20913696289062, 14.320022583007812, -61.011474609375, 32.78080749511719, 146.362548828125, 54.300819396972656, 5.15899658203125, 128.17886352539062, 60.179962158203125, 123.90518951416016, 84.125, 94.66786193847656, 68.43143463134766, 187.27059936523438, 115.61908721923828, 51.806427001953125, 71.62042236328125, 2.19287109375, 185.0072021484375, 70.53846740722656, 99.929931640625, 96.31991577148438, -15.577919006347656, 29.523963928222656, 62.942840576171875, 58.81132507324219, 42.65141296386719, 318.8421630859375, 144.5501708984375, -31.876251220703125, 153.57354736328125, -103.66511535644531, -6.8109130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000204.npy"}
{"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 71.71823120117188, "std": 115.15574645996094, "min": -230.92636108398438, "p10": -67.75863342285156, "median": 55.40771484375, "p90": 204.23424987792967, "max": 625.955078125, "pos_frac": 0.75, "sample": [250.86019897460938, 0.0, 11.667716979980469, 11.714971542358398, -96.19226837158203, 55.283294677734375, -5.601715087890625, -71.06878662109375, 43.749267578125, 19.037391662597656, 194.4715576171875, 132.97412109375, 40.27627182006836, 59.34120178222656, 17.195770263671875, 5.8335418701171875, 130.82080078125, 169.83212280273438, 26.25318145751953, 150.65383911132812, 9.94464111328125, 111.45870971679688, -78.63238525390625, 186.15533447265625, 137.8465576171875, 187.08697509765625, 23.363082885742188, 46.498199462890625, 127.96014404296875, 123.7037353515625, 144.39923095703125, 68.97702026367188, 9.385955810546875, 16.558135986328125, -85.931396484375, 17.684505462646484, 128.3776092529297, 141.3165283203125, 33.21126937866211, 12.067947387695312, 29.815780639648438, 52.471832275390625, -230.92636108398438, 70.88551330566406, 119.86859130859375, 185.0557861328125, 0.0, 68.694091796875, -56.793190002441406, -66.33999633789062, 7.686286926269531, 28.901596069335938, 78.09732055664062, 156.59039306640625, -0.5546875, -75.82684326171875, 237.53614807128906, 185.25038146972656, -48.65936279296875, 144.50680541992188, -159.34976196289062, -107.67630004882812, 10.048614501953125, 86.13998413085938, -4.3662109375, -40.91900634765625, 290.916259765625, 35.43724060058594, -54.868896484375, 127.3028564453125, 119.86672973632812, 203.5306396484375, -121.26742553710938, 205.87600708007812, 28.438125610351562, 25.585296630859375, 55.532135009765625, 171.98463439941406, 82.73980712890625, -18.34576416015625, -90.32405090332031, 369.9793701171875, 142.54690551757812, 156.12417602539062, 154.71432495117188, 11.733306884765625, -40.37115478515625, -82.08236694335938, 152.3486785888672, 122.56907653808594, 151.26416015625, 317.2299499511719, 175.5615234375, -58.197265625, 145.589599609375, 79.48133850097656, 218.42706298828125, 76.02503967285156, 99.66254425048828, -64.25222778320312, 625.955078125, 234.6395263671875, -11.134185791015625, -30.08441162109375, 234.202392578125, 92.77972412109375, -76.40611267089844, 125.75180053710938, 251.94937133789062, 162.81552124023438, 139.04464721679688, 23.778778076171875, -80.61988830566406, 127.07140350341797, 141.039306640625, 22.910614013671875, -13.498153686523438, 137.0924072265625, 24.255165100097656, 24.919662475585938, 25.2679443359375, 214.17282104492188, -60.35841369628906, 207.02447509765625, 0.8382720947265625, 162.81591796875, 0.0, 104.28038787841797], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000205.npy"}
{"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 55.590843200683594, "std": 112.79855346679688, "min": -225.8656005859375, "p10": -97.6871971130371, "median": 38.70165252685547, "p90": 204.67308349609374, "max": 307.4605712890625, "pos_frac": 0.6875, "sample": [167.7423095703125, 21.86119842529297, -114.13873291015625, 13.04180908203125, 218.42776489257812, -110.55936431884766, -68.25189208984375, 93.00128173828125, -23.041748046875, 38.00547790527344, 147.00157165527344, 145.57293701171875, 246.126953125, -50.620887756347656, -125.95166015625, 48.228179931640625, 307.4605712890625, 89.58753967285156, 100.88365173339844, 287.8685607910156, -9.783340454101562, 208.14898681640625, -12.762012481689453, -118.38204956054688, 142.20556640625, 110.8455810546875, 1.48944091796875, -74.44366455078125, 14.055091857910156, -7.3075103759765625, -2.13970947265625, 26.142333984375, -115.541259765625, -39.9521484375, 52.0299072265625, -21.326446533203125, 169.01394653320312, 236.7789764404297, 214.48046875, 20.36841583251953, 89.6531982421875, 93.2682876586914, 1.22857666015625, 16.31744384765625, 125.63153076171875, 183.27239990234375, -15.51934814453125, 305.670654296875, 278.4092712402344, 6.255126953125, 123.89077758789062, 139.142333984375, 74.5452651977539, 144.96795654296875, 19.554933547973633, -134.19647216796875, 181.892822265625, 248.96725463867188, -225.8656005859375, -9.812957763671875, 111.60479736328125, -63.76458740234375, -156.6771240234375, 140.95762634277344, 185.28427124023438, 63.23182678222656, -1.365875244140625, -37.1337890625, -25.21915626525879, 24.70330810546875, 64.48949432373047, 83.47003173828125, 176.012939453125, -149.8939208984375, 287.9141845703125, 44.16096496582031, 11.952392578125, 28.651626586914062, 86.30181884765625, 23.33922576904297, -68.25686645507812, -182.58505249023438, 246.01124572753906, 96.12890625, 123.05841064453125, 30.317047119140625, 46.62206268310547, -94.46996307373047, 41.257835388183594, 6.022041320800781, 123.34130859375, 122.2422103881836, 131.91937255859375, -14.549793243408203, -32.512939453125, 145.3267822265625, 22.79437255859375, -3.81561279296875, -13.510498046875, 16.338027954101562, 114.00570678710938, 39.3978271484375, 203.18341064453125, 192.36279296875, 58.44830322265625, 180.18209838867188, -182.39822387695312, 6.6811676025390625, -24.431121826171875, -34.81573486328125, -19.444259643554688, 276.6839294433594, 139.92955017089844, 53.77545166015625, -3.624481201171875, 149.54190063476562, 121.365478515625, -105.19407653808594, 187.87167358398438, 179.0194091796875, 49.819580078125, 7.96600341796875, -64.81207275390625, 100.20146179199219, 37.8013916015625, 22.132598876953125, 27.0594482421875, -110.21965026855469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000206.npy"}
{"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 56.53125, "std": 126.98229217529297, "min": -529.4523315429688, "p10": -70.13501434326172, "median": 42.66292190551758, "p90": 209.8006332397461, "max": 491.7392578125, "pos_frac": 0.703125, "sample": [219.81787109375, 122.80889892578125, 42.621246337890625, -41.405555725097656, 13.469955444335938, 106.25254821777344, 49.07586669921875, 210.05421447753906, 122.23309326171875, -16.176624298095703, -95.9486083984375, 135.18353271484375, -121.17828369140625, 23.40357208251953, 88.0616455078125, 227.74485778808594, 10.850818634033203, 133.12008666992188, 173.06381225585938, 31.67059326171875, 147.44857788085938, -69.28526306152344, -159.9183349609375, -16.461318969726562, 27.27469253540039, 147.56243896484375, 41.2373046875, -28.912368774414062, -261.9165954589844, -2.5715484619140625, 118.66117858886719, 20.89665985107422, 87.42303466796875, -100.22235107421875, 209.69195556640625, 137.9277801513672, 56.02362060546875, 29.349349975585938, 69.17179870605469, 183.388671875, -49.210723876953125, 6.739959716796875, 133.43630981445312, -16.0164794921875, -2.39044189453125, 16.33074951171875, 94.29817199707031, -30.393707275390625, 9.29754638671875, -7.2775421142578125, -529.4523315429688, -33.3001708984375, -15.687103271484375, -2.2437744140625, 204.101806640625, 0.0, 95.92523193359375, -15.494293212890625, 48.42962646484375, 231.56747436523438, 124.36483764648438, -51.627838134765625, 48.217552185058594, 59.793853759765625, 47.54730224609375, -43.82232666015625, 253.69610595703125, -199.31776428222656, 29.062240600585938, 21.934585571289062, 5.216552734375, -72.11776733398438, 110.41217041015625, 352.283447265625, 301.8702392578125, 320.937255859375, 63.44709777832031, 86.317138671875, 85.711181640625, 252.650146484375, 26.365554809570312, -136.99917602539062, -24.370574951171875, 80.57085418701172, 15.650026321411133, 491.7392578125, -8.833465576171875, 59.97450256347656, -90.57662963867188, 27.155670166015625, 24.71575927734375, 52.71858215332031, -41.745880126953125, 127.88142395019531, 195.84808349609375, 186.28692626953125, -63.11248779296875, -22.867034912109375, 14.76513671875, -126.664306640625, 133.0357666015625, 264.412109375, -36.03582763671875, 102.5102767944336, 12.012739181518555, 161.16860961914062, 9.92657470703125, -19.998809814453125, 121.76028442382812, 170.59445190429688, 42.70459747314453, 215.31439208984375, 22.131141662597656, 83.27914428710938, 41.8299560546875, 16.953826904296875, 191.8958740234375, 170.1708526611328, 214.9150390625, -100.36053466796875, 150.2799835205078, -211.73709106445312, 135.94622802734375, 208.41879272460938, 0.9397430419921875, 203.4391326904297, 50.73655700683594, 84.52496337890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000207.npy"}
{"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 70.49407958984375, "std": 114.38600158691406, "min": -185.07095336914062, "p10": -34.33326034545898, "median": 59.31986427307129, "p90": 186.4533264160156, "max": 745.5185546875, "pos_frac": 0.7421875, "sample": [17.2039794921875, 13.6116943359375, -41.69160461425781, 75.30863952636719, -43.0477294921875, -17.248252868652344, -8.017101287841797, 177.96255493164062, -9.254425048828125, 236.7589111328125, 56.1298942565918, 66.91580200195312, 100.45065307617188, 48.513214111328125, 75.92910766601562, 327.25927734375, 136.25564575195312, 41.95814514160156, 1.42999267578125, 350.417236328125, -162.34442138671875, 180.19451904296875, -41.86857604980469, 130.22967529296875, 31.19085693359375, 95.61502075195312, 196.9036865234375, 87.82284545898438, 35.42237091064453, -23.543968200683594, 286.77276611328125, 78.80976867675781, 73.58633422851562, 177.3394775390625, -36.60869598388672, 177.065673828125, -167.8294677734375, 131.07687377929688, 115.11553955078125, 40.533538818359375, 9.45551872253418, -175.7603759765625, 0.72454833984375, 91.96501159667969, 137.35305786132812, -28.32257080078125, 139.39401245117188, 28.066650390625, 0.0, 188.56353759765625, -185.07095336914062, -4.569557189941406, 44.350128173828125, -45.904876708984375, 72.50582885742188, 201.18212890625, 27.513580322265625, -20.112045288085938, 10.697242736816406, 249.9005126953125, -26.036895751953125, 124.75114440917969, 1.3313064575195312, 130.4453125, 46.21588134765625, -9.31793212890625, 45.49554443359375, 20.320266723632812, 84.9803466796875, 157.00521850585938, 745.5185546875, 0.0, 89.5584487915039, 218.803466796875, 71.40580749511719, 52.22601318359375, -12.097984313964844, 242.90719604492188, 184.16278076171875, 170.23117065429688, -34.266639709472656, 182.69427490234375, 185.5489501953125, 123.314208984375, 0.0680694580078125, 88.87519073486328, 34.318603515625, -9.679996490478516, 63.12876892089844, 79.06622314453125, -34.48870849609375, 90.59237670898438, 44.049041748046875, 151.30340576171875, 111.78263854980469, 93.0076904296875, 212.01419067382812, 22.958709716796875, 83.12399291992188, -9.6075439453125, 40.02250671386719, 26.923513412475586, -9.7327880859375, 38.86011505126953, 67.85391235351562, 62.50983428955078, -13.211349487304688, 163.1300048828125, -3.7165145874023438, 3.788177490234375, 76.32583618164062, 97.18331909179688, 150.76133728027344, -15.8948974609375, 26.11688232421875, -152.48245239257812, 26.848526000976562, 87.19506072998047, -108.79046630859375, 145.82916259765625, 216.4671630859375, 175.91070556640625, 43.201133728027344, 72.29061889648438, 177.97222900390625, -102.60736083984375, -4.3192138671875, 164.836669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000208.npy"}
{"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 42.52222442626953, "std": 108.4056396484375, "min": -281.58563232421875, "p10": -100.69842987060547, "median": 31.876571655273438, "p90": 167.4140441894531, "max": 326.544189453125, "pos_frac": 0.6640625, "sample": [178.13922119140625, 119.3260498046875, 145.82461547851562, 128.85914611816406, -78.1387939453125, -3.5537567138671875, 140.74526977539062, -281.58563232421875, -84.393798828125, 101.3197021484375, 165.161865234375, -26.88312530517578, 66.92550659179688, 5.332969665527344, 150.1798095703125, 164.87425231933594, -216.1461181640625, 18.55904197692871, 19.290374755859375, 141.3258056640625, 143.455078125, 29.04559326171875, 8.464187622070312, -105.5123519897461, 71.118896484375, 18.417095184326172, 59.13404846191406, 34.80194091796875, -84.22201538085938, 102.38652038574219, 130.1181640625, -167.4051513671875, -47.78106689453125, -161.28176879882812, -111.30340576171875, -90.614990234375, 272.4330749511719, -2.3277435302734375, 0.0, 2.5973358154296875, -78.39337158203125, 76.7786865234375, 172.66912841796875, -23.010650634765625, -79.99578857421875, 58.7841796875, 25.22704315185547, 16.958805084228516, -97.57330322265625, 42.498046875, 125.32284545898438, 108.32846069335938, 106.31869506835938, 31.024200439453125, 1.261566162109375, 6.446672439575195, 23.47186279296875, 24.808441162109375, 115.74187469482422, -5.887083053588867, 86.13034057617188, -203.90823364257812, 8.96307373046875, 282.1434326171875, 127.44174194335938, -99.87458801269531, -13.541259765625, 29.47119140625, 97.99163818359375, 149.18161010742188, 96.7845458984375, 198.92172241210938, 20.196395874023438, 82.30126953125, -51.21893310546875, 0.0, 32.72894287109375, 138.61068725585938, -1.1670684814453125, 101.22955322265625, 61.259368896484375, -15.08868408203125, 47.745880126953125, -54.52191162109375, -34.93168640136719, 178.8851318359375, 114.55780029296875, -51.699867248535156, -102.95611572265625, -111.25372314453125, 50.53424072265625, 64.03047180175781, -9.258941650390625, 122.13037109375, -7.12348747253418, -24.81689453125, 120.67193603515625, 117.35137939453125, 226.48471069335938, -102.6207275390625, -99.17694091796875, 4.8609466552734375, -128.34780883789062, -25.41973876953125, 8.647430419921875, -137.71478271484375, -2.167205810546875, 29.92743682861328, 133.00369262695312, 156.60543823242188, 110.1624755859375, 105.54473876953125, 187.60836791992188, 80.09925842285156, 12.59317398071289, 211.581787109375, 326.544189453125, 66.7744140625, 72.79429626464844, 191.8421630859375, 262.9779052734375, -126.23980712890625, 127.7176742553711, 104.01315307617188, 225.04232788085938, 162.13214111328125, -35.58522033691406, 137.7916259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000209.npy"}
{"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 61.55084991455078, "std": 107.51420593261719, "min": -253.56045532226562, "p10": -54.45604400634765, "median": 57.14292907714844, "p90": 203.98888244628907, "max": 324.5350341796875, "pos_frac": 0.703125, "sample": [8.31884765625, -18.66248321533203, 67.11654663085938, -7.089111328125, 226.3631591796875, 43.30107116699219, 179.64654541015625, 195.14129638671875, 203.92718505859375, 113.02473449707031, 29.352142333984375, -45.69999313354492, 12.110885620117188, 70.91970825195312, 187.94775390625, 5.539405822753906, 88.33226013183594, 89.06104278564453, 68.67233276367188, -7.999267578125, 31.446834564208984, 101.08352661132812, -53.22406005859375, -34.41743469238281, 58.68719482421875, 51.873207092285156, -8.37310791015625, 324.39678955078125, -175.23931884765625, 17.676422119140625, 92.43295288085938, 91.46934509277344, 9.014251708984375, -17.36590576171875, 183.86639404296875, 118.90225219726562, 25.2049560546875, 116.93035888671875, -1.597320556640625, 93.0491943359375, 26.251327514648438, 173.4004669189453, 5.51567268371582, -51.74156188964844, 82.96082305908203, 58.97540283203125, 56.835174560546875, -57.33067321777344, 204.13284301757812, 82.13353729248047, 77.46746826171875, -5.28594970703125, 182.72744750976562, 211.33544921875, 112.10364532470703, 156.5250244140625, 84.0198974609375, -253.56045532226562, 170.58734130859375, -18.13250732421875, 161.484619140625, 81.27200317382812, 253.69015502929688, 135.34181213378906, -43.05352783203125, 238.64198303222656, -33.98541259765625, 2.8427734375, 188.48828125, 117.23544311523438, -108.93978881835938, -150.23524475097656, 131.1514892578125, 59.89765930175781, 190.36306762695312, 324.5350341796875, 0.18951034545898438, 149.76683044433594, -93.26763916015625, -141.08700561523438, -13.38381576538086, 39.91349792480469, -7.981758117675781, 217.9169921875, -9.345222473144531, 61.258026123046875, 298.6788330078125, 57.45068359375, -15.32220458984375, -67.73289489746094, -83.14308166503906, 10.958793640136719, 4.838218688964844, -93.55380249023438, -2.8045730590820312, 47.325439453125, -36.15824890136719, 266.41363525390625, 67.74087524414062, 75.43339538574219, 156.0880126953125, 265.3252868652344, 35.82349395751953, 125.71646118164062, 294.47796630859375, 161.44583129882812, 16.330429077148438, 51.53363037109375, 10.833526611328125, -44.745147705078125, 82.74736022949219, 62.8861083984375, -3.2629547119140625, 169.0430908203125, 43.22032165527344, -135.15423583984375, 15.282806396484375, 216.29901123046875, 33.25433349609375, -114.14559936523438, 102.90350341796875, -47.1656494140625, 154.68283081054688, 63.908203125, -44.038482666015625, -10.6002197265625, -71.9698486328125, 170.89251708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000210.npy"}
{"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 79.933349609375, "std": 98.47713470458984, "min": -272.6819763183594, "p10": -23.776094818115233, "median": 70.80691528320312, "p90": 226.62977905273436, "max": 304.3135986328125, "pos_frac": 0.8046875, "sample": [-5.697242736816406, 85.04493713378906, 240.385498046875, 166.97842407226562, 170.84085083007812, 101.88966369628906, 97.77813720703125, 16.20171546936035, -5.99072265625, 61.7872314453125, 43.51091003417969, 0.570281982421875, 55.668670654296875, 45.4100341796875, -25.65192413330078, 271.7165832519531, 23.070716857910156, -1.1606330871582031, -272.6819763183594, -79.689453125, 136.39999389648438, 74.30964660644531, 32.69679260253906, -7.145965576171875, 94.19593048095703, 116.7293701171875, 81.83859252929688, -68.27987670898438, 63.1756591796875, -4.80035400390625, -19.844520568847656, 182.65992736816406, 30.257675170898438, 170.71652221679688, 66.46734619140625, 176.95101928710938, 111.18795776367188, -63.16030502319336, 4.062919616699219, 180.10244750976562, 40.16033935546875, -99.89105224609375, 27.07830810546875, 61.757537841796875, -37.04876708984375, 231.667236328125, 174.25555419921875, 127.60418701171875, 97.56126403808594, 23.94464874267578, 157.031005859375, 20.494354248046875, 46.23004150390625, 27.690231323242188, 304.3135986328125, 185.203857421875, 253.65631103515625, -103.17872619628906, 95.08624267578125, 96.71881103515625, -12.25335693359375, -7.5262451171875, 77.31846618652344, 293.70623779296875, 39.676971435546875, -29.439529418945312, -22.97216796875, 157.22116088867188, 133.91220092773438, 4.19401741027832, -83.7154541015625, 4.6419677734375, -8.641414642333984, 39.944114685058594, 6.877685546875, 132.89456176757812, 168.31332397460938, -39.39239501953125, 212.39874267578125, 163.13088989257812, 105.37387084960938, 85.37083435058594, 67.30418395996094, -10.500244140625, 178.21343994140625, 175.38131713867188, 101.71188354492188, 2.570556640625, 28.68744659423828, 0.8975753784179688, -46.45549011230469, 226.003173828125, 131.53497314453125, 21.259368896484375, 239.233642578125, 50.812889099121094, 300.3072509765625, 247.53561401367188, 20.816707611083984, 228.09185791015625, 133.87945556640625, 230.12319946289062, 116.00192260742188, -0.3261871337890625, -133.75457763671875, 21.41461181640625, 88.73046875, 62.5277099609375, 259.9943542480469, 132.70089721679688, 134.87384033203125, 10.201507568359375, 157.89877319335938, 82.5638427734375, 110.26190185546875, 56.317413330078125, 255.18692016601562, 94.467529296875, 196.80075073242188, 41.353546142578125, 61.73472595214844, 86.29186248779297, 15.172760009765625, 126.57600402832031, 35.10845184326172, 184.05459594726562, 79.68114471435547, 128.35374450683594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000211.npy"}
{"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 47.636680603027344, "std": 103.10509490966797, "min": -317.9579772949219, "p10": -56.61048278808593, "median": 45.66416358947754, "p90": 167.62321624755856, "max": 384.3727722167969, "pos_frac": 0.734375, "sample": [17.219951629638672, 75.91455078125, 52.8426513671875, -19.400352478027344, -27.804428100585938, -106.22105407714844, -14.37554931640625, 109.66697692871094, 71.3414306640625, -81.1337890625, 162.86676025390625, -32.99662780761719, 204.9737548828125, 3.025115966796875, 102.94195556640625, 19.5260009765625, 15.716203689575195, 138.71893310546875, 107.87190246582031, 105.80364990234375, 37.53324890136719, 43.131683349609375, -160.74493408203125, 45.92205810546875, 165.25425720214844, 71.94952392578125, 117.56097412109375, -35.816200256347656, 93.45954895019531, 29.507354736328125, -44.87107467651367, -69.79852294921875, 45.60551452636719, -14.545272827148438, 45.72281265258789, 77.0538330078125, 121.61065673828125, 259.61785888671875, 29.48150634765625, 129.02880859375, 160.19619750976562, -226.61676025390625, 16.9896240234375, 9.995428085327148, -62.559906005859375, 147.52423095703125, 51.197174072265625, 72.59228515625, -119.50254821777344, -116.83016967773438, 93.74004364013672, 90.69012451171875, -255.57147216796875, 163.9398193359375, 7.9896240234375, -26.30816650390625, -74.32010650634766, 273.2213134765625, 47.908538818359375, -18.65894317626953, -39.054908752441406, -52.57078552246094, 173.15078735351562, 25.0556640625, 4.713764190673828, 131.20779418945312, -37.23568344116211, 0.267486572265625, 71.07748413085938, 194.7333984375, 88.024169921875, 195.1173095703125, 6.1514434814453125, 103.74563598632812, 190.26612854003906, 73.1080322265625, 42.600135803222656, 205.007080078125, -26.09571075439453, 193.66213989257812, -30.038970947265625, 155.7056884765625, 5.3787689208984375, 56.78289794921875, -188.69313049316406, 34.116119384765625, 64.92916870117188, -49.925567626953125, -39.86834716796875, 2.1294021606445312, 130.1746826171875, 182.56243896484375, 69.13931274414062, -11.50994873046875, 152.065673828125, 84.89924621582031, 16.00787353515625, 116.97687530517578, -15.0430908203125, 42.678802490234375, 111.69843292236328, 272.17730712890625, 13.92486572265625, 0.945526123046875, 91.7989501953125, -317.9579772949219, 151.20993041992188, 92.81259155273438, 140.87551879882812, -42.537078857421875, 103.17001342773438, -54.06072998046875, 51.895965576171875, 12.517488479614258, 63.298919677734375, -63.46600341796875, 74.77059936523438, -18.162376403808594, 28.963733673095703, 37.5792236328125, 31.81219482421875, 78.00057983398438, 2.88232421875, 197.134765625, 14.69866943359375, 115.30775451660156, 384.3727722167969, 71.65390014648438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000212.npy"}
{"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 54.16126251220703, "std": 117.94307708740234, "min": -356.4413757324219, "p10": -76.9490852355957, "median": 41.036224365234375, "p90": 197.68509826660156, "max": 497.3504638671875, "pos_frac": 0.6953125, "sample": [84.95123291015625, 179.73703002929688, 16.29198455810547, 95.8961181640625, 11.522186279296875, 119.14727783203125, 147.4978485107422, 135.16677856445312, 24.57183837890625, 260.00006103515625, 148.0389862060547, 138.64984130859375, 81.06846618652344, 34.538490295410156, 7.819868087768555, 224.03009033203125, -28.847305297851562, 20.65972900390625, 36.85247802734375, 52.6302490234375, 63.94232177734375, 1.347686767578125, 233.44680786132812, -50.371337890625, 357.15631103515625, 189.5303955078125, 15.789962768554688, 58.574066162109375, 48.64434051513672, 32.2147216796875, 55.90544128417969, -356.4413757324219, -45.16487121582031, -24.78375244140625, 121.83827209472656, 0.884613037109375, 62.396461486816406, 79.12958526611328, 76.96426391601562, -106.5709228515625, 230.20692443847656, 194.1571502685547, -15.311447143554688, 72.79154968261719, -136.84835815429688, 315.61199951171875, 73.33566284179688, 14.878082275390625, -40.63690185546875, 72.04153442382812, 12.22012710571289, 223.53884887695312, 187.18569946289062, 42.1495361328125, 138.95346069335938, -54.34577941894531, -18.796615600585938, 14.997505187988281, -51.62596130371094, 197.08926391601562, 11.336517333984375, 452.237060546875, -7.7444610595703125, -13.592041015625, 46.6605224609375, 58.268310546875, 256.83123779296875, -12.435089111328125, -139.16058349609375, -1.93377685546875, 126.69728088378906, 0.53912353515625, 101.67355346679688, -87.10043334960938, -44.2105712890625, -1.2478103637695312, 89.30628967285156, 170.57769775390625, -11.245338439941406, 92.01974487304688, -10.590980529785156, 15.334972381591797, 3.311819076538086, -102.55133056640625, -45.390892028808594, 119.4398193359375, -171.90402221679688, 107.44140625, -3.81744384765625, 151.42254638671875, 17.070159912109375, 271.41925048828125, 82.54959869384766, 53.270263671875, 51.04986572265625, -122.40005493164062, 64.85638427734375, -107.35987854003906, -71.94024658203125, 497.3504638671875, -13.90362548828125, 36.69593811035156, 142.76300048828125, 54.032501220703125, -80.306640625, 199.07537841796875, 136.73086547851562, -44.40452575683594, 39.92291259765625, 0.0, 13.493179321289062, 160.19451904296875, 117.38357543945312, 205.15797424316406, 14.457111358642578, -128.68792724609375, 136.78091430664062, 5.177976608276367, -19.129806518554688, 52.4853515625, 23.166141510009766, 88.8104248046875, -76.10310363769531, 50.84761047363281, 85.92507934570312, -78.92304229736328, -50.873687744140625, -100.4142837524414], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000213.npy"}
{"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 69.63395690917969, "std": 101.2620849609375, "min": -220.79266357421875, "p10": -47.57189598083496, "median": 58.823638916015625, "p90": 221.08383789062498, "max": 413.61553955078125, "pos_frac": 0.75, "sample": [20.15399169921875, 100.54592895507812, 7.2296142578125, 15.22998046875, 105.61644744873047, 225.59225463867188, 72.97589111328125, -24.676834106445312, 51.6767578125, -37.397369384765625, 82.66986083984375, 158.85792541503906, 63.23028564453125, 111.57095336914062, -31.12213134765625, 90.28302001953125, -62.72027587890625, 188.82876586914062, 49.285186767578125, 102.1356201171875, 29.39868927001953, -111.944091796875, 34.02379608154297, 116.8958740234375, 259.45513916015625, 69.55471801757812, 82.29106903076172, -69.377685546875, 94.51171875, 98.99459075927734, -2.0389404296875, 18.839813232421875, 18.371826171875, -58.763893127441406, 115.82032775878906, 46.83984375, 368.5379638671875, 287.99468994140625, 258.668212890625, 36.462554931640625, 65.82035827636719, -62.877777099609375, -65.57073974609375, 72.29544067382812, 54.73486328125, -46.21164321899414, -9.93505859375, 254.49078369140625, 44.032012939453125, -1.2592582702636719, 75.05508422851562, 97.051025390625, -51.49266052246094, 175.6654052734375, 10.549217224121094, 169.0090789794922, -44.154510498046875, 3.20880126953125, -13.920501708984375, 115.5889892578125, 129.96905517578125, 182.5799560546875, 124.30232238769531, 30.6009521484375, 71.52493286132812, 33.82672119140625, 22.194091796875, -78.28229522705078, -50.745819091796875, 0.0, -20.168350219726562, 103.18814086914062, -2.39520263671875, 30.61456298828125, 149.59100341796875, 24.37786102294922, 74.689208984375, -50.882877349853516, 134.0806884765625, 114.34026336669922, -89.34278869628906, -220.79266357421875, 129.9537353515625, 87.03265380859375, 101.28532409667969, -43.009490966796875, 136.77316284179688, 22.874271392822266, -7.2730712890625, 27.292739868164062, -98.76406860351562, 95.69147491455078, 205.12844848632812, 204.48992919921875, 232.74948120117188, 56.22746276855469, 28.415756225585938, 77.59896850585938, -42.16668701171875, 61.41981506347656, 224.36431884765625, 156.711669921875, 26.94122314453125, 95.96612548828125, 15.181121826171875, 413.61553955078125, 260.33416748046875, 283.8612976074219, -5.17095947265625, 65.83392333984375, 265.126220703125, 24.3330078125, 34.828582763671875, -3.3507080078125, 94.21917724609375, 219.67791748046875, 146.16275024414062, 54.3043212890625, -8.64764404296875, 13.517578125, 114.08768463134766, 19.54236602783203, 114.54177856445312, 116.17750549316406, -17.11263084411621, 257.36273193359375, 91.75473022460938, 23.412109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000214.npy"}
{"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 44.80018615722656, "std": 112.18980407714844, "min": -215.43280029296875, "p10": -79.4102813720703, "median": 20.730224609375, "p90": 190.40330810546874, "max": 438.43115234375, "pos_frac": 0.65625, "sample": [32.57553482055664, -49.12757873535156, 46.450347900390625, 204.18038940429688, 242.10009765625, 149.54190063476562, 139.41302490234375, 155.51150512695312, 123.56851196289062, -7.630523681640625, 163.15658569335938, 88.48419189453125, 308.6047668457031, -114.23471069335938, 15.881683349609375, -13.23388671875, 178.4528045654297, 259.06939697265625, 438.43115234375, 152.9049072265625, 156.68283081054688, 2.903778076171875, 0.0, -87.58856201171875, 6.028980255126953, 10.69369125366211, 84.07850646972656, 160.95477294921875, 300.5503234863281, 8.208000183105469, 193.845947265625, 25.810150146484375, 19.76953125, -215.43280029296875, 11.827762603759766, 88.98123168945312, 151.5086669921875, -75.90530395507812, -37.803314208984375, 191.28668212890625, -9.79498291015625, -19.02960205078125, 7.14031982421875, -91.52299499511719, 5.789571762084961, 55.15757751464844, -75.27728271484375, -14.322341918945312, 207.8189697265625, 20.305465698242188, 15.723350524902344, 360.22802734375, 221.99557495117188, -17.2659912109375, -169.672119140625, -16.493755340576172, 73.37007141113281, -47.121307373046875, 0.0, 114.13739013671875, -38.60227966308594, 35.51124572753906, 51.95098876953125, -42.255706787109375, 57.649749755859375, 18.34478759765625, -98.50009155273438, -8.007072448730469, 161.21591186523438, 40.174163818359375, 17.448486328125, 30.17498207092285, -0.8354339599609375, 146.58062744140625, -183.92376708984375, 144.84564208984375, 53.2166748046875, -145.01171875, 21.237354278564453, 82.64785766601562, 78.29191589355469, 55.724578857421875, 65.53353881835938, -48.254180908203125, -120.13316345214844, -2.8703155517578125, 137.58709716796875, 9.449249267578125, 109.04583740234375, -150.37472534179688, -37.4552001953125, 24.80792236328125, -29.833267211914062, -27.012451171875, 22.490692138671875, 21.154983520507812, 13.761072158813477, 46.77586364746094, 116.63432312011719, 190.02471923828125, -66.869384765625, 99.0665283203125, -48.45951843261719, 4.015476226806641, -129.7781524658203, 8.290069580078125, -3.7209625244140625, 116.13609313964844, -48.33140563964844, 177.10198974609375, 23.82415771484375, 2.7003440856933594, 55.3046875, 9.78021240234375, 120.44917297363281, 91.40069580078125, -38.818603515625, 64.63267517089844, -137.51141357421875, -11.5296630859375, 235.18740844726562, -69.30108642578125, 67.6968994140625, 138.95751953125, 4.195709228515625, -62.10573959350586, -112.96697998046875, 294.1993713378906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000215.npy"}
{"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 63.12995147705078, "std": 119.19027709960938, "min": -269.0211486816406, "p10": -67.4413345336914, "median": 50.17616271972656, "p90": 197.22234039306642, "max": 511.341552734375, "pos_frac": 0.7265625, "sample": [-40.711509704589844, 173.7438507080078, 176.3363494873047, 111.34158325195312, 16.9517822265625, 173.20559692382812, -170.2120361328125, 61.95745849609375, 56.042236328125, -18.023460388183594, 185.36685180664062, 130.05731201171875, 83.22625732421875, -11.887107849121094, 51.35150146484375, 95.85678100585938, 511.341552734375, 3.8963394165039062, 144.14761352539062, 168.59439086914062, 58.82586669921875, -127.61273193359375, 32.35441589355469, -29.341278076171875, 12.512870788574219, 26.96392822265625, -90.3900146484375, -57.26780700683594, 40.7640380859375, 122.26171875, 31.26788330078125, -182.06658935546875, -66.41790771484375, 13.797027587890625, 69.60240173339844, 41.408363342285156, 189.19937133789062, -269.0211486816406, 303.8694152832031, 197.03565979003906, -25.300621032714844, 278.662109375, 151.79345703125, 186.9295196533203, 175.6104278564453, 169.96417236328125, 64.97406005859375, 131.92437744140625, -0.239349365234375, 30.015541076660156, 99.64886474609375, -24.288803100585938, 195.0479736328125, 184.05093383789062, 5.309028625488281, 81.71044921875, 49.000823974609375, 197.65792846679688, -84.2635498046875, 19.35858917236328, 0.0, 31.4691162109375, -65.25213623046875, 180.23928833007812, 38.556488037109375, 102.26806640625, 321.5243835449219, -112.69326782226562, -118.75942993164062, -13.70068359375, 115.4759521484375, 111.08013916015625, -137.65994262695312, 7.01422119140625, 191.828125, 236.51931762695312, -7.02362060546875, 201.16595458984375, 90.732421875, 34.31964111328125, -61.70191955566406, -29.299530029296875, 149.32037353515625, 45.843505859375, 324.3057861328125, 110.385498046875, -69.82933044433594, -49.9326171875, 209.2578125, 222.48828125, -110.97055053710938, -27.35687255859375, 40.1759033203125, 131.51385498046875, -26.627410888671875, 80.2718505859375, 23.0428466796875, -4.87969970703125, 135.52359008789062, 45.022735595703125, 276.3729248046875, 60.24789810180664, 0.0, 98.92755126953125, 154.158935546875, -171.83876037597656, 18.96258544921875, 120.7209701538086, 105.7442398071289, -11.74041748046875, 4.88575553894043, 91.70782470703125, 80.23162841796875, 76.961669921875, 221.91421508789062, 105.04322814941406, 2.412506103515625, 55.6619873046875, 4.4598388671875, 346.155517578125, 6.0052490234375, -207.08053588867188, 96.1640625, 88.44781494140625, 20.67279052734375, 9.2265625, -63.752410888671875, 38.4078369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000216.npy"}
{"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 76.84033203125, "std": 117.0416259765625, "min": -236.98724365234375, "p10": -47.78045043945311, "median": 58.886871337890625, "p90": 211.06259155273438, "max": 420.910888671875, "pos_frac": 0.7578125, "sample": [46.05436706542969, 84.16622924804688, -94.90228271484375, 54.6441650390625, 75.56109619140625, -44.934112548828125, 1.9948577880859375, 382.1620788574219, -68.47782897949219, 54.20599365234375, 177.1214141845703, 12.5804443359375, 73.10105895996094, 145.88302612304688, 20.9200439453125, 243.2666015625, 42.286033630371094, 122.60293579101562, 160.25363159179688, -26.230682373046875, 28.430633544921875, 53.034088134765625, 118.36124420166016, 159.11801147460938, 420.910888671875, 2.1888885498046875, 198.2783966064453, 91.49595642089844, 106.1756591796875, 240.34429931640625, 50.725341796875, 191.17596435546875, -17.914535522460938, 350.81842041015625, 342.6204833984375, -3.6905517578125, 323.6249084472656, 143.5430908203125, 195.966796875, 81.257568359375, -121.9512939453125, 301.3370361328125, 136.41390991210938, 193.148681640625, -19.835693359375, 112.26039123535156, 125.71448516845703, 1.8850250244140625, 35.5931396484375, 30.02142333984375, 11.488452911376953, -31.873863220214844, -30.111053466796875, 98.11767578125, 119.45613098144531, -122.48971557617188, 69.34011840820312, 191.98001098632812, 46.169342041015625, -70.29356384277344, 68.70994567871094, 7.47369384765625, 48.858123779296875, 172.75131225585938, 39.41156005859375, -5.195274353027344, 9.338325500488281, 159.668701171875, 5.705480575561523, 31.04241943359375, 158.650146484375, 218.32525634765625, 171.6372528076172, -10.368656158447266, 213.64337158203125, -95.88212585449219, 63.12957763671875, -15.611045837402344, 81.46243286132812, 237.005615234375, 31.70362091064453, 186.1314697265625, 48.31062316894531, -11.400070190429688, -144.5128173828125, 195.04940795898438, 171.58285522460938, 260.828857421875, 15.6591796875, -151.77383422851562, 80.10494995117188, 93.23248291015625, -54.421905517578125, 209.95654296875, 138.83026123046875, 86.18869018554688, -13.550209045410156, -13.640830993652344, 5.245025634765625, 78.65625, 47.89007568359375, 0.0, 32.526031494140625, 143.33572387695312, -175.50054931640625, 186.99795532226562, 189.75399780273438, 12.10870361328125, 93.38885498046875, 354.1905212402344, 129.7452392578125, -18.214614868164062, -1.3507080078125, 140.74871826171875, 142.83935546875, 41.58757019042969, 139.03001403808594, 49.611785888671875, -236.98724365234375, -114.90289306640625, 201.73568725585938, -128.32421875, 100.24990844726562, 6.981048583984375, 111.51885986328125, -41.728057861328125, 43.59967041015625, -2.3001441955566406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000217.npy"}
{"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 69.17512512207031, "std": 114.34358215332031, "min": -213.6647491455078, "p10": -52.43563880920409, "median": 56.18251037597656, "p90": 204.4835266113281, "max": 555.0714721679688, "pos_frac": 0.75, "sample": [39.02532958984375, 14.297500610351562, 95.86135864257812, 213.97811889648438, -76.00358581542969, 39.91410827636719, 17.842437744140625, 555.0714721679688, -9.186737060546875, 45.18435287475586, -151.00863647460938, -13.417678833007812, 198.1525115966797, 233.96597290039062, 142.23953247070312, -92.21307373046875, -56.0477294921875, 169.8593292236328, 1.9290771484375, 192.21546936035156, 0.011463165283203125, 122.28337097167969, 215.66494750976562, 153.90821838378906, 82.15167236328125, -8.607162475585938, -1.939910888671875, 2.68798828125, -26.468292236328125, 175.7189178466797, 86.87747192382812, 298.71075439453125, 2.8198318481445312, 180.26150512695312, 30.38958740234375, 182.02447509765625, 149.22454833984375, -71.47357177734375, 213.24978637695312, 120.99441528320312, 229.05755615234375, -50.88759994506836, 64.09506225585938, 113.12693786621094, 33.03364562988281, 132.636962890625, -9.863136291503906, -168.78143310546875, 15.445098876953125, 307.40875244140625, 145.23989868164062, 21.820281982421875, 175.21701049804688, -48.720008850097656, 184.27487182617188, 2.7914981842041016, 338.3358154296875, 85.34390258789062, 93.71620178222656, -99.7464599609375, 33.544708251953125, 26.44873046875, 102.32888793945312, 14.234542846679688, 0.0, 72.20144653320312, 127.14224243164062, 2.31634521484375, -35.4761962890625, 129.37643432617188, 154.92922973632812, 133.620849609375, 32.22257995605469, 131.14251708984375, 189.587646484375, 55.558563232421875, -86.29303741455078, -8.241348266601562, 53.5087890625, 35.400665283203125, 170.34033203125, 187.15814208984375, 210.94757080078125, 20.3826904296875, 48.536224365234375, 73.74453735351562, -83.38137817382812, -37.62944030761719, 133.3167724609375, 56.80645751953125, -38.04389953613281, 26.09235382080078, -22.052734375, -210.43115234375, 83.3922119140625, 83.27550506591797, 190.09022521972656, 345.34161376953125, -163.35003662109375, 126.333251953125, 3.293426513671875, 96.63662719726562, -26.284080505371094, 102.8792724609375, 81.98260498046875, 94.3619384765625, -213.6647491455078, 132.60226440429688, -3.311126708984375, 147.16806030273438, 23.77398681640625, 203.07916259765625, 103.441162109375, 64.83560180664062, -63.605712890625, 251.81292724609375, 104.93695068359375, 64.74017333984375, 10.340621948242188, -33.542724609375, -13.74481201171875, 14.44149398803711, 10.374847412109375, -18.871856689453125, 69.26904296875, 207.7603759765625, 32.330230712890625, 9.2630615234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000218.npy"}
{"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 73.6720199584961, "std": 120.59542846679688, "min": -331.80560302734375, "p10": -65.30285339355468, "median": 65.40491104125977, "p90": 251.39205932617188, "max": 420.1902770996094, "pos_frac": 0.765625, "sample": [-199.24755859375, 50.70082092285156, 79.92696380615234, 118.17805480957031, 308.73248291015625, 139.04388427734375, -65.72012329101562, 116.751220703125, 193.46066284179688, 54.5762939453125, 57.7445068359375, 6.010398864746094, -35.09613037109375, -127.05966186523438, -11.171379089355469, 48.601287841796875, 118.13995361328125, -29.3699951171875, 15.573654174804688, 190.55679321289062, 73.22508239746094, -0.9098720550537109, 50.3587646484375, 148.7479248046875, 105.47322845458984, 103.82205200195312, 80.52815246582031, 19.77713966369629, 90.53619384765625, 66.70864868164062, -138.47079467773438, 128.1957550048828, -168.07513427734375, 88.90412139892578, 37.116607666015625, 119.43928527832031, -65.1240234375, 145.85726928710938, 112.24334716796875, 204.3570556640625, 251.22698974609375, 267.95513916015625, 18.397836685180664, -79.49478149414062, 39.63111114501953, -69.54861450195312, -77.07284545898438, 85.68505859375, 135.38583374023438, 249.03070068359375, 0.0, 41.92417907714844, 21.630126953125, -14.6749267578125, -35.962135314941406, 70.0545654296875, -18.225494384765625, 140.0660400390625, 136.56155395507812, -23.579864501953125, 195.92916870117188, 338.36962890625, 15.714019775390625, -109.07199096679688, -10.197978973388672, 0.47528076171875, 7.5821533203125, 48.862613677978516, 7.453338623046875, 194.71388244628906, 16.7010498046875, 80.10504150390625, 34.13807678222656, 251.7772216796875, 82.51812744140625, -20.30487060546875, -99.51797485351562, -13.390869140625, 96.19305419921875, 293.1543884277344, 64.1011734008789, 251.99879455566406, 253.0179443359375, 119.12847900390625, 0.3543415069580078, 12.713134765625, 205.06649780273438, 0.5864734649658203, 310.4891662597656, 14.336181640625, 95.49795532226562, 13.986297607421875, 10.7281494140625, 44.48193359375, 25.354782104492188, 199.10189819335938, 129.6946563720703, 92.74746704101562, 105.369873046875, 230.42901611328125, 9.795112609863281, 265.05914306640625, 323.305419921875, -165.6078338623047, 252.64337158203125, 97.13543701171875, 124.54470825195312, -9.209548950195312, 118.19781494140625, 79.35369873046875, -9.898754119873047, 47.629547119140625, 80.67767333984375, -331.80560302734375, 243.24176025390625, 389.08135986328125, 29.633010864257812, 94.91862487792969, 182.81488037109375, 124.18807983398438, 17.21502685546875, -22.614227294921875, 113.54925537109375, 74.83787536621094, 420.1902770996094, 15.159942626953125, -14.995933532714844, -81.44235229492188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000219.npy"}
{"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 73.21768188476562, "std": 107.9820556640625, "min": -209.92291259765625, "p10": -62.8094497680664, "median": 61.89701843261719, "p90": 210.45888671874997, "max": 408.53399658203125, "pos_frac": 0.75, "sample": [201.20489501953125, 408.53399658203125, 169.50796508789062, 99.158935546875, -27.286672592163086, -16.9053955078125, 6.81085205078125, 173.46719360351562, 54.382164001464844, 116.21493530273438, 97.51924133300781, 48.608123779296875, 188.71380615234375, 187.11724853515625, -13.027496337890625, 131.1776123046875, 131.5406494140625, 241.75799560546875, 115.3994140625, 28.938766479492188, 38.911903381347656, 41.1402587890625, 6.199253082275391, 12.035453796386719, 111.30349731445312, 191.58767700195312, 40.27325439453125, 75.20863342285156, -77.57919311523438, -18.585494995117188, 92.79204559326172, 16.666000366210938, -32.44744873046875, -12.801498413085938, 135.28335571289062, -8.00701904296875, 45.12200927734375, 65.91696166992188, 11.76579475402832, 225.1732177734375, 176.78363037109375, 183.88800048828125, 221.01747131347656, -33.429931640625, 236.05999755859375, 178.51171875, -50.69941329956055, -209.92291259765625, -13.7899169921875, 39.7928466796875, 187.6695556640625, 109.57872772216797, -9.5687255859375, 295.26580810546875, 23.31688690185547, 52.484291076660156, -41.7781982421875, 9.820602416992188, -9.985595703125, 138.07577514648438, -69.87757873535156, 131.5569610595703, 186.67193603515625, 122.07383728027344, 35.971580505371094, -54.58235168457031, 1.2889404296875, 88.53387451171875, -93.73167419433594, 220.19851684570312, 207.20706176757812, 135.10977172851562, 45.710693359375, 217.6600341796875, 120.73220825195312, 67.94819641113281, -100.4207763671875, 54.75140380859375, 39.83343505859375, -12.23681640625, -32.75306701660156, 83.85966491699219, 28.36620330810547, 62.034454345703125, 74.00225830078125, 163.66470336914062, -62.09906005859375, 61.75958251953125, 282.4059143066406, 262.49786376953125, -127.09603881835938, 36.655609130859375, -33.0701904296875, -129.0205078125, 132.036376953125, 136.27597045898438, 237.4061279296875, 47.910133361816406, 87.968017578125, 22.82982635498047, 143.8070068359375, 116.31241607666016, -72.66921997070312, -149.84616088867188, 151.3926544189453, -24.39031982421875, 280.7017517089844, -79.42611694335938, 207.3726806640625, 225.66949462890625, 83.87664794921875, 57.208160400390625, 93.41351318359375, 13.471786499023438, 146.0079345703125, 37.89776611328125, 174.983642578125, 202.5784912109375, 148.0223388671875, 103.494140625, -205.28414916992188, 45.32574462890625, 170.25230407714844, 39.231903076171875, 17.36932373046875, -71.1628646850586, 114.802490234375, -64.46702575683594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000220.npy"}
{"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 59.12171173095703, "std": 121.61140441894531, "min": -234.84805297851562, "p10": -75.45525588989257, "median": 36.5355339050293, "p90": 219.30357818603514, "max": 409.5908508300781, "pos_frac": 0.671875, "sample": [-17.10931396484375, 101.41412353515625, 28.97802734375, -37.666595458984375, -20.806900024414062, 87.10162353515625, 14.065071105957031, 127.79415893554688, -131.8331756591797, -18.714582443237305, 12.034416198730469, 220.2457733154297, 23.5582275390625, 85.233154296875, -44.375701904296875, 126.82135009765625, 176.20242309570312, -77.50460052490234, 85.74462127685547, -7.601310729980469, 113.77066040039062, -56.29351806640625, 409.5908508300781, -99.73311614990234, 235.31124877929688, 269.59747314453125, 166.02597045898438, 308.0973815917969, 91.02058410644531, 70.38265991210938, 58.6793212890625, 41.466796875, 367.99609375, -191.01791381835938, -74.57696533203125, -16.21002197265625, -73.204833984375, 92.99798583984375, 0.539215087890625, 307.031005859375, -51.64598083496094, 47.52667236328125, -5.312591552734375, 31.95184326171875, -126.44209289550781, 25.45709228515625, 190.80288696289062, -69.74618530273438, 230.3135223388672, 20.59637451171875, -2.037761688232422, 193.3837890625, 16.20379638671875, 39.910369873046875, 0.3279132843017578, 132.32733154296875, -9.40753173828125, 34.255523681640625, 81.97590637207031, 168.66539001464844, -33.3587646484375, 131.4725341796875, -7.39129638671875, 24.48895263671875, 109.94281005859375, 108.59589385986328, 141.10781860351562, 106.43902587890625, 11.029075622558594, -20.336082458496094, 73.3150634765625, -82.78192138671875, 90.57681274414062, 193.67733764648438, 158.83294677734375, 31.873291015625, 233.10137939453125, 83.6275634765625, -66.82763671875, -234.84805297851562, 318.43914794921875, 170.77691650390625, -116.18890380859375, 50.51588439941406, -5.4346923828125, 307.35260009765625, 8.285064697265625, -4.06378173828125, 56.57354736328125, 109.89366149902344, 174.81829833984375, 123.81289672851562, 7.6638031005859375, 2.1240997314453125, -88.3739013671875, 0.77587890625, -149.87095642089844, 175.92181396484375, -106.556396484375, 142.59463500976562, 31.8184814453125, -193.25503540039062, -58.30529022216797, 218.8997802734375, 71.020751953125, 95.16384887695312, 185.37396240234375, 175.91656494140625, 172.85333251953125, 10.14849853515625, 34.28187561035156, 173.06936645507812, -61.709503173828125, -70.5289306640625, -0.5255279541015625, 2.7010726928710938, 197.56561279296875, -6.03021240234375, 277.01287841796875, -8.105743408203125, 64.3041763305664, 138.0296630859375, -64.1256103515625, 261.53271484375, -34.50274658203125, 38.78919219970703, -207.7581787109375, 188.18936157226562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000221.npy"}
{"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 60.34489822387695, "std": 108.09507751464844, "min": -273.44818115234375, "p10": -70.1607177734375, "median": 54.955726623535156, "p90": 203.75081787109374, "max": 332.221435546875, "pos_frac": 0.703125, "sample": [90.31777954101562, -17.477508544921875, 310.4276428222656, 116.27963256835938, -4.9486083984375, -39.3433837890625, 44.170318603515625, 104.80281066894531, 66.20916748046875, 49.855472564697266, 64.25711059570312, 332.221435546875, 282.56976318359375, 32.62208557128906, 57.6064453125, -85.03759002685547, 87.07024383544922, 202.57403564453125, 82.59588623046875, -20.6990966796875, 49.62095642089844, 14.006050109863281, -91.73724365234375, -21.7308349609375, 56.61883544921875, 240.21487426757812, 99.22970581054688, -69.00927734375, -158.85206604003906, -54.2069091796875, -63.88301086425781, 40.067535400390625, 262.73358154296875, -96.90765380859375, 9.03271484375, -33.84393310546875, -24.6341552734375, -0.89617919921875, 228.11326599121094, 284.51727294921875, -32.04402160644531, -134.82757568359375, 48.935302734375, 7.709175109863281, 254.19586181640625, 25.13775062561035, -64.75228881835938, 8.765396118164062, -273.44818115234375, 94.12054443359375, 30.20281982421875, 121.7799072265625, 4.711040496826172, 215.1432342529297, -57.788089752197266, 164.42633056640625, 77.363525390625, 173.97283935546875, 1.0273704528808594, 10.3875732421875, 137.38232421875, 74.88082885742188, 97.30728149414062, 129.082275390625, 41.027923583984375, 7.85015869140625, 138.87374877929688, 135.57333374023438, 113.95306396484375, 53.29261779785156, 163.0580291748047, 171.3044891357422, -72.847412109375, 137.04388427734375, -61.543365478515625, 140.7959442138672, 152.33651733398438, -74.5590591430664, 166.74082946777344, -37.02880859375, -22.36444854736328, 123.60252380371094, -0.49049949645996094, 112.67428588867188, -131.2378692626953, 0.17584228515625, 114.86160278320312, -47.384124755859375, -3.9598541259765625, 171.03012084960938, 176.0572052001953, 220.83074951171875, 132.38870239257812, 99.7079849243164, 13.607877731323242, -74.68177795410156, 51.17566680908203, 89.50010681152344, 75.25360107421875, 92.42388916015625, 191.3792724609375, 74.4097900390625, 83.93682861328125, 60.49664306640625, 206.49664306640625, -28.92542266845703, 38.1846923828125, 58.61082458496094, -44.208038330078125, 153.01473999023438, 52.614532470703125, 89.56195068359375, -123.56759643554688, 266.107666015625, -27.44072723388672, 45.80607604980469, -114.98190307617188, 255.1212158203125, -127.49124145507812, 125.7528076171875, 73.44696044921875, -16.07452392578125, 152.02056884765625, 196.13214111328125, 36.5557861328125, -10.594383239746094, 146.17941284179688, 4.360416412353516], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000222.npy"}
{"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 65.2840805053711, "std": 114.50830078125, "min": -229.5518035888672, "p10": -75.13264236450193, "median": 65.80786895751953, "p90": 209.67567901611326, "max": 386.8638610839844, "pos_frac": 0.75, "sample": [22.877906799316406, -16.507110595703125, 30.293472290039062, 97.48623657226562, 81.25631713867188, -85.51423645019531, -54.88506317138672, 36.81788635253906, 99.03960418701172, 300.2040710449219, 53.941070556640625, 269.35955810546875, 40.41926574707031, -26.7738037109375, 100.16476440429688, 96.9825439453125, 240.27294921875, -134.3919677734375, 99.18824768066406, -214.3714599609375, 161.60479736328125, 199.23614501953125, 32.332977294921875, 69.4710693359375, 49.64723205566406, -229.5518035888672, -12.752639770507812, 106.18243408203125, 84.89483642578125, -59.37176513671875, 33.534088134765625, 23.96112060546875, 157.16036987304688, 45.766143798828125, 208.30325317382812, 154.91998291015625, 194.30645751953125, -180.82510375976562, 128.53697204589844, 29.505355834960938, 15.493095397949219, 67.01092529296875, 62.53497314453125, 42.20348358154297, 224.54173278808594, 60.826904296875, 147.77317810058594, 0.186248779296875, 226.92660522460938, 197.59750366210938, 38.954254150390625, -20.68328857421875, 179.51397705078125, 297.97454833984375, 92.11760711669922, 161.64483642578125, -15.19219970703125, 212.8780059814453, 150.36373901367188, 115.48779296875, 56.57354736328125, 91.53312683105469, 99.536865234375, 168.0775146484375, 0.7175674438476562, 120.36177062988281, 3.59283447265625, 117.01202392578125, 168.66641235351562, 2.49859619140625, 100.06130981445312, 106.20944213867188, -22.364471435546875, 185.35513305664062, -120.00049591064453, -26.819623947143555, -25.41504669189453, -93.35171508789062, 127.515625, 108.03010559082031, 102.38108825683594, -6.592620849609375, 1.1640396118164062, 88.10279846191406, 247.00927734375, 70.35681915283203, 9.08514404296875, -70.68338775634766, 110.93453979492188, -16.297504425048828, -53.91069030761719, 76.10889434814453, 37.14271545410156, 251.94921875, 264.8626708984375, 386.8638610839844, 316.1377258300781, 64.60481262207031, 6.2541046142578125, 57.8663330078125, 10.378866195678711, 108.51197814941406, 110.34466552734375, -21.31623649597168, 173.85296630859375, -11.0328369140625, 132.1661376953125, 153.29373168945312, -4.567131042480469, -89.45661926269531, 231.15493774414062, -119.48027801513672, -2.560791015625, 43.75215148925781, 207.05047607421875, 74.139404296875, -129.5381622314453, -24.463882446289062, 114.01866149902344, 13.1500244140625, -146.7919921875, 116.97511291503906, 107.56500244140625, 20.729949951171875, 77.41246032714844, 38.11792755126953, -209.25506591796875, -219.79598999023438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000223.npy"}
{"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 71.7781982421875, "std": 111.3569564819336, "min": -162.06358337402344, "p10": -65.7580337524414, "median": 58.45191955566406, "p90": 219.51038818359373, "max": 378.789794921875, "pos_frac": 0.75, "sample": [216.59060668945312, 259.17474365234375, 26.9383544921875, 28.707901000976562, 32.819610595703125, -0.7818183898925781, -99.09457397460938, 120.37612915039062, 26.859130859375, 173.56533813476562, 207.88919067382812, 159.4536590576172, -26.65478515625, 40.58985900878906, 188.22412109375, 83.6578369140625, -90.07708740234375, 207.77001953125, 40.201629638671875, -63.7547607421875, 127.20303344726562, -88.20649719238281, 113.69294738769531, 236.03253173828125, 61.044281005859375, 24.591495513916016, 111.06565856933594, 110.79415893554688, 65.53094482421875, 68.86538696289062, 230.23294067382812, 82.284423828125, -32.6529541015625, 100.67507934570312, 215.8460693359375, -20.32720947265625, 70.34909057617188, 50.208675384521484, 180.4967041015625, 27.5943603515625, 6.9224395751953125, -46.512908935546875, 15.711967468261719, -22.759326934814453, 141.64120483398438, 112.78953552246094, 128.83001708984375, -154.76968383789062, 119.01202392578125, -25.291776657104492, 158.03887939453125, 76.41122436523438, 0.0, 213.69900512695312, 202.67115783691406, 16.19029426574707, 43.197906494140625, -147.94195556640625, 2.8514404296875, 55.051788330078125, 136.5316619873047, 22.27911376953125, 65.30671691894531, 305.1434326171875, -10.213134765625, 201.203125, -56.278350830078125, -103.17822265625, 23.682220458984375, -32.01295471191406, 300.911376953125, 317.478271484375, 8.684112548828125, 154.79278564453125, 55.85955810546875, -104.9927978515625, -26.162139892578125, 284.5138244628906, 69.0428466796875, 172.54559326171875, -6.6072540283203125, -162.06358337402344, -72.70944213867188, -11.887405395507812, 118.04716491699219, 233.524169921875, 90.48529052734375, -68.71078491210938, 317.02960205078125, -153.4763946533203, 181.15133666992188, 54.546112060546875, 71.02642822265625, 226.32321166992188, 45.7943115234375, 81.81796264648438, 116.2344970703125, 117.61224365234375, 94.48974609375, 31.7152099609375, -102.49667358398438, 109.0354995727539, -24.26068115234375, 109.9298095703125, 36.489646911621094, 3.3144378662109375, 282.2060852050781, -139.13565063476562, 378.789794921875, 82.81314086914062, 33.888946533203125, -64.49256896972656, -5.806459426879883, 66.35748291015625, -15.55694580078125, 47.66461181640625, 160.70748901367188, 14.132728576660156, 116.893310546875, 53.286865234375, 50.924072265625, 319.93865966796875, 36.457763671875, 19.94171142578125, 86.88165283203125, 136.758056640625, 33.877227783203125, 106.07080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000224.npy"}
{"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 65.46749877929688, "std": 123.85049438476562, "min": -265.9866943359375, "p10": -66.57723999023438, "median": 49.06171798706055, "p90": 239.07757263183592, "max": 460.11090087890625, "pos_frac": 0.6796875, "sample": [460.11090087890625, -35.0096435546875, 55.90997314453125, 204.79296875, 192.0399169921875, 31.141387939453125, -125.069580078125, -81.89944458007812, -73.9947509765625, 196.2509765625, 16.11077880859375, 238.5968017578125, 37.049835205078125, -48.36456298828125, 240.19937133789062, 72.3485107421875, 254.56655883789062, 157.26687622070312, 32.476531982421875, 172.26675415039062, -5.87451171875, -8.82073974609375, 15.724845886230469, -41.04467010498047, 124.25094604492188, -15.8909912109375, 248.6033935546875, 33.11225891113281, 79.4866943359375, 2.17352294921875, -48.69795227050781, -166.242431640625, 176.23214721679688, 86.32373046875, 22.773345947265625, 35.433876037597656, -0.18218994140625, -70.55337524414062, -5.24658203125, -40.258514404296875, 94.00057983398438, 317.01922607421875, 34.889801025390625, -130.00106811523438, -223.15512084960938, 94.91677856445312, 169.1637420654297, 34.57745361328125, 123.79723358154297, 93.8521728515625, -28.7435302734375, 64.196044921875, 40.825439453125, 87.10992431640625, -0.980194091796875, 192.4881591796875, -115.50909423828125, -67.8074951171875, 189.15988159179688, 158.3507537841797, 106.44378662109375, 331.6634521484375, 333.346923828125, 11.44171142578125, 127.89544677734375, -1.3509674072265625, -71.60943603515625, 123.97589111328125, 59.634925842285156, 200.3853759765625, 47.5223388671875, 142.900146484375, 124.5006103515625, 201.74082946777344, -6.9744873046875, 89.262451171875, 136.62680053710938, 288.1755065917969, -31.677734375, 115.6747055053711, -28.9381103515625, -239.09918212890625, -60.10661315917969, 35.23769760131836, 76.26420593261719, 22.534637451171875, 142.78768920898438, -3.9481658935546875, 67.92147827148438, -265.9866943359375, 2.84332275390625, -66.04998779296875, -23.1278076171875, 111.347412109375, 85.90383911132812, -15.07952880859375, 22.929428100585938, 66.95082092285156, 241.6302490234375, 65.42666625976562, 34.5838737487793, -18.885543823242188, 89.03155517578125, 361.11639404296875, -34.701873779296875, 144.36544799804688, 132.247314453125, -32.414642333984375, 18.370330810546875, 215.76605224609375, -56.85319519042969, 50.601097106933594, 158.9371337890625, 321.82373046875, 8.36419677734375, 88.61904907226562, 72.07110595703125, -47.146270751953125, 132.2572784423828, -10.6903076171875, 33.965362548828125, -40.347259521484375, 13.936920166015625, 81.617431640625, -103.58047485351562, 51.62992858886719, 299.4847106933594, 300.4071044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000225.npy"}
{"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 57.41851806640625, "std": 115.23440551757812, "min": -295.1125793457031, "p10": -58.622860717773406, "median": 49.02851104736328, "p90": 209.08018951416014, "max": 431.0688781738281, "pos_frac": 0.6640625, "sample": [32.06658935546875, 244.353759765625, -30.712646484375, 117.64788818359375, 161.3343505859375, 334.1456298828125, -8.243026733398438, -118.8323974609375, 53.51881408691406, 121.03935241699219, 140.22459411621094, 165.16371154785156, 11.5101318359375, 250.2902069091797, -26.82232666015625, 55.846187591552734, -39.220306396484375, 38.47802734375, 166.873046875, 88.88449096679688, 28.124435424804688, 20.187049865722656, 149.7682342529297, 87.99275970458984, 207.10337829589844, 57.8731689453125, -25.137962341308594, -295.1125793457031, 190.3103485107422, 149.56610107421875, -3.6220016479492188, 195.47531127929688, 57.592041015625, 193.83758544921875, 0.02850341796875, 91.15704345703125, 69.25641632080078, -199.95013427734375, 22.84893798828125, 116.20513916015625, -49.95269775390625, -1.8531570434570312, -0.9132461547851562, 0.18111038208007812, -4.33990478515625, 201.12387084960938, -14.923843383789062, -49.841957092285156, -171.5, 174.2078857421875, 227.4156036376953, -92.75128173828125, 221.28997802734375, 93.76872253417969, 33.55208206176758, 31.33380126953125, 44.85015869140625, 54.37396240234375, 13.513147354125977, -10.7796630859375, -4.7808380126953125, 130.44415283203125, 74.73564147949219, -149.6246337890625, 145.0427703857422, -15.466510772705078, 73.26514434814453, 431.0688781738281, -8.427419662475586, 57.390228271484375, -10.173736572265625, 19.422889709472656, 118.56303405761719, 58.43780517578125, 12.779850006103516, -36.408172607421875, 49.37968444824219, 187.61798095703125, 11.800331115722656, 214.44183349609375, 77.86354064941406, 55.07878112792969, 129.2132568359375, -42.53972625732422, 213.6927490234375, 49.88372802734375, 7.883544921875, 120.67483520507812, 82.11761474609375, -9.407623291015625, 125.822265625, 84.73211669921875, 201.93527221679688, -26.53509521484375, 214.11883544921875, -5.0142364501953125, -26.14111328125, 15.76068115234375, 64.32463073730469, 42.45701599121094, 143.7200927734375, 62.56158447265625, 214.04714965820312, -24.880706787109375, 1.317138671875, 48.677337646484375, 364.4462585449219, 97.4521484375, -33.400543212890625, -78.85324096679688, -141.5727996826172, -39.283172607421875, -3.4168930053710938, 115.14784240722656, 227.26068115234375, -29.652381896972656, 88.99533081054688, -38.582611083984375, -94.5189208984375, 349.9217529296875, 43.0777587890625, -161.1954345703125, -103.40426635742188, 104.34789276123047, -86.51406860351562, -101.82902526855469, 158.24264526367188, -37.775360107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000226.npy"}
{"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 75.01754760742188, "std": 118.51415252685547, "min": -207.09722900390625, "p10": -66.23000946044921, "median": 79.98970031738281, "p90": 206.0786682128906, "max": 390.6784973144531, "pos_frac": 0.703125, "sample": [-62.21263122558594, 247.6376190185547, 174.34048461914062, -42.15614318847656, 106.38566589355469, 17.09683609008789, 274.0479431152344, 152.4119873046875, -39.24823760986328, -71.97468566894531, 200.90725708007812, 123.88323974609375, 51.312530517578125, -3.487640380859375, -36.359344482421875, 136.1672821044922, 0.0, 190.05145263671875, -195.42529296875, -198.5432586669922, 161.12875366210938, -136.9356689453125, 158.2466583251953, 171.12530517578125, 218.63247680664062, 213.8209228515625, 164.274658203125, -150.9659423828125, 139.87353515625, -6.305419921875, -50.40870666503906, 156.74851989746094, 3.85650634765625, 108.56881713867188, 53.52691650390625, -47.89703369140625, 7.2396392822265625, 17.517601013183594, 33.47100830078125, 225.21713256835938, 188.1702880859375, 349.4599304199219, 63.68585205078125, 164.727783203125, 214.98388671875, -79.1558609008789, 145.89813232421875, 77.24508666992188, 4.72467041015625, 88.1942138671875, 259.0641784667969, 105.73387145996094, -18.21392822265625, -46.12860107421875, 104.556640625, 202.76055908203125, 54.9281005859375, 154.30010986328125, 135.42584228515625, 30.568912506103516, 277.7659912109375, -134.3369140625, 70.60379028320312, 163.43023681640625, -207.09722900390625, 115.69586181640625, 18.411914825439453, 165.96728515625, -22.771942138671875, 175.51727294921875, 78.28829956054688, 61.34602355957031, 132.8907470703125, -63.76800537109375, 350.61016845703125, 82.814453125, 144.864013671875, 72.5152587890625, -176.07763671875, 158.17555236816406, 390.6784973144531, 132.7131805419922, 172.18527221679688, 281.8863525390625, -17.5521240234375, 118.52365112304688, 71.97196960449219, 84.32888793945312, 69.76335144042969, -53.60772705078125, -82.320556640625, 145.33541870117188, -13.52239990234375, -119.45919799804688, 88.17068481445312, 191.12548828125, 160.1383514404297, 120.41220092773438, -10.5120849609375, 157.5621337890625, 46.70013427734375, 199.958251953125, -24.98319435119629, 92.72386169433594, 55.24822998046875, 178.2572021484375, -10.002252578735352, 2.09808349609375, 58.137245178222656, -3.798370361328125, 126.734619140625, -14.444076538085938, 226.83895874023438, 152.9254150390625, 55.3038330078125, 142.83900451660156, -18.58148193359375, 201.13442993164062, 167.0924072265625, -10.15447998046875, -201.07681274414062, -98.14779663085938, 81.69110107421875, 117.79193878173828, -13.52069091796875, -0.7547645568847656, 55.56504821777344, 17.5032958984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000227.npy"}
{"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 72.33233642578125, "std": 106.8985595703125, "min": -187.6973876953125, "p10": -51.90662155151367, "median": 59.115394592285156, "p90": 216.16878662109372, "max": 376.4518127441406, "pos_frac": 0.7265625, "sample": [32.26927185058594, -25.246307373046875, 193.992919921875, 117.43280029296875, -7.7602081298828125, 3.8492488861083984, 224.45770263671875, 54.496612548828125, 125.84454345703125, 145.1851043701172, 30.643810272216797, -51.796897888183594, -107.98641967773438, 146.03146362304688, 48.898773193359375, 35.558807373046875, -45.5242919921875, 254.29031372070312, -52.16264343261719, 171.42208862304688, 149.95257568359375, 44.669124603271484, -126.93414306640625, 15.060012817382812, 99.5802001953125, 83.87818908691406, 13.252243041992188, 26.819061279296875, -74.4700927734375, 61.65308380126953, 50.76287841796875, 26.691650390625, -31.017189025878906, 66.03564453125, 158.34939575195312, 183.3081817626953, 55.99591064453125, 78.97419738769531, 42.31329345703125, 11.585418701171875, 171.74539184570312, 17.875274658203125, -79.83914184570312, -61.2137451171875, 181.39398193359375, -20.799224853515625, -124.0870361328125, 72.49130249023438, 134.52615356445312, 135.33078002929688, 59.87384033203125, 223.03240966796875, 126.00938415527344, -5.57014274597168, 243.29315185546875, 173.71385192871094, 117.1895751953125, 73.5692138671875, 122.55078125, 376.4518127441406, -11.4744873046875, -30.81383514404297, -78.50448608398438, -5.658805847167969, 71.9190673828125, 6.382568359375, 0.0, 122.37283325195312, 60.08148193359375, -84.60762023925781, 53.504127502441406, 185.05343627929688, 46.29011535644531, 33.27593994140625, -4.552032470703125, 205.23495483398438, -104.60791015625, 82.69622802734375, -1.897735595703125, 176.47955322265625, -46.90460205078125, -50.295555114746094, 50.3873291015625, 74.30535888671875, 272.03680419921875, 315.7882080078125, -57.253692626953125, 82.5224609375, 264.4984436035156, 69.76242065429688, 63.3519287109375, 180.24005126953125, 58.35694885253906, 103.82180786132812, 164.35691833496094, -22.763275146484375, 37.076934814453125, 25.00739288330078, 44.99613952636719, 355.8285827636719, 31.135833740234375, -19.146575927734375, -24.690322875976562, 89.65509796142578, 75.5238265991211, 326.3681335449219, 123.98043823242188, -187.6973876953125, -0.04827880859375, -40.883514404296875, 5.2210540771484375, 268.59478759765625, 30.94024658203125, 81.24126434326172, 202.62767028808594, 244.80709838867188, 245.3592529296875, -0.054904937744140625, 4.480339050292969, -81.04592895507812, 63.0081787109375, 161.60443115234375, 175.07086181640625, 150.93272399902344, 213.22723388671875, 181.73703002929688, -4.329620361328125, 66.7315673828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000228.npy"}
{"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 84.5587158203125, "std": 114.77399444580078, "min": -198.33749389648438, "p10": -53.09383239746093, "median": 77.10332870483398, "p90": 238.28234252929684, "max": 348.5074768066406, "pos_frac": 0.7578125, "sample": [170.2325439453125, -12.4669189453125, 186.45816040039062, 150.4122314453125, -23.93988037109375, -149.72756958007812, 27.055099487304688, 88.02069091796875, 37.28057861328125, -142.98410034179688, 284.78045654296875, 154.4954833984375, 129.52740478515625, -66.51165771484375, 263.9236145019531, 243.29150390625, 88.21273803710938, -7.69439697265625, 4.685935974121094, 53.8104248046875, 187.15438842773438, 77.7570571899414, 321.9186706542969, 121.89083862304688, 31.247955322265625, -7.8648681640625, 33.398529052734375, 75.57058715820312, 145.08753967285156, 110.88298797607422, 9.387100219726562, 257.54644775390625, 236.13555908203125, 171.54205322265625, 63.792213439941406, 169.49053955078125, 130.0423583984375, -0.9498291015625, 234.18075561523438, -198.33749389648438, -5.8892822265625, -143.67123413085938, 76.44960021972656, 120.01524353027344, 19.911651611328125, 324.3890075683594, 229.97662353515625, 15.673553466796875, 108.17042541503906, 144.95802307128906, 183.05838012695312, -1.2114830017089844, 54.423500061035156, -9.443359375, 68.18929290771484, 140.47972106933594, 196.8243408203125, 8.659332275390625, 34.026336669921875, -48.26190185546875, -0.9456920623779297, -133.16168212890625, 213.68948364257812, 104.86767578125, -6.177894592285156, -109.7845230102539, 164.0230712890625, 68.4962158203125, 157.43556213378906, 49.6781005859375, 3.0669403076171875, -19.179649353027344, -8.024269104003906, -105.86335754394531, 137.6871337890625, 137.96685791015625, -10.277252197265625, 93.67616271972656, 115.24578857421875, 88.66180419921875, 325.22479248046875, 175.82769775390625, 168.0013427734375, 288.2041320800781, 19.067371368408203, 22.6400146484375, -182.14923095703125, 124.74119567871094, 254.86505126953125, 21.879615783691406, 211.32659912109375, 105.0586929321289, 14.550100326538086, 27.92742919921875, 21.20318603515625, 179.7677001953125, 95.02096557617188, 250.46136474609375, 43.748199462890625, 90.38789367675781, 69.4410400390625, 69.98139953613281, 222.89437866210938, 191.02896118164062, -51.510894775390625, 117.88119506835938, -95.69352722167969, -59.99981689453125, 75.98747253417969, 313.6435546875, 67.471923828125, -89.93388366699219, 105.28985595703125, 110.82757568359375, 126.51837158203125, 154.11370849609375, 200.61248779296875, -19.589569091796875, 55.40666198730469, -56.787353515625, 348.5074768066406, 29.19097900390625, 40.23370361328125, 146.1764678955078, 99.518310546875, -20.85548973083496, 311.23040771484375, -2.3679122924804688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000229.npy"}
{"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 76.95330810546875, "std": 125.35077667236328, "min": -356.10064697265625, "p10": -47.70450172424316, "median": 62.357330322265625, "p90": 219.52353363037108, "max": 581.268798828125, "pos_frac": 0.734375, "sample": [232.61260986328125, 63.501678466796875, 152.4703369140625, 273.776123046875, 217.4085693359375, 107.95419311523438, 141.55743408203125, -12.85205078125, 165.91650390625, 43.38153076171875, 70.95094299316406, 27.295013427734375, 50.64048767089844, -44.7958984375, -356.10064697265625, 266.0995788574219, 225.05947875976562, 0.13585281372070312, 10.960479736328125, 85.38421630859375, -2.826751708984375, 40.93358612060547, 277.381591796875, -75.89608764648438, 0.0, 217.80996704101562, -28.72265625, 42.79357147216797, -46.832916259765625, 37.40399169921875, 170.53640747070312, 218.3933563232422, -14.276290893554688, 161.1075897216797, 5.990634918212891, -29.673126220703125, 158.48406982421875, 204.43804931640625, 96.6378173828125, 105.76203918457031, 7.272195816040039, -71.7578125, 36.749420166015625, 65.52816772460938, 84.19886779785156, 138.04144287109375, 11.523811340332031, 45.681884765625, 165.55496215820312, -135.30514526367188, -9.72780990600586, 61.212982177734375, -150.7322998046875, -3.8348388671875, 166.7734375, -58.28154754638672, 144.44894409179688, -21.708648681640625, 198.55938720703125, 77.5869140625, -49.73820114135742, 98.20320129394531, 179.52035522460938, 116.85516357421875, -33.655029296875, 9.70318603515625, -38.80323791503906, 184.16452026367188, 3.6445159912109375, 223.66294860839844, 132.4534912109375, 26.929901123046875, -31.286094665527344, -65.52510833740234, 132.42425537109375, 8.83734130859375, -2.26605224609375, 19.00735855102539, 183.56494140625, 191.41458129882812, 82.37220764160156, 152.5745849609375, 155.34913635253906, 99.6210708618164, 64.86053466796875, -221.32647705078125, -0.3962860107421875, -78.07183837890625, 222.16061401367188, 106.495361328125, 267.72784423828125, 77.76596069335938, -54.647552490234375, 31.74468994140625, 147.59088134765625, 129.61831665039062, -0.7884330749511719, 26.196044921875, 59.72808074951172, 558.883056640625, 139.5610809326172, 2.4096508026123047, 111.35279846191406, 41.85936737060547, 41.55493927001953, 105.79673767089844, -18.22705078125, 581.268798828125, 89.793212890625, 28.066465377807617, 135.7098388671875, 304.34002685546875, 320.0513916015625, -31.178295135498047, 106.86703491210938, 93.29989624023438, 6.597890853881836, 210.987548828125, 336.86639404296875, -6.599388122558594, -56.629600524902344, 93.2467041015625, 2.278106689453125, 53.15557861328125, 39.10809326171875, 108.73733520507812, -45.07647705078125, -74.33050537109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000230.npy"}
{"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 61.55538558959961, "std": 108.24125671386719, "min": -241.5523681640625, "p10": -67.20279388427731, "median": 61.647010803222656, "p90": 200.24409790039059, "max": 379.44927978515625, "pos_frac": 0.7265625, "sample": [27.587417602539062, 73.80413818359375, -34.98114013671875, -102.28569793701172, 134.45587158203125, 170.8797149658203, 89.03253173828125, -35.16685485839844, -12.72247314453125, -129.11827087402344, 216.18145751953125, -57.14237976074219, 125.98321533203125, 101.84130859375, 12.238861083984375, 83.05455017089844, 13.056686401367188, 23.332672119140625, 14.478996276855469, 61.39442443847656, -153.40399169921875, -99.1927490234375, -27.403518676757812, 62.773101806640625, 59.62846374511719, 139.89910888671875, 21.086517333984375, 89.71168518066406, 38.354736328125, -91.53759765625, 213.18072509765625, 138.13430786132812, 22.70782470703125, 4.910564422607422, 44.99615478515625, 25.162391662597656, 379.44927978515625, 247.59832763671875, -58.955078125, 168.62353515625, 194.6998291015625, 149.1082763671875, 149.27438354492188, 25.77691650390625, 219.23410034179688, 94.67953491210938, -54.0545654296875, 116.18101501464844, -19.3427734375, 82.3641357421875, 105.55303955078125, 169.6036376953125, 99.65847778320312, -56.977142333984375, 117.36138916015625, 32.175323486328125, 186.55914306640625, -16.89556884765625, 54.19158172607422, 260.4324035644531, -155.58409118652344, 137.80784606933594, -11.24658203125, 10.853042602539062, -20.823726654052734, 52.673431396484375, 134.38739013671875, 22.332672119140625, 92.104736328125, 139.2146759033203, 153.17184448242188, 144.55245971679688, 222.00619506835938, -218.36483764648438, -241.5523681640625, 53.478729248046875, 16.845867156982422, -23.286773681640625, 135.1641845703125, 49.0384521484375, 33.31549072265625, 120.7833251953125, 69.74613952636719, -16.460693359375, -44.37202072143555, 28.823974609375, 13.39007568359375, 152.79953002929688, 61.89959716796875, 162.08877563476562, 63.217376708984375, 82.18746948242188, -15.029647827148438, 147.66038513183594, -23.66998291015625, 185.5642547607422, 74.60421752929688, 37.929290771484375, 317.2984619140625, -0.08087730407714844, -152.61407470703125, 70.3065185546875, 79.44056701660156, 215.16427612304688, 15.715713500976562, -28.27239990234375, 87.28608703613281, 224.79507446289062, 91.37332153320312, 41.95726013183594, -86.44746398925781, 15.184019088745117, 141.12176513671875, -86.98727416992188, 108.71463012695312, 233.10009765625, 126.22711181640625, 105.96774291992188, -22.343551635742188, 106.17080688476562, -129.26779174804688, 120.09732055664062, 278.2801818847656, -99.31103515625, 314.2308654785156, 0.0, -22.025482177734375, 75.544677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000231.npy"}
{"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 61.118553161621094, "std": 125.54875946044922, "min": -277.18695068359375, "p10": -67.7531829833984, "median": 45.980018615722656, "p90": 209.8050598144531, "max": 397.7674560546875, "pos_frac": 0.6953125, "sample": [-277.18695068359375, 100.94052124023438, -5.0892181396484375, 134.58697509765625, 66.40357971191406, 110.41195678710938, 59.12274169921875, 224.78579711914062, -44.22747802734375, 75.94198608398438, 122.82569885253906, 269.1572265625, 20.09930419921875, -49.25154113769531, -27.01715087890625, 31.024696350097656, -12.135345458984375, -100.5195541381836, 18.06439208984375, 90.98555755615234, 71.86962890625, 45.55156707763672, -238.68389892578125, 28.895965576171875, 316.51397705078125, 311.59185791015625, 113.40629577636719, 64.30450439453125, -43.539737701416016, 284.53369140625, 30.57391357421875, 28.19562530517578, -95.58236694335938, 110.8818359375, 397.7674560546875, 177.93353271484375, -54.5777587890625, 42.418548583984375, 134.0745849609375, 314.96575927734375, -36.402191162109375, -32.94831848144531, 3.319070816040039, -51.13543701171875, 0.8817901611328125, 15.370376586914062, 52.44873809814453, 71.6787109375, 169.20896911621094, 4.860713958740234, -25.09637451171875, -56.753387451171875, 183.18194580078125, 9.809967041015625, 65.79508209228516, 154.6866455078125, 111.27008056640625, 214.91329956054688, 20.117355346679688, 141.36793518066406, 46.408470153808594, 325.60980224609375, 198.7347412109375, -14.32977294921875, 110.76866149902344, 335.39605712890625, -5.38238525390625, 53.4166259765625, -9.70635986328125, 61.281646728515625, 94.49665832519531, 186.82086181640625, -0.4310302734375, -253.54061889648438, 251.7818603515625, 183.909423828125, 3.9926300048828125, 39.271240234375, 10.246185302734375, 83.5322265625, 8.440099716186523, -14.9412841796875, -51.2147216796875, 24.397918701171875, 108.57330322265625, -4.127998352050781, 26.74738121032715, 200.51583862304688, -23.0689697265625, -55.76068115234375, 183.75830078125, 17.481353759765625, 0.0, 119.8409423828125, 39.258628845214844, -12.433135986328125, 51.97901916503906, 119.91170501708984, 43.855194091796875, 114.33367919921875, 31.45191192626953, 199.931640625, 179.1622314453125, -93.41937255859375, 250.27496337890625, 203.70709228515625, 50.1943359375, -198.54653930664062, 207.61581420898438, 82.81109619140625, 16.853851318359375, 179.39199829101562, -142.16867065429688, -140.32827758789062, -23.200775146484375, -169.16082763671875, 279.2033386230469, -96.32976531982422, 87.31887817382812, -125.25494384765625, 183.579345703125, 122.22650146484375, -225.0527801513672, 199.29742431640625, 196.06439208984375, 120.75843811035156, -3.797210693359375, -15.825927734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000232.npy"}
{"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 77.69261169433594, "std": 116.60076141357422, "min": -268.95159912109375, "p10": -38.48995742797852, "median": 61.1346435546875, "p90": 237.01448974609374, "max": 490.62109375, "pos_frac": 0.75, "sample": [-122.34827423095703, 109.45891571044922, 49.759674072265625, 6.0008544921875, -24.670822143554688, 159.81640625, -10.72601318359375, 63.97705078125, 0.7865123748779297, -25.100265502929688, -15.7969970703125, 242.7840118408203, 200.32443237304688, 18.12060546875, -21.014801025390625, 22.416259765625, 93.63174438476562, 490.62109375, 183.1792449951172, 209.80563354492188, 303.7986145019531, 179.02490234375, 194.30364990234375, 159.11294555664062, -17.77472686767578, 93.92878723144531, 162.41134643554688, 138.66064453125, -4.013214111328125, -138.18753051757812, 122.99688720703125, 1.747314453125, 254.03579711914062, -13.020614624023438, 130.40719604492188, 76.873291015625, 17.37835693359375, -16.57628631591797, 141.67822265625, 279.73626708984375, -3.48162841796875, -2.3114376068115234, 31.37435531616211, -3.49072265625, -15.33135986328125, 235.13201904296875, 95.03654479980469, 41.47305679321289, -46.97552490234375, 114.74992370605469, 370.50726318359375, 200.21282958984375, 154.825439453125, 25.99555206298828, 62.99668884277344, 2.4322509765625, 274.607666015625, 61.672607421875, 38.2777099609375, 12.7279052734375, 101.474365234375, 154.5714111328125, -125.92803955078125, 78.18484497070312, 135.45391845703125, -40.33856201171875, 3.02020263671875, 91.31095886230469, -102.59573364257812, -57.34442138671875, 40.566009521484375, 220.08380126953125, 144.1151123046875, 112.52169799804688, 367.1307373046875, 202.44100952148438, 54.19610595703125, 49.99830627441406, 19.47655487060547, -47.600982666015625, 57.05847930908203, -53.178131103515625, 0.18133544921875, 177.7420654296875, -55.770782470703125, 121.88455200195312, 246.25244140625, 126.43911743164062, 60.5966796875, 176.59121704101562, 44.550071716308594, 44.616973876953125, -260.6500244140625, 146.44076538085938, -20.609268188476562, -17.44415283203125, 31.29020881652832, 162.06871032714844, -38.12493896484375, 251.59564208984375, -5.4532470703125, -17.792205810546875, -39.34166717529297, -268.95159912109375, 269.2499694824219, 116.05068969726562, 57.16835021972656, 203.92898559570312, 69.812744140625, 29.44781494140625, 31.026077270507812, 78.83270263671875, 300.3876647949219, 43.312042236328125, 130.0555419921875, 134.9219512939453, 9.377574920654297, 155.8660888671875, 63.33271026611328, 79.72904968261719, 100.73794555664062, 2.737377166748047, -35.1820068359375, 115.589111328125, 241.40692138671875, 100.48260498046875, 8.657241821289062, 19.017547607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000233.npy"}
{"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 64.3183822631836, "std": 126.0421371459961, "min": -225.5801544189453, "p10": -114.68190841674804, "median": 54.35906219482422, "p90": 221.82178649902343, "max": 415.91217041015625, "pos_frac": 0.6796875, "sample": [-87.21481323242188, -132.89947509765625, -18.721099853515625, 100.65373229980469, 138.58596801757812, 53.84928894042969, 221.81591796875, 84.60281372070312, -99.44894409179688, 307.3401794433594, 170.18521118164062, -216.83193969726562, 212.35604858398438, 159.02627563476562, -165.37847900390625, 98.34823608398438, 160.1538848876953, -42.806182861328125, 30.538238525390625, 149.87734985351562, 112.47721862792969, 0.890289306640625, 109.98028564453125, 201.5911865234375, 45.7777099609375, -118.17019653320312, 205.5458984375, 141.255615234375, 30.066673278808594, 34.740478515625, 17.221267700195312, 68.62115478515625, 115.31365966796875, 211.5625, 222.14398193359375, -157.93862915039062, 54.86883544921875, 69.89193725585938, -86.65394592285156, -21.82952880859375, 147.536865234375, 3.679351806640625, -14.863677978515625, -122.93098449707031, 126.03257751464844, -19.748321533203125, -6.221527099609375, 35.76429748535156, 25.353004455566406, 41.94532012939453, 18.983665466308594, 37.104736328125, 187.40463256835938, 117.46382141113281, -5.9102325439453125, 267.36248779296875, 231.4196319580078, 117.9285659790039, 154.4293212890625, 23.54937744140625, -20.679567337036133, 40.39049530029297, 289.3670654296875, 68.53890991210938, -7.414430618286133, 53.464630126953125, -168.95758056640625, 0.0, 90.6512451171875, -58.50028991699219, 47.807212829589844, 138.1399383544922, -25.3865966796875, 28.043746948242188, 200.5440673828125, 258.0036315917969, 279.3478088378906, 221.83547973632812, -168.04563903808594, 117.36068725585938, -15.056106567382812, 227.2919464111328, 119.49386596679688, -39.009521484375, 207.99623107910156, 50.27439880371094, 161.97787475585938, 59.571044921875, -204.56375122070312, 179.49798583984375, -121.06341552734375, 348.2774353027344, 69.9854736328125, 120.30804443359375, 102.60567474365234, -113.18692779541016, 155.64544677734375, 116.40130615234375, -36.44171905517578, -1.593994140625, 159.55381774902344, 0.0, 38.339210510253906, -12.942977905273438, 415.91217041015625, 284.265625, -104.52978515625, 55.9818115234375, 205.816162109375, 126.3406982421875, 145.73028564453125, -61.31146240234375, -225.5801544189453, 35.65838623046875, 156.57666015625, 148.950439453125, -43.96916198730469, -51.384124755859375, -120.34317016601562, 33.98992919921875, 127.09280395507812, -36.40263366699219, -133.10498046875, 176.15377807617188, -61.0299072265625, 288.2073059082031, 16.226348876953125, 147.96240234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000234.npy"}
{"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 70.67632293701172, "std": 114.27713012695312, "min": -242.62701416015625, "p10": -57.80996322631836, "median": 60.358219146728516, "p90": 216.822834777832, "max": 357.01190185546875, "pos_frac": 0.703125, "sample": [34.88409423828125, 21.806983947753906, -26.3382568359375, 103.86820983886719, 13.48282241821289, -164.95706176757812, -22.041149139404297, -58.90021514892578, 174.0694122314453, 46.7650146484375, 27.06151580810547, 316.81744384765625, 216.176513671875, 81.4100341796875, -8.645263671875, 139.68621826171875, 176.99819946289062, 193.3299560546875, 9.233217239379883, -39.242095947265625, 125.41737365722656, 93.15084838867188, 209.77056884765625, 168.28665161132812, 225.4275360107422, 144.63409423828125, 124.44143676757812, -87.18048095703125, 118.99655151367188, -21.411712646484375, 150.0140380859375, -12.37237548828125, 186.49569702148438, 236.820068359375, 44.240570068359375, 218.33091735839844, 354.96319580078125, 212.95681762695312, 44.666873931884766, 119.8717041015625, -58.874778747558594, 137.36428833007812, -66.18099212646484, -56.3741455078125, 226.56362915039062, 133.900390625, -242.62701416015625, -14.574764251708984, -123.15802001953125, 62.297340393066406, 273.2016906738281, 68.88876342773438, 8.138704299926758, -57.574066162109375, 127.84521484375, 86.44039916992188, -25.083511352539062, 154.46287536621094, -153.2325439453125, 140.07315063476562, -10.618640899658203, 48.37274169921875, 169.61680603027344, -21.932998657226562, 260.98614501953125, 22.543514251708984, -6.1043701171875, -150.4950714111328, -14.329559326171875, 177.210205078125, 23.72614288330078, 122.4505615234375, -2.414703369140625, 42.23908996582031, -0.9879188537597656, 18.20001220703125, 5.31927490234375, 357.01190185546875, 39.8924560546875, 133.6004638671875, 170.6765899658203, 40.018707275390625, -180.91748046875, 17.37236785888672, 54.42289733886719, -7.6683502197265625, 274.9488830566406, 211.5537872314453, 70.69683837890625, 4.8724212646484375, 226.2225341796875, 44.88343811035156, 122.6329574584961, 231.04473876953125, 121.21917724609375, 131.68075561523438, 1.939697265625, 18.847015380859375, 161.76416015625, 73.265869140625, -16.84238624572754, 210.8638916015625, -8.651906967163086, -46.4002685546875, 198.1136474609375, 69.76225280761719, 12.351905822753906, 77.10366821289062, 108.09063720703125, -3.1234130859375, 58.419097900390625, -6.804840087890625, -28.52381134033203, -156.11077880859375, 76.79325866699219, 154.87673950195312, 166.66748046875, -173.307861328125, -4.632637023925781, -58.360389709472656, 67.37054443359375, 134.21263122558594, 45.986175537109375, 249.96612548828125, 208.00643920898438, 133.65237426757812, 81.85806274414062, -23.0135555267334], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000235.npy"}
{"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 70.19259643554688, "std": 118.95829772949219, "min": -213.59130859375, "p10": -51.96551818847656, "median": 49.57305908203125, "p90": 245.2720031738281, "max": 372.3309326171875, "pos_frac": 0.7265625, "sample": [134.56619262695312, 26.49041748046875, -106.92581176757812, 51.881614685058594, 47.814178466796875, 22.397369384765625, 29.626266479492188, 324.6613464355469, -176.54019165039062, -47.30372619628906, -104.450927734375, 115.81745147705078, -133.45263671875, 178.66470336914062, 96.34991455078125, 190.95098876953125, -4.949108123779297, 161.75120544433594, 63.935577392578125, 122.91305541992188, 63.850555419921875, 9.636810302734375, 72.6241455078125, 35.106689453125, -50.8302001953125, 292.0472717285156, 43.697174072265625, 13.829700469970703, 13.335479736328125, 243.67138671875, 78.9791259765625, 38.559112548828125, 81.12532806396484, 7.733268737792969, 17.666885375976562, 75.45489501953125, -211.21014404296875, 291.2134704589844, 346.49945068359375, -15.788467407226562, 175.0037384033203, 99.29206085205078, 181.26992797851562, 38.033203125, 118.62908935546875, 15.666004180908203, -117.10189819335938, -33.79876708984375, 25.68353271484375, 0.0, 5.8482666015625, 9.357467651367188, 137.6683349609375, 198.72901916503906, 17.57075309753418, -54.614593505859375, -213.59130859375, 225.84317016601562, 21.8465576171875, -17.729095458984375, 35.708404541015625, 188.20504760742188, 68.60076904296875, 306.3178405761719, -21.855484008789062, -39.951416015625, 119.9244384765625, -91.9058837890625, 48.90338134765625, 15.107780456542969, -71.81452941894531, 120.15069580078125, 63.59070587158203, 188.965576171875, 33.12468719482422, 26.902084350585938, -122.11935424804688, -23.3409423828125, 234.97439575195312, 7.529088973999023, 121.75413513183594, 101.203369140625, -36.41888427734375, 127.9395980834961, 142.95639038085938, 91.92724609375, 26.359542846679688, 150.0335693359375, -15.556373596191406, -6.8253631591796875, 95.8125, 138.42242431640625, 123.43264770507812, -8.273117065429688, 8.781280517578125, 227.298583984375, 168.80859375, 54.58502197265625, 164.46429443359375, 260.38983154296875, 179.50949096679688, 50.24273681640625, -39.219573974609375, 10.27410888671875, 0.0, 66.66207885742188, -48.931182861328125, 249.00677490234375, -168.252685546875, 285.93475341796875, 96.95083618164062, -26.318252563476562, -26.24292755126953, 69.45018005371094, 207.77191162109375, 259.39801025390625, 372.3309326171875, 241.88299560546875, -4.849346160888672, -67.01222229003906, 306.1585388183594, 89.71854400634766, 279.79119873046875, 19.79107666015625, -41.9619140625, 255.93585205078125, -2.8383331298828125, 70.35010528564453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000236.npy"}
{"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 70.90882873535156, "std": 106.53488159179688, "min": -250.77978515625, "p10": -41.858100128173824, "median": 52.526920318603516, "p90": 228.65061950683594, "max": 378.5289611816406, "pos_frac": 0.734375, "sample": [193.44390869140625, 25.088912963867188, 15.775138854980469, 249.6905517578125, -37.2967529296875, 121.61895751953125, 29.663772583007812, 205.9002685546875, 119.69403076171875, 105.1131591796875, -24.625335693359375, 27.94097900390625, 177.80267333984375, -41.27167510986328, -6.320098876953125, -32.113800048828125, 87.34750366210938, -43.22642517089844, 357.3663024902344, 61.09349060058594, 108.26051330566406, 181.3664093017578, 211.53945922851562, 157.3604736328125, 41.148258209228516, 40.095947265625, 123.19277954101562, 0.0, 115.88668823242188, 31.555084228515625, 52.33429718017578, 120.68836975097656, -92.11456298828125, 32.56883239746094, 6.5052490234375, 41.31341552734375, -26.345245361328125, 28.17352294921875, -113.80880737304688, -13.775833129882812, 31.975753784179688, -8.42535400390625, 53.212120056152344, 97.30453491210938, 48.9517822265625, -60.474212646484375, 250.60293579101562, 41.906982421875, 59.508056640625, -3.9222984313964844, 90.94157409667969, 261.2797546386719, 94.64654541015625, 29.4293212890625, 105.83212280273438, 204.79507446289062, 161.8851318359375, 199.68203735351562, 5.0857086181640625, 57.69511413574219, 255.23443603515625, 59.52637481689453, 14.814422607421875, 56.201324462890625, 22.398880004882812, 20.116119384765625, 146.11715698242188, 187.8413848876953, -4.98779296875, -78.0484619140625, -6.588039398193359, 248.95108032226562, -1.5592041015625, 124.526611328125, 246.81683349609375, 52.72346878051758, -14.01708984375, 17.39630126953125, 318.35028076171875, 66.3380126953125, 94.27676391601562, 52.71954345703125, 306.4141845703125, -48.8087158203125, 173.6468048095703, 26.762969970703125, -0.8549118041992188, -28.861480712890625, 378.5289611816406, 8.17950439453125, 31.381343841552734, 216.73690795898438, -122.0841064453125, 10.4075927734375, -107.6927490234375, 77.39715576171875, -2.5299072265625, 23.104278564453125, 135.09463500976562, 48.890167236328125, 11.11285400390625, 59.41771697998047, 161.04476928710938, -8.504596710205078, 47.86284255981445, 21.565628051757812, 103.9442138671875, -99.76515197753906, -72.57614135742188, -54.131805419921875, 60.642333984375, 295.3004150390625, -44.991119384765625, 59.22383117675781, -8.846549987792969, 106.66586303710938, -250.77978515625, 92.03096008300781, 228.30926513671875, 168.5985565185547, 114.11148071289062, -18.482391357421875, -23.326583862304688, 229.44711303710938, 99.15412139892578, 255.1332550048828, 68.62893676757812, 110.13629150390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000237.npy"}
{"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 69.5528793334961, "std": 114.8326416015625, "min": -140.55438232421875, "p10": -62.30773849487304, "median": 55.4642219543457, "p90": 213.41791381835935, "max": 442.77032470703125, "pos_frac": 0.71875, "sample": [76.29269409179688, 65.60542297363281, -13.037826538085938, 33.173641204833984, 77.03240966796875, 234.92166137695312, 146.9496612548828, 134.17202758789062, -13.772125244140625, -140.55438232421875, 7.5291748046875, 141.43173217773438, 13.093643188476562, 78.47015380859375, 63.74671936035156, -133.79554748535156, 244.78271484375, 269.38726806640625, -90.72924041748047, 143.453369140625, 200.61801147460938, 201.82122802734375, 101.27264404296875, 167.93707275390625, 249.519287109375, 119.85646057128906, -75.50689697265625, 66.82650756835938, 40.210052490234375, 10.931747436523438, -64.65226745605469, -49.81713104248047, 394.05938720703125, -71.47824096679688, 18.744659423828125, 117.3590087890625, 36.20751953125, 212.29425048828125, 9.393013000488281, 113.250244140625, 129.27783203125, 49.13507080078125, 23.998565673828125, 442.77032470703125, 28.808349609375, 207.67257690429688, 115.4169921875, 87.0594482421875, 76.0277099609375, 94.27383422851562, 398.0531005859375, 299.86651611328125, -52.8455810546875, 19.676239013671875, 95.794677734375, 28.48931884765625, 32.743499755859375, -79.26188659667969, 141.61544799804688, 147.91282653808594, 11.066497802734375, 40.46807861328125, -42.755462646484375, -65.4136962890625, 224.090576171875, 59.42291259765625, 39.23457336425781, 170.3486785888672, 146.67401123046875, 75.1550064086914, -128.59353637695312, -104.882568359375, -31.912338256835938, 63.097869873046875, -10.83407974243164, 91.04550170898438, 195.374267578125, -90.9068603515625, 74.29731750488281, 63.98114013671875, -30.06414794921875, 290.10247802734375, 5.763889312744141, -24.730148315429688, -18.2899169921875, 50.3680419921875, 79.95589447021484, 107.91134643554688, 8.252262115478516, 64.12615966796875, 15.130840301513672, -25.179168701171875, -40.068389892578125, 109.9652099609375, 292.3678894042969, -28.452667236328125, 69.16073608398438, 25.894296646118164, 51.505531311035156, -61.302940368652344, 70.70552062988281, 201.34796142578125, 180.00787353515625, 21.710693359375, -12.628902435302734, -68.1817626953125, -77.97825622558594, 216.039794921875, -48.800048828125, -22.5615234375, 41.62586975097656, 120.3614501953125, -32.97990036010742, 39.06874084472656, 64.45318603515625, -41.69871520996094, 117.48127746582031, 193.76654052734375, 1.4856948852539062, 1.47216796875, 167.9693603515625, 417.37884521484375, 84.1134033203125, -43.803924560546875, -24.122844696044922, -10.914794921875, 209.51937866210938, -6.92822265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000238.npy"}
{"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 62.747100830078125, "std": 136.8614044189453, "min": -502.43328857421875, "p10": -59.93000946044921, "median": 46.378562927246094, "p90": 231.67969360351563, "max": 406.32794189453125, "pos_frac": 0.6796875, "sample": [9.402076721191406, -109.43238830566406, 322.3412170410156, -45.842376708984375, -27.677490234375, 23.501495361328125, -284.0875244140625, 67.43358612060547, 179.1705780029297, -35.368507385253906, 108.12298583984375, -502.43328857421875, -44.046905517578125, -13.609039306640625, 224.02139282226562, 92.3232421875, 231.638916015625, 234.84609985351562, -4.0899658203125, -34.03192138671875, 27.65911865234375, 403.688720703125, 5.808570861816406, 124.2620849609375, -47.086822509765625, 162.80654907226562, 37.439964294433594, 119.18852233886719, 94.17987060546875, 45.39176940917969, -28.07818603515625, 231.77484130859375, 187.72711181640625, -28.440528869628906, 374.7723388671875, -59.244293212890625, 47.3653564453125, 206.6876220703125, 355.40606689453125, 268.656494140625, 140.0118865966797, -125.29608154296875, 132.88694763183594, -73.59687805175781, 210.59255981445312, -61.53001403808594, -24.50439453125, 96.08758544921875, -23.927459716796875, -29.186920166015625, -102.09150695800781, 166.92654418945312, 25.638473510742188, -138.39703369140625, 64.78128051757812, 52.925048828125, 52.8828125, 97.73939514160156, -3.7564773559570312, -173.58743286132812, 36.34263610839844, 162.89056396484375, 10.606287002563477, 113.74253845214844, 131.5333251953125, 40.68743896484375, 121.09869384765625, -36.322021484375, 7.20294189453125, 194.728515625, 0.8588180541992188, 8.299514770507812, 145.25759887695312, 107.0296630859375, 77.20819091796875, 0.9989013671875, 121.51498413085938, -17.220260620117188, 16.984024047851562, -182.29087829589844, -14.73858642578125, -22.51392364501953, 343.7532043457031, -57.70794677734375, 55.9000244140625, 397.90460205078125, -199.48690795898438, 49.31256866455078, 57.85047912597656, -48.5408935546875, 146.64962768554688, -1.3193435668945312, -78.57750701904297, 164.4700927734375, 51.866302490234375, 20.777435302734375, 67.35874938964844, 97.47784423828125, 138.3680419921875, 63.049224853515625, -12.707832336425781, 250.8916015625, 53.50341796875, -47.35302734375, 104.088134765625, 10.447032928466797, -75.56048583984375, 72.95370483398438, 89.5133056640625, 386.83221435546875, 406.32794189453125, 100.38043212890625, 159.0564422607422, 305.1534729003906, 166.10887145996094, 0.0, 34.53358459472656, -3.816558837890625, 213.46424865722656, -0.285736083984375, 19.827239990234375, 16.06822967529297, 114.66000366210938, 177.2467041015625, 23.592529296875, -39.87342834472656, 4.246570587158203, 2.579864501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000239.npy"}
{"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 76.74000549316406, "std": 120.34369659423828, "min": -230.95574951171875, "p10": -44.24855422973632, "median": 64.56788635253906, "p90": 237.06110534667968, "max": 639.5399169921875, "pos_frac": 0.71875, "sample": [37.624576568603516, 148.51364135742188, 51.143951416015625, 137.8981475830078, -39.820091247558594, 98.96568298339844, 87.35357666015625, 86.18238830566406, 0.0, 160.3970947265625, 70.598876953125, -230.95574951171875, -74.53192138671875, -24.72918701171875, -70.41999053955078, 133.79901123046875, 164.36331176757812, 125.41014099121094, 34.334293365478516, -42.675254821777344, 6.263031005859375, 62.4503173828125, 140.3114013671875, 38.9052734375, -1.553070068359375, 63.132843017578125, 130.22686767578125, -66.31265258789062, 321.876953125, 260.9980773925781, 93.54632568359375, -16.25897216796875, 639.5399169921875, -34.57463073730469, -89.1966552734375, 6.827301025390625, 262.23797607421875, 172.0798797607422, 106.70162200927734, -13.254051208496094, 239.09933471679688, 96.95956420898438, 321.85650634765625, -22.196807861328125, 86.49810791015625, 91.55810546875, 230.66305541992188, 80.30415344238281, 193.6807861328125, 128.10044860839844, 225.55657958984375, 43.103851318359375, 0.0, -13.327888488769531, -205.47113037109375, 48.54058837890625, 16.28924560546875, 141.7421875, 37.939971923828125, -42.49371337890625, 69.43636322021484, 236.64471435546875, -28.7821044921875, -47.919586181640625, 222.60601806640625, 217.61851501464844, 72.6265869140625, 50.42652130126953, 238.03268432617188, 34.66261291503906, 255.54901123046875, -81.15679931640625, 0.0, 60.991180419921875, -27.98845672607422, -32.005645751953125, 328.14398193359375, -12.275726318359375, 0.099212646484375, 105.8323974609375, -53.34452819824219, 71.91555786132812, 139.2108154296875, 71.62892150878906, 95.34738159179688, 185.271728515625, -25.618804931640625, 252.38644409179688, -57.33851623535156, 297.810546875, 6.819160461425781, 103.3438720703125, 20.771240234375, 101.51422119140625, 219.48367309570312, 3.833251953125, -172.06268310546875, 168.9783935546875, 74.81687927246094, 76.95437622070312, 14.8270263671875, 32.57017517089844, 6.019523620605469, -13.888107299804688, 284.06536865234375, 44.010589599609375, 182.60231018066406, 164.45819091796875, 138.11917114257812, 49.235626220703125, 197.94398498535156, 18.961700439453125, 51.163330078125, -48.996315002441406, 74.21778869628906, -20.514495849609375, -148.77685546875, 130.31044006347656, -24.93011474609375, 262.8326416015625, 0.0, 83.53118896484375, 128.56072998046875, 25.014404296875, 30.24517822265625, -2.4364490509033203, 66.0029296875, 217.50341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000240.npy"}
{"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 58.10354995727539, "std": 123.34646606445312, "min": -299.50299072265625, "p10": -82.77634887695312, "median": 39.063209533691406, "p90": 217.0051300048828, "max": 393.9942626953125, "pos_frac": 0.6875, "sample": [50.50341796875, 89.38339233398438, 176.79685974121094, 53.629730224609375, 2.3220291137695312, -116.27281188964844, 51.0111083984375, -10.070943832397461, 83.69830322265625, 215.68057250976562, 354.05194091796875, 0.0, -32.449615478515625, 349.6396484375, 83.34626007080078, 393.9942626953125, 71.2479248046875, 14.53265380859375, 14.244873046875, 199.35366821289062, 27.088302612304688, 151.6571502685547, -14.027145385742188, 39.74169921875, 69.09780883789062, 183.18862915039062, -160.80010986328125, -0.0934906005859375, 142.96865844726562, 109.48764038085938, 159.9913330078125, -110.92794799804688, 295.8131103515625, -53.3529052734375, 69.59844970703125, -25.203887939453125, -40.295745849609375, -23.73089599609375, 15.042936325073242, 55.500762939453125, -54.15795135498047, 105.95968627929688, 31.039093017578125, 192.183837890625, 61.142822265625, 168.3062286376953, 122.25933837890625, -70.93576049804688, 119.30517578125, 31.818084716796875, 200.7834014892578, 22.71950340270996, 5.109245300292969, 97.0157470703125, -175.76560974121094, 11.217559814453125, 39.41099548339844, 40.26252746582031, -113.0889892578125, -44.760223388671875, -26.0440673828125, 220.09576416015625, 200.4097900390625, 88.61581420898438, -5.1551361083984375, 229.23251342773438, 16.376602172851562, -31.3419189453125, 141.19879150390625, 172.04690551757812, -76.57638549804688, 4.658905029296875, 357.1923828125, 147.68072509765625, 17.03765869140625, -270.5147705078125, 308.55712890625, 180.07080078125, -56.823486328125, -7.72674560546875, 22.670684814453125, 50.512874603271484, 263.4442138671875, 106.44943237304688, 43.96942138671875, -24.405532836914062, -81.77427673339844, 34.6165771484375, 65.73335266113281, 224.65536499023438, 33.28352355957031, -103.69415283203125, 33.44172668457031, -89.6773681640625, 224.76327514648438, 65.20726013183594, -6.2000885009765625, -115.48204040527344, 214.31219482421875, 5.934173583984375, 69.5526123046875, -299.50299072265625, 38.715423583984375, 24.771270751953125, 321.64227294921875, 133.38070678710938, 37.501136779785156, 18.400489807128906, 104.96914672851562, 172.2945556640625, -0.19903564453125, -143.1683349609375, -85.11451721191406, -9.936943054199219, 171.2868194580078, -7.422889709472656, 161.91159057617188, 66.19683837890625, -43.715370178222656, 210.58099365234375, 249.4647216796875, 70.37802124023438, 52.151885986328125, -104.09909057617188, 30.5311279296875, -78.084228515625, 19.872772216796875, -47.06903076171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000241.npy"}
{"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 65.20032501220703, "std": 114.26319885253906, "min": -175.623779296875, "p10": -53.099954986572264, "median": 42.25935363769531, "p90": 200.05322265624994, "max": 487.04290771484375, "pos_frac": 0.7265625, "sample": [132.61074829101562, 12.754249572753906, 83.95263671875, 29.195327758789062, 72.61331176757812, -53.01897430419922, 107.25927734375, 94.0037841796875, 7.274261474609375, 4.7928924560546875, 290.8141784667969, 122.45166015625, 31.448524475097656, -82.01287841796875, -110.88809204101562, 195.3953399658203, 118.89854431152344, 186.20462036132812, -80.73735046386719, 96.94892120361328, 12.327180862426758, 167.4278106689453, 372.9856872558594, 101.8011245727539, 90.13714599609375, 269.3611145019531, 13.762039184570312, 260.49176025390625, 58.1162109375, 5.218101501464844, 124.74691772460938, -46.1270751953125, -13.14434814453125, 214.88165283203125, 40.812713623046875, 318.2272644042969, 487.04290771484375, -5.4613037109375, -154.314208984375, 16.80547332763672, 27.660125732421875, -35.58026123046875, -29.57177734375, 86.62869262695312, -114.85089111328125, 18.66550064086914, 128.62075805664062, 38.499847412109375, 158.31082153320312, -68.80789184570312, -63.86027526855469, -50.927223205566406, -53.288909912109375, 41.95832824707031, -79.62339782714844, 19.213016510009766, 51.43267822265625, 99.27789306640625, -73.021728515625, 45.95503234863281, 124.03140258789062, 40.674495697021484, -51.93743896484375, 286.04010009765625, -4.267974853515625, -6.6934814453125, 193.8623046875, 478.0636901855469, 26.652923583984375, 50.349273681640625, 87.50967407226562, 89.50338745117188, -3.2498016357421875, 67.74742126464844, 141.365234375, 20.720367431640625, 98.52418518066406, 42.56037902832031, 37.28790283203125, -18.441631317138672, 34.36045837402344, 125.25003051757812, 328.7730712890625, 25.45934295654297, -44.339935302734375, 69.55450439453125, 13.457124710083008, 17.924728393554688, 113.30496215820312, 16.8270263671875, 227.3262939453125, 48.14715576171875, 1.424224853515625, 82.9720230102539, -32.41396713256836, 276.0858154296875, 74.98112487792969, 164.3204345703125, 134.48533630371094, -10.12542724609375, -25.3223876953125, 120.11709594726562, -42.82988739013672, -24.297927856445312, 183.2212371826172, -37.52012634277344, 43.71873474121094, -175.623779296875, 16.435394287109375, 37.03392028808594, -13.469268798828125, -43.66651916503906, 90.84634399414062, 110.66680908203125, 82.90963745117188, 126.97894287109375, 126.0166015625, 33.12175750732422, 25.943603515625, 184.07940673828125, 119.02932739257812, -20.345359802246094, 52.34556579589844, -126.6329345703125, 108.86188507080078, 168.13851928710938, -94.86279296875, 210.92161560058594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000242.npy"}
{"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 80.12322998046875, "std": 138.4698028564453, "min": -311.75958251953125, "p10": -49.22453613281249, "median": 67.53742980957031, "p90": 228.35937194824217, "max": 919.318603515625, "pos_frac": 0.7265625, "sample": [-139.6895751953125, -55.29905700683594, 112.79331970214844, -37.24263000488281, 242.62548828125, 150.45455932617188, 59.8018798828125, 257.8314208984375, 31.30841064453125, 143.8506317138672, 46.8228759765625, 129.62939453125, 0.36652565002441406, -84.41297912597656, 76.1182861328125, -2.53204345703125, -46.62117004394531, 98.99806213378906, 512.4805908203125, -36.03569793701172, 5.558753967285156, 40.47760009765625, 11.677734375, 20.10637664794922, 128.19522094726562, 87.70343017578125, 919.318603515625, 116.251220703125, -10.727142333984375, 61.806915283203125, 0.7394542694091797, 353.0616455078125, -22.793487548828125, 135.84205627441406, 217.75311279296875, 61.922119140625, 157.53652954101562, 23.695220947265625, 70.15194702148438, 389.47119140625, -84.30780029296875, 145.7982177734375, 56.18170166015625, 116.42294311523438, 34.312557220458984, 195.62600708007812, 316.429443359375, -30.4080810546875, 132.399658203125, 86.80711364746094, 124.33990478515625, -311.75958251953125, 27.47613525390625, 84.53582763671875, -32.93572998046875, -194.91552734375, -99.74603271484375, 4.475677490234375, -13.434234619140625, -1.1298828125, 200.39361572265625, 107.07418823242188, 161.76217651367188, -15.16375732421875, -28.9271240234375, 160.06497192382812, 48.96903991699219, -23.1048583984375, 137.051513671875, -61.1688232421875, 51.07432556152344, -102.8094482421875, -34.568603515625, 321.29937744140625, 92.54666137695312, 84.56382751464844, -6.742950439453125, 136.463134765625, 111.48233032226562, 18.385398864746094, -10.425472259521484, 62.596290588378906, 89.3619384765625, 176.97445678710938, 8.27496337890625, 0.0, 6.3681793212890625, 105.00469970703125, 233.717529296875, 226.06301879882812, 60.152496337890625, 79.69747924804688, 130.45034790039062, -36.02231216430664, 97.89073181152344, 280.0881652832031, 65.46540832519531, 138.93780517578125, -77.83121490478516, 6.829345703125, -69.39654541015625, -36.5728759765625, -64.30148315429688, 70.04544067382812, -18.105194091796875, 242.47943115234375, 187.3675537109375, 104.06419372558594, 8.363563537597656, 56.071685791015625, 170.66456604003906, -41.09856414794922, 107.16536712646484, -63.53839111328125, 27.75244903564453, 214.960693359375, 53.04199981689453, 279.37750244140625, 169.54010009765625, 109.24270629882812, 69.60945129394531, 82.27816772460938, 280.2039794921875, 154.1348876953125, 162.70645141601562, 143.8355712890625, 102.80865478515625, -34.33184814453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000243.npy"}
{"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 79.34622955322266, "std": 118.66773223876953, "min": -451.0274353027344, "p10": -44.12876205444336, "median": 83.01158142089844, "p90": 214.46965484619133, "max": 528.3904418945312, "pos_frac": 0.75, "sample": [-451.0274353027344, 126.80520629882812, 107.3394775390625, 5.11834716796875, 249.5380859375, 50.23396301269531, 334.00665283203125, -42.684173583984375, 118.88641357421875, 67.93936157226562, 82.80728149414062, 43.918304443359375, 63.562347412109375, 25.681594848632812, 54.42100524902344, 107.395751953125, 171.525390625, 129.38912963867188, 131.3790283203125, 152.09503173828125, 99.64543151855469, 103.13806915283203, -65.18463134765625, -6.235198974609375, 127.41316223144531, -13.34912109375, -80.1934814453125, 30.608360290527344, 163.3153076171875, -28.962493896484375, 44.99873352050781, -86.01663208007812, -19.233856201171875, -40.128395080566406, -175.06243896484375, 234.3666229248047, 342.717041015625, 57.62396240234375, 160.63412475585938, 158.31776428222656, 8.022994995117188, 108.5048828125, 309.6491394042969, 100.4910888671875, 186.427734375, 63.631500244140625, 1.6241455078125, 35.89649963378906, -27.510101318359375, -11.70770263671875, 83.21588134765625, 109.007080078125, 0.0, 135.43026733398438, 71.75677490234375, -156.39352416992188, 147.274169921875, 109.13211059570312, -35.70782470703125, 36.049400329589844, 94.43740844726562, -75.14112854003906, -43.748779296875, 165.11309814453125, 25.209060668945312, 82.26898193359375, 155.10140991210938, 262.372802734375, 528.3904418945312, 100.57916259765625, 193.06298828125, 102.27381896972656, -164.095458984375, 113.8480224609375, 105.81271362304688, -45.858673095703125, 130.38671875, -23.3433837890625, 73.187744140625, 129.4484405517578, -7.786224365234375, 142.6217041015625, 112.37245178222656, -68.58322143554688, 77.99169921875, 151.05560302734375, -15.6890869140625, 305.99346923828125, 268.74249267578125, -0.30859375, 54.278564453125, -17.79296875, 255.20590209960938, 122.6690673828125, 167.12423706054688, 238.7714385986328, 284.67236328125, 19.052913665771484, 205.9423828125, 265.4071044921875, 56.95460510253906, 135.48065185546875, 57.66050720214844, 46.64459991455078, 5.500823974609375, 118.57547760009766, 112.37078857421875, 35.277191162109375, 160.419189453125, 125.5063247680664, -90.9796142578125, 25.969390869140625, -36.14919662475586, 94.6854248046875, 92.91946411132812, 204.7366943359375, -21.256107330322266, -45.01538848876953, 28.807350158691406, -31.353530883789062, 137.8154296875, 67.97616577148438, 204.2918701171875, -50.389190673828125, 80.61956787109375, 159.10406494140625, 108.99972534179688, 88.56088256835938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000244.npy"}
{"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 79.7557373046875, "std": 113.9113998413086, "min": -210.94900512695312, "p10": -40.56087951660156, "median": 70.81748962402344, "p90": 222.7940002441406, "max": 364.6429138183594, "pos_frac": 0.7734375, "sample": [35.996826171875, -26.934005737304688, 136.40277099609375, 61.96131134033203, 29.271575927734375, 77.636474609375, 70.44464111328125, -9.566162109375, 3.653289794921875, 146.22955322265625, 54.401336669921875, 81.96597290039062, -11.685455322265625, 142.4852294921875, 345.74676513671875, -1.45648193359375, 210.9423828125, 199.97775268554688, 10.548736572265625, 262.93780517578125, -204.6059112548828, -140.89263916015625, -32.41007995605469, -42.23468017578125, 152.1002960205078, 83.27099609375, 0.0, 38.72503662109375, 218.93902587890625, -33.536865234375, 300.70245361328125, 152.1648406982422, 167.78997802734375, 134.70440673828125, 79.37677001953125, 177.23458862304688, 74.2832260131836, 178.10226440429688, 140.35159301757812, 94.06393432617188, 258.65582275390625, 66.67538452148438, 26.595733642578125, 209.800537109375, 37.363006591796875, 109.17794799804688, 45.8775634765625, 121.13693237304688, 67.50598907470703, -46.48353576660156, 9.18438720703125, 9.455034255981445, 35.08837890625, 364.6429138183594, 231.7889404296875, -27.84064483642578, 65.58355712890625, 176.3773956298828, 238.0970001220703, 303.3883056640625, 216.7564697265625, 32.21947479248047, 88.55648803710938, 13.0699462890625, 95.5792236328125, 145.56985473632812, 122.54049682617188, -130.34597778320312, 91.16717529296875, -156.7579803466797, 31.086196899414062, 109.21235656738281, -2.398286819458008, 119.70819091796875, 10.685455322265625, -91.16363525390625, -39.843536376953125, 6.220272064208984, 288.5480041503906, 179.60635375976562, -32.64622497558594, 170.69943237304688, 24.44720458984375, 62.62104034423828, 26.70794677734375, -143.85923767089844, 8.859375, 205.35916137695312, 147.0184326171875, -19.238677978515625, 234.1290283203125, 109.27255249023438, -52.791839599609375, 187.41763305664062, -8.628448486328125, 284.5508728027344, 191.8714599609375, 109.08056640625, 338.5797119140625, 55.44401550292969, 60.91339111328125, 105.15229797363281, 63.9581298828125, -210.94900512695312, -6.7679901123046875, 89.63397216796875, 142.91409301757812, 111.63873291015625, 131.25811767578125, 62.11326599121094, -16.720428466796875, -47.54273223876953, 151.5968017578125, 71.19033813476562, 244.6243896484375, 37.3507080078125, 46.33345031738281, 27.079208374023438, 98.38139343261719, -147.42355346679688, 194.7379150390625, 28.147140502929688, 178.89743041992188, 137.69204711914062, -192.9349365234375, 0.9396381378173828, 0.0, 156.45021057128906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000245.npy"}
{"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 54.75814437866211, "std": 128.42591857910156, "min": -283.0592956542969, "p10": -107.91642761230469, "median": 50.448638916015625, "p90": 229.65841369628905, "max": 389.9877624511719, "pos_frac": 0.671875, "sample": [7.3108673095703125, 129.57122802734375, 93.07699584960938, -84.5513916015625, 166.9573974609375, -107.2227783203125, 321.6209411621094, 29.9100341796875, 54.96981430053711, 389.9877624511719, -61.70097351074219, 68.56963348388672, -31.726715087890625, 71.12559509277344, 8.013313293457031, -11.550994873046875, -23.326980590820312, 159.8328857421875, 332.59893798828125, -174.39480590820312, 48.54351806640625, 296.07305908203125, 52.353759765625, 124.52225494384766, 58.135986328125, -69.40982055664062, 148.80746459960938, 4.633758544921875, -9.22174072265625, 16.318008422851562, 96.017578125, -46.03314208984375, 338.97589111328125, 60.919029235839844, -109.89990234375, -62.65791320800781, -166.1317138671875, 148.3350830078125, 94.1038818359375, 3.821901321411133, 95.4464111328125, -122.87603759765625, 83.04800415039062, 220.35348510742188, -93.90335083007812, -43.500732421875, 20.95384979248047, 0.286163330078125, -30.1204833984375, 66.6248779296875, 19.34991455078125, 152.7554931640625, -115.54010009765625, 76.49639892578125, 231.06689453125, -146.17616271972656, 165.4820556640625, -100.18367767333984, -283.0592956542969, 9.934427261352539, 334.00238037109375, -109.70562744140625, 86.4527587890625, -93.85214233398438, -157.04359436035156, 29.722084045410156, 86.8760986328125, 160.330078125, -6.7318878173828125, 220.82830810546875, 383.4410400390625, -9.27630615234375, -59.383270263671875, 6.854248046875, 12.49530029296875, 56.399627685546875, 251.4372100830078, -52.79449462890625, -152.7335205078125, 58.33100891113281, -96.98822021484375, -68.07516479492188, 78.063232421875, 66.48626708984375, 28.492076873779297, -33.7275390625, -193.89059448242188, -13.7430419921875, 113.3363037109375, 35.12421417236328, 43.488677978515625, 182.48373413085938, 91.2855453491211, -109.53494262695312, 35.49976348876953, 273.82049560546875, -63.48816680908203, 81.9517822265625, 0.0, 93.72125244140625, 229.05477905273438, 237.8511962890625, -56.47010803222656, 84.32391357421875, 260.53839111328125, 115.78585815429688, -17.81231689453125, 151.75592041015625, 105.12002563476562, 101.16548156738281, 121.42160034179688, 6.978118896484375, 227.09646606445312, 180.46414184570312, 73.70655822753906, 186.146484375, -43.493064880371094, -128.6302490234375, -31.8984375, 40.657501220703125, 266.359130859375, 104.025634765625, 138.3478546142578, 146.6224365234375, 71.06083679199219, 47.43849182128906, 9.568191528320312, 218.18655395507812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000246.npy"}
{"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 71.74578857421875, "std": 116.34754943847656, "min": -158.34121704101562, "p10": -60.559897613525386, "median": 60.282012939453125, "p90": 212.1671340942383, "max": 496.80059814453125, "pos_frac": 0.703125, "sample": [167.51771545410156, -94.51947021484375, 128.6285858154297, 139.07850646972656, -29.15631103515625, 53.729156494140625, 172.17422485351562, 124.0143814086914, 235.1914520263672, 17.352313995361328, 0.0, 226.37881469726562, 159.61123657226562, 12.1605224609375, 82.23655700683594, 269.6915283203125, -2.0002593994140625, 14.415496826171875, -20.659210205078125, 98.7003173828125, 12.552169799804688, 194.91326904296875, -12.66375732421875, -17.820396423339844, 78.33453369140625, 109.85289001464844, 85.5479736328125, -23.98748779296875, 64.929931640625, 173.431396484375, 92.74081420898438, 211.96987915039062, -12.966033935546875, 402.1637268066406, 147.93685913085938, -111.6552734375, -1.0682220458984375, 327.812255859375, -23.930419921875, 27.144302368164062, -63.064964294433594, -24.064163208007812, 25.30255126953125, 11.44390869140625, 163.32952880859375, 123.56103515625, 3.2976341247558594, 24.5096435546875, 7.870857238769531, -100.20941162109375, 21.900386810302734, 91.03532409667969, -27.490310668945312, 103.50555419921875, -134.68215942382812, 235.248779296875, 223.33816528320312, 110.78616333007812, 161.67770385742188, -30.567283630371094, -145.80271911621094, 66.90069580078125, 33.81079864501953, 386.21160888671875, 139.0498504638672, -59.486297607421875, 137.18836975097656, -101.40560913085938, -30.598800659179688, -85.16355895996094, 34.87458801269531, 496.80059814453125, -122.07998657226562, 27.818035125732422, 130.50967407226562, 25.574432373046875, 86.03233337402344, -0.34442138671875, 173.40478515625, 161.4749298095703, 168.29061889648438, 62.93824005126953, 297.03466796875, -18.50531005859375, -158.34121704101562, 12.080413818359375, 46.97624969482422, -12.173538208007812, 217.84149169921875, 88.57354736328125, 82.67413330078125, 210.53501892089844, -1.04669189453125, 97.2841796875, 55.105194091796875, 142.93226623535156, 116.32708740234375, 200.15487670898438, 211.68511962890625, -40.165557861328125, 50.70610809326172, 55.137367248535156, 52.24169921875, 115.58880615234375, 147.8391571044922, -17.438278198242188, 68.9913330078125, -85.18789672851562, 383.17071533203125, 89.524658203125, -113.58523559570312, 87.98827362060547, 57.62578582763672, 64.31475830078125, -2.372373580932617, -28.19969940185547, 134.16909790039062, -41.74513626098633, -127.3800048828125, -2.234130859375, 143.2071533203125, 87.04244995117188, 212.6273956298828, 3.6964569091796875, 131.57887268066406, 57.09417724609375, 15.773590087890625, 101.80355834960938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000247.npy"}
{"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 67.37956237792969, "std": 120.76802825927734, "min": -338.09613037109375, "p10": -82.25841751098632, "median": 61.325157165527344, "p90": 229.38685302734373, "max": 420.097900390625, "pos_frac": 0.7421875, "sample": [29.306785583496094, 235.41848754882812, 0.05870819091796875, 120.08865356445312, 254.684814453125, -16.7052001953125, 160.09210205078125, -108.83856201171875, 158.33380126953125, 29.983055114746094, 19.466712951660156, 53.65473937988281, 112.25650024414062, -7.063079833984375, -58.26145935058594, 49.406463623046875, 287.9554138183594, 12.360015869140625, -27.620986938476562, 26.601083755493164, -67.59378814697266, 233.3225555419922, 3.2975311279296875, 120.19680786132812, 261.7921447753906, 164.274658203125, 118.01095581054688, 59.994873046875, -18.904815673828125, 198.53497314453125, 125.22969055175781, -145.37713623046875, 236.78231811523438, -23.723648071289062, 9.88740348815918, 124.06625366210938, 259.736083984375, -338.09613037109375, -18.703548431396484, -80.80826568603516, -16.3736572265625, 8.921073913574219, 145.97166442871094, -113.31048583984375, -96.0062255859375, 227.7298583984375, 10.273876190185547, -99.41339111328125, 73.84672546386719, 46.199310302734375, 52.450164794921875, 340.5909729003906, 98.58157348632812, 132.3264923095703, 420.097900390625, -27.08624267578125, 27.4136962890625, 3.1520957946777344, 107.41903686523438, 116.73141479492188, 111.953369140625, 208.82460021972656, -24.4393310546875, -46.20811462402344, 163.48562622070312, 82.2965087890625, 259.4415283203125, 205.0802459716797, -41.549156188964844, 127.24392700195312, 128.5693359375, -76.71926879882812, 36.831661224365234, 46.312416076660156, 27.60345458984375, -133.81565856933594, 0.0, 142.11508178710938, 233.253173828125, 112.02587890625, 143.09805297851562, 120.6195068359375, 93.7885971069336, 164.2364501953125, 62.99726867675781, 240.51451110839844, -78.36761474609375, -99.9302978515625, 79.89363098144531, 152.35220336914062, 67.7882080078125, -19.28387451171875, 162.69107055664062, 118.53453063964844, 182.1661376953125, -139.34674072265625, 169.647216796875, 71.59663391113281, 37.671051025390625, 97.444580078125, 1.9945125579833984, -180.10003662109375, 217.0517578125, 80.42617797851562, 102.72976684570312, 223.24526977539062, 116.22775268554688, -85.64210510253906, 47.30419921875, -101.52079772949219, 35.88795471191406, 23.29669189453125, 93.59089660644531, 62.65544128417969, 414.863525390625, 133.15139770507812, 23.865325927734375, 151.48764038085938, 35.474822998046875, 64.58685302734375, -14.993888854980469, -41.02348327636719, 23.180068969726562, 1.922576904296875, 146.2654571533203, -211.16851806640625, 43.789825439453125, 13.005157470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000248.npy"}
{"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 58.26978302001953, "std": 132.5446014404297, "min": -462.0299072265625, "p10": -84.84311676025389, "median": 50.162841796875, "p90": 219.474658203125, "max": 429.2615966796875, "pos_frac": 0.6796875, "sample": [-96.22105407714844, 78.23245239257812, -76.06329345703125, -59.698638916015625, 10.559005737304688, 0.9858627319335938, -7.555419921875, -81.96376037597656, -160.72666931152344, 25.802810668945312, -29.615562438964844, 16.04999542236328, 131.74462890625, 190.24517822265625, -67.0179443359375, 237.43939208984375, 98.93325805664062, 125.45321655273438, 15.45501708984375, 182.15777587890625, 76.29415893554688, -462.0299072265625, 200.1917266845703, 53.241180419921875, -11.100494384765625, 373.28515625, 106.65859985351562, -105.43443298339844, -106.76498413085938, 80.64649963378906, 52.14378356933594, -64.61630249023438, 73.96875, -106.07583618164062, 41.93341064453125, 59.165916442871094, 233.4595947265625, 1.7614822387695312, 127.44754028320312, -134.16452026367188, 119.31767272949219, 207.9647216796875, -32.18901062011719, 198.52188110351562, 246.44229125976562, -33.7440185546875, -25.609771728515625, -18.742996215820312, 178.34530639648438, 26.113739013671875, -22.373565673828125, 203.46591186523438, 219.77206420898438, 177.27349853515625, 211.2364501953125, 131.9547119140625, 45.16001510620117, 71.54102325439453, -68.0733642578125, 21.621673583984375, 295.0765686035156, -63.11236572265625, -10.573150634765625, 1.173370361328125, 116.636474609375, -5.393180847167969, 135.925537109375, -26.0755615234375, 105.80719757080078, 60.92156982421875, -34.46293640136719, 16.08324432373047, 235.32858276367188, 133.7661895751953, 162.9985809326172, 102.7479248046875, 239.3255615234375, 170.10023498535156, 212.1425018310547, 146.28390502929688, 219.34719848632812, 138.11911010742188, 90.76537322998047, 47.91377258300781, 11.7423095703125, 101.89077758789062, -38.51936721801758, -80.5458984375, 13.37060546875, 219.00161743164062, 41.90278625488281, -71.794189453125, 252.91131591796875, 51.9100341796875, 136.97491455078125, 288.94256591796875, 132.57595825195312, 202.23573303222656, -91.56161499023438, -243.04660034179688, 429.2615966796875, -174.8349609375, 28.65559959411621, 23.11871337890625, 77.61632537841797, 223.49005126953125, -25.270904541015625, -189.16168212890625, -60.66912841796875, -69.63365173339844, -230.72952270507812, 12.730365753173828, 188.90969848632812, 10.818107604980469, 46.35333251953125, -27.835784912109375, 60.350555419921875, 48.4156494140625, 211.77230834960938, -76.796875, 203.229248046875, 141.12814331054688, 157.19363403320312, -8.871780395507812, -176.30715942382812, 12.2935791015625, 68.31021118164062, 253.95423889160156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000249.npy"}
{"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 82.12449645996094, "std": 140.8482208251953, "min": -274.4635009765625, "p10": -70.36655654907226, "median": 73.10792541503906, "p90": 261.06042175292964, "max": 426.60595703125, "pos_frac": 0.7265625, "sample": [76.14930725097656, -1.979766845703125, -253.5049591064453, 183.453125, 118.81500244140625, 222.66668701171875, 221.22232055664062, 102.19664764404297, 216.17140197753906, -60.69214630126953, 12.184913635253906, 85.17578887939453, 383.31353759765625, 282.79510498046875, 30.816619873046875, 415.3601379394531, 93.11355590820312, 98.27044677734375, 193.5906982421875, 388.8905029296875, 267.586669921875, 46.6700439453125, 35.62315368652344, 199.98582458496094, -72.72357177734375, -24.028472900390625, -274.4635009765625, 37.2779541015625, 15.07733154296875, 7.94732666015625, 225.43096923828125, -28.179473876953125, 170.1971435546875, 418.7436828613281, -66.36693572998047, 86.36358642578125, 11.08203125, 1.0395584106445312, 246.89324951171875, -145.30050659179688, -13.45980453491211, -153.4324188232422, 291.02191162109375, 121.47016906738281, 76.41592407226562, 73.19760131835938, -247.19287109375, -37.9404296875, -3.340810775756836, 245.5251922607422, 139.22933959960938, 78.88455200195312, 50.253265380859375, 52.532318115234375, 74.75210571289062, 57.7630615234375, 85.18836212158203, -54.91033935546875, -2.549468994140625, -81.18428039550781, 218.25233459472656, 60.36907958984375, 165.46002197265625, -78.30271911621094, 91.26029205322266, 246.5133056640625, 258.2634582519531, 73.01824951171875, -46.49258041381836, 227.33509826660156, 75.12960815429688, 196.72494506835938, 53.656585693359375, 131.9068603515625, 47.73150634765625, -46.464813232421875, 30.354705810546875, 212.4771728515625, 231.82504272460938, 27.127609252929688, 115.98162078857422, -39.2728271484375, 178.09454345703125, -220.87136840820312, 223.51348876953125, 89.48452758789062, 305.62060546875, -168.283935546875, 144.26742553710938, 243.54238891601562, 78.12460327148438, 377.365966796875, -69.35640716552734, 19.84857177734375, -206.930419921875, 45.62391662597656, 61.656280517578125, -62.07136535644531, -34.29486083984375, 270.2335205078125, 176.10589599609375, 152.00155639648438, 113.720458984375, 162.7277069091797, 61.979373931884766, 275.22430419921875, -104.79541015625, -38.6190185546875, 15.815948486328125, 13.71826171875, -0.3756103515625, 57.941253662109375, 426.60595703125, -15.286712646484375, 178.23007202148438, 142.84921264648438, 279.68927001953125, 33.186279296875, 36.229034423828125, 118.87689208984375, 2.2168121337890625, 164.36819458007812, 77.65249633789062, -79.10469055175781, -20.031341552734375, -21.5623779296875, 68.44662475585938, -13.385913848876953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000250.npy"}
{"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 75.82389831542969, "std": 143.2790069580078, "min": -193.39453125, "p10": -64.87655029296874, "median": 49.42414474487305, "p90": 212.4163070678711, "max": 801.57373046875, "pos_frac": 0.71875, "sample": [15.400222778320312, -30.4346923828125, 65.51876831054688, -77.83916473388672, 74.52432250976562, -60.697357177734375, 23.01873779296875, -128.02371215820312, 88.232177734375, 84.62396240234375, 45.672080993652344, 153.1986083984375, 173.71380615234375, 17.301910400390625, -7.1883544921875, 21.51929473876953, 36.020355224609375, -59.07386779785156, 121.48783111572266, 190.29876708984375, -39.70600891113281, 42.74089050292969, 24.939849853515625, 194.802978515625, 132.75985717773438, 201.56524658203125, -55.696533203125, 137.6147003173828, 195.9737548828125, 32.48974609375, 48.23992919921875, 51.6470947265625, 24.927162170410156, 161.272705078125, 81.76129150390625, -101.5246353149414, 264.6539306640625, 17.99301528930664, -24.695419311523438, -85.55579376220703, 109.69403076171875, 105.88943481445312, 135.78704833984375, 398.8770446777344, 62.367919921875, -30.968963623046875, -193.39453125, -92.97530364990234, 17.937898635864258, 116.63641357421875, -70.72792053222656, 72.93634033203125, 801.57373046875, 136.1820068359375, 100.92130279541016, 108.38311767578125, 18.0347900390625, 89.8961181640625, 49.09282684326172, -4.173164367675781, 196.8104248046875, -135.55227661132812, 22.840667724609375, 96.04507446289062, 80.88308715820312, -85.32010650634766, -12.669395446777344, 9.30331802368164, 61.69091796875, -17.818355560302734, 212.24044799804688, -62.36882019042969, 212.82664489746094, 14.638040542602539, 196.68409729003906, -31.62115478515625, 48.36616516113281, 24.621826171875, 38.640159606933594, 44.939971923828125, 46.87445068359375, 6.7277984619140625, 3.6337242126464844, 23.019119262695312, 125.63314056396484, 106.69017791748047, 124.9342041015625, -39.394775390625, -160.1113739013672, 52.74462890625, 43.715396881103516, 626.0445556640625, 184.605712890625, 82.05352783203125, 559.22412109375, -2.138143539428711, 453.9361572265625, 128.93453979492188, 169.88229370117188, -98.54730987548828, 212.0186767578125, 55.197052001953125, 231.4879150390625, -21.961212158203125, 263.2187194824219, 71.06988525390625, 170.027099609375, 43.90574645996094, 70.83192443847656, -44.95733642578125, -4.943023681640625, 176.56625366210938, -58.3758544921875, 53.47996520996094, -17.04059600830078, -81.73355102539062, -93.9146728515625, -27.294532775878906, -56.97686004638672, 249.2584228515625, 270.2528991699219, 89.91349792480469, 367.7647705078125, 255.02072143554688, -12.19842529296875, 49.755462646484375, 95.42471313476562, 60.57081604003906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000251.npy"}
{"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 71.3315200805664, "std": 127.52811431884766, "min": -297.6463317871094, "p10": -65.84519271850584, "median": 62.25157165527344, "p90": 224.48345794677735, "max": 423.41473388671875, "pos_frac": 0.71875, "sample": [229.03384399414062, 284.298583984375, 262.6562194824219, 225.41819763183594, -8.226119995117188, 26.834640502929688, -11.065238952636719, 4.07403564453125, 33.655128479003906, -89.61459350585938, 64.31005096435547, -248.48960876464844, 119.40594482421875, 83.04037475585938, 423.41473388671875, 92.67361450195312, 142.162353515625, 177.6322021484375, 163.8414306640625, 116.4381103515625, -1.3930320739746094, 242.22830200195312, 13.54803466796875, 395.36041259765625, -297.6463317871094, 1.6149673461914062, -2.803558349609375, 190.45632934570312, -14.492034912109375, -40.47706604003906, 72.7547836303711, 208.73837280273438, 106.0426025390625, 66.872314453125, -92.25911712646484, -16.6806640625, -291.8719787597656, -77.27753448486328, 62.087249755859375, 255.9127197265625, 136.4462890625, 14.446540832519531, 4.670631408691406, 60.74237060546875, -20.3665771484375, 81.48905944824219, 28.34735107421875, 7.762725830078125, 92.31759643554688, -60.16236114501953, 224.08285522460938, -9.7442626953125, 286.41412353515625, 182.9819793701172, 15.2763671875, 64.6566162109375, 166.45977783203125, -94.08001708984375, 115.30598449707031, 19.427886962890625, -39.34637451171875, -88.66780090332031, 68.22344970703125, 259.95697021484375, 33.74424743652344, 85.8077392578125, 195.37794494628906, 117.876953125, 176.470947265625, 240.59393310546875, -22.178955078125, 183.66598510742188, 144.68603515625, -4.461669921875, 59.80670166015625, 77.22415161132812, -53.27899169921875, 46.53038024902344, 220.07305908203125, 37.66551971435547, 152.59423828125, 85.67681884765625, 213.8681640625, 219.09298706054688, 58.80780029296875, -38.44036865234375, 0.0, 107.17031860351562, -201.95086669921875, 214.5813446044922, 0.6145668029785156, 178.93789672851562, 34.132904052734375, 54.87103271484375, 38.05389404296875, 193.0521240234375, -0.8251190185546875, 403.4469299316406, 76.82681274414062, 43.21380615234375, -121.1080322265625, 134.86517333984375, 134.431884765625, -57.455535888671875, 218.8424530029297, 45.9715576171875, 21.943023681640625, -181.10211181640625, 146.29217529296875, -20.4462890625, -16.646942138671875, -184.09893798828125, 217.38241577148438, -60.94561767578125, 24.35601806640625, 16.01812744140625, 34.460601806640625, 233.84634399414062, -81.83245849609375, 205.20907592773438, 87.219970703125, 182.57057189941406, 62.4158935546875, -7.70538330078125, -22.978607177734375, 111.75887298583984, 154.98529052734375, 90.00637817382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000252.npy"}
{"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 69.97415924072266, "std": 142.37460327148438, "min": -313.84991455078125, "p10": -78.43677673339843, "median": 52.00999450683594, "p90": 265.151089477539, "max": 479.8239440917969, "pos_frac": 0.6796875, "sample": [298.3771667480469, 6.0107421875, 92.69022369384766, 35.4822998046875, -2.51239013671875, -71.83467102050781, 1.2920684814453125, 459.6079406738281, 364.6051025390625, 27.389068603515625, 58.19403076171875, 86.45867156982422, -14.05584716796875, -77.7281494140625, 114.84808349609375, -149.82369995117188, 167.18829345703125, -110.15306091308594, 281.576904296875, -57.4930419921875, 152.8768310546875, 6.29901123046875, 120.42777252197266, 94.3781509399414, -19.3388671875, -313.84991455078125, 69.2677001953125, 225.64309692382812, 74.18575286865234, 88.22367095947266, 13.983369827270508, -98.83201599121094, -214.4449920654297, -4.6561279296875, 237.29644775390625, 140.58547973632812, -115.73484802246094, 50.999542236328125, -262.9881591796875, -14.169891357421875, 181.9310302734375, -1.270355224609375, 191.4150390625, 200.1209259033203, 73.3516845703125, -145.88401794433594, 130.65786743164062, 58.499114990234375, 24.302413940429688, 97.19692993164062, 372.5895080566406, 53.02044677734375, 212.2564697265625, 149.3539276123047, 120.1134033203125, 29.612548828125, -5.952568054199219, 27.717994689941406, -74.66305541992188, 62.24896240234375, 269.32769775390625, -96.82292175292969, -24.080673217773438, 237.42593383789062, 84.25973510742188, 89.63412475585938, 22.855438232421875, 18.889236450195312, 405.7547912597656, -167.3704833984375, 42.73326110839844, -57.883026123046875, -68.63124084472656, 131.46780395507812, -62.700103759765625, 3.9382553100585938, -121.09185791015625, -0.5228786468505859, -15.453811645507812, -11.49676513671875, 19.14095687866211, -9.267013549804688, 178.86224365234375, 89.60894775390625, 166.14630126953125, -21.47723388671875, 31.11767578125, -48.399078369140625, -132.22531127929688, 479.8239440917969, 64.58306884765625, -74.01226806640625, 116.07159423828125, -31.196136474609375, 30.05523681640625, 238.1127166748047, -80.09024047851562, 275.34722900390625, 159.2998809814453, 48.98716735839844, 235.9689178466797, 61.792137145996094, -39.77093505859375, -45.43829345703125, 187.4503173828125, 196.37184143066406, 85.79745483398438, 324.1286315917969, 111.62069702148438, 8.051685333251953, 69.2911376953125, 106.95925903320312, 110.9194107055664, -71.42131042480469, -15.823131561279297, 23.791244506835938, 54.078765869140625, 197.9565887451172, 130.159912109375, 290.4227294921875, 40.88056945800781, 450.68316650390625, 337.49725341796875, 163.07537841796875, -50.22711181640625, 36.15483093261719, 15.348060607910156, 263.3611145019531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000253.npy"}
{"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 78.08509063720703, "std": 132.2981414794922, "min": -180.39529418945312, "p10": -68.22195892333984, "median": 63.22199630737305, "p90": 237.49539184570312, "max": 557.614990234375, "pos_frac": 0.6953125, "sample": [211.447265625, 17.458480834960938, -180.39529418945312, 35.198020935058594, -150.63571166992188, 1.1545581817626953, 250.7498016357422, -17.92351531982422, 6.560308456420898, 6.454933166503906, -63.40176773071289, -56.752105712890625, -31.259368896484375, 43.9661865234375, -38.62908172607422, 165.7711181640625, 73.06317138671875, 110.46173095703125, -27.52105712890625, -81.294921875, -82.20465087890625, 198.40570068359375, 54.664794921875, 4.97265625, 43.076751708984375, 25.603134155273438, 206.84646606445312, 10.228790283203125, 142.67822265625, 63.0262451171875, 83.30999755859375, 395.6696472167969, -161.2901611328125, 176.38717651367188, 69.97865295410156, 80.740478515625, -65.43238830566406, 127.87174987792969, -17.976280212402344, -69.30284881591797, -61.759429931640625, 243.96763610839844, 119.8740234375, -143.48443603515625, -34.88203430175781, 63.417747497558594, 195.05593872070312, 276.5419616699219, 557.614990234375, 219.81423950195312, 38.12584686279297, 0.26464271545410156, 15.124523162841797, 215.16513061523438, 115.60107421875, -7.3564453125, 140.96527099609375, 224.37399291992188, 220.49618530273438, 66.02444458007812, 161.12603759765625, -85.50774383544922, 196.83056640625, 5.533393859863281, 97.50839233398438, 98.65802001953125, 87.23995971679688, 138.99249267578125, 344.32208251953125, 152.50765991210938, -22.726287841796875, 18.70159912109375, 162.02902221679688, 85.430908203125, 136.0624542236328, -30.674530029296875, -124.65869140625, 47.798583984375, -97.05634307861328, -44.60027313232422, -50.98004150390625, -9.278594970703125, 143.1295166015625, 286.64892578125, 133.21324157714844, 192.12515258789062, 145.73565673828125, 64.4989013671875, -5.75445556640625, 351.47210693359375, 9.83272933959961, 230.57138061523438, 236.751708984375, 192.74452209472656, 269.117919921875, 82.16339111328125, -0.05978202819824219, 456.8424987792969, -14.383148193359375, 21.39251708984375, -43.231292724609375, 138.49111938476562, 85.13211059570312, -3.9632930755615234, 0.0, -63.623687744140625, 239.23065185546875, 137.61793518066406, 67.04666137695312, 159.67698669433594, 26.29589080810547, -67.75872039794922, 151.8495330810547, 361.0026550292969, 72.00518798828125, -9.260223388671875, -163.24978637695312, -15.223220825195312, 184.50115966796875, 33.07640075683594, -89.2774658203125, -86.31497192382812, 36.79785919189453, 366.37322998046875, 53.179046630859375, 147.83120727539062, 159.5252685546875, 27.187210083007812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000254.npy"}
{"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 85.91107940673828, "std": 152.5751495361328, "min": -589.5400390625, "p10": -76.84444274902343, "median": 80.76709365844727, "p90": 267.72395019531245, "max": 544.9794311523438, "pos_frac": 0.734375, "sample": [-24.672380447387695, -72.86273193359375, 173.3726806640625, 323.30120849609375, 80.99394226074219, -297.609130859375, -116.79586791992188, 166.18307495117188, -39.84442138671875, 183.1660919189453, -46.41925048828125, -43.69013214111328, 27.424175262451172, 66.74896240234375, 544.9794311523438, 493.20751953125, -21.548133850097656, 130.14166259765625, 349.64794921875, 14.23284912109375, 335.3533020019531, -62.82319641113281, 116.6387939453125, 161.58682250976562, 84.91366577148438, 190.8900604248047, 88.40355682373047, 217.46694946289062, -21.98724365234375, 33.912689208984375, 118.9393310546875, 123.483154296875, 140.20614624023438, 57.924713134765625, 80.54024505615234, 17.417266845703125, 244.50918579101562, 148.52499389648438, -124.47842407226562, -29.434295654296875, 265.44476318359375, 199.72222900390625, 339.255615234375, 72.69378662109375, -191.51048278808594, -37.29576873779297, -104.03436279296875, 117.51263427734375, 234.1753387451172, 116.43540954589844, 184.8701171875, 241.9644775390625, 273.04205322265625, 77.42929077148438, 199.31201171875, -14.436004638671875, 78.45391845703125, -86.13510131835938, -232.4159393310547, 107.42898559570312, 54.196624755859375, 302.9258728027344, 71.28143310546875, 138.0167999267578, 62.3287353515625, 177.12643432617188, 0.2889213562011719, -174.40518188476562, 325.9705810546875, 25.726654052734375, 135.257568359375, -38.4796142578125, 4.671195983886719, -89.66845703125, -14.83795166015625, 115.64427185058594, 2.018939971923828, 176.58486938476562, 126.21685791015625, 194.58705139160156, 25.929794311523438, 112.65261840820312, -138.61123657226562, 282.8796691894531, 26.7816162109375, 260.2164001464844, 56.8505859375, 167.22128295898438, -9.035018920898438, 30.7608642578125, 190.0147705078125, 91.97308349609375, 184.54885864257812, 356.751220703125, 10.2208251953125, 309.7777099609375, 216.75811767578125, 68.30770874023438, 99.32498168945312, 60.93428039550781, 18.722164154052734, -97.5303955078125, -18.657020568847656, -30.427322387695312, 198.35601806640625, 175.5753173828125, -71.82981872558594, 260.27838134765625, 58.281341552734375, 149.674072265625, 9.44244384765625, -18.68768310546875, -98.90199279785156, 246.53353881835938, 295.99462890625, 16.522491455078125, -62.52471923828125, -71.31103515625, 85.15794372558594, 90.080810546875, 42.009071350097656, 166.9200439453125, -23.12347412109375, 208.06329345703125, -589.5400390625, 223.48114013671875, 34.563751220703125, 145.92694091796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000255.npy"}
{"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 78.49009704589844, "std": 150.02528381347656, "min": -659.7313232421875, "p10": -86.10993804931638, "median": 90.97138214111328, "p90": 242.2748077392578, "max": 509.7940979003906, "pos_frac": 0.75, "sample": [24.459243774414062, 131.33013916015625, 232.31716918945312, -67.941162109375, 235.78973388671875, 117.34088134765625, 150.33749389648438, 125.62579345703125, 161.74374389648438, 7.3473663330078125, 143.13856506347656, -75.86558532714844, -15.266143798828125, -127.52871704101562, 2.7927494049072266, 169.2989959716797, 310.6383056640625, -175.607666015625, 196.5937957763672, 250.1557159423828, 108.3089599609375, 232.06272888183594, -76.87641906738281, 246.14617919921875, 57.864013671875, 327.07733154296875, 350.2293701171875, -659.7313232421875, 243.60818481445312, 87.62408447265625, 241.86801147460938, 88.80317687988281, -55.34588623046875, 16.198638916015625, 111.33743286132812, -30.2255859375, -80.39447021484375, -226.24810791015625, 154.44509887695312, 54.358009338378906, 373.31414794921875, 122.34768676757812, 90.79627990722656, 152.72848510742188, 40.5244140625, 123.8489990234375, 23.470199584960938, -33.3670654296875, -99.44602966308594, 239.0670166015625, -11.631134033203125, 92.78910827636719, 203.13455200195312, 149.4930419921875, -153.32351684570312, 18.88613510131836, -207.5316162109375, -59.58074951171875, 118.80847930908203, 129.95269775390625, 240.42164611816406, -3.1487274169921875, 140.1087646484375, 181.98849487304688, 1.7182502746582031, 64.77255249023438, 85.30259704589844, -162.20120239257812, 126.7254638671875, 66.7503433227539, -64.03689575195312, -184.80245971679688, 233.3983154296875, 162.92355346679688, 190.1974334716797, 509.7940979003906, -55.56964111328125, 99.00811767578125, 14.68890380859375, 218.06719970703125, 212.01995849609375, 154.8546905517578, 244.8751678466797, 13.92193603515625, 112.0023193359375, 206.66796875, -120.60479736328125, 41.406333923339844, 47.71429443359375, 248.004150390625, -38.51432800292969, 174.38197326660156, 130.88572692871094, 416.24383544921875, -49.394073486328125, 243.2239990234375, 42.21076965332031, -2.699951171875, 30.18700408935547, 91.146484375, 82.627197265625, 20.203908920288086, 269.2287292480469, 125.326416015625, 204.52703857421875, 102.58489990234375, 117.42092895507812, 119.46383666992188, -155.0797882080078, -257.5386657714844, 50.055206298828125, 175.39878845214844, 202.26544189453125, 71.18142700195312, 18.214548110961914, -43.93794250488281, 130.97784423828125, -68.9462890625, -18.076515197753906, -138.9112091064453, 155.85491943359375, 16.82440185546875, 28.049896240234375, 120.84681701660156, 38.700439453125, 47.95116424560547, 74.49237060546875, 190.294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000256.npy"}
{"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 80.12361145019531, "std": 133.01927185058594, "min": -298.0881042480469, "p10": -58.83943328857421, "median": 59.612953186035156, "p90": 266.80272216796874, "max": 502.14617919921875, "pos_frac": 0.6796875, "sample": [59.92268371582031, 255.86260986328125, 260.1339416503906, 63.210655212402344, 58.73133850097656, 350.27435302734375, 44.60369873046875, 219.31707763671875, 193.3699951171875, 244.42758178710938, 0.978271484375, 282.038330078125, 70.02166748046875, 136.1962890625, 36.274322509765625, 26.737060546875, 122.69583129882812, -76.63784790039062, -110.50563049316406, 116.99887084960938, 279.31610107421875, -64.48460388183594, 46.58198547363281, 502.14617919921875, -30.602264404296875, -41.21893310546875, 126.1524658203125, 212.92999267578125, -2.908111572265625, -7.352266311645508, -102.84090423583984, -1.9510498046875, 91.32357788085938, 71.93731689453125, 116.40316772460938, 180.57623291015625, 250.5623779296875, 157.29103088378906, 203.77801513671875, 186.6899871826172, 59.30322265625, -17.238037109375, -298.0881042480469, 99.27676391601562, 285.2264404296875, -1.91961669921875, -8.054000854492188, 9.69329833984375, 68.94737243652344, -240.22943115234375, 105.77175903320312, 269.322021484375, 119.05840301513672, 15.840896606445312, 137.9520263671875, 77.91983032226562, 228.54852294921875, -78.12956237792969, 51.943878173828125, 203.58770751953125, -33.093482971191406, 294.59771728515625, 216.95077514648438, 56.890377044677734, -121.58879852294922, -39.74864196777344, 96.765380859375, -36.79212951660156, -14.882606506347656, 192.6920623779297, 46.768829345703125, 32.3677978515625, 60.10157775878906, 273.1093444824219, -5.464107513427734, 44.210296630859375, -195.88674926757812, -17.530624389648438, 146.96621704101562, 170.40512084960938, -1.6105499267578125, 127.38327026367188, 330.945068359375, -27.229827880859375, 77.25993347167969, 32.581146240234375, -9.07586669921875, 216.873779296875, 140.18264770507812, 14.632080078125, -6.61968994140625, 295.83209228515625, -165.84780883789062, 238.51144409179688, 20.65502166748047, 468.5750427246094, 277.9764404296875, 92.42138671875, -5.670166015625, -44.06672668457031, -112.76638793945312, 33.41290283203125, 156.61883544921875, 128.50186157226562, -18.538848876953125, 128.03628540039062, -5.692138671875, -0.484375, -140.18731689453125, 280.54534912109375, 29.07025146484375, 142.214599609375, -25.717559814453125, 265.7230224609375, 47.81622314453125, -32.400970458984375, 130.96533203125, -56.420074462890625, 4.743919372558594, 50.388427734375, 98.81968688964844, 67.78626251220703, -31.41279411315918, 113.44474792480469, -17.83087921142578, -68.77825927734375, 44.22459411621094, 185.47647094726562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000257.npy"}
{"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 85.8848876953125, "std": 129.6304473876953, "min": -287.14630126953125, "p10": -69.54698486328125, "median": 85.69732666015625, "p90": 253.8368408203125, "max": 502.28948974609375, "pos_frac": 0.7578125, "sample": [269.8819274902344, -4.014867782592773, 111.2532958984375, -14.661834716796875, 120.11753845214844, -139.97280883789062, 160.4190673828125, -69.26296997070312, 79.45928955078125, 502.28948974609375, 12.333892822265625, 157.20989990234375, 210.86737060546875, 42.060211181640625, 194.828857421875, 130.7927703857422, -15.153457641601562, 176.34707641601562, 128.20306396484375, 210.71499633789062, 387.5068359375, 259.196044921875, 20.331153869628906, 18.7459716796875, 269.2825927734375, -1.58856201171875, -287.14630126953125, 152.80914306640625, 143.57464599609375, -87.51484680175781, 197.79296875, 180.45132446289062, -72.59799194335938, 200.0186767578125, 280.0542907714844, 37.18280029296875, -4.217170715332031, -48.741851806640625, -10.09707260131836, 164.897216796875, 278.497802734375, 34.8282585144043, 5.816093444824219, 182.96400451660156, 100.43966674804688, 166.67062377929688, 154.02392578125, 84.6192626953125, 52.48175048828125, 122.1322021484375, 116.46826934814453, 4.546045303344727, 116.39556884765625, -15.371917724609375, -211.6632080078125, 147.1262969970703, 142.98675537109375, -2.8902587890625, -118.47613525390625, -77.94961547851562, -86.99916076660156, 56.47228240966797, 36.16949462890625, 62.9561767578125, 255.86288452148438, -19.75305938720703, 171.285400390625, 252.96853637695312, 5.572479248046875, 171.83547973632812, 111.57662963867188, 140.71624755859375, 9.814018249511719, 30.93939208984375, -49.024017333984375, 163.38861083984375, 337.2261047363281, -130.84417724609375, 191.65386962890625, 86.775390625, 4.61726188659668, -50.7313232421875, 125.66778564453125, 131.476318359375, -39.83082580566406, 18.746559143066406, 166.99075317382812, 166.8194580078125, 367.1800842285156, 108.54037475585938, -66.56092834472656, -48.37043762207031, 16.3521728515625, 5.198326110839844, 37.51600646972656, 12.6461181640625, 195.15130615234375, 168.92962646484375, -56.516319274902344, 11.995223999023438, -201.5130615234375, -17.502227783203125, 60.37262725830078, 21.26910400390625, 247.70404052734375, 59.635894775390625, 116.59843444824219, 110.74203491210938, 299.6499938964844, 191.92628479003906, 109.06817626953125, -116.19239807128906, 386.73992919921875, 91.67233276367188, 173.20635986328125, 316.01568603515625, 69.4726333618164, 65.29464721679688, 57.707122802734375, 105.25178527832031, -151.69161987304688, 155.6488037109375, 63.700950622558594, 169.69940185546875, 13.993896484375, 54.08726501464844, 189.20840454101562, -70.20968627929688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000258.npy"}
{"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 83.54182434082031, "std": 140.20999145507812, "min": -264.88739013671875, "p10": -67.63598480224609, "median": 79.46730422973633, "p90": 247.52655487060545, "max": 783.7145385742188, "pos_frac": 0.703125, "sample": [143.1337890625, 7.5261077880859375, 283.425537109375, 106.41285705566406, -77.82658386230469, 72.237548828125, 180.50282287597656, 196.65167236328125, 119.5322265625, -23.30963134765625, 110.877197265625, 151.66647338867188, 150.8749542236328, 312.8561096191406, 72.07233428955078, -41.98699951171875, 3.712432861328125, -110.30032348632812, 53.65887451171875, 123.65467834472656, -48.131439208984375, 232.39883422851562, 201.4459228515625, 35.50420379638672, 308.02764892578125, 60.20166015625, -29.711578369140625, 157.01138305664062, -112.25042724609375, -65.63121032714844, 2.3429641723632812, 35.29886245727539, -47.027130126953125, 154.70404052734375, 166.8480224609375, 4.740142822265625, -17.4105224609375, -0.33347320556640625, 59.462554931640625, 105.78880310058594, -66.406005859375, 94.793212890625, 160.9246826171875, 41.98638916015625, -26.094970703125, 12.5025634765625, 48.481529235839844, 244.38421630859375, -8.8575439453125, 101.34149169921875, 134.00144958496094, 56.02696228027344, 273.53887939453125, -200.2860107421875, 104.8973388671875, -81.9820556640625, 96.95321655273438, 289.5932312011719, 216.6607208251953, 25.548851013183594, 154.4315185546875, 66.09194946289062, -15.8734130859375, 86.69705963134766, -33.99449157714844, -27.628787994384766, 245.47732543945312, -207.47341918945312, -50.57882308959961, 230.1378173828125, 271.7633361816406, 68.6356201171875, -34.62672424316406, 195.02613830566406, 270.5474853515625, -0.097015380859375, 98.29806518554688, 0.0, -4.507843017578125, 198.50762939453125, 117.18867492675781, 148.1539306640625, -70.50593566894531, 110.35769653320312, 151.5806884765625, 118.37091064453125, 156.26187133789062, -182.99993896484375, -8.238166809082031, 147.0638427734375, -264.88739013671875, -37.225799560546875, 309.23565673828125, 783.7145385742188, 50.7052001953125, 14.9256591796875, 14.648971557617188, -47.429840087890625, 187.0389404296875, 91.04827880859375, 116.12552642822266, 163.91891479492188, -110.66207885742188, 397.7637634277344, -6.99798583984375, 1.532135009765625, 169.64614868164062, -83.54083251953125, 130.92892456054688, 212.02154541015625, 104.28057861328125, 252.30809020996094, 169.984619140625, 58.7305908203125, 28.31540870666504, -43.13426208496094, 95.06741333007812, 56.880157470703125, 130.97267150878906, -52.425537109375, 237.4792938232422, -126.0308837890625, -106.5806884765625, 263.33544921875, 0.7144622802734375, 437.80975341796875, 172.30633544921875, 166.10348510742188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000259.npy"}
{"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 34.1196174621582, "std": 124.68858337402344, "min": -265.34210205078125, "p10": -120.51822509765624, "median": 15.400140762329102, "p90": 210.80394287109374, "max": 347.6614074707031, "pos_frac": 0.609375, "sample": [129.78216552734375, -46.92767333984375, 189.2819061279297, -29.141708374023438, -150.37918090820312, 49.43151092529297, -41.005615234375, 266.1361999511719, -5.2334747314453125, 156.72616577148438, -86.74066162109375, 28.5830078125, 0.0, -33.615234375, 59.843658447265625, 39.243324279785156, 180.6440887451172, 292.12652587890625, -45.545738220214844, 28.65716552734375, 304.668701171875, -265.34210205078125, 15.18460464477539, 190.99066162109375, 259.8175048828125, -170.59112548828125, 22.137821197509766, -23.785648345947266, 77.050048828125, 141.71279907226562, 57.94427490234375, -134.24102783203125, 120.89389038085938, 8.948820114135742, 38.036376953125, 69.23851013183594, -174.10223388671875, -118.14019775390625, -185.95872497558594, 14.596023559570312, 38.626434326171875, 94.435791015625, 67.2120361328125, -61.1495361328125, -221.25863647460938, 107.9592514038086, -5.313549041748047, -55.33504867553711, -60.434661865234375, 13.160032272338867, -78.16192626953125, -87.835693359375, 221.76608276367188, 177.76260375976562, 11.131732940673828, 8.222686767578125, 347.6614074707031, -64.5819091796875, 263.4144287109375, 135.37652587890625, -94.49749755859375, 15.051250457763672, 215.3307342529297, -5.748931884765625, 43.94605255126953, -140.06942749023438, 43.84043502807617, 188.26263427734375, -12.3265380859375, 326.73712158203125, 67.67042541503906, 209.96044921875, -157.37924194335938, 9.5552978515625, 56.75285339355469, -34.234954833984375, -35.02657699584961, 205.939208984375, 1.34381103515625, -70.25543212890625, -182.51153564453125, -13.902130126953125, 346.5111999511719, -7.0592803955078125, 10.66143798828125, 111.25140380859375, -11.910171508789062, -4.0335540771484375, -48.63810729980469, 15.615676879882812, -61.543426513671875, 11.476654052734375, -61.13580322265625, 257.46636962890625, 63.802520751953125, 44.031219482421875, -46.653663635253906, 126.97914123535156, -24.24798583984375, 28.47674560546875, 212.7720947265625, 289.6310729980469, 96.65298461914062, 16.925949096679688, 30.534912109375, -139.24444580078125, 2.2866973876953125, 123.8399658203125, 35.118812561035156, 137.78179931640625, -17.204498291015625, 124.95794677734375, 107.62942504882812, -17.1773681640625, -126.06695556640625, 70.12149047851562, -222.69216918945312, -105.29501342773438, 13.729156494140625, 48.372589111328125, 141.50982666015625, -104.9427490234375, 24.6253662109375, 16.008411407470703, 142.91574096679688, -83.36540985107422, 69.23844909667969, 5.5750732421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000260.npy"}
{"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 67.44648742675781, "std": 142.527099609375, "min": -280.7972412109375, "p10": -103.94131774902343, "median": 50.55539131164551, "p90": 259.77687072753906, "max": 438.8138732910156, "pos_frac": 0.6875, "sample": [10.93646240234375, -280.7972412109375, -72.79203796386719, -113.02536010742188, 88.97158813476562, 58.213409423828125, 181.079833984375, -42.99983215332031, 80.14227294921875, 64.84770202636719, 124.42250061035156, 0.0, -45.01313781738281, -194.2039794921875, 31.608083724975586, 104.12326049804688, 24.778038024902344, 87.25018310546875, 283.6382141113281, 79.34112548828125, 255.2841033935547, -155.43133544921875, 200.9842987060547, -92.84600830078125, 122.69749450683594, 29.518775939941406, -121.31413269042969, 322.3760986328125, -49.43861389160156, -161.1704864501953, 110.30393981933594, 221.86509704589844, 384.72296142578125, 3.8512344360351562, 158.08700561523438, 275.21722412109375, 27.233787536621094, 3.4747314453125, -262.4915466308594, 25.78338623046875, 50.42898941040039, 50.681793212890625, 241.92645263671875, -56.5543212890625, 155.65850830078125, -49.30979919433594, -0.4022998809814453, 303.97613525390625, -72.69733428955078, 41.47835159301758, 105.11566162109375, -18.37060546875, 159.393798828125, 232.14810180664062, 257.3857421875, 50.0, 12.442558288574219, -156.99185180664062, 334.48956298828125, 153.24420166015625, 385.7109375, -28.614776611328125, 10.3759765625, -46.3878173828125, 122.11178588867188, 59.439781188964844, 40.04541015625, -48.12384796142578, 217.62889099121094, 47.99385070800781, 151.98419189453125, 349.34912109375, 120.6771240234375, 260.5636291503906, 187.9441375732422, 116.07952880859375, -100.04815673828125, 100.27960205078125, 19.335418701171875, 42.843406677246094, 136.07684326171875, -15.078353881835938, 80.20553588867188, -159.14337158203125, 228.42031860351562, 140.11837768554688, 137.4857177734375, 75.20449829101562, 215.87005615234375, -121.1531982421875, -45.02885437011719, -22.174102783203125, -68.5810546875, 2.526153564453125, -195.85589599609375, 259.7345275878906, 176.32186889648438, 245.09149169921875, 146.74948120117188, -135.3680419921875, -46.71510314941406, 3.1690635681152344, 30.34451675415039, 170.34906005859375, -25.983261108398438, 373.516845703125, -44.81278991699219, 312.7781066894531, 246.88314819335938, 13.756385803222656, 259.87567138671875, -43.26142883300781, 53.04237365722656, -199.11851501464844, 76.22039794921875, 2.643829345703125, -17.20062255859375, 124.9755859375, 26.845306396484375, 101.7071762084961, 91.80213928222656, 438.8138732910156, 80.88264465332031, -53.90167236328125, 21.735456466674805, -64.45021057128906, 65.55354309082031, -20.18407440185547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000261.npy"}
{"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 91.61766052246094, "std": 126.10179901123047, "min": -209.8546142578125, "p10": -58.76340789794922, "median": 76.93147277832031, "p90": 266.0780029296875, "max": 368.9700012207031, "pos_frac": 0.75, "sample": [271.7321472167969, 57.72564697265625, 141.62518310546875, 60.801177978515625, -25.66632080078125, 97.56084442138672, 199.4326171875, 342.99578857421875, 79.17984008789062, 198.2879638671875, 200.1771240234375, -14.8433837890625, 237.95901489257812, 46.65876770019531, -45.868560791015625, 0.10086441040039062, 326.1624450683594, 146.91793823242188, -58.96260070800781, 90.23302459716797, 125.1654052734375, -60.710693359375, -138.8039093017578, 261.6202392578125, 37.191650390625, 220.57479858398438, 29.625350952148438, 38.316497802734375, 70.44281005859375, 284.25927734375, 76.6734619140625, 24.813034057617188, 95.70314025878906, -47.614288330078125, 8.205520629882812, -138.51463317871094, 52.36102294921875, 34.99462890625, -65.04513549804688, 177.7158203125, 141.85342407226562, -7.446197509765625, 85.1578369140625, -159.96197509765625, 257.87615966796875, -29.949859619140625, 122.30520629882812, -12.89520263671875, 8.88629150390625, -131.09768676757812, 187.89764404296875, 264.6354064941406, 368.9700012207031, 101.1327896118164, -93.05413818359375, 40.99638366699219, 7.657806396484375, 109.8031005859375, -57.509521484375, 47.39872741699219, 210.97052001953125, -16.43804931640625, 77.18948364257812, 152.86685180664062, -66.64400482177734, 3.57354736328125, 184.2215576171875, 122.95118713378906, 289.6215515136719, 190.34112548828125, 28.385650634765625, -97.16455078125, 206.16018676757812, 37.769256591796875, 241.78985595703125, -47.562469482421875, 31.234466552734375, 261.97027587890625, 18.88232421875, 66.77792358398438, -131.78237915039062, -23.128196716308594, 0.0, 120.09700775146484, 295.45843505859375, 4.842987060546875, 127.27059936523438, 66.01957702636719, 12.417743682861328, -4.0920867919921875, 248.94692993164062, 98.54153442382812, 270.37322998046875, 45.51763916015625, 269.4440612792969, 247.18402099609375, -31.07562255859375, 19.14373779296875, -20.319839477539062, 284.6596374511719, -12.594593048095703, -55.33184814453125, 122.9271240234375, 80.42959594726562, -58.67803955078125, 178.2659912109375, 87.61676025390625, 71.3828125, 250.97659301757812, 251.91404724121094, 50.03520202636719, 86.27030944824219, 356.483154296875, 197.34219360351562, 197.67422485351562, 195.78924560546875, 335.1184997558594, -69.08866882324219, 248.85354614257812, 138.2289581298828, 96.09002685546875, -209.8546142578125, 148.39297485351562, 60.507598876953125, 148.28721618652344, 326.5760498046875, 30.47564697265625, -15.28497314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000262.npy"}
{"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 76.94769287109375, "std": 136.11199951171875, "min": -286.4401550292969, "p10": -84.42914428710937, "median": 73.19920349121094, "p90": 235.94460754394527, "max": 440.1968994140625, "pos_frac": 0.7265625, "sample": [334.96893310546875, 227.25094604492188, 246.79287719726562, 66.88677978515625, 13.355052947998047, 77.2713623046875, -48.46099853515625, 50.622955322265625, 126.55469512939453, 267.9718017578125, 5.463766098022461, 133.8282928466797, 93.85504150390625, 17.13568115234375, -71.977294921875, 31.966659545898438, -3.3513240814208984, -52.33624267578125, 75.95111083984375, -18.68048095703125, 219.85845947265625, 31.777191162109375, 33.593421936035156, 343.3985595703125, 2.11944580078125, 72.66696166992188, 231.29534912109375, 127.57869720458984, -53.15484619140625, -172.09112548828125, -286.4401550292969, 165.09536743164062, 248.36053466796875, 356.5777893066406, 178.9456787109375, -29.228134155273438, 150.78759765625, 46.537689208984375, -30.93463134765625, 165.2548065185547, 38.55528259277344, -4.4676513671875, 81.45208740234375, 440.1968994140625, 360.1045227050781, 9.713787078857422, 171.3385009765625, -96.02742004394531, 286.46533203125, 138.31179809570312, 178.40646362304688, -148.23568725585938, -104.02494812011719, 141.62521362304688, 179.85623168945312, 163.46536254882812, -202.330322265625, 82.94775390625, 376.7213439941406, 250.57154846191406, 0.0, 156.8369140625, -219.96517944335938, 17.632577896118164, 10.72100830078125, 30.6448974609375, 17.2579345703125, -21.507843017578125, 79.17672729492188, 84.88481140136719, 101.71684265136719, -53.94727325439453, -61.56244659423828, 77.0747299194336, 9.483612060546875, 73.7314453125, 133.16485595703125, 202.8123779296875, 133.408203125, -23.309326171875, -170.56602478027344, 60.55833435058594, -158.24578857421875, 64.78518676757812, -43.75517272949219, 2.9898948669433594, -140.0758514404297, 171.75634765625, 217.86480712890625, 14.117355346679688, -28.7236328125, 165.51876831054688, 15.890806198120117, -249.30307006835938, -6.538059234619141, -45.35638427734375, -91.25, 197.74725341796875, 85.12841796875, -189.574951171875, 38.61140441894531, -81.50592041015625, 0.0, 67.3070297241211, 150.76925659179688, 56.903228759765625, 399.0587158203125, 164.00567626953125, 209.86129760742188, 118.93370056152344, 195.85311889648438, 164.6431884765625, 173.3394012451172, 0.0, 97.6845703125, 196.21839904785156, 129.17726135253906, 202.3594512939453, 133.0560760498047, 65.32067108154297, 120.88917541503906, 212.08029174804688, 186.0238800048828, -19.115997314453125, 252.1458740234375, 60.77021789550781, 70.9505615234375, 141.02755737304688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000263.npy"}
{"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 94.46124267578125, "std": 142.98802185058594, "min": -298.722900390625, "p10": -58.256318664550776, "median": 85.05634307861328, "p90": 267.58436889648436, "max": 731.5313720703125, "pos_frac": 0.71875, "sample": [3.6019039154052734, 161.19558715820312, 179.46554565429688, 86.05560302734375, 192.30731201171875, 256.7430725097656, 110.3270034790039, 108.62290954589844, -17.623655319213867, -61.056427001953125, -172.0886688232422, 121.802734375, 112.78353881835938, 0.816497802734375, 138.71392822265625, 128.9721221923828, 225.73178100585938, 179.81834411621094, -106.9498519897461, -15.959503173828125, 130.83901977539062, 191.66900634765625, -1.408447265625, -19.754024505615234, -15.013961791992188, 156.93255615234375, 200.77685546875, 108.06095886230469, 24.19189453125, 268.8048400878906, -24.183319091796875, -66.60408020019531, 311.24749755859375, 731.5313720703125, 326.7640686035156, 0.0, -27.631317138671875, 413.24755859375, -2.2803802490234375, 32.68236541748047, 33.134765625, 188.90692138671875, -46.40216064453125, 149.30987548828125, 36.438995361328125, 442.11962890625, -298.722900390625, 267.0613098144531, -134.174072265625, 125.35385131835938, 16.06817626953125, -44.65997314453125, 257.66259765625, 98.7823486328125, 84.51495361328125, -140.94158935546875, 203.6943817138672, 202.84735107421875, 9.97740364074707, 72.18135070800781, 128.29931640625, 33.110877990722656, 120.62442779541016, 67.76103973388672, 17.36761474609375, 65.75662231445312, -31.8822021484375, 3.1993026733398438, 161.75091552734375, 260.26776123046875, 152.3928680419922, -6.0418853759765625, -187.9385986328125, 9.77547836303711, 38.448341369628906, -41.21333312988281, -23.340564727783203, 85.59773254394531, 143.61532592773438, 249.14202880859375, 245.19845581054688, -63.461212158203125, 40.584041595458984, 222.3418426513672, 7.33428955078125, 299.44964599609375, -71.35032653808594, 0.0, 47.041259765625, 99.43843078613281, -107.19699096679688, 72.48968505859375, 102.55853271484375, 199.70413208007812, 54.9637451171875, 144.44427490234375, -2.864154815673828, 81.79058837890625, 122.203125, 66.53573608398438, 315.25653076171875, -7.5010986328125, -66.8787841796875, -59.8349609375, -34.76959228515625, 147.23593139648438, 6.5382080078125, 92.14138793945312, 143.3353271484375, 303.6504821777344, 107.90898132324219, 335.48712158203125, 127.22418212890625, 178.42979431152344, 49.441375732421875, 451.36761474609375, -38.469390869140625, 54.554595947265625, 217.16732788085938, -36.18439483642578, -57.57975769042969, 148.63198852539062, 36.908111572265625, 125.02023315429688, 351.3944091796875, -12.843496322631836, 286.4147644042969, 222.78842163085938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000264.npy"}
{"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 100.4276123046875, "std": 156.75071716308594, "min": -298.2342529296875, "p10": -80.17026977539062, "median": 101.97520065307617, "p90": 296.9203826904297, "max": 585.6392211914062, "pos_frac": 0.6796875, "sample": [4.809408187866211, 172.64541625976562, 487.40911865234375, -71.10137939453125, 66.94903564453125, 0.06097412109375, 193.0492401123047, 57.093292236328125, 152.8078155517578, -0.6711883544921875, 209.97511291503906, 187.5614776611328, 208.52420043945312, -7.53961181640625, 224.04412841796875, -35.486083984375, 234.31344604492188, 585.6392211914062, 125.03543090820312, 147.55538940429688, 355.205078125, 458.9804382324219, 275.7521057128906, 127.5943832397461, 334.87274169921875, 0.56256103515625, 248.83578491210938, 218.36114501953125, 235.05361938476562, 34.15362548828125, 31.34062957763672, 165.2623291015625, 270.49798583984375, 208.39251708984375, 301.82147216796875, 134.91375732421875, 107.51864624023438, 62.71685791015625, 134.3140869140625, 219.01889038085938, -3.58056640625, 77.42828369140625, -97.4300537109375, 198.0733642578125, -9.392623901367188, 104.08787536621094, 21.6710205078125, -43.748779296875, 6.096893310546875, 133.53079223632812, -5.5135498046875, -126.86567687988281, -105.36279296875, -1.389312744140625, -3.7183990478515625, 290.56451416015625, 188.32281494140625, 121.54248046875, -41.68206787109375, -63.374786376953125, 99.8625259399414, 172.70709228515625, -104.931396484375, 235.4539794921875, -86.1136474609375, -23.58551025390625, 240.2481689453125, 176.66598510742188, 195.6564178466797, -37.95301055908203, -48.051483154296875, 267.577880859375, 372.7071533203125, 85.04954528808594, 0.0, -25.127948760986328, -194.41647338867188, -237.1395263671875, 359.9573974609375, 392.3309020996094, 311.28192138671875, 68.00921630859375, 58.81077575683594, 37.80647277832031, 198.3567657470703, 312.1913146972656, 0.0, -14.838310241699219, -298.2342529296875, 187.07846069335938, 158.011474609375, -77.62310791015625, -57.251708984375, 17.773963928222656, 294.8199157714844, 35.47058868408203, 69.256591796875, 293.5684814453125, -45.19000244140625, -109.31997680664062, -21.977783203125, 109.41073608398438, 127.28670501708984, 133.10488891601562, -67.47439575195312, -127.87684631347656, -215.4958953857422, 240.9815673828125, 172.709716796875, -160.08499145507812, -24.703994750976562, 130.4818572998047, 139.95245361328125, -7.645454406738281, -61.72383117675781, 349.97796630859375, 34.20657730102539, 256.37408447265625, 130.0831756591797, 165.62449645996094, 308.8143310546875, -51.95404052734375, 96.19877624511719, 66.86492919921875, 32.360511779785156, -154.86981201171875, 181.2407989501953, 284.88848876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000265.npy"}
{"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 98.95314025878906, "std": 135.7544403076172, "min": -283.87518310546875, "p10": -33.177059173583984, "median": 78.96714782714844, "p90": 281.47138671875, "max": 546.997802734375, "pos_frac": 0.7734375, "sample": [79.25140380859375, 160.25628662109375, 281.3377380371094, 161.17623901367188, -24.314300537109375, -123.78189086914062, 126.04173278808594, -18.58514404296875, -7.520214080810547, 20.239822387695312, -11.386566162109375, 154.9892578125, -13.169998168945312, 53.6878662109375, 157.14466857910156, -15.870574951171875, -41.359619140625, -48.75556564331055, 169.1275634765625, -283.87518310546875, 301.986572265625, 62.32184600830078, 58.543426513671875, 41.51239013671875, 99.673583984375, 55.809295654296875, -39.572296142578125, 226.40435791015625, 546.997802734375, 246.97470092773438, 110.1532974243164, 349.64501953125, 87.39230346679688, 92.12948608398438, 12.443695068359375, 354.1627197265625, -133.353759765625, 56.39275360107422, 72.79295349121094, 221.52035522460938, 75.43238830566406, -2.9852294921875, 12.748123168945312, 112.08218383789062, 130.80465698242188, -19.392913818359375, 42.9390869140625, 424.23992919921875, 27.4132080078125, 91.81625366210938, 122.3157958984375, 166.05404663085938, -161.37245178222656, -153.07545471191406, -5.430294036865234, 3.13397216796875, -6.953878402709961, 103.18939208984375, 50.4306640625, 119.85531616210938, 182.50177001953125, 99.03047180175781, 43.046234130859375, 9.415252685546875, 5.538505554199219, 352.02337646484375, 249.92474365234375, 177.9082489013672, 21.962738037109375, 92.911865234375, 263.99395751953125, 297.27850341796875, 123.7872314453125, 85.7615966796875, 152.5364990234375, 444.7950439453125, 69.25628662109375, 4.84063720703125, 3.7626953125, -32.241493225097656, -203.51687622070312, 37.59759521484375, 62.709503173828125, 383.64508056640625, 111.78631591796875, 198.54153442382812, 78.68289184570312, 124.89138793945312, 385.30352783203125, 75.11129760742188, 170.84609985351562, 54.1007080078125, -30.284666061401367, 9.8609619140625, 106.23532104492188, 358.0154724121094, 182.14093017578125, 193.65414428710938, 48.96368408203125, -16.942764282226562, 0.1805419921875, 123.33486938476562, 146.937744140625, 135.97979736328125, 44.703758239746094, -2.283203125, -9.950054168701172, -40.922630310058594, 51.4630126953125, -46.4188232421875, 173.6104736328125, 281.7832336425781, 411.69024658203125, 30.834732055664062, 198.88925170898438, -35.36004638671875, 227.9244384765625, 31.12281036376953, 42.54153823852539, 260.8681945800781, 242.9033203125, -44.81282043457031, -19.459915161132812, 82.3904037475586, 182.1821746826172, 108.45857238769531, 104.9498291015625, 271.2752380371094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000266.npy"}
{"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 74.13900756835938, "std": 150.2488250732422, "min": -353.39105224609375, "p10": -92.4622817993164, "median": 68.41185760498047, "p90": 269.63832702636716, "max": 601.3193969726562, "pos_frac": 0.7109375, "sample": [11.240251541137695, 113.85203552246094, 184.90953063964844, 54.733642578125, 601.3193969726562, 141.4014434814453, -23.902732849121094, 66.31904602050781, 172.649658203125, -37.56866455078125, -14.149642944335938, 95.405517578125, -23.3294677734375, 201.59323120117188, -7.77470588684082, 30.1068115234375, 58.40111541748047, 135.9217529296875, 103.972900390625, 147.704833984375, -85.22694396972656, 275.0205078125, -21.77911376953125, -45.441986083984375, -19.719375610351562, -118.39736938476562, 83.23944091796875, 179.0369873046875, 147.451416015625, -331.5889892578125, 36.496826171875, 22.034439086914062, 77.49188232421875, -17.869537353515625, 77.62901306152344, 135.85467529296875, -3.380298614501953, 269.4197998046875, 251.09988403320312, 150.8934783935547, -63.96882629394531, 315.40667724609375, 24.703659057617188, 153.83749389648438, 89.51174926757812, 393.4918212890625, 94.91693115234375, 27.91815185546875, 361.30712890625, 138.4555206298828, -49.55868148803711, -92.761962890625, 99.53012084960938, 145.74964904785156, 44.954742431640625, 363.76123046875, 215.2703094482422, 153.06488037109375, 270.1482238769531, 140.46078491210938, 318.794921875, 81.692138671875, -353.39105224609375, 16.617828369140625, 143.76426696777344, 62.990814208984375, 102.9794921875, -154.29037475585938, 162.59262084960938, 47.22943115234375, 70.50466918945312, 131.7022705078125, 275.5960998535156, -277.14385986328125, -136.34231567382812, 195.67486572265625, 297.70068359375, 274.2096862792969, -23.534175872802734, 7.476072311401367, -139.17288208007812, -24.74517822265625, -47.751190185546875, 135.44940185546875, 12.162689208984375, 206.07119750976562, -186.9964599609375, 240.69313049316406, 61.41375732421875, 56.051422119140625, 35.54908752441406, -29.232147216796875, 17.722198486328125, -57.22760009765625, 61.73187255859375, 207.564208984375, 199.1100311279297, 93.43502044677734, 292.70220947265625, -43.129241943359375, 150.02987670898438, 105.58984375, -47.144622802734375, 30.54168701171875, -3.3250350952148438, 32.52203369140625, 126.8956298828125, 100.78971862792969, 16.44750213623047, 591.1407470703125, -192.63812255859375, -92.33384704589844, 70.97245788574219, 106.37725830078125, -20.2098388671875, 0.2369842529296875, 221.3857421875, 143.93338012695312, -93.60208129882812, -65.96273803710938, 170.48822021484375, 95.20486450195312, 39.938232421875, 3.33648681640625, 86.75067138671875, -238.9605712890625, -120.64718627929688, 34.54052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000267.npy"}
{"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 93.08436584472656, "std": 140.3735809326172, "min": -230.55923461914062, "p10": -61.199316406249984, "median": 82.2119140625, "p90": 268.60500183105466, "max": 477.96142578125, "pos_frac": 0.7265625, "sample": [179.00975036621094, 17.39349365234375, 9.296295166015625, 103.0113525390625, -26.955078125, -45.644500732421875, 44.259891510009766, -15.580108642578125, -9.148468017578125, -127.00189971923828, -11.332984924316406, 352.9902648925781, 105.68228149414062, 173.7371826171875, 42.6470947265625, 71.9183349609375, 477.96142578125, -9.394783020019531, 11.3968505859375, 139.3055419921875, 237.6714324951172, 135.46026611328125, 203.95291137695312, -195.32644653320312, 450.37158203125, 126.6026611328125, 8.428316116333008, 105.78155517578125, 162.99752807617188, 45.02216339111328, 61.585227966308594, 167.9110107421875, 154.20257568359375, 21.934280395507812, -74.60133361816406, -13.66278076171875, -36.575233459472656, -108.11453247070312, 24.598281860351562, 267.1845703125, 203.51492309570312, 72.47322082519531, 87.0594482421875, 162.2667236328125, 77.09246826171875, 27.7220458984375, 223.35093688964844, -139.91195678710938, -109.72509765625, 195.01248168945312, 217.22576904296875, 79.87026977539062, 233.98489379882812, 352.1961669921875, -28.74005126953125, 28.123626708984375, 368.8397216796875, -106.67362976074219, 86.07684326171875, -230.55923461914062, 295.7316589355469, 88.37504577636719, 252.9217071533203, -55.10906982421875, 4.743499755859375, 180.172119140625, 205.56040954589844, 150.37783813476562, -41.05552673339844, -36.07562255859375, 301.54058837890625, 239.53700256347656, 291.1634521484375, 109.44723510742188, 61.837066650390625, -39.02789306640625, 89.13140869140625, 209.31494140625, 242.19195556640625, 71.477783203125, 49.398223876953125, 182.33892822265625, 212.2125244140625, -190.65919494628906, 82.4114990234375, 91.4944839477539, -186.10443115234375, 321.611083984375, 127.80172729492188, 62.4864501953125, 181.6632080078125, 271.9193420410156, 310.8563537597656, 87.14593505859375, 2.99969482421875, 232.27581787109375, 114.6063232421875, 18.759231567382812, 110.03900146484375, 256.91815185546875, 101.13037109375, -2.6055679321289062, 82.0123291015625, 25.330764770507812, 10.214828491210938, 369.83111572265625, -25.0906982421875, -70.4608154296875, -4.12652587890625, -11.2972412109375, 247.48529052734375, -1.6944580078125, 202.66073608398438, 66.62992858886719, 9.98052978515625, 103.15385437011719, -148.81272888183594, -57.2301025390625, 394.568359375, 131.0033721923828, -7.561946868896484, -189.80023193359375, 211.23605346679688, 205.48341369628906, 57.93701171875, 241.60565185546875, -17.3905029296875, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000268.npy"}
{"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 66.193603515625, "std": 141.08511352539062, "min": -410.9677734375, "p10": -112.44222564697266, "median": 52.321218490600586, "p90": 246.17959899902343, "max": 453.03973388671875, "pos_frac": 0.6875, "sample": [-234.28189086914062, 96.7890396118164, 111.17413330078125, -54.29832458496094, 158.35166931152344, -22.913787841796875, -85.44673156738281, 168.01564025878906, 123.91561889648438, 59.69775390625, 110.29447937011719, 61.556976318359375, 46.54252624511719, 185.49937438964844, -20.5738525390625, 178.636962890625, 15.474380493164062, -5.536163330078125, 289.8253479003906, 133.06338500976562, 271.8295593261719, 142.98367309570312, 65.58976745605469, 303.97491455078125, 132.41412353515625, -80.03777313232422, 234.2442626953125, 188.7283935546875, 37.9381103515625, 155.7913818359375, 49.49786376953125, 296.6060485839844, 232.70147705078125, -117.05078125, 116.39425659179688, 0.0, 156.48101806640625, -50.369873046875, 49.697166442871094, 245.9959716796875, -150.32839965820312, 65.69955444335938, 316.2784118652344, 336.11614990234375, 30.32989501953125, -28.8834171295166, 151.59744262695312, 0.0, -111.99476623535156, 37.57537841796875, 151.20655822753906, -29.405548095703125, 300.1686706542969, -213.96951293945312, -26.407867431640625, 323.37841796875, 177.68902587890625, 55.00867462158203, -156.22998046875, -7.642421722412109, 343.693115234375, 68.92636108398438, 99.68643188476562, -52.8582763671875, 184.3109893798828, 31.777740478515625, 2.21661376953125, -33.07664489746094, 236.78515625, 34.50543975830078, 115.92636108398438, 9.64300537109375, 0.0, 453.03973388671875, 45.36094665527344, 91.18307495117188, -209.39553833007812, -52.70004653930664, 346.57177734375, 165.11978149414062, 14.02679443359375, -81.21807861328125, 197.7210693359375, 50.055110931396484, 1.80499267578125, 73.49765014648438, 75.63583374023438, 101.66375732421875, 57.15496826171875, 48.41624450683594, 32.79901123046875, 154.25933837890625, 0.253509521484375, -127.21176147460938, 107.02108001708984, -113.48629760742188, -98.34358215332031, 229.7678680419922, -62.4635009765625, 3.368194580078125, 186.89154052734375, 246.60806274414062, -7.500335693359375, -181.9262237548828, -2.006481170654297, 23.518844604492188, -28.405532836914062, 166.79586791992188, -122.787109375, 234.309814453125, -52.018798828125, 186.6221923828125, 45.9302978515625, 54.58732604980469, 102.00480651855469, -123.31344604492188, -211.65341186523438, -42.86079406738281, 4.127838134765625, 18.319236755371094, 173.025146484375, 270.6826477050781, 25.862716674804688, -410.9677734375, -60.3883056640625, 58.96430969238281, 238.807373046875, 164.73077392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000269.npy"}
{"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 84.95326232910156, "std": 164.63746643066406, "min": -322.1661376953125, "p10": -98.05679702758788, "median": 86.6656494140625, "p90": 272.41156005859375, "max": 680.8733520507812, "pos_frac": 0.7109375, "sample": [-322.1661376953125, 141.71823120117188, 176.2242431640625, -0.09447479248046875, -104.00029754638672, 469.994384765625, -188.8209228515625, 154.85467529296875, 175.97885131835938, -116.97067260742188, 220.89022827148438, 224.94512939453125, 98.21295166015625, 35.54277038574219, 191.53404235839844, 99.93695068359375, 1.228607177734375, 268.4808654785156, -26.225746154785156, -82.0179443359375, -95.50958251953125, -59.211387634277344, 141.13076782226562, 74.38754272460938, -25.521072387695312, -9.16424560546875, 211.07717895507812, 172.43719482421875, -55.21661376953125, 11.468786239624023, 25.33831787109375, 315.0272216796875, 86.75933837890625, 4.155803680419922, 241.85037231445312, 46.510406494140625, -263.63177490234375, 120.68878173828125, -193.16046142578125, 248.61009216308594, 16.406219482421875, 109.06494140625, 69.95857238769531, 16.179763793945312, 225.5517578125, 20.7469482421875, -22.013011932373047, 507.1102294921875, 117.6089859008789, -113.89675903320312, -246.99688720703125, 371.11865234375, 130.72564697265625, 299.2813720703125, -64.79061889648438, 74.10438537597656, 132.75030517578125, -120.515380859375, 116.0972900390625, 57.54969787597656, -12.20703125, 38.68669128417969, 41.0821533203125, 10.086067199707031, 325.6278991699219, 119.04132080078125, 277.1759948730469, 163.70216369628906, 167.070068359375, -41.689727783203125, 5.895538330078125, 131.09637451171875, -257.7983093261719, 220.68096923828125, -54.1380615234375, 300.01513671875, 247.8553924560547, 12.190942764282227, 318.3673400878906, -283.45391845703125, 103.51742553710938, 36.480133056640625, 18.6512451171875, 130.08731079101562, -27.763526916503906, 85.29988098144531, 137.04611206054688, 83.18449401855469, 57.614341735839844, -71.20880889892578, -56.98980712890625, -58.6982421875, 341.42584228515625, 291.721923828125, -47.77287292480469, 123.54144287109375, 270.3696594238281, 169.27886962890625, 159.62591552734375, -29.91864013671875, 108.05014038085938, 240.1578369140625, -8.2977294921875, 167.57318115234375, -4.1642303466796875, -246.427490234375, 81.650390625, -24.6458740234375, 183.90011596679688, 8.873138427734375, -3.18560791015625, 269.76397705078125, 91.5843505859375, 185.10101318359375, 199.72821044921875, 203.10211181640625, 179.56585693359375, 102.42416381835938, 33.55029296875, 527.157958984375, -74.39410400390625, 680.8733520507812, 138.57977294921875, 188.6375274658203, 123.1859130859375, 86.57196044921875, 109.28369140625, -232.2764892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000270.npy"}
{"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 78.04963684082031, "std": 139.6155242919922, "min": -281.167236328125, "p10": -76.65537109374999, "median": 61.16162300109863, "p90": 268.7468719482422, "max": 502.4595031738281, "pos_frac": 0.7265625, "sample": [49.1689453125, -111.98455810546875, 18.19293212890625, 19.0489501953125, 49.31683349609375, -48.47621154785156, 17.954734802246094, -35.14268493652344, 75.35540771484375, 17.981399536132812, 92.16986083984375, 65.8302001953125, 166.5499267578125, 161.39735412597656, -123.40908813476562, 304.6546325683594, 151.03125, 153.466796875, 82.44444274902344, -85.44158935546875, 54.64488983154297, 56.51371765136719, 277.3537902832031, 119.70658874511719, 12.011093139648438, -32.35350799560547, -281.167236328125, 82.74111938476562, 69.39796447753906, 19.856903076171875, 301.7726745605469, 301.7650146484375, -4.292724609375, 148.33526611328125, 53.18463897705078, 38.307708740234375, 113.8958740234375, 75.22545623779297, -14.796966552734375, -74.79301452636719, 404.2508850097656, 47.940887451171875, 10.844451904296875, 196.55072021484375, 192.08551025390625, -18.825836181640625, 92.549560546875, -5.001983642578125, 114.93449401855469, 137.30935668945312, 86.45768737792969, 191.40213012695312, 139.56590270996094, 83.58163452148438, 502.4595031738281, 303.5713806152344, -11.927581787109375, 176.67263793945312, 31.189208984375, 199.98196411132812, 68.28440856933594, -20.234405517578125, -12.849090576171875, 399.8161315917969, 110.32295989990234, -53.198944091796875, -36.741790771484375, 412.37799072265625, 51.718597412109375, -5.11175537109375, 194.3020477294922, 228.23641967773438, -56.8123779296875, 84.62826538085938, 9.627471923828125, 65.28158569335938, -90.46151733398438, 68.09088134765625, 54.441986083984375, 24.621261596679688, -50.384307861328125, 132.48184204101562, 111.98721313476562, -151.98092651367188, 401.4443359375, 39.706787109375, -116.8007583618164, -27.972442626953125, 39.033897399902344, -51.459075927734375, -43.15306091308594, 259.2901306152344, 235.69140625, -158.70498657226562, -14.108383178710938, 120.55130004882812, 142.3328857421875, 210.4232940673828, 42.10813903808594, 222.26112365722656, 110.14028930664062, 94.45716857910156, -247.83792114257812, 268.95184326171875, 268.6590270996094, -34.13465881347656, -122.25100708007812, 169.04299926757812, 84.96438598632812, 167.0711669921875, 95.5257568359375, 44.46954345703125, 141.7916259765625, 13.658950805664062, -25.22271728515625, 36.66654968261719, 325.7948913574219, 216.75775146484375, -263.4781494140625, -147.503662109375, 57.04166030883789, -81.00086975097656, 443.25543212890625, 157.528564453125, 10.725200653076172, 95.66061401367188, 36.82768249511719, 20.691375732421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000271.npy"}
{"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 85.47587585449219, "std": 161.02877807617188, "min": -403.045654296875, "p10": -78.14648132324218, "median": 89.52149963378906, "p90": 263.75471191406245, "max": 706.529052734375, "pos_frac": 0.6875, "sample": [-33.21003723144531, 0.0, 157.47314453125, 50.88551330566406, -403.045654296875, 104.44183349609375, 513.9541015625, -32.9945068359375, 287.80633544921875, -78.11184692382812, -9.937973022460938, -4.996747970581055, 177.98855590820312, 226.98919677734375, 96.2847900390625, 310.02154541015625, 251.5048828125, 4.13763427734375, 97.49655151367188, 31.07811737060547, 366.6788024902344, 34.69925308227539, 10.8648681640625, -104.28253173828125, 144.26112365722656, -11.8817138671875, 27.004215240478516, 221.60745239257812, 0.0117645263671875, 11.647918701171875, 115.91070556640625, 546.136474609375, 164.67987060546875, -20.294471740722656, 706.529052734375, 170.53521728515625, 246.7877197265625, 184.818359375, 60.7332763671875, -7.894493103027344, 97.65892791748047, -200.89544677734375, -27.5714111328125, 224.78228759765625, 84.23580932617188, 138.5941162109375, -24.533721923828125, -73.06982421875, 175.4573974609375, -6.530363082885742, 398.83935546875, 208.20428466796875, -29.078472137451172, 123.28005981445312, 213.5416717529297, 329.7564697265625, -186.5786895751953, 111.6942138671875, 176.9329833984375, 20.798526763916016, -6.516042709350586, 74.41598510742188, 73.4075927734375, -54.659271240234375, 26.871559143066406, -3.323977470397949, 301.69580078125, 122.99945068359375, 153.04733276367188, 40.07647705078125, 357.1805725097656, -109.36196899414062, 0.880889892578125, 101.84124755859375, -380.8900146484375, -35.352294921875, 94.80718994140625, 9.633148193359375, -261.87640380859375, 223.532958984375, 245.84765625, 219.33248901367188, 81.63562774658203, 285.23480224609375, 9.2762451171875, -220.23471069335938, 28.420345306396484, 122.40087890625, 96.52447509765625, -1.3787078857421875, -27.122215270996094, 120.52679443359375, 117.50631713867188, -134.473388671875, -124.08507537841797, 149.976318359375, 132.60769653320312, 258.7364501953125, -12.57989501953125, 13.580812454223633, 0.0, -58.0992431640625, 153.0591278076172, -78.227294921875, -10.757293701171875, -29.421630859375, 194.99240112304688, 59.75348663330078, 116.7588119506836, 126.52407836914062, 140.1435546875, 275.4639892578125, -156.797607421875, -0.6675567626953125, 129.50726318359375, -131.87344360351562, 95.03706359863281, 148.50656127929688, -19.27998161315918, 13.440185546875, 333.2669372558594, 198.6104736328125, 155.87158203125, 38.6829833984375, 147.55123901367188, 243.0947265625, 205.3885498046875, 188.40269470214844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000272.npy"}
{"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 86.65823364257812, "std": 155.9329071044922, "min": -239.44107055664062, "p10": -94.8839790344238, "median": 85.60308837890625, "p90": 282.08828430175777, "max": 557.3064575195312, "pos_frac": 0.6875, "sample": [117.30142211914062, 334.97393798828125, 145.98974609375, 154.45216369628906, 235.9754180908203, -17.262474060058594, 479.2596435546875, 33.20166015625, 216.74783325195312, 82.9706802368164, -87.28994750976562, 20.482040405273438, -133.29348754882812, 137.8191680908203, 275.03375244140625, -200.57760620117188, 169.076416015625, 389.2666015625, 13.88201904296875, 29.11444091796875, -86.9766845703125, 13.172937393188477, 223.85858154296875, 59.739898681640625, 230.72787475585938, 40.1851806640625, -239.44107055664062, 128.6915740966797, 276.8674621582031, 0.9898529052734375, 21.97087860107422, -208.91409301757812, -55.821380615234375, 278.2569885253906, -112.60338592529297, -78.11151123046875, 40.485504150390625, -201.1981201171875, -40.50285339355469, 153.68841552734375, -62.44011688232422, -136.355712890625, -29.162818908691406, 114.3536376953125, 115.3433837890625, 87.62652587890625, 92.2188949584961, -20.95222282409668, -64.44840240478516, 18.9681396484375, 92.18392944335938, -78.73080444335938, 407.8840637207031, 49.57550048828125, 118.89332580566406, -7.012054443359375, -131.98867797851562, -12.698883056640625, 94.56219482421875, 119.673828125, -170.4581298828125, 315.36981201171875, 48.40949249267578, 199.05841064453125, -13.855682373046875, 11.0755615234375, 134.680908203125, -60.55584716796875, 142.0547637939453, 17.463775634765625, 287.5630798339844, -10.25726318359375, -32.500457763671875, 18.443572998046875, 202.6282958984375, 297.7707214355469, -34.37896728515625, 196.19683837890625, 163.61782836914062, 248.43222045898438, -71.55770874023438, 212.02325439453125, 187.47076416015625, -62.44793701171875, 117.46685791015625, -21.24481964111328, 154.89578247070312, 97.47225189208984, 416.6136779785156, 166.17880249023438, 94.38885498046875, -38.344398498535156, 33.08403015136719, 205.15264892578125, -199.34765625, 105.16705322265625, -130.54498291015625, 149.9858856201172, 346.230224609375, -50.60479736328125, 461.5769348144531, 110.59596252441406, 182.37567138671875, 296.7165832519531, 279.741943359375, -8.61245346069336, 58.546295166015625, 137.94949340820312, 186.00338745117188, -16.357086181640625, 28.470199584960938, 28.91583251953125, -4.5586395263671875, 557.3064575195312, 114.85551452636719, 192.28070068359375, -48.35540771484375, 276.7276306152344, 134.7965087890625, 3.8766021728515625, 50.873619079589844, 224.67596435546875, 83.57965087890625, 245.01309204101562, 306.0799560546875, -115.37298583984375, 248.6497802734375, -206.57669067382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000273.npy"}
{"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 104.03874206542969, "std": 152.0389862060547, "min": -333.3472595214844, "p10": -65.41287498474118, "median": 74.82292938232422, "p90": 286.5829162597656, "max": 523.4346923828125, "pos_frac": 0.7890625, "sample": [-8.974590301513672, 195.6453857421875, -110.80462646484375, 66.94374084472656, -57.244632720947266, 96.03448486328125, 176.56640625, 387.55889892578125, 285.79766845703125, 6.039436340332031, 158.33053588867188, 155.088134765625, 455.937255859375, 72.82939147949219, -6.2423095703125, 127.91984558105469, 337.7677001953125, 502.17535400390625, -130.69528198242188, 2.5868988037109375, 66.11154174804688, 279.6732482910156, 335.231201171875, 288.4151611328125, 45.30760192871094, 60.31439208984375, -163.63470458984375, 27.411590576171875, 4.7920074462890625, 69.23312377929688, 17.45452880859375, 167.6370849609375, 178.60499572753906, 191.2442626953125, 210.16177368164062, 299.38140869140625, 86.17507934570312, 92.22978210449219, 32.88555908203125, 213.85064697265625, 80.4405517578125, 16.127771377563477, -37.064239501953125, -16.75257110595703, -13.647705078125, 79.59353637695312, 213.3275146484375, -45.501068115234375, 42.585052490234375, 230.18955993652344, 9.0712890625, 75.853759765625, 109.38731384277344, -156.56698608398438, -109.69506072998047, 63.36549377441406, 53.926116943359375, 213.46237182617188, 2.1561622619628906, 222.6123046875, -9.588489532470703, 31.986103057861328, 107.31755065917969, 71.03321838378906, -279.8290100097656, 49.44023132324219, -6.998321533203125, -12.876100540161133, 493.35858154296875, 369.8407287597656, 59.34681701660156, -333.3472595214844, 218.7037353515625, 262.8076477050781, 140.94488525390625, 75.15863037109375, -93.54766845703125, 305.1881103515625, 523.4346923828125, 219.91726684570312, 384.3779296875, 109.10995483398438, 55.1439208984375, 250.0927734375, -115.81344604492188, 270.08746337890625, 268.1062316894531, 160.03314208984375, 238.31729125976562, 242.25094604492188, 150.617431640625, 73.48771667480469, -143.0895538330078, 120.07232666015625, 254.1715087890625, 141.66197204589844, 120.21646118164062, 395.5343933105469, 255.758056640625, 3.865264892578125, -13.07196044921875, 20.97247314453125, 51.78729248046875, -1.82171630859375, -84.47210693359375, 37.41025924682617, 14.358749389648438, 240.75686645507812, 55.389923095703125, 164.25030517578125, 80.744140625, 185.77481079101562, 27.6546630859375, 35.92732238769531, 17.7467041015625, -9.17938232421875, 83.44720458984375, -27.538909912109375, -115.25552368164062, 66.97976684570312, 12.191963195800781, -156.94366455078125, 146.68222045898438, 8.66856575012207, 74.48722839355469, 200.63009643554688, 282.7344970703125, 241.7430419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000274.npy"}
{"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 108.05087280273438, "std": 146.73318481445312, "min": -258.7895202636719, "p10": -46.1420539855957, "median": 74.85879898071289, "p90": 290.30410156249997, "max": 528.0047607421875, "pos_frac": 0.765625, "sample": [225.29803466796875, -13.37542724609375, -34.652645111083984, -20.55316162109375, 146.88775634765625, 19.055450439453125, 516.9923095703125, 28.2227783203125, 130.19076538085938, 0.53350830078125, 39.3839111328125, 454.4640197753906, -12.29229736328125, 306.7520446777344, 0.0, 39.46771240234375, 134.71173095703125, 45.65437316894531, 11.243818283081055, 9.1898193359375, 0.0, 0.0, 174.9846954345703, 213.42547607421875, 29.116195678710938, -4.673126220703125, 162.36578369140625, 146.8004150390625, 9.877197265625, -28.737533569335938, 378.91326904296875, 46.65393829345703, 129.28652954101562, 147.16607666015625, 262.420166015625, 173.53472900390625, 29.204622268676758, 12.415817260742188, -108.99746704101562, 116.1388168334961, 473.84991455078125, 70.5015869140625, 247.87181091308594, 187.10894775390625, -126.54220581054688, -10.312068939208984, -78.15188598632812, 220.51382446289062, 248.76283264160156, -107.19815063476562, 158.27737426757812, -258.7895202636719, 4.5871124267578125, -17.22552490234375, 60.24561309814453, 79.033447265625, 4.2960052490234375, 21.3824462890625, 479.1319885253906, 0.0, 57.26220703125, -85.36053466796875, 91.48477172851562, 528.0047607421875, 76.40303039550781, 350.8323669433594, 317.8372802734375, 105.79730224609375, -69.22969055175781, 0.0, -22.836082458496094, 220.11578369140625, 77.34878540039062, 243.33261108398438, -46.13811492919922, 210.41070556640625, 452.3436279296875, 299.44622802734375, 198.75942993164062, 5.862663269042969, 138.64561462402344, 215.66552734375, 105.03567504882812, 284.45758056640625, 73.31456756591797, 7.559328079223633, 206.96575927734375, 286.38604736328125, -48.64073181152344, 183.87765502929688, -43.803314208984375, 51.42378234863281, 56.233978271484375, 98.50921630859375, 269.7691955566406, 4.35809326171875, 11.682403564453125, -68.37297058105469, 308.345947265625, -46.1512451171875, 5.2924346923828125, 263.91259765625, 93.7562255859375, 26.956283569335938, 433.6337890625, 190.00125122070312, -7.2681732177734375, 122.45516204833984, 33.750885009765625, -107.223876953125, 185.85305786132812, -64.86177062988281, 202.33944702148438, 51.67913818359375, 118.92935180664062, 5.612125396728516, 112.872802734375, 178.30081176757812, 260.8983154296875, 243.69920349121094, 8.3836669921875, 263.8679504394531, 53.307098388671875, -68.29544067382812, 184.175048828125, 0.76983642578125, 183.97369384765625, 172.38800048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000275.npy"}
{"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 102.97511291503906, "std": 155.9966278076172, "min": -446.1894226074219, "p10": -33.05928497314453, "median": 69.99708557128906, "p90": 296.37336425781245, "max": 623.8349609375, "pos_frac": 0.7890625, "sample": [537.861328125, 205.14932250976562, 173.74334716796875, -25.7828369140625, 154.56524658203125, 44.46484375, 84.62075805664062, 15.116241455078125, 94.5267333984375, 16.07991600036621, 166.26382446289062, -68.48941040039062, 40.007843017578125, -20.322509765625, -170.06109619140625, 90.71749877929688, 155.3846435546875, 17.605926513671875, -20.52618408203125, 230.29080200195312, 191.3966064453125, -35.70960998535156, 1.9563102722167969, 236.76223754882812, 152.72921752929688, -144.8111572265625, -391.06561279296875, -18.85247802734375, -60.59716796875, 152.05130004882812, 80.31759643554688, 10.268585205078125, 623.8349609375, 158.88153076171875, 50.1705322265625, 39.06402587890625, 132.3780517578125, 3.6449642181396484, 97.01763916015625, 123.08929443359375, 56.1375732421875, 42.279052734375, 229.78787231445312, 5.60968017578125, 59.24064636230469, 382.5589904785156, 57.134307861328125, 293.6065673828125, 69.3941650390625, 26.504364013671875, 35.142982482910156, 220.8568115234375, 173.1151123046875, 60.895660400390625, 266.2698974609375, 59.178016662597656, 8.713226318359375, 148.80029296875, 334.05450439453125, -91.0704345703125, 0.0, -29.279296875, 138.70994567871094, -12.597023010253906, -10.323753356933594, 27.941070556640625, 148.45721435546875, 235.25009155273438, -40.281646728515625, 22.3994140625, -20.031047821044922, 41.421539306640625, 28.9482421875, 335.2984924316406, 135.90179443359375, 302.8292236328125, 422.2066345214844, -8.206924438476562, 332.2946472167969, 79.47689819335938, 215.03378295898438, 285.3969421386719, 56.296844482421875, 220.99960327148438, 51.04411315917969, 65.98538208007812, -446.1894226074219, 155.18661499023438, 35.36225891113281, 62.2105712890625, 6.705417633056641, 281.38946533203125, 247.6851806640625, 22.95703125, 77.68319702148438, 166.32669067382812, 16.984527587890625, 348.51904296875, 72.45146179199219, 470.005859375, 279.6864013671875, 234.05552673339844, 349.64862060546875, 48.30029296875, -23.880035400390625, 124.56610107421875, -31.923431396484375, -10.740242004394531, 69.81317138671875, 192.542236328125, -1.6046066284179688, 180.00204467773438, 71.36721801757812, 179.5623779296875, 102.74392700195312, -171.933349609375, 175.79302978515625, 70.18099975585938, 187.46380615234375, -98.19760131835938, 169.70159912109375, 36.5494384765625, 446.5912170410156, -136.35040283203125, -43.4132080078125, 405.16619873046875, 215.32684326171875, 55.393043518066406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000276.npy"}
{"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 68.21009826660156, "std": 160.31195068359375, "min": -330.52117919921875, "p10": -130.85187835693358, "median": 49.69580078125, "p90": 219.11047363281247, "max": 618.612548828125, "pos_frac": 0.65625, "sample": [179.45326232910156, -17.82745361328125, 29.594196319580078, 148.486083984375, 601.775390625, -330.52117919921875, -47.6451416015625, -2.43157958984375, 104.4508056640625, -52.94395446777344, 228.59747314453125, 101.90000915527344, 58.66499328613281, 175.22921752929688, -106.1241683959961, 229.96218872070312, 26.925127029418945, -196.159423828125, 179.27047729492188, 173.5599365234375, 48.56459045410156, -12.319746017456055, 50.25660705566406, 131.7313232421875, 99.70028686523438, 255.52642822265625, -19.43602752685547, -138.1353759765625, 166.9927978515625, 46.68206787109375, 49.13499450683594, -253.421875, -12.951740264892578, -191.39404296875, -39.19219207763672, 74.77302551269531, 151.30335998535156, 33.69623565673828, 185.151611328125, 133.56500244140625, -294.110107421875, -120.06948852539062, -129.42095947265625, -95.40777587890625, 26.963531494140625, 16.828094482421875, 76.689453125, 163.43760681152344, 115.02197265625, 122.68389892578125, 151.984619140625, -31.460771560668945, 2.0533981323242188, 618.612548828125, 188.880615234375, 0.0, -58.90887451171875, 169.54446411132812, 92.38075256347656, 103.5570068359375, 387.87969970703125, 214.31448364257812, -203.22616577148438, -5.3495635986328125, -36.275054931640625, 38.1322021484375, 164.15240478515625, -160.92295837402344, 403.31683349609375, 398.61212158203125, -8.946964263916016, -53.63348388671875, 290.511962890625, 82.505615234375, -5.958953857421875, -221.94146728515625, 162.19418334960938, -58.7171630859375, -74.61567687988281, 0.9127540588378906, 48.86456298828125, 94.70265197753906, -152.6045684814453, -46.093017578125, 117.19390869140625, 184.41348266601562, 90.14517974853516, 213.447265625, -134.19068908691406, 383.990966796875, 27.428863525390625, 64.548828125, -18.749343872070312, -47.44891357421875, 247.86276245117188, -5.4779052734375, 31.4185791015625, 215.04461669921875, 168.10643005371094, 397.1590270996094, 186.8878173828125, 101.69326782226562, 42.236480712890625, 186.58355712890625, 143.69635009765625, 198.05923461914062, 48.830108642578125, 425.11309814453125, 108.34683227539062, 0.0, 202.72235107421875, 182.34942626953125, 39.29473114013672, -121.8572998046875, -27.428817749023438, 165.01104736328125, -212.00238037109375, 176.0880126953125, 147.3939208984375, 40.046016693115234, -24.216949462890625, -232.18374633789062, -11.948715209960938, 4.039577484130859, 101.26280212402344, 142.23045349121094, 155.4961700439453, 6.738134384155273], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000277.npy"}
{"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 57.88999938964844, "std": 191.34628295898438, "min": -321.3428955078125, "p10": -138.98670196533203, "median": 18.352535247802734, "p90": 291.5165618896484, "max": 939.4137573242188, "pos_frac": 0.5625, "sample": [393.91094970703125, 30.628820419311523, -196.74179077148438, 144.36843872070312, -48.763702392578125, 143.92544555664062, 157.62783813476562, -38.3992919921875, -35.18269729614258, 290.1278991699219, 173.51376342773438, 19.95043182373047, 173.33863830566406, 154.2629852294922, 147.48504638671875, 169.94720458984375, -112.09518432617188, -117.01304626464844, 212.7773895263672, -82.23635864257812, -102.60942840576172, 108.5906982421875, -6.767791748046875, -76.85896301269531, -51.319488525390625, 470.13165283203125, 544.3584594726562, 29.433879852294922, 16.603302001953125, -321.3428955078125, -188.62115478515625, 939.4137573242188, 16.754638671875, -21.99078369140625, 221.82977294921875, 27.028457641601562, 10.960744857788086, -116.12298583984375, 317.517333984375, -81.52074432373047, 206.18121337890625, 129.38751220703125, 199.76943969726562, -0.12811279296875, 36.359920501708984, 102.48028564453125, -101.35784912109375, 220.1636962890625, 173.14346313476562, -82.99371337890625, 10.778858184814453, 405.361572265625, -3.846996307373047, -142.01040649414062, -102.01254272460938, -141.57009887695312, -11.328397750854492, 498.25946044921875, -223.99366760253906, -183.79356384277344, 37.94309997558594, 148.46002197265625, -54.06890869140625, -7.9515380859375, -47.623992919921875, 244.173095703125, -121.67428588867188, 186.08111572265625, 206.82691955566406, 126.992431640625, 294.75677490234375, 299.2340087890625, 13.664834976196289, -136.99903869628906, 179.35845947265625, -207.2627716064453, -13.857728958129883, 22.93415069580078, 75.9573974609375, 155.83270263671875, -299.56817626953125, -137.87953186035156, -6.433725357055664, 279.33013916015625, 208.87564086914062, 82.85589599609375, -38.488983154296875, 144.1448516845703, 71.26318359375, 84.56299591064453, -22.417526245117188, -80.7042236328125, 137.0560302734375, 525.1068115234375, -67.6549072265625, 318.9178771972656, -218.02764892578125, -24.614105224609375, 88.05439758300781, -0.549041748046875, -9.53863525390625, 62.90177917480469, 52.846923828125, -128.3272705078125, -299.6264343261719, 86.0089111328125, 147.97607421875, -49.24614715576172, -35.120391845703125, 93.0367660522461, 148.44248962402344, 25.052478790283203, -53.1270751953125, 639.144287109375, 4.997692108154297, -30.28295135498047, 14.438901901245117, -160.06704711914062, -27.379348754882812, -263.15301513671875, 10.193862915039062, -6.080169677734375, 24.66619873046875, 157.29586791992188, -95.27130126953125, 90.49099731445312, -45.15081024169922, 302.4393005371094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000278.npy"}
{"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 106.81974792480469, "std": 170.02053833007812, "min": -289.15228271484375, "p10": -88.1832405090332, "median": 86.7470932006836, "p90": 296.82197265625, "max": 778.254150390625, "pos_frac": 0.7578125, "sample": [92.40380859375, 286.35205078125, 296.79791259765625, 99.27162170410156, 3.1710052490234375, -103.0712890625, 79.32327270507812, 308.8930969238281, 44.45376205444336, 167.91522216796875, 14.96795654296875, 239.09136962890625, 161.9510498046875, -16.66668701171875, 12.34033203125, 166.44223022460938, 247.0335693359375, 148.44815063476562, 45.235595703125, 252.142333984375, 50.99595642089844, -18.446792602539062, 110.40740966796875, 2.62353515625, 139.01458740234375, 7.059539794921875, -42.46819305419922, 51.2940673828125, -201.12542724609375, 28.443740844726562, 176.8789825439453, 268.88134765625, 203.07891845703125, 127.07867431640625, 55.8126220703125, -33.407135009765625, 33.00868225097656, 88.52772521972656, -289.15228271484375, 21.467620849609375, 69.27247619628906, 82.42469787597656, 110.02194213867188, 418.572998046875, 85.79388427734375, 228.82188415527344, 8.11260986328125, 22.932708740234375, 69.7550048828125, -19.89263916015625, 313.3924560546875, 298.3520812988281, -76.46771240234375, 20.19283676147461, 165.61026000976562, -234.83160400390625, 299.1549072265625, -168.64906311035156, 196.73727416992188, 265.4696960449219, -124.28308868408203, 52.0467529296875, 198.2216033935547, 181.2406005859375, -132.87783813476562, -30.448699951171875, -107.68426513671875, 175.229248046875, 26.682697296142578, 282.8472900390625, -19.045867919921875, 660.3983154296875, 64.28555297851562, -39.007568359375, 8.239313125610352, -195.84469604492188, -40.6834716796875, 605.189208984375, 258.3572998046875, 25.45135498046875, -141.94955444335938, 93.00283813476562, 327.7506103515625, 778.254150390625, 256.86956787109375, 102.08526611328125, 103.28134155273438, -21.18364715576172, 143.44064331054688, 230.05381774902344, 87.48283386230469, -95.493896484375, 16.810516357421875, 209.8804931640625, -6.082035064697266, -47.40777587890625, 234.0181884765625, 20.863203048706055, 161.41470336914062, 169.25872802734375, 207.0047607421875, 269.3567810058594, -85.05010223388672, 89.39935302734375, 192.23370361328125, 382.687744140625, 220.66726684570312, 36.949310302734375, -10.324798583984375, 173.1013641357422, -64.44308471679688, 92.6300048828125, 333.4189758300781, 216.12445068359375, 31.437469482421875, 86.0113525390625, -112.0538330078125, 216.87167358398438, -150.72113037109375, -6.056434631347656, 559.8524169921875, 106.78634643554688, 290.8720703125, -48.479164123535156, 276.87359619140625, 60.201141357421875, 58.787811279296875, 296.87811279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000279.npy"}
{"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 84.96122741699219, "std": 146.68304443359375, "min": -348.94268798828125, "p10": -64.02732696533201, "median": 74.44647216796875, "p90": 260.2154296875, "max": 527.9676513671875, "pos_frac": 0.7265625, "sample": [256.65362548828125, 45.927818298339844, -16.495773315429688, 109.89312744140625, -22.114606857299805, -175.65228271484375, 191.16932678222656, 235.06414794921875, 178.16864013671875, 47.08966064453125, -4.455230712890625, 119.18629455566406, 75.71846771240234, 260.174560546875, -7.2747802734375, -58.70551300048828, 133.98626708984375, 18.987548828125, 333.001708984375, -35.88592529296875, 75.8134765625, 262.185546875, -348.94268798828125, 175.9482879638672, -173.57034301757812, 328.38641357421875, 1.1576461791992188, 60.82275390625, 73.21640014648438, 350.75537109375, 0.0, 176.49615478515625, 269.16448974609375, -123.3653564453125, 74.30734252929688, 216.427978515625, 22.226211547851562, 129.10870361328125, 24.688081741333008, -22.7537841796875, 119.2984619140625, 95.0145492553711, 33.31399917602539, 4.64105224609375, -2.4128570556640625, 151.5693359375, 115.41952514648438, -37.89872741699219, 205.40541076660156, -9.57611083984375, 175.948486328125, 99.96158599853516, 13.6815185546875, 58.69801712036133, -27.720535278320312, 215.643310546875, 111.30615234375, 22.31573486328125, -16.8575439453125, 260.310791015625, 298.63970947265625, 98.94131469726562, 164.11668395996094, 27.593772888183594, -30.439411163330078, -40.69786834716797, -105.4166259765625, 217.63536071777344, 56.235015869140625, 166.4788818359375, 488.3037109375, -76.44489288330078, 220.565673828125, 250.68612670898438, 214.286376953125, -292.70843505859375, 188.06927490234375, 13.594390869140625, -31.7335205078125, 173.6033935546875, 96.39796447753906, 193.23306274414062, 65.4495849609375, 115.14154052734375, 384.9881896972656, 37.307342529296875, 73.024169921875, -196.1990509033203, 153.63516235351562, -23.762786865234375, 202.21551513671875, -8.9356689453125, -8.08938980102539, -206.06402587890625, 31.891799926757812, 274.9768981933594, -94.20213317871094, 186.16021728515625, 158.9345703125, 123.3033447265625, 167.13284301757812, 55.84901428222656, 498.71337890625, -17.163238525390625, 87.93405151367188, -76.69876098632812, 6.6424560546875, 94.35200500488281, 527.9676513671875, 65.43505859375, 178.08351135253906, 95.96006774902344, 20.057758331298828, 184.84762573242188, -236.60894775390625, 75.4906005859375, 152.43441772460938, 229.64739990234375, 23.87847900390625, -218.68934631347656, 15.10791015625, 74.58560180664062, 67.17559814453125, -9.518299102783203, 80.35481262207031, 31.752788543701172, 302.15325927734375, -7.12750244140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000280.npy"}
{"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 102.15110778808594, "std": 158.79429626464844, "min": -280.7315368652344, "p10": -68.1113800048828, "median": 80.1039810180664, "p90": 293.5957977294922, "max": 900.364990234375, "pos_frac": 0.703125, "sample": [151.86798095703125, 116.2276840209961, -108.30209350585938, 44.165435791015625, 236.66665649414062, 21.8345947265625, 299.7257385253906, 27.28656005859375, 179.3531494140625, 145.37469482421875, 247.358154296875, -8.684906005859375, -18.16015625, 333.2547302246094, -27.216781616210938, 137.67681884765625, 83.02383422851562, 320.1014404296875, 71.12107849121094, 13.262130737304688, 253.18145751953125, 29.964523315429688, 77.18412780761719, -61.781219482421875, 124.3905029296875, 38.78656768798828, -86.0692138671875, 381.0554504394531, 72.87667846679688, 95.00315856933594, 0.0, 145.3563232421875, 63.789642333984375, 377.6910095214844, -3.525146484375, 316.8343505859375, 276.5058288574219, -99.02914428710938, 170.55722045898438, 105.12948608398438, -280.7315368652344, 11.462677001953125, 193.80596923828125, -24.51258087158203, 0.0, -36.755126953125, -9.3330078125, 136.1739501953125, 33.291259765625, 64.58891296386719, 241.35574340820312, 13.071197509765625, -12.456184387207031, 147.1985321044922, 10.25616455078125, 287.1049499511719, -53.2314453125, 93.02958679199219, 33.23974609375, 93.78458404541016, 298.1302795410156, 183.33384704589844, -67.43023681640625, 1.8739280700683594, 205.00367736816406, 112.1231689453125, 225.1985321044922, 1.983123779296875, -120.10580444335938, -27.47393798828125, -69.70071411132812, 191.99212646484375, -97.527099609375, 241.3006134033203, 296.5821838378906, -58.145904541015625, 202.30621337890625, 170.31878662109375, 188.4573974609375, 85.78804016113281, 6.530494689941406, 147.9039306640625, 900.364990234375, 575.4727783203125, 397.2203063964844, -5.841346740722656, -14.903762817382812, 109.2662124633789, 153.2531280517578, 11.726203918457031, 265.4893798828125, 51.5849609375, 95.30416870117188, 25.655113220214844, 167.32901000976562, -31.75115966796875, 26.883056640625, 292.31591796875, 434.2030029296875, -15.235107421875, -83.14987182617188, 119.1798095703125, 0.0, 421.607177734375, -96.99142456054688, -84.59928894042969, 53.091514587402344, -66.5836181640625, 239.44613647460938, 164.10232543945312, 150.24838256835938, 131.77830505371094, 212.55926513671875, 84.35491943359375, 233.52944946289062, -90.252197265625, -63.668365478515625, 249.44308471679688, -44.980262756347656, 230.77232360839844, 29.820648193359375, -41.30084228515625, -71.03390502929688, 240.27996826171875, 205.36941528320312, -157.0480194091797, 69.63043212890625, -2.22564697265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000281.npy"}
{"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 81.4150161743164, "std": 151.8291778564453, "min": -285.11785888671875, "p10": -101.2623291015625, "median": 72.29769897460938, "p90": 261.90822753906247, "max": 569.02392578125, "pos_frac": 0.703125, "sample": [243.3785400390625, -30.68243408203125, 51.107513427734375, 569.02392578125, -26.990325927734375, 25.00749969482422, -194.61563110351562, 339.6146240234375, 134.76254272460938, -99.38323211669922, -43.44770812988281, 112.17675018310547, 11.944183349609375, -42.070770263671875, 145.1988525390625, 16.45446014404297, 228.76031494140625, -246.36087036132812, 243.69207763671875, 174.8498077392578, 35.352142333984375, -34.693939208984375, -58.90653610229492, 184.26287841796875, 77.42855834960938, 419.5426330566406, 226.85533142089844, 162.039306640625, -61.95648193359375, 30.06732177734375, -46.66082763671875, 196.64028930664062, 175.5029296875, 76.12689208984375, 1.2069091796875, -5.1290283203125, 161.33370971679688, -7.536439895629883, 89.89962768554688, 254.294921875, 93.57978820800781, -2.833984375, 273.0509033203125, -72.53584289550781, 312.9817199707031, -115.67839050292969, -272.2958984375, 156.26438903808594, 141.2523193359375, 142.17828369140625, 10.66986083984375, 66.97604370117188, -31.4383544921875, 46.31608581542969, -107.6649169921875, 263.94781494140625, 82.71157836914062, 71.70474243164062, -72.98540496826172, -136.92178344726562, -26.0653076171875, 56.5198974609375, -12.613677978515625, 219.73809814453125, 199.55087280273438, 79.9471435546875, 94.87837982177734, -105.64688873291016, 104.69035339355469, 114.93060302734375, -285.11785888671875, -238.08175659179688, 149.35964965820312, 92.36328125, 72.89065551757812, 265.4608459472656, 321.93505859375, 4.251960754394531, 282.439208984375, -263.374755859375, -152.7893524169922, -1.4917411804199219, 102.31524658203125, 236.35922241210938, -118.3255615234375, 38.23326110839844, 25.999313354492188, -0.6131591796875, 221.90640258789062, 233.9578857421875, 246.89028930664062, 61.380401611328125, 30.616790771484375, 124.09307861328125, -30.1763916015625, 69.70559692382812, 77.60110473632812, 107.73442077636719, -3.9890708923339844, 25.64227294921875, 249.08261108398438, 478.68280029296875, 261.03411865234375, -0.7979049682617188, 90.19046020507812, 22.873291015625, -180.35601806640625, 89.44473266601562, 370.08770751953125, 142.29693603515625, 208.37847900390625, 50.650909423828125, 66.723876953125, 48.58293914794922, -36.404014587402344, 222.724609375, 336.73504638671875, 195.874267578125, 428.2381286621094, 44.92230224609375, 69.117431640625, -16.201217651367188, 89.39665222167969, 4.2528839111328125, 163.39767456054688, -29.097763061523438, 154.54782104492188, 136.26824951171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000282.npy"}
{"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 88.3296127319336, "std": 157.93617248535156, "min": -278.9112243652344, "p10": -87.78116989135742, "median": 61.62265396118164, "p90": 269.08287658691404, "max": 606.1727294921875, "pos_frac": 0.6796875, "sample": [-94.2313232421875, 121.08309936523438, 224.37831115722656, -94.9002685546875, 10.045286178588867, -120.87985229492188, 379.35552978515625, -56.788429260253906, 462.6849365234375, -84.95201110839844, -117.25970458984375, 162.37420654296875, 183.41262817382812, 14.865850448608398, -27.283489227294922, -82.42037963867188, -32.140625, 208.0405731201172, 20.245758056640625, -54.69781494140625, 587.579345703125, 330.25982666015625, 109.54833984375, -87.0023193359375, 366.3643493652344, 309.3692932128906, 216.010009765625, -25.940460205078125, 202.92428588867188, 180.05154418945312, 210.5286865234375, -47.01820373535156, -99.47813415527344, 216.48587036132812, 88.54164123535156, 248.01966857910156, -23.974380493164062, 4.112470626831055, 296.7948913574219, -15.258285522460938, -59.8138427734375, 268.0809326171875, -47.062347412109375, 4.531982421875, 46.31281280517578, 253.78958129882812, 49.752166748046875, -44.868255615234375, 116.91554260253906, 427.2352294921875, 52.011260986328125, 130.1020050048828, 222.070556640625, -40.80377960205078, 518.8412475585938, -3.1279296875, -20.03420639038086, 138.763427734375, 66.102783203125, 81.03107452392578, 249.34893798828125, 427.445556640625, 245.62030029296875, 143.42941284179688, 120.363037109375, 10.1573486328125, 150.5323486328125, 52.899444580078125, 256.1714782714844, 62.56416320800781, -13.348388671875, 3.9459686279296875, -89.5984878540039, 188.619140625, -98.89959716796875, 63.1513671875, 63.62660217285156, 106.47660827636719, 0.0, -31.83563232421875, -32.01630401611328, 45.76710510253906, 167.9251708984375, -2.0550537109375, 114.2706298828125, -44.28094482421875, 55.710052490234375, -91.97421264648438, -153.8934326171875, 70.44845581054688, 30.655776977539062, -278.9112243652344, 188.201904296875, -58.07159423828125, 100.99932861328125, 89.32064819335938, 183.9881591796875, 139.2471923828125, -200.4014129638672, 106.99484252929688, 7.871694564819336, 38.27276611328125, 606.1727294921875, 60.68114471435547, 124.4676513671875, -33.446250915527344, 140.5958251953125, -112.91363525390625, 51.506866455078125, -165.58151245117188, 216.71566772460938, 35.66143798828125, -15.408515930175781, 13.59242057800293, 0.0, 246.14398193359375, 111.70866394042969, 473.9991455078125, 133.37969970703125, 28.395801544189453, -34.60943603515625, 115.18952941894531, 107.32916259765625, 271.4207458496094, 4.309814453125, 8.44915771484375, 108.48028564453125, 172.53060913085938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000283.npy"}
{"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 88.77078247070312, "std": 180.68136596679688, "min": -381.8451843261719, "p10": -139.0985137939453, "median": 82.0682144165039, "p90": 308.70740966796876, "max": 689.864501953125, "pos_frac": 0.6953125, "sample": [-102.15281677246094, 114.802734375, -168.1949005126953, -233.284912109375, 97.51470947265625, 53.088043212890625, 308.57696533203125, 362.8892822265625, 29.815032958984375, 47.90779113769531, 4.68812370300293, 269.5694580078125, 297.22882080078125, 62.478675842285156, 375.68280029296875, 248.05340576171875, 31.749053955078125, 142.29071044921875, 323.0931396484375, 81.03731536865234, 141.25115966796875, 622.0462646484375, -44.876739501953125, 191.85623168945312, -39.0865592956543, 155.30435180664062, 151.03103637695312, -200.20611572265625, 53.99599075317383, -54.107391357421875, -96.822021484375, 270.66748046875, 185.44598388671875, -191.10446166992188, 90.73419189453125, 100.85000610351562, 178.31787109375, 195.4046630859375, 154.57925415039062, -198.1524200439453, -150.59844970703125, 59.84370422363281, 366.534912109375, 90.2286376953125, -193.37376403808594, 167.03567504882812, 6.2093963623046875, 689.864501953125, 197.351318359375, 282.4888916015625, 83.07220458984375, -87.29647827148438, 265.94598388671875, -243.30029296875, 326.64837646484375, 181.02490234375, -8.3819580078125, -22.32071304321289, 262.61328125, 556.365478515625, 185.26287841796875, 197.8882293701172, 136.032958984375, -83.67784118652344, 127.70449829101562, 111.26995849609375, 349.3497314453125, 49.51324462890625, 239.30172729492188, 10.666839599609375, 362.9691467285156, 14.717765808105469, 2.5361709594726562, -11.3685302734375, 4.4340667724609375, 425.1904602050781, 64.81414794921875, 15.194313049316406, 109.01303100585938, 110.77346801757812, 300.0052490234375, 120.10223388671875, 102.63037872314453, 211.84530639648438, -313.5289001464844, -70.96371459960938, -9.166065216064453, -127.87884521484375, -114.7371826171875, -9.6390380859375, 166.52964782714844, -28.10833740234375, 81.06422424316406, 144.83963012695312, -9.060604095458984, -52.20709228515625, 91.442138671875, -258.6192932128906, 1.69482421875, 61.45606994628906, 302.16802978515625, -139.30908203125, 261.2833557128906, 164.04776000976562, 27.668136596679688, 167.036376953125, -381.8451843261719, 57.37855529785156, -26.723388671875, 125.32123565673828, 309.01177978515625, 283.12640380859375, 69.07026672363281, -14.181488037109375, -200.3825225830078, -22.02685546875, 14.70172119140625, -85.45623016357422, -11.06454086303711, -18.091796875, 180.52850341796875, 137.55296325683594, 220.33441162109375, 376.7795104980469, 152.9691925048828, -55.35852813720703, -139.00827026367188, 53.92430114746094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000284.npy"}
{"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 59.75489807128906, "std": 169.8118896484375, "min": -383.5426025390625, "p10": -134.37408142089845, "median": 42.06298828125, "p90": 272.24884033203125, "max": 598.3583984375, "pos_frac": 0.65625, "sample": [47.04754638671875, 83.3154296875, 168.938720703125, 40.9119873046875, 253.38714599609375, 158.53421020507812, 146.8394775390625, -85.42857360839844, 58.7568359375, 90.78662109375, 148.38409423828125, -176.5489501953125, 61.22288513183594, -53.007354736328125, -82.09210205078125, 73.40019226074219, 36.46168518066406, 24.9403076171875, -77.91448974609375, 187.90570068359375, -47.8106689453125, 33.6197509765625, -26.016281127929688, 192.26194763183594, 40.028350830078125, 83.59666442871094, 413.7135314941406, 204.15377807617188, 106.79315185546875, 427.74273681640625, 298.92547607421875, -135.01309204101562, 26.523208618164062, -107.2286376953125, 365.35528564453125, 380.9866943359375, 99.73834228515625, -210.87506103515625, -160.08984375, -124.49444580078125, -94.86672973632812, -121.8121337890625, -147.30728149414062, 20.937286376953125, -8.162004470825195, 83.32855224609375, 93.8812255859375, 43.173583984375, 51.943939208984375, 258.7164611816406, 328.8341979980469, 0.8369255065917969, -34.51371765136719, -55.897743225097656, 520.553955078125, -2.925262451171875, -94.62293243408203, -40.05054473876953, 237.6437530517578, -21.393264770507812, -145.8030548095703, 330.72808837890625, 136.18856811523438, 275.51214599609375, 110.69512939453125, -251.197509765625, 68.67059326171875, -19.894960403442383, 10.747299194335938, 232.18389892578125, -67.51541137695312, 240.14599609375, 212.84056091308594, 3.16876220703125, 76.84598541259766, 14.316482543945312, 40.952392578125, -18.662208557128906, 29.18218994140625, 53.8057861328125, 63.72636413574219, 158.791015625, 17.88818359375, 78.40116882324219, -118.92037963867188, 242.41204833984375, 246.59536743164062, 190.4326171875, 598.3583984375, 270.85028076171875, 131.82472229003906, 153.63778686523438, 14.519950866699219, -361.73577880859375, -66.68991088867188, -97.09814453125, 50.2218017578125, 306.50177001953125, -60.418914794921875, 24.7108154296875, 169.9924774169922, -134.1002197265625, 136.67840576171875, -60.11383056640625, -383.5426025390625, -236.5205078125, 10.19012451171875, 29.058414459228516, 246.0867919921875, 93.62896728515625, -290.5633850097656, -115.97703552246094, 129.67877197265625, -104.52146911621094, -51.40315246582031, -208.21591186523438, 37.89030075073242, 69.46414184570312, 316.0916748046875, 63.092193603515625, 37.57244873046875, -36.3846435546875, -38.584075927734375, 375.6167297363281, -169.82089233398438, 189.28271484375, 146.16693115234375, 264.9128723144531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000285.npy"}
{"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 88.30255126953125, "std": 166.27940368652344, "min": -271.62353515625, "p10": -76.3698547363281, "median": 71.65084838867188, "p90": 278.60321960449215, "max": 847.6080932617188, "pos_frac": 0.6796875, "sample": [-31.384521484375, 75.25726318359375, 64.60137939453125, -21.775604248046875, 50.8062744140625, -13.345466613769531, 56.9156494140625, 12.286170959472656, 27.978759765625, 142.370361328125, 173.15206909179688, 94.7435302734375, 98.3222885131836, 233.03533935546875, -138.10226440429688, 97.6884765625, 120.88684844970703, 178.73748779296875, 169.82290649414062, 190.5440673828125, -51.94610595703125, 252.26358032226562, 203.14349365234375, 58.03515625, 77.35611724853516, 89.62728881835938, -55.90234375, -2.5288963317871094, 151.26931762695312, -42.26494598388672, -224.06915283203125, 95.15974426269531, 306.5783386230469, 277.7359313964844, -12.53692626953125, 95.1346206665039, 201.81369018554688, 20.31103515625, 83.2225570678711, 129.96888732910156, 44.106109619140625, 65.84822082519531, 203.26968383789062, -271.62353515625, 152.3486328125, -94.7691650390625, 122.98486328125, 669.1019897460938, 32.0491943359375, -33.78173828125, 80.29916381835938, -100.81210327148438, 125.08602905273438, 58.406471252441406, 149.47222900390625, 226.3438720703125, -8.959579467773438, 405.4710998535156, -57.588157653808594, 306.30877685546875, -68.48443603515625, 0.0, -17.54296875, -155.04034423828125, 7.587013244628906, 165.9947509765625, -62.4171142578125, -14.773468017578125, 68.04443359375, 95.62967681884766, -220.96612548828125, 194.31622314453125, 41.5306396484375, 164.12432861328125, -2.1979713439941406, -26.08050537109375, 395.675537109375, 94.91605377197266, -25.869802474975586, 48.038970947265625, 181.34722900390625, -9.245357513427734, 6.491733551025391, 295.04132080078125, 386.61676025390625, 173.9229736328125, 125.51542663574219, 123.77458190917969, 185.87362670898438, -156.6627197265625, 50.996864318847656, 241.95654296875, 119.14544677734375, 467.80841064453125, 239.97613525390625, 175.85797119140625, 64.64624786376953, -19.453113555908203, -118.51484680175781, 52.73602294921875, 847.6080932617188, 59.864654541015625, 161.4244384765625, 192.43453979492188, -7.55029296875, -152.6907958984375, 112.23269653320312, 95.69656372070312, -11.02813720703125, 47.8642463684082, -242.7098388671875, 450.0511474609375, -7.159149169921875, -118.66741943359375, -48.576904296875, 546.7997436523438, -170.72769165039062, -16.815704345703125, 102.1048583984375, 13.847122192382812, 89.15733337402344, 280.62689208984375, -24.052274703979492, 3.5165863037109375, 298.1756591796875, -19.123458862304688, 112.44329833984375, 129.1893310546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000286.npy"}
{"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 79.77581787109375, "std": 169.35079956054688, "min": -427.609130859375, "p10": -117.1626037597656, "median": 53.29206848144531, "p90": 293.45243835449213, "max": 557.9708862304688, "pos_frac": 0.6640625, "sample": [158.5050048828125, 8.867809295654297, -110.76083374023438, 10.728816986083984, -155.0784454345703, -109.11003112792969, 255.84957885742188, 272.3663635253906, 254.28456115722656, 243.25775146484375, 46.0068359375, -7.3040618896484375, 38.40991973876953, 105.11164093017578, 240.43035888671875, 15.7353515625, -160.86624145507812, 13.764068603515625, 52.76255798339844, 121.34027099609375, -33.559967041015625, 270.08746337890625, 557.9708862304688, -34.060211181640625, 234.89590454101562, -427.609130859375, -10.454437255859375, 226.197021484375, 212.386474609375, 102.05645751953125, 53.82157897949219, 519.9122314453125, -3.8109359741210938, 307.04541015625, -149.20730590820312, -80.68293762207031, -58.73652648925781, 183.42974853515625, 17.9232120513916, 188.2860565185547, 124.89250946044922, 135.59738159179688, 149.06414794921875, 170.3116455078125, 290.4284973144531, -24.334156036376953, 251.333984375, 116.06031799316406, 259.1431884765625, 167.39373779296875, -24.90594482421875, 322.4195556640625, 253.82254028320312, 138.64248657226562, 117.08538818359375, 94.50946044921875, 48.57176208496094, 49.499542236328125, 364.3426208496094, -86.346435546875, 209.38217163085938, -103.22476196289062, 344.6207580566406, 16.653600692749023, 24.239356994628906, -47.129852294921875, 137.72628784179688, 334.8817443847656, 174.2835693359375, 228.54708862304688, -316.3021240234375, -2.2134475708007812, 328.09405517578125, 179.43084716796875, 113.5477294921875, 96.0760498046875, 45.247100830078125, 252.17825317382812, -74.74853515625, 44.5263557434082, 139.18698120117188, -97.53302001953125, -85.70448303222656, 52.45880889892578, -168.3201446533203, -63.423606872558594, -147.9864044189453, 275.0845947265625, -132.10006713867188, 106.04377746582031, 429.37457275390625, 373.08343505859375, -236.9658203125, -75.51766967773438, -82.18597412109375, 40.680023193359375, 42.29671096801758, 2.5692787170410156, -11.4986572265625, 107.14303588867188, -24.916778564453125, 146.16650390625, 62.238983154296875, 19.995361328125, -313.23834228515625, -45.3682861328125, 185.07620239257812, -211.36529541015625, 318.51025390625, 46.07488250732422, 277.6148681640625, -7.777961730957031, 211.7174530029297, 300.50830078125, 332.03558349609375, -27.2652587890625, 58.165008544921875, 152.3082275390625, 145.19512939453125, -0.7056541442871094, -15.228116989135742, 35.43132400512695, -203.06134033203125, -184.928466796875, -4.725616455078125, 112.13563537597656, -66.64222717285156, 173.12770080566406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000287.npy"}
{"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 75.76020050048828, "std": 159.2097625732422, "min": -426.4767150878906, "p10": -112.55722427368163, "median": 59.56513595581055, "p90": 280.3228485107422, "max": 548.0299682617188, "pos_frac": 0.6953125, "sample": [183.38836669921875, -32.45790481567383, -109.9381103515625, 51.268402099609375, 113.89053344726562, 50.7493896484375, 74.33309936523438, -152.1240234375, 64.67752075195312, 116.175048828125, 191.932861328125, 336.66253662109375, -104.8780517578125, 139.891845703125, 187.3021240234375, -181.86363220214844, 39.8985595703125, 154.19107055664062, 50.88604736328125, 0.0, -7.6683349609375, -14.91680908203125, -117.49514770507812, 83.82722473144531, 31.76848602294922, 17.8961181640625, 115.93841552734375, -123.491943359375, -160.3167724609375, 99.070556640625, 325.7969055175781, 242.05361938476562, 71.9979019165039, -33.33538818359375, 26.517837524414062, 0.0, -74.9465560913086, -10.848228454589844, 127.9393310546875, 79.34072875976562, -82.84486389160156, 255.55242919921875, -87.46688842773438, -119.07246398925781, -5.559541702270508, 319.81884765625, 224.1119384765625, 228.73141479492188, -241.5111083984375, -1.0987281799316406, 281.7241516113281, -91.85405731201172, 230.48046875, 0.452484130859375, 68.31953430175781, -169.6268310546875, -28.70144271850586, 63.76800537109375, 132.09617614746094, -133.9952392578125, 508.3092041015625, 81.71232604980469, 129.51104736328125, -201.05543518066406, 262.72235107421875, 42.6103515625, -110.44097137451172, -426.4767150878906, 331.10333251953125, 59.817588806152344, 385.467041015625, -30.928863525390625, -53.83673095703125, 122.19758605957031, 88.92340087890625, -43.448028564453125, 103.8121337890625, 20.7987060546875, -108.132080078125, 50.858665466308594, -101.16737365722656, 332.5576171875, 25.85277557373047, -8.337249755859375, -53.772216796875, 7.997810363769531, 176.5933074951172, 548.0299682617188, 260.66510009765625, 164.76760864257812, 190.250244140625, 13.980255126953125, 155.42333984375, 4.733325958251953, 180.68569946289062, 51.532012939453125, 205.28485107421875, -138.0308380126953, 501.42791748046875, 14.5821533203125, 96.45941162109375, -19.478302001953125, 27.978607177734375, 364.00738525390625, 43.05431365966797, 59.31268310546875, 75.96561431884766, -69.92922973632812, 101.18649291992188, 58.48162841796875, -128.08517456054688, 116.72396087646484, 82.77084350585938, 48.1195068359375, 151.69662475585938, 29.87664794921875, 6.0093994140625, 77.44964599609375, 147.2672119140625, 279.7222900390625, 78.73780059814453, 329.71795654296875, 483.2445068359375, 227.15805053710938, 247.03871154785156, 20.5147705078125, 247.18475341796875, 72.10140991210938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000288.npy"}
{"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 51.90618896484375, "std": 148.5804901123047, "min": -307.3839416503906, "p10": -126.64831085205078, "median": 44.62575912475586, "p90": 253.21331176757812, "max": 441.38421630859375, "pos_frac": 0.6640625, "sample": [-117.98895263671875, -17.68414306640625, 17.6986083984375, 81.50511932373047, -72.65216064453125, 17.812835693359375, 71.18597412109375, 19.516571044921875, 98.6802978515625, -85.555419921875, 308.9183349609375, 28.848251342773438, 8.564878463745117, 35.18310546875, 102.70779418945312, -6.592193603515625, -307.3839416503906, 140.40460205078125, -52.251434326171875, -40.5150146484375, 146.21746826171875, 197.84970092773438, 185.6321563720703, 98.02937316894531, 95.4276123046875, -109.64852905273438, -161.14480590820312, 0.4869232177734375, -3.68218994140625, 10.887786865234375, -51.866973876953125, 255.8922119140625, -111.02578735351562, 141.4805145263672, 35.57618713378906, 0.0, 59.04668426513672, 130.10662841796875, 112.18350219726562, 86.818115234375, 37.79585266113281, -102.08575439453125, -167.60873413085938, -126.03155517578125, 176.01458740234375, -103.86868286132812, 188.1942138671875, 56.890625, 94.62733459472656, 77.74203491210938, -246.4884033203125, 216.6094970703125, -251.37881469726562, -104.81494140625, 52.26838684082031, 21.970462799072266, -72.42914581298828, 29.265079498291016, -129.1259765625, 208.66610717773438, -100.9608154296875, 186.5058135986328, 148.65377807617188, 152.30335998535156, 100.57135009765625, -98.99969482421875, -33.107879638671875, -81.75841522216797, -94.76634216308594, 28.22418975830078, 337.72064208984375, -127.4881591796875, 68.32815551757812, 237.16213989257812, 141.63729858398438, 57.66961669921875, 6.636955261230469, 441.38421630859375, 52.052642822265625, 154.98883056640625, -23.946746826171875, -62.040771484375, -149.96258544921875, 259.154541015625, -126.28837585449219, 45.326454162597656, -8.409896850585938, 286.6834716796875, -233.54730224609375, 253.27389526367188, 7.887359619140625, 422.02642822265625, 274.0152893066406, -10.8115234375, 180.22616577148438, 98.19180297851562, 207.7054443359375, 0.527740478515625, 200.70709228515625, 43.92506408691406, 283.4199523925781, 232.96063232421875, 84.60943603515625, 180.5516357421875, -204.38363647460938, 133.81251525878906, 58.152618408203125, -204.48275756835938, 32.366546630859375, 391.0302734375, -183.69525146484375, 167.62750244140625, 246.1029815673828, -126.19674682617188, 280.0248718261719, -209.0443115234375, -60.146148681640625, 260.79791259765625, 6.730873107910156, 253.18734741210938, 49.262939453125, 92.37030029296875, 185.5279541015625, -101.45823669433594, 83.26705932617188, 18.596336364746094, 30.65228271484375, 216.0645751953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000289.npy"}
{"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 84.98925018310547, "std": 172.2332000732422, "min": -427.0901794433594, "p10": -116.63733978271485, "median": 84.83675384521484, "p90": 298.4318237304687, "max": 644.57177734375, "pos_frac": 0.7109375, "sample": [394.8389587402344, 23.3292236328125, 26.97369384765625, 344.13067626953125, 90.8724365234375, -138.72442626953125, 7.1900482177734375, 168.83819580078125, -10.57403564453125, 110.26229858398438, 533.4036865234375, 223.468505859375, 130.0791015625, -86.84991455078125, 271.0334777832031, 106.29339599609375, 220.73626708984375, 71.43182373046875, 400.98126220703125, 175.4287109375, 155.36270141601562, 14.317131042480469, 68.14202880859375, -58.091522216796875, 356.72967529296875, -135.346923828125, -55.82566833496094, -321.7547607421875, -9.139877319335938, 4.327789306640625, 644.57177734375, 31.054183959960938, 116.173828125, 131.37213134765625, 362.91888427734375, 546.3816528320312, 207.60894775390625, 16.726451873779297, -32.8321533203125, -220.69476318359375, 132.57200622558594, 168.52777099609375, -16.015106201171875, 130.64810180664062, 173.25381469726562, 15.742759704589844, 6.54290771484375, -208.105224609375, -25.1531925201416, 106.00070190429688, -252.6512451171875, 115.97918701171875, 175.9288330078125, -9.245796203613281, 108.53874969482422, -59.53466796875, -131.58782958984375, 119.38287353515625, 50.050048828125, 93.64356994628906, 147.81515502929688, 44.236122131347656, -130.71273803710938, 177.91552734375, 143.66192626953125, 168.6708984375, -4.2313690185546875, 71.7459716796875, 162.44485473632812, 312.499755859375, -117.20941162109375, 76.80966186523438, -28.095672607421875, 194.75930786132812, -35.014404296875, 292.4027099609375, 22.926950454711914, -170.48760986328125, 5.264984130859375, 240.52694702148438, -38.9683837890625, -25.705215454101562, -116.39216613769531, 246.52027893066406, 85.90724182128906, 164.98568725585938, 348.3412170410156, 15.15216064453125, -259.33929443359375, 72.38441467285156, 108.385009765625, -74.84617614746094, 75.7322998046875, 7.873443603515625, -63.42376708984375, 15.133232116699219, -106.68704223632812, -21.906417846679688, 149.67617797851562, 228.51272583007812, 92.09349060058594, -162.33392333984375, 182.611572265625, 330.50701904296875, 546.3179931640625, 64.14773559570312, 423.8397216796875, 181.003662109375, 249.92822265625, 99.89713287353516, -93.38604736328125, 118.46408081054688, 221.461181640625, 151.2552490234375, 1.956512451171875, -20.8836669921875, -23.00481414794922, 94.3743896484375, -427.0901794433594, 173.96560668945312, 92.128173828125, 81.16571044921875, 142.15414428710938, -98.66043853759766, 7.4269866943359375, 83.76626586914062, 167.602783203125, 230.9873046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000290.npy"}
{"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 109.19486999511719, "std": 156.46743774414062, "min": -296.8673095703125, "p10": -79.80052795410155, "median": 107.19392395019531, "p90": 312.2867004394531, "max": 571.8372192382812, "pos_frac": 0.75, "sample": [311.7822265625, -20.252010345458984, 340.65496826171875, 93.8995361328125, 199.74185180664062, 152.16062927246094, 182.91976928710938, 345.08270263671875, 157.08929443359375, 105.67752075195312, 115.83920288085938, 42.29283142089844, 176.20687866210938, -76.54547119140625, 297.5033874511719, 31.336624145507812, 531.50732421875, -27.178070068359375, 122.416748046875, 214.7445068359375, 122.7587890625, 85.7218017578125, 235.12615966796875, -28.568084716796875, -171.9288787841797, 184.70404052734375, -3.20379638671875, -36.26679229736328, 166.56173706054688, -102.60028076171875, -140.52317810058594, 50.01811218261719, -87.39566040039062, 47.66969299316406, -296.8673095703125, 188.512451171875, 142.17724609375, 65.66455078125, 62.99908447265625, 82.78916931152344, 231.79296875, -6.7015228271484375, -9.747894287109375, -47.854713439941406, 465.1032409667969, 244.00997924804688, 3.4463882446289062, 313.46380615234375, 91.28082275390625, -31.024276733398438, 99.54766845703125, 163.26959228515625, 159.0010986328125, -50.920196533203125, 352.234375, 343.27734375, -50.54937744140625, -4.978240966796875, -154.8164825439453, 141.79830932617188, 211.4149169921875, 50.583709716796875, 95.17626953125, -16.29296875, 333.68878173828125, -258.9915771484375, 85.75114440917969, -18.573928833007812, -106.15643310546875, 145.642333984375, -130.40182495117188, 206.96597290039062, 2.4880313873291016, 336.2940673828125, 115.27073669433594, 262.863525390625, 30.72730255126953, 292.04815673828125, 233.2794952392578, -135.1815185546875, 143.44102478027344, 2.4384536743164062, 119.720703125, -50.35565185546875, 108.7103271484375, 389.6673278808594, 114.36531066894531, -37.9510498046875, 0.0, 13.141258239746094, 91.64198303222656, 179.39822387695312, 109.23162841796875, 16.357864379882812, 284.4052429199219, 44.618812561035156, 210.43524169921875, 170.67584228515625, 194.853515625, -144.68588256835938, 124.27667236328125, 571.8372192382812, 92.46844482421875, 353.55841064453125, 38.9459228515625, 240.226806640625, 38.009857177734375, 141.62759399414062, 552.45263671875, 63.3582763671875, -124.44528198242188, 184.23883056640625, 204.07470703125, 15.27267837524414, 208.24493408203125, 49.380401611328125, 124.565673828125, 1.3270587921142578, 243.7056884765625, 240.64019775390625, 102.55075073242188, 117.77471923828125, -141.20855712890625, -7.899408340454102, 244.60975646972656, 170.75714111328125, 10.857025146484375, 303.1670837402344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000291.npy"}
{"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 90.7706298828125, "std": 173.43377685546875, "min": -422.6458740234375, "p10": -84.9298858642578, "median": 74.23554229736328, "p90": 287.41490783691404, "max": 670.4733276367188, "pos_frac": 0.7421875, "sample": [44.45843505859375, 8.467620849609375, 70.17314147949219, -3.0718994140625, 156.79026794433594, 149.53472900390625, 99.1390380859375, 421.448486328125, 105.65673828125, 338.238037109375, 670.4733276367188, 61.561798095703125, -65.64129638671875, -91.48330688476562, 145.70220947265625, 157.157958984375, -28.533124923706055, 5.235532760620117, 54.803192138671875, 1.8582763671875, 119.08956909179688, -26.051513671875, 91.15292358398438, 136.53097534179688, 249.28292846679688, 116.11468505859375, 286.1243896484375, 290.4261169433594, 37.02752685546875, 15.074085235595703, 420.32470703125, -14.657035827636719, 25.483789443969727, 5.4965057373046875, 253.18368530273438, 85.71756744384766, 73.88119506835938, 45.879150390625, -323.82958984375, 108.46741485595703, 167.6220703125, -177.06317138671875, 246.85958862304688, 131.09603881835938, -6.10833740234375, 226.93336486816406, -9.167411804199219, -223.81430053710938, 28.78441619873047, 284.41741943359375, -240.76007080078125, 508.22991943359375, -44.97438049316406, -11.6776123046875, -44.863525390625, -27.715774536132812, -19.38995361328125, 74.58988952636719, 15.185546875, 85.91508483886719, 140.94110107421875, 70.93463134765625, 50.153472900390625, -93.14715576171875, 55.2425537109375, 143.23663330078125, 137.54010009765625, 137.11822509765625, 583.797607421875, 420.95318603515625, 29.14849853515625, 112.07376098632812, -21.749267578125, 45.0732421875, -24.484439849853516, 510.401611328125, 44.48839569091797, 368.7345886230469, 87.1606216430664, -145.44351196289062, -138.9058837890625, -293.8170166015625, 195.5858154296875, 277.0364074707031, 261.5179138183594, 35.17173767089844, 223.84039306640625, -8.5582275390625, 16.465505599975586, 272.5420227050781, -82.12127685546875, 23.16619873046875, 109.86659240722656, 84.8668212890625, -154.54505920410156, 95.7442855834961, -2.0650596618652344, -45.31422424316406, 18.61786460876465, 160.6942138671875, 81.6964111328125, 230.31459045410156, 25.919265747070312, 21.713930130004883, 141.40939331054688, 150.9447021484375, 151.33682250976562, 2.1299896240234375, 95.50648498535156, 70.50225830078125, 18.595916748046875, 492.55810546875, 141.35015869140625, -21.190216064453125, 276.28765869140625, 158.60687255859375, 227.23733520507812, -236.13726806640625, 92.26217651367188, 39.98930358886719, 295.3366394042969, -422.6458740234375, -231.0391845703125, 229.31787109375, 315.09857177734375, 172.841552734375, 147.5166015625, -11.568361282348633], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000292.npy"}
{"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 92.34884643554688, "std": 158.71588134765625, "min": -266.02520751953125, "p10": -97.42460479736327, "median": 71.43595504760742, "p90": 288.6437255859375, "max": 515.041015625, "pos_frac": 0.6875, "sample": [219.53565979003906, 54.415870666503906, 310.9053039550781, 210.3177490234375, 96.25314331054688, -210.20608520507812, 55.136566162109375, 114.66082763671875, -46.23883056640625, 139.5273895263672, 480.8165588378906, 360.556884765625, 263.93438720703125, -67.21112060546875, 34.88175964355469, 7.7286529541015625, -1.2409210205078125, 7.434356689453125, 30.899673461914062, 109.46792602539062, 39.074913024902344, -84.85528564453125, 487.990478515625, 50.864593505859375, 94.44157409667969, 0.0, 295.1744384765625, 138.8136749267578, 222.41448974609375, 92.43387603759766, 91.99215698242188, -53.76922607421875, -95.62837219238281, -101.61581420898438, 201.27239990234375, -135.25843811035156, -73.5838623046875, 277.8026428222656, 141.38479614257812, 26.238113403320312, 173.41683959960938, -3.620616912841797, 233.370361328125, 120.19189453125, -21.6627197265625, -187.85479736328125, 285.8448486328125, 202.31854248046875, 515.041015625, 72.4104995727539, -0.0234832763671875, 170.0101318359375, 83.61918640136719, 133.62109375, -133.14065551757812, 350.6667785644531, -84.354248046875, -18.86956787109375, -28.41071319580078, 60.339935302734375, 7.381233215332031, 405.1249084472656, 175.32766723632812, -35.6741943359375, -188.77593994140625, -108.30844116210938, 39.80924606323242, -266.02520751953125, 108.9276123046875, -1.1878814697265625, -189.93202209472656, -174.35511779785156, 251.43202209472656, 241.321533203125, -190.64781188964844, 70.46141052246094, 197.1549072265625, 186.41738891601562, -14.69329833984375, 62.563690185546875, 47.16705322265625, 137.00543212890625, 145.17160034179688, 128.25613403320312, 178.95391845703125, 40.47528076171875, 22.14410400390625, 208.4906005859375, 208.1053466796875, -108.44886016845703, -9.19854736328125, 392.6793212890625, 109.59469604492188, 250.47039794921875, 157.11996459960938, 131.9276885986328, 134.2970733642578, 9.385612487792969, 44.294708251953125, -9.231225967407227, 252.5589141845703, 489.9683837890625, 375.3243408203125, 261.83758544921875, 213.444091796875, 261.1968994140625, 48.01507568359375, 223.99072265625, -144.64659118652344, -10.331592559814453, -83.25149536132812, -24.14410400390625, -2.192474365234375, 333.80242919921875, 348.56103515625, 285.60107421875, 29.10650634765625, 29.417617797851562, 184.91871643066406, -75.8014907836914, -20.517852783203125, -53.900115966796875, 57.55402374267578, 82.22123718261719, 6.459598541259766, -26.098358154296875, 148.71961975097656, 124.1795654296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000293.npy"}
{"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 80.16456604003906, "std": 175.4775390625, "min": -532.7521362304688, "p10": -76.71023178100586, "median": 61.64799499511719, "p90": 281.0738861083984, "max": 788.0152587890625, "pos_frac": 0.6640625, "sample": [224.11563110351562, 147.4195556640625, 225.60336303710938, -100.60964965820312, 67.34228515625, -26.18340301513672, -1.8273811340332031, -102.3450927734375, 134.709716796875, 345.374267578125, 101.3619384765625, 120.04827880859375, 81.96673583984375, -75.7769775390625, 130.77389526367188, 16.17822265625, 109.04267883300781, 129.51364135742188, 283.7177429199219, 100.46612548828125, 148.45602416992188, 173.85110473632812, 24.049575805664062, -27.16259765625, 5.16448974609375, 261.67413330078125, -42.36534118652344, -128.29188537597656, 124.90289306640625, 18.972557067871094, 457.8851318359375, -52.039825439453125, 34.43128967285156, -220.843994140625, -16.485992431640625, 229.29232788085938, -532.7521362304688, -198.47601318359375, 172.8359375, -31.272872924804688, -67.31036376953125, 250.31524658203125, -43.38922119140625, -44.05620574951172, 242.61264038085938, 283.27874755859375, 162.7128143310547, 125.33033752441406, -78.88782501220703, 120.99099731445312, 282.23486328125, 13.164039611816406, 403.178466796875, 788.0152587890625, 109.64263153076172, 627.4149780273438, 169.96282958984375, -65.88876342773438, 54.804443359375, 2.2970199584960938, 56.9012451171875, 80.617919921875, 34.254669189453125, 354.4786376953125, -164.28982543945312, 40.533077239990234, 180.4304656982422, 122.19303894042969, 6.9893341064453125, -26.6673583984375, 67.18017578125, 450.5132751464844, 118.7818603515625, 0.0, 280.5763244628906, 253.17071533203125, -40.878326416015625, -68.26724243164062, 27.81032943725586, 34.26824951171875, -3.9186744689941406, 236.30096435546875, -9.419794082641602, -242.46925354003906, -57.927001953125, 328.30377197265625, -31.415557861328125, 28.29987335205078, 58.434906005859375, -60.184234619140625, 64.861083984375, 53.102935791015625, -50.51347351074219, -203.25048828125, -69.53926849365234, 120.06332397460938, 27.052520751953125, -45.67425537109375, 41.2886962890625, -176.6165008544922, 251.10679626464844, 149.66854858398438, 273.50811767578125, 143.8965301513672, 113.64822387695312, 194.2449493408203, -370.0106201171875, -52.9554443359375, 332.6302490234375, -0.5967025756835938, 27.69756317138672, 446.7905578613281, 185.68496704101562, 144.439453125, 100.50546264648438, 74.27545166015625, 0.7190399169921875, -34.449424743652344, 151.734130859375, -72.3449478149414, 177.14825439453125, -26.656982421875, 176.12770080566406, -19.345396041870117, 146.33697509765625, -131.52947998046875, 170.3332977294922, 243.94175720214844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000294.npy"}
{"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 82.98999786376953, "std": 158.7803497314453, "min": -268.9536437988281, "p10": -90.62001037597656, "median": 45.29022216796875, "p90": 262.1526733398437, "max": 747.7430419921875, "pos_frac": 0.6875, "sample": [264.46270751953125, 157.91250610351562, -204.32083129882812, 19.818893432617188, -4.5448150634765625, 7.726593017578125, 51.577369689941406, 176.66201782226562, -183.51174926757812, -223.41122436523438, 25.335960388183594, -33.750343322753906, 259.8602600097656, 48.24749755859375, -37.286773681640625, -35.02173614501953, -237.53085327148438, 177.33981323242188, -109.52767944335938, 1.8104934692382812, 101.0802001953125, 252.44854736328125, -68.632568359375, -268.9536437988281, -32.0299072265625, 63.113746643066406, 70.1942138671875, -173.59164428710938, 11.820404052734375, 136.1765594482422, 26.007659912109375, 173.8043975830078, 271.0255126953125, 284.95623779296875, 171.7604522705078, -17.592620849609375, 45.363311767578125, 74.95050048828125, 239.15213012695312, 24.213958740234375, 374.57373046875, -13.343757629394531, 272.48388671875, 194.081298828125, -89.67037963867188, 39.156585693359375, 5.450675964355469, 239.74105834960938, 203.61328125, 205.03240966796875, -135.95298767089844, -55.099403381347656, 36.71783447265625, 747.7430419921875, 249.15138244628906, 210.89642333984375, -138.9673614501953, 92.01822662353516, 127.04144287109375, 193.4483642578125, 62.17315673828125, -14.836181640625, 45.217132568359375, 39.792442321777344, 251.23507690429688, -45.511878967285156, -12.855911254882812, 45.07818603515625, -0.32509613037109375, 47.12548828125, 26.918739318847656, 24.00836181640625, 5.5338897705078125, 376.8895263671875, -54.212493896484375, 186.42015075683594, 243.31222534179688, 125.76683807373047, 15.229827880859375, 137.18377685546875, 73.08856201171875, 170.95462036132812, -76.67424774169922, 85.68299865722656, -2.2238922119140625, 73.82940673828125, 0.0, 182.5825653076172, 88.5032958984375, 22.7808837890625, 489.0225830078125, 255.26409912109375, 156.30450439453125, 40.55413818359375, 164.64013671875, -12.065048217773438, -22.39202880859375, 236.284423828125, -53.50550842285156, 213.7492218017578, 181.01759338378906, 4.7354583740234375, -12.234245300292969, -42.77227783203125, 291.9722900390625, -166.98220825195312, 148.45693969726562, 26.729934692382812, -10.355789184570312, -142.10292053222656, -97.14149475097656, 463.5246276855469, 261.16265869140625, 176.22824096679688, 19.007553100585938, 334.3829345703125, 22.810192108154297, -16.910919189453125, 65.515380859375, -92.8358154296875, -84.70913696289062, -39.802001953125, 191.53756713867188, 341.5169677734375, 227.76071166992188, 445.4683837890625, 43.010406494140625, 227.9732666015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000295.npy"}
{"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 107.07852172851562, "std": 182.18301391601562, "min": -424.7726745605469, "p10": -110.85243530273436, "median": 91.57671356201172, "p90": 340.88836364746095, "max": 626.3002319335938, "pos_frac": 0.71875, "sample": [32.43743133544922, 25.546297073364258, 89.4912109375, -15.554367065429688, 236.06503295898438, 9.191802978515625, -44.724578857421875, -33.273277282714844, -0.478912353515625, 16.257896423339844, 125.79196166992188, -158.67254638671875, 450.2461853027344, 271.457763671875, -1.85430908203125, 141.20831298828125, 238.1399383544922, 208.19696044921875, 223.83657836914062, -108.84686279296875, 434.70263671875, 27.654449462890625, 73.7529296875, 204.26438903808594, 90.07952880859375, 212.25003051757812, 1.9650993347167969, 150.22445678710938, 257.6237487792969, -177.25274658203125, 8.547134399414062, 9.482925415039062, 344.1517333984375, -25.86530303955078, -81.14413452148438, 194.04541015625, 323.46734619140625, -56.32783508300781, 222.10142517089844, 18.453468322753906, -242.3173828125, 162.4031982421875, 269.43975830078125, 176.8431854248047, 49.95252990722656, -115.5321044921875, 197.25469970703125, -194.25103759765625, 248.9688720703125, 84.36843872070312, 499.13140869140625, 156.2085418701172, -85.61737060546875, 166.2928466796875, 626.3002319335938, 262.7395935058594, 7.632148742675781, 364.97943115234375, -65.251708984375, 272.12939453125, -169.20632934570312, 217.8572998046875, 137.96734619140625, -153.96490478515625, 380.0289306640625, 92.28947448730469, 165.27850341796875, 280.86602783203125, -188.94967651367188, 198.176025390625, 344.73028564453125, 75.0219955444336, -317.7573547363281, 309.9641418457031, -29.3125, 165.9156036376953, 339.4897766113281, 443.55792236328125, -205.12901306152344, 147.29010009765625, 211.1727294921875, 539.0528564453125, -51.99781799316406, -75.76260375976562, 384.29541015625, 180.2918701171875, -90.29524230957031, -424.7726745605469, 161.21914672851562, -14.5546875, 201.4891357421875, -54.822784423828125, 226.45867919921875, -52.693939208984375, 218.691650390625, 0.0, 84.70001220703125, 154.01730346679688, 149.09854125976562, 246.87060546875, 90.86395263671875, 92.841796875, 372.75372314453125, 297.0946960449219, 77.05171966552734, 33.2852783203125, 82.30636596679688, 38.25732421875, -29.49310302734375, 337.38836669921875, 223.91018676757812, -6.753654479980469, 34.544227600097656, -282.4429931640625, -8.861419677734375, 189.35354614257812, 163.9185791015625, 41.25465393066406, 22.446685791015625, 270.79449462890625, 13.301567077636719, 109.20068359375, 65.90966796875, 254.75418090820312, 89.78034973144531, 377.24847412109375, -147.1146240234375, -102.43046569824219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000296.npy"}
{"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 83.7637710571289, "std": 152.9705352783203, "min": -227.8441162109375, "p10": -75.0080192565918, "median": 56.04084396362305, "p90": 273.00660400390626, "max": 606.385986328125, "pos_frac": 0.671875, "sample": [453.6732177734375, 209.5187530517578, 213.98892211914062, 207.06103515625, 74.42465209960938, 146.81585693359375, 392.9852600097656, 334.1289978027344, 289.84423828125, -92.59274291992188, -160.1531219482422, -12.771087646484375, -28.406761169433594, 8.250213623046875, 273.4122314453125, -188.90884399414062, -14.16258430480957, 50.598907470703125, 508.0726318359375, 228.62254333496094, 26.433181762695312, 81.25172424316406, -78.07776641845703, 227.77386474609375, 64.34088134765625, 54.1888427734375, -59.0399169921875, -211.44046020507812, 36.67828369140625, -44.66332244873047, -32.95225524902344, 140.92926025390625, 141.97195434570312, -22.39148712158203, -70.49063110351562, -9.689544677734375, 198.91104125976562, 116.55465698242188, 217.12191772460938, -129.73651123046875, 272.832763671875, -73.69241333007812, 105.77586364746094, -35.689453125, 2.292724609375, -33.967254638671875, -203.78121948242188, 169.03750610351562, -84.0675048828125, 233.4468994140625, -48.588531494140625, -44.87139892578125, -16.12872314453125, 40.61004638671875, 29.71746826171875, 253.04537963867188, 191.6843719482422, 35.57665252685547, -26.77198028564453, -227.8441162109375, 110.64876556396484, 89.21835327148438, -224.69003295898438, -106.94459533691406, 33.448333740234375, 137.92138671875, -58.681610107421875, 21.6351318359375, 386.02044677734375, -20.047927856445312, 290.7183837890625, 189.6220703125, 187.47613525390625, 167.57754516601562, 280.47381591796875, -138.98089599609375, 34.1745491027832, 179.8170166015625, 13.31304931640625, -11.568876266479492, -218.71267700195312, 155.32504272460938, 216.08795166015625, -35.69841003417969, 106.84751892089844, 43.039398193359375, 94.1307373046875, 17.85076904296875, 228.223876953125, 123.73880004882812, 144.12445068359375, 18.547210693359375, 35.58550262451172, 115.07267761230469, 32.70111083984375, 221.71275329589844, 108.39360046386719, 259.2962646484375, 57.892845153808594, 189.60260009765625, 366.7024841308594, 108.9078369140625, -34.408843994140625, 163.69151306152344, -36.42456817626953, 91.113037109375, 48.3433837890625, -6.4837646484375, 95.06932830810547, 248.52371215820312, 337.6217041015625, 372.59490966796875, -56.75962829589844, -58.283203125, 11.771240234375, -11.891258239746094, -29.22552490234375, 136.67413330078125, 194.9320068359375, 44.090362548828125, 82.303466796875, 40.0223388671875, 268.68572998046875, -49.995208740234375, -15.70709228515625, 147.75924682617188, 100.14350891113281, 606.385986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000297.npy"}
{"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 96.22921752929688, "std": 174.04383850097656, "min": -363.8456115722656, "p10": -100.20215148925779, "median": 77.97760009765625, "p90": 322.90557861328125, "max": 660.709716796875, "pos_frac": 0.6796875, "sample": [233.62989807128906, 207.93722534179688, -42.81890869140625, 84.38180541992188, -282.83148193359375, -148.2589111328125, -120.957275390625, -179.9937744140625, -28.2476806640625, 289.2145690917969, 133.6009521484375, -58.87406921386719, -14.001480102539062, 83.65264892578125, 31.771255493164062, -8.397003173828125, 217.64471435546875, 199.1181640625, 252.76470947265625, 191.00555419921875, 207.5972900390625, 296.03314208984375, 55.53802490234375, 379.05731201171875, -165.797607421875, -63.303955078125, 250.51303100585938, -228.77374267578125, 166.04379272460938, 247.7470703125, 435.5124206542969, -2.3287353515625, 135.79481506347656, 53.3267822265625, 108.2874755859375, 47.43536376953125, 65.86384582519531, 240.85501098632812, 321.3834228515625, 0.33713531494140625, 139.25912475585938, 63.806854248046875, 167.07293701171875, -211.13140869140625, 262.5093688964844, 432.80511474609375, 87.06402587890625, 137.34881591796875, -175.66952514648438, -126.79757690429688, -35.39154815673828, 473.0682373046875, 142.88153076171875, 159.58737182617188, -194.73822021484375, 64.56278991699219, 108.36268615722656, 258.98748779296875, 107.83624267578125, 276.4912414550781, 424.31475830078125, -363.8456115722656, 52.97796630859375, -141.16192626953125, 14.18806266784668, 229.45761108398438, 190.22137451171875, 14.204833984375, -48.612213134765625, 0.0, 43.23564529418945, 20.8597412109375, 660.709716796875, -11.688568115234375, 109.26094055175781, -31.644065856933594, 150.98193359375, 140.7659149169922, 52.30429458618164, 179.50189208984375, -82.27027893066406, -10.894563674926758, 24.269041061401367, 184.8367156982422, 208.94168090820312, 61.748138427734375, 335.75701904296875, 390.70623779296875, 389.4635009765625, 141.46038818359375, 326.457275390625, -26.9755859375, -38.75624084472656, -47.7294921875, -6.232353210449219, 129.48577880859375, 51.5826416015625, 7.786712646484375, -33.05320739746094, 72.30255126953125, -91.30709838867188, 85.701904296875, 145.12359619140625, 257.33404541015625, 266.224365234375, 168.86572265625, -11.730316162109375, 151.09170532226562, 442.1932373046875, 196.05136108398438, 393.0760803222656, 16.5118408203125, 28.598285675048828, 58.504112243652344, 175.51873779296875, -23.10992431640625, -60.56500244140625, 317.6198425292969, -213.81036376953125, 462.76165771484375, -5.522998809814453, 162.38253784179688, 136.34140014648438, 8.923675537109375, -39.51116943359375, -85.2225341796875, -28.91925811767578, -90.08189392089844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000298.npy"}
{"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 71.8763427734375, "std": 149.8361053466797, "min": -358.10565185546875, "p10": -76.07451019287107, "median": 50.43840026855469, "p90": 248.36773681640622, "max": 552.5098876953125, "pos_frac": 0.7265625, "sample": [167.16758728027344, 406.3135986328125, 67.0125961303711, 70.12814331054688, 29.6766357421875, -161.27667236328125, 143.6978759765625, -64.15673828125, 87.063232421875, -312.339111328125, 552.5098876953125, 310.37353515625, 120.29204559326172, 128.27313232421875, -63.5252685546875, 51.87139892578125, -53.19891357421875, 140.00384521484375, -5.547149658203125, -358.10565185546875, 299.7315368652344, -5.65338134765625, 236.5528564453125, -12.679229736328125, 124.3560791015625, -61.370880126953125, -204.96006774902344, 20.7882080078125, 66.84528350830078, -4.8645782470703125, 303.91888427734375, 5.30780029296875, 184.65969848632812, 25.977134704589844, -331.17193603515625, 245.11679077148438, 194.11923217773438, -17.227203369140625, 31.989036560058594, 73.969482421875, 177.33163452148438, 514.31591796875, 18.03504180908203, -18.43044662475586, 13.762832641601562, 25.05145263671875, 130.38894653320312, -40.816307067871094, 28.444412231445312, -28.436546325683594, -163.61468505859375, 64.61013793945312, 110.2115707397461, 46.9468994140625, 50.489501953125, -71.16061401367188, 221.00006103515625, -155.82470703125, 179.482421875, 31.78398895263672, 116.22445678710938, 99.60516357421875, 12.59710693359375, 94.90069580078125, 8.571674346923828, 314.88543701171875, 19.547698974609375, -34.283363342285156, 44.27679443359375, 50.387298583984375, 191.60787963867188, 30.892887115478516, 103.62579345703125, -7.643890380859375, 139.63735961914062, 114.22179412841797, 112.1805419921875, 48.08430480957031, 38.22697448730469, 165.5269775390625, 29.093353271484375, 149.77545166015625, 32.143592834472656, 39.2371826171875, -39.65321350097656, 39.049652099609375, 444.1767883300781, 173.7978515625, 119.83592224121094, -87.54026794433594, -17.619171142578125, 67.78836059570312, 241.86924743652344, -65.62322998046875, -121.12869262695312, -4.6497955322265625, 64.01978302001953, 84.74971771240234, 240.03866577148438, -108.40485382080078, -116.43841552734375, 32.76129150390625, 30.854110717773438, 26.688705444335938, -4.755218505859375, 478.6681823730469, 71.15219116210938, 268.18243408203125, 255.95327758789062, -1.762847900390625, 306.4358215332031, 14.97268295288086, 131.6644287109375, -226.07257080078125, 18.78936767578125, 167.4993896484375, 82.9923095703125, 54.195281982421875, 162.390380859375, 289.8509216308594, -110.83914947509766, -31.888641357421875, 221.807373046875, 174.37469482421875, 85.86053466796875, 212.44476318359375, 15.84765625, 79.301513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000299.npy"}
{"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 80.2927017211914, "std": 169.56658935546875, "min": -401.28497314453125, "p10": -120.15323638916013, "median": 66.0776138305664, "p90": 265.04175720214846, "max": 608.0199584960938, "pos_frac": 0.7109375, "sample": [254.11813354492188, 97.70211791992188, 19.465965270996094, -201.96664428710938, 31.59154510498047, -142.36135864257812, 266.1669006347656, -59.1546630859375, 178.12579345703125, -14.08056640625, 255.5523681640625, 313.5528564453125, 158.04345703125, 2.64300537109375, 29.995756149291992, 592.3628540039062, -262.34063720703125, 72.17001342773438, 38.1424560546875, -37.762939453125, 440.0522155761719, 185.52471923828125, 40.301971435546875, 154.31422424316406, 245.50860595703125, 149.15335083007812, 200.89474487304688, 29.51178741455078, 71.4388427734375, -112.36509704589844, 182.15631103515625, 69.9530029296875, 67.53963470458984, -43.54119873046875, 265.77642822265625, -105.13114929199219, -350.4720458984375, 110.74700927734375, 12.219240188598633, 264.7268981933594, 206.420654296875, -42.777557373046875, -216.46923828125, 16.9267578125, 336.28631591796875, 212.6671142578125, -109.91377258300781, 64.61559295654297, 100.06028747558594, 373.22283935546875, 40.578433990478516, -99.88111877441406, -163.76541137695312, 258.12689208984375, 35.03144836425781, 12.551929473876953, 165.8663330078125, 63.526580810546875, 170.740234375, -15.673500061035156, 161.76092529296875, 57.9810791015625, -201.09625244140625, 191.765380859375, 38.773223876953125, 27.86370849609375, -16.593505859375, 125.39337158203125, 115.84593200683594, 252.99481201171875, 95.59478759765625, 41.21684265136719, 6.457849502563477, -3.8132781982421875, -27.89539337158203, 13.41571044921875, 331.0958251953125, 101.3489990234375, 43.14970397949219, 170.81500244140625, 461.9228820800781, 113.2293930053711, 92.19329833984375, 83.17178344726562, -18.659561157226562, 202.8436279296875, -14.30853271484375, 0.0, -25.29925537109375, 36.72941589355469, 357.77294921875, 99.41107177734375, -401.28497314453125, 169.27517700195312, 137.86776733398438, 401.94561767578125, 93.30908966064453, 18.28851318359375, 257.75823974609375, 157.3106689453125, 110.11502075195312, 382.2279052734375, 38.66047668457031, -19.484832763671875, -142.27572631835938, 260.9154968261719, 237.51788330078125, -138.3255615234375, -193.588623046875, -276.21893310546875, 131.65704345703125, -8.180032730102539, 64.18922424316406, -37.01251220703125, 144.7481689453125, 126.32156372070312, 213.36801147460938, -54.45710754394531, -111.79533386230469, -184.44044494628906, 234.94497680664062, 146.4120330810547, -25.36920166015625, -80.10968017578125, 172.81866455078125, 608.0199584960938, 15.866897583007812, 32.972198486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000300.npy"}
{"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 80.64683532714844, "std": 152.41665649414062, "min": -348.93878173828125, "p10": -72.00808410644531, "median": 58.81925582885742, "p90": 291.9726989746094, "max": 501.3045654296875, "pos_frac": 0.6875, "sample": [-8.872528076171875, 399.0711364746094, 105.68838500976562, 163.9478759765625, 30.09912872314453, 359.2393493652344, 13.14889907836914, -69.90316772460938, 63.07940673828125, 1.1056785583496094, 3.9058570861816406, -58.611083984375, 164.65545654296875, 56.544525146484375, 66.92138671875, 39.93470001220703, 27.878707885742188, -79.69027709960938, 304.366943359375, 238.35812377929688, 59.499839782714844, 65.16357421875, 211.03106689453125, 57.753082275390625, 79.81668090820312, 30.17327880859375, 7.139617919921875, 501.3045654296875, 154.16172790527344, 121.09524536132812, 75.86785888671875, -26.91595458984375, 0.0, 96.8892822265625, 111.87095642089844, 147.75680541992188, -20.5338134765625, 466.3223876953125, 11.038505554199219, 218.12164306640625, -1.7637271881103516, 201.42205810546875, 331.53680419921875, 144.2525634765625, -172.21575927734375, 90.65570068359375, -133.60879516601562, 175.790283203125, -108.86972045898438, 47.3924560546875, 201.33822631835938, -39.4083251953125, -14.192474365234375, 290.912353515625, 240.5496826171875, 97.53667449951172, 113.490478515625, 241.47854614257812, -61.72222900390625, 417.80230712890625, -54.51922607421875, 157.83673095703125, -6.678741455078125, 63.6334228515625, -42.358642578125, 1.3358955383300781, 163.79373168945312, -348.93878173828125, -272.7192687988281, 6.471427917480469, 5.156303405761719, 222.91539001464844, -64.64761352539062, 19.55080795288086, 155.68666076660156, 58.138671875, 176.50567626953125, -27.04119873046875, 243.97723388671875, -212.7589111328125, 48.9141845703125, -39.90033721923828, 359.7673645019531, 159.76129150390625, 271.6688232421875, 16.44482421875, -41.765132904052734, -31.76788330078125, 134.7911376953125, -96.70252990722656, 376.4031982421875, 459.29931640625, -31.648345947265625, -61.50965881347656, 200.155029296875, -68.81898498535156, 16.993202209472656, 103.27420043945312, 92.81187438964844, -26.7933349609375, 236.16360473632812, 170.85910034179688, 294.44683837890625, -191.74127197265625, 137.7891082763672, 193.96560668945312, 84.9532470703125, 52.23393249511719, 306.43365478515625, -81.46638488769531, 119.19490051269531, 3.0101699829101562, 355.8892822265625, -117.32763671875, -44.47843933105469, 165.4384307861328, -217.70028686523438, -34.576416015625, 16.457733154296875, 171.98294067382812, 232.4979705810547, -76.9195556640625, -27.6973876953125, -2.4814987182617188, 141.21157836914062, 95.17083740234375, -0.6766033172607422, 2.641876220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000301.npy"}
{"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 102.34239959716797, "std": 158.54037475585938, "min": -283.4809875488281, "p10": -57.71110305786132, "median": 88.53132629394531, "p90": 310.0191772460937, "max": 589.7647705078125, "pos_frac": 0.703125, "sample": [152.85452270507812, 234.475341796875, 272.8924560546875, 154.37294006347656, 62.833255767822266, 157.70880126953125, 0.0, 66.07269287109375, 45.89764404296875, 51.147003173828125, -224.496826171875, 362.9404296875, 266.7393493652344, 275.114501953125, -116.32492065429688, 144.06558227539062, -239.99139404296875, 102.64727783203125, -20.509002685546875, -61.599273681640625, 507.63714599609375, 243.81903076171875, 392.9403076171875, -17.536865234375, 134.450439453125, 44.51971435546875, 282.13604736328125, 76.02050018310547, -47.6072998046875, 131.30015563964844, 199.92291259765625, 303.8951416015625, 171.677001953125, 256.9101867675781, 128.71615600585938, -66.4578857421875, 427.52752685546875, 136.53823852539062, 224.91798400878906, 62.82958984375, -10.44671630859375, -44.439910888671875, -218.4537353515625, 101.87554931640625, 176.7403106689453, 125.70292663574219, 280.22503662109375, 125.49322509765625, 229.89935302734375, 339.263916015625, -41.65936279296875, 172.8091278076172, 93.72896575927734, 25.19842529296875, 130.08367919921875, -29.8104248046875, 70.7275390625, 62.0302734375, 212.41412353515625, -11.572830200195312, 176.1104736328125, 109.32014465332031, 156.51611328125, 98.87161254882812, 88.83203125, -29.94708251953125, -231.34463500976562, 8.773967742919922, 0.488311767578125, -14.46841049194336, 192.29562377929688, 106.2183837890625, 112.15353393554688, -82.36843872070312, 409.432373046875, 212.36468505859375, -27.101600646972656, 207.5072021484375, 241.56301879882812, -36.58076477050781, 31.34002685546875, 14.021759033203125, 325.65185546875, -283.4809875488281, 589.7647705078125, 80.99429321289062, -13.44085693359375, -12.952781677246094, -131.9378662109375, -18.898391723632812, 117.34961700439453, -30.14886474609375, -145.160888671875, 34.84027099609375, -11.525211334228516, 403.7552490234375, 31.094375610351562, 353.266357421875, 265.35821533203125, -56.220970153808594, 31.8658447265625, -29.89111328125, -44.574127197265625, 80.63262176513672, 296.5755310058594, -7.402019500732422, 88.23062133789062, 255.43014526367188, 25.874296188354492, 180.5116729736328, -61.188079833984375, 8.008407592773438, 117.35577392578125, -48.554901123046875, 216.93942260742188, -106.52775573730469, 324.30859375, 3.33306884765625, 292.2772521972656, -9.147125244140625, 62.72642517089844, 344.55224609375, 203.42015075683594, 109.75575256347656, -10.537948608398438, 10.694091796875, 436.32489013671875, 15.72064208984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000302.npy"}
{"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 90.06295776367188, "std": 149.99276733398438, "min": -388.47344970703125, "p10": -89.32413177490233, "median": 74.80796813964844, "p90": 273.1098327636719, "max": 612.1708984375, "pos_frac": 0.7578125, "sample": [54.142303466796875, 36.683563232421875, -245.600830078125, -250.2040252685547, 297.2476501464844, 67.53789520263672, 270.7352294921875, 340.8506774902344, 172.919921875, 177.2479705810547, -99.35919189453125, -25.541160583496094, 227.98733520507812, 268.5152587890625, 72.224365234375, -60.519775390625, -165.16116333007812, 43.327945709228516, -11.031084060668945, 183.28253173828125, 10.08447265625, -20.50128173828125, 66.78533172607422, 54.185516357421875, 74.34234619140625, 133.4130859375, 74.1910629272461, 217.23907470703125, 291.9039306640625, 55.016029357910156, 278.65057373046875, 177.44091796875, -183.64329528808594, 289.73614501953125, -110.48153686523438, 52.29864501953125, -25.881072998046875, 75.41574096679688, 0.0, 22.03990936279297, 3.2484588623046875, 40.54046630859375, -32.73173522949219, 55.080322265625, 52.90085220336914, 35.517578125, 215.28192138671875, 76.89694213867188, 178.98057556152344, 188.24757385253906, 180.4529571533203, 3.00360107421875, 143.7015380859375, -260.8997497558594, 267.369873046875, 344.0889587402344, 75.27359008789062, 3.3549423217773438, 136.38427734375, 161.02455139160156, 4.239662170410156, -66.61524963378906, -22.3045654296875, -37.1827392578125, -126.987548828125, 147.74652099609375, 99.10049438476562, 26.094375610351562, 346.1261291503906, 172.3833465576172, 241.2572021484375, -29.306686401367188, -49.46929931640625, -85.02339172363281, 170.25888061523438, 189.45541381835938, 22.7698974609375, 192.30889892578125, 247.66116333007812, 329.560302734375, 219.38232421875, -388.47344970703125, 226.78018188476562, -125.98047637939453, 108.83939361572266, 88.83763122558594, 317.95281982421875, 117.5526123046875, -5.584136962890625, 116.215087890625, 189.62847900390625, -137.57080078125, 204.80694580078125, 55.79389953613281, 257.6708984375, -0.26849365234375, 42.08599090576172, -45.75196075439453, -30.45330810546875, 160.03334045410156, -144.300048828125, -172.876708984375, 112.9971923828125, 15.50628662109375, 36.882057189941406, 92.0020523071289, 180.78021240234375, 223.26754760742188, 343.76055908203125, 196.81991577148438, 70.60348510742188, 193.06317138671875, 18.229930877685547, 89.72085571289062, 11.905227661132812, 260.96258544921875, 207.20445251464844, -41.048583984375, 117.05856323242188, 149.50314331054688, 612.1708984375, 23.14208984375, 364.8218994140625, 7.2726287841796875, 133.13043212890625, 45.971771240234375, 150.8558349609375, 329.8424377441406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000303.npy"}
{"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 98.43850708007812, "std": 172.043212890625, "min": -306.41156005859375, "p10": -93.9564971923828, "median": 75.55384063720703, "p90": 296.24471130371086, "max": 680.0532836914062, "pos_frac": 0.71875, "sample": [0.0, 49.8206787109375, -70.72140502929688, 29.693214416503906, -210.86209106445312, 7.54840087890625, -111.0737075805664, 165.35003662109375, 118.609130859375, -121.5228271484375, 141.54051208496094, 97.40228271484375, 123.74528503417969, -17.907272338867188, -173.6468505859375, 235.36019897460938, -5.779815673828125, 253.1420135498047, 213.98281860351562, 15.8590087890625, 131.43148803710938, 59.07693862915039, 56.05853271484375, -115.35758972167969, 129.09844970703125, 236.39735412597656, 165.2733154296875, 150.14654541015625, 84.43052673339844, 2.50860595703125, -6.747711181640625, 218.0859832763672, 188.67669677734375, 24.489181518554688, 459.3752746582031, 272.6728210449219, 148.78196716308594, -33.16144561767578, 401.1034240722656, 139.23138427734375, 147.7879638671875, 100.6111068725586, 398.8326110839844, -80.87872314453125, 232.51043701171875, 36.05888366699219, -125.85118103027344, 30.10589599609375, -114.9169921875, 213.46270751953125, 414.0250549316406, -71.68940734863281, 108.16278076171875, -168.85113525390625, 0.0, 218.1578826904297, -139.99566650390625, -0.1601409912109375, 4.292327880859375, 167.00070190429688, -75.46514892578125, 2.6041717529296875, 35.077606201171875, 230.42855834960938, 72.54774475097656, 43.655372619628906, -306.41156005859375, 49.967041015625, 312.6413269042969, 557.8004760742188, -104.69158935546875, 93.94610595703125, 186.47860717773438, -85.21337890625, -66.03802490234375, 579.65625, 260.4552307128906, -15.300655364990234, 197.58074951171875, -115.49728393554688, 388.282958984375, 125.26741027832031, 37.47435760498047, -26.438383102416992, 633.137451171875, -19.02960205078125, 680.0532836914062, 8.3525390625, 196.27197265625, -144.6610107421875, 92.08538818359375, -5.274557113647461, 11.867645263671875, 99.6678466796875, 585.5374755859375, 36.554359436035156, 14.799041748046875, 132.20623779296875, 167.86073303222656, 103.81497192382812, 21.564208984375, 161.28594970703125, 66.29222106933594, 59.49766540527344, 54.664039611816406, 138.97244262695312, 72.10836791992188, 356.1941833496094, 158.57203674316406, 289.21759033203125, 156.70616149902344, 78.5599365234375, -87.518798828125, 39.092437744140625, 204.44281005859375, 91.41616821289062, -66.7564697265625, -89.35574340820312, -10.684844970703125, 325.3836364746094, 92.77957153320312, 176.99072265625, -82.86373901367188, 149.8597412109375, -32.817657470703125, 185.70115661621094, 36.365325927734375, 261.63104248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000304.npy"}
{"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 150.35633850097656, "std": 182.65370178222656, "min": -390.04791259765625, "p10": -27.04372100830078, "median": 124.865478515625, "p90": 409.217788696289, "max": 883.541259765625, "pos_frac": 0.8515625, "sample": [206.42945861816406, 138.2960968017578, 43.84119415283203, 293.63226318359375, 24.891357421875, -3.09295654296875, 58.73169708251953, -121.38118743896484, 72.42472839355469, -15.65911865234375, 123.33074951171875, 70.64556884765625, -42.62335205078125, 203.81735229492188, 13.226211547851562, 35.13287353515625, 72.20184326171875, 67.12765502929688, 263.4905700683594, 32.87448501586914, 102.00531005859375, 420.0975036621094, 284.63055419921875, 108.19900512695312, 15.955947875976562, 21.410964965820312, 93.4957275390625, -2.8119354248046875, 42.00396728515625, 75.2109375, 69.52508544921875, 235.74029541015625, 165.45211791992188, 238.6177978515625, 278.0602111816406, 63.023223876953125, 72.81687927246094, -29.74291229248047, 285.77655029296875, 183.563720703125, -201.17681884765625, 141.31454467773438, 34.23589324951172, 201.08761596679688, 601.121826171875, 264.138671875, 583.7435913085938, 240.50180053710938, 119.48095703125, 111.40599060058594, -162.6917724609375, -167.71917724609375, 213.73724365234375, 468.2929992675781, 306.4664306640625, 135.46502685546875, 883.541259765625, 92.412841796875, 287.8699951171875, 147.09136962890625, 181.90036010742188, 143.92579650878906, 77.01031494140625, 267.3499755859375, 71.81204223632812, -25.886924743652344, 121.98101806640625, -7.136466979980469, 574.5535888671875, 243.1350555419922, 2.62701416015625, 313.22100830078125, -169.2559051513672, 402.7994079589844, 161.43072509765625, -390.04791259765625, 28.889114379882812, 131.5181427001953, -85.66909790039062, 126.40020751953125, -40.966339111328125, 466.94879150390625, 271.5329895019531, 62.44245147705078, 112.12169647216797, 196.67018127441406, -75.072265625, 568.1689453125, 24.08831787109375, 52.045013427734375, 404.5550537109375, 155.11505126953125, 91.83477783203125, 181.127197265625, 225.14501953125, 134.0047607421875, 425.50225830078125, 289.8260498046875, 573.8192138671875, 138.3330078125, 226.13455200195312, 36.15203857421875, 275.7437744140625, 186.60678100585938, 320.914306640625, 174.28131103515625, 97.93377685546875, 43.36181640625, 188.50198364257812, 173.10406494140625, 219.2196044921875, -138.55892944335938, 115.34078216552734, 4.08135986328125, 92.80369567871094, 81.17162322998047, 239.02731323242188, 47.28876495361328, 115.97927856445312, 53.270477294921875, 463.532958984375, 160.96466064453125, 424.51947021484375, 251.55801391601562, 447.42535400390625, -171.04087829589844, -1.10821533203125, 128.91262817382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000305.npy"}
{"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 88.58779907226562, "std": 151.86984252929688, "min": -215.28628540039062, "p10": -101.062451171875, "median": 67.33971786499023, "p90": 295.29445800781247, "max": 572.1976318359375, "pos_frac": 0.71875, "sample": [95.22671508789062, -35.65863037109375, 60.34736633300781, 257.30657958984375, -142.83132934570312, 459.35394287109375, 136.03787231445312, 18.706314086914062, -164.26019287109375, 301.53289794921875, -180.7340545654297, 57.2108154296875, 348.36676025390625, 89.02078247070312, -90.10588073730469, -11.30987548828125, -109.7005615234375, 306.7523193359375, -26.938560485839844, 44.009521484375, 75.57614135742188, 130.0001983642578, -105.9580078125, 3.53704833984375, -139.10989379882812, 270.9154968261719, 383.157958984375, 167.77194213867188, 79.85885620117188, 25.45697021484375, 124.17108154296875, 67.71441650390625, -75.138671875, 174.07354736328125, 119.57705688476562, 154.04019165039062, 104.81591796875, 364.7191162109375, 149.77658081054688, 58.43048095703125, 16.849700927734375, 60.83620834350586, -5.77532958984375, 44.81596374511719, 134.22845458984375, 124.39826965332031, 18.614776611328125, 63.05059814453125, 55.655242919921875, -205.501220703125, -171.82589721679688, 252.3421630859375, 24.751678466796875, 287.4239501953125, -86.3990478515625, -215.28628540039062, 26.24847412109375, 224.31768798828125, -60.83000946044922, 277.710693359375, -42.1939697265625, 0.63238525390625, 287.40460205078125, -1.000457763671875, 217.5193328857422, 28.417387008666992, 106.02079772949219, 22.798755645751953, 322.93621826171875, -100.09700012207031, 66.96501922607422, -188.66055297851562, 69.6134033203125, 101.6424560546875, 292.732666015625, -0.5661544799804688, 137.169677734375, 36.57905197143555, -3.6925048828125, -53.406005859375, 158.398193359375, 83.34707641601562, 113.95037841796875, 176.3282470703125, 276.573974609375, -66.70896911621094, 72.9417724609375, 172.4232177734375, 63.8260498046875, 282.1385498046875, 184.63339233398438, 301.27197265625, 175.7840576171875, -63.343994140625, -12.411224365234375, 185.5571746826172, 123.114501953125, -13.684173583984375, 285.051025390625, 3.006601333618164, 450.70501708984375, 165.869873046875, 572.1976318359375, 246.07437133789062, 23.83800506591797, 82.08979797363281, 78.30018615722656, -3.827850341796875, 6.1056365966796875, -151.61404418945312, -60.14638900756836, 8.40594482421875, 187.50772094726562, 335.119384765625, 105.2525634765625, 347.2529296875, -141.9056854248047, -27.7027587890625, -57.38104248046875, 51.3653564453125, 5.381927490234375, 217.52096557617188, -103.31517028808594, -4.657707214355469, 63.171287536621094, 246.23681640625, 309.17291259765625, 173.86212158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000306.npy"}
{"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 89.41481018066406, "std": 148.9650115966797, "min": -300.576416015625, "p10": -84.46238403320312, "median": 77.16853713989258, "p90": 266.2088684082031, "max": 461.8968505859375, "pos_frac": 0.7734375, "sample": [148.93887329101562, 373.32794189453125, 73.57501220703125, 271.50872802734375, -5.33251953125, -21.646821975708008, 35.37966537475586, 175.7984619140625, 162.74172973632812, 262.93853759765625, -258.58209228515625, 27.816749572753906, 184.61117553710938, -214.78721618652344, -84.0093994140625, 139.91168212890625, 461.8968505859375, 273.2791748046875, 86.64578247070312, 91.41644287109375, 449.1084899902344, -247.42440795898438, 72.04043579101562, -2.4280853271484375, 263.9375, 53.46258544921875, 53.21173095703125, 159.42507934570312, 13.87994384765625, 44.490264892578125, 98.65904235839844, 77.79901123046875, 282.0738220214844, 207.29888916015625, 68.4788818359375, -24.59912109375, 209.423828125, 117.9112548828125, 165.45611572265625, 132.45867919921875, -42.10919189453125, 42.20851135253906, 61.63452911376953, -23.47772216796875, 27.58935546875, -50.08527374267578, 50.47047424316406, -22.199787139892578, 39.14849853515625, 223.44287109375, 11.0556640625, 76.63949584960938, 185.082763671875, 62.79118347167969, 54.28239440917969, 194.2833709716797, 152.09371948242188, 235.7000274658203, 90.11753845214844, -45.82311248779297, 38.3865966796875, 141.89962768554688, -201.1591796875, 70.03720092773438, 27.693721771240234, -174.56796264648438, 260.1422119140625, 346.9293518066406, 77.69757843017578, 3.0369110107421875, 19.47933006286621, 10.101966857910156, 9.91156005859375, -112.7279052734375, -82.99017333984375, 4.545623779296875, 100.47882080078125, 295.016357421875, 114.7786865234375, 109.02603149414062, 123.48704528808594, -300.576416015625, 4.047473907470703, 167.15078735351562, 188.67486572265625, 286.4854736328125, 48.674346923828125, 250.24276733398438, 37.76622009277344, 114.32293701171875, 129.18247985839844, -182.49432373046875, 102.85615539550781, 320.5457458496094, 102.33139038085938, 4.387969970703125, 155.20794677734375, -8.243843078613281, -170.25845336914062, 191.74041748046875, -85.51934814453125, 366.8201904296875, 47.436798095703125, 258.7353820800781, -278.2401123046875, -111.83843994140625, -32.223297119140625, 262.375, 238.02108764648438, -29.330528259277344, -76.38226318359375, 44.004150390625, 458.432861328125, 142.82669067382812, 139.52078247070312, 176.73361206054688, 149.41822814941406, -107.49270629882812, 44.279266357421875, 224.06460571289062, 42.141326904296875, 27.29167938232422, 232.12686157226562, 181.14962768554688, 213.210693359375, -7.899635314941406, 379.36834716796875, 219.87921142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000307.npy"}
{"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 63.276702880859375, "std": 163.28646850585938, "min": -458.6989440917969, "p10": -129.78673706054684, "median": 57.15200424194336, "p90": 254.5478744506836, "max": 759.5361328125, "pos_frac": 0.6796875, "sample": [272.9486083984375, 90.2655258178711, -70.7320556640625, 101.4526596069336, 158.384033203125, -104.54022216796875, -9.21728515625, 156.5885009765625, 102.09713745117188, 3.608367919921875, 13.07794189453125, 169.57562255859375, -181.7781219482422, -240.96475219726562, -39.26519775390625, 65.98419189453125, 83.0843505859375, 56.36271667480469, 209.87777709960938, 161.73614501953125, 243.86892700195312, 83.06146240234375, 97.92599487304688, -30.464637756347656, 71.93167114257812, 155.42803955078125, -458.6989440917969, -6.863517761230469, 66.18756103515625, 270.2067565917969, 185.24729919433594, -49.6605224609375, -24.993408203125, 145.18014526367188, -167.67681884765625, -99.16329956054688, 295.92022705078125, 41.935203552246094, 76.84359741210938, -4.170074462890625, 5.4331512451171875, 46.46556854248047, 335.5552978515625, -123.50048828125, -287.416259765625, 51.232879638671875, -10.92364501953125, 155.10369873046875, 4.973108291625977, 227.40675354003906, 22.70928955078125, 33.141395568847656, -8.8096923828125, 237.6330108642578, -356.6277770996094, 82.34991455078125, 254.44183349609375, -24.29351043701172, -219.259521484375, 149.37582397460938, -16.615394592285156, -1.67022705078125, -49.6064453125, -223.47186279296875, 462.5043640136719, 36.570281982421875, -2.5557422637939453, -176.793701171875, 142.66368103027344, 14.427513122558594, 67.915771484375, 254.79530334472656, 314.70440673828125, 98.73115539550781, 148.63833618164062, 97.1240234375, 395.96697998046875, 206.70880126953125, 759.5361328125, -32.000732421875, 48.277923583984375, 173.5059814453125, -5.648933410644531, -185.36941528320312, 149.21664428710938, -78.24179077148438, 142.85671997070312, 239.31951904296875, -144.45465087890625, 114.51258087158203, 56.07696533203125, 41.279727935791016, -39.481689453125, 91.320068359375, 57.94129180908203, 128.74783325195312, 286.82562255859375, 314.4687805175781, 59.837982177734375, 168.09933471679688, -13.861968994140625, 151.6405029296875, 381.4762878417969, -48.49101257324219, 43.060585021972656, 26.00299072265625, 88.0396728515625, 14.539798736572266, 92.44639587402344, 23.720916748046875, -246.81182861328125, 55.882843017578125, 79.89306640625, 44.66880798339844, 12.968231201171875, -102.46881103515625, 6.3935089111328125, 138.0030517578125, -178.74618530273438, -11.27008056640625, -32.96649169921875, 139.1082000732422, 234.05194091796875, 118.98269653320312, 146.5731201171875, -101.92926025390625, 357.48468017578125, 98.78054809570312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000308.npy"}
{"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 112.13523864746094, "std": 174.41683959960938, "min": -324.5345764160156, "p10": -65.54586105346678, "median": 96.09910583496094, "p90": 323.5437377929687, "max": 739.8079833984375, "pos_frac": 0.7421875, "sample": [-23.367996215820312, 93.83183288574219, 301.135986328125, 262.547607421875, -48.36475372314453, -49.34123229980469, 197.0390625, -9.467052459716797, 167.76168823242188, 103.53778076171875, 152.24102783203125, 2.2603607177734375, -34.73988342285156, 89.78732299804688, -72.29024505615234, 166.56857299804688, 18.713623046875, -73.34037780761719, 35.994895935058594, 228.01150512695312, 0.725982666015625, 138.0904541015625, -6.1234130859375, -324.5345764160156, 33.1234130859375, -27.791290283203125, -44.86315155029297, 381.4869689941406, 101.5537109375, 178.072998046875, -41.902191162109375, 249.70611572265625, 82.61529541015625, 0.130767822265625, -7.3184967041015625, -55.862823486328125, 108.521484375, -97.27809143066406, 435.9346923828125, -234.53857421875, 349.5898742675781, 116.98114013671875, 29.73670196533203, 45.14988708496094, 398.06964111328125, -85.68746185302734, -41.14619445800781, 53.4080810546875, -89.6407470703125, -7.458404541015625, 270.1278076171875, -0.7309360504150391, 77.45648193359375, -176.1324462890625, 122.62860107421875, 101.40411376953125, -39.440826416015625, 17.02496337890625, -4.31732177734375, 139.49639892578125, 327.73638916015625, -0.793121337890625, 193.41705322265625, 500.33984375, -52.28688049316406, 554.41259765625, -62.65541076660156, 56.326141357421875, 262.49200439453125, 121.53732299804688, -94.07179260253906, 257.7608337402344, 0.0649871826171875, 701.4910888671875, 92.8599853515625, 120.65596008300781, 103.99920654296875, 177.22727966308594, 49.990840911865234, 411.1337890625, 129.3846435546875, 313.003662109375, 63.33244323730469, 132.51211547851562, 98.36637878417969, 117.73150634765625, 4.641876220703125, 84.455322265625, 430.241943359375, 31.53691864013672, 186.7947998046875, 22.147247314453125, 15.964645385742188, -167.13296508789062, 220.99801635742188, 321.74688720703125, 211.4758758544922, 82.55713653564453, 106.89346313476562, 26.23896026611328, -147.5643310546875, 116.8946533203125, 179.72158813476562, 237.0341796875, 210.61221313476562, -89.55078125, 225.915771484375, 152.4847412109375, 117.41256713867188, -258.9106750488281, 230.13284301757812, 494.63140869140625, 7.3571319580078125, 236.83236694335938, 74.56068420410156, 281.2970886230469, 140.56317138671875, 739.8079833984375, 38.8287353515625, 267.86395263671875, 307.28350830078125, 136.48828125, 218.66195678710938, -34.602294921875, 203.75955200195312, 53.550323486328125, 19.140520095825195, 383.78582763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000309.npy"}
{"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 108.12503051757812, "std": 169.22793579101562, "min": -291.5356750488281, "p10": -80.15418472290038, "median": 99.53433227539062, "p90": 319.06050109863276, "max": 748.292724609375, "pos_frac": 0.7265625, "sample": [453.38372802734375, 218.22360229492188, -24.08969497680664, 415.5545654296875, -1.172515869140625, 222.460205078125, 130.36062622070312, 187.3323211669922, 227.09884643554688, 43.220916748046875, 176.27183532714844, 457.3333740234375, 186.5848846435547, 209.5972442626953, 227.66912841796875, 294.9249267578125, 71.62710571289062, -31.829797744750977, 332.013427734375, -170.88699340820312, 82.53471374511719, 126.44052124023438, -19.712661743164062, 151.93435668945312, 206.5229034423828, 103.8797607421875, 175.97406005859375, -219.353271484375, 131.61972045898438, 55.799713134765625, 191.46461486816406, -8.180408477783203, -18.59088134765625, 404.03961181640625, -28.61383056640625, -79.1083984375, 100.42024230957031, 171.1444091796875, 41.27435302734375, -104.6263427734375, -2.3941574096679688, -17.78863525390625, 132.302734375, 204.951171875, -102.26386260986328, -55.18304443359375, 259.5762023925781, 127.561279296875, 82.08851623535156, 142.7917022705078, 21.27130126953125, 34.422027587890625, 222.03543090820312, 11.057220458984375, 63.23863220214844, -2.859954833984375, 748.292724609375, 219.29507446289062, -13.07098388671875, -238.72250366210938, 10.644966125488281, 375.9155578613281, 99.68643188476562, 20.97589111328125, -291.5356750488281, -36.36851501464844, -71.40118408203125, 313.5092468261719, 89.38577270507812, 12.547096252441406, -91.76302337646484, -124.38314819335938, -65.65046691894531, 146.95526123046875, 222.29481506347656, 231.554443359375, 196.53475952148438, 398.08038330078125, 282.5859680175781, 11.060256958007812, -235.2998046875, 136.50344848632812, 234.15240478515625, 115.19866943359375, 21.998336791992188, 70.98220825195312, 269.0128479003906, -8.422637939453125, 6.278192520141602, 80.66152954101562, 17.47247314453125, 37.19921875, 179.43336486816406, 109.67041015625, -22.431854248046875, -23.981979370117188, 142.82061767578125, 261.03466796875, -59.383026123046875, 154.83387756347656, 194.80056762695312, 267.96417236328125, -59.373779296875, -188.01776123046875, 264.8985595703125, 152.86358642578125, -35.43513488769531, -153.50286865234375, -194.82749938964844, 77.4049072265625, -82.59435272216797, 51.6400146484375, 99.38223266601562, 195.15548706054688, 179.47738647460938, 8.55865478515625, 354.93365478515625, 304.31829833984375, 83.3968276977539, 0.300537109375, 88.6090087890625, 9.765621185302734, 464.3923034667969, 348.5837097167969, 139.7047119140625, 259.47430419921875, 345.1496276855469, 517.4774780273438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000310.npy"}
{"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 112.09170532226562, "std": 201.39013671875, "min": -365.43707275390625, "p10": -116.81330795288085, "median": 108.39195251464844, "p90": 319.5373596191406, "max": 1252.5028076171875, "pos_frac": 0.75, "sample": [66.04098510742188, -100.03561401367188, 225.794677734375, 127.17263793945312, 327.1494445800781, -24.29762840270996, 482.431884765625, 65.63633728027344, 283.0859375, 195.25018310546875, 286.74481201171875, 165.6112060546875, 50.973106384277344, -18.54914093017578, -101.355712890625, -316.01397705078125, 93.0908203125, 83.30168914794922, 1.4339408874511719, 13.55120849609375, 223.5047607421875, 135.89297485351562, 35.6724853515625, 60.12968444824219, 15.919448852539062, -89.99333190917969, 182.37002563476562, -26.921676635742188, 107.22267150878906, 102.4412841796875, -23.528038024902344, -328.9892578125, -365.43707275390625, 115.40440368652344, 303.3556213378906, 435.38037109375, -54.22894287109375, 3.0762939453125, 132.79806518554688, 147.16220092773438, -150.36927795410156, 229.0071258544922, 260.9230041503906, 357.46343994140625, 18.800914764404297, 310.95257568359375, -207.56732177734375, 175.41500854492188, -243.21051025390625, 188.06695556640625, 109.06132507324219, 68.3133544921875, 95.60037231445312, 257.04168701171875, 158.814453125, 43.9053955078125, 157.72023010253906, 15.34271240234375, -43.599365234375, 188.5164794921875, 596.4406127929688, 268.79876708984375, 324.55609130859375, 150.87429809570312, 381.25335693359375, -38.3109130859375, 100.60028076171875, 22.899459838867188, 119.83739471435547, 117.73724365234375, 266.8266296386719, -57.04190444946289, 173.87677001953125, 41.97308349609375, 162.79678344726562, 107.72257995605469, 277.10784912109375, 195.67572021484375, -36.193450927734375, 374.0567321777344, 332.22943115234375, 43.33056640625, -59.166473388671875, 0.6207809448242188, -29.65350341796875, -114.16085815429688, 317.386474609375, 567.6936645507812, 397.14935302734375, 250.50360107421875, 216.56777954101562, 60.029541015625, 161.72958374023438, 73.19305419921875, 225.38555908203125, -134.6491241455078, 1252.5028076171875, -37.7872314453125, -190.5372772216797, 252.78663635253906, 20.993419647216797, 184.23794555664062, -137.0921630859375, 168.1138916015625, 93.43951416015625, -15.32863998413086, 205.87136840820312, 408.23748779296875, -123.00235748291016, 179.0732421875, 77.34927368164062, 37.058837890625, 5.7725830078125, 183.667236328125, -281.8828125, -75.78721618652344, 185.12176513671875, 220.7881317138672, 146.6126251220703, 21.617767333984375, -126.47831726074219, -17.76457977294922, 190.704345703125, 279.644287109375, 168.1616973876953, 199.81484985351562, 185.71768188476562, -182.01210021972656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000311.npy"}
{"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 96.67532348632812, "std": 220.8535614013672, "min": -555.1392822265625, "p10": -113.55183105468748, "median": 36.945823669433594, "p90": 329.192056274414, "max": 834.4077758789062, "pos_frac": 0.6875, "sample": [303.708251953125, -19.393096923828125, 123.47003173828125, 323.2936706542969, 92.93515014648438, -555.1392822265625, 589.71435546875, -25.481170654296875, 22.670326232910156, 15.456283569335938, 237.094970703125, -109.37113952636719, -127.18673706054688, 196.00543212890625, 227.3214111328125, 834.4077758789062, 147.70655822753906, 176.8418731689453, -167.86355590820312, -127.55416870117188, -6.068603515625, 39.33940124511719, 255.64715576171875, 8.78369140625, -28.231529235839844, 1.7103118896484375, 60.6263427734375, 257.4312744140625, -47.6375732421875, -30.719200134277344, 521.5850830078125, 94.1573257446289, -16.09991455078125, 794.5787963867188, 163.3726806640625, -98.18901062011719, 270.5090637207031, -6.2576751708984375, 34.55224609375, 277.47637939453125, 188.10812377929688, 724.2883911132812, -28.53015899658203, 150.38031005859375, 33.825660705566406, 173.75442504882812, 150.33197021484375, 276.54705810546875, 260.19287109375, 8.23187255859375, 312.14410400390625, 265.0213317871094, -23.126129150390625, 20.86376190185547, 14.88702392578125, 0.344818115234375, 20.262557983398438, -50.295623779296875, 5.4433441162109375, 438.7400207519531, 15.533248901367188, -7.54388427734375, 486.9757995605469, -5.5799560546875, -32.217864990234375, -488.0394287109375, 131.8000946044922, 47.879486083984375, -86.97457885742188, 28.321868896484375, 14.8189697265625, 69.41384887695312, 120.95051574707031, 9.32614517211914, -33.74444580078125, 209.1021728515625, 27.07349395751953, 342.9549560546875, 48.942787170410156, 185.85513305664062, 236.71875, 217.94406127929688, -174.65692138671875, 20.69554901123047, -25.901260375976562, 0.0, 344.1396484375, 168.73318481445312, -74.15289306640625, -327.490234375, -14.898681640625, -9.5540771484375, 90.65863037109375, 121.54718017578125, 227.28692626953125, 17.047142028808594, -394.4140625, 10.883291244506836, 553.2567749023438, -186.75953674316406, 41.21942138671875, 22.62200927734375, 117.3828125, 223.41058349609375, 99.29144287109375, -165.477783203125, 395.5455322265625, 154.9476318359375, 448.8028259277344, 727.3118286132812, 198.5723876953125, -44.659149169921875, 268.62603759765625, -5.560100555419922, -123.30677795410156, 144.52822875976562, 211.79339599609375, -101.78504943847656, -190.96240234375, 206.05349731445312, -284.1212463378906, 310.76702880859375, -70.59481811523438, 63.15496826171875, 174.1418914794922, 5.031402587890625, 6.883996963500977, 8.271148681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000312.npy"}
{"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 93.75078582763672, "std": 218.2593231201172, "min": -325.45330810546875, "p10": -141.857763671875, "median": 72.25006484985352, "p90": 327.4017059326172, "max": 1569.802978515625, "pos_frac": 0.6953125, "sample": [388.3024597167969, 529.7998657226562, -21.297607421875, -98.18302917480469, 17.2110595703125, 19.980941772460938, 19.82012939453125, 155.64309692382812, 321.46368408203125, -110.379150390625, -81.68829345703125, -192.65960693359375, -142.54998779296875, 453.1888122558594, -50.20354461669922, 384.382568359375, 75.33798217773438, 173.36264038085938, 14.13629150390625, -259.7529296875, -26.844085693359375, 55.6959228515625, -2.774150848388672, 143.76791381835938, 141.29095458984375, -78.11822509765625, -242.5218505859375, 77.38986206054688, 45.220550537109375, 294.81494140625, 188.28338623046875, 321.43511962890625, 34.110626220703125, 183.6710205078125, -141.56109619140625, 197.9520263671875, 51.39244842529297, -50.3487548828125, 337.87628173828125, 155.55226135253906, -301.59625244140625, 316.62310791015625, 384.9260559082031, -18.142288208007812, -70.08296966552734, 106.89561462402344, 85.36279296875, -154.54299926757812, 21.419967651367188, 382.097900390625, 142.2132110595703, -4.18116569519043, -143.4215087890625, 23.50444984436035, 52.84075927734375, -32.99534225463867, 530.9808349609375, 105.47039031982422, 35.756988525390625, -6.432220458984375, 178.66012573242188, 94.21063232421875, 175.47576904296875, 21.30133819580078, 72.23651885986328, -43.23826599121094, -15.7835693359375, 168.80844116210938, -78.2514877319336, -66.75680541992188, 315.31793212890625, 84.93141174316406, 326.4930725097656, 85.9140625, -39.98651123046875, 52.38507080078125, 131.92706298828125, 260.97406005859375, -210.51528930664062, 122.56898498535156, 13.821975708007812, -51.700042724609375, 73.14126586914062, -214.99642944335938, 77.37493896484375, 108.38207244873047, 8.580718994140625, -11.992156982421875, 18.58312225341797, 89.21583557128906, 102.91796875, 1569.802978515625, 146.59890747070312, 109.86468505859375, 146.43838500976562, 239.5987091064453, 91.82038879394531, 329.5218505859375, -223.62539672851562, -53.359771728515625, 112.4151611328125, 145.21942138671875, 72.26361083984375, 303.78753662109375, 52.902587890625, 70.30836486816406, 29.946752548217773, -114.8580093383789, 449.6734924316406, 442.5323486328125, 456.82684326171875, 232.41201782226562, 49.23681640625, 58.411041259765625, 27.354053497314453, -53.04777526855469, 191.7670135498047, 288.3604736328125, 36.98333740234375, -325.45330810546875, -56.99542236328125, -238.5247802734375, -156.99078369140625, 217.42031860351562, 119.69549560546875, 235.8983154296875, 175.67019653320312, 231.32470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000313.npy"}
{"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 89.52448272705078, "std": 159.42791748046875, "min": -361.43243408203125, "p10": -74.00508346557616, "median": 73.15666198730469, "p90": 267.5857269287109, "max": 629.9464721679688, "pos_frac": 0.703125, "sample": [46.90553283691406, 240.33737182617188, 233.3787384033203, 116.76266479492188, 229.53244018554688, 101.26568603515625, -130.91091918945312, 61.029815673828125, -68.85140991210938, -20.90643310546875, -289.470947265625, 243.44921875, 594.61376953125, 294.0458679199219, 210.53265380859375, -361.43243408203125, -115.3714599609375, 166.65670776367188, 28.032882690429688, 216.17086791992188, 404.3963928222656, -123.45635986328125, 18.833980560302734, 116.33755493164062, -52.09992980957031, 93.23533630371094, 316.4385986328125, 102.01873779296875, 60.219757080078125, 439.0828857421875, 278.4738464355469, -104.73335266113281, -2.859466552734375, -36.23216247558594, 98.44984436035156, 36.815185546875, 184.82681274414062, 265.52801513671875, 80.90380859375, 405.704345703125, -73.42449951171875, -55.127838134765625, 115.64739990234375, 160.14544677734375, 43.63182830810547, 218.42605590820312, 272.3870544433594, 144.76626586914062, 63.800048828125, -37.29193115234375, 45.3377685546875, 43.419761657714844, 11.430526733398438, 191.52593994140625, 47.15581512451172, 102.73855590820312, 124.44955444335938, 71.10198974609375, 55.60017395019531, -14.847015380859375, 189.6033172607422, 152.98101806640625, 175.29483032226562, 0.0, -172.79803466796875, -251.496826171875, -81.65521240234375, 147.7928466796875, 16.726654052734375, -10.774627685546875, -12.60845947265625, -46.89862060546875, 93.5614013671875, 263.7496643066406, -96.3668212890625, 35.18583679199219, -39.469451904296875, 310.21527099609375, -53.559814453125, -229.93972778320312, 212.783935546875, 182.31533813476562, 58.87705993652344, 342.66021728515625, 0.0, 146.64578247070312, 18.089935302734375, 184.59011840820312, 221.3026123046875, 39.46464538574219, 156.07699584960938, 67.99786376953125, 255.4232177734375, 43.5255126953125, -28.16217041015625, 371.74658203125, 227.33837890625, 39.63523864746094, 106.20465087890625, 24.30572509765625, 223.02023315429688, 19.5499267578125, -27.472076416015625, -241.9920654296875, 91.96054077148438, -39.003631591796875, 162.0076904296875, 192.17620849609375, 96.72586059570312, 133.33863830566406, 401.05047607421875, -65.50374603271484, -71.47088623046875, 111.36373901367188, -12.524322509765625, 629.9464721679688, 171.31015014648438, -75.35977935791016, 25.851219177246094, 152.74822998046875, 75.21133422851562, 38.66563415527344, -54.529510498046875, 149.55807495117188, -47.302490234375, 248.26370239257812, 212.2715301513672, -9.61911392211914], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000314.npy"}
{"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 117.00270080566406, "std": 171.53550720214844, "min": -319.5525817871094, "p10": -73.52851257324218, "median": 103.54548645019531, "p90": 331.7963989257812, "max": 640.4940185546875, "pos_frac": 0.7734375, "sample": [-105.71609497070312, 49.93059539794922, 10.19921875, 28.319290161132812, -104.73844909667969, -60.90826416015625, -72.3084716796875, 103.00613403320312, 159.38189697265625, -56.324462890625, 120.91641998291016, 4.993499755859375, 258.2260437011719, 140.96112060546875, 60.316375732421875, 34.583892822265625, 80.61444091796875, 149.408203125, 107.89093017578125, -169.25376892089844, 17.75229835510254, 12.696044921875, 7.0463409423828125, -188.9027862548828, 295.0018310546875, 305.39276123046875, 144.51576232910156, 198.19467163085938, 196.4381103515625, 220.3212127685547, 120.79962158203125, 107.39877319335938, -44.83575439453125, 421.74517822265625, 192.00621032714844, 228.04348754882812, 145.1373291015625, 370.5621643066406, 42.68278503417969, -47.46502685546875, -31.780670166015625, -82.55557250976562, 34.54339599609375, 640.4940185546875, 27.786758422851562, -7.99407958984375, 48.4261474609375, -129.24716186523438, -39.38616943359375, 275.2213439941406, 177.6378936767578, -57.014713287353516, 483.2906494140625, 243.15188598632812, 308.970703125, 451.8189392089844, 83.17291259765625, 89.06794738769531, 325.22174072265625, 151.0648193359375, 162.92050170898438, 79.8477783203125, -7.732372283935547, 38.56794738769531, 261.3959655761719, 283.5881652832031, 149.63327026367188, 239.82081604003906, -76.37527465820312, 280.8687744140625, -319.5525817871094, 156.62432861328125, 630.03369140625, 176.681396484375, 347.13726806640625, -66.07440185546875, -187.89065551757812, 293.65069580078125, 21.88718605041504, 445.9405517578125, 384.03741455078125, -68.49819946289062, 92.46670532226562, 259.252685546875, 251.4393310546875, 59.134735107421875, 166.62648010253906, 157.43321228027344, 10.98455810546875, 240.4537353515625, 0.2855377197265625, 72.58233642578125, -275.6839294433594, -65.19650268554688, 109.70208740234375, 508.55670166015625, 214.94989013671875, 22.572603225708008, -101.87509155273438, 230.01654052734375, -35.490875244140625, 29.00677490234375, 237.16439819335938, 106.41902923583984, -80.57881164550781, -100.88331604003906, -66.89111328125, 454.372314453125, 13.672353744506836, 131.38812255859375, 125.6590576171875, 87.83589172363281, 157.94432067871094, 428.5303039550781, 46.62908935546875, 104.0848388671875, 269.10791015625, -30.9385986328125, 84.8702392578125, 81.15180969238281, 192.5430908203125, 45.491546630859375, 235.13658142089844, 142.32264709472656, 181.3285675048828, 371.6932678222656, 95.28228759765625, 13.359012603759766], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000315.npy"}
{"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 98.01518249511719, "std": 173.32513427734375, "min": -381.926513671875, "p10": -116.12301788330078, "median": 72.08008575439453, "p90": 342.3649658203125, "max": 606.5385131835938, "pos_frac": 0.7578125, "sample": [269.4974365234375, 40.611568450927734, 18.82025146484375, 201.445068359375, 27.990234375, 78.864501953125, -13.683624267578125, 405.8919982910156, 35.94456481933594, 286.4844970703125, -142.42904663085938, -160.94186401367188, 88.2216796875, -161.85137939453125, -69.05963134765625, 135.0417022705078, 54.4417724609375, -195.19415283203125, 86.72576904296875, 168.36953735351562, 565.3583984375, 252.15249633789062, -72.56195068359375, -4.680335998535156, 189.21728515625, 347.21026611328125, 46.203369140625, 260.8282470703125, 198.60934448242188, 589.3018188476562, 50.81348419189453, 274.6658935546875, -54.910919189453125, 153.70013427734375, 156.1683349609375, 79.09034729003906, 96.55455780029297, 385.467041015625, 125.70831298828125, 38.12986755371094, -133.72021484375, 19.182571411132812, 41.224609375, 153.324462890625, -18.92609405517578, 333.304443359375, 9.159538269042969, 154.11016845703125, -12.5562744140625, 71.70365905761719, 26.691253662109375, -8.65191650390625, -150.621337890625, -305.61419677734375, -58.389892578125, 135.2379150390625, -16.009857177734375, 50.17784118652344, 82.41888427734375, -11.720138549804688, 72.12106323242188, 391.2724304199219, 236.5904541015625, 419.1040954589844, 183.42416381835938, 31.9833984375, -239.32305908203125, 606.5385131835938, 137.28794860839844, 289.2864990234375, 120.41909790039062, -159.13746643066406, 43.24627685546875, 127.57247924804688, 94.56767272949219, 97.03887939453125, 50.70831298828125, 347.201904296875, 20.198898315429688, 191.5271759033203, 386.9853820800781, 126.90276336669922, 280.04583740234375, -96.58145904541016, 14.759416580200195, 25.2333984375, 115.640869140625, 340.2919921875, -381.926513671875, 145.37664794921875, -111.81085205078125, 42.3258056640625, 0.0, -197.28573608398438, 189.90985107421875, 283.4847412109375, 160.19384765625, 27.34295654296875, 4.4305572509765625, 287.2613220214844, 56.54437255859375, 72.03910827636719, 12.374805450439453, -21.125045776367188, -40.902313232421875, 146.79769897460938, 423.6145935058594, 29.679489135742188, 144.5390625, 128.042236328125, 408.08319091796875, 12.9825439453125, 354.88665771484375, -117.7025146484375, 49.30812072753906, 138.8455810546875, -115.44609069824219, -193.46630859375, 227.0660400390625, 56.864715576171875, 205.80984497070312, 34.301109313964844, 9.705810546875, 52.80059814453125, 219.00338745117188, 123.9019775390625, 260.0409240722656, -29.398712158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000316.npy"}
{"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 93.41114807128906, "std": 143.5781707763672, "min": -423.88336181640625, "p10": -42.30798645019531, "median": 79.93685531616211, "p90": 275.58888854980466, "max": 509.50567626953125, "pos_frac": 0.75, "sample": [156.049072265625, -147.8546142578125, 0.0, 83.5733871459961, 292.23193359375, -84.02971649169922, 358.16668701171875, 204.12606811523438, 202.89031982421875, 187.356689453125, 21.25177001953125, 122.11674499511719, 229.11843872070312, -10.601364135742188, 275.4482727050781, -143.36837768554688, -72.17698669433594, 243.03903198242188, 196.2264404296875, 103.8143081665039, 209.63665771484375, -0.34521484375, 221.5691680908203, 92.093505859375, 23.62574005126953, 75.29185485839844, 11.555610656738281, 0.34932518005371094, -19.964385986328125, 66.708740234375, -25.754770278930664, 154.45196533203125, 457.9014587402344, -29.89678955078125, -41.897857666015625, 26.069549560546875, 139.10089111328125, 251.77244567871094, 25.041275024414062, -80.11746215820312, 203.01463317871094, -76.90672302246094, 473.06463623046875, -26.454010009765625, -24.410247802734375, 334.019287109375, 181.064208984375, 217.62982177734375, 17.344444274902344, 0.0, 11.863792419433594, 141.55401611328125, 136.6262664794922, 324.16729736328125, -147.12925720214844, -12.835952758789062, 120.6055908203125, 119.86349487304688, 34.33567810058594, 11.1695556640625, 99.84811401367188, 199.85549926757812, 273.27862548828125, 9.646564483642578, 0.0, 4.36602783203125, 25.4222412109375, 42.35200500488281, 123.56112670898438, 275.9169921875, 183.44847106933594, -34.85850524902344, -34.36174011230469, -423.88336181640625, 320.08740234375, 509.50567626953125, 346.55426025390625, -73.685791015625, 37.08467102050781, -203.0504913330078, -20.43255615234375, 194.1719512939453, 202.95030212402344, 119.03369903564453, 45.131103515625, 38.6429443359375, 122.34859466552734, 33.83892822265625, 175.84811401367188, 361.19183349609375, 1.6893310546875, 192.14077758789062, 144.77978515625, -43.26495361328125, 96.24176025390625, -3.604978561401367, -10.20831298828125, 208.42214965820312, 161.53041076660156, 244.0174560546875, -106.17378234863281, 159.1207275390625, 194.4527587890625, 76.30032348632812, -247.85617065429688, 96.63275146484375, 69.29156494140625, 177.42578125, -4.059087753295898, -14.35028076171875, 14.106109619140625, 41.993003845214844, 158.71697998046875, 25.519607543945312, 96.66336059570312, 85.15911865234375, 48.19371795654297, 1.836538314819336, 301.0476379394531, 102.12519073486328, 30.375329971313477, 103.85865783691406, 19.11822509765625, 157.982666015625, 54.97156524658203, 180.86366271972656, 0.0579986572265625, 374.5399169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000317.npy"}
{"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 114.99917602539062, "std": 165.18643188476562, "min": -276.61285400390625, "p10": -63.38826599121092, "median": 98.10998916625977, "p90": 343.2532043457031, "max": 707.81494140625, "pos_frac": 0.78125, "sample": [213.84349060058594, 192.4083251953125, 124.56951141357422, 181.5052032470703, 23.961090087890625, 233.21157836914062, 5.9677734375, 93.82064819335938, 707.81494140625, 388.71038818359375, 47.196990966796875, -128.55307006835938, 322.5077209472656, 86.56666564941406, 309.54595947265625, 4.345184326171875, 134.05722045898438, 43.8126220703125, 38.67637634277344, 116.91201782226562, -46.31292724609375, -104.98974609375, 45.038330078125, 129.86883544921875, -25.631195068359375, 81.84149169921875, 207.18414306640625, 167.802490234375, -59.6583251953125, 349.90435791015625, 149.75698852539062, -87.27249145507812, -26.133209228515625, 15.591354370117188, -99.69071960449219, -80.4903564453125, 3.164020538330078, 283.4230651855469, 446.733642578125, 152.8746337890625, 38.7103271484375, 202.8271484375, 461.715576171875, 74.38775634765625, 30.614456176757812, 24.01678466796875, -233.43505859375, 8.575790405273438, 18.012969970703125, 165.8953857421875, -73.791015625, 111.913818359375, 109.4031982421875, 164.56707763671875, 181.1818389892578, 340.4027099609375, 12.307014465332031, 231.2223663330078, 334.63720703125, -25.0858154296875, 48.214359283447266, 28.893585205078125, 404.77850341796875, 126.29289245605469, -47.28007888793945, 52.15985107421875, 122.29150390625, 84.12930297851562, 429.1548156738281, -96.94525146484375, 212.37661743164062, 84.19525146484375, -20.436798095703125, 80.44666290283203, 235.00006103515625, 146.4068603515625, -52.882110595703125, 137.38504028320312, 132.64715576171875, 23.5235595703125, 194.2435302734375, 74.21588134765625, 330.49725341796875, 115.34513854980469, 75.67549133300781, -57.37591552734375, 225.53274536132812, 300.82012939453125, -126.90386199951172, -42.765380859375, 19.36151123046875, 176.41921997070312, 119.158447265625, 102.39933013916016, 143.64508056640625, 120.29193115234375, 410.7122802734375, 65.796142578125, 537.992431640625, -36.593292236328125, 120.86022186279297, 284.326904296875, 545.8641357421875, 157.32858276367188, -276.61285400390625, 245.63372802734375, 107.08551025390625, 30.2952880859375, 9.647994995117188, -72.09146118164062, 373.2157287597656, 110.2738037109375, 59.57672119140625, 72.87277221679688, 11.559837341308594, 448.75994873046875, 130.45175170898438, 362.64013671875, -54.44384765625, 224.40774536132812, -162.95440673828125, -32.7574462890625, 87.77043151855469, 231.55838012695312, 202.3212127685547, -2.751495361328125, -195.72930908203125, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000318.npy"}
{"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 105.83576202392578, "std": 146.7064666748047, "min": -533.3651733398438, "p10": -63.86032409667969, "median": 117.57242202758789, "p90": 267.6774536132812, "max": 537.87255859375, "pos_frac": 0.7890625, "sample": [197.21127319335938, -40.275421142578125, 291.0020751953125, 4.391082763671875, 128.38311767578125, 348.4922790527344, 262.55621337890625, -63.6842041015625, 507.3442077636719, 316.0841064453125, -49.33056640625, -21.40229034423828, 59.823883056640625, 154.4569091796875, 316.8692626953125, 229.07559204101562, -44.0836181640625, 0.0, 11.79583740234375, 129.48013305664062, 37.35129165649414, 121.582275390625, 95.61015319824219, 68.32518005371094, 216.08456420898438, 216.14816284179688, 12.021209716796875, 12.708511352539062, 1.80810546875, 150.6563720703125, 160.7276611328125, -21.568641662597656, 221.61276245117188, 91.16300201416016, 145.29690551757812, 239.81875610351562, -4.2331695556640625, 537.87255859375, 160.11611938476562, 100.89586639404297, 42.705665588378906, 122.3277587890625, 150.9554443359375, -120.46615600585938, 95.43864440917969, 145.71575927734375, 137.32037353515625, 178.90121459960938, -120.18878173828125, 248.7283935546875, 49.36279296875, 361.7669677734375, 188.6259307861328, -171.21652221679688, 107.892333984375, 111.67282104492188, 58.11651611328125, -32.55414581298828, 154.2409210205078, 166.67567443847656, -233.09130859375, 179.2216339111328, 154.89942932128906, -55.2933349609375, 118.93621063232422, 35.46678161621094, 279.62701416015625, -114.76902770996094, 169.36984252929688, 161.25204467773438, -99.84586334228516, 164.56121826171875, 251.55426025390625, 110.5858154296875, -17.17828369140625, -79.79165649414062, 58.59747314453125, 225.5768280029297, 351.9736633300781, -82.525146484375, 228.8672637939453, 168.80450439453125, 206.980224609375, 429.98211669921875, 128.07244873046875, 27.849929809570312, 424.885498046875, 344.3861083984375, -64.27127075195312, 37.516937255859375, 10.361572265625, 103.63992309570312, 38.04791259765625, 42.7742919921875, 260.82220458984375, 23.389144897460938, 165.53909301757812, 282.75689697265625, -108.50497436523438, 160.8314208984375, 61.5911865234375, -28.84765625, 178.71372985839844, 84.19631958007812, 116.20863342285156, 153.29827880859375, 171.79249572753906, 150.10052490234375, 168.64707946777344, -43.16694259643555, 67.1619873046875, -11.888853073120117, -130.21078491210938, 255.6014404296875, -533.3651733398438, 204.48330688476562, 141.98956298828125, 175.24578857421875, -102.05077362060547, 45.9036865234375, 149.1395263671875, 51.76214599609375, 100.09929656982422, 4.7232818603515625, 64.3726806640625, 135.52294921875, 74.58428955078125, 171.30169677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000319.npy"}
{"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 102.767578125, "std": 179.47215270996094, "min": -320.0545654296875, "p10": -87.03452377319334, "median": 72.98417663574219, "p90": 318.986978149414, "max": 726.4624633789062, "pos_frac": 0.734375, "sample": [73.1112060546875, 14.09050178527832, -68.11026000976562, -311.6057434082031, 213.9471435546875, 251.19094848632812, -2.08953857421875, 56.057403564453125, 474.2610778808594, 182.8790283203125, -26.301361083984375, 123.02032470703125, -132.62847900390625, 78.2147216796875, 273.10443115234375, 644.9982299804688, -43.29925537109375, 12.648067474365234, 16.13043212890625, 82.09681701660156, -1.7796249389648438, 228.3760986328125, 285.63702392578125, -50.581634521484375, 61.504127502441406, -97.78262329101562, 177.126953125, 257.33148193359375, 10.400970458984375, -193.1734619140625, 223.18707275390625, 130.33026123046875, 376.65380859375, 59.23388671875, 36.9224853515625, 409.304443359375, 72.060546875, 217.4241485595703, 122.29180908203125, 1.3980789184570312, -5.742910385131836, 13.207515716552734, 206.82601928710938, -95.75727081298828, 280.5558776855469, 382.91461181640625, 49.817626953125, 239.24526977539062, 76.603271484375, 280.517822265625, -74.15252685546875, 11.08197021484375, -108.54815673828125, -16.51898193359375, 240.3393096923828, 72.85714721679688, 192.01654052734375, 5.9860992431640625, 100.06846618652344, 323.6087341308594, 167.20281982421875, 70.09039306640625, -35.18798828125, 317.0062255859375, 230.3148193359375, 94.82260131835938, 362.5966491699219, -41.0037841796875, 259.1383056640625, 351.60784912109375, 67.9541244506836, 1.208099365234375, 172.06460571289062, -9.300689697265625, -220.18157958984375, 85.35720825195312, 22.351531982421875, 198.42041015625, 93.66064453125, 285.7939453125, 225.04708862304688, 26.63481903076172, -4.132417678833008, 74.78704833984375, -67.02586364746094, 170.4277801513672, 1.6806182861328125, 23.552703857421875, 141.6612548828125, -43.35475158691406, 26.600906372070312, -32.8717041015625, 182.023681640625, -83.29620361328125, 82.45355224609375, -64.67811584472656, 122.4207763671875, 4.909889221191406, -151.95059204101562, 207.67733764648438, 592.3470458984375, -4.267578125, 331.98870849609375, 150.67015075683594, 30.06243896484375, 365.064208984375, 236.73703002929688, -103.1233901977539, 50.9202880859375, -191.51840209960938, 0.1161041259765625, 234.2666015625, -320.0545654296875, -27.300621032714844, 666.0794677734375, 96.77536010742188, 298.66162109375, -70.683837890625, 80.09693908691406, 79.70243835449219, 67.18096923828125, 115.64187622070312, 66.40779113769531, -118.17787170410156, 726.4624633789062, 22.344078063964844, -169.00811767578125, 217.86741638183594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000320.npy"}
{"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 75.57539367675781, "std": 175.4705352783203, "min": -372.28759765625, "p10": -121.40768280029296, "median": 36.41254425048828, "p90": 306.57692871093747, "max": 832.98681640625, "pos_frac": 0.6796875, "sample": [234.61001586914062, -124.33139038085938, 113.97944641113281, -281.9102783203125, 30.418701171875, 249.67709350585938, 11.2120361328125, -208.4423828125, -112.55166625976562, 208.399169921875, 222.57351684570312, 76.43208312988281, -27.414443969726562, 19.113128662109375, 372.6708984375, 83.704345703125, 0.41860198974609375, 251.42539978027344, 345.87921142578125, 33.9515380859375, -11.16749382019043, -229.54208374023438, -25.619285583496094, 28.31756591796875, 365.06768798828125, -52.081085205078125, -50.40101623535156, 34.464691162109375, 200.13746643066406, -123.897705078125, 61.05548858642578, 162.12857055664062, -39.55847930908203, 131.19070434570312, 200.92218017578125, -204.84323120117188, -52.5467529296875, 96.20806121826172, 117.19775390625, 93.8292236328125, 153.25767517089844, 34.864990234375, -40.03424072265625, 10.588043212890625, -195.416015625, 364.41058349609375, 16.77935791015625, 369.10406494140625, 352.76080322265625, 23.054916381835938, 148.59780883789062, 79.58074951171875, -29.164222717285156, 172.32171630859375, 13.881423950195312, -21.329879760742188, 281.53057861328125, -74.77227783203125, 168.09750366210938, -42.675262451171875, 61.740150451660156, -59.95008087158203, 90.68678283691406, 286.4334716796875, 334.1812744140625, 242.6471710205078, 15.101959228515625, 25.21826171875, 34.57403564453125, -148.17568969726562, 18.044677734375, 173.94140625, -107.53199768066406, 249.4730224609375, -36.55238342285156, -65.35139465332031, 253.2498779296875, 832.98681640625, 76.0028076171875, 66.24102783203125, 74.478515625, -180.852783203125, 141.76333618164062, 315.58660888671875, 2.7017822265625, 142.22042846679688, 12.987930297851562, 302.71563720703125, -278.8901062011719, 184.19369506835938, -20.894290924072266, 149.9530029296875, -138.7554931640625, 442.36688232421875, -79.73881530761719, 580.3659057617188, 6.45538330078125, 85.0758056640625, 248.63729858398438, 124.43194580078125, 149.97218322753906, -21.53912353515625, -24.78497314453125, 105.23271179199219, 6.576446533203125, 175.41293334960938, -51.21244430541992, -14.419809341430664, 20.573585510253906, -120.34053039550781, -163.9824981689453, -40.62774658203125, 43.46968078613281, -78.20549011230469, 22.396522521972656, 286.6680908203125, -372.28759765625, 319.9519348144531, 69.57550048828125, 37.96009826660156, 38.64642333984375, 137.898193359375, 173.54788208007812, -20.557838439941406, 423.28607177734375, 117.0919189453125, 0.0, 9.46832275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000321.npy"}
{"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 88.03853607177734, "std": 183.09329223632812, "min": -602.5040893554688, "p10": -69.25960998535156, "median": 52.58604431152344, "p90": 286.8599853515625, "max": 714.4212646484375, "pos_frac": 0.6484375, "sample": [-602.5040893554688, 51.8228759765625, 7.12017822265625, 168.37428283691406, -2.393035888671875, 47.455474853515625, 156.4025421142578, -170.71893310546875, 235.59263610839844, 400.56524658203125, -122.94485473632812, 22.795989990234375, 245.69064331054688, 32.5794677734375, 242.3848419189453, 234.7267608642578, -38.123779296875, 66.41546630859375, 15.835609436035156, 336.1509094238281, 239.32537841796875, -169.06295776367188, 179.48587036132812, -45.376312255859375, -35.88520812988281, 198.0850830078125, -40.724700927734375, 250.92005920410156, 91.46794128417969, 216.5830078125, -23.924240112304688, 32.15791702270508, 30.025665283203125, -24.759552001953125, 167.30474853515625, 80.84564208984375, -6.8250274658203125, 208.13900756835938, -74.4979248046875, 228.78338623046875, 164.06344604492188, 175.27301025390625, 53.349212646484375, -66.97540283203125, 221.43975830078125, -42.129852294921875, 169.57046508789062, -67.01461791992188, 390.92572021484375, -45.59532928466797, 113.4964599609375, -20.101409912109375, 287.09881591796875, 584.5911865234375, -20.36651611328125, 246.490966796875, 76.98639678955078, -35.517791748046875, 116.30120849609375, 22.13262939453125, 211.59613037109375, -57.1519775390625, 268.7062072753906, 269.02886962890625, -37.086334228515625, 225.30307006835938, -146.68634033203125, 180.16265869140625, 92.68461608886719, -61.88832092285156, 272.34344482421875, 392.72393798828125, 70.32073974609375, 133.99505615234375, 19.905853271484375, 0.0, -143.72055053710938, 117.630126953125, 41.834747314453125, -230.2337646484375, 425.6242370605469, -10.412174224853516, 36.242431640625, 427.5679931640625, -187.6688690185547, 218.72674560546875, 178.2337646484375, -30.70501708984375, 141.88156127929688, 93.51667785644531, 286.75762939453125, -34.496917724609375, -42.41770935058594, -183.60589599609375, -40.71527099609375, 50.126556396484375, -56.057861328125, 39.8450927734375, -467.3179626464844, 213.7559814453125, 43.3145751953125, -26.740509033203125, -2.72467041015625, 342.8448486328125, 155.12860107421875, -124.00173950195312, -53.35211181640625, 496.74603271484375, 418.16632080078125, 6.8529510498046875, -21.304092407226562, -38.89918518066406, 133.83021545410156, 178.32266235351562, -199.44253540039062, 98.35592651367188, -50.672393798828125, 32.4896240234375, 184.108642578125, 8.393455505371094, 126.77267456054688, 354.1404113769531, 212.51707458496094, -12.27716064453125, 48.45086669921875, 714.4212646484375, 242.619384765625, 169.20751953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000322.npy"}
{"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 70.4774169921875, "std": 180.4899444580078, "min": -328.57244873046875, "p10": -151.06874389648436, "median": 69.84213256835938, "p90": 253.5990524291992, "max": 870.4642333984375, "pos_frac": 0.6796875, "sample": [94.31791687011719, 39.484161376953125, -248.02252197265625, -46.164031982421875, 170.8902130126953, 157.9663848876953, -1.412139892578125, -125.6195297241211, 157.374755859375, 263.2239990234375, -167.01553344726562, 377.1082763671875, 65.3267822265625, -59.7222900390625, -57.1568603515625, 44.53520965576172, 221.2207489013672, -67.215087890625, -246.8985595703125, -254.60955810546875, 274.908203125, 207.42111206054688, -69.71987915039062, 72.3267822265625, 99.51335144042969, 539.5419921875, 79.9107666015625, 49.05595779418945, -252.1492156982422, 120.39334106445312, -56.79119873046875, -220.66815185546875, -0.6401596069335938, 127.42536163330078, 5.219429016113281, 119.72079467773438, 103.66667175292969, -20.646900177001953, 85.6666259765625, 170.3480682373047, 67.35748291015625, 209.34799194335938, 116.33779907226562, 525.193359375, 260.405029296875, 20.171279907226562, -187.0918731689453, -328.57244873046875, 151.16250610351562, 5.9737701416015625, 9.806589126586914, 227.85720825195312, -170.88201904296875, 31.172344207763672, 173.4222412109375, 65.07546997070312, 135.1744384765625, 209.4969940185547, 223.24667358398438, -52.02595520019531, 298.94146728515625, 90.537841796875, 208.4573974609375, -50.2332763671875, 202.74560546875, -19.686996459960938, 215.54232788085938, -72.92913818359375, -219.82196044921875, 238.5626220703125, 47.59832763671875, 20.20390510559082, 157.23004150390625, 36.73394775390625, 250.6822052001953, 101.86270141601562, 195.19415283203125, 60.24235534667969, 207.62936401367188, -30.80682373046875, 280.8378601074219, -269.41046142578125, 181.19325256347656, 18.75433349609375, 89.3406982421875, 474.0102844238281, -284.85650634765625, 128.1031494140625, -148.02935791015625, 34.171485900878906, 176.8017120361328, 126.35614013671875, 27.164047241210938, 118.18489074707031, 80.85906982421875, -148.0015106201172, 84.01254272460938, 28.51885986328125, 113.07418823242188, -99.02841186523438, -26.0732421875, 21.88885498046875, -81.78656005859375, 180.92062377929688, -36.2396240234375, 218.57386779785156, 93.21967315673828, -60.770904541015625, 110.2525634765625, 870.4642333984375, 125.97406005859375, -105.37156677246094, 126.48872375488281, 583.3818969726562, -111.12854766845703, -158.16064453125, -21.425949096679688, 34.933563232421875, 114.77433013916016, -45.69764709472656, -77.79383087158203, 56.839752197265625, 95.43563842773438, 278.5182800292969, 222.28750610351562, -135.11251831054688, 49.025184631347656, 302.2052001953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000323.npy"}
{"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 119.06622314453125, "std": 165.3207244873047, "min": -328.0999450683594, "p10": -36.44272766113281, "median": 112.49603271484375, "p90": 352.29444885253906, "max": 573.7601318359375, "pos_frac": 0.796875, "sample": [547.2505493164062, 573.7601318359375, 24.33355712890625, 79.18689727783203, 60.65789794921875, 166.895263671875, 38.343658447265625, 147.2060089111328, 263.2276611328125, 146.2703857421875, 70.62835693359375, -10.098464965820312, 47.477264404296875, 365.20458984375, 356.8729248046875, 178.26742553710938, 297.68841552734375, 44.985836029052734, -328.0999450683594, 0.0, 33.062744140625, -3.1263427734375, 139.11080932617188, 21.89706802368164, 39.5859260559082, 125.22714233398438, 141.16807556152344, -228.00802612304688, 144.91668701171875, 40.00410461425781, 7.595222473144531, 350.3322448730469, 390.7451171875, 321.57568359375, 76.03587341308594, 291.2039794921875, -70.66380310058594, 121.6837158203125, 161.92755126953125, 0.0, 8.449630737304688, 103.308349609375, 521.65185546875, 158.67904663085938, 80.66232299804688, 280.05413818359375, 0.0, -70.37471008300781, 187.5118408203125, -193.751220703125, -33.78169250488281, 271.49468994140625, -12.85760498046875, 27.38641357421875, -14.02740478515625, 10.349769592285156, 139.003173828125, 213.65716552734375, 242.27359008789062, 144.05517578125, -221.6672821044922, 403.91259765625, 59.552978515625, 288.55224609375, -21.753273010253906, -42.65180969238281, 222.01296997070312, 136.41445922851562, 68.58139038085938, -49.080482482910156, 186.37466430664062, 79.50297546386719, 8.6978759765625, 179.24853515625, -106.16110229492188, 3.3702392578125, 82.59977722167969, 73.53390502929688, -10.860816955566406, 401.4271240234375, 410.78533935546875, 13.70450210571289, 147.38583374023438, 72.79409790039062, 554.15283203125, 362.69488525390625, 178.95565795898438, 231.10986328125, 57.803375244140625, -6.484375, 237.37489318847656, 182.24215698242188, -196.27212524414062, 205.70947265625, 28.123138427734375, 141.3809814453125, 288.1749572753906, 129.217041015625, 196.40994262695312, 126.88623046875, -182.64773559570312, -147.07138061523438, 7.93109130859375, 15.567359924316406, -302.8395080566406, 271.70294189453125, 296.6666259765625, 246.07684326171875, 25.0712890625, 75.875244140625, 382.318603515625, 217.30364990234375, 35.88520812988281, 62.052215576171875, -27.864471435546875, 368.32904052734375, 205.49346923828125, 237.7725830078125, 26.239713668823242, 192.2165069580078, 151.9241485595703, -20.01373291015625, 24.589630126953125, 225.73178100585938, 80.86387634277344, 166.16510009765625, 121.90130615234375, 141.40188598632812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000324.npy"}
{"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 101.17729187011719, "std": 205.92294311523438, "min": -634.5320434570312, "p10": -117.1172119140625, "median": 70.58073425292969, "p90": 350.5997436523437, "max": 1047.3648681640625, "pos_frac": 0.7421875, "sample": [135.03662109375, 51.935791015625, -106.90678405761719, 10.9378662109375, 123.7373046875, 37.51422119140625, 27.540283203125, 354.78179931640625, 304.9256896972656, 36.550872802734375, 0.701416015625, 30.160736083984375, 144.3630828857422, 167.99246215820312, 154.91848754882812, 53.58284378051758, 178.18019104003906, 268.08660888671875, 147.98699951171875, 773.062744140625, 154.33494567871094, 521.5599365234375, -78.82649230957031, -9.718255996704102, 200.96890258789062, 171.00860595703125, 72.51800537109375, 153.42218017578125, 438.8819885253906, 172.71524047851562, 31.69029998779297, -30.324844360351562, 115.98097229003906, 99.42860412597656, 152.18426513671875, 35.39637756347656, 157.5582733154297, 429.81182861328125, 60.87687683105469, 143.31185913085938, -31.955062866210938, -298.71807861328125, 286.1195068359375, 42.187950134277344, -634.5320434570312, 2.798858642578125, 457.74920654296875, -253.8532257080078, 295.0589599609375, -11.71795654296875, 13.988876342773438, 26.01226806640625, 203.53253173828125, 81.64108276367188, 50.13554382324219, 105.22653198242188, -89.4229736328125, -170.7137908935547, 42.377685546875, -153.02044677734375, 23.69537353515625, 300.8927917480469, -45.81159973144531, -25.631011962890625, 117.75089263916016, 114.74967956542969, -116.50778198242188, -76.37310791015625, 267.43414306640625, -88.3834228515625, 132.29827880859375, 183.4024658203125, 1047.3648681640625, 64.81727600097656, -38.8798828125, 50.59881591796875, -42.25114440917969, 19.40357208251953, 46.298095703125, 122.95755004882812, -150.17459106445312, -156.32025146484375, 47.344512939453125, 456.2054443359375, -180.11404418945312, 193.99322509765625, 272.08807373046875, 415.70843505859375, 348.80743408203125, -118.53921508789062, 1.2944793701171875, 117.75813293457031, 375.1190185546875, 29.384124755859375, 6.604681015014648, -259.5419006347656, -65.86414337158203, 386.1671142578125, 21.70916748046875, -153.5834197998047, 237.89306640625, 64.31631469726562, 172.50222778320312, 226.173583984375, 214.37380981445312, -66.94699096679688, 131.4161834716797, 144.7635498046875, 68.64346313476562, 18.087890625, -142.00100708007812, 293.413330078125, 174.71298217773438, 137.58297729492188, -2.506732940673828, 235.03182983398438, 260.2137145996094, 404.51007080078125, -341.4379577636719, 138.88851928710938, 50.255767822265625, -37.634002685546875, 300.0624084472656, -14.966888427734375, -14.5859375, 464.7162780761719, 89.30108642578125, 247.27732849121094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000325.npy"}
{"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 104.01132202148438, "std": 179.2697296142578, "min": -333.2185974121094, "p10": -125.4618537902832, "median": 90.65824508666992, "p90": 287.7516235351562, "max": 770.2141723632812, "pos_frac": 0.7265625, "sample": [159.38137817382812, 234.53208923339844, 34.546966552734375, 770.2141723632812, 52.267822265625, 391.3870544433594, -107.20632934570312, 4.3349609375, 224.87619018554688, -150.81982421875, 234.2451171875, 133.14425659179688, 59.63920593261719, 200.19717407226562, 158.99386596679688, -97.94502258300781, 42.39643478393555, 503.69134521484375, 513.02001953125, 10.991317749023438, 165.53976440429688, 293.58447265625, 75.07051086425781, -104.00192260742188, 272.2608642578125, -33.53623962402344, 285.2518310546875, 360.903076171875, 77.79376220703125, 121.98069763183594, -24.656524658203125, 63.758575439453125, 166.92837524414062, 200.6220703125, -110.72125244140625, -99.53506469726562, 71.46249389648438, 117.13741302490234, -180.85595703125, 278.1395568847656, 60.79071044921875, -10.759712219238281, -21.29345703125, 264.0680236816406, 16.272613525390625, 62.64220428466797, -192.23585510253906, 64.8902359008789, -40.03497314453125, 425.4336242675781, 122.74310302734375, 97.901123046875, 535.45849609375, 144.2169647216797, 146.19219970703125, -241.11688232421875, -96.73434448242188, 15.50537109375, 182.13348388671875, -68.10871887207031, 285.14727783203125, 66.20988464355469, 63.1510009765625, 244.9571075439453, 0.0, -167.939453125, 22.79004669189453, 0.9002685546875, 34.27422332763672, 37.32611083984375, -2.0596160888671875, 271.80194091796875, 156.45367431640625, 270.5204772949219, 190.12867736816406, -177.82595825195312, -31.110870361328125, -24.822586059570312, 193.86260986328125, 147.40037536621094, -13.037261962890625, 378.541748046875, 33.230865478515625, 109.76333618164062, 97.4046401977539, 192.03683471679688, -245.04336547851562, -170.27438354492188, 213.01564025878906, 3.2890491485595703, 213.95831298828125, 87.62359619140625, 231.70068359375, 25.648406982421875, 93.6928939819336, 38.37811279296875, 195.15492248535156, 58.79107666015625, -34.16522216796875, 260.2657775878906, 223.9947052001953, 0.0, 133.41213989257812, 157.3269805908203, 179.20028686523438, 160.32386779785156, 85.5775146484375, 274.60931396484375, 121.66352844238281, 0.0, 259.99078369140625, -333.2185974121094, -126.2508544921875, -125.12371063232422, 268.6488037109375, -188.58941650390625, 225.35824584960938, -199.94183349609375, 392.6199951171875, 160.03607177734375, 96.84442901611328, 39.890594482421875, -18.14111328125, -141.28184509277344, 298.3961181640625, 420.43109130859375, 482.16796875, 277.383544921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000326.npy"}
{"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 111.1228256225586, "std": 183.35459899902344, "min": -299.2763671875, "p10": -115.9921432495117, "median": 103.82366943359375, "p90": 300.7556610107422, "max": 1068.275634765625, "pos_frac": 0.78125, "sample": [-109.90568542480469, 164.87213134765625, 198.61782836914062, 55.41270446777344, 248.3389892578125, -299.2763671875, 237.83712768554688, 90.31935119628906, 111.90946960449219, 177.90786743164062, 114.93144989013672, 200.4000244140625, 32.5975341796875, -219.464111328125, 104.5133285522461, -43.63946533203125, -63.30279541015625, 21.6036376953125, 260.47808837890625, 435.08441162109375, 206.23574829101562, 276.32525634765625, 262.88958740234375, 83.94679260253906, 57.475830078125, -203.92465209960938, 75.57492065429688, 181.80548095703125, 19.67083740234375, 73.66837310791016, 57.0333251953125, 273.97735595703125, -68.29170227050781, 150.35525512695312, 93.34979248046875, 338.09271240234375, 357.782470703125, 110.2838134765625, 559.6826171875, 20.458831787109375, 2.780517578125, -130.19387817382812, 17.246856689453125, 193.6221466064453, 131.97491455078125, 66.81619262695312, 277.651611328125, 157.54171752929688, 225.19866943359375, 38.388702392578125, 133.2508544921875, -13.029541015625, -31.478225708007812, -214.88946533203125, 21.702430725097656, 56.637420654296875, 323.6118469238281, 50.47773742675781, 179.0728759765625, 244.03216552734375, -97.71160125732422, 256.2952575683594, 324.8799743652344, 201.46832275390625, 297.7784118652344, 276.17767333984375, 366.2533264160156, 63.47808837890625, 100.8641357421875, -205.13864135742188, 92.06759643554688, -50.575042724609375, 1.388153076171875, -184.77342224121094, -65.18661499023438, 24.942611694335938, 193.8684539794922, 185.10699462890625, 380.49822998046875, 88.034423828125, -26.170989990234375, 113.38275146484375, 106.35137939453125, 102.22409057617188, 117.59779357910156, 248.659912109375, -10.5233154296875, 0.0, 238.44854736328125, 71.863525390625, 16.15899658203125, 1068.275634765625, 330.6837463378906, 103.3480224609375, 112.49058532714844, 202.53616333007812, 266.4333801269531, 211.90887451171875, 307.70257568359375, 130.40968322753906, 157.099853515625, 74.47409057617188, -132.0553741455078, -199.90985107421875, 121.738525390625, 29.941055297851562, 197.54275512695312, 66.82587432861328, 104.29931640625, 84.02032470703125, -221.69732666015625, 200.60076904296875, 268.2225341796875, 36.341766357421875, -147.29486083984375, 379.52801513671875, -202.76611328125, 248.98260498046875, 149.84146118164062, -66.41172790527344, 608.889892578125, 223.2018280029297, -20.100234985351562, -75.55123901367188, -247.14981079101562, 67.5069580078125, 70.01947021484375, 110.03945922851562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000327.npy"}
{"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 105.92172241210938, "std": 170.74020385742188, "min": -327.8228759765625, "p10": -104.50511474609375, "median": 104.40217971801758, "p90": 298.8164367675781, "max": 638.8599853515625, "pos_frac": 0.7578125, "sample": [27.979110717773438, 113.82302856445312, 144.79931640625, 112.91680908203125, 347.98614501953125, 116.7562255859375, -1.42645263671875, 286.60986328125, 217.4758758544922, 102.52384185791016, 35.560150146484375, 7.0925750732421875, 205.83921813964844, 106.280517578125, 24.592987060546875, -5.818233489990234, 450.3931884765625, 44.55963134765625, 57.38995361328125, 111.23686981201172, 183.05494689941406, 215.91287231445312, 23.98187255859375, -8.070632934570312, 129.87020874023438, 225.52508544921875, -104.26710510253906, 154.04385375976562, 222.86309814453125, 112.0238265991211, -72.52029418945312, 238.425048828125, 34.557586669921875, 459.5250549316406, 492.62615966796875, 65.09944915771484, 233.50701904296875, 188.95748901367188, 638.8599853515625, 3.7173004150390625, 106.29022216796875, 295.88153076171875, 159.74639892578125, -105.06047058105469, 82.20466613769531, -81.33621215820312, -46.549072265625, 117.14761352539062, -110.92184448242188, 36.407379150390625, 32.93994140625, 71.3763427734375, -207.67617797851562, 363.88104248046875, 92.13796997070312, 134.1773681640625, -0.2261962890625, 83.58865356445312, 117.40721893310547, -140.83958435058594, 20.37554931640625, 38.163551330566406, 181.10980224609375, 163.9317626953125, 244.16497802734375, 373.14581298828125, 305.66455078125, -327.8228759765625, 116.54203796386719, -108.3890609741211, 13.630172729492188, 107.48876953125, 208.74801635742188, 89.87029266357422, 277.6315612792969, 587.0347900390625, 264.3528747558594, -9.442047119140625, 39.50238037109375, -99.68606567382812, -79.52716064453125, 25.425735473632812, 76.504150390625, 176.7813720703125, -34.70458984375, 58.718109130859375, 162.79473876953125, -12.101547241210938, 167.4503631591797, -297.0403747558594, -65.07363891601562, 276.74005126953125, 128.1814422607422, 64.6357421875, -28.753082275390625, -121.01287841796875, 265.9492492675781, 89.08128356933594, -152.86692810058594, 175.16444396972656, 116.65684509277344, 76.80123901367188, 232.4954376220703, 25.38372802734375, -180.70819091796875, 592.642822265625, 372.7935791015625, -128.86077880859375, 65.09530639648438, -44.42115783691406, 107.97735595703125, 107.37005615234375, 267.7561340332031, -236.80743408203125, 11.79058837890625, 54.65190124511719, 134.4349822998047, 218.46084594726562, -5.511287689208984, -131.26785278320312, 27.822525024414062, 143.26864624023438, 376.7698059082031, -51.37432861328125, 120.515380859375, 457.8399658203125, 279.941650390625, 239.25946044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000328.npy"}
{"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 114.81500244140625, "std": 221.80157470703125, "min": -304.7942199707031, "p10": -123.87532424926758, "median": 78.16324615478516, "p90": 387.58659057617183, "max": 952.699462890625, "pos_frac": 0.734375, "sample": [-225.02972412109375, -120.32594299316406, -190.1861572265625, 350.70538330078125, 65.62474060058594, 159.8598175048828, 65.58058166503906, 108.92831420898438, -29.17298126220703, 97.5263671875, 79.61471557617188, 75.71930694580078, -244.0523681640625, 2.273406982421875, 3.5593719482421875, 233.672607421875, -19.632293701171875, -83.66986083984375, 168.8466033935547, 37.60479736328125, -96.12802124023438, 897.68896484375, 280.07568359375, 169.1051025390625, 271.9737854003906, 359.9819641113281, -160.31947326660156, 135.41015625, 448.4328308105469, 112.75537109375, 129.28614807128906, 77.32208251953125, -133.456298828125, 186.34451293945312, -108.0865478515625, 37.752227783203125, 181.21237182617188, 190.48291015625, 105.77938842773438, 15.415069580078125, 284.70562744140625, 952.699462890625, 200.6273193359375, -111.16851806640625, 144.91226196289062, 59.328857421875, -143.1734161376953, 393.6552734375, -68.50509643554688, -279.16375732421875, 15.498695373535156, 423.9971923828125, -231.58538818359375, 8.366485595703125, 367.44671630859375, 172.0142822265625, -27.80322265625, 147.69342041015625, 110.82926177978516, -123.58877563476562, 148.111572265625, 291.00616455078125, 45.206787109375, -18.216140747070312, 27.869606018066406, -55.293975830078125, 25.335540771484375, 72.19033813476562, 228.4918212890625, 274.02978515625, 89.9368896484375, 0.44720458984375, 199.27590942382812, 782.2547607421875, 113.2224349975586, 163.35543823242188, 271.88824462890625, -89.47000122070312, 8.648370742797852, 294.6071472167969, 208.28402709960938, 102.878173828125, -304.7942199707031, -20.64270782470703, -70.46063232421875, 430.71539306640625, 316.1805725097656, 7.925628662109375, 261.2082824707031, 168.60519409179688, 265.05908203125, -59.141357421875, 85.62770080566406, 756.8818359375, 124.21501159667969, 13.3746337890625, 548.6160888671875, 385.21795654296875, -21.305801391601562, 94.9688720703125, 73.79325866699219, 64.12980651855469, 213.49334716796875, 67.95751953125, -87.54306030273438, -124.54393768310547, 420.34808349609375, 197.41635131835938, 393.1134033203125, 694.102783203125, -164.71876525878906, 72.867431640625, 159.1419219970703, -237.72134399414062, 478.419921875, -97.97428894042969, 126.30401611328125, -28.715606689453125, 53.04206848144531, 9.50885009765625, 30.61164093017578, 12.983779907226562, 40.90243148803711, 72.41761779785156, 288.5384826660156, 79.00440979003906, -180.7562255859375, -31.37609100341797], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000329.npy"}
{"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 107.8102035522461, "std": 185.3886260986328, "min": -344.3221130371094, "p10": -67.95709686279297, "median": 109.12045669555664, "p90": 282.5986083984375, "max": 1050.32275390625, "pos_frac": 0.75, "sample": [-269.2659912109375, 156.25973510742188, 30.779525756835938, -344.3221130371094, 80.33773803710938, 37.7369384765625, -4.7708740234375, 586.6712646484375, 240.26092529296875, 119.75778198242188, -8.6978759765625, 166.44622802734375, 52.104156494140625, -340.20367431640625, 197.0028076171875, 202.74147033691406, 623.938232421875, 208.33126831054688, 203.4957275390625, 90.14706420898438, 136.02989196777344, -109.73234558105469, -340.85760498046875, -7.10186767578125, 9.794784545898438, 195.492431640625, 44.28564453125, 115.75593566894531, -5.441034317016602, -139.81642150878906, 23.067520141601562, 72.21772766113281, 168.167724609375, 155.206787109375, 115.50415802001953, 95.15936279296875, 38.5328369140625, -19.2392578125, -44.12237548828125, 231.88992309570312, 318.9759521484375, 209.2333984375, 245.138671875, 127.135009765625, -67.29328918457031, 217.55322265625, 9.284383773803711, 406.4908142089844, 107.71979522705078, 172.950439453125, -13.097854614257812, 196.48486328125, 283.2049560546875, -173.57546997070312, 185.55764770507812, 34.55061340332031, 45.81024169921875, 87.93325805664062, 185.28463745117188, 87.39830017089844, 193.801513671875, 282.3387451171875, -13.137725830078125, 207.52389526367188, -42.12127685546875, 128.93856811523438, 275.34234619140625, -17.409912109375, 14.838409423828125, -40.34342956542969, 205.30450439453125, -260.32373046875, 305.3096008300781, 119.2236328125, 3.85235595703125, 174.15582275390625, 18.23326873779297, 236.9449462890625, 10.688262939453125, -21.62994384765625, 110.5211181640625, 297.57720947265625, 97.03445434570312, -95.08807373046875, 132.34567260742188, 123.63853454589844, 130.44000244140625, 358.9188232421875, 650.0042724609375, 106.86620330810547, 169.3790283203125, 204.61798095703125, 132.5254669189453, 205.36224365234375, 130.5218505859375, 22.920242309570312, 136.71368408203125, 23.269277572631836, -12.360549926757812, 181.62979125976562, 258.8603515625, 102.3385009765625, -188.32623291015625, -38.98906707763672, -113.21902465820312, 430.9181213378906, 70.2120361328125, 159.6392364501953, -39.87013626098633, 200.41665649414062, 238.67349243164062, 38.497684478759766, 115.84249114990234, -73.0701904296875, -7.138671875, 309.5933837890625, 272.5314025878906, 36.85334014892578, 15.93170166015625, 408.8548583984375, 0.201507568359375, 1050.32275390625, -32.43229675292969, -34.13836669921875, -69.5059814453125, 198.80255126953125, 40.69596862792969, 128.5592041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000330.npy"}
{"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 97.24219512939453, "std": 200.94190979003906, "min": -421.72662353515625, "p10": -122.06964797973632, "median": 85.57939147949219, "p90": 334.72282714843743, "max": 947.6246948242188, "pos_frac": 0.7265625, "sample": [100.11449432373047, 254.8162841796875, 154.9495391845703, -379.05755615234375, 315.71209716796875, 127.54112243652344, 119.82415771484375, 105.18701171875, 120.56210327148438, 114.76181030273438, 4.575897216796875, -40.45356750488281, 421.0183410644531, -31.608978271484375, 947.6246948242188, -219.05503845214844, 281.9825439453125, 45.51470947265625, -287.4572448730469, -176.24298095703125, 94.18746948242188, 15.939453125, -119.80870056152344, -29.236251831054688, -56.24083709716797, 28.65459442138672, 4.4216766357421875, 504.9697570800781, 82.48443603515625, 132.21437072753906, 14.305023193359375, 146.7724609375, -136.8828125, 197.78543090820312, 101.28472900390625, 254.34829711914062, 118.42742919921875, -68.96580505371094, 243.98428344726562, 317.61346435546875, -127.3451919555664, 7.360687255859375, 43.78828811645508, -339.22723388671875, 71.169189453125, 23.707931518554688, 502.6329040527344, -6.556640625, -5.95587158203125, 0.0, 97.7586669921875, 30.286575317382812, 111.84689331054688, -140.27011108398438, 1.85699462890625, 330.1298828125, 40.463409423828125, 210.40835571289062, 88.67434692382812, 116.36065673828125, 121.201904296875, 60.50341796875, -181.21932983398438, 17.64801025390625, 318.560546875, 121.57244873046875, 297.8234558105469, 286.65911865234375, -6.6583251953125, -53.485595703125, 6.36883544921875, -30.645263671875, 480.57513427734375, -89.35641479492188, 168.9626007080078, 1.3018817901611328, -207.35377502441406, 49.555511474609375, 119.25260925292969, -74.55796813964844, 23.53448486328125, 158.22848510742188, 305.58135986328125, 155.93167114257812, -247.77603149414062, 205.63653564453125, 197.63827514648438, 397.9102783203125, -181.7235870361328, 345.439697265625, 304.4270324707031, 100.69268798828125, 175.00045776367188, 569.7198486328125, 9.166610717773438, 351.4626770019531, 345.4919128417969, -26.630027770996094, -27.802978515625, 122.52146911621094, 33.717041015625, -82.28631591796875, 306.8214111328125, 102.5118408203125, -111.40837860107422, 197.65089416503906, 567.733642578125, 19.97093391418457, 75.30232238769531, 103.50808715820312, 76.75404357910156, -48.24931335449219, 150.12103271484375, 18.793548583984375, 149.0260009765625, 207.99069213867188, -12.79022216796875, 622.7752685546875, 27.408702850341797, 156.94898986816406, -421.72662353515625, 10.48931884765625, 384.22235107421875, 223.03863525390625, 24.00152587890625, -20.7860107421875, 121.96269226074219, -11.318588256835938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000331.npy"}
{"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 107.56364440917969, "std": 213.2957305908203, "min": -584.1437377929688, "p10": -131.3519332885742, "median": 97.02605438232422, "p90": 378.59475097656247, "max": 855.936279296875, "pos_frac": 0.6875, "sample": [-19.981821060180664, 53.5599365234375, 27.02398681640625, 718.1740112304688, 187.37042236328125, 93.07424926757812, -44.885528564453125, 382.8254699707031, 208.5240478515625, 390.8371276855469, 87.87258911132812, -67.25973510742188, 70.4971694946289, 116.44844818115234, 153.10130310058594, -85.435791015625, -388.0765380859375, 41.9127197265625, 197.72906494140625, 57.84009552001953, 260.41265869140625, 75.69184875488281, 457.3990173339844, 10.861515045166016, 30.60277557373047, 149.01266479492188, -136.49868774414062, 185.94058227539062, -24.456836700439453, 138.4935302734375, 146.96588134765625, 204.328857421875, -154.2437744140625, 26.106597900390625, 136.64120483398438, 111.55964660644531, 215.695556640625, 376.7815856933594, 204.73147583007812, 256.510986328125, 275.3888244628906, 308.7118835449219, 240.2025146484375, -6.212593078613281, 27.1663818359375, 258.83148193359375, 201.81552124023438, 288.0991516113281, -17.354873657226562, 100.97785949707031, 516.1405029296875, 855.936279296875, -77.49446105957031, -80.07965087890625, 281.34228515625, 397.10528564453125, -170.182373046875, 57.446693420410156, 336.3058776855469, -341.3309326171875, 257.30615234375, 332.40625, 328.38812255859375, -0.095947265625, -129.53721618652344, 129.77224731445312, 184.46896362304688, 12.66033935546875, -6.744848251342773, -150.38778686523438, 45.999053955078125, 519.4342041015625, -13.876953125, 214.35304260253906, 41.952911376953125, 497.5135498046875, 200.689697265625, 176.96337890625, -211.8502197265625, 165.6075439453125, 63.9759521484375, 73.21128845214844, -64.28999328613281, -59.037200927734375, 206.43218994140625, -267.9476318359375, 258.0724792480469, -4.507080078125, -8.43603515625, 134.14662170410156, 167.14495849609375, 23.196401596069336, -24.849990844726562, 166.7733154296875, 261.3350830078125, -29.32012939453125, 170.2229461669922, -135.58627319335938, -221.173583984375, -66.202880859375, 399.7553405761719, 310.48382568359375, -584.1437377929688, -261.4210510253906, 423.24322509765625, 357.2476806640625, 219.34756469726562, -103.82286071777344, 197.2573699951172, 27.47650146484375, 49.185943603515625, 109.65145874023438, 91.30415344238281, 513.8304443359375, 163.75210571289062, 171.11163330078125, -43.78631591796875, 139.04074096679688, -14.036199569702148, 137.11669921875, 80.83560180664062, 445.5032958984375, -32.1693115234375, -129.28424072265625, 0.0, -19.232666015625, -444.5406494140625, 19.7529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000332.npy"}
{"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 89.71623229980469, "std": 182.77024841308594, "min": -402.3160705566406, "p10": -126.42008056640626, "median": 65.6189956665039, "p90": 308.8730560302734, "max": 630.6014404296875, "pos_frac": 0.6875, "sample": [306.0602722167969, -402.3160705566406, 36.55732345581055, 518.8591918945312, 157.06207275390625, -370.8421936035156, 130.3204345703125, 155.46121215820312, 33.3394775390625, 438.4794921875, -70.88185119628906, 19.6082763671875, -151.36181640625, 165.96160888671875, 287.058349609375, -101.44744873046875, 233.74435424804688, 59.661163330078125, 181.5382080078125, 224.008544921875, -30.460861206054688, -76.19849395751953, 169.57342529296875, 160.94354248046875, 152.0689697265625, -53.47313690185547, 208.87408447265625, 315.43621826171875, 357.275146484375, 79.69442749023438, 43.794525146484375, 174.260009765625, 54.92462158203125, 99.4371337890625, -36.29405212402344, 21.772815704345703, 614.629150390625, -140.77621459960938, 263.2239074707031, 4.476898193359375, 120.77923583984375, 97.89959716796875, 169.49276733398438, 8.2900390625, 243.96282958984375, -40.0208740234375, 4.969221115112305, 44.805816650390625, 66.8206787109375, -188.52993774414062, -111.76666259765625, 262.3938293457031, 0.0, 34.38507080078125, -47.280242919921875, 206.24017333984375, 95.27911376953125, 570.3113403320312, 74.41708374023438, 85.142578125, 17.400634765625, 228.58201599121094, -190.3990020751953, 472.4080505371094, 34.6884765625, -85.4923095703125, 152.6919708251953, 282.136474609375, 218.2613525390625, 94.17333984375, -196.11373901367188, 368.47589111328125, 275.0629577636719, -86.92965698242188, -30.250686645507812, 287.0439453125, -50.09552001953125, -15.02935791015625, 93.56771850585938, 201.8922119140625, 111.7589111328125, 16.65673065185547, 30.17803955078125, 230.62713623046875, -191.83074951171875, 486.3797607421875, 100.299560546875, -38.9256591796875, 1.21966552734375, 123.7115478515625, -13.272384643554688, 361.814453125, -65.31114196777344, 123.09892272949219, 55.953643798828125, -41.96234130859375, 60.429718017578125, 62.95806884765625, -166.14157104492188, 207.76280212402344, -18.944488525390625, 224.41627502441406, -132.68206787109375, -96.41712951660156, 630.6014404296875, 16.340768814086914, 41.08221435546875, 301.49603271484375, 114.82827758789062, 435.908203125, 64.41731262207031, -26.094738006591797, 214.76287841796875, -126.1962890625, 85.0926513671875, -33.8529052734375, -185.87513732910156, -18.209022521972656, -43.34124755859375, 72.39727783203125, 164.33135986328125, -126.9422607421875, -163.62181091308594, 117.05281829833984, -4.823486328125, 346.8957214355469, 126.88345336914062, 45.047386169433594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000333.npy"}
{"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 99.57649230957031, "std": 184.149658203125, "min": -375.9969177246094, "p10": -137.5749267578125, "median": 85.26023864746094, "p90": 352.0155090332031, "max": 555.0452880859375, "pos_frac": 0.75, "sample": [352.1501770019531, 173.65921020507812, -170.85369873046875, -327.16534423828125, 0.0, -45.82440185546875, -20.87017822265625, -170.7353973388672, -9.710407257080078, -19.777023315429688, 87.66195678710938, 181.07765197753906, 295.4962158203125, 231.00704956054688, 43.757568359375, 233.59210205078125, -62.94679260253906, 429.8067321777344, 84.76760864257812, 64.86993408203125, 74.02560424804688, 219.249755859375, 330.8575439453125, 155.37203979492188, -94.10015869140625, 351.9577941894531, 92.31893920898438, 103.4375, -278.148681640625, -12.156295776367188, 102.7730712890625, -53.421470642089844, 8.24044418334961, 175.08489990234375, 169.24888610839844, 107.397216796875, 76.72669219970703, 82.72854614257812, -137.37576293945312, 266.8202209472656, -190.13821411132812, 43.54522705078125, 16.9453125, 17.616943359375, 110.88838195800781, 236.006591796875, 16.83111572265625, -156.98370361328125, 178.01901245117188, -45.51080322265625, 59.046356201171875, -66.38674926757812, 453.8802185058594, 23.62896728515625, 404.8816223144531, 206.02835083007812, 0.7686557769775391, 90.49578857421875, 44.719512939453125, 86.83383178710938, 15.48037338256836, 91.83927154541016, 314.14605712890625, 100.6541748046875, -294.8134765625, -48.57670593261719, 439.9188232421875, -204.24102783203125, 103.16390991210938, 128.22186279296875, 76.84100341796875, 14.471954345703125, -138.03964233398438, 300.5522766113281, 288.2257080078125, 129.42152404785156, 26.875274658203125, 166.206787109375, 13.7679443359375, 175.0889892578125, 268.14385986328125, 179.37274169921875, -2.024139404296875, 93.32742309570312, 60.78289794921875, 510.01727294921875, 11.988815307617188, 273.9900207519531, 386.1797180175781, 73.63470458984375, 320.3104248046875, 211.5323486328125, 555.0452880859375, -67.47010803222656, 3.041566848754883, 36.91679382324219, 124.38591766357422, 78.14439392089844, 447.87701416015625, 473.8298645019531, 207.1905517578125, 7.638214111328125, 48.913448333740234, -375.9969177246094, 142.14077758789062, -57.269805908203125, -104.01983642578125, 269.0065612792969, 212.01686096191406, 238.724365234375, -242.86412048339844, 511.1653137207031, 202.56753540039062, -11.312820434570312, 81.19839477539062, -300.54071044921875, 85.30892944335938, 85.2115478515625, -174.08102416992188, 184.5247802734375, 114.29031372070312, 379.7686462402344, 268.5160217285156, 82.54931640625, 267.8875732421875, 4.699127197265625, -105.3397216796875, 363.5498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000334.npy"}
{"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 96.80036926269531, "std": 190.86941528320312, "min": -357.12481689453125, "p10": -113.63577880859374, "median": 70.96828842163086, "p90": 320.2912139892578, "max": 816.551513671875, "pos_frac": 0.671875, "sample": [107.500244140625, -270.5974426269531, 56.48939514160156, -310.9096984863281, 98.34478759765625, 120.02713012695312, 235.17263793945312, 21.4901123046875, 216.55581665039062, 37.3106689453125, 367.1457214355469, -163.79898071289062, 222.77301025390625, 38.35651397705078, 16.079376220703125, 74.35443115234375, 196.97959899902344, -16.32843780517578, -70.951416015625, -312.94635009765625, -83.14422607421875, 392.2810363769531, -3.311859130859375, 61.61057662963867, -8.92291259765625, -109.57858276367188, 106.94352722167969, -150.03359985351562, -238.50946044921875, 564.9537353515625, -142.93157958984375, 231.33724975585938, -77.07110595703125, 244.72137451171875, 158.114990234375, 552.243408203125, 31.434951782226562, 432.0107421875, 208.5616455078125, 39.272865295410156, 816.551513671875, 162.5968780517578, 318.3968505859375, 134.84352111816406, 232.013427734375, 296.098388671875, -30.77465057373047, 218.76792907714844, 230.23272705078125, 271.157470703125, -91.39134216308594, 71.58828735351562, 57.33789825439453, 177.6204833984375, 182.54107666015625, 225.90052795410156, 323.8814697265625, 186.4072265625, 98.07667541503906, 7.57696533203125, 111.72607421875, -197.45303344726562, -55.4993896484375, 173.85406494140625, -76.17169189453125, 286.91546630859375, -55.27997589111328, 318.7525329589844, 205.7470703125, 186.21495056152344, -4.9086456298828125, 40.954071044921875, 227.2822265625, -47.263946533203125, 308.1961364746094, 223.7696533203125, 21.657485961914062, -357.12481689453125, 97.98889923095703, 40.725242614746094, 20.07476806640625, 198.30738830566406, -50.350494384765625, 117.62709045410156, 558.3908081054688, -195.42843627929688, -98.5743408203125, 49.28620910644531, 309.375, 70.3482894897461, -123.10256958007812, 14.347488403320312, -16.385940551757812, 27.931015014648438, -74.24243927001953, -48.20634460449219, 134.30929565429688, 17.124114990234375, 338.7099609375, -131.19102478027344, 414.04541015625, -36.542938232421875, 253.64703369140625, 249.1705322265625, 330.26171875, -53.878204345703125, 0.0, 203.39251708984375, 252.1986083984375, 270.44000244140625, -69.6641845703125, 188.6156768798828, 353.8404541015625, -205.92147827148438, -24.002098083496094, 448.873779296875, -24.011831283569336, -7.830451965332031, -23.09379005432129, 6.0029296875, 111.22257995605469, 171.86106872558594, 23.907379150390625, -46.9049072265625, 5.654045104980469, -38.36846923828125, 94.62200927734375, 212.0246124267578], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000335.npy"}
{"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 102.76631164550781, "std": 215.0046844482422, "min": -432.98663330078125, "p10": -125.15399780273438, "median": 64.20690155029297, "p90": 355.1061462402343, "max": 1121.202880859375, "pos_frac": 0.6796875, "sample": [540.0157470703125, 11.5904541015625, 233.7576904296875, -3.61444091796875, 108.71763610839844, 276.7055358886719, -14.638397216796875, -8.220001220703125, 217.45046997070312, 383.15179443359375, -69.0936279296875, 151.4284210205078, 206.62217712402344, 16.738750457763672, 1.3108024597167969, 130.54620361328125, 158.89768981933594, 174.39080810546875, 193.68753051757812, 529.179443359375, 521.5358276367188, -138.3212890625, -103.38768768310547, 197.4461212158203, -190.21609497070312, 24.542877197265625, 42.378570556640625, 78.5120849609375, 1.9376220703125, 167.46343994140625, 90.90057373046875, 163.5747528076172, 25.21038055419922, 369.59637451171875, 316.4891052246094, 395.4286193847656, 121.39157104492188, 548.5870361328125, 93.1604995727539, 33.868408203125, -432.98663330078125, 226.0807342529297, -122.19329833984375, 123.87554931640625, 133.8837890625, -80.88296508789062, 3.6436100006103516, 470.083251953125, -222.10894775390625, -29.862407684326172, -39.55207061767578, 16.485931396484375, -55.130859375, 304.58306884765625, 322.00433349609375, -185.5484619140625, -36.817138671875, -76.89274597167969, 106.25090789794922, -104.39307403564453, -60.946044921875, 13.46087646484375, -124.770263671875, 288.7241516113281, -174.029296875, 24.491729736328125, 163.13690185546875, -158.30038452148438, 59.00465393066406, 173.83534240722656, 128.00222778320312, 5.489292144775391, 158.24429321289062, 224.64846801757812, -42.59436798095703, 19.09759521484375, -75.50616455078125, 1.01953125, 41.09257507324219, -2.765228271484375, 15.708036422729492, 98.85345458984375, -14.38348388671875, 123.248046875, 318.7432861328125, 139.9827423095703, 155.95849609375, 368.7296142578125, 34.32334899902344, 151.26321411132812, -95.69635009765625, -15.7147216796875, 304.220947265625, 349.26751708984375, 968.6841430664062, 447.39923095703125, 332.1558837890625, 7.57025146484375, 158.885498046875, 243.0833740234375, 122.38754272460938, 455.80511474609375, -34.4285888671875, -129.45281982421875, -126.04937744140625, 1121.202880859375, 60.550506591796875, 198.6153564453125, 266.27392578125, -181.61398315429688, 67.86329650878906, -81.17777252197266, -93.88382720947266, 83.34380340576172, 207.8459930419922, -47.09482192993164, 327.8819580078125, -62.412628173828125, -170.67718505859375, 0.0, 28.686767578125, 54.7432861328125, 128.96856689453125, 135.09799194335938, -183.67205810546875, -127.51181030273438, 94.48165893554688, -34.55121612548828], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000336.npy"}
{"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 113.45880126953125, "std": 179.56288146972656, "min": -334.2056579589844, "p10": -107.86724395751952, "median": 111.98973083496094, "p90": 298.1266235351562, "max": 666.0599365234375, "pos_frac": 0.7734375, "sample": [83.70806884765625, -196.71348571777344, 45.22023010253906, 195.47476196289062, 18.552852630615234, -177.40631103515625, 169.22149658203125, 270.140869140625, 237.13818359375, 599.9324340820312, 37.510589599609375, 142.38858032226562, 39.50390625, 216.8514404296875, 666.0599365234375, 245.60931396484375, 141.40155029296875, 70.70909118652344, -213.14208984375, 78.17436218261719, -334.2056579589844, -82.46900177001953, 84.91143798828125, -136.86000061035156, 189.9100799560547, 109.18533325195312, 99.58251953125, 231.84176635742188, 42.562538146972656, 271.8177795410156, 79.50439453125, 309.8279113769531, 140.5794677734375, 390.44915771484375, 261.91943359375, -36.96028137207031, 203.14321899414062, 62.50965118408203, -16.1671142578125, 77.94155883789062, -88.5042724609375, -58.81834411621094, 102.94010925292969, 149.92108154296875, 118.99581909179688, 444.9600524902344, -140.35894775390625, -117.84225463867188, 66.4049072265625, 126.87701416015625, 28.95758056640625, 172.4075469970703, 280.1388854980469, 269.03814697265625, 171.8280029296875, -82.37361145019531, 146.83935546875, 484.1160888671875, 55.936622619628906, 146.89910888671875, -32.720550537109375, 144.63893127441406, -82.92732238769531, 345.80157470703125, 157.42820739746094, -45.970794677734375, 290.0066833496094, 177.66299438476562, -103.59223937988281, 15.07476806640625, 268.63671875, 272.93206787109375, 103.67108154296875, 117.23681640625, 147.775146484375, 28.686004638671875, -80.6356201171875, -265.1507568359375, 202.7740478515625, 63.027313232421875, -15.655538558959961, 241.712646484375, 53.56886291503906, 88.1142578125, 218.9321746826172, 293.1117858886719, 13.892404556274414, 155.9761962890625, 114.79412841796875, 18.649507522583008, 190.44802856445312, 259.96978759765625, 144.76173400878906, 312.8423156738281, -151.86346435546875, 363.87054443359375, 106.64187622070312, 95.67529296875, 6.592685699462891, 264.74298095703125, 96.62295532226562, -88.06134033203125, 42.42711639404297, 257.601318359375, -137.57382202148438, -161.97291564941406, 571.0340576171875, 120.58628845214844, 36.29424285888672, 53.086570739746094, -67.29718017578125, 0.19979095458984375, 629.1820678710938, 313.72161865234375, -81.10888671875, 234.15512084960938, -230.95806884765625, 125.62207794189453, -268.98486328125, 27.010780334472656, 249.1597900390625, 236.6136016845703, -66.93460083007812, 325.84698486328125, 275.58709716796875, 146.77255249023438, 278.10467529296875, 207.13299560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000337.npy"}
{"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 113.66871643066406, "std": 189.289794921875, "min": -258.1979064941406, "p10": -80.32608337402343, "median": 84.87208938598633, "p90": 343.5090209960937, "max": 698.1466064453125, "pos_frac": 0.7109375, "sample": [307.40618896484375, -174.92108154296875, 11.5338134765625, -29.75006103515625, 606.7702026367188, 693.79638671875, 67.94490051269531, -2.548095703125, -31.42162322998047, -194.23980712890625, 109.9166259765625, 330.828125, 38.26409912109375, 144.97293090820312, 698.1466064453125, 159.1629180908203, -93.77944946289062, 114.43582153320312, -4.4893798828125, -224.4417724609375, 264.1904602050781, 67.68157958984375, 9.380300521850586, 129.4124755859375, 64.27706909179688, 300.8094177246094, 268.7415771484375, 67.70269012451172, -88.47006225585938, -30.22801971435547, 119.3367919921875, 61.9005126953125, 92.79669189453125, 140.4810028076172, -21.437576293945312, 134.09498596191406, 427.21697998046875, 127.52719116210938, 232.6256561279297, 145.49234008789062, -23.202667236328125, 79.71630859375, 189.99972534179688, 171.34164428710938, 147.46044921875, 27.405517578125, -30.4471435546875, 380.3688049316406, 320.2398681640625, 199.0093994140625, 134.5360107421875, -67.48995971679688, 360.2222900390625, -75.8236312866211, 89.92147827148438, 32.44226837158203, -78.34417724609375, 210.08993530273438, -233.42425537109375, 535.3358154296875, 296.8274841308594, 301.5850830078125, 189.5303955078125, 77.49740600585938, 236.0325927734375, 33.93656921386719, -74.46260070800781, 15.763519287109375, 18.459625244140625, -31.78032684326172, 120.38834381103516, 264.9679260253906, 596.8592529296875, 37.29888916015625, -9.3016357421875, -12.13210678100586, 13.329265594482422, 199.0558624267578, 336.34619140625, 33.08873748779297, 33.40155029296875, 1.6352272033691406, 242.6389923095703, 79.82270050048828, 201.83538818359375, -56.7196044921875, 114.23841094970703, 2.07025146484375, -41.52838134765625, 268.7395324707031, 8.793670654296875, 381.8384704589844, 29.713165283203125, -232.97047424316406, -6.5053253173828125, 438.85394287109375, 147.36239624023438, 249.8822021484375, -183.58624267578125, 109.01367950439453, -41.39701843261719, -30.332321166992188, -221.69705200195312, 297.39837646484375, -57.207237243652344, 498.8928527832031, 243.73626708984375, 114.61323547363281, -156.015380859375, 254.1595458984375, 125.58612060546875, 232.40708923339844, 13.625421524047852, 140.90411376953125, -27.3751220703125, 283.08837890625, 387.7042236328125, 202.5756378173828, 57.030914306640625, -31.84649658203125, -258.1979064941406, -95.05380249023438, 28.5125732421875, -70.14768981933594, 265.63800048828125, 149.27914428710938, -84.95053100585938, 486.36663818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000338.npy"}
{"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 127.68128967285156, "std": 203.14109802246094, "min": -302.3391418457031, "p10": -139.24908294677732, "median": 120.26541137695312, "p90": 357.1535614013672, "max": 724.508544921875, "pos_frac": 0.71875, "sample": [-60.26197814941406, -171.3447265625, 177.63229370117188, 226.04013061523438, 56.57733154296875, -59.792327880859375, 99.34182739257812, -147.99099731445312, 120.3812255859375, -82.04190063476562, 494.9187927246094, 109.2174072265625, -26.8597412109375, 94.12406921386719, 355.4230041503906, -19.153564453125, 126.91978454589844, 119.44577026367188, 100.34104919433594, -75.48178100585938, 273.0504455566406, -84.40542602539062, 419.57958984375, 221.39508056640625, 0.0, 724.508544921875, 25.585678100585938, 267.1561279296875, 0.0, 101.057373046875, 168.07733154296875, -52.31329345703125, 184.44192504882812, 194.2049560546875, -167.444091796875, 133.61416625976562, 177.6251220703125, 274.7478332519531, 68.76203918457031, -44.496177673339844, 28.334228515625, 274.7932434082031, -36.91937255859375, 312.1791076660156, 158.7061767578125, 81.85664367675781, 241.41705322265625, -67.71441650390625, 163.80267333984375, 484.3582763671875, 58.224151611328125, -286.30859375, 299.3658752441406, 140.07130432128906, 217.37332153320312, 285.7317810058594, 48.89002227783203, 178.07110595703125, 103.87815856933594, 275.49188232421875, 401.5596618652344, -21.61524200439453, 34.340118408203125, -172.9959716796875, -1.2703666687011719, 266.84393310546875, -135.50254821777344, 211.95767211914062, 633.88720703125, 74.21697998046875, 331.037109375, 12.612640380859375, -249.89462280273438, 347.9722595214844, 25.993743896484375, 211.55517578125, 281.67626953125, 228.47525024414062, -101.24334716796875, 17.883522033691406, -50.21331787109375, 132.33778381347656, 214.90261840820312, -64.69842529296875, 543.1795654296875, 344.62445068359375, 317.056640625, 0.0, 152.78720092773438, 229.35565185546875, 183.4144287109375, 191.56768798828125, 79.981689453125, 237.02096557617188, 333.697265625, -35.130645751953125, 186.79302978515625, 123.50698852539062, 303.0885925292969, 272.620361328125, 411.811279296875, -157.6072998046875, 70.79559326171875, 120.14959716796875, 107.4068603515625, 136.37481689453125, 142.880859375, 48.03302001953125, 0.7345123291015625, -230.5608673095703, 361.1915283203125, 652.5598754882812, -208.2860107421875, 3.7741241455078125, 107.17041015625, -2.545257568359375, 240.38653564453125, -162.1279296875, 483.0157470703125, -302.3391418457031, -271.6802062988281, 656.7263793945312, 235.83425903320312, 105.927734375, 220.44102478027344, -45.903289794921875, -248.46719360351562, 489.94097900390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000339.npy"}
{"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 94.97323608398438, "std": 229.23382568359375, "min": -781.9817504882812, "p10": -127.96461181640625, "median": 79.0782470703125, "p90": 409.0798706054687, "max": 631.7116088867188, "pos_frac": 0.6953125, "sample": [575.5441284179688, 0.406951904296875, -654.9418334960938, 182.80963134765625, 17.037841796875, 623.58056640625, 13.720428466796875, -86.54513549804688, -306.46929931640625, 401.17791748046875, -129.86175537109375, 490.2378234863281, -110.66558837890625, 50.259979248046875, 0.0, 204.2903289794922, 18.478790283203125, -657.7359008789062, 92.5718994140625, 239.5336456298828, 144.3468017578125, 195.45457458496094, -113.32731628417969, 202.10598754882812, -169.5467987060547, 100.46586608886719, -51.68426513671875, 163.112548828125, 124.22468566894531, 26.92132568359375, 447.62890625, -12.923469543457031, 195.46026611328125, 468.4755859375, 50.486572265625, -71.909423828125, -96.96330261230469, -289.66851806640625, 186.66148376464844, 185.08425903320312, 312.56512451171875, 197.45849609375, -133.5869140625, 199.15554809570312, -184.8330535888672, 360.6287841796875, -55.36338806152344, 151.8431854248047, 57.440643310546875, -34.540985107421875, 202.75250244140625, 631.7116088867188, -36.20086669921875, -131.3466796875, 19.91413116455078, 7.56353759765625, 360.2964172363281, 423.5555419921875, 523.2805786132812, -39.31178283691406, 118.10310363769531, -152.149658203125, 407.6124267578125, 0.0, 412.50390625, 121.4759521484375, 337.7655334472656, 62.555442810058594, -82.99713134765625, 76.56201171875, -45.29107666015625, 0.06813240051269531, 91.53395080566406, -190.91201782226562, 454.470947265625, 115.23818969726562, 174.34707641601562, -40.31378173828125, 149.99505615234375, -110.74298095703125, 513.383544921875, 454.17962646484375, -781.9817504882812, -94.4460678100586, 46.282501220703125, 275.08599853515625, 69.60092163085938, -106.88552856445312, 263.5184326171875, 70.81552124023438, 57.45989990234375, -26.872779846191406, -68.77267456054688, 52.05235290527344, 166.68887329101562, 251.73797607421875, 135.79248046875, 16.43402099609375, 151.47882080078125, 189.90975952148438, 313.35601806640625, 32.983741760253906, 347.0921936035156, 162.55007934570312, 403.0614013671875, -116.03128051757812, 218.7423858642578, -116.98591613769531, 263.60894775390625, 26.2159423828125, 226.27496337890625, 103.96896362304688, 19.222240447998047, 111.77645111083984, 462.1458740234375, 212.68310546875, 81.764404296875, 98.8510513305664, 81.594482421875, 56.880096435546875, -51.328277587890625, 71.2607421875, -127.15155029296875, -18.932722091674805, -153.51255798339844, 115.94369506835938, 313.1033935546875, 1.3299732208251953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000340.npy"}
{"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 93.44046020507812, "std": 235.43585205078125, "min": -547.9067993164062, "p10": -202.9694641113281, "median": 94.66284561157227, "p90": 360.2470214843748, "max": 1000.3543090820312, "pos_frac": 0.6875, "sample": [24.53290557861328, 213.979736328125, 123.7154541015625, 129.44500732421875, 81.2578125, 158.4573211669922, 141.56407165527344, 70.3619384765625, 261.8470458984375, -121.66256713867188, 1000.3543090820312, 167.3461456298828, 213.32034301757812, 1.01104736328125, 460.64031982421875, 424.4239501953125, -101.4521484375, 105.5118408203125, -5.314300537109375, 14.686075210571289, 290.11065673828125, 218.56536865234375, 196.354248046875, -547.9067993164062, 777.16796875, 156.8442840576172, 101.41268920898438, 113.13595581054688, -323.34930419921875, 26.30560302734375, 416.970947265625, 418.3575439453125, -242.47467041015625, 150.48866271972656, -199.94329833984375, 212.84283447265625, 147.95751953125, -141.06155395507812, -8.753921508789062, -68.33246612548828, 148.39581298828125, 125.05406188964844, 240.1512451171875, -37.853424072265625, 169.1563262939453, -60.35546875, 215.20440673828125, 109.23670959472656, 10.569091796875, 101.3277816772461, 84.61236572265625, 114.19091796875, 632.5698852539062, -9.99102783203125, 174.46920776367188, 157.10031127929688, 102.33473205566406, 11.700042724609375, 32.84368896484375, -210.030517578125, 4.09173583984375, -52.8802490234375, 522.7755126953125, 22.974252700805664, 154.3209228515625, -321.3614196777344, -292.7908630371094, -11.043792724609375, 197.59097290039062, 18.96435546875, 135.90505981445312, 277.245361328125, 74.80465698242188, 87.99790954589844, 109.2367935180664, 230.40707397460938, 435.4156188964844, 10.138824462890625, -104.12274169921875, -210.58059692382812, 237.61448669433594, 103.71694946289062, 252.44830322265625, -219.37442016601562, 335.936767578125, 16.82366943359375, 65.29168701171875, 173.98236083984375, -20.0457763671875, 316.8317565917969, -14.41851806640625, -78.118408203125, 258.24078369140625, 171.37454223632812, -320.27447509765625, -5.89166259765625, -168.82968139648438, -22.076114654541016, 127.99954223632812, 282.4967346191406, -19.892059326171875, 331.30078125, 64.8011474609375, 263.57763671875, -3.169525146484375, 327.5616455078125, 111.35928344726562, -104.07954406738281, -355.8636779785156, -240.66136169433594, 205.7870635986328, -109.50396728515625, -133.2987060546875, 475.80780029296875, -118.15103149414062, -24.306427001953125, 167.4971466064453, 513.5414428710938, 72.51290893554688, 655.1281127929688, 62.6973876953125, 38.980010986328125, 44.66349792480469, -513.9393920898438, 597.5474243164062, 43.1412353515625, -336.3257141113281, -8.557056427001953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000341.npy"}
{"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 97.88438415527344, "std": 180.8232421875, "min": -400.0798645019531, "p10": -110.97467193603515, "median": 76.70146560668945, "p90": 326.6327453613281, "max": 727.1881713867188, "pos_frac": 0.71875, "sample": [-16.097412109375, -36.75483703613281, 47.15625, 82.78903198242188, 225.93344116210938, 203.95846557617188, -28.084379196166992, -221.3856658935547, -110.81204223632812, 117.09967041015625, 214.93667602539062, 93.11151123046875, -161.09402465820312, 158.48846435546875, 332.5281677246094, 300.1536560058594, -160.9229736328125, 727.1881713867188, -94.47212219238281, -134.3135223388672, 456.35748291015625, 149.44766235351562, 83.87701416015625, 25.34937286376953, -85.89042663574219, 41.71873474121094, 127.570068359375, -177.85430908203125, 183.7445068359375, 76.8670654296875, 100.04993438720703, 329.11688232421875, 43.83983612060547, 167.07095336914062, -14.81280517578125, 31.75164794921875, 158.00469970703125, 188.28961181640625, -124.82005310058594, 144.3310546875, 9.030941009521484, 166.98941040039062, 29.4306640625, -45.8087158203125, -164.718017578125, -108.43280029296875, -220.11474609375, 87.37176513671875, 556.1843872070312, -62.6326904296875, 65.90731048583984, -11.008453369140625, -29.01507568359375, 294.16455078125, 107.12730407714844, 286.29638671875, 68.18896484375, 50.83013916015625, 379.5671691894531, 360.938720703125, -32.1873779296875, 710.765625, 207.11331176757812, 252.31356811523438, 424.88592529296875, -111.35414123535156, 21.13287353515625, 0.0, 126.13180541992188, 55.6829833984375, 122.64596557617188, -34.004764556884766, 226.97891235351562, 16.619813919067383, -63.24895477294922, 95.8603515625, 225.97467041015625, 124.81366729736328, 289.8413391113281, 23.42205238342285, 29.64013671875, 56.16217041015625, 80.19178771972656, 68.505615234375, -90.87213134765625, 73.33060455322266, -251.04010009765625, -44.755218505859375, 284.12127685546875, 23.147966384887695, 173.7453155517578, 57.477874755859375, 106.78140258789062, 39.021240234375, 76.5358657836914, 180.61190795898438, 168.71871948242188, 130.35467529296875, -60.575889587402344, -52.041290283203125, 153.2657470703125, 150.33389282226562, 19.094528198242188, 526.3349609375, 147.37496948242188, 237.36605834960938, -21.58929443359375, 129.17129516601562, 242.6749267578125, -161.92767333984375, 443.36444091796875, 180.277587890625, -164.51150512695312, 359.9559326171875, 57.179168701171875, 300.3038330078125, 72.6044921875, 14.475067138671875, 223.17489624023438, 135.24920654296875, 72.25332641601562, -52.122657775878906, -400.0798645019531, 325.568115234375, 218.4832000732422, 4.363771438598633, 330.55322265625, -10.15228271484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000342.npy"}
{"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 80.28048706054688, "std": 166.0921630859375, "min": -330.468505859375, "p10": -131.19703369140623, "median": 78.46748352050781, "p90": 306.2083984375, "max": 558.090087890625, "pos_frac": 0.625, "sample": [-162.45852661132812, -0.64697265625, 306.08740234375, 5.024866104125977, -7.6513671875, 149.83978271484375, 92.5625, -32.86163330078125, -46.19366455078125, 108.0458984375, 277.26171875, -106.71664428710938, -113.62142944335938, -48.94647216796875, -40.87353515625, 332.01422119140625, 58.79737091064453, 78.87649536132812, -52.491355895996094, -59.373191833496094, 203.13421630859375, 63.670501708984375, 228.02731323242188, 78.0584716796875, 199.96771240234375, 106.88526916503906, 92.69889831542969, 49.922393798828125, -13.04998779296875, 249.7145538330078, 212.218994140625, -262.91436767578125, -16.067626953125, -52.725608825683594, -134.57650756835938, 254.1112060546875, -166.66510009765625, 341.69207763671875, 35.264129638671875, 318.67987060546875, -30.3448486328125, 379.8359375, 77.74407958984375, 135.7039794921875, 0.0, -150.6751708984375, 356.4310302734375, 212.73011779785156, 316.58209228515625, 381.297119140625, 85.53192138671875, 60.10718536376953, -66.33273315429688, -4.381914138793945, 254.79547119140625, -142.1221923828125, 86.51737976074219, 182.13076782226562, 12.063232421875, 124.08448791503906, 159.73312377929688, 162.12774658203125, 80.99819946289062, -35.613983154296875, 64.84260559082031, 293.6842346191406, -152.58010864257812, 558.090087890625, 47.725830078125, -330.468505859375, -129.74868774414062, -142.20416259765625, 271.5496520996094, -122.89523315429688, -139.43310546875, 78.98370361328125, -55.85693359375, 282.185791015625, 136.18020629882812, 85.76486206054688, 167.6557159423828, 208.26869201660156, 73.62370300292969, 98.25503540039062, 158.5564727783203, 191.205810546875, 64.346435546875, 143.72976684570312, -102.92080688476562, -89.40142822265625, 5.365169525146484, 454.987548828125, 529.060302734375, 259.561279296875, 111.9205322265625, 94.65165710449219, 81.29733276367188, 215.29141235351562, 121.552978515625, 167.56504821777344, 202.30441284179688, 268.0769348144531, -30.71990966796875, 306.49072265625, -238.265625, 153.7742919921875, -23.631744384765625, 141.69509887695312, 3.389984130859375, -146.42593383789062, 39.48133850097656, 261.9034423828125, 168.81610107421875, -33.87872314453125, -4.02471923828125, -28.745025634765625, -36.63945007324219, -56.29205322265625, 239.31997680664062, 328.2773132324219, -253.63392639160156, -56.663238525390625, 114.02890014648438, -41.67893981933594, -17.3807373046875, 315.7088928222656, -72.54486083984375, -90.8976821899414], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000343.npy"}
{"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 89.33435821533203, "std": 180.71604919433594, "min": -328.40966796875, "p10": -129.70629348754883, "median": 88.61961364746094, "p90": 305.28263854980463, "max": 551.0800170898438, "pos_frac": 0.6796875, "sample": [85.13894653320312, -34.492340087890625, 230.28387451171875, 116.24606323242188, -207.47955322265625, 289.52093505859375, 455.72650146484375, 128.16799926757812, 253.34413146972656, -8.28570556640625, 147.73902893066406, 7.83935546875, 31.61736297607422, -62.749237060546875, 270.7357177734375, 297.82720947265625, 179.54544067382812, 136.60646057128906, -323.3623962402344, 266.2449645996094, -83.49160766601562, 46.38976287841797, -117.501220703125, 55.784698486328125, 551.0800170898438, 266.5971984863281, -50.01289367675781, 285.4900207519531, 130.95535278320312, -13.92437744140625, 216.3330078125, -61.4815673828125, 75.95285034179688, 12.501571655273438, -257.09185791015625, 77.57841491699219, 233.172607421875, 124.9015884399414, 88.70220947265625, 8.659912109375, -267.5557861328125, 165.62966918945312, 61.186927795410156, -328.40966796875, -74.29779815673828, -21.017288208007812, -97.13092041015625, 4.364585876464844, 312.9034729003906, 226.59373474121094, -230.27435302734375, -193.9525146484375, 246.92001342773438, -225.1531219482422, 106.6629638671875, -99.44024658203125, 180.32505798339844, -108.01821899414062, 173.62130737304688, 226.41644287109375, 245.4938201904297, 145.45285034179688, 396.04974365234375, 302.1914978027344, 488.2672119140625, -84.49668884277344, -134.129150390625, 493.804443359375, -107.05078125, -162.2659912109375, -87.44384765625, 34.20135498046875, 112.75390625, 397.7035827636719, -122.81317138671875, 220.76614379882812, -38.8125, 251.32034301757812, 17.6007080078125, -33.91194152832031, 223.11277770996094, 96.3673095703125, 178.197509765625, 167.7728271484375, -5.191192626953125, 87.62606811523438, -145.43478393554688, 246.63580322265625, 41.3739013671875, 67.12480163574219, 312.49530029296875, 118.6566162109375, 179.7510986328125, -109.53863525390625, 43.758583068847656, 103.67153930664062, 273.3516845703125, 15.880546569824219, 118.5113525390625, -162.81838989257812, -16.328144073486328, 267.6605529785156, -117.8477783203125, 329.6910400390625, -3.425872802734375, 53.4197998046875, 246.86758422851562, 138.867431640625, -61.86000061035156, 27.102264404296875, 88.53701782226562, -16.43603515625, 15.8707275390625, 487.211669921875, 162.8035125732422, 71.8719482421875, 166.91458129882812, -127.81078338623047, 207.65234375, 333.6218566894531, 424.430419921875, 183.6688232421875, 170.08978271484375, 126.86616516113281, 114.48171997070312, -30.774559020996094, 316.1163635253906, -220.63565063476562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000344.npy"}
{"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 88.30291748046875, "std": 161.09083557128906, "min": -272.8204040527344, "p10": -125.9023712158203, "median": 79.5311279296875, "p90": 298.29283447265624, "max": 600.0364379882812, "pos_frac": 0.734375, "sample": [41.6788330078125, 159.6631317138672, 309.7210693359375, -230.0884246826172, 307.5238037109375, 250.5948486328125, 105.0726318359375, -47.973480224609375, 6.9090576171875, 0.4853973388671875, -36.563011169433594, -14.556121826171875, 17.619140625, 44.857177734375, -145.99298095703125, -150.21194458007812, 600.0364379882812, 103.65037536621094, 34.850128173828125, -25.250518798828125, 138.93460083007812, -115.1356201171875, 303.77423095703125, 174.5505828857422, -3.9343395233154297, 164.95651245117188, 265.18682861328125, 434.39312744140625, -195.99215698242188, 7.8548583984375, -14.721466064453125, 15.653778076171875, -10.958419799804688, 62.21955871582031, 25.318328857421875, 159.34381103515625, 83.50396728515625, 370.49713134765625, 46.6688232421875, 213.35147094726562, 152.33102416992188, 8.937028884887695, -47.888214111328125, 82.61526489257812, 110.88558959960938, -23.85333251953125, 202.52609252929688, 31.250905990600586, 336.699462890625, 266.4188232421875, 159.0087432861328, 245.03704833984375, 160.3629150390625, 210.1374053955078, 126.80278015136719, 97.88374328613281, 32.99542236328125, -139.530517578125, 29.007171630859375, 23.44641876220703, 69.25897216796875, 82.93179321289062, -224.37203979492188, 59.097198486328125, 84.19721984863281, 246.3912353515625, -272.8204040527344, 110.57052612304688, -147.50485229492188, -144.4454345703125, 93.0746841430664, 206.91885375976562, 178.48214721679688, 110.11827087402344, 76.44699096679688, -150.37071228027344, 321.329345703125, -109.8696060180664, 123.85275268554688, -48.41199493408203, 362.87164306640625, 526.718017578125, 125.38990783691406, -258.810546875, 218.58059692382812, 139.9811553955078, -83.90902709960938, 353.44287109375, 163.48687744140625, 152.04031372070312, 551.720703125, 204.17501831054688, 193.9107666015625, -120.06173706054688, 3.60888671875, 43.11058807373047, 251.2223663330078, 295.94366455078125, 20.467147827148438, 2.6373748779296875, 174.1480712890625, -62.50933837890625, 128.00787353515625, -155.538818359375, 9.772308349609375, 117.75503540039062, -42.19119644165039, 119.78341674804688, -24.418365478515625, 5.8997802734375, -30.8056640625, 181.33596801757812, -21.2646484375, 204.48289489746094, 381.83123779296875, 217.974365234375, 60.3487548828125, 176.74539184570312, 0.0, 55.62363052368164, 70.37361145019531, 153.34042358398438, -13.209266662597656, 2.031064987182617, 284.2724304199219, 5.57647705078125, 128.12396240234375, -158.68218994140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000345.npy"}
{"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 113.39187622070312, "std": 199.47740173339844, "min": -406.83721923828125, "p10": -97.83772659301758, "median": 93.55172729492188, "p90": 304.5365753173828, "max": 905.9039916992188, "pos_frac": 0.75, "sample": [-284.892578125, 11.045642852783203, 24.90981674194336, -144.14431762695312, 43.64495849609375, 162.3914794921875, 539.9140014648438, 289.76739501953125, 55.62683868408203, -52.20979309082031, 296.9424743652344, -190.41648864746094, 164.66717529296875, -38.40325927734375, 161.01104736328125, -66.89523315429688, -16.026077270507812, 269.869140625, 163.57958984375, 343.4207763671875, 17.26641845703125, 216.4154510498047, 155.72341918945312, 17.77410888671875, 25.20676040649414, 251.69964599609375, 305.75946044921875, 282.3489990234375, 94.21939086914062, -136.54933166503906, 68.30703735351562, 180.84820556640625, 19.788969039916992, 121.55691528320312, -35.953128814697266, 428.5489501953125, 151.37185668945312, 159.69876098632812, 9.331491470336914, 195.9039306640625, 91.2640380859375, 18.196517944335938, -314.021484375, 56.403289794921875, -406.83721923828125, -9.7987060546875, -98.34310913085938, -88.2054443359375, 76.07208251953125, 0.2879791259765625, 150.98370361328125, 86.91793823242188, 205.43870544433594, -53.70697784423828, 119.6365966796875, 97.3209228515625, 649.4691162109375, 115.16459655761719, -66.04824829101562, 273.75396728515625, 201.62344360351562, 160.406982421875, 547.07568359375, 50.72923278808594, 10.284072875976562, 116.53721618652344, 274.307373046875, -98.32598876953125, 290.43426513671875, 249.12307739257812, 224.53768920898438, 146.09759521484375, 53.230838775634766, 377.8839111328125, 761.4306640625, 302.65472412109375, -189.44387817382812, 199.3726806640625, 104.70736694335938, 29.6729736328125, 38.1280517578125, -299.2516174316406, 298.7845458984375, 71.59686279296875, -97.62847137451172, -7.180717468261719, -85.55189514160156, 53.316444396972656, -25.73516845703125, 187.96044921875, 243.13137817382812, 278.43792724609375, -155.18115234375, 202.26425170898438, 68.07894897460938, -165.98643493652344, 176.5548553466797, -388.0535888671875, 77.8254165649414, 327.943359375, -30.652494430541992, 50.763580322265625, 100.52279663085938, 304.0124816894531, 85.89437866210938, 197.59547424316406, 264.39556884765625, 379.1751708984375, 51.43949890136719, 92.88406372070312, 0.0, 121.32453918457031, -34.252716064453125, 62.08860778808594, 456.4468688964844, 905.9039916992188, 172.38059997558594, 149.38394165039062, -5.459381103515625, -57.29945373535156, 261.70001220703125, 294.2240295410156, -40.6785888671875, 240.93331909179688, 366.76727294921875, 73.66659545898438, 39.08438491821289, 233.10284423828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000346.npy"}
{"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 90.40385437011719, "std": 173.2226104736328, "min": -579.53564453125, "p10": -81.05123291015624, "median": 79.71556091308594, "p90": 309.53114624023436, "max": 724.9971923828125, "pos_frac": 0.7265625, "sample": [59.20037078857422, 306.64569091796875, -43.894012451171875, 184.5706329345703, 2.38018798828125, 53.325233459472656, 58.791297912597656, 324.2927551269531, 222.88580322265625, 426.0994873046875, -256.96502685546875, -7.79931640625, 213.47897338867188, 47.166778564453125, 202.81710815429688, -215.18931579589844, -579.53564453125, -64.21713256835938, 288.7420654296875, 544.8374633789062, 259.27587890625, -22.696319580078125, 181.33071899414062, 40.342002868652344, -31.961753845214844, 62.658935546875, 359.7817687988281, 225.37640380859375, -1.6553192138671875, 49.35980224609375, 49.337158203125, 103.273193359375, 277.3373107910156, 98.40365600585938, 312.18011474609375, 10.817230224609375, 287.8292541503906, 31.401893615722656, 81.209228515625, -71.24632263183594, -23.5325927734375, 308.3958740234375, 364.7700500488281, 82.12350463867188, 136.42413330078125, -4.2666015625, -9.388404846191406, -0.5523681640625, 94.43356323242188, -83.020751953125, -30.979949951171875, 6.2687225341796875, 129.0466766357422, 172.86239624023438, -145.4366455078125, 90.74620056152344, 30.620361328125, 210.33157348632812, 427.6617431640625, -93.12187194824219, 89.986328125, 4.190948486328125, -40.55478286743164, 3.557249069213867, 85.23455810546875, 89.01193237304688, 134.30380249023438, 156.286376953125, 256.13775634765625, 109.2393569946289, 162.79144287109375, -32.90582275390625, -304.90948486328125, 46.75261688232422, 319.2172546386719, -28.668506622314453, -234.26708984375, 70.47394561767578, 14.418212890625, -7.29876708984375, -38.414344787597656, 166.65447998046875, 724.9971923828125, 64.63992309570312, 204.51425170898438, -156.82470703125, 179.86984252929688, 280.80889892578125, 154.6387939453125, 3.8970947265625, -249.94964599609375, 92.87359619140625, -80.2071533203125, 215.2540283203125, 94.73359680175781, -180.12890625, 23.1158447265625, 10.089752197265625, 184.52371215820312, 116.0067138671875, -88.75244140625, -200.03823852539062, 353.0274963378906, 243.3114013671875, 0.0, 98.344970703125, 25.876312255859375, 143.59817504882812, 221.38311767578125, 155.1131591796875, 382.2981872558594, 60.585899353027344, 203.61083984375, 328.95416259765625, -17.739013671875, 105.05691528320312, 7.2306365966796875, 42.7950439453125, -42.44233703613281, 78.22189331054688, 160.48196411132812, 288.3304138183594, 320.6700134277344, 147.16098022460938, 4.62567138671875, -76.74188232421875, 142.91183471679688, 46.354705810546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000347.npy"}
{"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 84.6531982421875, "std": 189.79534912109375, "min": -335.352294921875, "p10": -90.53924407958982, "median": 46.67082214355469, "p90": 347.40118713378905, "max": 1015.71435546875, "pos_frac": 0.6328125, "sample": [79.49984741210938, 135.69332885742188, 17.216745376586914, 15.80078125, -18.75628662109375, -48.7926025390625, 334.70477294921875, -30.510608673095703, 237.94776916503906, 251.7491455078125, -268.1080322265625, 358.8988037109375, 212.04913330078125, 60.638702392578125, 106.7197265625, 34.57562255859375, 312.3826599121094, -28.6212158203125, -5.673797607421875, 0.0, 48.1265869140625, 395.7900695800781, 62.788597106933594, -7.199138641357422, -26.9674072265625, 487.3936767578125, -140.59097290039062, -4.2723388671875, -6.688201904296875, 278.83056640625, 228.79263305664062, -39.084232330322266, 15.561798095703125, 5.541267395019531, 177.1363525390625, 78.32080078125, 202.19020080566406, 53.41398620605469, 231.12405395507812, 153.91986083984375, -2.619832992553711, -69.43212890625, -304.64031982421875, 47.50482177734375, -34.551422119140625, 245.9133758544922, 64.0166015625, -116.92349243164062, 92.80641174316406, 163.0464324951172, -9.433212280273438, 152.4639892578125, 71.25213623046875, 55.09990310668945, -39.06512451171875, 101.34445190429688, 130.76467895507812, 372.2298889160156, 346.67578125, -111.53892517089844, -53.806243896484375, -112.42657470703125, -174.33761596679688, 61.51176452636719, 0.6556034088134766, 18.4495849609375, 495.1920471191406, 0.8731613159179688, 175.59652709960938, 34.2442626953125, 406.840087890625, -69.43807983398438, -125.87814331054688, -19.247833251953125, 252.17698669433594, -82.712158203125, 73.8369140625, -147.42527770996094, -77.98046875, 47.73822021484375, 348.323974609375, 51.001220703125, 167.27938842773438, 530.792724609375, -16.13409423828125, 377.7657470703125, -244.50027465820312, -103.22784423828125, -152.42022705078125, -47.19842529296875, 42.96931457519531, 135.31991577148438, 154.10675048828125, -81.98296356201172, 443.1986083984375, -30.28680419921875, 45.836822509765625, -2.5657119750976562, -64.16845703125, 100.99063110351562, 238.35403442382812, 3.013427734375, 1.9315338134765625, 97.30889892578125, -10.099166870117188, 17.131698608398438, 135.70770263671875, 146.83795166015625, 11.08203125, 446.54180908203125, 177.18077087402344, -35.07318115234375, -0.8504791259765625, -40.048858642578125, -85.10127258300781, 173.79969787597656, 615.1109619140625, 62.82208251953125, 80.28376770019531, 34.443206787109375, -73.12451171875, -8.249519348144531, 130.29830932617188, 31.18475341796875, 347.0057067871094, 192.33595275878906, -335.352294921875, 1015.71435546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000348.npy"}
{"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 129.43206787109375, "std": 204.0281524658203, "min": -903.5386352539062, "p10": -51.941631317138665, "median": 102.17575073242188, "p90": 387.7575012207031, "max": 773.8838500976562, "pos_frac": 0.7890625, "sample": [180.46054077148438, 96.20643615722656, 179.36129760742188, -282.5255126953125, 35.7474365234375, 10.615814208984375, 580.6079711914062, 337.6580505371094, 57.60132598876953, 384.63677978515625, 131.522705078125, -202.78909301757812, 102.90538024902344, 206.8900146484375, 172.91802978515625, 102.16632080078125, 506.52215576171875, 37.998870849609375, 395.0391845703125, 322.0732421875, 221.57623291015625, 83.02117919921875, 7.09136962890625, -22.22900390625, -11.4759521484375, 12.094329833984375, -150.96490478515625, 23.555747985839844, 370.296142578125, 284.9349670410156, 52.43171691894531, 199.1641387939453, 126.65841674804688, 32.53515625, 184.7264404296875, 45.18994140625, 259.3273010253906, -109.8052978515625, 57.65192413330078, 184.175048828125, 41.232234954833984, 59.64961242675781, 167.79074096679688, 206.7824249267578, 32.255210876464844, 191.992431640625, 445.77227783203125, 408.5502014160156, 200.33734130859375, 161.27835083007812, -29.767608642578125, -8.936418533325195, 56.518463134765625, 673.8134765625, 285.4622802734375, 13.435531616210938, 7.338958740234375, 128.69882202148438, -91.01162719726562, 197.04876708984375, 0.0, 257.572509765625, -17.2401123046875, 283.515380859375, 220.336669921875, 79.453857421875, 41.350555419921875, 415.88348388671875, 99.00628662109375, 17.839950561523438, -36.148284912109375, -270.3394775390625, -44.046730041503906, 379.35302734375, 67.7174072265625, 317.03533935546875, -113.99679565429688, 223.70504760742188, -49.038368225097656, 102.7536849975586, 15.172515869140625, 293.6228942871094, 196.12188720703125, 232.00494384765625, 224.86181640625, 69.6075439453125, 373.12744140625, 96.65098571777344, 639.267333984375, -38.322967529296875, 149.98532104492188, 82.1302490234375, 344.36065673828125, 80.43997955322266, -106.1099853515625, 193.4873046875, 396.0259094238281, 43.543975830078125, 46.521636962890625, -102.44108581542969, 95.13031005859375, 95.84819030761719, 30.290191650390625, 223.19342041015625, -903.5386352539062, 102.1851806640625, 19.274734497070312, -1.7642593383789062, 12.171890258789062, 129.160400390625, -7.57818603515625, 228.9744873046875, -58.715911865234375, 399.4608154296875, 410.02325439453125, 352.1421203613281, 113.4443359375, -71.31640625, 480.29296875, 773.8838500976562, -43.4932861328125, 146.66583251953125, 175.54290771484375, 116.69290161132812, -147.5284423828125, -27.07208251953125, 135.05917358398438, 230.29010009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000349.npy"}
{"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 103.38925170898438, "std": 194.61402893066406, "min": -301.17779541015625, "p10": -115.20919418334961, "median": 88.98295593261719, "p90": 349.70638427734366, "max": 685.1998901367188, "pos_frac": 0.6875, "sample": [312.96942138671875, 126.70576477050781, 7.354911804199219, -48.00355529785156, 15.723094940185547, 63.56428527832031, -69.65617370605469, 173.70309448242188, -114.59845733642578, 172.0554656982422, 103.1475830078125, 29.873794555664062, 248.24594116210938, -41.297760009765625, 90.04833984375, 578.6277465820312, 180.36050415039062, 367.62158203125, 186.02532958984375, -163.24949645996094, 237.33428955078125, 81.33209228515625, -52.135223388671875, -94.33114624023438, 61.89103698730469, 95.83871459960938, 460.26263427734375, -20.8521728515625, 255.666259765625, 299.1451416015625, 265.44696044921875, -43.62896728515625, 55.683509826660156, -3.812469482421875, -271.65338134765625, -3.9367141723632812, 117.56106567382812, 75.14034271240234, 93.21826171875, 685.1998901367188, 7.151397705078125, 427.66571044921875, 79.1912841796875, -171.1182861328125, 285.53314208984375, 29.82708740234375, -66.84042358398438, 156.1408233642578, -95.43365478515625, -116.63424682617188, 240.61207580566406, 268.87921142578125, 629.3966064453125, 178.72547912597656, 118.19500732421875, 317.2708740234375, 375.0445556640625, 87.91757202148438, 259.0474853515625, 506.1025390625, -75.0992431640625, -91.7823486328125, 171.96243286132812, 70.00485229492188, -87.78981018066406, -53.090309143066406, 254.3023681640625, 214.21511840820312, -203.4649658203125, 19.154949188232422, 179.34291076660156, 258.10400390625, 14.112762451171875, 30.709991455078125, -26.468307495117188, 80.77499389648438, -81.74495697021484, 101.7564697265625, 0.372222900390625, 178.22373962402344, 272.6649169921875, 474.1094970703125, 18.052513122558594, 55.44964599609375, 57.0382080078125, 162.0038299560547, 32.6348876953125, 342.0284423828125, -13.928451538085938, -38.84765625, 188.30810546875, 268.7799072265625, 107.29446411132812, 131.7795867919922, 380.4694519042969, -42.13189697265625, 110.26052856445312, 148.4641571044922, 534.6516723632812, -102.5377197265625, 192.2081756591797, -301.17779541015625, 129.27755737304688, -126.0936279296875, -266.19647216796875, 237.13124084472656, -119.59063720703125, 54.67732238769531, -96.92892456054688, 370.021728515625, -250.57550048828125, 21.58233642578125, 154.21353149414062, 277.61724853515625, -151.1671905517578, 159.8463134765625, -272.0928649902344, 248.16802978515625, 597.2761840820312, -98.06893920898438, 134.24574279785156, -21.725555419921875, -29.58135986328125, -213.52383422851562, 172.8211212158203, -31.723617553710938, 176.9656524658203, 216.81268310546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000350.npy"}
{"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 58.05055618286133, "std": 192.90078735351562, "min": -609.544189453125, "p10": -152.70466003417968, "median": 34.00126647949219, "p90": 327.2001525878906, "max": 684.271484375, "pos_frac": 0.6015625, "sample": [0.0, 271.0633544921875, -0.96990966796875, 323.81195068359375, 36.2451171875, 385.77264404296875, 40.56829833984375, 158.27481079101562, 295.90606689453125, 20.334808349609375, 48.639404296875, -169.99349975585938, 182.20748901367188, 45.4189453125, 409.84600830078125, -150.37403869628906, -27.991823196411133, 265.6651306152344, -176.77749633789062, 66.3753662109375, 31.580963134765625, 117.07252502441406, -129.40704345703125, 442.81768798828125, 484.22271728515625, 25.91199493408203, -27.812726974487305, -16.7587890625, -141.42172241210938, -87.93672943115234, 684.271484375, 337.8362121582031, -64.30419921875, -94.03711700439453, 81.73196411132812, -27.317718505859375, -10.288846969604492, 223.40045166015625, -94.28802490234375, -42.73638153076172, -191.26214599609375, 26.7508544921875, 95.16757202148438, 170.3077392578125, 23.7132568359375, 167.87335205078125, 91.00362396240234, 200.3763885498047, -27.914825439453125, -212.83123779296875, -141.21206665039062, -230.5467529296875, -38.332000732421875, 31.1263427734375, 56.3826904296875, 20.48931884765625, 219.98892211914062, -89.82733154296875, -66.16439819335938, 6.29595947265625, 256.29412841796875, 318.4363098144531, 59.57274627685547, 335.10595703125, 341.33697509765625, -360.46588134765625, 160.9145965576172, 102.94828796386719, 75.14898681640625, -133.89064025878906, -15.565460205078125, 46.27436065673828, -215.24688720703125, -158.1427764892578, 425.5968933105469, -609.544189453125, 120.41064453125, -9.434478759765625, -48.416168212890625, 146.94573974609375, 0.0, 5.10772705078125, 126.69168090820312, 76.37649536132812, 233.1465606689453, -0.22526168823242188, -7.8178558349609375, 425.4861755371094, 160.68817138671875, -50.13568115234375, 106.44918823242188, 22.684707641601562, 75.57928466796875, -38.19289016723633, 0.0, 210.03497314453125, 297.73321533203125, -55.84710693359375, 44.20246887207031, 45.77459716796875, 13.15667724609375, -277.2713623046875, 118.03116607666016, 266.3775634765625, -70.30914306640625, 31.757415771484375, 356.9151611328125, 20.541168212890625, 374.6123352050781, -213.1013946533203, -7.405187606811523, 141.19808959960938, -61.17303466796875, 41.09275817871094, 62.25440216064453, 279.0798034667969, 152.0269775390625, -92.59201049804688, -77.22567749023438, -181.83102416992188, 89.79605102539062, 288.327880859375, -506.084716796875, 95.15521240234375, -136.63873291015625, -125.16776275634766, 344.3776550292969, 160.61070251464844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000351.npy"}
{"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 120.97248840332031, "std": 187.25779724121094, "min": -347.0133361816406, "p10": -95.05133132934567, "median": 126.44475173950195, "p90": 360.2365509033203, "max": 645.911865234375, "pos_frac": 0.75, "sample": [103.2449951171875, 196.21337890625, 262.6524658203125, -74.4791259765625, 161.69522094726562, 443.1434020996094, 152.97467041015625, 163.2997589111328, 167.29266357421875, 32.972412109375, -61.0010986328125, 127.16251373291016, -282.3721923828125, 102.75306701660156, 471.1856689453125, 50.23399353027344, 120.6217041015625, 480.9229736328125, 303.0340576171875, 168.25021362304688, 372.9084777832031, 1.7943248748779297, -267.4930419921875, 580.2085571289062, 645.911865234375, -194.55572509765625, 17.978668212890625, 199.80728149414062, 48.740966796875, 191.66419982910156, 2.59918212890625, -81.82818603515625, 157.5855712890625, 190.73394775390625, 354.80572509765625, 157.0801544189453, -8.59381103515625, -11.1705322265625, -11.59619140625, 136.94154357910156, 245.46646118164062, -147.251220703125, 37.93116760253906, 175.32998657226562, -347.0133361816406, -47.404083251953125, 0.0, 181.5823974609375, 31.313499450683594, 265.390380859375, 49.521575927734375, 495.6910400390625, 24.869659423828125, 142.88214111328125, 350.8094482421875, 163.44189453125, 138.05764770507812, 0.0, 192.69009399414062, 167.2462158203125, 120.44338989257812, -87.52400970458984, 46.665313720703125, 239.5655517578125, -253.6427001953125, -168.37466430664062, 48.84821319580078, 244.46502685546875, 107.00323486328125, 76.22491455078125, -16.629722595214844, 491.56298828125, 224.739990234375, -23.208084106445312, -29.609514236450195, -181.4686279296875, 281.85296630859375, 82.98236846923828, -198.04345703125, 39.424896240234375, 385.7951965332031, 257.0736999511719, 291.04937744140625, 290.868408203125, 18.6927490234375, 158.58432006835938, -23.858551025390625, -309.5152587890625, 0.1952972412109375, 182.3802490234375, -42.08135986328125, -63.30219268798828, 292.67877197265625, 56.58396530151367, 183.58291625976562, 290.1953125, 226.8265380859375, 161.069091796875, 78.0648193359375, 436.6669921875, 156.3161163330078, -22.903060913085938, 402.763671875, 245.56427001953125, 218.07186889648438, -163.747802734375, 347.95068359375, 118.57919311523438, 246.90753173828125, 66.86216735839844, 125.72698974609375, 83.52056121826172, 523.5330200195312, 29.2149658203125, -115.68850708007812, -30.888473510742188, 164.7737579345703, 452.47003173828125, 212.56976318359375, 256.769287109375, -37.35296630859375, -112.61508178710938, 150.25564575195312, 312.3365173339844, 4.8484954833984375, 144.21461486816406, 21.27965545654297, 72.43887329101562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000352.npy"}
{"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 106.32084655761719, "std": 191.4430389404297, "min": -423.74755859375, "p10": -108.90603561401366, "median": 101.28435516357422, "p90": 347.27607116699215, "max": 650.6934814453125, "pos_frac": 0.71875, "sample": [171.09445190429688, 87.2353515625, 195.87347412109375, 135.31756591796875, 70.38401794433594, 206.75961303710938, -116.83528137207031, 65.5455322265625, 119.9949951171875, -5.8359375, -137.16403198242188, 127.1895751953125, -257.8651123046875, 385.4349365234375, -165.91207885742188, 85.0289306640625, -115.16360473632812, -50.4410400390625, 47.8653564453125, -175.55743408203125, 5.8091278076171875, 205.56076049804688, -20.3740234375, -241.8186798095703, 105.09999084472656, -33.33331298828125, 288.1898193359375, 168.2139892578125, 503.068359375, -20.565086364746094, 625.08642578125, 267.7300109863281, 154.26034545898438, -41.680694580078125, 165.316650390625, -423.74755859375, 480.3713684082031, 229.09930419921875, 227.72320556640625, 360.04095458984375, 205.568359375, -92.27392578125, 153.18258666992188, 401.7443542480469, 93.57582092285156, -15.6968994140625, 170.53775024414062, 650.6934814453125, 478.70941162109375, 34.64390563964844, 642.4395141601562, 141.24008178710938, -314.60882568359375, 121.27033996582031, 92.35137939453125, 427.83404541015625, 10.992761611938477, 118.50574493408203, 47.93356704711914, 253.98658752441406, 152.718994140625, -3.156097412109375, 25.609603881835938, -126.35113525390625, 92.46588134765625, 119.58363342285156, -11.881851196289062, -65.99798583984375, 117.7564697265625, 60.10418701171875, 96.99986267089844, 11.57806396484375, 355.3873596191406, 205.02923583984375, 343.7998046875, 53.2755126953125, 34.364768981933594, 258.81695556640625, -41.199981689453125, -104.25579071044922, 142.11508178710938, -80.00634765625, 11.051895141601562, 150.76187133789062, 161.36785888671875, 595.8704833984375, 343.6554260253906, 127.69770812988281, -80.83747863769531, -106.2242202758789, 1.6873741149902344, 97.46871948242188, 182.25982666015625, 110.58955383300781, 156.7384490966797, 108.7135009765625, -14.570676803588867, -82.54247283935547, -87.99952697753906, -30.196533203125, 222.776611328125, -31.749160766601562, 88.64166259765625, 42.88988494873047, 322.70245361328125, 144.15386962890625, 431.18896484375, 113.0311279296875, 323.39593505859375, -227.45799255371094, 60.064369201660156, 52.36711120605469, 167.15774536132812, 125.10552978515625, 9.706268310546875, 0.0, 208.77996826171875, 220.98614501953125, -273.8433532714844, -170.114990234375, 37.820159912109375, -53.956939697265625, 11.3521728515625, 181.87130737304688, 229.65243530273438, 198.96478271484375, 316.50146484375, 275.20330810546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000353.npy"}
{"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 82.8077392578125, "std": 191.2166290283203, "min": -345.5440673828125, "p10": -126.74190063476561, "median": 55.4246826171875, "p90": 339.34287109375, "max": 708.8582763671875, "pos_frac": 0.6640625, "sample": [39.608543395996094, -130.00332641601562, -109.10037231445312, 368.5348205566406, 229.58319091796875, 534.3778686523438, 131.94039916992188, -19.216156005859375, -30.397216796875, 66.86903381347656, 185.17086791992188, 112.8115234375, 82.00240325927734, 70.10963439941406, 418.3081359863281, 50.58965301513672, -97.39555358886719, -71.02456665039062, -120.15861511230469, 24.743438720703125, 14.533979415893555, 49.48863220214844, 337.83148193359375, -164.15164184570312, 268.6646728515625, 64.53280639648438, -68.04991912841797, 57.89268493652344, 198.01025390625, 570.6889038085938, -84.83636474609375, -36.5919189453125, -49.39107131958008, 708.8582763671875, 145.70986938476562, 68.57846069335938, 57.81608581542969, -31.0994873046875, 28.69378662109375, 342.86944580078125, -2.47216796875, -73.70416259765625, 100.90029907226562, 19.215728759765625, 109.66873168945312, 356.1855773925781, -125.34414672851562, 15.957218170166016, -199.70986938476562, 164.11849975585938, 130.11199951171875, 352.46063232421875, 265.119140625, 482.9582214355469, 183.53021240234375, 324.7221374511719, 252.26954650878906, 95.29238891601562, -17.83642578125, 29.228973388671875, 212.99884033203125, -40.365509033203125, -190.22711181640625, -235.0574951171875, 51.47782897949219, 53.03327941894531, 138.03286743164062, 644.7789916992188, 38.5291748046875, 0.22127532958984375, 216.3011474609375, -60.55589294433594, 170.01480102539062, 5.589832305908203, 279.2419738769531, 114.51904296875, 111.39704895019531, -17.84937286376953, -74.73922729492188, 10.825325012207031, 227.38507080078125, -170.22409057617188, 398.8272705078125, 74.47720336914062, -57.414764404296875, 58.127105712890625, 165.96307373046875, 223.02642822265625, 33.97285461425781, -20.352745056152344, 126.47799682617188, -263.2922058105469, 273.78570556640625, -22.38542938232422, 247.55398559570312, 134.3068389892578, -32.550498962402344, 7.294281005859375, -55.03633117675781, -147.6475372314453, 208.29978942871094, 19.10736083984375, -108.6075439453125, 79.99482727050781, 169.54656982421875, 124.21389770507812, -55.6807861328125, -103.30401611328125, 192.74020385742188, 582.8878784179688, 214.5870361328125, -0.42777442932128906, -220.68109130859375, 305.38128662109375, -297.61077880859375, 220.507568359375, -345.5440673828125, 187.51705932617188, -253.79266357421875, 29.52154541015625, 146.39794921875, 63.512943267822266, -163.03887939453125, 23.04291534423828, 7.24981689453125, -112.25669860839844, 375.518798828125, -0.2295379638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000354.npy"}
{"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 111.39378356933594, "std": 231.8589630126953, "min": -933.2530517578125, "p10": -95.08444976806639, "median": 111.91909790039062, "p90": 339.42028198242184, "max": 1227.17041015625, "pos_frac": 0.78125, "sample": [78.28462219238281, -265.0029296875, 284.142578125, 207.7969207763672, -18.231689453125, -2.6260604858398438, 63.61859130859375, -111.1651611328125, 320.010498046875, 115.4649658203125, 482.1917724609375, 16.261695861816406, 53.7098388671875, 180.6756134033203, -15.242158889770508, 157.91531372070312, 202.47251892089844, 250.784912109375, -0.5529022216796875, 16.43438720703125, -317.5527648925781, 14.486968994140625, -43.539329528808594, 699.0368041992188, 35.57110595703125, 215.57391357421875, 15.211978912353516, 41.79083251953125, 62.200927734375, 180.80001831054688, 149.70504760742188, -43.9364013671875, 102.22171020507812, 1227.17041015625, 349.86590576171875, 136.392822265625, 241.15655517578125, -19.19474983215332, 40.77813720703125, 21.051498413085938, 200.72901916503906, 164.6861572265625, 128.8017578125, 93.2498779296875, -313.7405090332031, 72.01458740234375, 115.26449584960938, 14.959945678710938, 327.2916259765625, 61.22509765625, 223.44500732421875, 134.0442352294922, 318.8651123046875, 50.504669189453125, 126.25701904296875, 111.61834716796875, 47.13218688964844, 53.03794860839844, 12.4537353515625, -92.6260986328125, 271.1021728515625, 53.717838287353516, -174.0325927734375, -4.9810028076171875, 114.43970489501953, -252.75086975097656, 60.51825714111328, 427.8353271484375, 231.09381103515625, 224.82778930664062, 2.613882064819336, 219.84658813476562, -297.29638671875, 201.94558715820312, -56.17882537841797, 18.213287353515625, 462.6826477050781, 76.31796264648438, 448.4194641113281, 164.5699005126953, 128.08212280273438, 276.8873291015625, 15.0087890625, -58.0172119140625, 221.5299530029297, -933.2530517578125, 153.19107055664062, -117.01814270019531, 229.90765380859375, 151.2317657470703, 278.2465515136719, -2.051849365234375, -517.043212890625, 107.8428955078125, 133.64703369140625, 74.37057495117188, 84.65205383300781, 145.91635131835938, -349.9416198730469, 126.27001953125, 112.7100830078125, -27.74573516845703, 253.85797119140625, 132.64218139648438, 580.7276611328125, 273.6332092285156, 338.5593566894531, 426.5629577636719, -40.8978271484375, 15.4600830078125, 120.29574584960938, 413.56268310546875, -100.82060241699219, 154.02798461914062, 290.4755859375, 229.22531127929688, 532.2157592773438, 75.12198638916016, 29.239273071289062, 168.76174926757812, -75.604248046875, 112.2198486328125, 29.6121826171875, 486.0643310546875, 78.10186767578125, -273.74163818359375, 341.4291076660156, 265.39422607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000355.npy"}
{"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 80.67941284179688, "std": 158.84400939941406, "min": -338.6517028808594, "p10": -110.04147491455078, "median": 52.307830810546875, "p90": 293.88030700683595, "max": 606.6076049804688, "pos_frac": 0.703125, "sample": [105.29129028320312, -92.72032165527344, 249.08241271972656, -5.82012939453125, 170.619873046875, -159.267578125, -48.20281982421875, -111.46920776367188, 1.277639389038086, 91.8873291015625, 10.91632080078125, 8.384796142578125, 87.17520141601562, 160.3336181640625, -11.90532112121582, 173.9609375, 22.687725067138672, 87.60990905761719, 64.58123779296875, -17.182220458984375, -133.4180908203125, 181.1912384033203, 22.291427612304688, 270.2460632324219, -5.829217910766602, 606.6076049804688, -102.01678466796875, 19.931625366210938, 4.223236083984375, 162.09368896484375, -15.21051025390625, 147.18182373046875, 236.09432983398438, 27.18353271484375, 362.103515625, 506.1983947753906, 131.54742431640625, 195.36737060546875, 132.42042541503906, 347.1309814453125, -110.73828125, 49.293182373046875, -22.145843505859375, -26.077312469482422, 6.6724700927734375, 189.65203857421875, -68.78817749023438, 326.0135803222656, 316.92034912109375, 33.174842834472656, -17.679031372070312, 22.38372802734375, 313.12799072265625, 110.86798095703125, -5.700410842895508, 483.366455078125, -122.59538269042969, 210.23336791992188, 127.48294830322266, 50.588653564453125, 144.7625732421875, -135.91119384765625, 46.449249267578125, 93.1336669921875, 105.72770690917969, 105.92689514160156, 236.1392822265625, -43.247161865234375, -109.74284362792969, 42.211700439453125, 11.023384094238281, 59.278411865234375, 186.55709838867188, 174.04931640625, 5.11083984375, 4.280902862548828, 24.51038360595703, -84.15682983398438, 96.02375793457031, -203.21461486816406, 10.463409423828125, -121.25517272949219, 372.53515625, 277.47015380859375, 51.437164306640625, 264.7735595703125, 295.7091979980469, 69.63053131103516, -58.44453430175781, 6.99853515625, -3.1613616943359375, -117.56060028076172, -90.70564270019531, -338.6517028808594, 293.09649658203125, 98.11532592773438, 65.78765869140625, 0.0, 74.92236328125, 149.61627197265625, 212.1011199951172, 18.551063537597656, -10.93134880065918, 135.5772705078125, -13.936891555786133, 53.178497314453125, 78.28327941894531, 86.98387908935547, 6.0776214599609375, 188.9631805419922, -22.874664306640625, 528.6045532226562, 183.85317993164062, 6.38519287109375, -157.81407165527344, 107.89219665527344, 219.71884155273438, 86.5715560913086, 281.48883056640625, 326.5733337402344, -212.21405029296875, 65.57099914550781, 43.695068359375, -253.96484375, 366.6927185058594, -17.781837463378906, -26.720062255859375, 238.1190643310547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000356.npy"}
{"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 106.86528778076172, "std": 202.50279235839844, "min": -352.7088317871094, "p10": -130.79234161376954, "median": 86.33831787109375, "p90": 348.32843017578125, "max": 730.5111083984375, "pos_frac": 0.6796875, "sample": [138.70391845703125, 482.5823974609375, -31.8492488861084, 188.72805786132812, 119.66371154785156, 155.01495361328125, 0.0, -56.03297424316406, 250.0792999267578, -59.57090759277344, 277.55859375, -198.21058654785156, -130.1795654296875, 396.27935791015625, 624.3402099609375, 217.861328125, -82.10917663574219, 310.2491455078125, 241.9801483154297, -20.962646484375, 302.06787109375, 87.07171630859375, 112.84201049804688, 428.808837890625, 404.04364013671875, -182.93740844726562, 719.3468017578125, 288.92535400390625, 269.5251770019531, 285.3465576171875, 20.51318359375, 0.166351318359375, -134.39779663085938, 34.843868255615234, 185.3409423828125, 143.59666442871094, -0.8224639892578125, -56.21686553955078, 293.1078796386719, 129.05194091796875, 4.709407806396484, 382.646484375, 132.79469299316406, 266.13665771484375, 134.04843139648438, -352.7088317871094, -120.95367431640625, -151.0192108154297, 730.5111083984375, -113.503173828125, -183.3144989013672, -30.485084533691406, 17.75951385498047, -43.587432861328125, 177.24346923828125, 263.8617248535156, 273.49029541015625, 395.0578918457031, 223.03887939453125, -119.37832641601562, 142.67556762695312, -100.60574340820312, 96.10733032226562, 43.2230224609375, 164.54547119140625, -7.749900817871094, 194.83749389648438, 283.3875732421875, 51.78839111328125, 234.27999877929688, -132.22215270996094, -317.8537292480469, 10.949935913085938, 63.1640625, 38.038116455078125, 147.36891174316406, 163.88006591796875, -127.25347900390625, 59.758697509765625, -39.50958251953125, -292.5076904296875, -180.0968017578125, 330.16912841796875, -10.20361328125, 306.1824951171875, 107.64505004882812, 168.4017333984375, -54.1939697265625, 85.60491943359375, -4.25762939453125, -240.78225708007812, 67.30010986328125, 114.65745544433594, 43.14099884033203, 277.95428466796875, 219.6261444091797, -29.416259765625, 52.1707763671875, 0.0, 352.1832580566406, -60.35963821411133, 0.0, 143.32940673828125, 31.634246826171875, 102.66217041015625, 32.45428466796875, 399.3193359375, 318.7786865234375, 179.8355712890625, -14.70819091796875, 342.0785217285156, -269.82403564453125, 676.1588134765625, 263.0372314453125, 22.535354614257812, 25.8778076171875, 24.470489501953125, 120.4456787109375, 235.0373077392578, 257.8028564453125, 346.6763610839844, 6.50273323059082, -51.512046813964844, -141.95382690429688, 357.2519226074219, -61.14997863769531, 5.2008819580078125, 64.08805847167969], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000357.npy"}
{"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 126.11814880371094, "std": 196.3557586669922, "min": -430.66961669921875, "p10": -92.70542907714842, "median": 129.44845581054688, "p90": 358.72520751953124, "max": 719.829833984375, "pos_frac": 0.7109375, "sample": [-62.3316650390625, -53.41654586791992, 80.14825439453125, 227.43988037109375, 25.861083984375, 23.906166076660156, -10.14862060546875, 195.23031616210938, 253.11376953125, 355.6734619140625, -32.153995513916016, 182.53338623046875, 274.2607116699219, 104.60691833496094, 247.47613525390625, 229.6357879638672, 218.57635498046875, 365.845947265625, 346.3927001953125, 128.541748046875, -90.24007415771484, 403.91009521484375, -1.42962646484375, -28.769622802734375, 221.9051513671875, -28.68817901611328, 184.90496826171875, -15.16302490234375, 369.6271667480469, -5.85675048828125, 65.32391357421875, 277.69207763671875, 9.437759399414062, 205.71310424804688, 22.89031982421875, -34.856109619140625, 455.5557861328125, -69.042236328125, 681.0936279296875, 222.7596435546875, -146.40692138671875, 327.638671875, 156.17526245117188, 311.6084899902344, -11.42852783203125, 210.35311889648438, 119.99834442138672, 331.3465576171875, 293.85662841796875, 176.8546142578125, -119.76608276367188, -57.92015075683594, -176.57955932617188, 158.28164672851562, 388.25146484375, -8.374176025390625, 271.40533447265625, 284.75439453125, -120.93797302246094, 379.7838134765625, 103.505615234375, -395.2008361816406, 266.8505859375, -73.36685943603516, 13.837081909179688, 276.15191650390625, 35.1571044921875, 60.10907745361328, 10.590057373046875, -17.980255126953125, 247.99075317382812, 101.58116149902344, 57.28227996826172, 51.34196472167969, 2.960693359375, -430.66961669921875, 159.1289520263672, 191.41937255859375, -205.52322387695312, 317.08172607421875, 148.16481018066406, -127.91104125976562, -12.56500244140625, 140.197998046875, 547.715576171875, 37.9085693359375, 280.86651611328125, 65.78387451171875, 160.94570922851562, 280.53369140625, 98.83915710449219, 339.3927001953125, 719.829833984375, 316.29949951171875, 156.02410888671875, 549.1786499023438, 16.467864990234375, 86.95941162109375, 25.4573974609375, 153.80014038085938, 300.601318359375, 56.256103515625, 170.17160034179688, 231.0105743408203, 371.2874755859375, 289.7496032714844, 345.53802490234375, -290.79327392578125, -98.45792388916016, -319.60040283203125, -32.72736358642578, 14.893768310546875, 257.9402160644531, 225.9193878173828, 256.93963623046875, -204.24423217773438, -100.21463012695312, -22.317350387573242, -56.51506042480469, 151.80535888671875, 0.0, -3.855499267578125, 456.07958984375, 132.7415771484375, 370.5898742675781, -30.578750610351562, 130.35516357421875, 67.55715942382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000358.npy"}
{"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 103.51795959472656, "std": 198.298095703125, "min": -473.202392578125, "p10": -104.9788688659668, "median": 97.04011917114258, "p90": 323.0026092529297, "max": 684.1563110351562, "pos_frac": 0.734375, "sample": [-82.37950897216797, -473.202392578125, 234.69793701171875, 530.0635986328125, 500.9598083496094, -138.74400329589844, 141.0535430908203, 50.05729675292969, 79.5767822265625, 142.19644165039062, 207.56515502929688, 260.2437438964844, 119.83024597167969, -54.610355377197266, -268.9072570800781, -27.228515625, 176.87332153320312, 377.933349609375, 19.2791748046875, 108.00228881835938, 165.07586669921875, -121.17611694335938, 288.03143310546875, 675.8689575195312, -250.04257202148438, 127.44991302490234, -16.104736328125, 243.98822021484375, 368.36083984375, -381.5705261230469, 169.06768798828125, 286.09332275390625, -10.979400634765625, 128.53878784179688, -10.171873092651367, 157.4571533203125, -33.88737487792969, 243.1717529296875, 181.24276733398438, 221.434326171875, 80.5606689453125, -201.06785583496094, 81.29718017578125, -21.186798095703125, 220.63510131835938, 12.899658203125, -35.65925598144531, 128.6507568359375, 201.91943359375, -46.807350158691406, 142.90994262695312, -13.08416748046875, 335.7228088378906, 9.152690887451172, 54.77850341796875, -51.32914733886719, 580.8233642578125, -429.3985595703125, 180.29466247558594, 92.91558837890625, 320.47955322265625, 43.4739990234375, -234.8621826171875, 254.2603759765625, 93.85052490234375, -104.8189468383789, 14.04632568359375, 58.10064697265625, 407.99017333984375, 232.8526611328125, 11.859619140625, 12.21710205078125, -105.35202026367188, 24.84912109375, -31.659500122070312, 157.8975372314453, 119.87901306152344, 136.35592651367188, 184.39492797851562, 35.36376953125, -64.56610107421875, 24.501907348632812, 15.162811279296875, 207.3629150390625, 318.357421875, 17.659378051757812, 189.03253173828125, 89.8577880859375, 162.29443359375, 163.82122802734375, 428.13665771484375, 185.3546905517578, 463.1885986328125, 100.2297134399414, 73.27294921875, 56.255279541015625, -67.64939880371094, 246.6458740234375, -10.346670150756836, -33.42924499511719, 54.05029296875, 136.77490234375, 153.73361206054688, -194.58514404296875, 120.14373779296875, -333.27294921875, 179.66738891601562, -34.02142333984375, 121.6826171875, 671.600341796875, 61.67002868652344, 32.15702819824219, -19.093902587890625, 44.05060577392578, 31.97430419921875, 257.5429992675781, 684.1563110351562, 236.5794677734375, 76.3238525390625, -157.88206481933594, 47.4400634765625, 328.8897399902344, 133.66815185546875, 145.2528839111328, 234.85150146484375, 129.05079650878906, 270.23883056640625, -17.803070068359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000359.npy"}
{"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 100.06249237060547, "std": 244.51092529296875, "min": -1048.8095703125, "p10": -87.69568481445312, "median": 74.72069549560547, "p90": 321.5905334472656, "max": 1295.55859375, "pos_frac": 0.7265625, "sample": [327.26873779296875, 50.600547790527344, -167.47573852539062, 289.15643310546875, -154.61099243164062, -11.903648376464844, -164.436767578125, 97.41094970703125, 1295.55859375, -167.58984375, 0.53289794921875, 11.803695678710938, 68.86697387695312, 335.28515625, -27.1302490234375, 282.6279296875, 95.1231460571289, 202.97979736328125, -86.384521484375, 501.48583984375, 320.8453674316406, 30.824920654296875, 59.4251708984375, 254.76918029785156, 8.103515625, 33.7071533203125, -39.80735778808594, 79.52532958984375, -54.186279296875, 228.41213989257812, 72.77880859375, 344.75372314453125, -83.67782592773438, 178.82855224609375, -0.1829833984375, 6.7554931640625, -198.9312744140625, 82.518310546875, -104.11383056640625, 95.49555969238281, 260.7833251953125, 110.28207397460938, 327.60015869140625, -12.563079833984375, 197.76910400390625, 136.67433166503906, -35.215728759765625, 315.7464599609375, -31.0447998046875, 85.9947509765625, -44.1318359375, 158.40203857421875, 251.29470825195312, -26.311431884765625, 807.4791259765625, 43.096954345703125, 323.3292541503906, -90.75506591796875, 120.93341064453125, 65.39717102050781, 256.85821533203125, 264.48907470703125, 212.43197631835938, 186.71636962890625, 66.02572631835938, 39.635738372802734, 255.77056884765625, 93.71832275390625, 189.70921325683594, -71.45263671875, -181.9711151123047, 198.38641357421875, -5.779058456420898, 165.08953857421875, -27.708740234375, 11.465858459472656, 198.2117156982422, 436.63629150390625, 195.43490600585938, 316.00152587890625, 76.66258239746094, 494.3834228515625, 48.10211181640625, 7.582973480224609, -1048.8095703125, 263.46258544921875, 343.94677734375, -83.92486572265625, -75.5054931640625, 11.211013793945312, 189.5255126953125, 152.0413818359375, 400.11016845703125, 39.79206085205078, 6.198198318481445, 89.6947021484375, 36.289581298828125, -580.3829345703125, 56.754241943359375, 21.318031311035156, 53.394989013671875, 0.0, 218.71286010742188, 129.34506225585938, 297.5963439941406, 78.29953002929688, 101.3011474609375, 1037.628662109375, -28.261764526367188, 92.83074951171875, 223.1187744140625, -31.102081298828125, -6.250244140625, 144.48892211914062, 40.01740264892578, -122.47634887695312, -214.4694366455078, -516.63037109375, 0.0, 33.05201721191406, 56.141143798828125, 116.2164306640625, 245.99093627929688, 63.369903564453125, 43.854248046875, 111.96967315673828, 156.14227294921875, 175.82200622558594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000360.npy"}
{"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 132.3245086669922, "std": 220.88223266601562, "min": -434.26007080078125, "p10": -101.34534912109375, "median": 105.8908462524414, "p90": 445.1708709716797, "max": 949.534423828125, "pos_frac": 0.734375, "sample": [137.4143524169922, 2.64483642578125, -1.4102783203125, 136.70428466796875, 114.08029174804688, -4.70745849609375, 9.738761901855469, -116.18759155273438, 271.04052734375, 86.11343383789062, -32.85601043701172, 39.38311767578125, 5.912040710449219, 258.2532958984375, -4.220451354980469, 534.191650390625, 81.25662231445312, 197.48435974121094, -135.9896240234375, 55.29804992675781, 482.63592529296875, 407.7581787109375, -31.97344970703125, 190.4675750732422, -42.30116271972656, 311.2522888183594, 200.6559600830078, 162.2836456298828, 60.085731506347656, 447.5830078125, 444.905029296875, 227.91555786132812, 96.55824279785156, 225.81460571289062, -34.981353759765625, 228.44076538085938, -331.7091064453125, 221.41680908203125, 202.3134002685547, 339.9363098144531, 258.47808837890625, 698.7198486328125, -279.3619384765625, -35.47797393798828, 208.90460205078125, -69.88720703125, -108.1209716796875, 216.44113159179688, 138.4745330810547, 29.322776794433594, -20.83885383605957, 106.9896240234375, 93.0550537109375, 0.0, 44.22825622558594, 140.07281494140625, 192.6007537841797, 599.927734375, 34.31305694580078, 118.04635620117188, 334.806884765625, 422.61309814453125, 19.165590286254883, 471.27105712890625, 266.48187255859375, -103.195068359375, 89.76348876953125, 60.4161376953125, 213.35870361328125, -157.50244140625, 189.8001708984375, -0.702880859375, -67.05950927734375, 80.99664306640625, -38.24203109741211, 0.0, 127.09051513671875, 53.24102783203125, -10.493362426757812, 67.22650146484375, 448.2243957519531, 162.93399047851562, 210.28790283203125, 132.6134033203125, -421.9919128417969, 157.99142456054688, 46.49591064453125, -157.10992431640625, 546.4376220703125, 191.43438720703125, -177.55165100097656, 104.79206848144531, -100.5526123046875, 175.1937255859375, 527.6912841796875, 237.1355743408203, 60.013282775878906, 97.789306640625, 145.27955627441406, 206.38433837890625, 393.61724853515625, 1.4898548126220703, 526.1632690429688, 122.72683715820312, 156.63259887695312, 949.534423828125, 231.14883422851562, -19.462615966796875, 270.90472412109375, 860.9442749023438, 254.70648193359375, 32.088233947753906, -273.8697814941406, 390.5576171875, 445.7911682128906, 179.41432189941406, 15.91937255859375, 324.1988525390625, 72.35851287841797, 32.78367614746094, 142.05816650390625, -11.644851684570312, 4.994354248046875, -13.839736938476562, -178.8245849609375, -3.8388137817382812, 43.625877380371094, -434.26007080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000361.npy"}
{"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 100.11392211914062, "std": 171.33460998535156, "min": -324.698486328125, "p10": -78.45950927734374, "median": 73.58692169189453, "p90": 298.7163604736328, "max": 934.857666015625, "pos_frac": 0.75, "sample": [-12.866928100585938, 161.746337890625, -41.50161361694336, 186.76681518554688, 159.1583251953125, 85.45782470703125, 298.3778991699219, 206.7351837158203, 69.50128173828125, 374.8756408691406, 118.15960693359375, -324.698486328125, -44.73529052734375, 19.575939178466797, 284.89739990234375, -13.606311798095703, 188.73394775390625, 68.0067138671875, 197.4005584716797, 232.16342163085938, -112.10183715820312, 323.31591796875, 418.44573974609375, 175.07830810546875, 178.63177490234375, 277.32354736328125, 103.65422058105469, 10.397262573242188, 50.17686462402344, 129.0500030517578, -44.308197021484375, -75.76484680175781, 186.26869201660156, -105.71041107177734, 51.29730987548828, -19.354217529296875, -6.6806182861328125, -22.53750228881836, 220.5621337890625, -138.30873107910156, 934.857666015625, 19.08319091796875, 351.0030212402344, 214.28375244140625, 287.12274169921875, 334.5849914550781, 190.6676788330078, 48.320709228515625, -33.2122802734375, 141.86917114257812, -59.60760498046875, 262.0274353027344, 84.059814453125, 13.1722412109375, 4.02044677734375, 23.0977783203125, -5.31365966796875, 311.8530578613281, 8.33587646484375, 147.8983612060547, 60.525146484375, 11.0882568359375, 88.583984375, 263.55877685546875, 108.00010681152344, 26.15673828125, 53.27643585205078, 100.49728393554688, -41.570892333984375, 275.83843994140625, 754.7353515625, 120.07627868652344, 10.730453491210938, 28.848663330078125, 101.1543197631836, 219.21307373046875, -183.71963500976562, -42.34473419189453, 233.995361328125, 311.4676513671875, -203.34498596191406, 170.78909301757812, -2.5412063598632812, 133.115966796875, -33.62078857421875, -65.873046875, -155.40997314453125, 361.9781494140625, 21.7139892578125, 12.859344482421875, 73.14291381835938, 38.531166076660156, 9.268989562988281, 158.00071716308594, 12.6270751953125, 98.24021911621094, 190.22657775878906, 17.147323608398438, -45.17817687988281, 138.26107788085938, -70.09521484375, 201.29141235351562, 336.2367858886719, 68.05059814453125, 44.28826141357422, -122.81854248046875, -84.74705505371094, 223.59603881835938, 65.77716064453125, 123.93600463867188, 223.28846740722656, 28.537473678588867, 180.577392578125, 195.3884735107422, 64.67819213867188, -229.85427856445312, 40.772674560546875, 408.82830810546875, 132.27499389648438, 74.03092956542969, -157.3228759765625, 255.67462158203125, 99.56076049804688, 115.53286743164062, -199.95062255859375, -94.65179443359375, 299.506103515625, 70.44146728515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000362.npy"}
{"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 114.80848693847656, "std": 211.86314392089844, "min": -319.20849609375, "p10": -108.83223495483396, "median": 110.7639389038086, "p90": 330.9008575439453, "max": 1070.6429443359375, "pos_frac": 0.703125, "sample": [230.1959228515625, -31.272377014160156, -26.79400634765625, 399.2186279296875, 190.25457763671875, 103.487060546875, -6.923929214477539, -100.488525390625, -24.08050537109375, 575.0247192382812, -317.9307556152344, 61.44903564453125, 215.2657928466797, 158.3201446533203, -42.895973205566406, -278.4314270019531, -5.375396728515625, 534.1336059570312, 18.91754150390625, -76.5281982421875, 36.41839599609375, 288.9506530761719, 221.480712890625, -47.065185546875, 115.04342651367188, 1070.6429443359375, 427.30450439453125, 188.9058837890625, 70.04731750488281, -53.52367401123047, 139.10610961914062, 12.09906005859375, 6.8914031982421875, 49.576568603515625, 183.24546813964844, 8.31348991394043, -69.6179428100586, 178.74627685546875, 329.2664489746094, -45.9097900390625, 170.7459259033203, 88.75423431396484, 121.61627197265625, 209.6562042236328, 254.64105224609375, 203.17120361328125, -144.0736846923828, -2.45648193359375, 111.48558044433594, 113.15884399414062, 210.8377685546875, 45.586280822753906, 776.080810546875, 215.85333251953125, -172.9278564453125, 120.9586181640625, 20.467681884765625, 251.09622192382812, 87.51617431640625, 707.42431640625, 79.56573486328125, 217.27447509765625, 95.33393859863281, 254.33499145507812, 266.80987548828125, 99.32427978515625, 99.54165649414062, 177.2115020751953, 160.81045532226562, -14.832071304321289, 302.8559265136719, -74.80801391601562, 237.25103759765625, 334.7144775390625, -36.12541198730469, 111.9324951171875, -89.4085693359375, -68.20793151855469, 4.8811187744140625, 138.9169921875, -18.254806518554688, 287.83428955078125, -163.07052612304688, -66.15109252929688, 93.31885528564453, -164.60606384277344, -181.07861328125, -43.0684814453125, 7.60479736328125, 205.9638671875, 537.5094604492188, -217.45594787597656, 324.084716796875, -122.06986236572266, -319.20849609375, 224.42886352539062, 195.88478088378906, -97.12286376953125, 303.2213134765625, 376.61614990234375, 35.29931640625, 213.34622192382812, -305.3331604003906, 259.5810546875, 510.73199462890625, 120.01296997070312, 124.65283966064453, 410.5834655761719, 243.21282958984375, 390.6513977050781, -103.15896606445312, -209.68075561523438, -97.2630615234375, 41.04602813720703, 68.8681640625, 202.27777099609375, 23.23919677734375, 146.7653045654297, 256.1201477050781, 222.8382568359375, 157.11151123046875, 235.06243896484375, 127.77426147460938, 73.55715942382812, -101.39010620117188, 148.7266845703125, 110.04229736328125, -146.00631713867188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000363.npy"}
{"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 111.43631744384766, "std": 235.66957092285156, "min": -419.0240173339844, "p10": -141.47686614990235, "median": 61.30961608886719, "p90": 425.6511077880859, "max": 860.0673828125, "pos_frac": 0.7109375, "sample": [526.2347412109375, 167.2891082763672, 174.7227783203125, 99.46046447753906, 269.9800720214844, 30.767593383789062, 398.00042724609375, 21.07354736328125, 38.901641845703125, -105.78176879882812, 104.94522857666016, 19.313072204589844, -33.762542724609375, -53.51885986328125, -64.8753662109375, 15.15484619140625, 768.1243896484375, -138.18133544921875, 153.8213348388672, 323.10626220703125, -292.4349365234375, 48.29951477050781, 138.7010955810547, -205.95138549804688, 243.45037841796875, -419.0240173339844, 241.66680908203125, 160.92507934570312, 17.216156005859375, 89.51644897460938, 709.7192993164062, 17.826339721679688, -206.0196533203125, 9.543254852294922, 57.36907958984375, 236.5610809326172, 57.24348449707031, 120.89199829101562, -3.6343536376953125, 326.86175537109375, 59.38153076171875, 10.7841796875, -261.7594299316406, 445.7809143066406, 99.19115447998047, 231.60263061523438, -143.30091857910156, -6.56689453125, 59.468658447265625, 166.32574462890625, 194.36949157714844, 171.57540893554688, 11.623764038085938, 506.1148681640625, 71.75366973876953, -83.33026123046875, 366.4969482421875, -19.643274307250977, -294.5177001953125, 742.0109252929688, 167.16622924804688, 102.8037109375, 17.0753173828125, -22.892318725585938, 465.9691162109375, 176.97308349609375, 16.267578125, 555.648681640625, 27.533599853515625, 41.5380859375, 81.077392578125, 342.9450988769531, 211.70001220703125, -138.347412109375, 146.86849975585938, -42.48583984375, 322.7796630859375, 98.99652099609375, -145.4769287109375, 417.0240478515625, 192.39599609375, 62.95556640625, 31.111778259277344, -105.65826416015625, -92.54278564453125, 160.9901123046875, 90.22056579589844, 456.22906494140625, 31.715652465820312, 756.7470703125, 291.72021484375, 7.218170166015625, -140.69512939453125, 17.557098388671875, 159.56735229492188, 510.0437316894531, -139.32835388183594, -132.5660858154297, 142.2984619140625, 390.31884765625, -16.012054443359375, 125.43072509765625, -258.44781494140625, 265.686279296875, 243.87136840820312, -286.58587646484375, 396.04278564453125, 59.663665771484375, -10.798095703125, -108.15350341796875, 105.22710418701172, -57.766510009765625, 8.451988220214844, 379.8575744628906, 35.55890655517578, 161.37721252441406, -238.5670166015625, 54.31077575683594, 860.0673828125, -46.84980392456055, 99.77337646484375, -32.26020050048828, 97.49333190917969, -42.49072265625, -167.1588134765625, 248.77569580078125, 665.9241943359375, -202.90048217773438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000364.npy"}
{"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 99.90142059326172, "std": 209.37985229492188, "min": -415.517822265625, "p10": -144.81806640624998, "median": 72.28655242919922, "p90": 357.96404113769523, "max": 704.7825317382812, "pos_frac": 0.6875, "sample": [0.0, 57.64126205444336, 140.24630737304688, 99.7863540649414, 489.16717529296875, 289.3133239746094, 101.8046875, 81.1226806640625, 0.0, 107.67364501953125, 704.296142578125, 141.23605346679688, 345.44061279296875, 414.6546630859375, 74.4649429321289, -3.6533050537109375, 20.23272705078125, 79.1440200805664, -75.84881591796875, 297.4288330078125, 36.46473693847656, 268.522705078125, -68.25357055664062, -18.75519561767578, 157.64645385742188, 87.65872192382812, -131.68873596191406, 262.06787109375, 221.16162109375, 704.7825317382812, 97.30242156982422, -65.17266082763672, -89.43524169921875, 121.08428192138672, -198.617431640625, -76.48431396484375, 112.98687744140625, 138.3131103515625, 147.85040283203125, 157.73068237304688, -27.375497817993164, 176.17337036132812, 223.76611328125, -138.88592529296875, -166.42742919921875, 337.4935302734375, -101.0443115234375, 205.5279998779297, -19.917156219482422, 157.54437255859375, 174.26063537597656, -222.2962646484375, 64.29180908203125, -206.404052734375, 42.39447021484375, 149.289306640625, 228.72027587890625, 572.1408081054688, 371.2633361816406, 24.826690673828125, 56.01959228515625, 429.91424560546875, -158.65972900390625, 36.94207763671875, 39.03014373779297, -69.64849853515625, 88.68992614746094, 143.36422729492188, 137.20645141601562, 352.26434326171875, 270.3938293457031, -23.352935791015625, 104.52532958984375, -76.00762939453125, -29.983184814453125, 198.23953247070312, 254.344482421875, -245.31497192382812, 646.2386474609375, -44.476318359375, 456.83367919921875, 8.780441284179688, 339.89288330078125, 67.5866470336914, 222.29293823242188, 67.64129638671875, 472.388916015625, 494.36962890625, 41.118919372558594, -124.88013458251953, 1.7386531829833984, 23.49150848388672, -2.617767333984375, 321.5019836425781, 8.108955383300781, 65.18435668945312, -46.54737091064453, -108.65634155273438, 72.82374572753906, -16.39495849609375, 314.931396484375, -318.5331115722656, 14.67486572265625, 57.99420166015625, 330.0088195800781, 0.0, 53.335975646972656, -60.84931182861328, 71.74935913085938, 254.845947265625, -158.6927490234375, 121.14476013183594, -13.842094421386719, -290.02642822265625, -218.21514892578125, -247.23037719726562, 474.9293212890625, 315.3443603515625, -351.41070556640625, 21.894287109375, 37.62620544433594, -415.517822265625, 449.19110107421875, 138.4598388671875, 12.599609375, 349.8739318847656, 189.18077087402344, 104.86862182617188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000365.npy"}
{"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 105.48709106445312, "std": 193.33697509765625, "min": -573.0150146484375, "p10": -103.88524551391602, "median": 95.2503890991211, "p90": 316.4000610351562, "max": 870.4755249023438, "pos_frac": 0.7265625, "sample": [15.349288940429688, -19.79669189453125, 312.5049133300781, 181.26412963867188, 205.24331665039062, -55.53564453125, 391.8298645019531, 164.7801513671875, -70.29975891113281, 236.40609741210938, 321.7123107910156, 252.0278778076172, 248.7252197265625, -15.186767578125, -6.814643859863281, 238.6100616455078, 7.060150146484375, 314.1233825683594, -137.05056762695312, 117.280029296875, -268.8729248046875, 271.52496337890625, 64.10896301269531, 228.12249755859375, 115.38037109375, 444.3304443359375, 199.3094940185547, 171.92279052734375, 379.8477478027344, 288.6147766113281, -15.798126220703125, 103.45541381835938, 110.1358642578125, -104.89019775390625, 225.78521728515625, -275.66363525390625, -302.7757568359375, 520.953125, 93.03292846679688, -60.757568359375, 19.940692901611328, 12.790679931640625, 64.79031372070312, -134.7606658935547, -2.318450927734375, 162.83358764648438, 379.1919860839844, -194.61895751953125, 178.25985717773438, -51.3016357421875, 10.74359130859375, 87.28459930419922, 307.67364501953125, 198.4781494140625, 43.56927490234375, 137.78985595703125, 106.61837768554688, 412.18475341796875, 253.03640747070312, 17.4571533203125, -204.05929565429688, 179.7821044921875, 175.8203125, 140.14756774902344, 14.040351867675781, 248.0694580078125, 14.304950714111328, 276.8687744140625, 38.39464569091797, -573.0150146484375, 149.02658081054688, 628.365234375, -32.97080993652344, 0.0, 25.071136474609375, 226.2208251953125, 186.02642822265625, 387.04119873046875, 24.94482421875, -9.035572052001953, -175.86798095703125, 70.05615234375, 296.9500427246094, 234.60491943359375, -23.656509399414062, 155.88308715820312, 77.10256958007812, 250.61880493164062, 0.8255615234375, 56.039268493652344, 127.04743957519531, -210.61883544921875, 173.66384887695312, 38.58531188964844, 44.61629867553711, 189.91807556152344, 126.22161865234375, -80.08172607421875, 436.864501953125, 224.85301208496094, 192.6551513671875, 80.23165893554688, 63.75428771972656, 220.75018310546875, 176.3595428466797, 97.46784973144531, 268.00860595703125, 88.95467376708984, 452.5354309082031, 10.458158493041992, -134.76275634765625, -12.833864212036133, 391.4395446777344, -31.07366943359375, -89.5243148803711, -41.392913818359375, -103.45455169677734, -212.41107177734375, 62.34064483642578, 179.99627685546875, 178.08279418945312, 39.57781982421875, -76.11946105957031, 176.3131103515625, 1.25030517578125, -75.76324462890625, 870.4755249023438, -79.2803955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000366.npy"}
{"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 119.64830017089844, "std": 199.640869140625, "min": -293.1473388671875, "p10": -121.09154968261718, "median": 95.96142196655273, "p90": 366.33881835937495, "max": 998.70947265625, "pos_frac": 0.7421875, "sample": [218.11770629882812, 158.08627319335938, -131.82826232910156, 307.54290771484375, -16.219161987304688, 116.39593505859375, 31.760128021240234, -61.85919189453125, 30.057891845703125, 69.43264770507812, 125.57794189453125, -32.309532165527344, 105.17092895507812, 79.52381896972656, -64.3392333984375, 375.08856201171875, -117.10125732421875, 387.528564453125, -78.23077392578125, -21.557281494140625, 23.128204345703125, 378.6858215332031, 258.4200134277344, -169.7139892578125, 26.943607330322266, 279.90673828125, -220.0228729248047, -15.13494873046875, 96.04963684082031, -74.39883422851562, -242.2271728515625, 134.89968872070312, 17.116064071655273, 29.992889404296875, 275.8001403808594, -293.1473388671875, 58.556976318359375, 134.46197509765625, 164.31890869140625, -57.295379638671875, 337.359375, 95.87320709228516, 60.69873809814453, 188.8446044921875, 325.77374267578125, 67.24136352539062, -183.99349975585938, 76.369140625, 46.65861511230469, -241.0275115966797, 492.843994140625, 81.20193481445312, -62.75419616699219, -6.34100341796875, 438.4926452636719, 288.0707092285156, 267.32568359375, 197.08041381835938, -133.2913055419922, -96.64498901367188, 131.04196166992188, -120.9822998046875, 416.5506286621094, 168.86831665039062, 54.272335052490234, 166.31036376953125, -54.85188293457031, 158.86026000976562, 713.6021728515625, 251.19384765625, 184.21136474609375, 307.08282470703125, 41.008026123046875, 220.4559326171875, 339.43182373046875, 362.58892822265625, 56.749786376953125, 317.8686828613281, 127.47389221191406, 236.96136474609375, 157.41058349609375, 491.42596435546875, 264.27496337890625, 500.078125, -22.781494140625, 341.50689697265625, 175.82742309570312, 12.42169189453125, -255.19699096679688, 106.38697814941406, 146.0758514404297, 250.48861694335938, 23.81658935546875, -171.46923828125, 150.39341735839844, -7.6355133056640625, -122.53470611572266, 328.36297607421875, 84.36417388916016, 572.6451416015625, 181.9270782470703, 998.70947265625, 185.12969970703125, 181.92095947265625, 17.41363525390625, 128.65151977539062, 93.337890625, 218.1959228515625, 101.55963134765625, 242.6832275390625, 65.50401306152344, -42.64041519165039, 117.889892578125, -200.66868591308594, 468.2000732421875, 71.53081512451172, 1.7974853515625, 76.71976470947266, -32.32403564453125, 55.92578125, 198.88772583007812, 32.46537780761719, 58.1507568359375, 138.47039794921875, 418.0464782714844, -121.34646606445312, -20.877777099609375, 48.20259094238281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000367.npy"}
{"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 142.49530029296875, "std": 183.44033813476562, "min": -261.78533935546875, "p10": -56.70344314575195, "median": 128.70037841796875, "p90": 376.7054260253906, "max": 671.6422119140625, "pos_frac": 0.8125, "sample": [44.1483154296875, 312.4792175292969, 129.701171875, 180.59844970703125, -16.441375732421875, 507.3765563964844, 14.845565795898438, -261.78533935546875, 82.98150634765625, -1.392242431640625, 616.3810424804688, -102.47120666503906, 217.32481384277344, 559.6101684570312, 174.72189331054688, 323.2899169921875, 384.12921142578125, 130.52032470703125, 671.6422119140625, 90.20817565917969, 14.989212036132812, 0.15716552734375, 6.1070556640625, 21.637184143066406, 13.92767333984375, 71.84588623046875, 127.6995849609375, 38.139251708984375, -124.72227478027344, 304.50860595703125, 74.52964782714844, 154.070556640625, -116.78104400634766, 142.33993530273438, 64.35955810546875, -230.54571533203125, -93.334716796875, 545.02587890625, 125.848876953125, 193.2891845703125, 146.08363342285156, 232.8719482421875, 74.13522338867188, 209.797607421875, 29.429840087890625, 9.25244140625, -13.445144653320312, 100.06913757324219, 106.5986328125, -224.18531799316406, 43.62939453125, 244.4138946533203, 2.378662109375, 9.134445190429688, 57.1951904296875, -15.293411254882812, 262.025634765625, 224.17987060546875, -17.33819580078125, 162.33163452148438, -31.093002319335938, 159.85418701171875, 272.02947998046875, 514.7437744140625, 311.2647399902344, 228.52557373046875, -139.04376220703125, 288.80743408203125, 230.07623291015625, -9.356292724609375, -62.63996887207031, 314.3532409667969, 386.88616943359375, 98.95277404785156, 345.21893310546875, 347.60711669921875, 273.41693115234375, 33.1591796875, 240.5069580078125, 255.8760528564453, 28.3665771484375, 73.2816162109375, 137.07717895507812, 177.41912841796875, -54.159217834472656, 74.59375, 34.513458251953125, 164.77700805664062, 153.2996368408203, 415.0562744140625, -34.06224822998047, 36.704856872558594, 391.23382568359375, -47.33960723876953, 166.32891845703125, 109.4941177368164, 87.1613540649414, 42.329833984375, 85.98466491699219, 148.720947265625, 252.41751098632812, -28.323360443115234, 606.8489379882812, 333.299072265625, -123.60479736328125, 241.3502197265625, -116.97555541992188, 243.71951293945312, 373.5238037109375, 276.39068603515625, 79.894775390625, 94.24270629882812, 144.36973571777344, 46.03907775878906, 81.18296813964844, 146.2550811767578, 154.3077392578125, 194.6484375, 157.1705322265625, 147.33584594726562, 263.83172607421875, -246.84866333007812, -88.34762573242188, 670.271240234375, 260.0880126953125, 440.7801513671875, 126.5975341796875, 224.7811279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000368.npy"}
{"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 97.02571105957031, "std": 168.32565307617188, "min": -337.5498046875, "p10": -93.91104736328124, "median": 71.98382186889648, "p90": 326.0992218017578, "max": 577.3079223632812, "pos_frac": 0.7265625, "sample": [212.80555725097656, 177.31271362304688, 177.44943237304688, 57.364990234375, 3.798145294189453, 132.49246215820312, -14.457023620605469, 48.93701171875, 149.5537567138672, 70.4674072265625, 67.07904052734375, 124.216552734375, -159.759521484375, 239.60357666015625, 325.4790344238281, 90.89654541015625, 62.60089111328125, 152.7323455810547, 135.16851806640625, 327.54632568359375, 252.4300537109375, -44.28590393066406, 24.96253204345703, 193.8177490234375, -122.24432373046875, -144.11911010742188, 40.894378662109375, -52.02197265625, 244.46636962890625, 36.00410461425781, 145.90277099609375, 109.2991943359375, 99.2618408203125, 404.5738525390625, -36.43408203125, 357.7243347167969, -71.63877868652344, 298.44036865234375, 573.3267822265625, 7.3965606689453125, 34.677978515625, -91.346435546875, 389.56817626953125, -67.06588745117188, 16.285858154296875, -100.65313720703125, -26.42545509338379, 131.05194091796875, 172.17727661132812, 577.3079223632812, -31.507156372070312, -83.57838439941406, 189.17132568359375, 18.4896240234375, 235.97735595703125, 337.9515075683594, -35.777854919433594, -264.70257568359375, 486.91143798828125, 40.11907958984375, -99.8951416015625, 14.666030883789062, 1.1283035278320312, 304.3438720703125, 429.6210632324219, -1.5362815856933594, 264.6837463378906, 71.69182586669922, 58.08563232421875, -26.769805908203125, 366.44775390625, 74.21345520019531, 92.64676666259766, -0.49981689453125, 79.637451171875, 306.0204772949219, 312.76373291015625, 5.06640625, -337.5498046875, 156.55201721191406, 79.9171371459961, -192.17022705078125, -88.66526794433594, -39.9356689453125, -188.66952514648438, 65.28665924072266, 267.34844970703125, 403.24029541015625, 108.9886474609375, 0.0, 77.12940979003906, 384.09820556640625, -110.83619689941406, 4.851776123046875, 80.3572769165039, 17.10296630859375, 21.404403686523438, 28.899063110351562, 168.804931640625, 238.9491424560547, -76.68510437011719, 173.09471130371094, 134.6285400390625, 182.07522583007812, 139.15768432617188, 16.369529724121094, 496.6803894042969, 23.945297241210938, 232.72314453125, -129.65573120117188, 30.918869018554688, -3.601715087890625, 67.3570556640625, 123.17129516601562, -17.988296508789062, -52.8974609375, 133.09039306640625, 155.03875732421875, 173.17620849609375, -130.11898803710938, 71.59585571289062, 236.79788208007812, 72.27581787109375, -85.86680603027344, 271.7835693359375, 224.08848571777344, 154.6480712890625, -221.608154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000369.npy"}
{"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 122.04708099365234, "std": 221.02272033691406, "min": -640.7973022460938, "p10": -89.02361068725584, "median": 80.86456298828125, "p90": 375.15350341796875, "max": 980.61669921875, "pos_frac": 0.75, "sample": [-85.03756713867188, -358.8812561035156, 167.40948486328125, 94.70982360839844, 379.6024169921875, 254.37841796875, -147.80499267578125, -31.53759765625, 232.17474365234375, -53.510345458984375, 92.42744445800781, 55.100341796875, -28.168991088867188, 45.4193115234375, 257.0502624511719, 9.91192626953125, 18.21954345703125, 727.2542114257812, -78.0167465209961, 203.05484008789062, 577.8081665039062, 134.00868225097656, 92.0421142578125, 286.2341613769531, -71.73992919921875, -17.694564819335938, 4.056339263916016, 407.3213806152344, -26.75390625, 84.8399658203125, 15.05877685546875, -171.307861328125, 255.55650329589844, 30.44278335571289, 82.54287719726562, 46.1954345703125, 438.51513671875, -63.133056640625, 4.736419677734375, 205.25933837890625, 178.513671875, -273.96575927734375, 427.93804931640625, -3.6645355224609375, -56.08889389038086, 25.879196166992188, 363.675048828125, 191.2537841796875, 50.069580078125, -98.32437896728516, 54.126678466796875, 23.65528106689453, 79.18624877929688, 185.6629180908203, 363.65948486328125, -125.57290649414062, 252.90625, 138.57989501953125, 423.33349609375, -640.7973022460938, -6.642822265625, 541.0936889648438, -14.626251220703125, 84.617919921875, 7.327508926391602, 69.29118347167969, 574.8760986328125, 233.30592346191406, 179.41453552246094, 14.376615524291992, 42.436676025390625, 38.325042724609375, 141.54556274414062, 272.2196350097656, 373.246826171875, 66.06292724609375, 269.9756164550781, 862.30029296875, 287.4914855957031, 456.4051818847656, 348.03472900390625, 59.95213317871094, -1.981964111328125, 301.19952392578125, 352.76739501953125, 199.73226928710938, -124.66108703613281, 53.00140380859375, -352.2764892578125, 207.55300903320312, 172.96067810058594, 199.80621337890625, -134.10543823242188, 130.00091552734375, 980.61669921875, -26.85094451904297, 215.0132598876953, -55.919464111328125, 154.82669067382812, 6.46600341796875, -35.111690521240234, 319.1916809082031, 93.82791137695312, 297.044189453125, -107.08523559570312, 34.34562683105469, 325.2322082519531, 119.28067779541016, 241.27316284179688, 234.96502685546875, 398.81402587890625, -221.93389892578125, 12.422637939453125, -26.66387939453125, 61.88859558105469, 225.423095703125, 229.9896240234375, 27.151676177978516, 212.741943359375, 77.21905517578125, 9.40570068359375, 48.959999084472656, 50.38627624511719, 257.9176025390625, -66.15361785888672, -189.60902404785156, 180.9055633544922, 3.2481689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000370.npy"}
{"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 73.13240814208984, "std": 189.04937744140625, "min": -666.584228515625, "p10": -115.6605331420898, "median": 66.17034912109375, "p90": 296.5264953613281, "max": 694.6041870117188, "pos_frac": 0.671875, "sample": [154.47149658203125, 46.050384521484375, 200.2081298828125, 337.55340576171875, -7.204280853271484, -18.570785522460938, 339.87847900390625, 92.59564208984375, -2.9632568359375, 103.2603759765625, 26.643753051757812, -290.9719543457031, -96.41592407226562, 437.5775451660156, 88.45883178710938, 71.2359619140625, -139.399169921875, 94.25941467285156, 60.958984375, -95.22552490234375, -163.00592041015625, 269.88909912109375, -44.66853332519531, 144.841796875, -35.30729675292969, 214.54168701171875, 9.758035659790039, 115.511962890625, -105.48683166503906, 197.53460693359375, 109.49111938476562, 419.99639892578125, -19.712600708007812, -68.69305419921875, -356.8429260253906, 290.96075439453125, -456.3411560058594, 109.61585998535156, 26.379562377929688, -241.481689453125, 223.276611328125, 99.34555053710938, -31.298324584960938, -40.9251708984375, 136.3150634765625, 80.52200317382812, 52.112060546875, 0.0, -164.2279052734375, 45.20636749267578, 308.18377685546875, 174.29486083984375, -13.40478515625, -666.584228515625, -271.5165100097656, 17.4644775390625, 201.99484252929688, 291.530517578125, -18.18499755859375, 105.57799530029297, -45.12521743774414, 111.01495361328125, 71.35470581054688, 532.0731201171875, 24.93426513671875, 112.414306640625, 79.61740112304688, 33.2144775390625, 108.70877838134766, -202.9392852783203, 694.6041870117188, 61.19586181640625, 280.27105712890625, 43.02655029296875, 144.200927734375, -5.165336608886719, -12.169868469238281, 81.20489501953125, 141.30609130859375, -46.03965759277344, -99.27334594726562, -57.656005859375, -2.708648681640625, 234.54722595214844, 119.95362091064453, 42.243316650390625, -316.846923828125, 125.17776489257812, 117.59942626953125, 23.457717895507812, 21.949142456054688, -190.1962890625, 249.0108642578125, 196.8343048095703, 235.8974609375, 41.781089782714844, 313.32098388671875, 188.54696655273438, 158.68551635742188, 414.0216369628906, 95.47322845458984, -66.49235534667969, 321.62213134765625, 201.04904174804688, 198.34146118164062, 137.5537109375, -83.1124267578125, 60.692413330078125, 114.73516082763672, 71.14483642578125, 51.75175476074219, -177.007568359375, 16.9561767578125, 8.227397918701172, -1.3370361328125, 157.1402587890625, 248.2574920654297, 226.43923950195312, 508.47467041015625, 381.1292724609375, -52.422332763671875, -59.24407958984375, -52.82740783691406, -28.87494659423828, 1.1408309936523438, 277.63116455078125, 53.07585906982422, 378.31793212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000371.npy"}
{"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 124.14448547363281, "std": 227.07737731933594, "min": -642.404052734375, "p10": -103.15381927490233, "median": 101.87370300292969, "p90": 347.23333435058595, "max": 989.1293334960938, "pos_frac": 0.75, "sample": [80.24462890625, 5.53753662109375, 100.5010986328125, -30.459186553955078, -6.9293365478515625, 30.56689453125, 107.18585205078125, -37.31146240234375, 103.24630737304688, 712.0384521484375, 14.361412048339844, 294.41583251953125, 66.30364990234375, 243.0601348876953, -92.89740753173828, 57.79876708984375, -10.160018920898438, 543.6094970703125, 35.74079895019531, -8.746116638183594, 901.094482421875, 33.89916229248047, 172.8209991455078, 46.78631591796875, -385.8124084472656, 484.8259582519531, 310.6624755859375, 75.37625122070312, 254.7693328857422, 163.10833740234375, -125.35104370117188, -32.66552734375, 226.85772705078125, -201.46392822265625, 48.645294189453125, 20.25946044921875, 121.60797119140625, 54.98143005371094, 123.99600982666016, 117.18920135498047, -65.76238250732422, 588.94677734375, -200.3467254638672, 186.20278930664062, 389.2684326171875, -642.404052734375, 257.5025634765625, 416.8722839355469, 124.96025848388672, 346.9760437011719, -25.45416259765625, -17.14324951171875, 37.46026611328125, -22.017974853515625, -10.711273193359375, 48.260711669921875, 19.701690673828125, 253.28347778320312, 65.46993255615234, 219.17251586914062, 224.37649536132812, 109.493408203125, 16.82183837890625, 159.63018798828125, 22.926082611083984, 106.92881774902344, 294.58221435546875, 22.29876708984375, 495.37274169921875, 38.38481903076172, 39.51214599609375, 164.24688720703125, 187.89950561523438, 330.8880615234375, 407.21136474609375, 322.0042724609375, -237.69427490234375, 196.8031463623047, 342.9504089355469, -197.76559448242188, 320.42529296875, 82.21186065673828, 269.4200439453125, 35.88885498046875, -126.55381774902344, -167.84912109375, 114.54833984375, 226.040771484375, -2.5044193267822266, 154.52053833007812, -99.30989074707031, 62.59326171875, 248.00570678710938, 306.09661865234375, 111.69384765625, -304.08111572265625, -186.1887664794922, 264.77532958984375, 97.20014953613281, -74.15958404541016, 160.322509765625, 47.6959228515625, 2.36688232421875, 21.259078979492188, 230.6465606689453, 287.5403137207031, -308.9048156738281, 376.97125244140625, 821.5220336914062, 272.2957763671875, 199.02035522460938, 989.1293334960938, 66.22480773925781, -16.349868774414062, 158.06454467773438, 185.67510986328125, -0.9287014007568359, 114.414306640625, -112.12298583984375, 306.6927490234375, 19.634521484375, 347.83367919921875, -24.732051849365234, 248.29086303710938, -38.74744415283203, 137.02288818359375, 153.75439453125, 276.3240051269531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000372.npy"}
{"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 119.14329528808594, "std": 197.30563354492188, "min": -425.163818359375, "p10": -131.49320678710936, "median": 108.37509155273438, "p90": 362.31721801757806, "max": 777.9617919921875, "pos_frac": 0.78125, "sample": [44.03924560546875, 82.05082702636719, 242.54891967773438, 241.09609985351562, 7.039897918701172, -329.65203857421875, 63.411895751953125, 289.2623596191406, 87.91668701171875, 354.923095703125, 14.947265625, -138.084716796875, 128.5515899658203, 356.626708984375, 102.822509765625, 431.63079833984375, 50.004459381103516, 62.086944580078125, -103.28488159179688, 278.04754638671875, 280.29827880859375, 219.31617736816406, 298.69384765625, 142.77081298828125, -94.7451171875, 152.86912536621094, 22.3284912109375, -260.080078125, 67.01861572265625, 260.0447692871094, 34.05902862548828, -310.3587646484375, 169.707275390625, 24.298828125, 288.6282653808594, 204.97705078125, 405.755615234375, -34.342315673828125, 236.53460693359375, 157.244873046875, 309.24908447265625, 215.05613708496094, 234.6182861328125, 26.046661376953125, 14.827880859375, 255.55078125, 46.09187316894531, -297.7514343261719, 397.81561279296875, 233.3011474609375, 5.7662200927734375, 230.54415893554688, -209.0933837890625, -125.34271240234375, 42.7213134765625, -156.79083251953125, 229.32781982421875, -128.66827392578125, 184.46612548828125, 165.8291015625, 3.746335983276367, -425.163818359375, 357.29571533203125, -40.35063171386719, 244.35064697265625, 46.246734619140625, -85.5649185180664, 315.32257080078125, 41.899444580078125, 97.36701965332031, 87.25772094726562, 184.3155059814453, 79.03681182861328, 115.81175231933594, 120.64452362060547, 55.497406005859375, 113.92767333984375, -32.8355712890625, 71.05635070800781, 15.752357482910156, -262.8646240234375, -180.9006805419922, 374.0340576171875, -187.30233764648438, 241.1155548095703, 777.9617919921875, 217.14927673339844, 265.04278564453125, 255.00762939453125, -41.502685546875, 80.25360107421875, 282.0122985839844, 64.41082763671875, 116.59175109863281, 499.2193298339844, 266.95050048828125, -60.45344543457031, 260.154296875, 14.591121673583984, -25.36090087890625, 195.52207946777344, 69.49586486816406, 385.8553771972656, 517.322021484375, 154.065185546875, 32.990692138671875, 97.90365600585938, -10.121444702148438, 440.3620300292969, 189.31280517578125, 24.125751495361328, -49.44359588623047, 452.68804931640625, 418.8509521484375, 59.067962646484375, -293.7525329589844, 252.5729217529297, 48.80482482910156, -37.228759765625, 229.22901916503906, -4.571638107299805, 210.39581298828125, -165.11053466796875, 156.26687622070312, 523.391357421875, 388.35601806640625, 175.96823120117188, 261.7294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000373.npy"}
{"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 104.19580078125, "std": 191.40301513671875, "min": -408.73114013671875, "p10": -76.24683227539062, "median": 64.90299224853516, "p90": 386.14757690429684, "max": 709.055419921875, "pos_frac": 0.65625, "sample": [184.40350341796875, 0.0, 83.78726196289062, 478.3158874511719, -35.3807373046875, -21.63836669921875, 31.400436401367188, 48.301483154296875, 592.3298950195312, 279.2894287109375, 27.275104522705078, 396.05035400390625, 476.5613708496094, -46.84513854980469, 421.54638671875, 40.821136474609375, 49.16792297363281, 166.8041534423828, 160.88168334960938, 108.00588989257812, -37.293373107910156, -52.031341552734375, 382.5668029785156, -32.271240234375, 330.0909118652344, -57.8162841796875, 102.87136840820312, 5.964630126953125, -4.268163681030273, -134.93106079101562, -30.28209686279297, 67.69471740722656, 95.68069458007812, 43.83747863769531, -61.0201416015625, 123.19554901123047, 247.2886962890625, -78.48696899414062, 310.6860046386719, 310.99615478515625, 206.47055053710938, -58.743988037109375, 709.055419921875, -19.612586975097656, 190.92230224609375, 8.86181640625, -6.481479644775391, 174.21987915039062, -404.1977844238281, -33.20799255371094, 223.66928100585938, -198.76052856445312, 166.90753173828125, 18.949172973632812, 161.876220703125, -96.6329345703125, 443.0404052734375, 165.69366455078125, 394.5027160644531, -23.35767364501953, 177.69703674316406, 251.88909912109375, 304.675048828125, -19.715438842773438, 88.98466491699219, 111.83364868164062, 335.0001220703125, 168.9305419921875, -27.09295654296875, 258.8958740234375, 120.12741088867188, 94.1759033203125, 69.11082458496094, 128.17967224121094, -315.6605224609375, -23.15869140625, -109.716552734375, 17.50372314453125, 415.74188232421875, 170.79586791992188, 268.2105712890625, -150.166748046875, 198.32293701171875, 174.6538848876953, 0.0, 21.159366607666016, 36.415103912353516, 33.9534912109375, -127.64224243164062, 62.11126708984375, 46.069664001464844, 71.39564514160156, 299.26409912109375, 243.21356201171875, -1.15966796875, 116.22442626953125, 429.3896179199219, 89.34130859375, 397.8115539550781, 458.91534423828125, 2.55474853515625, -26.900001525878906, 7.8583526611328125, -34.42503356933594, 36.85256576538086, -67.5316390991211, 262.9573974609375, 348.0223388671875, -209.84019470214844, -20.53852081298828, -88.1934814453125, -54.795166015625, -16.54241943359375, 26.285186767578125, 176.544921875, -15.188507080078125, 308.5240478515625, -75.28677368164062, -140.962158203125, 80.48478698730469, 605.623779296875, 240.043212890625, 40.0615234375, 223.85205078125, 295.453369140625, -27.744625091552734, -13.780548095703125, -408.73114013671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000374.npy"}
{"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 86.01385498046875, "std": 182.47894287109375, "min": -322.1925048828125, "p10": -129.77535705566407, "median": 59.74365234375, "p90": 323.308609008789, "max": 582.6103515625, "pos_frac": 0.671875, "sample": [57.45347595214844, 37.7234001159668, -16.771926879882812, -212.91542053222656, 136.79034423828125, 146.67709350585938, 159.32876586914062, -257.50079345703125, 27.757984161376953, 46.36676025390625, 262.3079833984375, 123.65257263183594, 387.1156005859375, 298.3134765625, -13.9771728515625, 2.84521484375, -136.8594970703125, 252.95506286621094, 141.08450317382812, 43.06773376464844, 292.52008056640625, 318.7173767089844, -45.240325927734375, 42.04826354980469, 96.41275024414062, 227.15890502929688, 89.010009765625, -43.88623809814453, -129.2888641357422, 171.143798828125, 112.20121765136719, 38.47406005859375, -63.72747802734375, -263.95648193359375, -35.609981536865234, 312.57452392578125, 168.02139282226562, 250.84005737304688, 106.91402435302734, -208.83987426757812, -25.908905029296875, 114.14541625976562, -65.93359375, 10.190216064453125, -174.7037353515625, 582.6103515625, 224.3455810546875, 173.3028564453125, -52.28424072265625, -46.586456298828125, -186.83421325683594, 3.872711181640625, 420.80877685546875, 66.01673126220703, 32.48462677001953, -49.810546875, 57.74029541015625, 75.07077026367188, -122.449462890625, -118.45477294921875, 342.5588073730469, 476.4710693359375, -50.79644775390625, 489.97979736328125, -36.180824279785156, 79.3326416015625, -78.77716064453125, 66.73601531982422, 553.5867919921875, -19.612171173095703, -76.35568237304688, 21.670238494873047, 270.36553955078125, 98.66840362548828, -25.384429931640625, 255.697509765625, 440.33685302734375, -67.60614013671875, 104.68051147460938, -130.91050720214844, 217.6534423828125, 61.74700927734375, 5.538970947265625, -197.4312744140625, 161.7790985107422, 75.20054626464844, -44.869873046875, 185.13970947265625, 116.32318115234375, 266.5537109375, -83.91988372802734, 236.3740234375, 334.021484375, -46.06646728515625, 416.04833984375, 388.297607421875, -322.1925048828125, 218.36862182617188, 228.4036102294922, 366.8901672363281, -223.56243896484375, -197.48684692382812, 82.92611694335938, -100.74353790283203, 196.54150390625, -19.639923095703125, 292.98199462890625, 281.616943359375, -111.38018798828125, 54.148681640625, -8.06243896484375, 284.61541748046875, 207.9933624267578, 177.87545776367188, 207.0843505859375, 36.873443603515625, 21.56355094909668, 7.638236999511719, 2.8908653259277344, -192.51715087890625, 249.70968627929688, 6.668556213378906, 411.0576477050781, 6.743408203125, 89.98129272460938, 194.971435546875, 2.475555419921875, -90.03726196289062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000375.npy"}
{"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 113.07284545898438, "std": 212.39544677734375, "min": -422.5770263671875, "p10": -95.27032470703125, "median": 72.0814208984375, "p90": 392.68732910156245, "max": 742.616455078125, "pos_frac": 0.6953125, "sample": [-419.01776123046875, 350.11212158203125, 285.48773193359375, 486.9169616699219, 204.326416015625, 742.616455078125, 36.1185302734375, 31.780502319335938, 59.23480224609375, -7.936676025390625, 346.27142333984375, -83.166015625, -213.45095825195312, 144.8028564453125, -219.97824096679688, 104.65400695800781, -65.88353729248047, 16.8719482421875, 41.7332763671875, 431.42962646484375, 388.253173828125, -208.47280883789062, 99.79817962646484, 72.70123291015625, 585.02783203125, -137.99313354492188, -229.09791564941406, 230.95770263671875, -14.82586669921875, 293.03387451171875, 207.85028076171875, 162.2939910888672, 200.1149444580078, 435.13671875, 82.46672058105469, -28.8131103515625, 221.69415283203125, -62.87975311279297, 641.57373046875, -164.205810546875, 0.0, 42.42334747314453, 224.74261474609375, 13.161857604980469, 63.572113037109375, 533.3037109375, 89.77713012695312, 279.2867431640625, 185.59532165527344, 19.75006103515625, -274.49267578125, 269.40264892578125, 123.4979248046875, 28.53173828125, 405.1078796386719, 223.98880004882812, 265.55230712890625, 161.96697998046875, -51.67475891113281, -11.7064208984375, 232.56886291503906, 20.96429443359375, 738.2776489257812, 236.56390380859375, 42.87432861328125, 236.701416015625, 258.569580078125, 28.39354705810547, -94.63330078125, -28.01171875, 122.69793701171875, -96.7567138671875, 4.45989990234375, 71.20906066894531, 28.294282913208008, -35.3037109375, -149.89181518554688, -17.764591217041016, -315.750244140625, 175.75115966796875, -42.8519287109375, 230.84219360351562, 294.5062561035156, 117.75286865234375, 71.46160888671875, 61.95539855957031, 45.28003692626953, -422.5770263671875, 8.2421875, -40.927154541015625, 267.3161315917969, -41.11457824707031, 158.0565948486328, -19.09288787841797, 12.425716400146484, 267.0768127441406, -103.65547180175781, -80.73553466796875, 42.587493896484375, 48.30030059814453, -4.677978515625, 135.58551025390625, -59.012298583984375, 86.53888702392578, 254.5899658203125, -2.7391395568847656, 174.10011291503906, 194.38551330566406, 617.2561645507812, 574.215087890625, 68.01251220703125, -90.39669799804688, -70.5191650390625, 97.53741455078125, 377.6767883300781, -17.509788513183594, 94.771728515625, -66.06280517578125, 256.3802795410156, 33.85619354248047, -5.565876007080078, 365.06060791015625, 403.03369140625, 101.32977294921875, 297.48590087890625, 101.56135559082031, 427.177490234375, 155.865234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000376.npy"}
{"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 75.96499633789062, "std": 214.91195678710938, "min": -896.6996459960938, "p10": -124.77696990966797, "median": 42.34353256225586, "p90": 308.1388275146484, "max": 1002.6849365234375, "pos_frac": 0.65625, "sample": [349.10516357421875, 75.28359985351562, 131.784912109375, 26.867355346679688, 171.9505615234375, 43.15031433105469, -19.4609375, 107.16213989257812, 173.70016479492188, -146.8453369140625, 134.33871459960938, -75.4759521484375, 204.17788696289062, -113.8809814453125, 151.39402770996094, 43.47845458984375, 306.54656982421875, -123.099609375, -64.02130126953125, 0.42950439453125, 81.79348754882812, 11.415908813476562, 4.866950988769531, 24.152376174926758, -58.723548889160156, -19.89377212524414, 310.2796936035156, 151.83767700195312, 263.5074462890625, -46.726898193359375, -92.06745910644531, 59.43511962890625, -39.035430908203125, -460.306396484375, 231.97560119628906, 159.36480712890625, 42.727996826171875, -6.9580078125, -138.0804443359375, 124.62274169921875, 165.74365234375, 240.68246459960938, 6.096405029296875, -896.6996459960938, 103.6104507446289, 13.1397705078125, 472.41973876953125, 131.15350341796875, -28.178512573242188, -87.81884765625, 10.00006103515625, 302.54296875, 11.497550964355469, 58.91902160644531, 159.2496337890625, -179.44178771972656, -294.8788146972656, 1.6538009643554688, 41.959068298339844, 429.9625244140625, 221.963134765625, 108.959228515625, -54.977203369140625, -132.096923828125, -210.6986083984375, 307.2178955078125, -6.069244384765625, 0.0, -24.225830078125, 63.487586975097656, 122.04330444335938, -95.12708282470703, 47.613677978515625, 301.4084777832031, -5.757537841796875, 140.10693359375, 1.4650421142578125, -74.80596923828125, 12.042144775390625, 183.9090576171875, 346.78570556640625, 267.488037109375, -84.8726806640625, -122.82615661621094, -58.796783447265625, -9.8575439453125, 36.80040740966797, 138.74752807617188, 435.3398132324219, 378.1121826171875, 198.7821044921875, -94.44711303710938, 226.1929931640625, 703.0377197265625, 28.35418701171875, 209.78228759765625, -3.27001953125, 0.998565673828125, 59.599853515625, 245.22360229492188, 189.1768798828125, -282.6446838378906, -128.90609741210938, 196.88943481445312, 451.9967956542969, -64.17903900146484, 144.35122680664062, -18.845809936523438, 266.25738525390625, 218.56069946289062, 141.5590057373047, 6.838706970214844, 299.0971374511719, 1002.6849365234375, 330.81317138671875, 220.3675537109375, 27.3780517578125, -236.08038330078125, -45.649444580078125, 307.2213134765625, -7.540122985839844, 21.76495361328125, -326.69989013671875, 389.49951171875, -128.69081115722656, -119.87788391113281, 12.19842529296875, 405.9591979980469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000377.npy"}
{"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 106.44680786132812, "std": 207.44158935546875, "min": -525.5523681640625, "p10": -111.00729980468749, "median": 89.85449600219727, "p90": 382.74446411132806, "max": 849.8759155273438, "pos_frac": 0.6875, "sample": [63.32666015625, -51.11798095703125, -51.91162109375, 301.4017028808594, 8.29840087890625, 110.28273010253906, -52.89556884765625, 32.25494384765625, 173.01034545898438, 369.64031982421875, 7.907508850097656, -69.70028686523438, 132.03195190429688, 192.71031188964844, 398.2279052734375, -369.4573974609375, -525.5523681640625, 151.3873291015625, 610.2733764648438, -220.10324096679688, 224.48834228515625, 353.08087158203125, -36.5406494140625, -195.7093048095703, -70.40029907226562, 435.00469970703125, 537.2308349609375, 147.2291717529297, 113.07492065429688, 217.05804443359375, 78.73612976074219, 222.2971649169922, 200.90927124023438, -107.12158203125, 173.40188598632812, 88.28251647949219, -84.82986450195312, 77.5716552734375, -62.89128875732422, 239.674072265625, 0.0, 413.2893981933594, -74.49563598632812, 471.10699462890625, 669.6963500976562, -83.98480224609375, -28.729103088378906, 324.92364501953125, 4.10791015625, 282.1585693359375, 197.99884033203125, 171.02890014648438, -80.6812973022461, 313.77130126953125, -22.62432861328125, 91.42647552490234, -283.5041809082031, 80.91215515136719, -183.64337158203125, 849.8759155273438, 420.6187744140625, 1.1881179809570312, 1.4634475708007812, 94.170654296875, 470.61029052734375, 183.47174072265625, 35.20429992675781, 112.027587890625, 439.74432373046875, 154.5738525390625, 189.35882568359375, 138.20077514648438, 6.776123046875, 36.51390075683594, 1.9985389709472656, 223.7476806640625, -47.05964660644531, -152.44888305664062, 74.30157470703125, -38.8330078125, 166.35386657714844, 9.51141357421875, 27.5228271484375, 131.5967559814453, -49.242340087890625, -9.83441162109375, 332.46142578125, 209.32583618164062, 181.13644409179688, -187.279296875, 240.3021240234375, 257.7545471191406, -29.341995239257812, 121.69183349609375, 234.75054931640625, 65.51277160644531, -92.68017578125, 103.05232238769531, 68.514404296875, -132.7767333984375, -120.073974609375, 312.3797607421875, 74.59011840820312, -76.14463806152344, -138.31153869628906, -4.98663330078125, -6.251953125, 200.7030029296875, 16.489898681640625, 176.11294555664062, 70.39581298828125, 508.50341796875, -10.262781143188477, -98.13250732421875, 223.13720703125, 376.10870361328125, 408.47802734375, 265.2713317871094, 170.94400024414062, 265.9656066894531, -249.16143798828125, 101.81190490722656, 187.80880737304688, 115.5189208984375, 178.66656494140625, -169.35714721679688, 81.08624267578125, -103.25503540039062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000378.npy"}
{"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 96.81559753417969, "std": 185.62545776367188, "min": -322.460205078125, "p10": -143.49624633789062, "median": 86.63621139526367, "p90": 331.80529174804684, "max": 641.8836669921875, "pos_frac": 0.703125, "sample": [-29.595699310302734, -29.533920288085938, 26.63141632080078, -13.473770141601562, 2.1786727905273438, 96.7237548828125, -1.4931182861328125, 156.57080078125, 38.031463623046875, 59.89513397216797, 96.72918701171875, 105.64297485351562, 11.029884338378906, 87.83206176757812, -11.777557373046875, 54.514068603515625, -6.727893829345703, -146.5435791015625, 285.3123474121094, 107.89967346191406, 228.34420776367188, 251.60086059570312, 144.70822143554688, -116.25128173828125, 50.75841522216797, -75.3702392578125, 372.2776794433594, 126.04248046875, -274.4169921875, 22.95178985595703, 72.91802978515625, 244.51025390625, 290.1328125, 204.60958862304688, 146.6002960205078, -76.91110229492188, 192.3182830810547, 641.8836669921875, 209.15728759765625, 71.56488037109375, 422.4140625, -218.0491943359375, 2.113250732421875, -108.66029357910156, 167.06329345703125, 251.10540771484375, -26.00273895263672, -153.9595489501953, -197.14215087890625, 119.182373046875, -25.25927734375, -175.88482666015625, 37.2452392578125, -274.3218994140625, -77.79745483398438, 182.42889404296875, 264.7690124511719, 59.29278564453125, 86.25695037841797, 93.8277587890625, 162.19052124023438, -104.50555419921875, 17.26507568359375, 188.2729949951172, 288.1952819824219, 161.4268798828125, 208.3528594970703, 324.7435302734375, 9.67169189453125, -138.80514526367188, -76.27218627929688, 0.0, -200.37745666503906, 177.55780029296875, 184.5657958984375, 247.86627197265625, 212.8367919921875, 303.392578125, 342.5658264160156, 478.6393127441406, 74.21923828125, 152.23025512695312, 37.335418701171875, -90.75241088867188, 351.8490295410156, 225.59982299804688, -2.3431243896484375, 143.79403686523438, -166.18789672851562, 237.7138671875, 221.92007446289062, 18.851058959960938, 349.35400390625, 37.13326644897461, 277.0991516113281, -38.138671875, 41.61006164550781, 87.01547241210938, -118.87334442138672, 418.4642333984375, 350.5101623535156, -322.460205078125, -65.65478515625, 87.15771484375, 199.21856689453125, 288.53466796875, 423.69512939453125, -142.19024658203125, 279.35595703125, 327.1936340332031, 352.906494140625, 239.40167236328125, 631.5567626953125, 209.5050048828125, 40.45360565185547, -165.36346435546875, 28.860153198242188, 13.527322769165039, 56.528839111328125, 149.656005859375, -210.7603302001953, 188.29502868652344, -138.8902587890625, 46.553863525390625, -182.87240600585938, 194.58468627929688, -114.06689453125, 535.7864379882812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000379.npy"}
{"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 112.0062026977539, "std": 232.12940979003906, "min": -644.522705078125, "p10": -99.66706085205077, "median": 67.57272338867188, "p90": 407.98875732421874, "max": 1072.233642578125, "pos_frac": 0.671875, "sample": [186.98187255859375, 316.3793640136719, 248.8100128173828, 132.78395080566406, -16.829132080078125, 53.98388671875, -644.522705078125, 81.38742065429688, 20.611228942871094, -48.75579833984375, 169.72286987304688, 177.97088623046875, 7.4700927734375, -230.46278381347656, 123.2928466796875, 14.496498107910156, 569.7637939453125, 7.77130126953125, 423.663818359375, 25.65082550048828, 0.0, 209.20445251464844, -0.6794052124023438, 75.1473388671875, 99.4710693359375, 153.7122802734375, -212.42201232910156, 456.772705078125, 36.794189453125, 68.68597412109375, 113.1265869140625, 251.71197509765625, 91.9781494140625, 153.13778686523438, 66.45947265625, -147.1926727294922, 268.9698181152344, -104.66407775878906, 133.6620330810547, -32.579498291015625, 218.880859375, 422.4426574707031, 419.1856384277344, -43.245025634765625, -151.00064086914062, -54.0950927734375, 83.69818115234375, 347.39935302734375, 1072.233642578125, -47.97503662109375, -132.854248046875, 116.59442138671875, 186.69271850585938, -97.52548217773438, 96.87124633789062, -19.331558227539062, 3.6259765625, 51.9710693359375, 380.11712646484375, 348.71112060546875, 126.45819091796875, 735.4417724609375, 229.5366973876953, 288.4985046386719, 246.0213623046875, 258.7740478515625, -250.0540771484375, -50.7359619140625, -5.09759521484375, -72.65392303466797, 405.56451416015625, 583.9578857421875, 21.38663673400879, 183.5050811767578, 155.75888061523438, -85.50299835205078, 433.2301330566406, -191.84823608398438, -3.126434326171875, 38.01799011230469, -20.92197036743164, 850.2672119140625, 8.4605712890625, 355.89080810546875, 230.45654296875, -18.60198211669922, 36.518218994140625, 161.41702270507812, 234.00796508789062, 21.88219451904297, 226.08914184570312, 270.50408935546875, -48.56280517578125, 174.93411254882812, -10.843391418457031, -199.76608276367188, -12.752532958984375, -454.7113037109375, 706.0162353515625, 359.49493408203125, -2.4626502990722656, -62.70628356933594, 103.73847961425781, 166.53656005859375, 57.87223815917969, 24.312973022460938, 226.48382568359375, 214.43588256835938, -47.83885192871094, 8.939746856689453, -25.09912109375, -64.23716735839844, 12.749618530273438, 61.12502670288086, 97.42122650146484, -113.3460693359375, 10.06207275390625, 183.65731811523438, -36.487884521484375, 324.27301025390625, 586.04443359375, 15.2811279296875, -61.019805908203125, 87.77642822265625, -75.02098083496094, -22.26195526123047, 413.64532470703125, -167.85557556152344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000380.npy"}
{"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 128.8124542236328, "std": 227.75582885742188, "min": -366.5635681152344, "p10": -83.51104049682617, "median": 88.25669860839844, "p90": 356.3911437988281, "max": 1110.8099365234375, "pos_frac": 0.6875, "sample": [166.10324096679688, 526.5693359375, 208.3524169921875, -103.06932830810547, 354.48089599609375, -68.23880004882812, 225.16610717773438, 827.268310546875, 225.27711486816406, 89.52963256835938, 153.6953125, -5.858734130859375, 402.19293212890625, 320.24212646484375, 433.6941833496094, 344.907958984375, 0.0, 472.1279296875, 18.245758056640625, 180.42694091796875, 1073.1552734375, 160.82977294921875, -155.0973358154297, 203.84515380859375, 279.48468017578125, 111.02088928222656, 42.79778289794922, -296.17041015625, 0.0, 43.0770263671875, 75.526123046875, 165.3763427734375, -366.5635681152344, 250.54885864257812, 27.1507568359375, 75.65672302246094, 40.74334716796875, 303.2667541503906, -3.9857177734375, 3.1324462890625, 301.76416015625, 130.09686279296875, -307.28125, 176.83042907714844, 514.3353271484375, -18.925506591796875, 325.4614562988281, 39.821998596191406, -41.495079040527344, 183.53024291992188, 172.51754760742188, 209.13682556152344, 224.0289764404297, 8.545486450195312, 138.51181030273438, 338.01513671875, 250.1962890625, 310.7181396484375, 805.7103271484375, 147.8951416015625, 75.05033874511719, -72.27691650390625, 261.1659240722656, -29.015625, -40.869049072265625, 71.50263977050781, 227.13092041015625, -48.259765625, -70.79104614257812, -42.989898681640625, 159.15695190429688, -26.17532730102539, -14.066436767578125, 316.87518310546875, -36.6708984375, 82.20681762695312, -115.87031555175781, -61.49482727050781, 56.79345703125, -37.15436935424805, 120.74176025390625, -35.02599334716797, 182.435791015625, 225.22802734375, 293.3982849121094, -37.25274658203125, -128.98541259765625, -44.393592834472656, 277.35369873046875, 29.543060302734375, 86.9837646484375, -131.14349365234375, 165.27264404296875, 137.81704711914062, 54.33283996582031, -112.055908203125, -13.547279357910156, 97.5628662109375, -13.615371704101562, -152.4837646484375, 35.339107513427734, -81.5956039428711, 27.05023193359375, -30.089675903320312, -195.21530151367188, 456.1939697265625, 67.9215087890625, 100.67637634277344, -87.98039245605469, 70.16552734375, 143.46707153320312, 562.0938110351562, 130.56637573242188, 323.96514892578125, 236.64163208007812, 34.95646667480469, 53.188087463378906, 65.50843811035156, 360.848388671875, -69.64430236816406, 139.50390625, 1110.8099365234375, 97.2446517944336, -128.28237915039062, -29.32586669921875, 406.2283630371094, 186.6953125, 130.32574462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000381.npy"}
{"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 136.54391479492188, "std": 194.79705810546875, "min": -320.421875, "p10": -104.03470382690429, "median": 138.26412200927734, "p90": 366.844497680664, "max": 709.3609008789062, "pos_frac": 0.75, "sample": [-57.307891845703125, 90.56890869140625, 267.95562744140625, 28.984710693359375, 306.88519287109375, 188.7803955078125, 703.135498046875, -126.74838256835938, 178.61207580566406, 76.88349151611328, 195.6571044921875, -320.421875, 39.79901123046875, 208.8592071533203, 128.0272216796875, 24.476593017578125, 158.57730102539062, -22.31414794921875, 264.32757568359375, -148.17910766601562, 88.34173583984375, 14.157272338867188, 420.64373779296875, -70.7742919921875, 21.56890106201172, 176.88572692871094, 67.6787109375, -101.19136047363281, 239.76885986328125, 147.70555114746094, 0.52642822265625, 275.71868896484375, -111.0850830078125, 3.64837646484375, 124.90045166015625, 240.43698120117188, 178.34384155273438, -1.4255828857421875, -48.68017578125, 271.1413269042969, 143.88238525390625, 77.42556762695312, 257.31951904296875, 130.33514404296875, 378.09442138671875, 153.5727081298828, 295.8612060546875, 284.2677307128906, 251.54165649414062, 61.209686279296875, 709.3609008789062, 139.25411987304688, 181.75436401367188, 320.53973388671875, -168.06478881835938, 137.2741241455078, 267.4441833496094, 250.08721923828125, 24.27630615234375, -17.879058837890625, -69.14834594726562, -70.40777587890625, 386.7422180175781, 178.4309539794922, 69.08781433105469, 6.910770416259766, 305.99505615234375, 607.11181640625, 400.41632080078125, 16.38751220703125, 357.32965087890625, 126.636474609375, -26.012924194335938, 73.9727783203125, 324.6390380859375, 204.99838256835938, -147.14028930664062, 89.58877563476562, 228.28335571289062, 323.88348388671875, 151.63873291015625, -60.064117431640625, 223.65328979492188, 49.141624450683594, -9.167945861816406, 143.61282348632812, -101.4091796875, 462.35498046875, -77.88276672363281, 362.0231018066406, 502.090576171875, 61.64459228515625, 117.66661071777344, 379.5516357421875, 382.6500244140625, -260.99041748046875, 192.28094482421875, -169.31143188476562, 105.0179443359375, -80.08575439453125, 253.9027099609375, 353.63470458984375, 79.98744201660156, 314.64825439453125, 72.11907958984375, -244.68011474609375, -10.043144226074219, 287.7093811035156, -110.16092681884766, 152.14453125, 160.52352905273438, -34.11212158203125, 247.50453186035156, 85.7158203125, 36.806488037109375, -12.805221557617188, -249.86898803710938, 303.3006896972656, -123.0993423461914, 199.08863830566406, 200.46697998046875, -146.2041015625, 629.6748657226562, 283.2502746582031, 199.9051971435547, 272.413330078125, 556.0166015625, -44.76318359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000382.npy"}
{"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 117.7965087890625, "std": 219.26699829101562, "min": -621.139892578125, "p10": -101.00270309448241, "median": 105.35885238647461, "p90": 387.92659606933586, "max": 1040.951416015625, "pos_frac": 0.6875, "sample": [46.543426513671875, 164.9742431640625, 370.094482421875, 204.18997192382812, -54.83882141113281, 49.2705078125, -73.49179077148438, -124.1639404296875, -0.03851318359375, -22.719573974609375, 223.45538330078125, 13.936050415039062, 219.23361206054688, 623.6788330078125, 0.0, -365.4100341796875, 148.79107666015625, -212.58721923828125, 144.9981689453125, -330.166748046875, 63.3533935546875, 525.71728515625, 283.05047607421875, 302.6346435546875, -52.0460205078125, 266.6414794921875, 39.180328369140625, -80.83319854736328, -0.7774887084960938, -52.54366683959961, 280.540283203125, 1040.951416015625, 105.57111358642578, -96.88294219970703, 405.98370361328125, 170.29986572265625, 259.9601745605469, 476.8280029296875, 91.0992431640625, 161.74795532226562, 39.81072235107422, 12.307548522949219, 286.5660095214844, 308.06341552734375, -139.45909118652344, -4.047637939453125, 403.0743713378906, 219.49261474609375, 137.27548217773438, 11.376220703125, 124.98757934570312, 52.298736572265625, -76.62129974365234, 233.63223266601562, 165.6690673828125, -271.4283447265625, -11.177787780761719, 206.7586212158203, -10.313848495483398, -97.32186889648438, 194.58511352539062, 241.10028076171875, -0.5781841278076172, 463.72406005859375, -621.139892578125, 58.482757568359375, 120.12493896484375, 105.92520141601562, -240.7390594482422, 162.00436401367188, 47.625518798828125, 205.71502685546875, 592.7237548828125, 232.16354370117188, 163.27081298828125, 67.28681182861328, 37.94976806640625, 3.35125732421875, 202.10150146484375, 285.2406921386719, 127.46353149414062, 105.14659118652344, 118.00921630859375, 487.50689697265625, 293.723388671875, 381.4346923828125, 37.964691162109375, -5.931980133056641, 87.20842742919922, -91.09844970703125, 193.94891357421875, -16.742610931396484, 677.3582153320312, -81.66294860839844, -35.542449951171875, -3.902008056640625, -214.9441680908203, 259.2158203125, 232.84735107421875, 326.4869689941406, 236.1217041015625, 156.00747680664062, 276.33221435546875, 95.25759887695312, -3.42816162109375, 72.45417785644531, 108.0120849609375, -195.06069946289062, 0.0, -0.8840351104736328, 473.28900146484375, -109.59131622314453, 129.480224609375, 45.58697509765625, 83.45852661132812, -150.06092834472656, 319.15887451171875, 136.38885498046875, 487.259033203125, -93.52255249023438, 39.640380859375, 16.05715560913086, 506.232177734375, -112.0342025756836, 162.817626953125, -16.517547607421875, 209.32041931152344, 199.60086059570312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000383.npy"}
{"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 105.52330017089844, "std": 189.60009765625, "min": -315.6916198730469, "p10": -124.73311767578122, "median": 103.14936828613281, "p90": 343.5583862304687, "max": 734.94091796875, "pos_frac": 0.6796875, "sample": [310.2327880859375, -193.42376708984375, 7.812530517578125, -33.3133544921875, -49.6864013671875, 303.6570129394531, 152.10043334960938, 369.6370849609375, 250.25770568847656, 34.87451934814453, 185.89053344726562, -56.759490966796875, 195.4350128173828, -192.07614135742188, 281.0635986328125, -202.09417724609375, 297.405029296875, 49.593116760253906, 308.00079345703125, -110.43862915039062, 1.7295494079589844, 312.747314453125, 186.95602416992188, -64.98287963867188, 59.7308349609375, 442.89111328125, 35.666412353515625, 171.64791870117188, 58.46875, -40.092529296875, 18.341733932495117, -26.88421630859375, 146.48028564453125, 103.02816772460938, -62.307098388671875, 111.0702133178711, 60.52165222167969, 164.97006225585938, 36.482452392578125, 102.86421203613281, -145.56576538085938, -8.643524169921875, -6.568687438964844, -44.61438751220703, 230.95506286621094, 255.76187133789062, 341.9617614746094, -182.3602294921875, -315.6916198730469, 106.16263580322266, 112.52462768554688, 224.01141357421875, -12.123117446899414, 103.27056884765625, -239.05712890625, 174.65748596191406, 381.8143615722656, 224.9212646484375, -98.1217041015625, 77.39596557617188, -42.195953369140625, -62.52099609375, 164.56549072265625, 240.52293395996094, 524.9146118164062, 488.713623046875, 108.200927734375, 42.613037109375, 322.0771789550781, -60.49969482421875, 127.97193908691406, 16.641647338867188, 383.3472900390625, 199.2127685546875, 158.26605224609375, 230.09463500976562, -39.924346923828125, -35.28462219238281, -19.123764038085938, 383.048828125, 44.5345458984375, -157.80807495117188, 138.03623962402344, 179.14837646484375, 309.5051574707031, 734.94091796875, 168.914306640625, -115.80484008789062, 116.28913879394531, 123.99273681640625, 210.08755493164062, 79.03363037109375, 155.60986328125, 704.4561767578125, -169.98443603515625, -194.31497192382812, 77.6334228515625, 433.7286071777344, 132.04750061035156, -30.932518005371094, 5.493133544921875, -159.47726440429688, 206.103515625, -110.0491943359375, -295.87774658203125, 29.722951889038086, -6.4809417724609375, 15.996881484985352, 162.30856323242188, -53.39208984375, 472.312255859375, -21.06915283203125, 234.21673583984375, 78.69154357910156, -26.415910720825195, 45.916748046875, 147.00045776367188, 347.2838439941406, 185.19110107421875, 192.6651611328125, 226.6828155517578, -35.26637268066406, -29.275997161865234, 178.05780029296875, 440.1575622558594, 296.09112548828125, 149.9234619140625, -147.47296142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000384.npy"}
{"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 87.740478515625, "std": 178.63735961914062, "min": -297.6834716796875, "p10": -103.30339889526365, "median": 56.13296127319336, "p90": 303.26441650390626, "max": 926.5250244140625, "pos_frac": 0.6796875, "sample": [-145.35983276367188, 16.01498031616211, 302.35595703125, 152.57879638671875, 57.0123291015625, -116.23553466796875, 318.84869384765625, 147.89541625976562, 199.9049072265625, 531.9171142578125, -100.25845336914062, -133.64190673828125, -69.19841003417969, 126.84735107421875, 39.79229736328125, 41.891845703125, 138.30908203125, 293.033935546875, 196.24551391601562, -6.588165283203125, 276.322021484375, -84.25616455078125, 1.98516845703125, -169.4392852783203, -35.163238525390625, 163.5148468017578, 208.9282989501953, -1.209991455078125, 36.27301025390625, 128.40841674804688, -12.6378173828125, 319.00146484375, 199.83877563476562, -180.68612670898438, 333.57171630859375, 426.38873291015625, -11.82232666015625, -0.126861572265625, 144.92442321777344, 333.9441833496094, -18.497772216796875, 42.274986267089844, 382.713134765625, 220.3271026611328, 243.7789306640625, -27.670272827148438, 287.4310302734375, 400.1535339355469, -264.7401428222656, 140.230712890625, 926.5250244140625, 183.282470703125, 293.234130859375, -75.3437728881836, 11.751052856445312, 65.09403991699219, 169.53366088867188, -11.255456924438477, -260.8818664550781, 51.27909851074219, -3.71881103515625, 414.3620910644531, -121.690185546875, 490.6167297363281, 239.5579833984375, 55.492919921875, -5.573394775390625, 48.917938232421875, -78.52104187011719, 192.5098876953125, 60.01850128173828, 75.68219757080078, -56.14996337890625, 67.60028076171875, 48.022125244140625, -134.75173950195312, 175.57998657226562, -50.2012939453125, 275.5445861816406, -45.9114990234375, 137.39772033691406, 75.71205139160156, 233.05087280273438, 36.74748992919922, 138.28408813476562, 36.18379211425781, -243.0097198486328, 116.82034301757812, -297.6834716796875, 36.41229248046875, 238.14222717285156, -51.26027297973633, 26.421875, 127.3897705078125, 499.28363037109375, 19.52911376953125, 0.0, 101.00794982910156, 134.15594482421875, 49.49293518066406, -66.7779312133789, 134.6285858154297, 103.0364990234375, -54.2109375, 166.91781616210938, 56.77300262451172, 82.1834716796875, -7.310489654541016, 34.762359619140625, 305.3841552734375, 215.45068359375, 0.0, 12.918060302734375, -282.30072021484375, 71.57318115234375, 16.91717529296875, -89.52725219726562, 4.5540771484375, -28.915557861328125, 175.18548583984375, 6.8033905029296875, -110.40827178955078, -80.70068359375, 85.86578369140625, 23.926254272460938, 189.83251953125, 289.75372314453125, 84.65560913085938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000385.npy"}
{"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 140.36346435546875, "std": 210.67881774902344, "min": -302.1082763671875, "p10": -82.88205566406249, "median": 92.86178970336914, "p90": 412.0774780273437, "max": 822.1102294921875, "pos_frac": 0.75, "sample": [90.42453002929688, -105.79624938964844, 35.15522766113281, 94.07456970214844, 221.6240234375, 74.84417724609375, 67.514404296875, 392.6680908203125, 28.912673950195312, 214.354248046875, 239.36611938476562, 539.0227661132812, -246.38360595703125, 107.23126220703125, 570.613037109375, -137.5526885986328, 34.965087890625, -79.68170166015625, 11.606353759765625, 684.8380737304688, 201.73643493652344, 262.4734802246094, 80.61099243164062, 123.33758544921875, -1.208251953125, -146.9064483642578, 58.52680969238281, -91.75497436523438, 270.2652587890625, 115.81854248046875, 63.00494384765625, 124.11978149414062, 240.35543823242188, 19.74676513671875, -40.47511291503906, 76.36758422851562, -46.06904602050781, 15.568939208984375, 232.64649963378906, 320.44940185546875, 407.1754150390625, 257.535400390625, 61.72212219238281, 109.05081176757812, 148.30259704589844, 355.0627746582031, 551.5015869140625, -269.1627197265625, 86.02406311035156, 339.0440673828125, 153.014404296875, -31.0460205078125, -72.76348876953125, -193.86312866210938, 272.0893249511719, 123.47857666015625, -32.942832946777344, 0.0, 31.270538330078125, -78.11463165283203, 27.40771484375, -42.3939208984375, 305.29541015625, 117.60137939453125, 11.416282653808594, 49.409515380859375, 108.623779296875, 65.83087158203125, 249.1464080810547, -228.54896545410156, 372.7291259765625, 79.66525268554688, 310.05078125, 366.7720947265625, 822.1102294921875, 235.724609375, 5.803459167480469, -8.392868041992188, 212.67434692382812, 116.03350830078125, 389.4092712402344, 168.99676513671875, -302.1082763671875, 77.60797119140625, 54.623626708984375, 245.26315307617188, 241.20297241210938, -24.80279541015625, 39.017303466796875, -90.34954833984375, -117.35873413085938, -32.75035095214844, 57.20135498046875, 0.0, 515.8084716796875, 113.35577392578125, 67.9613037109375, 53.409820556640625, 257.97119140625, 258.81298828125, 423.515625, 384.0178527832031, 312.03143310546875, 720.1905517578125, 57.901329040527344, 450.3457336425781, 172.5101318359375, -20.61541748046875, -32.010986328125, 236.39569091796875, 124.143798828125, 205.13619995117188, 579.1951293945312, -60.383514404296875, -155.5403289794922, 241.62564086914062, -45.03684997558594, 31.2691650390625, 1.5384750366210938, 628.4420776367188, 182.28256225585938, 401.0150451660156, 91.64900970458984, -210.13427734375, 466.86065673828125, 0.0, 491.7665710449219, 204.38739013671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000386.npy"}
{"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 103.26756286621094, "std": 217.38230895996094, "min": -814.8128662109375, "p10": -119.26919708251953, "median": 105.70172882080078, "p90": 330.5018737792969, "max": 1012.75146484375, "pos_frac": 0.7109375, "sample": [193.2053680419922, -113.12926483154297, -15.192047119140625, 107.45220184326172, 51.959869384765625, 263.63592529296875, 251.64169311523438, -20.721649169921875, -580.8162231445312, -208.74493408203125, 0.0, 269.2117919921875, -150.327880859375, 315.11749267578125, 419.37567138671875, 1012.75146484375, 30.34801483154297, -64.983154296875, -94.92219543457031, 89.7862548828125, -156.68943786621094, 308.53204345703125, 258.4525146484375, -196.52349853515625, 127.11445617675781, 344.0074462890625, -46.312957763671875, 108.73460388183594, 275.47723388671875, 45.827674865722656, 285.8628845214844, 318.50067138671875, 225.1944580078125, 317.2911071777344, 262.355224609375, 120.45828247070312, 17.54888916015625, 157.52590942382812, -66.53394317626953, 2.305572509765625, -117.2176513671875, 272.1230773925781, 41.87571716308594, -54.826934814453125, 276.5590515136719, -7.673164367675781, 119.47933959960938, 108.98367309570312, 5.076650619506836, 40.766998291015625, 128.36895751953125, 443.02618408203125, -243.0194091796875, 4.64453125, -52.206787109375, 207.8228302001953, 120.84454345703125, -149.1563720703125, 238.47177124023438, 58.689422607421875, 349.1201171875, 134.0164794921875, 237.97967529296875, 42.97393798828125, 120.34091186523438, -814.8128662109375, 89.39923095703125, 175.07852172851562, -73.63555908203125, -117.18669128417969, -137.33938598632812, 55.282012939453125, 93.37847900390625, 224.04428100585938, 366.60003662109375, 98.16361999511719, 342.7165832519531, 107.22789764404297, -34.44708251953125, 9.262847900390625, 243.9928741455078, -10.1805419921875, 319.63336181640625, -68.4210205078125, 186.66896057128906, 366.8711242675781, 99.70033264160156, 29.887420654296875, -51.41650390625, 241.783935546875, -124.05613708496094, 62.383155822753906, -39.21913146972656, 68.03875732421875, 342.89337158203125, 201.73663330078125, 138.41668701171875, -46.1075439453125, 322.9090270996094, 433.5546875, 59.92039108276367, 163.27935791015625, -36.371826171875, 338.4862976074219, -13.371192932128906, -185.15655517578125, -504.7882995605469, 117.92092895507812, 104.1755599975586, 619.6227416992188, -57.0438232421875, 156.54852294921875, 112.63893127441406, 242.17916870117188, 330.34515380859375, -21.385658264160156, 76.50191497802734, 289.0070495605469, 273.38720703125, 110.82974243164062, 21.736236572265625, -250.40383911132812, 321.5142822265625, 329.91351318359375, 24.585968017578125, 303.7143249511719, 64.95164489746094, 330.8675537109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000387.npy"}
{"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 71.39234161376953, "std": 210.20689392089844, "min": -443.7137145996094, "p10": -173.66323242187497, "median": 48.03253936767578, "p90": 340.3490905761719, "max": 808.506103515625, "pos_frac": 0.6328125, "sample": [32.130126953125, -25.04010009765625, 211.83154296875, 0.0, 670.7813720703125, 191.0023193359375, -45.461883544921875, 297.0130615234375, -90.42369079589844, 265.2379150390625, -303.8385009765625, 167.00750732421875, -200.0642852783203, -2.577728271484375, -20.241683959960938, 342.23687744140625, 95.23182678222656, -128.62081909179688, 200.841064453125, 8.3878173828125, 532.4017333984375, 437.91949462890625, -292.6881408691406, -44.89471435546875, -4.180084228515625, -443.7137145996094, 79.84561920166016, 271.234375, 10.885986328125, 198.66412353515625, -266.0377197265625, -108.39306640625, 0.0, 77.97052001953125, 37.26359558105469, -61.504669189453125, -96.798583984375, 44.05670166015625, 60.03448486328125, 64.2730941772461, 79.76895141601562, 44.578895568847656, 511.4034729003906, 202.1035919189453, -242.14608764648438, 47.0059814453125, 143.16354370117188, 15.074874877929688, -55.51043701171875, 38.57334899902344, -2.5409317016601562, 55.177490234375, 400.936279296875, -378.96441650390625, 100.16650390625, 91.73050689697266, 253.33709716796875, 172.08963012695312, -62.39129638671875, -59.427764892578125, -99.27889251708984, 130.63262939453125, 165.21331787109375, -162.34849548339844, 100.8961410522461, -299.41168212890625, 201.44747924804688, -129.70767211914062, 436.6652526855469, -0.10141181945800781, 278.476806640625, 339.5400390625, 223.3614501953125, 171.77645874023438, 71.79458618164062, 251.80221557617188, 60.555419921875, 480.36224365234375, 158.082275390625, -148.33868408203125, 56.80812072753906, 371.7488708496094, 250.3578643798828, -88.17843627929688, 20.10826873779297, 182.56863403320312, -85.31611633300781, 203.1014862060547, 2.93585205078125, 152.171630859375, -229.27447509765625, 153.5990447998047, -308.31695556640625, -210.83599853515625, -116.38953399658203, 481.2871398925781, 808.506103515625, 115.03800201416016, 1.6424942016601562, 74.50457000732422, 27.798965454101562, 34.6776123046875, 290.8673095703125, 413.74664306640625, -12.506103515625, 128.24635314941406, -121.75442504882812, -18.33251953125, -366.05377197265625, -8.165146827697754, 231.64840698242188, 269.0301513671875, -211.66329956054688, -88.24057006835938, 32.144989013671875, -22.234264373779297, 394.1490478515625, 48.08412170410156, 6.40814208984375, 159.8829345703125, -42.526039123535156, -132.20135498046875, 107.12265014648438, 148.83663940429688, 47.98095703125, 142.94644165039062, 245.16102600097656, -150.22262573242188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000388.npy"}
{"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 114.30616760253906, "std": 219.76608276367188, "min": -414.7886657714844, "p10": -115.1572280883789, "median": 76.06527709960938, "p90": 432.560952758789, "max": 740.26171875, "pos_frac": 0.65625, "sample": [-49.61109924316406, -21.5989990234375, 10.440719604492188, 88.46148681640625, 184.20111083984375, 177.36273193359375, 58.48980712890625, 57.187652587890625, -111.73133850097656, 482.97412109375, 202.9822998046875, 404.5098876953125, -101.80010986328125, 300.766845703125, -79.1717529296875, 6.9075927734375, -84.72711181640625, 8.535186767578125, -50.642974853515625, 686.1904296875, -95.3348617553711, 101.54715728759766, 226.576904296875, 348.37835693359375, -51.25845718383789, 28.473968505859375, -141.23114013671875, 178.69769287109375, 187.05418395996094, 740.26171875, -6.0413818359375, -269.8782958984375, 66.82022094726562, -178.22496032714844, -26.713836669921875, -262.944091796875, -222.03778076171875, 123.08686828613281, 442.7005920410156, 477.0731506347656, 333.5023193359375, 0.0, 57.65025329589844, 85.52398681640625, 375.4814147949219, -69.3685302734375, 519.8717651367188, 210.49696350097656, 314.3885498046875, 503.8528137207031, 85.65086364746094, 198.13827514648438, 76.75799560546875, 53.3812255859375, 259.13983154296875, 114.136962890625, 124.84500122070312, 483.33612060546875, -56.39373779296875, -95.5272216796875, 21.160293579101562, -16.81145477294922, 555.6334838867188, 336.65576171875, -197.77496337890625, 60.80830383300781, 263.83489990234375, 428.21539306640625, 136.05563354492188, 117.29571533203125, 40.167449951171875, -65.7513427734375, 96.99716186523438, 328.4056396484375, -29.87677001953125, 314.925537109375, 15.92755126953125, 83.7841796875, -93.09555053710938, 75.37255859375, 290.8159484863281, 223.84793090820312, 341.0329284667969, 736.1953735351562, -35.64618682861328, 641.8882446289062, -242.7238311767578, 204.37998962402344, -45.23060607910156, -152.13751220703125, 200.5538330078125, 45.382537841796875, 246.36874389648438, -29.65863800048828, -194.73202514648438, 42.04072570800781, 193.86790466308594, -36.17376708984375, -123.15097045898438, 166.18975830078125, -91.47137451171875, 323.002685546875, -54.4739990234375, -6.408271789550781, 151.87164306640625, 335.2139587402344, 0.0, 186.1971435546875, 164.40261840820312, 244.61837768554688, 93.45720672607422, -64.31382751464844, -103.28656005859375, -103.77203369140625, 456.59002685546875, 339.73040771484375, 262.6297607421875, 150.1005401611328, -146.5506134033203, 17.37537384033203, 54.01658630371094, 231.88766479492188, -254.60028076171875, -26.508956909179688, 3.3767852783203125, -414.7886657714844, 71.06930541992188, 455.1863098144531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000389.npy"}
{"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 111.90046691894531, "std": 192.04000854492188, "min": -421.76483154296875, "p10": -111.65430908203125, "median": 100.78089904785156, "p90": 337.7410430908203, "max": 628.3814697265625, "pos_frac": 0.6953125, "sample": [273.757568359375, -72.18663024902344, 36.03924560546875, 212.85302734375, -132.37503051757812, 93.91462707519531, 37.024131774902344, 44.85626220703125, 528.3720703125, 143.98587036132812, -63.937591552734375, 44.154388427734375, -25.65172576904297, 23.252197265625, 341.5353088378906, 315.3443298339844, -83.45223236083984, 249.55650329589844, 327.6021728515625, 161.61410522460938, 356.40191650390625, -6.2897796630859375, 294.4051818847656, 35.68437194824219, 300.99774169921875, -34.71441650390625, -127.64227294921875, 276.7169189453125, 219.37249755859375, 167.82144165039062, 336.11492919921875, 141.0625762939453, 51.23356628417969, 254.18228149414062, -19.49822998046875, -3.4052734375, 17.360660552978516, 154.98716735839844, -136.95172119140625, 86.65412902832031, -177.00604248046875, 296.1060791015625, -141.80169677734375, 183.04647827148438, 62.854000091552734, -169.61270141601562, -76.73435974121094, -318.0299072265625, 206.91107177734375, -30.2384033203125, -117.67437744140625, 243.93560791015625, -30.381088256835938, 52.5738525390625, 628.3814697265625, -161.27584838867188, -16.49761962890625, 187.18502807617188, 163.79112243652344, -65.5980224609375, 67.36148071289062, 521.2239990234375, 300.9396057128906, -80.6174545288086, 365.3586120605469, 95.22276306152344, 589.9890747070312, 294.75445556640625, -71.82522583007812, 399.44403076171875, 62.01409912109375, 283.6636657714844, -109.07427978515625, 0.0, 120.51995849609375, -7.982391357421875, 223.1343994140625, -300.800048828125, 260.0169677734375, 247.46795654296875, -204.511962890625, 240.02456665039062, 223.03573608398438, -421.76483154296875, -69.28817749023438, 295.3409423828125, 576.3112182617188, 311.9730224609375, 78.08599853515625, -265.4205627441406, 9.417415618896484, 343.221923828125, -16.6240234375, 221.4833984375, 124.88363647460938, 69.53614044189453, -80.05963134765625, 368.64642333984375, 144.53558349609375, -78.4664306640625, 18.79575538635254, 59.20204162597656, 544.3187255859375, 210.9072265625, -93.69218444824219, 185.036865234375, 407.3316650390625, 12.52777099609375, 134.77980041503906, 127.061279296875, -43.324737548828125, 159.3009033203125, 63.046844482421875, 86.53839111328125, 109.20346069335938, -4.3726806640625, 154.2183837890625, 155.99627685546875, 162.09765625, 293.91900634765625, 106.33903503417969, 128.56494140625, 82.8697509765625, 324.75518798828125, 197.46627807617188, -102.254150390625, 7.2206268310546875, 163.54727172851562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000390.npy"}
{"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 123.5382080078125, "std": 217.09947204589844, "min": -358.22119140625, "p10": -110.52243347167968, "median": 106.73062896728516, "p90": 358.5581359863281, "max": 878.993408203125, "pos_frac": 0.7265625, "sample": [31.315887451171875, -250.9599151611328, 63.397239685058594, 56.875064849853516, 200.17178344726562, 89.75634765625, -44.273681640625, 29.795913696289062, -108.17005920410156, -0.9226284027099609, 91.61805725097656, 66.19287109375, -69.86430358886719, -20.235214233398438, 277.29229736328125, 123.49969482421875, 101.1849136352539, 237.84988403320312, -116.01130676269531, 45.16045379638672, -256.5316162109375, 108.72296905517578, 701.058349609375, 799.13330078125, -56.02851867675781, 81.658203125, 230.67337036132812, -19.11773681640625, -8.102550506591797, -26.3341064453125, -79.694091796875, -358.22119140625, 219.059814453125, 180.266845703125, 122.007080078125, 324.8144226074219, 29.791778564453125, 5.70936393737793, -261.91033935546875, 511.1680908203125, 80.00301361083984, 265.8547058105469, 140.6112823486328, 878.993408203125, 249.9417724609375, 60.23295593261719, 237.9578094482422, 145.8734588623047, -59.54977035522461, -37.627044677734375, 358.5259094238281, 161.10682678222656, -82.03077697753906, 242.64016723632812, -16.180206298828125, 169.92051696777344, 108.52522277832031, 551.3795776367188, 258.074462890625, -8.547393798828125, 282.7469177246094, -264.65625, 130.62448120117188, 35.08647918701172, 126.15069580078125, 501.40692138671875, 555.4761352539062, 15.434955596923828, 170.5673065185547, 39.45439147949219, 22.99957275390625, 305.72882080078125, 29.52410888671875, 155.9546356201172, 215.47262573242188, -53.730712890625, 454.65594482421875, 110.001953125, -246.0570068359375, 303.03045654296875, 78.16427612304688, 206.02975463867188, 296.9066467285156, 446.64385986328125, 133.583740234375, -18.775543212890625, -63.81153106689453, 303.300048828125, 336.27056884765625, -23.17034912109375, -12.15533447265625, 111.92402648925781, -79.56353759765625, 328.961669921875, 307.68023681640625, 126.25320434570312, 26.03485107421875, 230.86306762695312, 290.03717041015625, 191.18359375, -355.71234130859375, -201.23406982421875, 135.81637573242188, -144.03424072265625, 22.883651733398438, 82.87997436523438, 37.67701721191406, 126.94253540039062, 302.98248291015625, 362.3589172363281, 37.724395751953125, 209.72190856933594, 307.3811950683594, 60.813690185546875, 133.6494140625, 298.13507080078125, 104.93603515625, 358.6333312988281, 666.9605102539062, 242.67254638671875, -124.836181640625, -239.24063110351562, 86.98431396484375, 92.37501525878906, -21.489761352539062, 151.88088989257812, -279.48187255859375, 491.77392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000391.npy"}
{"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 134.77157592773438, "std": 195.4231414794922, "min": -722.0026245117188, "p10": -39.84673156738281, "median": 129.9314727783203, "p90": 330.37689208984375, "max": 726.0489501953125, "pos_frac": 0.7578125, "sample": [146.03768920898438, 193.57821655273438, 370.7450256347656, 241.27737426757812, 212.1460418701172, 8.04345703125, 123.70654296875, 259.4368591308594, 103.907470703125, 332.5972900390625, -171.14102172851562, 88.78329467773438, 261.5320129394531, 100.36566162109375, -28.576904296875, -58.135292053222656, -722.0026245117188, 395.699462890625, 108.06552124023438, -15.139640808105469, 264.6930847167969, -18.016216278076172, -197.33047485351562, 176.55694580078125, 172.02947998046875, 228.32720947265625, 0.0, -143.913330078125, 152.24050903320312, 604.89599609375, -34.20643615722656, 194.85769653320312, 303.59521484375, 383.82244873046875, 123.13096618652344, 177.472900390625, 588.617431640625, 79.24073791503906, 106.15875244140625, -58.88568115234375, 384.11248779296875, 225.7286376953125, 13.556648254394531, 84.155517578125, -39.661529541015625, 261.98126220703125, 12.58856201171875, -40.27886962890625, 151.3463897705078, 85.06534576416016, 222.75238037109375, 272.1510314941406, 319.7288513183594, 183.1217041015625, 655.0913696289062, 171.9260711669922, -14.494842529296875, 256.3836669921875, 13.68951416015625, -21.383056640625, -35.63002014160156, -18.722923278808594, 50.39601135253906, 311.602783203125, 298.97576904296875, 90.8436279296875, -37.362091064453125, 246.447998046875, 663.805908203125, -23.42462158203125, 323.7799377441406, 315.5926513671875, 526.9368896484375, 185.22207641601562, 165.32955932617188, 341.68475341796875, 143.24037170410156, -254.21714782714844, -28.742141723632812, 13.90908432006836, 74.72989654541016, 252.14532470703125, -65.83041381835938, 185.293212890625, 96.21730041503906, 274.7992858886719, 43.437530517578125, 70.26217651367188, 40.52778625488281, 46.47442626953125, 37.7960205078125, 0.0, 168.22296142578125, 31.35687255859375, 154.72747802734375, 273.18060302734375, 116.6983642578125, 170.42623901367188, 214.97601318359375, -40.28050231933594, 283.5175476074219, -15.495599746704102, 122.61994934082031, 19.90582847595215, 180.4600830078125, 272.49853515625, -9.378875732421875, 230.43426513671875, 209.6116485595703, 42.329376220703125, 74.16220092773438, 145.08963012695312, 95.57268524169922, -30.99163818359375, 418.325439453125, 2.89569091796875, 726.0489501953125, -68.8165283203125, -360.7870178222656, 311.910888671875, 136.15640258789062, 247.65652465820312, -218.8673858642578, 242.52203369140625, -38.372772216796875, 202.17401123046875, 329.42529296875, 23.5509033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000392.npy"}
{"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 95.44483184814453, "std": 198.29588317871094, "min": -489.30523681640625, "p10": -127.75753631591795, "median": 69.48172760009766, "p90": 352.92045288085933, "max": 829.032958984375, "pos_frac": 0.734375, "sample": [10.69017219543457, 59.776039123535156, -22.897232055664062, -27.522491455078125, 48.029579162597656, -121.1668701171875, 480.374267578125, 412.66290283203125, 281.9928894042969, 26.2921142578125, 92.20721435546875, 15.011322021484375, 116.89109802246094, 14.394790649414062, 32.13897705078125, 75.6207275390625, 11.442516326904297, 70.03080749511719, 252.40664672851562, 228.05987548828125, 146.03228759765625, 96.446044921875, 272.1822814941406, 107.42984008789062, -250.68701171875, 381.5437927246094, 191.16366577148438, 423.8463439941406, 42.9178466796875, 141.9442901611328, 446.52734375, 8.65521240234375, -108.91043090820312, 150.29690551757812, -24.31427001953125, 132.30825805664062, -93.53831481933594, 256.13409423828125, 133.68927001953125, 114.10404968261719, 528.7647705078125, -222.43325805664062, 147.90927124023438, 11.3936767578125, 108.42559814453125, 231.5843505859375, -98.81756591796875, -17.165130615234375, -149.56826782226562, 208.03323364257812, 544.3124389648438, 90.13434600830078, 9.296737670898438, 107.05287170410156, 146.5784912109375, 64.36688232421875, 8.080411911010742, -171.4244842529297, 193.32423400878906, 52.993316650390625, -489.30523681640625, 115.74066162109375, 175.45455932617188, -54.2298583984375, 89.7044906616211, 76.62490844726562, -152.88079833984375, 233.99093627929688, -308.3873291015625, 25.8206787109375, -262.7252197265625, 68.93264770507812, 56.23667907714844, 829.032958984375, 46.42536163330078, 38.156097412109375, 562.7724609375, -197.57113647460938, 424.23516845703125, -70.90089416503906, 60.34327697753906, -6.265678405761719, 350.39947509765625, -204.0442352294922, 101.6943130493164, 66.54175567626953, 248.34674072265625, -11.984481811523438, 261.2619323730469, 178.8692626953125, -149.17471313476562, 553.441650390625, 266.1689758300781, 280.6981201171875, 588.5789794921875, 94.85955810546875, -13.700469970703125, -86.7755126953125, 79.85400390625, -328.409423828125, -5.196537017822266, 358.802734375, 140.7935791015625, 10.507797241210938, 257.11968994140625, 244.7887420654297, 97.41389465332031, 0.0, 67.40901184082031, 119.6077880859375, 243.8392333984375, 4.121429443359375, 50.7335205078125, 7.25616455078125, -9.8785400390625, -143.13575744628906, 198.58248901367188, 10.688528060913086, -100.17681884765625, -11.855361938476562, 142.0340576171875, -120.69802856445312, 158.3291015625, 191.46511840820312, 4.6894073486328125, -35.29058837890625, 60.72840881347656, 319.3802795410156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000393.npy"}
{"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 101.28084564208984, "std": 236.19216918945312, "min": -450.460205078125, "p10": -110.95886535644532, "median": 55.187950134277344, "p90": 380.9990051269531, "max": 1185.94384765625, "pos_frac": 0.6015625, "sample": [202.770751953125, -103.47122192382812, 179.6759033203125, 113.3226318359375, 1185.94384765625, 336.0150451660156, -53.7637939453125, -72.85408020019531, 31.265289306640625, 140.33712768554688, 121.2178955078125, -19.308868408203125, 55.44512939453125, -95.69149780273438, -20.30120849609375, -89.79197692871094, 279.3749084472656, -192.06436157226562, 507.17413330078125, -18.115564346313477, 54.16107177734375, 391.5540771484375, 350.9301452636719, 5.891902923583984, -6.683065414428711, -111.66567993164062, 71.20697021484375, 185.8953857421875, 244.1297607421875, 84.80511474609375, -221.08441162109375, 0.0, -11.881805419921875, -36.28950500488281, 721.381103515625, -9.02337646484375, -119.25318908691406, 35.009857177734375, -165.11489868164062, 161.37660217285156, 403.28955078125, 362.01214599609375, 150.49227905273438, -292.32098388671875, 29.777576446533203, 216.2740478515625, 249.49452209472656, 35.3306999206543, 447.54290771484375, -32.70751953125, -155.96975708007812, -12.957916259765625, 167.56182861328125, -103.874267578125, -49.52398681640625, 361.4452209472656, -96.14418029785156, -28.308349609375, 92.80271911621094, -52.874237060546875, 178.19732666015625, -68.40496826171875, -3.177825927734375, 24.319786071777344, 371.03289794921875, 25.290124893188477, -47.0032958984375, 126.93988037109375, -81.78813934326172, 243.207763671875, 76.84025573730469, 662.9801635742188, 536.323974609375, 4.766105651855469, 638.8989868164062, -55.41265869140625, -450.460205078125, -108.65191650390625, -29.892303466796875, 355.89935302734375, -46.120887756347656, -93.343017578125, 354.0574645996094, 4.642547607421875, -203.63320922851562, 116.390625, 110.16259765625, 158.76983642578125, 180.12950134277344, -16.9415283203125, -15.61883544921875, -58.5217399597168, 193.41696166992188, 137.30836486816406, -110.65594482421875, 148.28855895996094, 24.812408447265625, 189.54827880859375, 139.9520721435547, 194.29925537109375, -209.31895446777344, 364.4047546386719, 151.0887451171875, -85.92007446289062, 178.8084716796875, 533.9887084960938, -367.8720397949219, 4.99822998046875, 246.8413848876953, 171.576171875, 200.76092529296875, 54.93077087402344, 421.2479553222656, 751.7138061523438, 226.48822021484375, -95.55859375, 422.82177734375, -0.662841796875, 90.39382934570312, 146.68499755859375, 77.78460693359375, 376.47540283203125, -106.25689697265625, 198.423828125, -182.81396484375, -365.30340576171875, -90.55331420898438, 238.0584716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000394.npy"}
{"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 135.872802734375, "std": 228.5165252685547, "min": -302.31768798828125, "p10": -74.0267395019531, "median": 90.01191711425781, "p90": 370.8287445068359, "max": 1176.8427734375, "pos_frac": 0.7578125, "sample": [173.5679931640625, 416.0028076171875, -47.42986297607422, -33.321876525878906, 118.75347900390625, -32.369659423828125, 0.54473876953125, 349.2340087890625, 1176.8427734375, 66.55697631835938, 47.80730438232422, 160.9490966796875, 88.03009033203125, 60.68040466308594, -2.509765625, 303.8121032714844, 60.82734680175781, -65.73150634765625, 35.71685791015625, 132.328125, 45.301177978515625, 420.12860107421875, 111.0873031616211, 46.44212341308594, 292.7606506347656, 221.46832275390625, 116.3624267578125, -12.916187286376953, 234.813720703125, 705.431396484375, 96.36663055419922, 263.2345275878906, -194.77386474609375, -22.177488327026367, 26.879013061523438, 207.11972045898438, 127.96255493164062, 142.7115478515625, 2.7794570922851562, 86.09449768066406, 180.0452423095703, 2.5686264038085938, -199.2901611328125, 133.72372436523438, 240.88327026367188, 40.22126770019531, -34.36445617675781, -62.02958297729492, 173.5330047607422, 276.47174072265625, 248.29571533203125, -228.41653442382812, 33.531707763671875, 277.02069091796875, 470.74365234375, 27.9700927734375, 200.13592529296875, 252.20310974121094, 1.51629638671875, -59.960906982421875, -92.46591186523438, -120.57014465332031, 312.91522216796875, 45.02099609375, 351.5143737792969, -24.137939453125, -162.9138641357422, -162.87716674804688, -14.638031005859375, 19.761871337890625, -12.5758056640625, 48.89353942871094, -5.790397644042969, -199.77813720703125, 55.430419921875, 175.38040161132812, 291.92791748046875, 145.96458435058594, 7.76617431640625, 206.4447021484375, 72.636474609375, 375.0196228027344, 833.00732421875, 277.3869934082031, 252.60601806640625, 112.32232666015625, 31.77410888671875, -256.5985107421875, 113.4429931640625, 281.482666015625, 57.030216217041016, 125.98353576660156, -96.45401000976562, -30.91463279724121, -302.31768798828125, 12.653518676757812, 359.02001953125, 369.03265380859375, 284.31329345703125, 423.7033386230469, -24.828765869140625, 279.94873046875, 87.13116455078125, 122.43733215332031, 208.07244873046875, 67.60966491699219, 531.58642578125, 68.97219848632812, 101.88006591796875, -52.616912841796875, 45.321990966796875, 260.7023010253906, 721.4876098632812, -262.80419921875, -111.265625, -66.12423706054688, 89.69049072265625, 90.33334350585938, 279.97027587890625, 732.0009155273438, 923.322509765625, 476.654541015625, 173.36895751953125, 124.4189453125, 51.635223388671875, 197.25167846679688, 70.31221008300781, 142.6759033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000395.npy"}
{"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 103.7474594116211, "std": 198.5272674560547, "min": -374.8706359863281, "p10": -129.7048568725586, "median": 92.30332946777344, "p90": 362.7626251220703, "max": 760.406494140625, "pos_frac": 0.671875, "sample": [166.82049560546875, 329.1526794433594, -26.45916748046875, 760.406494140625, 297.55584716796875, -35.00274658203125, -20.840316772460938, 100.92798614501953, -374.8706359863281, 294.4066467285156, 103.16963195800781, -285.83868408203125, 24.300567626953125, 241.13101196289062, -33.95936584472656, -5.866790771484375, 366.3313903808594, -105.14285278320312, 6.920127868652344, 434.03704833984375, 75.21735382080078, 153.65170288085938, -72.342529296875, 605.0639038085938, -193.65122985839844, 26.434722900390625, 292.31890869140625, 188.50814819335938, 91.89718627929688, 76.8600845336914, 92.70947265625, 108.24932098388672, 241.67141723632812, 277.6103210449219, -100.22024536132812, -23.383148193359375, 110.27409362792969, -51.518341064453125, -168.2770233154297, 221.47869873046875, 139.04031372070312, 192.99253845214844, 344.30145263671875, 236.1187286376953, 717.3568115234375, 44.25115966796875, 17.327617645263672, 63.387115478515625, -9.25537109375, 239.73280334472656, 182.79302978515625, 468.4457092285156, 290.3268127441406, 83.34904479980469, 108.17940521240234, 91.80264282226562, 153.87550354003906, -105.31489562988281, 78.2349853515625, -36.58939743041992, 87.88441467285156, -45.396820068359375, -108.14656066894531, -17.522308349609375, 57.447845458984375, -248.6118927001953, 172.3577880859375, -58.7554931640625, 298.7478942871094, -163.84519958496094, 167.58132934570312, -33.061920166015625, 22.35089874267578, -237.0308837890625, 191.94717407226562, 180.88287353515625, 194.06500244140625, 157.89303588867188, -53.40423583984375, -197.13180541992188, 98.567138671875, 122.89742279052734, -28.2730712890625, 105.8223876953125, 135.591064453125, 164.43795776367188, 77.71536254882812, 162.4219970703125, -15.404104232788086, 321.3264465332031, -10.11846923828125, 495.31103515625, -207.44119262695312, -51.42091369628906, 43.02236557006836, 39.43220520019531, 317.2486572265625, 136.81027221679688, -30.7891845703125, 465.1429138183594, 400.03900146484375, 378.3931579589844, 68.268798828125, -255.98095703125, -134.00225830078125, 372.1879577636719, -64.98567962646484, 361.233154296875, 163.4283447265625, 91.62225341796875, 0.0, 115.16336059570312, 180.13479614257812, 551.25634765625, 207.93991088867188, 108.10202026367188, 0.0, 131.66616821289062, 143.9761962890625, -235.63232421875, 82.32681274414062, -127.86311340332031, 48.35723876953125, 379.52154541015625, 215.57070922851562, -190.31692504882812, -107.9537124633789, 196.58245849609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000396.npy"}
{"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 93.48770141601562, "std": 219.29412841796875, "min": -381.94757080078125, "p10": -143.1965148925781, "median": 68.1259765625, "p90": 336.27125244140626, "max": 943.7335205078125, "pos_frac": 0.671875, "sample": [55.129486083984375, -51.9842529296875, -92.70265197753906, 44.39948272705078, 247.608154296875, -53.23655700683594, -83.35964965820312, 547.2080078125, -58.637229919433594, 74.8770751953125, 5.677642822265625, 88.52642822265625, -111.75003051757812, 103.46881103515625, 374.11004638671875, 131.66993713378906, 26.132537841796875, 209.0579833984375, 265.08453369140625, 265.64886474609375, -320.9331359863281, 139.47544860839844, -7.4937744140625, 60.13116455078125, 69.78057861328125, 137.28271484375, 0.0, 18.54107666015625, 463.1939392089844, 117.21804809570312, 0.0, -258.6103515625, 334.17254638671875, -137.44436645507812, 31.83294677734375, 9.927154541015625, 122.14031982421875, 337.2681884765625, -118.08700561523438, 79.74234008789062, 311.7933349609375, 231.20135498046875, 334.01739501953125, 324.6115417480469, -23.64141845703125, -89.55471801757812, 197.05029296875, 479.06365966796875, 254.28125, -154.04360961914062, 943.7335205078125, -81.2821044921875, 200.0927734375, -249.8026123046875, -261.9158935546875, -302.2898864746094, 65.49905395507812, -244.3815155029297, 70.14450073242188, -242.218017578125, -173.03179931640625, 335.843994140625, 634.6033935546875, 113.02268981933594, -60.569549560546875, 54.98176193237305, 280.26531982421875, 59.16157531738281, -138.54776000976562, -6.2989044189453125, 370.07220458984375, 331.39788818359375, 386.2539978027344, 194.99127197265625, 133.89974975585938, 238.88323974609375, 277.97186279296875, 66.47137451171875, -212.56259155273438, 27.744140625, 216.88758850097656, 203.9891357421875, 821.073486328125, 517.5964965820312, 44.186187744140625, -17.920211791992188, 23.680160522460938, -26.797607421875, -14.532928466796875, 161.64923095703125, 100.05426025390625, -96.19461059570312, 130.8120880126953, 74.77093505859375, 243.07421875, 99.668701171875, 185.2915496826172, 43.227203369140625, 265.72076416015625, 97.90116882324219, 36.9591064453125, -118.412109375, 118.93267822265625, 20.280200958251953, 526.6270751953125, 260.2098388671875, 7.935707092285156, -72.10977172851562, -19.688316345214844, -129.8232421875, -381.94757080078125, 188.19281005859375, 240.24618530273438, 30.12969970703125, 88.58236694335938, 92.035888671875, -218.9169921875, 36.77976989746094, 134.703857421875, 57.84185791015625, -91.59661865234375, -332.6793212890625, 297.59442138671875, 129.09188842773438, -120.67117309570312, -54.0506591796875, -24.164459228515625, 448.22589111328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000397.npy"}
{"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 79.31932067871094, "std": 258.77362060546875, "min": -1044.945556640625, "p10": -209.98691406249998, "median": 87.65488815307617, "p90": 377.9693328857422, "max": 677.6749267578125, "pos_frac": 0.65625, "sample": [149.95556640625, 188.12689208984375, 183.77073669433594, -24.329315185546875, 515.5834350585938, -140.51177978515625, 677.6749267578125, 5.08038330078125, 149.62974548339844, 654.566162109375, -283.4563293457031, -116.89605712890625, -524.0726318359375, 163.3578338623047, 38.853843688964844, 198.45237731933594, 311.5072326660156, -16.853103637695312, 471.34124755859375, 368.690673828125, 390.7250061035156, -119.20924377441406, 160.49331665039062, 263.1761474609375, 138.39630126953125, 176.95004272460938, 215.507080078125, 575.06884765625, 81.9501953125, 261.510986328125, -24.973052978515625, 101.09239959716797, 339.60296630859375, -277.82696533203125, -259.5527648925781, 111.50724792480469, 140.69830322265625, 100.10899353027344, 153.27813720703125, 63.80820083618164, 9.374984741210938, -16.620849609375, 12.409317016601562, 311.0245056152344, -20.026344299316406, -298.70599365234375, 93.18339538574219, 55.724220275878906, 33.468719482421875, 95.1724853515625, -102.66607666015625, 46.684356689453125, 5.307106018066406, 56.650054931640625, 203.0670166015625, 181.15130615234375, -11.128110885620117, -203.8394775390625, -67.66096496582031, 0.8539180755615234, 218.7034454345703, 78.288330078125, 422.91766357421875, -403.5054626464844, 221.88336181640625, -40.17732238769531, -88.70915222167969, -149.94189453125, -90.48377990722656, 383.9532775878906, 306.5362548828125, 333.998046875, 157.4167022705078, 223.93685913085938, -296.5748596191406, -224.3309326171875, -4.135150909423828, 197.6767578125, -716.2200927734375, -26.307235717773438, 596.356689453125, 329.05682373046875, 16.964752197265625, 154.93893432617188, 71.87974548339844, 46.164642333984375, -150.9482421875, 145.7649383544922, -124.8441162109375, -506.19732666015625, 254.60012817382812, 278.46051025390625, -160.58984375, -78.38739013671875, -177.607421875, -264.28814697265625, 330.3607177734375, -82.32757568359375, -127.54879760742188, 237.1353302001953, -15.485427856445312, -170.92965698242188, 248.47802734375, 201.4141082763672, 118.751953125, 82.12638092041016, 264.477783203125, 446.88134765625, 386.004150390625, -172.21774291992188, -1044.945556640625, 634.6759033203125, 247.5062255859375, -133.81033325195312, 375.40478515625, 139.6602783203125, -95.90585327148438, -238.07269287109375, 54.674591064453125, 223.04888916015625, 302.502685546875, 17.35223388671875, 9.775718688964844, 419.38134765625, 258.6441955566406, 237.05491638183594, 147.40060424804688, -31.052947998046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000398.npy"}
{"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 106.72332763671875, "std": 208.6949462890625, "min": -375.506103515625, "p10": -151.27959442138672, "median": 85.11331176757812, "p90": 363.2023071289062, "max": 781.1573486328125, "pos_frac": 0.671875, "sample": [40.76762390136719, 551.5342407226562, 229.62466430664062, 498.9649658203125, 77.29620361328125, -54.81154251098633, 62.01252746582031, 118.74191284179688, 165.09100341796875, 442.8484191894531, 345.1209411621094, 13.832473754882812, -282.0593566894531, 424.375732421875, 48.49647521972656, 657.3299560546875, -86.86892700195312, -25.828460693359375, 54.71226501464844, 85.2562255859375, -185.59133911132812, -39.72454833984375, -32.47222900390625, 253.9835205078125, -5.69964599609375, -66.51251220703125, 51.29754638671875, -222.51309204101562, -375.506103515625, 312.75738525390625, 197.26364135742188, 289.69317626953125, -19.415687561035156, 170.03350830078125, 124.74116516113281, 24.996627807617188, 202.47186279296875, -166.07359313964844, -59.25236511230469, 330.78167724609375, 150.65057373046875, 351.8601989746094, 358.96868896484375, 409.69049072265625, -62.7686767578125, 100.31568145751953, 280.9963073730469, -149.8353271484375, 218.53829956054688, 84.97039794921875, -44.056884765625, 5.9556121826171875, -7.6406097412109375, 107.07940673828125, 536.5126342773438, -49.971527099609375, 98.05280303955078, -0.2274169921875, -83.4291000366211, -113.16081237792969, -86.49676513671875, 379.5062255859375, -81.83143615722656, 63.8590087890625, 176.3648681640625, -12.234127044677734, 450.65972900390625, 105.95626831054688, 142.28746032714844, 242.6595458984375, 781.1573486328125, 181.7310791015625, -78.94630432128906, -131.71429443359375, -252.00494384765625, -178.3866424560547, 2.2452392578125, 159.5997314453125, 150.89569091796875, 108.28144073486328, 123.30447387695312, 260.69793701171875, 224.72418212890625, 195.78387451171875, 301.63427734375, 708.1748046875, 32.89068603515625, -154.64955139160156, 217.28216552734375, 25.50225830078125, 196.44769287109375, 251.84841918945312, 6.419281005859375, 147.193359375, 180.8120880126953, 43.40162658691406, -231.18414306640625, -1.505584716796875, 373.08074951171875, 248.19140625, 243.97528076171875, 198.95843505859375, 226.03213500976562, -159.93914794921875, -33.9136962890625, 296.50848388671875, -186.31939697265625, 309.1890869140625, 139.461181640625, 16.939651489257812, 562.037841796875, -180.24307250976562, -214.10983276367188, 249.23194885253906, 22.11322593688965, 39.53752136230469, -142.3134765625, 312.0686340332031, 30.318801879882812, 152.92868041992188, -26.18353271484375, -17.71392822265625, 90.4737548828125, -66.1252670288086, -124.9573974609375, 68.53387451171875, 36.21250915527344, 200.04733276367188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000399.npy"}
{"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 102.36151123046875, "std": 210.2042999267578, "min": -472.427490234375, "p10": -139.85802764892577, "median": 95.65238952636719, "p90": 372.90104370117183, "max": 675.8822021484375, "pos_frac": 0.6875, "sample": [46.084686279296875, -17.975112915039062, -96.36221313476562, 96.47137451171875, 121.1488037109375, 422.5426025390625, 116.54731750488281, -26.651229858398438, 56.51490783691406, 207.90020751953125, 70.82505798339844, 245.20294189453125, -376.5380554199219, -137.21353149414062, 366.5906066894531, 21.98162841796875, 136.11322021484375, 127.40621948242188, 25.414993286132812, 238.59652709960938, 195.31732177734375, -138.81712341308594, -31.984098434448242, 86.02725219726562, 146.79638671875, 647.5081787109375, -315.17779541015625, 207.89227294921875, -82.410400390625, 222.7083740234375, 136.5558624267578, 162.03797912597656, -117.84933471679688, 74.11581420898438, 142.50132751464844, 163.7337646484375, 344.74212646484375, 291.91021728515625, -45.20623779296875, 138.94818115234375, -314.432373046875, 400.8937072753906, 267.8418884277344, 476.41790771484375, 39.576507568359375, 675.8822021484375, -63.90924072265625, 86.67080688476562, 152.18646240234375, 227.47775268554688, 73.85614013671875, 236.82705688476562, 354.860595703125, -142.28680419921875, 18.41259765625, 22.632080078125, 114.29286193847656, 164.14208984375, 385.2706298828125, -107.21150207519531, -46.448081970214844, 466.118896484375, 255.31674194335938, 312.5051574707031, 92.84136962890625, -128.86370849609375, 213.8961181640625, 131.50393676757812, 610.4741821289062, 207.35321044921875, 202.04632568359375, 65.25244903564453, -472.427490234375, -55.90455627441406, -94.01679992675781, -291.37591552734375, 401.18792724609375, 613.645263671875, 132.1599884033203, -35.216064453125, -19.723190307617188, 402.91864013671875, 94.83340454101562, 27.19073486328125, -32.819480895996094, 245.87939453125, 114.47632598876953, -305.4529724121094, 310.3209228515625, 59.696144104003906, 61.07156753540039, 21.66168212890625, -8.912277221679688, -185.0834197998047, 17.242523193359375, 156.19277954101562, 30.6510009765625, 300.8199157714844, -14.425615310668945, -216.98248291015625, 132.89404296875, 234.3427276611328, 114.38748168945312, -7.4429931640625, 40.94293212890625, 94.1185302734375, 210.1066436767578, -22.702823638916016, -158.32620239257812, 283.23577880859375, 191.063232421875, -53.86079406738281, 267.42193603515625, -120.2987060546875, 178.531982421875, 158.72805786132812, 262.78277587890625, -294.4916076660156, 471.01519775390625, 367.59979248046875, 399.5749206542969, 92.32940673828125, 0.0, -220.9119873046875, -172.61328125, -35.27093505859375, -134.09127807617188, 240.25155639648438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000400.npy"}
{"epoch": 0.837696335078534, "step": 401, "batch_size": 128, "mean": 109.63511657714844, "std": 209.6938934326172, "min": -525.814453125, "p10": -136.96511840820312, "median": 104.76068115234375, "p90": 373.93363647460933, "max": 811.6365966796875, "pos_frac": 0.7109375, "sample": [-57.844871520996094, -108.63754272460938, 531.615966796875, -234.36883544921875, 53.8338623046875, 403.2366638183594, -1.50799560546875, 171.20494079589844, -160.68447875976562, -106.00971984863281, 294.5096435546875, 190.34506225585938, 169.62741088867188, 239.8896484375, 36.995025634765625, -15.042266845703125, 8.4378662109375, 243.3059844970703, 1.426361083984375, 39.187744140625, -9.188743591308594, 169.0143280029297, -42.611610412597656, -165.20684814453125, -373.401123046875, 487.8424377441406, 330.1593017578125, 24.651296615600586, 175.5936279296875, 186.117431640625, 175.0791015625, 443.2261962890625, 111.07998657226562, 63.563720703125, -1.17095947265625, 48.8037109375, 260.616943359375, 273.3248596191406, 193.21115112304688, 118.18484497070312, 515.6839599609375, 46.38458251953125, 225.55584716796875, 258.9981689453125, 182.70578002929688, 78.38796997070312, -84.47380828857422, 73.58326721191406, 215.70263671875, 382.99615478515625, 71.77679443359375, 195.02056884765625, 132.95474243164062, 73.26032257080078, -407.1014709472656, 225.93194580078125, 68.20245361328125, 137.2911376953125, 9.942733764648438, 45.22064971923828, -165.54592895507812, 225.11062622070312, 432.62066650390625, 497.0414733886719, 44.078521728515625, 241.4647979736328, -134.57980346679688, -57.70438766479492, 289.5995178222656, -107.47802734375, -43.79841232299805, -187.04080200195312, 322.10382080078125, 34.2164306640625, 195.4580078125, -81.58551025390625, 434.35906982421875, 811.6365966796875, 150.4697265625, 1.3840255737304688, 101.464111328125, -279.2680358886719, 197.64151000976562, 141.87841796875, 118.83834838867188, 192.8489990234375, 51.045166015625, 40.928314208984375, 108.0572509765625, 159.12310791015625, 179.72775268554688, 204.9750213623047, -27.1380615234375, -179.38763427734375, 15.002708435058594, -145.64805603027344, -142.53085327148438, 382.01239013671875, 592.014892578125, 112.950927734375, 21.1529541015625, -27.27703857421875, 289.5504150390625, 210.28512573242188, 370.4713134765625, 193.4097900390625, -85.35467529296875, 124.28176879882812, -75.94552612304688, 353.82183837890625, -47.38664245605469, 36.49041748046875, -70.25387573242188, -3.279743194580078, 77.35356903076172, 47.65354919433594, -42.42572021484375, 115.14688873291016, 226.7615509033203, -146.87254333496094, 305.4490966796875, 262.65582275390625, 272.85467529296875, -21.08300018310547, 747.702880859375, -525.814453125, -96.89289093017578, 178.0579833984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000401.npy"}
{"epoch": 0.8397905759162304, "step": 402, "batch_size": 128, "mean": 105.5995864868164, "std": 225.99078369140625, "min": -356.9195556640625, "p10": -114.2592788696289, "median": 81.67096710205078, "p90": 332.67609863281245, "max": 1004.23095703125, "pos_frac": 0.6875, "sample": [-57.740177154541016, -49.462703704833984, -83.86000061035156, 154.109375, 114.49456787109375, -302.91705322265625, -220.1346435546875, -298.1233215332031, 74.81793212890625, -356.9195556640625, 2.9053955078125, 839.13818359375, 118.33352661132812, -72.83380126953125, 14.472282409667969, 453.51776123046875, -57.45744323730469, 10.7703857421875, -111.5626220703125, -25.52899169921875, 202.38284301757812, 37.41859436035156, 84.78752136230469, 207.66836547851562, -275.22930908203125, 9.829591751098633, 64.29046630859375, 384.4804382324219, -251.44268798828125, 78.55441284179688, 136.3292236328125, -115.260498046875, -1.6383819580078125, 454.89825439453125, -71.1458740234375, 0.0, 963.16357421875, 300.3228759765625, -50.377166748046875, 232.76669311523438, 174.12408447265625, 125.06219482421875, 157.59593200683594, -13.2388916015625, 50.96131896972656, 241.4434814453125, -104.58116149902344, -192.4290008544922, 160.1407470703125, 141.92269897460938, 67.6649169921875, -164.1549072265625, -315.4373779296875, 202.88421630859375, 330.3890380859375, 138.2955780029297, 557.9586791992188, 200.78201293945312, 24.36468505859375, 338.0125732421875, -3.859844207763672, 1004.23095703125, 248.53677368164062, 114.87301635742188, 203.37586975097656, -168.4602508544922, 43.079681396484375, -21.10137939453125, 182.0589599609375, -225.87582397460938, 264.6093444824219, 359.5570068359375, 94.15095520019531, -65.14544677734375, 178.57281494140625, 104.44338989257812, 627.422119140625, -6.4074249267578125, 208.99005126953125, 125.6168212890625, 339.8519287109375, 92.07168579101562, 118.33615112304688, 250.04922485351562, -64.65292358398438, 92.70632934570312, 281.96856689453125, 160.96563720703125, -113.83018493652344, 294.98291015625, 122.54009246826172, -0.969696044921875, -11.228790283203125, 15.020988464355469, -14.23992919921875, 204.689453125, 87.05574035644531, 301.60211181640625, 21.3114013671875, 111.86895751953125, 173.3037109375, 198.9293975830078, -18.423614501953125, 168.04220581054688, 781.5488891601562, -138.42611694335938, 95.51510620117188, -55.18292236328125, 35.55499267578125, 30.36419677734375, -11.0030517578125, 167.63589477539062, 36.0970458984375, 562.4376220703125, 16.21044921875, 30.79058837890625, -59.712646484375, 295.52813720703125, 75.90974426269531, 25.138031005859375, 147.32843017578125, 174.35235595703125, -20.056171417236328, 191.11248779296875, 3.9161376953125, 74.6956787109375, 297.4216613769531, 19.372461318969727], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000402.npy"}
{"epoch": 0.8418848167539267, "step": 403, "batch_size": 128, "mean": 151.53472900390625, "std": 203.861572265625, "min": -290.87200927734375, "p10": -56.48677291870115, "median": 111.18880844116211, "p90": 403.1025451660156, "max": 1010.9657592773438, "pos_frac": 0.8125, "sample": [60.40068054199219, 190.38787841796875, 67.23251342773438, 672.871337890625, 32.751007080078125, 193.5286865234375, 40.2955322265625, -4.942363739013672, 102.34091186523438, -125.84780883789062, 74.59671020507812, 411.3891296386719, 154.83837890625, -28.5140380859375, 100.88142395019531, 59.05731201171875, 69.11854553222656, 229.81027221679688, 84.7440185546875, 25.937889099121094, 328.72979736328125, -40.927276611328125, -262.19775390625, 245.23895263671875, 329.0936279296875, 263.74505615234375, 276.77801513671875, 115.552734375, 107.35687255859375, 301.2259521484375, 272.02392578125, -50.38446044921875, -22.274269104003906, 127.45272827148438, 5.613718032836914, 235.14828491210938, 42.21343994140625, 278.70220947265625, 0.0, -175.21649169921875, 9.193122863769531, 327.12689208984375, 332.69207763671875, 131.8133544921875, 404.9731140136719, 332.3795166015625, 265.31842041015625, 63.441986083984375, 467.9123840332031, 233.81231689453125, 81.1142578125, 77.51826477050781, -25.722412109375, 34.73358154296875, 237.13946533203125, 171.55718994140625, 44.957763671875, -149.48651123046875, 226.357666015625, -116.41705322265625, -17.290496826171875, 146.57455444335938, -157.4065704345703, 1010.9657592773438, 141.59701538085938, 78.23993682861328, 108.2220458984375, -70.72550201416016, 72.88004302978516, 18.82891845703125, 43.21337890625, 296.67156982421875, 83.19987487792969, 663.9332275390625, 0.23394775390625, 111.84535217285156, 452.3603515625, 386.0771179199219, 216.68746948242188, 157.462646484375, 110.53226470947266, 156.5347900390625, 382.1302795410156, 321.2174987792969, 478.70538330078125, -81.52706909179688, 51.14695739746094, 441.60894775390625, 128.7003173828125, 106.528564453125, 673.3292846679688, 74.25674438476562, 787.4389038085938, 315.44891357421875, 13.490234375, 190.50836181640625, -202.12298583984375, 302.11358642578125, 245.17462158203125, 156.67872619628906, 396.8099365234375, 55.83678436279297, 179.8639678955078, 2.6104736328125, -290.87200927734375, 402.3008728027344, -102.5018310546875, 42.76451110839844, -4.631906509399414, 302.3739929199219, -81.693115234375, 82.70044708251953, 56.49139404296875, 118.44469451904297, -105.61175537109375, 458.0188903808594, 56.81315612792969, -1.0290374755859375, 23.917736053466797, 30.026187896728516, -45.517578125, 147.73133850097656, 203.6343994140625, 135.64443969726562, 124.1217041015625, 131.60751342773438, 510.77349853515625, 231.184814453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000403.npy"}
{"epoch": 0.8439790575916231, "step": 404, "batch_size": 128, "mean": 114.07733154296875, "std": 231.84805297851562, "min": -609.71435546875, "p10": -112.88077697753906, "median": 99.6270523071289, "p90": 368.53032531738273, "max": 959.8197021484375, "pos_frac": 0.7265625, "sample": [384.74237060546875, 198.08453369140625, 88.89984130859375, 334.0655822753906, -273.1201171875, 388.8370056152344, 221.41720581054688, 90.94296264648438, -162.85882568359375, 0.0, -107.71405029296875, 339.97283935546875, 34.706321716308594, 12.636016845703125, -609.71435546875, 293.9332275390625, -10.4783935546875, 355.3265380859375, 29.13233184814453, -42.346923828125, 568.388427734375, 17.791595458984375, 19.325592041015625, 38.93646240234375, 61.55393981933594, 242.05307006835938, -499.2936706542969, 482.08245849609375, -59.80780029296875, 173.1360626220703, -3.615142822265625, -95.50128173828125, 136.84298706054688, 87.6083984375, -76.78534698486328, 37.62853240966797, -91.35481262207031, 145.4282989501953, 723.912841796875, -9.3709716796875, 194.5098876953125, 9.468002319335938, -169.98431396484375, -166.164794921875, 115.48193359375, 9.013519287109375, 265.2047119140625, 118.279052734375, 18.297840118408203, 113.23974609375, -19.8603515625, 16.808013916015625, 361.5823059082031, -124.946533203125, 116.5047607421875, 317.60394287109375, -113.81219482421875, 61.99610900878906, 108.31114196777344, 294.6014404296875, 200.35943603515625, 128.94744873046875, -98.21049499511719, 136.3272705078125, -80.43991088867188, 82.11901092529297, 443.478515625, 40.7249755859375, -89.44529724121094, 235.5286407470703, 179.49951171875, 29.59609031677246, 78.17947387695312, 42.974281311035156, 164.65716552734375, -70.18865966796875, -238.50979614257812, -238.8780517578125, 242.39743041992188, 89.26881408691406, 111.58509826660156, 27.201156616210938, 113.27322387695312, 123.1817626953125, 115.25552368164062, -112.48159790039062, -19.374984741210938, -97.95306396484375, 218.4554443359375, -401.24346923828125, 30.999839782714844, 429.6383972167969, 239.70465087890625, 571.3211669921875, -99.97056579589844, -39.966064453125, 46.24781799316406, 122.24729919433594, 115.67247009277344, 215.09039306640625, -102.05484008789062, 147.12261962890625, 325.335205078125, 256.9144287109375, 358.5382385253906, 73.92340087890625, 15.303047180175781, 127.00115966796875, -319.45928955078125, 815.8892822265625, 289.08599853515625, 335.2888488769531, 327.95166015625, 561.5252075195312, 135.501953125, 90.76535034179688, 154.80010986328125, -59.28730010986328, 623.752685546875, 325.2997741699219, 110.58013916015625, -120.98890686035156, 64.29988098144531, 213.529052734375, 261.44122314453125, 385.63330078125, 265.55865478515625, 959.8197021484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000404.npy"}
{"epoch": 0.8460732984293193, "step": 405, "batch_size": 128, "mean": 125.53387451171875, "std": 210.91763305664062, "min": -407.3255615234375, "p10": -86.68281173706053, "median": 95.5351333618164, "p90": 426.7509368896484, "max": 788.0036010742188, "pos_frac": 0.7265625, "sample": [54.30108642578125, -71.50982666015625, 24.483245849609375, 0.0, 23.49273681640625, -69.6627197265625, 258.474853515625, 123.5010986328125, 225.755126953125, 101.49520874023438, 27.108551025390625, -57.66259765625, 88.22986602783203, 154.62921142578125, 177.6994171142578, -139.39224243164062, -190.56698608398438, 89.41424560546875, 249.2530517578125, 427.2828369140625, 16.255645751953125, 182.18936157226562, 27.312875747680664, -62.328269958496094, 147.5737762451172, -15.930564880371094, -176.05221557617188, 123.1978759765625, 556.1168823242188, 141.97320556640625, 173.95962524414062, -20.514137268066406, 151.84011840820312, 17.503387451171875, 556.4595947265625, 213.03704833984375, 192.34323120117188, 30.69940185546875, 7.725046157836914, 60.14282989501953, -27.037296295166016, 599.1612548828125, 255.47885131835938, -7.989397048950195, 48.207672119140625, 788.0036010742188, 506.0638122558594, 6.653631210327148, 230.37335205078125, -270.36004638671875, 355.88787841796875, 178.73248291015625, 525.846923828125, 117.06346130371094, 91.00173950195312, -56.82586669921875, 84.77780151367188, 663.6883544921875, 175.4451446533203, -118.871337890625, -93.6127700805664, -13.295097351074219, 177.43545532226562, -255.30966186523438, 195.213623046875, -21.966598510742188, 493.8339538574219, -83.71282958984375, 20.34490966796875, 214.13232421875, 64.74281311035156, 303.099853515625, 185.41116333007812, 267.2580261230469, 292.08099365234375, 539.3974609375, -407.3255615234375, 181.68572998046875, 2.4589767456054688, 183.2655792236328, 358.26959228515625, 12.903356552124023, -107.97412109375, -243.77780151367188, 58.2025146484375, 44.23878479003906, 27.965438842773438, 412.08721923828125, 8.666961669921875, 211.42691040039062, 370.12091064453125, 27.52587127685547, -186.38699340820312, 120.83538818359375, 426.5229797363281, 344.7091369628906, -12.295761108398438, -17.6019287109375, 246.94146728515625, 173.8142547607422, -343.6277160644531, 242.58840942382812, 239.83853149414062, 554.3919677734375, -16.56780242919922, 89.67578887939453, 132.32540893554688, 269.6460266113281, 287.75103759765625, 5.386474609375, 125.55487060546875, 145.2798309326172, 224.83090209960938, -23.741422653198242, -65.70208740234375, -19.385406494140625, 17.512954711914062, -1.1945858001708984, -50.480865478515625, 52.27556228637695, 200.83624267578125, 431.62542724609375, 375.611083984375, -81.939208984375, 100.06852722167969, 573.7545166015625, -181.68182373046875, 297.2386779785156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000405.npy"}
{"epoch": 0.8481675392670157, "step": 406, "batch_size": 128, "mean": 81.9852294921875, "std": 178.8982391357422, "min": -494.0307312011719, "p10": -123.30378646850586, "median": 64.31588745117188, "p90": 326.32835388183594, "max": 585.4082641601562, "pos_frac": 0.734375, "sample": [227.53146362304688, 12.537612915039062, -123.69851684570312, 245.31689453125, -123.19329833984375, 215.4058837890625, 128.399658203125, 234.16355895996094, 166.86886596679688, 338.4151306152344, 9.096656799316406, 7.876178741455078, 15.546142578125, 384.98944091796875, 178.41650390625, -413.7423095703125, 450.30364990234375, -22.75342559814453, 153.0585174560547, 163.74551391601562, -62.135589599609375, -494.0307312011719, 53.37615966796875, 170.41397094726562, 53.066253662109375, 113.1151123046875, 32.719398498535156, 202.34814453125, 116.36105346679688, 167.3870849609375, 338.20245361328125, 361.0048828125, -211.8607177734375, 1.145721435546875, 388.1722412109375, -104.05453491210938, 158.7033233642578, 4.5376434326171875, 28.215194702148438, 3.685272216796875, -96.5313720703125, -116.59249877929688, -117.69914245605469, -140.2681884765625, 118.61415100097656, 125.61117553710938, 44.848114013671875, 223.72486877441406, 284.8905334472656, 177.91729736328125, -20.323806762695312, 206.08035278320312, 329.1070556640625, -35.81452941894531, -365.0499267578125, 36.29144287109375, 160.75970458984375, 379.4704284667969, -123.56159210205078, 8.8082275390625, 22.277008056640625, 417.3946533203125, -178.44049072265625, 166.4102783203125, 255.96656799316406, 141.37109375, 76.13192749023438, -94.11746215820312, 75.31451416015625, 297.81927490234375, 325.1374816894531, 101.60375213623047, 39.444610595703125, 20.3956298828125, 221.56478881835938, 103.11180114746094, 34.982933044433594, -18.25970458984375, 20.03357696533203, 399.8666076660156, 217.78787231445312, 13.992603302001953, -227.69677734375, 35.77466583251953, 331.1025390625, 346.5018005371094, 146.4967041015625, -115.68217468261719, 196.97314453125, 26.352020263671875, 127.94143676757812, 97.49296569824219, 156.83602905273438, 273.0015869140625, -23.9381103515625, 314.56646728515625, 244.73046875, -172.24916076660156, 137.223876953125, 52.4761962890625, -11.13592529296875, 7.971527099609375, 585.4082641601562, -26.9158935546875, 4.780517578125, 133.9171142578125, 20.858367919921875, 259.63092041015625, -100.46150207519531, 109.74386596679688, 264.26055908203125, -35.561370849609375, -74.62420654296875, 22.21600341796875, 255.49459838867188, -3.2417144775390625, 30.58908462524414, -102.06884765625, -112.72412109375, 79.11456298828125, 22.5594482421875, 5.583099365234375, -285.2676086425781, -168.44503784179688, 308.0678405761719, -126.06820678710938, 98.54043579101562, 75.255615234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000406.npy"}
{"epoch": 0.8502617801047121, "step": 407, "batch_size": 128, "mean": 121.75782012939453, "std": 226.59677124023438, "min": -546.9071044921875, "p10": -94.32433624267577, "median": 84.90425109863281, "p90": 414.6848144531249, "max": 909.7606811523438, "pos_frac": 0.671875, "sample": [8.58188247680664, 189.52566528320312, -30.837646484375, -118.07444763183594, 334.0342102050781, 16.038772583007812, 217.2247314453125, 453.47735595703125, -79.521484375, 109.65922546386719, -30.118858337402344, 29.041534423828125, -83.51432037353516, -3.9188232421875, -316.39642333984375, 265.22625732421875, -25.308837890625, -78.65109252929688, 61.70416259765625, -9.134124755859375, 222.70147705078125, -19.73883056640625, -38.678924560546875, 144.04603576660156, 323.0268859863281, 24.310623168945312, -70.04243469238281, 116.83275604248047, -269.65460205078125, -60.629180908203125, -98.57392883300781, 70.68212890625, -81.36196899414062, 6.073974609375, 156.54998779296875, 274.85797119140625, 87.47549438476562, 312.13262939453125, -61.31810760498047, 45.3768310546875, 225.597900390625, 132.72760009765625, 86.39382934570312, -17.970252990722656, -92.50308227539062, 366.789794921875, 208.62643432617188, 113.25578308105469, 256.40966796875, 126.2320556640625, -9.486572265625, 198.39718627929688, 6.964508056640625, 641.5223999023438, -44.623779296875, 513.2310791015625, -38.8162841796875, 68.75244140625, 3.26971435546875, 179.79937744140625, 47.33515167236328, 142.08140563964844, 375.272216796875, -12.040678024291992, 161.7974853515625, 8.585906982421875, 77.64544677734375, -361.69866943359375, 243.17816162109375, 63.22430419921875, 83.4146728515625, -178.93618774414062, 440.29052734375, 191.79275512695312, 403.7109375, -30.494342803955078, 61.623809814453125, 202.81912231445312, 0.0, 179.37295532226562, 381.7261962890625, -114.08287048339844, 394.83062744140625, -166.41824340820312, 305.7467346191406, -546.9071044921875, 667.8819580078125, 161.31500244140625, -79.7489013671875, 170.19671630859375, 650.466064453125, 3.9769287109375, 38.829315185546875, 89.03974914550781, 239.2684326171875, -25.72216796875, -31.53619384765625, 9.255706787109375, 95.5614013671875, 155.19024658203125, 462.67474365234375, -213.02781677246094, 230.92593383789062, -236.48687744140625, 187.081787109375, -100.31405639648438, 330.7533264160156, 466.3479919433594, 518.4825439453125, 71.83908081054688, 396.4490966796875, 351.87640380859375, 909.7606811523438, -130.44410705566406, 531.02294921875, -68.6916275024414, 100.857666015625, -31.70158576965332, 159.92483520507812, 701.78173828125, 203.56930541992188, 568.134521484375, 142.1915283203125, 50.14435958862305, -70.347900390625, 244.11874389648438, 392.5574645996094, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000407.npy"}
{"epoch": 0.8523560209424084, "step": 408, "batch_size": 128, "mean": 120.32861328125, "std": 199.1055908203125, "min": -439.3919677734375, "p10": -109.2465003967285, "median": 123.85470581054688, "p90": 347.03063964843744, "max": 847.7227783203125, "pos_frac": 0.7265625, "sample": [-112.06392669677734, 75.4107666015625, -121.44831848144531, 116.69384765625, 154.55029296875, 158.2699432373047, -71.81094360351562, 422.0164794921875, 159.75186157226562, 225.32168579101562, -72.14393615722656, 146.11251831054688, 258.58056640625, 44.2186279296875, 226.018310546875, 228.7689208984375, 278.6237487792969, 15.11968994140625, -20.30242919921875, -97.56842803955078, -138.61395263671875, -53.7878303527832, 153.2093505859375, -138.79702758789062, 847.7227783203125, 90.6573257446289, 695.4091186523438, 131.01556396484375, 27.13470458984375, 53.0887451171875, 26.242431640625, 403.6083984375, 262.77728271484375, 85.48818969726562, -298.4198913574219, 154.26824951171875, 526.380615234375, 173.28353881835938, 247.63429260253906, 85.82235717773438, 91.46969604492188, -168.74005126953125, -327.66229248046875, 137.07220458984375, 259.1582946777344, 78.32767486572266, -44.52851867675781, -34.494224548339844, -439.3919677734375, 406.7308654785156, -3.143402099609375, 180.80682373046875, 111.92083740234375, -42.01179504394531, 177.6402587890625, 236.54354858398438, 158.15472412109375, 281.83453369140625, -108.03903198242188, 110.26080322265625, 338.955810546875, 343.3890380859375, 356.87896728515625, 141.09080505371094, 104.794921875, 217.22006225585938, -63.91461181640625, 115.37294006347656, 210.3885498046875, -34.76129150390625, 235.20513916015625, 15.664413452148438, 224.63980102539062, -9.618408203125, 59.77996826171875, 392.5550231933594, -116.91635131835938, 234.91998291015625, 64.88067626953125, 134.8900146484375, 197.47882080078125, 224.37765502929688, -46.14484405517578, 78.65887451171875, -49.21966552734375, 181.55372619628906, 402.817138671875, 190.31105041503906, 160.12655639648438, -303.949462890625, 174.37515258789062, 148.45045471191406, 355.5277099609375, 133.53900146484375, 636.3779296875, -83.33106994628906, -149.49578857421875, 245.5023193359375, 246.66864013671875, 180.19525146484375, -362.05413818359375, 286.6562805175781, 64.55514526367188, 100.53716278076172, 195.71417236328125, -52.367218017578125, 15.737335205078125, 103.20770263671875, 268.70269775390625, -68.20223999023438, -19.72186279296875, 74.45353698730469, 6.067234039306641, 0.0, 167.75637817382812, 307.82427978515625, 406.538818359375, -19.013498306274414, -27.129074096679688, -132.176513671875, 96.62902069091797, 56.4100341796875, 217.81710815429688, 238.08877563476562, 191.0615234375, 286.5008239746094, 65.18023681640625, 633.901123046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000408.npy"}
{"epoch": 0.8544502617801047, "step": 409, "batch_size": 128, "mean": 137.17941284179688, "std": 372.37542724609375, "min": -676.9859619140625, "p10": -112.63756408691405, "median": 110.55915832519531, "p90": 367.1034545898437, "max": 3649.0908203125, "pos_frac": 0.7265625, "sample": [248.42169189453125, 83.4759292602539, 213.7761993408203, -80.43630981445312, 220.19183349609375, -676.9859619140625, 212.7240447998047, 578.769287109375, -29.5841064453125, 478.90179443359375, 111.82070922851562, 116.74229431152344, 211.75343322753906, 76.87933349609375, -40.1749267578125, 377.43170166015625, 99.71711730957031, 140.54949951171875, 185.7091522216797, -104.57763671875, -117.45196533203125, 154.65052795410156, -5.928400039672852, 348.8052062988281, 165.55584716796875, -103.90457153320312, 146.19769287109375, -235.77011108398438, 270.98779296875, -493.777099609375, 650.1080322265625, 232.37545776367188, 30.2891845703125, 109.297607421875, 11.2191162109375, 127.83660888671875, 135.0543670654297, 52.681427001953125, 322.2998046875, 397.9761047363281, 75.70536804199219, 415.5518493652344, -320.8985595703125, 194.36361694335938, -8.593368530273438, 36.0030517578125, 74.70379638671875, 189.46548461914062, 89.09573364257812, 252.38885498046875, 301.6397705078125, 320.37396240234375, 362.67706298828125, -6.2905731201171875, 288.3885498046875, -12.4886474609375, -208.830810546875, 123.35064697265625, 36.441680908203125, -155.19569396972656, 305.9922790527344, 11.6356201171875, 114.1701889038086, 117.00421142578125, 302.880126953125, -110.57424926757812, 3.44354248046875, -106.51217651367188, -25.162277221679688, -292.912353515625, 132.4228515625, -1.74102783203125, -54.138641357421875, 316.6612548828125, 21.851280212402344, 391.87841796875, 3649.0908203125, 639.7378540039062, 41.602684020996094, 159.42633056640625, 132.90768432617188, -9.733444213867188, 1.033843994140625, 0.0, 1.7923355102539062, 445.31805419921875, 249.67648315429688, 317.41278076171875, 252.74960327148438, -233.61285400390625, -124.94805145263672, 251.11282348632812, 89.52557373046875, 32.20537567138672, 1.9215087890625, -2.4286117553710938, 12.811126708984375, 467.49395751953125, 181.5543670654297, 57.54547119140625, -10.785263061523438, 212.97996520996094, 133.15914916992188, 560.2220458984375, 176.54953002929688, -24.133010864257812, -58.43495178222656, -157.08218383789062, 232.27862548828125, 98.73929595947266, 221.72935485839844, -11.019790649414062, -131.8914794921875, 2.3912887573242188, 216.55813598632812, 123.34234619140625, 61.523033142089844, 310.2669677734375, 164.20208740234375, -7.3574066162109375, 429.597900390625, 33.98402404785156, 8.900650024414062, 165.32896423339844, -166.15032958984375, 61.95500183105469, 266.13629150390625, 233.42092895507812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000409.npy"}
{"epoch": 0.856544502617801, "step": 410, "batch_size": 128, "mean": 106.36808013916016, "std": 212.80514526367188, "min": -340.8705749511719, "p10": -148.88847045898436, "median": 77.80393981933594, "p90": 389.5383728027343, "max": 728.448486328125, "pos_frac": 0.703125, "sample": [206.200439453125, 103.05863189697266, 82.25225830078125, -118.71734619140625, 72.41150665283203, 414.88726806640625, -18.884662628173828, 330.1441345214844, -121.73486328125, 244.1443634033203, 432.8065490722656, 42.97491455078125, 174.58114624023438, -213.52752685546875, 224.8896484375, -340.8705749511719, -12.7030029296875, 167.5690460205078, 492.3672180175781, 191.80416870117188, -42.377197265625, -143.55380249023438, 222.75357055664062, 7.211158752441406, 257.712158203125, -309.8680419921875, -12.965255737304688, 728.448486328125, 52.332183837890625, 252.9799041748047, 86.011962890625, -43.324668884277344, 111.29132080078125, 131.04522705078125, -9.405097961425781, 308.9599304199219, 42.369083404541016, 75.32748413085938, 43.4727783203125, 465.06134033203125, 188.85833740234375, 153.8956298828125, 50.017608642578125, 477.9365234375, -237.63638305664062, 211.64071655273438, -77.28775024414062, 41.855194091796875, -80.39913940429688, 56.93560791015625, 483.07373046875, 711.70166015625, 109.78516387939453, -290.32586669921875, 68.53910827636719, 70.1119384765625, -174.76901245117188, 277.5072937011719, 45.34358215332031, 86.103759765625, -241.16696166992188, 129.36737060546875, 504.0933837890625, 193.08563232421875, -17.959447860717773, 73.91616821289062, -103.07192993164062, 29.67987060546875, 1.6914596557617188, 41.986053466796875, 88.71829223632812, 132.423583984375, -19.609039306640625, -81.7413330078125, -229.7730255126953, 0.0, 323.3655700683594, 129.35787963867188, 378.674560546875, 278.63690185546875, 148.10333251953125, 13.932861328125, 18.923004150390625, -93.84072875976562, -247.74172973632812, 90.73098754882812, 25.370323181152344, -62.19189453125, -228.43960571289062, 234.05572509765625, -250.3085479736328, -16.27385711669922, -50.784149169921875, 197.42205810546875, -53.48272705078125, 224.17852783203125, -232.05274963378906, 255.21176147460938, 81.23148345947266, 48.34965515136719, 104.18560791015625, 222.72796630859375, 169.54043579101562, -15.1666259765625, 165.19888305664062, 273.9270935058594, 681.5385131835938, 190.33480834960938, 665.6363525390625, 80.2803955078125, 324.6226501464844, 7.79681396484375, 299.37493896484375, 55.952781677246094, -121.44149780273438, 173.4794921875, 258.2642822265625, -161.33602905273438, 528.5847778320312, 222.41024780273438, 5.660686492919922, 326.5706787109375, 234.5831298828125, -28.601104736328125, 8.15765380859375, 12.256439208984375, 475.40704345703125, -8.924121856689453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000410.npy"}
{"epoch": 0.8586387434554974, "step": 411, "batch_size": 128, "mean": 126.44950103759766, "std": 204.7080841064453, "min": -266.01824951171875, "p10": -59.814833831787105, "median": 93.45581817626953, "p90": 358.8613098144531, "max": 1366.4302978515625, "pos_frac": 0.765625, "sample": [-3.274076461791992, 251.86544799804688, 310.81683349609375, 52.4080810546875, 253.44659423828125, 42.8199462890625, 6.58917236328125, 67.18194580078125, 77.803955078125, 183.11221313476562, 437.0234375, 186.04061889648438, 36.59736251831055, -52.37310791015625, -114.183837890625, -49.162200927734375, 435.520751953125, 12.233322143554688, 220.037109375, 26.447052001953125, 95.92971801757812, 282.86065673828125, 148.57284545898438, 151.45993041992188, 104.73973083496094, 487.9230651855469, -50.988861083984375, 1366.4302978515625, 178.16314697265625, -22.6123046875, 344.1947326660156, 22.84067726135254, -149.05763244628906, 479.64215087890625, 126.79608917236328, 357.04620361328125, 304.9766845703125, 223.003173828125, 72.54888916015625, 1.5729827880859375, 306.096435546875, 109.7137451171875, -229.3787078857422, 286.6176452636719, 55.70538330078125, 21.7025146484375, 272.88763427734375, 138.00033569335938, -55.7142333984375, 5.857151031494141, 364.2018737792969, 125.7818603515625, -46.021881103515625, 5.464630126953125, 24.493234634399414, 380.4017333984375, -129.48941040039062, 91.10418701171875, 71.3583984375, -32.6551513671875, -20.33990478515625, 109.62628173828125, 51.15576171875, -250.4644775390625, 160.98321533203125, -3.365509033203125, 184.30926513671875, 325.0221862792969, 277.2041320800781, 147.0546875, -5.4442596435546875, 295.1307067871094, 191.16592407226562, 383.56536865234375, -58.355934143066406, 211.10476684570312, 36.03570556640625, -81.68850708007812, 559.3663330078125, 132.34613037109375, 89.07441711425781, -266.01824951171875, 181.097412109375, -6.797021865844727, 49.436676025390625, -12.816482543945312, -40.795318603515625, 22.040298461914062, 183.26260375976562, 393.7296447753906, 232.3367156982422, -18.395858764648438, 20.161468505859375, 132.50701904296875, 157.17007446289062, 115.81635284423828, -79.40121459960938, 223.43475341796875, 58.1395263671875, -116.81306457519531, 290.69635009765625, 35.82057189941406, -161.81851196289062, -32.238861083984375, 52.466651916503906, 76.19378662109375, 44.460174560546875, 149.1195068359375, 95.80744934082031, 227.52711486816406, 102.74681854248047, 103.46746826171875, 267.4158630371094, 24.70801544189453, 147.76318359375, -70.8468017578125, 80.00100708007812, 363.0965576171875, 33.746063232421875, 251.45925903320312, 412.94195556640625, 857.5973510742188, 54.04498291015625, 250.96090698242188, 99.18801879882812, -182.00088500976562, -63.21893310546875, 39.7283935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000411.npy"}
{"epoch": 0.8607329842931937, "step": 412, "batch_size": 128, "mean": 105.11750793457031, "std": 214.35826110839844, "min": -426.875244140625, "p10": -125.4246109008789, "median": 82.49504089355469, "p90": 360.5970397949218, "max": 824.9537963867188, "pos_frac": 0.6796875, "sample": [-59.44384765625, 530.301025390625, -2.6801910400390625, 348.3233642578125, 163.780517578125, 232.44354248046875, 275.898193359375, 249.00759887695312, 11.7962646484375, -157.58514404296875, -129.48110961914062, 102.44329833984375, 2.2576141357421875, -33.00164031982422, 94.59820556640625, -248.81304931640625, -119.85665893554688, -371.93084716796875, -110.70489501953125, 127.21244812011719, 30.523345947265625, 19.375564575195312, 99.53143310546875, 155.59716796875, 380.20819091796875, -44.45770263671875, 329.23919677734375, 322.26629638671875, -202.03366088867188, 86.182861328125, 184.9296875, 179.231689453125, 148.3297119140625, 108.80972290039062, 237.88931274414062, 277.32635498046875, 392.59210205078125, 136.52532958984375, 151.8353271484375, 162.92892456054688, 397.11041259765625, -155.96826171875, 211.56210327148438, -42.4205322265625, -75.69139862060547, -51.83343505859375, 19.76202392578125, -81.55001831054688, -182.70159912109375, 2.5775146484375, 352.1922607421875, 500.3746643066406, 60.49945068359375, -274.987060546875, 242.96800231933594, -70.15731811523438, 10.140533447265625, 73.94764709472656, 145.36785888671875, -109.80718994140625, -43.915618896484375, 74.68095397949219, 286.53350830078125, 100.68121337890625, 78.80722045898438, 681.1119384765625, 45.21894836425781, 722.6572875976562, 476.2774353027344, 198.906982421875, -84.54925537109375, 305.70184326171875, -426.875244140625, 12.164886474609375, 57.64988708496094, 76.84750366210938, 273.97088623046875, 117.74029541015625, -26.68389892578125, -252.32470703125, 118.41891479492188, 120.20723724365234, -17.592559814453125, 73.89732360839844, 146.29208374023438, -207.01004028320312, 548.47216796875, 100.01482391357422, 724.056884765625, -71.06282043457031, -23.499114990234375, 259.3121337890625, 329.2000732421875, -151.25128173828125, -150.19735717773438, -1.302886962890625, -66.940673828125, 168.91717529296875, 9.0977783203125, 137.3775177001953, 77.5389404296875, -4.214597702026367, 824.9537963867188, 161.8828887939453, 162.09844970703125, 55.941627502441406, 75.42613220214844, 121.01513671875, -123.68611145019531, -103.88827514648438, 327.30853271484375, 77.16522216796875, 0.0, 455.3830261230469, -43.653076171875, 202.9476318359375, -54.55914306640625, 116.39236450195312, 2.707366943359375, 130.21002197265625, 127.46265411376953, 240.2288818359375, -108.29228210449219, 476.31982421875, 63.80513000488281, -18.134323120117188, 170.23257446289062, 288.6575012207031], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000412.npy"}
{"epoch": 0.86282722513089, "step": 413, "batch_size": 128, "mean": 96.06249237060547, "std": 203.04502868652344, "min": -464.91143798828125, "p10": -140.69446411132807, "median": 76.86106872558594, "p90": 327.4828643798828, "max": 773.8713989257812, "pos_frac": 0.6796875, "sample": [91.14417266845703, -10.478126525878906, 37.965476989746094, 18.655311584472656, 20.514892578125, 197.23951721191406, -12.4056396484375, -215.9368896484375, -302.7167663574219, 43.657684326171875, 314.1508483886719, 289.674560546875, 311.7327880859375, -91.41812133789062, 66.40879821777344, 134.96002197265625, 225.77337646484375, 278.7357177734375, 255.32211303710938, -375.7222900390625, 147.1331787109375, 219.14175415039062, 154.89285278320312, 18.3038330078125, -67.93923950195312, -41.94744873046875, 64.31605529785156, 198.943603515625, 74.56539916992188, -95.6561279296875, 243.74639892578125, 76.66708374023438, -182.85450744628906, 196.25689697265625, 47.70635986328125, 292.7079772949219, 773.8713989257812, 83.95687866210938, 328.53741455078125, 90.61465454101562, 157.87200927734375, -5.521026611328125, -128.07406616210938, 256.4681701660156, 246.45892333984375, -38.998687744140625, -11.5628662109375, 153.95159912109375, 196.88873291015625, 119.49237060546875, 26.34185791015625, 46.137451171875, -60.64308166503906, 24.841888427734375, -28.268310546875, -208.26589965820312, 575.7323608398438, -222.16920471191406, 193.61715698242188, 74.27884674072266, 327.0309143066406, 164.50852966308594, -73.18994140625, 135.3341064453125, 121.19821166992188, 346.63330078125, -177.48233032226562, 279.093505859375, 54.627288818359375, -464.91143798828125, 265.5286865234375, 75.40139770507812, -24.33116912841797, 663.9881591796875, 115.56029510498047, -28.725265502929688, -5.9600830078125, 353.806884765625, 19.52386474609375, -192.9686279296875, -1.5265350341796875, 58.8402099609375, -114.4925537109375, 319.5528259277344, 203.2279052734375, -216.83255004882812, 142.2313232421875, 169.09915161132812, -14.879453659057617, -56.702117919921875, 359.8998107910156, -8.094675064086914, -126.61666870117188, 114.2783203125, 519.2586669921875, -170.14205932617188, 14.193359375, -108.31977844238281, 77.0550537109375, 97.37007141113281, 121.44804382324219, -100.42549133300781, 74.44509887695312, 257.8245849609375, 299.7222900390625, 167.9976806640625, -5.623199462890625, 90.24215698242188, 169.73468017578125, 234.07122802734375, -53.89872741699219, 153.294677734375, 32.55895233154297, 332.0245361328125, 31.813201904296875, -225.1368408203125, 328.743408203125, 186.64511108398438, -80.85842895507812, 745.5049438476562, 161.32041931152344, 389.4026184082031, 84.88681030273438, -90.93827819824219, 483.9479675292969, 237.36099243164062, -224.48312377929688, 13.50738525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000413.npy"}
{"epoch": 0.8649214659685864, "step": 414, "batch_size": 128, "mean": 127.18183898925781, "std": 213.23272705078125, "min": -389.8258056640625, "p10": -87.50524902343749, "median": 103.01643371582031, "p90": 369.63390502929684, "max": 1050.48486328125, "pos_frac": 0.7421875, "sample": [82.85958099365234, -216.26126098632812, 0.8331069946289062, 216.04550170898438, -30.58733367919922, 99.820556640625, 76.87591552734375, -82.45465087890625, -78.04299926757812, -39.83917236328125, -39.87359619140625, 73.97532653808594, 602.2945556640625, 128.79949951171875, 79.64549255371094, -12.898723602294922, 270.78118896484375, -135.9616241455078, 346.2780456542969, -10.375865936279297, -296.076904296875, -121.5526123046875, 352.41143798828125, 11.965576171875, 120.50665283203125, -110.31684112548828, -5.38397216796875, 224.7127227783203, 239.80484008789062, 241.49839782714844, 28.91097640991211, -11.13531494140625, 101.86917114257812, 216.22430419921875, -284.1651916503906, 319.4368896484375, 197.44009399414062, -218.45098876953125, 373.22930908203125, 246.39344787597656, -28.032623291015625, 307.7835388183594, 223.8636474609375, 113.89178466796875, 55.51374816894531, 748.6156616210938, 299.68719482421875, 30.508174896240234, 86.636474609375, 461.45623779296875, -53.049285888671875, 26.474166870117188, 14.199167251586914, 540.433837890625, 133.1304168701172, -140.26959228515625, 388.049072265625, 398.2662658691406, 213.53952026367188, 139.9525146484375, 104.1636962890625, 178.81773376464844, 78.07963562011719, 261.35870361328125, 69.974609375, -78.24513244628906, 307.27557373046875, 135.62060546875, 115.66278076171875, 87.61466979980469, 208.8543701171875, -34.235740661621094, 23.360565185546875, 293.69384765625, 104.54066467285156, 377.27734375, 160.06161499023438, -28.831985473632812, 180.97271728515625, -46.430763244628906, -41.5233154296875, 23.3896484375, -389.8258056640625, 1050.48486328125, 761.1387939453125, 70.33631896972656, 105.30499267578125, 100.7038345336914, -16.655105590820312, 15.012451171875, -251.00430297851562, 192.65802001953125, 230.08688354492188, 284.7897033691406, 34.28401565551758, 308.53009033203125, 2.2209930419921875, 339.1510009765625, -175.08580017089844, 190.2283477783203, -31.007064819335938, 124.195068359375, 272.20489501953125, 68.79682159423828, -16.307756423950195, 120.64247131347656, 127.56414794921875, 107.41871643066406, 567.561279296875, 289.0179138183594, 15.65283203125, 130.09555053710938, 368.093017578125, 130.87432861328125, 8.705142974853516, 67.63327026367188, -141.4009246826172, 413.4664611816406, 6.554962158203125, 255.22113037109375, 1.11737060546875, -16.52850341796875, -99.28997802734375, 81.72369384765625, 182.467041015625, 621.61083984375, 219.31689453125, 150.18008422851562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000414.npy"}
{"epoch": 0.8670157068062827, "step": 415, "batch_size": 128, "mean": 119.06481170654297, "std": 187.2147216796875, "min": -311.93359375, "p10": -98.25188827514648, "median": 92.553955078125, "p90": 364.6597900390625, "max": 732.1036987304688, "pos_frac": 0.7265625, "sample": [36.09039306640625, 402.16015625, 281.45904541015625, -37.46779251098633, -196.14715576171875, -19.349624633789062, 182.12423706054688, 139.44419860839844, 37.49156188964844, -48.76331329345703, -46.725341796875, -36.39737319946289, 173.7779083251953, 251.8148193359375, 11.596885681152344, -96.51898956298828, 732.1036987304688, 39.282318115234375, 109.80157470703125, 74.3126449584961, 195.96556091308594, -6.429443359375, 288.26141357421875, 184.3938751220703, 145.08114624023438, 122.39028930664062, -10.991668701171875, 101.24395751953125, 424.900146484375, 168.12033081054688, -23.68408203125, 261.66644287109375, 455.26995849609375, 43.80548095703125, -150.6287841796875, 89.95001220703125, 72.32818603515625, 592.9571533203125, -93.24530029296875, 52.515533447265625, -41.68262481689453, 190.94851684570312, 39.55665588378906, 25.754608154296875, 31.520042419433594, -232.47174072265625, 303.42816162109375, 263.940185546875, 225.8697509765625, -62.06349182128906, -156.20884704589844, 359.6142272949219, -91.25494384765625, -78.98529052734375, 250.06863403320312, 294.6949462890625, -23.16522216796875, 350.75408935546875, 200.8687744140625, -18.054901123046875, 21.14031982421875, 87.48109436035156, 64.67568969726562, 60.895751953125, 383.93939208984375, -39.74774169921875, 178.86529541015625, 250.6650390625, -0.67950439453125, 1.9491405487060547, -255.41265869140625, 218.44573974609375, 433.7336730957031, 254.740478515625, -15.833099365234375, 1.96368408203125, 60.358062744140625, 238.50787353515625, 52.453800201416016, -137.0731201171875, 191.19964599609375, 38.0118408203125, 126.99078369140625, -258.61187744140625, 330.5693359375, -102.29531860351562, -155.75888061523438, 179.12380981445312, 85.80307006835938, 145.5777587890625, 71.545166015625, 48.67584228515625, 312.3912048339844, 407.04412841796875, 198.348876953125, -8.781280517578125, 185.280517578125, 126.96414184570312, 420.093505859375, -311.93359375, 219.1956787109375, 174.7777099609375, 359.00592041015625, -147.7001953125, 59.6793212890625, 442.2485046386719, -208.2963104248047, -31.8707275390625, 565.4227294921875, 225.4550018310547, 362.99774169921875, 94.2786865234375, 96.14448547363281, 74.37371063232422, 9.580997467041016, 292.1463623046875, 145.67088317871094, 368.53790283203125, 288.5399169921875, 90.8292236328125, -104.84368896484375, 349.164306640625, 243.3050537109375, -81.39222717285156, 98.74679565429688, 204.23690795898438, 12.996002197265625, 408.6661376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000415.npy"}
{"epoch": 0.8691099476439791, "step": 416, "batch_size": 128, "mean": 132.89117431640625, "std": 199.07589721679688, "min": -389.4073791503906, "p10": -102.77026672363282, "median": 121.50231170654297, "p90": 370.0603942871094, "max": 869.713623046875, "pos_frac": 0.734375, "sample": [383.7237548828125, 324.46343994140625, 529.497314453125, 10.506103515625, -6.851570129394531, 74.31570434570312, -74.37115478515625, 76.16841888427734, -72.27394104003906, 152.2129364013672, -34.14813232421875, 345.00567626953125, 106.57803344726562, 78.16001892089844, -76.529296875, 91.41825103759766, 339.643310546875, 339.8860778808594, 423.2437744140625, 32.16680908203125, 208.91485595703125, 298.16119384765625, -139.37583923339844, 371.48419189453125, -102.74420166015625, 869.713623046875, -35.345794677734375, 103.79180908203125, 11.424308776855469, 34.346221923828125, 312.2050476074219, 457.0204772949219, 369.4501953125, 158.65774536132812, 132.19546508789062, 202.72769165039062, -389.4073791503906, 297.78070068359375, 430.3702697753906, -30.82720947265625, 309.9930419921875, 69.55130004882812, -0.4477996826171875, -337.2698974609375, -196.749755859375, 461.8655090332031, -13.180435180664062, 332.98590087890625, 185.8485107421875, 224.75042724609375, 15.4754638671875, 466.29522705078125, 63.80401611328125, 63.65795135498047, -95.05946350097656, 342.8135986328125, 351.25933837890625, 201.62881469726562, -212.39535522460938, 123.40560913085938, 164.2803955078125, 176.71563720703125, -39.411895751953125, 132.698486328125, 318.9329528808594, 171.15792846679688, 70.01271057128906, 150.7252197265625, -16.285247802734375, 202.4568634033203, -17.596206665039062, -102.83108520507812, 171.88055419921875, 506.72412109375, 443.1914978027344, 33.38905715942383, 38.84722900390625, 161.0653076171875, 248.86480712890625, -139.17666625976562, -46.930908203125, 26.494384765625, 258.759033203125, -72.15763854980469, -30.389007568359375, 8.801239013671875, -20.61383056640625, 105.98468017578125, -250.6588592529297, 11.69757080078125, 230.79266357421875, 309.6902770996094, 47.315948486328125, 327.90130615234375, -151.16259765625, 181.9102020263672, 522.8425903320312, 13.214691162109375, 193.6784210205078, 229.97390747070312, -143.642333984375, 228.04261779785156, 219.09637451171875, 70.59677124023438, 174.157470703125, -121.92373657226562, 239.76519775390625, 47.130043029785156, 331.81488037109375, 190.45565795898438, -14.047523498535156, 339.03375244140625, 92.85832214355469, -226.6412353515625, -20.94635009765625, 119.59901428222656, 243.61721801757812, 377.4061279296875, 189.02197265625, -175.08413696289062, 1.8157958984375, -43.37158203125, 276.477294921875, 116.80176544189453, 355.43701171875, 56.84693908691406, 203.4000244140625, 347.9754638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000416.npy"}
{"epoch": 0.8712041884816754, "step": 417, "batch_size": 128, "mean": 113.9468994140625, "std": 229.28038024902344, "min": -367.09051513671875, "p10": -120.97269287109374, "median": 86.82389831542969, "p90": 351.3926391601562, "max": 1293.05810546875, "pos_frac": 0.7109375, "sample": [171.63291931152344, 330.0722961425781, 295.27899169921875, 59.86100769042969, -75.88224792480469, 698.3504638671875, 23.702880859375, -259.97845458984375, 445.85382080078125, 256.1141662597656, 21.38848876953125, 59.07208251953125, 158.5540771484375, 189.69061279296875, -367.09051513671875, 56.34129333496094, 15.410354614257812, -127.82868957519531, -282.3004150390625, 86.877197265625, 489.59759521484375, -291.73712158203125, 624.45166015625, 75.31332397460938, -32.407432556152344, -29.749908447265625, 64.9404296875, 55.2256965637207, -204.68753051757812, 691.85986328125, 287.7607421875, -46.3505859375, 226.28228759765625, 178.7918701171875, 161.14865112304688, 629.842041015625, 158.71484375, -75.49716186523438, 246.14736938476562, 161.89697265625, 234.70498657226562, 86.77059936523438, 83.64653015136719, 100.63409423828125, 111.92849731445312, 230.93942260742188, 46.594818115234375, 108.199951171875, 37.06623840332031, 347.228271484375, 170.45782470703125, 9.262598037719727, -50.927581787109375, -51.69953918457031, -293.1606750488281, 221.09828186035156, -72.7440185546875, 75.11111450195312, 118.791015625, -52.22943115234375, 56.8304443359375, 332.0782470703125, -22.75899314880371, -23.27758026123047, 176.70071411132812, -116.72743225097656, 301.0470275878906, 39.065765380859375, -351.9809265136719, -1.2165298461914062, -118.42337036132812, 298.1265563964844, 442.6988525390625, 144.97772216796875, 237.86248779296875, 73.8114013671875, -3.5888214111328125, 322.24578857421875, 0.0, 61.87872314453125, 133.6714630126953, 0.0, 139.5916748046875, 126.49757385253906, 116.23359680175781, 94.03045654296875, 41.18614196777344, -3.3617095947265625, -105.93698120117188, 8.170745849609375, -136.50436401367188, 285.4863586425781, 211.93814086914062, 331.5514221191406, 1293.05810546875, 130.77944946289062, -277.8584899902344, 65.07484436035156, 553.687255859375, 193.98667907714844, 234.5308837890625, -126.92111206054688, -322.08306884765625, 4.861396789550781, 153.26661682128906, 125.12814331054688, 361.1094970703125, 1.762786865234375, -97.64996337890625, -11.979461669921875, 5.0078125, -48.4801025390625, 32.82341766357422, 386.27947998046875, 101.01666259765625, 187.96038818359375, -49.488677978515625, 326.784912109375, 401.6866455078125, 2.9201393127441406, 432.19793701171875, 133.718017578125, 216.85377502441406, -161.45350646972656, 216.1317596435547, 114.67559814453125, 328.17889404296875, -2.604248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000417.npy"}
{"epoch": 0.8732984293193717, "step": 418, "batch_size": 128, "mean": 121.0215835571289, "std": 249.7872314453125, "min": -667.8074340820312, "p10": -127.54127044677735, "median": 104.98632049560547, "p90": 384.10387573242184, "max": 1473.17333984375, "pos_frac": 0.7265625, "sample": [-307.07391357421875, 333.3901672363281, 49.76434326171875, 276.20849609375, 87.95703125, -125.29222106933594, 1473.17333984375, 295.174072265625, -667.8074340820312, -415.0213928222656, 106.40177917480469, 100.76054382324219, 128.659423828125, 31.407058715820312, 319.760498046875, 218.7657012939453, 351.1460266113281, -221.0687713623047, -248.85403442382812, 121.63960266113281, 689.280029296875, 141.1702880859375, -20.5130615234375, 57.297218322753906, -72.5230712890625, 70.09176635742188, -151.82119750976562, 404.1588134765625, 68.46563720703125, 593.3910522460938, 381.6118469238281, 128.52349853515625, 504.89056396484375, 237.36441040039062, 151.3334503173828, 669.7174072265625, 0.2544403076171875, 171.49636840820312, -101.51165771484375, 208.53805541992188, -91.2802963256836, 251.71209716796875, -57.214752197265625, 264.3041076660156, 44.56703186035156, -28.31817626953125, -126.60240173339844, 91.24658966064453, 318.4700927734375, 146.95938110351562, -18.95904541015625, 781.8207397460938, 31.559722900390625, 165.40696716308594, 22.889785766601562, -15.06817626953125, 55.57232666015625, 118.40008544921875, 541.26953125, 80.93594360351562, -91.75921630859375, 27.318450927734375, 104.08590698242188, 238.01223754882812, 430.22271728515625, -12.68243408203125, 225.73939514160156, -55.477455139160156, 74.2529296875, 176.730712890625, -20.6343994140625, 105.88673400878906, 109.2894058227539, 113.8563003540039, 97.8568115234375, 186.34548950195312, 379.2623291015625, -202.5089111328125, 304.1025390625, -129.73196411132812, 130.12844848632812, 160.16842651367188, -222.1810302734375, 33.792938232421875, 309.3940124511719, 413.5699462890625, 92.51998901367188, -32.55339431762695, 192.7269287109375, -39.185302734375, 289.1029052734375, -173.22003173828125, 55.916168212890625, 193.09854125976562, 227.708984375, 152.8509979248047, 312.7179870605469, 165.0709228515625, 70.82077026367188, 221.73336791992188, -124.5455322265625, -0.1751861572265625, 172.30198669433594, 46.63637161254883, -311.1517333984375, 112.3895263671875, 50.71826171875, 223.13516235351562, -290.49432373046875, 119.8006591796875, 129.7694091796875, 143.40789794921875, 61.445037841796875, 41.67814636230469, 162.64585876464844, 201.5220947265625, 584.76513671875, 35.21722412109375, 40.60590362548828, 389.9186096191406, -1.33612060546875, -44.40049743652344, -80.3355712890625, -148.43682861328125, 57.77880859375, 148.06521606445312, 621.005126953125, -83.49444580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000418.npy"}
{"epoch": 0.875392670157068, "step": 419, "batch_size": 128, "mean": 109.97142791748047, "std": 242.36204528808594, "min": -941.6162109375, "p10": -165.17926788330078, "median": 102.46807861328125, "p90": 357.84825439453124, "max": 919.3839111328125, "pos_frac": 0.703125, "sample": [222.92782592773438, -173.81137084960938, -102.72738647460938, 307.28900146484375, 137.52223205566406, 388.809326171875, 38.774375915527344, 146.6402587890625, 192.73167419433594, -341.5734558105469, -25.331302642822266, 99.974609375, -246.89108276367188, 208.09750366210938, 42.93096160888672, 85.47564697265625, 246.33087158203125, 140.73568725585938, 149.83370971679688, 225.8818359375, 267.4619140625, 81.94622802734375, 219.267822265625, 231.49520874023438, 357.037353515625, 98.7003173828125, 45.077362060546875, 28.86895751953125, -56.3267822265625, -283.8921813964844, 414.56414794921875, 184.85446166992188, -33.22092819213867, -30.67645263671875, -4.626220703125, -18.201934814453125, 320.5644226074219, -6.4725189208984375, 123.05419921875, -458.31427001953125, 283.87115478515625, -156.22962951660156, 138.0838623046875, 104.9615478515625, 601.9351196289062, 44.3621826171875, 375.0784606933594, -286.99383544921875, 74.83642578125, 208.55825805664062, 120.5831298828125, 483.498779296875, -78.53838348388672, 165.874267578125, 181.35549926757812, 111.6480712890625, 919.3839111328125, 280.4434814453125, 48.74847412109375, 245.77798461914062, -279.25433349609375, 84.97152709960938, 123.29178619384766, -15.37392807006836, 359.7403564453125, -163.75404357910156, 345.2642822265625, 565.6284790039062, 65.23432922363281, 231.8247833251953, -246.44032287597656, 308.7770080566406, 299.42510986328125, 17.256500244140625, 63.81956481933594, 0.0, -7.044647216796875, -601.3630981445312, 51.752197265625, 0.0, 66.06741333007812, 143.3169708251953, 213.77093505859375, 186.8940887451172, -11.332626342773438, -8.703903198242188, -57.61961364746094, 319.8642578125, 209.34234619140625, -86.43379211425781, 223.98025512695312, 69.43841552734375, 202.62017822265625, -26.693206787109375, -56.94612121582031, 701.44580078125, 199.92568969726562, 270.646728515625, 199.6959228515625, 643.2802124023438, 369.3367614746094, 58.04254150390625, 535.67333984375, -64.98796081542969, 181.550537109375, -168.50479125976562, 229.31301879882812, 297.28070068359375, 94.54345703125, 333.3724365234375, -45.27288818359375, -346.544921875, 243.7833251953125, 315.1478271484375, -941.6162109375, 56.12965393066406, -225.25018310546875, 50.97315216064453, -3.57208251953125, 97.01806640625, 150.2283935546875, 354.7047119140625, 65.6832275390625, 42.8675537109375, 169.07833862304688, -19.402450561523438, 519.49658203125, 28.93408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000419.npy"}
{"epoch": 0.8774869109947644, "step": 420, "batch_size": 128, "mean": 142.38693237304688, "std": 263.4527587890625, "min": -705.81689453125, "p10": -139.08905792236328, "median": 122.34814071655273, "p90": 439.0421752929687, "max": 1130.82275390625, "pos_frac": 0.71875, "sample": [25.359676361083984, 160.20941162109375, 268.8914794921875, -317.4732666015625, -182.5707244873047, 133.6223907470703, 96.43408203125, 370.3558349609375, -211.8052978515625, 225.457763671875, 224.14889526367188, 8.4852294921875, -239.12234497070312, 68.96188354492188, 101.20619201660156, 59.376739501953125, 206.22068786621094, -46.27519226074219, 115.58822631835938, -66.73173522949219, 504.4313659667969, 120.9083480834961, -28.41101837158203, 119.95843505859375, -35.16096496582031, 909.3045654296875, 173.61465454101562, -23.108421325683594, 222.6865997314453, 552.8037109375, -62.265625, 451.3536682128906, 351.13970947265625, 240.87362670898438, 116.28375244140625, 318.111083984375, 252.88970947265625, 1130.82275390625, 340.2042236328125, 110.36517333984375, 219.9681396484375, 16.869674682617188, -76.62452697753906, 829.3477783203125, 134.20477294921875, 245.60470581054688, 329.0751953125, -42.05326843261719, 22.082962036132812, 48.82391357421875, -118.55757141113281, 230.25372314453125, 221.0941619873047, 463.88275146484375, 156.1192626953125, 146.8943634033203, 0.0, 36.34075927734375, -469.2437744140625, -151.82101440429688, 238.1115264892578, 219.4668426513672, -32.792633056640625, -40.88500213623047, -138.3418731689453, 123.78793334960938, 170.71780395507812, 279.519287109375, 532.412353515625, 237.47152709960938, -6.181884765625, -14.682968139648438, 23.19891357421875, 250.0422821044922, -221.05963134765625, 606.6982421875, -52.24042510986328, 179.6383514404297, 449.7254638671875, 16.673492431640625, -47.09577941894531, 35.801570892333984, 343.402099609375, -140.83248901367188, 457.56158447265625, 115.76467895507812, -119.70892333984375, 46.335880279541016, 378.07781982421875, 255.17129516601562, -196.41360473632812, 11.543212890625, 972.64306640625, 230.33233642578125, 102.68061828613281, 169.06283569335938, 99.879150390625, 233.70071411132812, 297.9886474609375, -13.6693115234375, 378.92864990234375, 284.8439636230469, -560.29296875, 230.63442993164062, 39.93334197998047, 65.52313232421875, 112.930908203125, -27.98248291015625, -705.81689453125, -89.33843231201172, 400.99822998046875, 353.54168701171875, 161.4359893798828, 284.57586669921875, 286.4991455078125, 23.40924072265625, 321.77569580078125, -231.1458740234375, 371.1380615234375, 283.7462158203125, 511.65240478515625, 434.463623046875, -105.21868896484375, 155.18215942382812, 41.378509521484375, 391.7987060546875, -167.24383544921875, -84.73553466796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000420.npy"}
{"epoch": 0.8795811518324608, "step": 421, "batch_size": 128, "mean": 120.446044921875, "std": 201.2406005859375, "min": -824.7244262695312, "p10": -79.92758407592773, "median": 103.55493927001953, "p90": 348.20244140624993, "max": 759.6285400390625, "pos_frac": 0.7421875, "sample": [280.3162841796875, 2.3097991943359375, 173.55859375, -78.22286987304688, 18.66802978515625, 53.38008117675781, 51.49853515625, 52.2908935546875, -193.02587890625, -269.3406982421875, 223.07394409179688, 63.002777099609375, -65.54466247558594, 48.92730712890625, -21.908950805664062, 246.01226806640625, 80.5867919921875, -127.40180969238281, 167.0831298828125, 218.6771697998047, 72.17671203613281, -4.4925689697265625, 74.55136108398438, 58.72516632080078, 131.38177490234375, 759.6285400390625, 237.7603759765625, 92.91583251953125, 102.70176696777344, 246.75335693359375, 0.0, -824.7244262695312, 160.51560974121094, -93.99547576904297, 40.26841735839844, 18.54400634765625, 321.8865966796875, 262.07464599609375, 117.83570861816406, 195.78924560546875, 290.7990417480469, 296.0288391113281, 108.73101806640625, -11.294246673583984, -22.492355346679688, 205.39163208007812, 270.3041687011719, -46.30328369140625, 367.45013427734375, 139.23187255859375, 216.16110229492188, -197.0452423095703, 280.4880676269531, 176.39645385742188, 92.28421020507812, 723.02587890625, 38.689857482910156, 372.27838134765625, -16.431396484375, 378.0443115234375, 96.045654296875, -37.14939880371094, 320.99212646484375, 315.9920349121094, 217.1876220703125, -30.243812561035156, 232.42723083496094, 155.1485595703125, 445.6956481933594, 539.3231811523438, 318.5011901855469, 520.5032958984375, 104.40811157226562, 125.65457153320312, 56.692283630371094, 306.6051330566406, 81.15376281738281, 64.88623046875, 140.6543426513672, 139.10220336914062, 281.26312255859375, 25.16564178466797, 159.3834228515625, 379.40869140625, -14.97003173828125, 53.001068115234375, 503.2421875, -97.21588134765625, 137.57037353515625, -83.9052505493164, 146.7366180419922, 166.95689392089844, 431.0283203125, 235.02694702148438, 21.6755428314209, 37.101715087890625, 135.110595703125, -89.14460754394531, -136.02267456054688, 11.398513793945312, -205.320068359375, -22.989898681640625, 91.5203857421875, 339.95343017578125, 68.387939453125, 75.78115844726562, 182.51690673828125, 12.6650390625, 192.75958251953125, -253.72698974609375, -48.197540283203125, 205.86297607421875, 240.27114868164062, 150.7519073486328, -310.0350341796875, 273.8631896972656, 176.7515106201172, 0.0, -23.19573974609375, -21.155319213867188, 531.5943603515625, 385.1279296875, 128.65255737304688, 0.0, -19.140640258789062, 37.97088623046875, 299.36907958984375, -45.3126220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000421.npy"}
{"epoch": 0.881675392670157, "step": 422, "batch_size": 128, "mean": 113.52565002441406, "std": 218.0114288330078, "min": -362.5631408691406, "p10": -138.77519073486326, "median": 83.91642761230469, "p90": 357.87584838867184, "max": 1163.03662109375, "pos_frac": 0.6640625, "sample": [30.8359375, -362.5631408691406, 115.81637573242188, -157.97422790527344, 154.17315673828125, -203.6842803955078, -79.37228393554688, -192.53314208984375, 122.68325805664062, -19.868194580078125, 263.7955627441406, 331.41357421875, 188.10498046875, 216.51837158203125, -33.154052734375, -53.883544921875, -14.2509765625, -22.19598388671875, -178.274658203125, 173.2614288330078, 190.34300231933594, 55.854400634765625, -40.9136962890625, -52.61796569824219, -24.484893798828125, 1163.03662109375, -123.176513671875, 255.60948181152344, 179.10337829589844, 518.2623291015625, 53.43650436401367, -75.06854248046875, 29.466110229492188, 626.1414794921875, -58.747283935546875, 178.98504638671875, 14.765872955322266, 18.359725952148438, 249.88113403320312, -4.835845947265625, -93.72181701660156, 343.5482177734375, -3.390777587890625, 96.28778076171875, 92.24905395507812, -286.20013427734375, -47.344970703125, 327.7604675292969, 382.9441223144531, 241.27886962890625, 367.72314453125, 119.62030029296875, 177.53073120117188, 87.90333557128906, 477.90545654296875, 173.21340942382812, -131.88844299316406, 169.229736328125, -310.314697265625, 111.18264770507812, 598.78125, -3.8818016052246094, -31.12384033203125, -187.73001098632812, 317.537841796875, 73.554443359375, 45.754669189453125, 112.39662170410156, -70.85537719726562, 341.29937744140625, 178.17422485351562, -104.81826782226562, 391.4827575683594, 47.4127197265625, 126.525634765625, -111.59552764892578, 72.55619812011719, 53.22601318359375, 330.46051025390625, 43.00324249267578, 299.935302734375, -58.706687927246094, 287.55322265625, 128.28045654296875, 252.3028564453125, 165.32546997070312, 181.9110107421875, -36.474273681640625, 353.65557861328125, 77.17146301269531, 7.703651428222656, 19.902587890625, 337.2668762207031, -84.0462646484375, -7.846805572509766, 218.92529296875, -2.6226806640625, -41.55412292480469, 375.36883544921875, -60.86597442626953, 243.44322204589844, 514.05859375, 448.1392822265625, 195.98553466796875, -154.84426879882812, 200.45245361328125, 210.59259033203125, -165.18695068359375, 68.77838134765625, 471.552734375, 0.0, -230.63134765625, 315.1271667480469, 287.607666015625, 87.8712158203125, 78.31710815429688, -281.203369140625, 278.60040283203125, 78.7862548828125, 79.96163940429688, -163.1429443359375, 200.33908081054688, 41.414756774902344, 591.311767578125, 154.053955078125, 225.26705932617188, 76.64535522460938, 314.8749084472656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000422.npy"}
{"epoch": 0.8837696335078534, "step": 423, "batch_size": 128, "mean": 129.40359497070312, "std": 213.39352416992188, "min": -507.94500732421875, "p10": -123.8750274658203, "median": 133.72410583496094, "p90": 362.07841796875, "max": 1000.9871215820312, "pos_frac": 0.7421875, "sample": [-39.03825378417969, 1000.9871215820312, -306.7445068359375, -131.98806762695312, 361.79083251953125, 1.0675048828125, 163.77430725097656, -21.908538818359375, -24.39239501953125, 250.62484741210938, 80.3739013671875, 9.932159423828125, 86.3988037109375, 92.87440490722656, -149.8804931640625, -46.016357421875, 298.6376953125, 553.4039306640625, 177.645751953125, 185.23397827148438, 64.92417907714844, 37.2706298828125, 56.50718688964844, 285.12493896484375, -163.47894287109375, 58.4320068359375, 244.0216064453125, 230.22518920898438, 374.70330810546875, 240.34112548828125, -24.521453857421875, 94.146484375, 159.7901611328125, 175.41668701171875, -7.492401123046875, 386.0321044921875, 406.0501708984375, 79.04736328125, -184.50921630859375, 238.55746459960938, 100.7798080444336, 264.19915771484375, -18.2978515625, 274.75640869140625, 70.65670776367188, -35.97663116455078, 222.07676696777344, 260.639892578125, 109.883544921875, -157.84075927734375, 326.9794616699219, -249.26950073242188, -464.5686950683594, 236.9906005859375, 302.3167724609375, 82.66513061523438, -118.57183837890625, -103.17277526855469, 1.8955078125, 190.95162963867188, 72.71900939941406, 55.38153076171875, 290.5968017578125, 188.9164581298828, -189.27706909179688, 186.61761474609375, -60.77778625488281, 215.20999145507812, -48.429969787597656, 220.47952270507812, 178.32200622558594, 216.82215881347656, 145.7667236328125, 82.19143676757812, 151.63644409179688, 127.06784057617188, -170.34112548828125, 262.22503662109375, 351.958251953125, -2.362091064453125, 59.93048095703125, 504.2033996582031, 175.65980529785156, 413.7134704589844, 244.22760009765625, 221.92306518554688, 341.3837890625, 184.04156494140625, -17.56304931640625, 483.19012451171875, 362.74945068359375, 51.843658447265625, 100.72991943359375, -266.2581481933594, 385.4129638671875, 289.9223937988281, -15.716453552246094, 25.85919189453125, 222.40255737304688, 283.08782958984375, -47.111175537109375, 253.46920776367188, 853.97900390625, 154.4352569580078, 307.9930419921875, -507.94500732421875, 40.448150634765625, 279.7905578613281, 189.68441772460938, 93.94489288330078, -138.81732177734375, 12.778642654418945, 29.74078369140625, 140.38037109375, 1.4009132385253906, 250.21463012695312, 369.2987976074219, 41.259429931640625, 540.0571899414062, 24.012481689453125, 191.36907958984375, 279.25128173828125, 252.02603149414062, -120.39801025390625, 272.19122314453125, -43.714263916015625, -5.149042129516602, -66.855712890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000423.npy"}
{"epoch": 0.8858638743455497, "step": 424, "batch_size": 128, "mean": 116.56576538085938, "std": 195.89158630371094, "min": -416.2470703125, "p10": -113.317610168457, "median": 111.82642364501953, "p90": 334.98440246582027, "max": 969.3406372070312, "pos_frac": 0.7578125, "sample": [183.9977569580078, 410.67584228515625, 180.0087432861328, -48.86419677734375, 43.72907638549805, 6.206939697265625, -53.57666015625, -39.9605712890625, 50.94525146484375, 52.5562744140625, -6.913116455078125, -166.36489868164062, 572.693359375, 331.8953857421875, -362.0489501953125, 190.37513732910156, -24.01806640625, 202.1171875, 162.6029052734375, 233.56129455566406, -33.896392822265625, 70.4010009765625, 215.29132080078125, 482.6634826660156, 71.54141235351562, 89.12577819824219, 56.847862243652344, -179.62283325195312, 148.401123046875, 135.77244567871094, 12.071807861328125, 451.14202880859375, 151.61495971679688, 50.45262145996094, 201.86273193359375, 216.55935668945312, 307.2199401855469, 296.0374450683594, 276.83184814453125, -64.03753662109375, 36.48919677734375, 233.07765197753906, 342.1921081542969, -359.39215087890625, -8.631156921386719, 244.6060791015625, 245.55859375, 262.2098388671875, -180.98187255859375, 56.59466552734375, -136.61473083496094, 199.9569091796875, -148.82562255859375, 201.094970703125, 223.9915771484375, 565.177734375, -135.44578552246094, 219.57418823242188, 254.97317504882812, -17.29095458984375, -87.23013305664062, -46.0986328125, 181.38381958007812, 268.8212585449219, 619.0252685546875, 178.07818603515625, 366.0662841796875, 75.16259765625, -416.2470703125, 33.41796875, -72.12109375, 453.4278869628906, 107.39495849609375, 111.44671630859375, 347.69073486328125, 159.81320190429688, -160.52163696289062, 324.09393310546875, 98.14695739746094, 122.89834594726562, 134.149169921875, 311.314697265625, 18.797767639160156, 53.60044860839844, 157.38719177246094, -24.641042709350586, 53.15509033203125, 0.0, 112.20613098144531, 185.33819580078125, 11.4127197265625, 438.83197021484375, 204.77493286132812, 0.939849853515625, -28.114391326904297, 160.00576782226562, 122.91790771484375, 187.81680297851562, 209.91818237304688, 37.570587158203125, 39.236942291259766, 71.69976806640625, 281.4077453613281, 134.5537872314453, 249.8360595703125, -148.87969970703125, -64.60165405273438, -52.453094482421875, -194.1987762451172, 43.25450897216797, 69.10926818847656, -103.8341064453125, 154.9010009765625, 231.07595825195312, -231.28704833984375, 10.37498664855957, 25.389089584350586, 174.38653564453125, 38.743316650390625, 969.3406372070312, 111.10467529296875, 129.47311401367188, 82.8701171875, 165.4619140625, 236.64007568359375, 123.19078063964844, 37.67361831665039, 349.7025146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000424.npy"}
{"epoch": 0.8879581151832461, "step": 425, "batch_size": 128, "mean": 114.23272705078125, "std": 215.0497283935547, "min": -380.2456359863281, "p10": -133.94042205810547, "median": 94.64501953125, "p90": 369.51432495117183, "max": 1162.7481689453125, "pos_frac": 0.6875, "sample": [234.4822998046875, 344.6352233886719, 96.95712280273438, 8.117233276367188, 2.939208984375, 150.94891357421875, 96.08486938476562, 250.54656982421875, 93.20516967773438, -251.88037109375, 382.2532653808594, 65.9556884765625, 400.970458984375, -31.625335693359375, 55.78033447265625, 3.704591751098633, 154.64662170410156, 502.96435546875, 9.329095840454102, 45.08837890625, 202.0087890625, 217.53860473632812, 162.4875946044922, 6.676883697509766, 66.14617919921875, -48.36444854736328, 7.858768463134766, 123.80010986328125, -0.93902587890625, -6.25909423828125, 79.8360595703125, -58.819183349609375, -190.54940795898438, 77.76536560058594, 320.69140625, -11.874090194702148, 192.70114135742188, 445.5732421875, 43.017425537109375, 325.35955810546875, -60.00152587890625, 460.2589111328125, 201.723388671875, 214.99969482421875, 205.0484619140625, -20.643596649169922, -290.40655517578125, -97.525390625, -252.56930541992188, 415.22991943359375, 132.35690307617188, 188.7935028076172, -183.6153564453125, 201.59588623046875, 256.9392395019531, 101.83688354492188, 250.41677856445312, 364.0547790527344, -21.69659423828125, 0.6442718505859375, -272.05816650390625, 332.27142333984375, 168.84487915039062, -51.622589111328125, 252.56695556640625, -61.66608428955078, 153.71128845214844, -12.442293167114258, 410.10516357421875, 123.93450927734375, 230.61090087890625, 70.27554321289062, 311.6834716796875, -81.85913848876953, -143.376708984375, -23.35986328125, -138.26641845703125, 322.6160888671875, 37.08582305908203, 197.78662109375, -0.9539985656738281, 354.6736755371094, 103.03790283203125, 403.77618408203125, 187.4586181640625, -137.17921447753906, 11.246597290039062, -89.43928527832031, -22.96440887451172, -49.37249755859375, -14.59228515625, -8.323375701904297, 74.27234649658203, -380.2456359863281, 148.1534423828125, 9.9793701171875, 153.261962890625, -254.1578369140625, 203.37066650390625, -119.0081787109375, 257.69482421875, 334.6636047363281, 416.5404052734375, 206.11630249023438, 123.46510314941406, 131.49966430664062, 786.3013916015625, 61.45628356933594, -132.5523681640625, 220.21612548828125, 0.0, -8.46026611328125, 223.74179077148438, 191.928466796875, 1162.7481689453125, -12.810832977294922, 51.198970794677734, 406.01556396484375, -252.05648803710938, -5.2618408203125, 278.46435546875, 636.7723388671875, 136.02369689941406, 209.157470703125, 230.77194213867188, -167.87823486328125, 70.71983337402344, 28.276275634765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000425.npy"}
{"epoch": 0.8900523560209425, "step": 426, "batch_size": 128, "mean": 120.25263977050781, "std": 207.5271453857422, "min": -677.3546142578125, "p10": -95.70395507812499, "median": 98.53670501708984, "p90": 386.35715942382814, "max": 818.4581909179688, "pos_frac": 0.7265625, "sample": [207.95474243164062, -46.97322082519531, 5.821748733520508, -79.3768310546875, 95.23727416992188, -136.5955047607422, 113.36563110351562, -9.01156997680664, 159.00836181640625, -14.347938537597656, -677.3546142578125, -38.6075439453125, -164.45712280273438, 213.3557891845703, 191.5192108154297, 136.14808654785156, 217.530029296875, 107.437255859375, 148.3577117919922, 108.5010986328125, -287.3882141113281, 25.197265625, 305.9249572753906, 145.16445922851562, 48.39582061767578, -12.9017333984375, 283.1177978515625, 377.508544921875, 178.37115478515625, 496.53985595703125, 125.85566711425781, 115.5721435546875, 557.0164794921875, 156.35198974609375, 446.64007568359375, 19.730356216430664, 15.637435913085938, 227.78219604492188, 234.34588623046875, 161.94677734375, 105.10629272460938, 324.96282958984375, 17.243179321289062, -30.298545837402344, 385.30340576171875, 421.6695251464844, 22.34222412109375, 573.8643188476562, 185.1319580078125, -93.260009765625, -122.32192993164062, 206.79202270507812, 21.767127990722656, -36.618316650390625, 189.3934326171875, 507.43634033203125, -174.79837036132812, 44.15227508544922, 524.9501342773438, 52.7412109375, 378.5143737792969, 286.7755432128906, -221.62789916992188, -115.91612243652344, 8.148284912109375, 353.68023681640625, 366.8653869628906, 95.06048583984375, 94.64759063720703, 65.6796875, 100.56494140625, 171.96109008789062, -77.50132751464844, 211.4158935546875, 96.50846862792969, 376.03485107421875, -210.89169311523438, 73.05271911621094, -12.927993774414062, 122.13401794433594, 54.766578674316406, 274.961669921875, 284.49493408203125, 54.638145446777344, 462.5801696777344, 58.81024169921875, -8.26611328125, 13.584121704101562, 112.1493148803711, 313.36456298828125, 62.483154296875, 78.257568359375, -37.4030647277832, -279.1720886230469, 364.4219055175781, 467.50665283203125, -5.66766357421875, 147.73590087890625, 2.4171905517578125, -25.509244918823242, 221.58673095703125, 47.91851806640625, 0.0, 57.44745635986328, -4.532379150390625, 818.4581909179688, 388.81591796875, -61.87115478515625, 154.825927734375, 176.40562438964844, 195.0260009765625, 46.6640625, -238.2784423828125, -91.68489837646484, -9.9139404296875, 146.74435424804688, -101.406494140625, 373.7716064453125, 389.76715087890625, 319.84039306640625, 3.7774429321289062, -67.07693481445312, -170.85577392578125, -69.7686767578125, 297.0818786621094, 219.49539184570312, 47.17034912109375, 460.71630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000426.npy"}
{"epoch": 0.8921465968586387, "step": 427, "batch_size": 128, "mean": 170.94326782226562, "std": 256.38348388671875, "min": -592.8646240234375, "p10": -65.52594604492185, "median": 172.06693267822266, "p90": 491.911993408203, "max": 1057.365478515625, "pos_frac": 0.7578125, "sample": [71.25326538085938, 450.45147705078125, 183.13821411132812, -458.9478759765625, 60.22259521484375, 215.3837127685547, 180.63385009765625, 449.1336669921875, 267.1451721191406, 230.94842529296875, 86.018310546875, 192.1299591064453, 315.5772705078125, -25.429317474365234, 468.3885803222656, 583.409423828125, 263.3251647949219, -18.16216278076172, 315.5799560546875, 600.340087890625, 260.4203186035156, 139.8327178955078, 24.925460815429688, -24.39501953125, 625.97119140625, 60.489166259765625, 355.9985656738281, 411.727294921875, 40.89011001586914, 191.33770751953125, 580.9296875, 332.88922119140625, -40.79364013671875, 73.06671142578125, 155.1236572265625, 143.35272216796875, 93.27136993408203, 6.385160446166992, -34.60784912109375, -494.65576171875, 344.20843505859375, 109.48284912109375, -213.95323181152344, 136.85476684570312, -61.2178955078125, 206.7001495361328, -27.05361557006836, 609.348388671875, -54.60987854003906, 255.969970703125, 188.56283569335938, 177.6322021484375, 102.66917419433594, 219.0184326171875, 339.5840148925781, 886.4769287109375, 217.244384765625, -143.67010498046875, -55.890228271484375, -143.99609375, 166.86155700683594, 178.67120361328125, 453.58154296875, -3.87164306640625, 117.36273193359375, 126.05305480957031, 11.905029296875, 222.08395385742188, 242.5906219482422, -227.28680419921875, -592.8646240234375, 378.34320068359375, -1.927581787109375, -586.9055786132812, 1057.365478515625, 28.280637741088867, -56.95759582519531, 204.56686401367188, 334.27838134765625, -13.539848327636719, 335.0635986328125, -162.3947296142578, 194.7991943359375, -95.93936157226562, -154.72816467285156, 326.21295166015625, 297.9097595214844, 364.4317321777344, 425.6014709472656, 251.898193359375, 103.123291015625, 204.68881225585938, -2.205078125, 23.160980224609375, 581.8533325195312, 539.166259765625, 477.3846435546875, 556.162353515625, -33.06329345703125, 25.287029266357422, 111.59500122070312, 238.12298583984375, 231.98410034179688, 103.03703308105469, 191.00344848632812, 525.8091430664062, 32.1463623046875, -75.57806396484375, 72.86537170410156, 177.27230834960938, 665.62158203125, 30.897308349609375, 106.36839294433594, 97.5816650390625, 314.20379638671875, -181.677490234375, 258.7176818847656, 66.18882751464844, -19.66619873046875, -30.8892822265625, 464.1963806152344, 306.239990234375, -22.344970703125, 10.462928771972656, 144.35531616210938, 438.58306884765625, 559.4097900390625, 341.16827392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000427.npy"}
{"epoch": 0.8942408376963351, "step": 428, "batch_size": 128, "mean": 147.7657470703125, "std": 222.7470703125, "min": -543.3193359375, "p10": -88.82394104003906, "median": 112.05220413208008, "p90": 449.0981384277342, "max": 913.9177856445312, "pos_frac": 0.765625, "sample": [135.34776306152344, -88.58706665039062, -175.48287963867188, 634.8250732421875, 581.1845703125, -161.6466064453125, 119.4254150390625, 355.97796630859375, -52.431640625, -22.639450073242188, 913.9177856445312, 80.236328125, 296.88946533203125, 141.746337890625, -22.0802001953125, 155.23098754882812, 41.8043212890625, 12.129150390625, 48.03038024902344, 70.97314453125, 210.26861572265625, -6.304779052734375, -302.0289306640625, 491.07513427734375, 525.9371948242188, 493.57415771484375, 335.7093811035156, 106.6713638305664, 165.39779663085938, 117.43304443359375, 135.61642456054688, 425.0440368652344, 21.09198570251465, 25.470611572265625, -13.3707275390625, 158.46875, -11.543354034423828, 223.92282104492188, 44.655609130859375, 255.25390625, 279.60382080078125, -21.373668670654297, -140.9959716796875, 161.33181762695312, 253.13836669921875, 294.7450866699219, -148.59609985351562, -34.031375885009766, 32.475341796875, 37.520362854003906, 211.26829528808594, 308.98406982421875, 214.01080322265625, -40.04075622558594, 168.07891845703125, 88.325439453125, 415.1814880371094, 65.12127685546875, -543.3193359375, 489.5573425292969, -306.55975341796875, 538.772705078125, 129.69927978515625, 319.632568359375, 369.0859680175781, 28.3548583984375, 428.44610595703125, 228.3751220703125, 559.6102294921875, 259.55963134765625, 16.1715087890625, 59.22434997558594, 100.70185852050781, 321.2579345703125, 366.32647705078125, 8.65234375, 78.24418640136719, 60.150909423828125, 100.47332763671875, 42.08155822753906, 66.04869079589844, 44.40180969238281, 184.83224487304688, 261.1953430175781, -156.11619567871094, -242.68304443359375, -39.15034484863281, 7.8262939453125, 8.77655029296875, 354.98114013671875, 484.18157958984375, 421.3968505859375, 266.55718994140625, 254.618408203125, 187.64642333984375, -210.14892578125, -89.37664794921875, 136.23257446289062, 256.126953125, 0.0, 250.4847412109375, 699.8006591796875, 369.31378173828125, 153.4654998779297, 238.14422607421875, 39.66120147705078, 99.9942626953125, 197.076904296875, 46.00236892700195, 141.49191284179688, 51.848388671875, 21.235092163085938, -10.53181266784668, -6.744697570800781, 252.68618774414062, 664.466064453125, 178.8408203125, 25.6954345703125, 169.80050659179688, -59.8597412109375, -96.52265930175781, -14.7603759765625, 434.0623779296875, -23.788265228271484, -103.28195190429688, 81.66650390625, 591.0859985351562, 58.89418411254883], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000428.npy"}
{"epoch": 0.8963350785340314, "step": 429, "batch_size": 128, "mean": 118.47213745117188, "std": 249.25392150878906, "min": -828.6990966796875, "p10": -119.27809906005858, "median": 85.24443435668945, "p90": 412.7733612060547, "max": 731.5309448242188, "pos_frac": 0.6796875, "sample": [12.07830810546875, -828.6990966796875, -352.9567565917969, 177.70204162597656, 423.720947265625, 251.91897583007812, 92.1702651977539, 47.3841552734375, 32.747406005859375, 417.1347961425781, -62.818077087402344, -216.34266662597656, -34.77447509765625, 117.8299560546875, 375.1895751953125, 279.19964599609375, 54.68585205078125, 731.5309448242188, -86.10516357421875, 137.108154296875, 172.4678497314453, 264.6518859863281, -85.9330825805664, 351.2284851074219, -94.38684844970703, 603.5053100585938, 104.25384521484375, 259.21905517578125, 96.68008422851562, 311.9803466796875, -181.47384643554688, 159.73919677734375, 98.95730590820312, 31.857208251953125, 462.8236999511719, 227.60562133789062, -135.79702758789062, 25.770751953125, 297.66595458984375, 616.140625, 281.2999267578125, 273.5337219238281, 55.27912902832031, -34.38141632080078, -12.683486938476562, 14.496170043945312, 218.24380493164062, 22.10150146484375, 60.99708557128906, -95.2503662109375, -2.802387237548828, 312.55072021484375, 161.94610595703125, 252.74374389648438, 69.39666748046875, 196.50042724609375, 694.5933227539062, -135.1651611328125, -99.69688415527344, 187.808349609375, -30.37677001953125, 118.66891479492188, 276.71990966796875, 556.193359375, -114.3828125, -724.5509643554688, 710.6978759765625, 174.220458984375, 256.428955078125, 300.7238464355469, -66.61236572265625, 163.24560546875, -343.67608642578125, -3.3761444091796875, 460.83453369140625, -445.16363525390625, 357.5002136230469, -28.679702758789062, 60.90374755859375, 209.02218627929688, 39.43357849121094, 0.0, 392.07965087890625, 406.3075866699219, 341.7453308105469, 155.397216796875, 237.4048614501953, -36.004295349121094, -23.603607177734375, 17.69913673400879, 330.64068603515625, -149.66787719726562, 407.92010498046875, -41.65858459472656, -115.14653015136719, 498.1680908203125, 8.849414825439453, -57.908241271972656, 61.878631591796875, -90.99935913085938, -128.91842651367188, 189.814697265625, -168.82720947265625, 170.2226104736328, -87.13427734375, 410.9041748046875, 78.318603515625, 444.177734375, -77.76856994628906, 220.57522583007812, 726.7264404296875, 287.59490966796875, -27.43707275390625, 323.9968566894531, 37.8643798828125, 56.90838623046875, 17.14126205444336, 77.23797607421875, 217.98056030273438, 356.505859375, 56.30393981933594, 173.083984375, -7.4214935302734375, -93.21832275390625, 184.21978759765625, -16.143142700195312, 67.00103759765625, -173.32498168945312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000429.npy"}
{"epoch": 0.8984293193717278, "step": 430, "batch_size": 128, "mean": 116.77705383300781, "std": 214.22494506835938, "min": -336.863525390625, "p10": -141.16856079101564, "median": 97.0669059753418, "p90": 385.2143371582031, "max": 723.4768676757812, "pos_frac": 0.6796875, "sample": [-100.57403564453125, -231.62445068359375, -108.73930358886719, 135.66004943847656, 19.274375915527344, -30.7052001953125, -142.69464111328125, 168.99420166015625, -12.73046875, -12.278457641601562, 95.9432373046875, -11.864700317382812, 308.17742919921875, -282.306884765625, 96.46090698242188, -185.75125122070312, 707.0806884765625, 273.6162109375, 28.193843841552734, 274.7235107421875, 110.05426025390625, -145.18177795410156, 717.265869140625, 130.330322265625, 213.04852294921875, -28.071746826171875, 140.679931640625, 0.0, 381.036376953125, -194.117919921875, 16.64892578125, 16.09527587890625, 272.4775390625, 147.9552764892578, 163.8924560546875, 0.0, 27.383468627929688, 50.07530212402344, 60.8475341796875, 259.96795654296875, -100.99041748046875, 332.5705871582031, 57.210540771484375, 84.46160888671875, 382.0103759765625, 29.564529418945312, -111.6339340209961, 177.10423278808594, -167.28375244140625, 194.1336669921875, 97.67290496826172, 143.40414428710938, 545.6918334960938, 113.71722412109375, 177.74166870117188, 231.281494140625, 129.08876037597656, -111.085205078125, 424.428466796875, -172.22930908203125, 463.92193603515625, 610.3174438476562, 452.55584716796875, 206.04669189453125, 96.05168151855469, 461.0200500488281, 239.9393310546875, 175.01144409179688, -2.890806198120117, 247.31536865234375, 147.788330078125, 332.1148986816406, 279.37939453125, 576.1150512695312, 46.03563690185547, 34.38724899291992, -140.5145263671875, 185.73858642578125, -60.08522033691406, 4.087432861328125, 213.8011474609375, 113.34771728515625, -1.3655166625976562, 184.3515167236328, -196.8575439453125, 39.7725830078125, 212.50283813476562, 392.69024658203125, 0.0, -57.29632568359375, 102.49321746826172, -108.0975341796875, 59.974884033203125, -87.927490234375, 723.4768676757812, -16.227529525756836, 236.67144775390625, 397.36883544921875, 10.992029190063477, -35.48638916015625, 622.2819213867188, 254.28387451171875, 154.8024444580078, -231.9872589111328, -35.604827880859375, 52.25184631347656, 204.19151306152344, -40.131874084472656, 370.78558349609375, 374.5653076171875, -336.863525390625, -140.113525390625, 379.32208251953125, 9.492950439453125, 244.78024291992188, 132.62152099609375, 42.02867126464844, 234.35772705078125, -8.519683837890625, -283.2018737792969, -46.91851806640625, -204.28610229492188, 186.9200439453125, 238.96395874023438, 183.1710968017578, 36.378692626953125, -74.62338256835938, 303.8908386230469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000430.npy"}
{"epoch": 0.900523560209424, "step": 431, "batch_size": 128, "mean": 97.6541976928711, "std": 207.04298400878906, "min": -626.1039428710938, "p10": -129.17711486816407, "median": 72.29627990722656, "p90": 376.2922729492187, "max": 728.6592407226562, "pos_frac": 0.7109375, "sample": [-62.95158386230469, -184.4791259765625, 235.63101196289062, 7.200347900390625, 34.8712158203125, 317.90899658203125, -79.83547973632812, 141.0445556640625, -116.55718994140625, 24.1837158203125, 34.11981201171875, -13.349456787109375, 86.2489013671875, 68.52224731445312, -316.358154296875, 381.2210693359375, -138.121337890625, 404.9156494140625, 243.91635131835938, 11.412315368652344, -33.785888671875, 152.39950561523438, 23.542449951171875, -200.79348754882812, 84.60311889648438, 78.2696533203125, -126.85699462890625, -626.1039428710938, 350.29461669921875, 389.13751220703125, 136.271240234375, 173.94996643066406, 47.73258972167969, -175.74395751953125, -320.21533203125, 716.4694213867188, -134.59072875976562, -13.6324462890625, -147.20989990234375, -34.19891357421875, 109.99698638916016, -139.71856689453125, 15.222648620605469, 137.25421142578125, 316.00640869140625, 366.48687744140625, 103.186279296875, -26.749588012695312, -24.788219451904297, -239.6585693359375, -89.38324737548828, 75.41552734375, 19.43658447265625, 73.74102783203125, -15.252593994140625, 155.32644653320312, 93.74710083007812, 269.34295654296875, 185.357666015625, 38.387603759765625, 28.0667724609375, 129.51441955566406, 366.3160400390625, 76.45758056640625, 87.34722900390625, 24.549468994140625, 155.192626953125, 135.60430908203125, -114.8561019897461, -108.72457885742188, 665.2364501953125, 37.71100997924805, 105.5029296875, 303.649658203125, 143.85617065429688, 323.9884338378906, 63.83306884765625, 415.0445251464844, 450.67010498046875, 444.3216552734375, 253.1798095703125, 555.475830078125, 89.99397277832031, -205.46331787109375, 6.221473693847656, -1.768768310546875, 268.506591796875, 271.2270812988281, -65.58425903320312, 47.85560607910156, 473.4996337890625, 33.19096755981445, 154.08294677734375, 95.59226989746094, 10.646430969238281, 618.2240600585938, 78.02569580078125, 374.179931640625, 107.9705810546875, 70.85153198242188, 393.305908203125, 127.78803253173828, 1.01348876953125, 67.3753662109375, 84.00164794921875, 33.762367248535156, 272.94818115234375, 262.4693603515625, 31.715023040771484, -110.72769165039062, 67.74955749511719, -17.895706176757812, 145.474853515625, 223.32809448242188, -2.474365234375, -1.598358154296875, -62.46907043457031, -5.9443359375, -34.449974060058594, 121.09623718261719, -170.71363830566406, -44.28076171875, 34.79380798339844, 195.94949340820312, 19.895660400390625, 728.6592407226562, 188.78228759765625, 138.55328369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000431.npy"}
{"epoch": 0.9026178010471204, "step": 432, "batch_size": 128, "mean": 119.30191040039062, "std": 230.1117706298828, "min": -443.884765625, "p10": -115.4490768432617, "median": 88.28446960449219, "p90": 419.079296875, "max": 969.556884765625, "pos_frac": 0.6875, "sample": [15.852813720703125, 19.6170654296875, 136.6844482421875, 629.2421264648438, 377.46099853515625, 230.4959716796875, 137.78067016601562, -0.2250518798828125, -416.326904296875, -295.52630615234375, 969.556884765625, 108.24661254882812, 281.17303466796875, 38.223876953125, 43.191619873046875, 49.10002899169922, 95.03073120117188, 416.82904052734375, 105.07022094726562, 23.93305206298828, 62.216163635253906, -207.04217529296875, -214.48226928710938, -165.79185485839844, -443.884765625, -10.24748420715332, -59.9832763671875, 281.074951171875, -15.448432922363281, 222.77166748046875, 119.892578125, 493.6231384277344, 244.38339233398438, 419.2845458984375, -295.20794677734375, -128.9495086669922, 392.15167236328125, -45.710540771484375, 15.9478759765625, -34.172393798828125, 61.316078186035156, 163.3389892578125, -138.93450927734375, 11.710660934448242, -46.36243438720703, 65.39347839355469, 202.7933349609375, 176.65301513671875, -273.57635498046875, 25.9468994140625, 205.90589904785156, 156.39437866210938, -78.31455993652344, 710.640625, -2.2958984375, -81.021484375, -7.597511291503906, 128.6045379638672, 75.92669677734375, 15.641502380371094, 82.41311645507812, 23.970535278320312, 303.4541015625, 4.145843505859375, 230.52081298828125, 22.496139526367188, 304.6413269042969, 484.6041259765625, 458.06768798828125, 156.12493896484375, -55.631004333496094, 115.43280029296875, 515.1110229492188, -33.2567138671875, 31.00164794921875, 508.65533447265625, 252.3706817626953, 212.8485107421875, 518.96826171875, 418.9913330078125, -11.44891357421875, 144.77120971679688, 367.5568542480469, -109.66317749023438, 219.91885375976562, -14.302490234375, 162.3233642578125, 217.9579315185547, -4.9630126953125, 294.06927490234375, -2.52392578125, -188.33860778808594, 0.0, 275.701171875, 74.81842041015625, -104.53948974609375, -151.05960083007812, -11.25885009765625, -25.023956298828125, -26.201171875, 141.03616333007812, -38.46342086791992, -76.03948974609375, 206.285888671875, 119.18382263183594, 1.6065597534179688, 310.7704772949219, 244.6162109375, 184.25738525390625, 170.915771484375, 4.58441162109375, 306.1451110839844, 99.37603759765625, 680.9219360351562, -134.83340454101562, 122.177978515625, 605.13037109375, 352.86529541015625, 20.851369857788086, 94.15582275390625, 130.2377471923828, 147.8387908935547, -74.86676025390625, 32.77338409423828, 795.6463012695312, 159.1534423828125, 98.10322570800781, -64.50961303710938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000432.npy"}
{"epoch": 0.9047120418848168, "step": 433, "batch_size": 128, "mean": 105.96646118164062, "std": 198.00413513183594, "min": -301.12579345703125, "p10": -121.10669784545897, "median": 96.64894104003906, "p90": 331.7701721191406, "max": 940.6431884765625, "pos_frac": 0.734375, "sample": [12.36442756652832, -233.71755981445312, 3.30096435546875, 379.8904113769531, 37.45695495605469, 5.06439208984375, 227.94776916503906, -47.70526123046875, -9.253694534301758, 98.99429321289062, 263.32269287109375, -33.576263427734375, 221.45838928222656, 175.78536987304688, 150.73300170898438, 57.7991943359375, -164.7950439453125, 132.025634765625, 162.227783203125, 150.0389404296875, 60.00482177734375, 331.0362854003906, 121.05046081542969, -24.355300903320312, 9.404129028320312, -105.81634521484375, 204.49658203125, 46.6769905090332, 497.2608947753906, 155.3575439453125, 174.31103515625, 33.223594665527344, 88.42828369140625, 143.63037109375, -68.0125732421875, 115.7156982421875, 302.16961669921875, 101.80641174316406, -147.51504516601562, 204.8560791015625, -134.812255859375, -108.06126403808594, 639.6072998046875, 434.6363830566406, 40.926483154296875, 240.5609130859375, -70.937744140625, -98.45480346679688, 94.3035888671875, 54.62565994262695, 206.10833740234375, 195.5757598876953, 333.4825744628906, 80.48410034179688, 102.61428833007812, -163.56680297851562, 319.2049255371094, 246.66943359375, 149.90745544433594, 23.721223831176758, -219.2447509765625, 26.205520629882812, 195.97503662109375, 123.40975952148438, 223.41969299316406, 57.51953125, 174.98971557617188, -258.00994873046875, 268.79315185546875, 247.43881225585938, -89.60360717773438, 17.74022674560547, 147.05844116210938, -33.365516662597656, 82.85211181640625, 123.29855346679688, 135.08038330078125, 180.82012939453125, 114.7397689819336, -205.09442138671875, -41.131011962890625, -182.61972045898438, 288.0062561035156, 126.85903930664062, 45.89483642578125, -301.12579345703125, 187.97181701660156, -136.30673217773438, 633.7667236328125, 66.27801513671875, 79.6884765625, -19.98863983154297, -105.83856201171875, -28.314987182617188, 104.5478744506836, 81.41670989990234, 176.27383422851562, 434.31170654296875, 233.15054321289062, -60.23004150390625, 151.2678680419922, -65.7321548461914, 8.224601745605469, 940.6431884765625, 71.0638427734375, 6.375701904296875, 654.5936279296875, 324.5751953125, 580.29833984375, 21.391895294189453, 348.8244323730469, 112.62896728515625, -46.6636962890625, -62.99464416503906, -268.2323303222656, -73.95770263671875, 253.8302459716797, 127.1297836303711, 165.91793823242188, 364.9728088378906, 212.50228881835938, 156.8995361328125, -183.11007690429688, -115.2328872680664, 45.35624694824219, 47.734588623046875, 411.0120849609375, 58.06613540649414], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000433.npy"}
{"epoch": 0.9068062827225131, "step": 434, "batch_size": 128, "mean": 99.86966705322266, "std": 217.83363342285156, "min": -725.1683959960938, "p10": -116.2211441040039, "median": 86.58119201660156, "p90": 345.6726379394531, "max": 1176.657958984375, "pos_frac": 0.6796875, "sample": [-36.183380126953125, 31.689651489257812, -13.52978515625, 56.44004821777344, 188.58590698242188, 241.91136169433594, 111.5260238647461, 189.32696533203125, 76.12139892578125, 313.633056640625, -97.46382904052734, 343.29144287109375, 61.4117431640625, 12.200225830078125, -183.66250610351562, 19.1038818359375, -96.00161743164062, 329.69482421875, -80.73739624023438, -9.469879150390625, 217.20657348632812, 662.5460205078125, 6.55120849609375, 476.74951171875, 283.2589111328125, -253.25347900390625, 0.0, -92.66964721679688, -101.238525390625, 27.286529541015625, 428.5082092285156, 208.03289794921875, 13.771270751953125, 105.15658569335938, 119.56781005859375, -155.84747314453125, 77.8326416015625, 159.39898681640625, 13.7010498046875, 303.6719055175781, 177.5953369140625, 495.44140625, 223.347412109375, -297.61541748046875, 125.63975524902344, -150.47525024414062, -3.59649658203125, 369.1103820800781, 193.20602416992188, 195.7281494140625, 140.89248657226562, 376.3406982421875, -56.917442321777344, 284.697265625, 281.78338623046875, 82.78115844726562, 89.91484069824219, -48.022247314453125, 87.83444213867188, 135.13296508789062, 142.96926879882812, 48.778961181640625, 265.61517333984375, -32.45535659790039, -59.51171875, -152.99716186523438, -226.61329650878906, 359.2884521484375, 41.537841796875, -40.07654571533203, 264.3365478515625, -144.79246520996094, 23.36199188232422, 239.37765502929688, -112.12457275390625, 156.57977294921875, 228.44033813476562, 589.9133911132812, 475.0576477050781, 132.252685546875, 177.97198486328125, -273.6208190917969, 164.42428588867188, 381.7718200683594, 138.425537109375, 209.04669189453125, 220.91864013671875, -725.1683959960938, 120.4239501953125, -122.859619140625, 166.9496612548828, -113.37608337402344, -105.01748657226562, 0.0, 217.7794189453125, 164.79031372070312, -86.60455322265625, 1176.657958984375, 297.3487243652344, 43.878055572509766, 234.32443237304688, 10.626113891601562, -76.16290283203125, 277.0296630859375, -14.063236236572266, 72.38330078125, 5.933845520019531, -158.04452514648438, 356.2061767578125, 170.10963439941406, 100.0728759765625, 5.94300651550293, 0.0, 304.1312255859375, 16.890716552734375, -47.48681640625, -19.193328857421875, 227.62619018554688, 133.6422119140625, -31.749755859375, -340.20751953125, 62.48826599121094, 85.32794189453125, -56.04139709472656, -96.1177978515625, 351.228759765625, 129.96571350097656, 164.83731079101562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000434.npy"}
{"epoch": 0.9089005235602095, "step": 435, "batch_size": 128, "mean": 136.4564971923828, "std": 235.37542724609375, "min": -381.3842468261719, "p10": -49.927670288085935, "median": 103.95057678222656, "p90": 339.0746948242187, "max": 1475.905029296875, "pos_frac": 0.7578125, "sample": [269.0457763671875, 102.53558349609375, -32.78942108154297, 172.7359161376953, 127.56246948242188, 544.0608520507812, 329.3450927734375, -19.33489990234375, 144.4817657470703, -4.0255126953125, 55.00881576538086, 11.289543151855469, -381.3842468261719, 232.24331665039062, 48.13477325439453, 4.20513916015625, 87.98486328125, 140.90338134765625, 709.7562255859375, 197.05328369140625, 149.74163818359375, 9.653020858764648, 419.1601257324219, 0.0, -4.907346725463867, -77.564453125, 96.96221923828125, 154.80599975585938, 194.3770751953125, 22.68236541748047, 185.84007263183594, 209.3509979248047, 171.4267578125, 75.17008209228516, -0.9543228149414062, 367.48529052734375, 213.13671875, 327.6507873535156, 179.15457153320312, -96.88079833984375, 321.61981201171875, 234.82894897460938, 273.6864013671875, 82.33740234375, 234.1329803466797, -5.234130859375, 216.79843139648438, -107.57427978515625, 199.6837158203125, 105.36557006835938, 90.65411376953125, 1067.3984375, 222.05807495117188, 417.15850830078125, -3.419811248779297, 198.96678161621094, 41.776580810546875, 6.8162689208984375, 173.10662841796875, -53.40583801269531, -25.957290649414062, 319.35955810546875, 85.4039306640625, -4.1094512939453125, 83.17784881591797, 16.268798828125, 392.0413513183594, 124.46798706054688, 7.232099533081055, 238.7481231689453, 78.1656494140625, -16.7252197265625, 119.81658935546875, 361.777099609375, 215.90890502929688, 109.99136352539062, 183.8507843017578, 157.4166259765625, -48.43702697753906, 102.1307373046875, -276.35394287109375, 316.146484375, 1150.1953125, 250.3305206298828, -200.370849609375, 18.923828125, 218.76422119140625, -1.8836383819580078, -10.138664245605469, 161.7279052734375, -57.291019439697266, 112.7491455078125, 128.26080322265625, 1475.905029296875, 117.69479370117188, -44.04167175292969, 267.441650390625, -152.25009155273438, -172.7530517578125, 405.5081787109375, 79.65577697753906, 369.37548828125, 4.736457824707031, -43.735374450683594, 294.3888244628906, -336.39263916015625, 56.268707275390625, 70.48043823242188, 171.2853546142578, 40.65740203857422, -18.386550903320312, 170.31668090820312, 54.2000732421875, 113.41429138183594, 27.43798828125, 433.4150390625, 44.156524658203125, 57.129638671875, -42.784423828125, 40.90228271484375, 93.4918212890625, 211.70816040039062, 133.35714721679688, -106.90812683105469, -67.83071899414062, 123.35415649414062, 78.62872314453125, 153.1573486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000435.npy"}
{"epoch": 0.9109947643979057, "step": 436, "batch_size": 128, "mean": 96.58380889892578, "std": 255.36936950683594, "min": -420.2515869140625, "p10": -153.23690490722655, "median": 68.81588745117188, "p90": 348.11074523925777, "max": 1940.321533203125, "pos_frac": 0.6953125, "sample": [10.6651611328125, -317.0194091796875, 417.7663269042969, -54.067230224609375, 3.190662384033203, -34.399932861328125, 128.36837768554688, -35.94080352783203, -92.46380615234375, 1940.321533203125, -112.42999267578125, 272.5746765136719, 118.7582778930664, 135.947998046875, -157.6156005859375, 662.2889404296875, 146.17279052734375, -145.97813415527344, -282.4299011230469, -336.17254638671875, -87.30795288085938, 42.342247009277344, 102.05076599121094, 288.62591552734375, 439.254638671875, -11.051681518554688, 150.23773193359375, 57.7574462890625, 316.72113037109375, 32.7392578125, 228.10308837890625, 40.64396667480469, -34.638519287109375, 58.34393310546875, 215.50389099121094, 383.90771484375, 122.2999267578125, 258.3000183105469, 185.8724365234375, 63.924102783203125, 245.19155883789062, -181.49974060058594, 199.21185302734375, 73.7642822265625, 173.78689575195312, 55.91424560546875, 15.238723754882812, -261.9896545410156, -69.29840087890625, 446.1099853515625, 231.1046142578125, -94.10263061523438, -65.4986572265625, -109.829833984375, 163.53701782226562, 13.804710388183594, 334.8750915527344, -223.82662963867188, 46.3441162109375, 436.6524963378906, 290.21661376953125, 100.87454223632812, 68.48934936523438, -77.23079681396484, 303.8534240722656, 327.79620361328125, 15.569549560546875, -147.10638427734375, 160.23123168945312, 501.0032653808594, 21.055328369140625, 362.06573486328125, 273.9284362792969, 78.41940307617188, 164.67410278320312, 66.93194580078125, 41.98383712768555, 74.02639770507812, 68.43679809570312, 17.633392333984375, 396.6090087890625, 69.14242553710938, 97.96243286132812, -3.8499374389648438, 157.86767578125, 10.43145751953125, 89.93342590332031, 110.77445220947266, 55.79264831542969, -174.66500854492188, 725.08203125, 151.36924743652344, -13.781705856323242, -312.4217529296875, -62.41009521484375, 92.1416015625, 302.2977294921875, -73.42376708984375, -15.670848846435547, -420.2515869140625, -12.839019775390625, -182.78521728515625, 73.8016357421875, 155.61349487304688, 168.70448303222656, 81.29180908203125, -256.54486083984375, 129.7352752685547, 467.5821533203125, 5.330085754394531, -151.36032104492188, 117.92911529541016, -1.374481201171875, 342.1300354003906, -100.42362976074219, 194.2985382080078, -17.0030517578125, 452.6247253417969, 104.8495101928711, 132.3382568359375, 92.91217041015625, 98.3773193359375, -11.582427978515625, 49.474365234375, -177.821044921875, 144.96304321289062, 8.071762084960938, 7.9947509765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000436.npy"}
{"epoch": 0.9130890052356021, "step": 437, "batch_size": 128, "mean": 142.06787109375, "std": 239.6721954345703, "min": -517.35791015625, "p10": -130.35510482788084, "median": 113.74642944335938, "p90": 422.1601409912107, "max": 1017.4478759765625, "pos_frac": 0.7421875, "sample": [718.103759765625, 83.0281982421875, 160.86810302734375, 182.8577880859375, -137.02020263671875, 1017.4478759765625, -21.13330078125, 169.1641845703125, 227.58740234375, 103.53076171875, 555.2919311523438, 179.453369140625, 79.99267578125, 177.9222412109375, -18.342559814453125, 359.4617919921875, -111.13311767578125, 108.40034484863281, 106.1820068359375, 104.05953979492188, 81.74755859375, 818.2479248046875, 822.0289306640625, 476.43994140625, 114.84324645996094, 392.91278076171875, -27.096637725830078, 126.87689208984375, 140.9080810546875, 85.95414733886719, -54.42626953125, 183.510986328125, 39.59661865234375, 68.95198059082031, 322.3904113769531, -4.157665252685547, -222.36334228515625, 306.03045654296875, -84.83021545410156, 61.35125732421875, 497.564208984375, 262.17852783203125, 113.96826171875, 515.0269775390625, -186.79977416992188, 35.77911376953125, 67.80288696289062, 0.324951171875, 210.230712890625, 290.393798828125, 142.70599365234375, 398.8973693847656, -89.36432647705078, 387.6139831542969, 8.198360443115234, -25.21820068359375, -259.5262451171875, 2.400543212890625, 573.7802734375, 221.41470336914062, 59.749053955078125, 63.63201904296875, 199.43157958984375, 337.98541259765625, 303.477783203125, -229.55499267578125, -11.866249084472656, 348.3028869628906, -166.38253784179688, -127.4986343383789, -1.712066650390625, -29.720306396484375, -517.35791015625, 76.39398193359375, -37.439483642578125, 506.6702880859375, 9.56501579284668, 180.4794921875, -217.82977294921875, 0.0, 260.35626220703125, -139.1776580810547, 99.79441833496094, 166.43014526367188, 250.21011352539062, 140.28768920898438, -119.147216796875, 197.14927673339844, 23.77978515625, 229.96218872070312, 295.886962890625, 141.21682739257812, 6.760061264038086, 38.647918701171875, -105.84503173828125, -20.79473876953125, 209.40261840820312, 319.03363037109375, 247.93902587890625, 61.864891052246094, 397.864990234375, 247.913330078125, 230.47607421875, 610.5308837890625, 225.32254028320312, 80.59414672851562, 256.1096496582031, -174.5799560546875, 640.8970336914062, -329.7155456542969, 38.170623779296875, -296.7611083984375, 168.015869140625, -310.9708251953125, 109.01266479492188, 306.45452880859375, 15.94207763671875, 292.1885070800781, 113.52459716796875, 515.8602294921875, 23.010360717773438, -38.4156494140625, 285.8160705566406, 228.1292724609375, 0.0, 159.56472778320312, 180.95416259765625, 296.68255615234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000437.npy"}
{"epoch": 0.9151832460732985, "step": 438, "batch_size": 128, "mean": 60.66808319091797, "std": 221.74827575683594, "min": -848.633056640625, "p10": -178.46074829101562, "median": 39.53957557678223, "p90": 301.2504028320312, "max": 939.202392578125, "pos_frac": 0.59375, "sample": [221.23446655273438, -328.726806640625, 210.60687255859375, -100.18344116210938, -35.67641067504883, 268.6748352050781, 369.9138488769531, 197.42221069335938, 0.0, 33.20228958129883, 522.0723876953125, -197.97979736328125, 88.29393768310547, 231.66993713378906, 48.7152099609375, 94.46339416503906, 69.4283447265625, 4.323974609375, 144.38555908203125, -62.67589569091797, -215.98947143554688, -78.23980712890625, -167.1383819580078, -66.48746490478516, 109.11154174804688, -64.32469177246094, 177.9468994140625, 85.75797271728516, -39.752197265625, 203.122314453125, 164.16603088378906, 673.571533203125, 136.18695068359375, 136.68435668945312, 217.00909423828125, 661.9932861328125, -164.0329132080078, -95.05715942382812, 51.965789794921875, -140.33282470703125, -231.81369018554688, -171.3348846435547, -32.281005859375, -45.707359313964844, 203.78561401367188, -78.17303466796875, 3.03533935546875, 17.02496337890625, -266.3961181640625, 13.759979248046875, -11.094970703125, -298.5705261230469, 1.3303680419921875, -79.68423461914062, -28.3604736328125, 187.76275634765625, 94.387939453125, 232.47210693359375, 86.138916015625, 124.16459655761719, 58.95201873779297, 17.9029541015625, -11.951133728027344, -185.27053833007812, 13.837692260742188, -25.764209747314453, 100.2935791015625, 94.45357513427734, 245.0732879638672, -131.5266571044922, 175.2955322265625, -30.94489288330078, 401.4435729980469, -175.54226684570312, 383.3270263671875, 504.711669921875, -90.20819091796875, -124.03844451904297, 303.9696960449219, 235.0032196044922, -198.23526000976562, 224.39639282226562, -81.83750915527344, -168.04080200195312, 316.12152099609375, -370.87762451171875, 216.4786376953125, 313.0904541015625, -848.633056640625, 225.8153839111328, 217.9440155029297, 117.65270233154297, 132.33248901367188, 73.36599731445312, 165.81207275390625, 195.59478759765625, 47.36177062988281, 504.1960754394531, 329.71337890625, -200.6802978515625, -34.32624053955078, -83.0704345703125, 184.9736328125, -101.87037658691406, -2.6893844604492188, -146.5772705078125, 30.109893798828125, -26.987350463867188, -136.29058837890625, 130.58575439453125, 300.0849914550781, 9.557588577270508, -134.98886108398438, 213.7799835205078, 0.0, 25.385040283203125, -7.605344772338867, 155.49505615234375, -205.2851104736328, -8.370925903320312, 220.7596435546875, -247.50804138183594, 7.669092178344727, 45.876861572265625, 939.202392578125, 144.93052673339844, 299.94927978515625, 136.36431884765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000438.npy"}
{"epoch": 0.9172774869109948, "step": 439, "batch_size": 128, "mean": 119.6225814819336, "std": 197.16513061523438, "min": -471.4423522949219, "p10": -101.29385375976562, "median": 113.60671997070312, "p90": 349.870947265625, "max": 684.6478271484375, "pos_frac": 0.734375, "sample": [-179.0015869140625, -97.4117431640625, -224.55557250976562, 631.4411010742188, -9.10125732421875, 141.35757446289062, 319.84649658203125, 62.06946563720703, 301.9093322753906, 315.1509704589844, -17.570539474487305, 116.45114135742188, -77.07294464111328, 201.9698486328125, 320.993896484375, -88.443359375, 10.56256103515625, 216.74928283691406, 15.969039916992188, -100.67120361328125, 235.88916015625, 263.92266845703125, 233.21026611328125, 321.2669372558594, 48.434295654296875, 301.768798828125, 212.26708984375, 246.49398803710938, 121.62374877929688, 57.90766906738281, -76.76419830322266, 122.99879455566406, -45.17197799682617, 422.3194885253906, 144.69064331054688, 167.18655395507812, 408.48065185546875, 211.07931518554688, 556.4860229492188, 106.42626953125, -64.65829467773438, 63.353519439697266, 416.24749755859375, 432.9660949707031, -102.7467041015625, 440.72894287109375, 193.375244140625, 80.6031494140625, -33.94036865234375, 110.76229858398438, -278.81475830078125, 124.8458023071289, -303.95819091796875, 182.37899780273438, 342.2499084472656, 304.4603271484375, 266.54766845703125, -107.50220489501953, 44.12966537475586, 0.0, 92.80557250976562, 0.0, 23.88446044921875, -55.378143310546875, 266.97808837890625, 567.1774291992188, 104.83233642578125, 101.9296875, -68.75498962402344, -17.4219970703125, 62.85693359375, 295.54205322265625, -147.80581665039062, -185.39218139648438, -53.02727508544922, 168.20724487304688, 70.41393280029297, 89.2757568359375, -471.4423522949219, 85.80438232421875, -212.71697998046875, 137.59466552734375, 68.05323791503906, 17.901824951171875, 344.9901428222656, 56.38446044921875, 235.673828125, 14.716156005859375, -295.75091552734375, 177.24229431152344, 211.4571990966797, -175.058837890625, 0.0, 234.8541259765625, 364.58892822265625, 244.66329956054688, 361.2594909667969, 215.37086486816406, 151.98385620117188, 81.19610595703125, 11.11431884765625, 31.547882080078125, -65.25994873046875, 16.446533203125, 137.483642578125, 87.55215454101562, 92.9228515625, 271.36199951171875, 133.12107849121094, 222.140380859375, 47.709075927734375, 191.80047607421875, 190.1219482421875, 617.6732177734375, 240.42408752441406, 0.0, -250.69491577148438, 265.43023681640625, 35.1904296875, 684.6478271484375, 368.77459716796875, 121.08268737792969, 173.32687377929688, -59.162109375, 268.0687255859375, 0.0, 136.24508666992188, 145.54537963867188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000439.npy"}
{"epoch": 0.9193717277486911, "step": 440, "batch_size": 128, "mean": 94.38501739501953, "std": 190.95184326171875, "min": -395.96478271484375, "p10": -107.93103942871093, "median": 78.67613220214844, "p90": 327.8423278808594, "max": 731.3079833984375, "pos_frac": 0.6953125, "sample": [-277.74505615234375, 0.0, 120.9764404296875, -79.27371215820312, 19.81171226501465, 29.6048583984375, -13.23785400390625, 82.58934020996094, -250.0845184326172, 132.9054718017578, 178.5010986328125, -29.99407958984375, -39.8734130859375, 22.0023193359375, 185.68609619140625, 123.01431274414062, 19.098052978515625, 291.20770263671875, 86.70011138916016, 207.14544677734375, 160.53155517578125, 165.58963012695312, -12.86517333984375, 34.233795166015625, 54.43896484375, -106.35824584960938, -395.96478271484375, 86.85433959960938, -301.9811096191406, 13.20770263671875, 407.16876220703125, 346.83892822265625, 64.95355987548828, -12.37469482421875, -342.25537109375, -55.44873046875, 269.5472717285156, -216.17437744140625, 0.0, 199.80792236328125, -293.9200439453125, 7.04534912109375, -17.17425537109375, 39.626529693603516, 282.8604431152344, -11.928535461425781, -97.81643676757812, 22.567604064941406, 65.60220336914062, 280.0906066894531, 95.28205871582031, -5.617530822753906, 261.6043701171875, -125.71511840820312, 114.2967529296875, 278.9894104003906, -10.319509506225586, 8.218269348144531, -111.60089111328125, 90.68092346191406, -349.0074768066406, 26.9512939453125, 338.8369445800781, -55.465179443359375, 310.4715576171875, 60.697174072265625, 691.3081665039062, 37.80708312988281, 133.615966796875, 157.70150756835938, 17.67320442199707, -44.116943359375, 127.07030487060547, 235.099609375, 376.69171142578125, -93.6068344116211, -73.44895935058594, -49.08427047729492, 156.87478637695312, 228.76446533203125, 54.7242431640625, -28.14862060546875, 324.78204345703125, 329.9543151855469, 344.3960876464844, -25.644775390625, 96.78292846679688, 184.525634765625, 109.30865478515625, 30.718069076538086, 291.796875, -13.216060638427734, 246.9962158203125, -29.1019287109375, 313.00054931640625, 73.26336669921875, -46.10960388183594, 10.923431396484375, 402.167236328125, 731.3079833984375, 151.1412353515625, 137.6383514404297, 40.00208282470703, 177.3483428955078, 155.1995849609375, 105.266357421875, 76.37045288085938, 610.7553100585938, 204.53024291992188, 242.0245361328125, 315.6678771972656, 220.78585815429688, -39.94757080078125, 327.781005859375, 119.58984375, 333.1204833984375, 410.42041015625, 19.4423828125, 283.25262451171875, 80.9818115234375, -113.1971206665039, 327.98541259765625, -222.83119201660156, 171.36874389648438, 24.744903564453125, 237.81129455566406, -140.96646118164062, 146.17794799804688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000440.npy"}
{"epoch": 0.9214659685863874, "step": 441, "batch_size": 128, "mean": 140.8197479248047, "std": 199.2602081298828, "min": -278.37872314453125, "p10": -93.2917915344238, "median": 128.06190490722656, "p90": 386.95143432617186, "max": 675.3955078125, "pos_frac": 0.75, "sample": [-56.72932434082031, 28.716552734375, -44.61566162109375, 251.59576416015625, -15.239208221435547, -278.37872314453125, 528.1326904296875, 395.60491943359375, 234.2294464111328, -0.28021240234375, -255.8509521484375, 575.6348876953125, 118.23888397216797, 251.9765625, 316.2091979980469, 34.720550537109375, 201.83474731445312, 244.650390625, 58.25574493408203, 222.62313842773438, 125.1854248046875, 201.57778930664062, 230.26260375976562, 112.02264404296875, 383.2427978515625, -270.7307434082031, 228.04541015625, 238.9711151123047, 127.61676025390625, 112.17501831054688, 360.75042724609375, 437.7034912109375, 86.41252136230469, -174.09182739257812, 126.43653869628906, 338.511962890625, 137.45635986328125, 327.29986572265625, 335.6526184082031, -135.3515625, 259.944580078125, 252.9063720703125, -83.47901916503906, -164.026611328125, 374.1959228515625, 5.871471405029297, 128.50704956054688, 166.6231231689453, 101.3150634765625, 175.35357666015625, 165.61911010742188, 95.92938232421875, 178.97711181640625, -17.634681701660156, 44.10292053222656, 129.56362915039062, -49.65362548828125, 27.75482177734375, 116.31663513183594, -152.21002197265625, -40.047332763671875, 15.119155883789062, -19.53191375732422, 423.09747314453125, 36.46038818359375, 0.0, 2.253509521484375, -30.436172485351562, 95.41858673095703, 351.5289611816406, 374.8861083984375, -159.67935180664062, 4.080665588378906, 71.41722869873047, 0.0, 85.11683654785156, 227.08010864257812, 224.72991943359375, 133.62632751464844, 510.0346984863281, 140.4752960205078, 185.3558349609375, 146.43991088867188, -254.33779907226562, 193.40567016601562, -84.21533203125, 199.72140502929688, 182.4384765625, -124.89299011230469, 110.6746826171875, 35.80206298828125, 134.48297119140625, 240.60214233398438, 251.89810180664062, 38.231746673583984, 70.29335021972656, 210.54652404785156, -111.96468353271484, 516.5721435546875, -33.59259033203125, 118.53814697265625, -214.931640625, 484.27117919921875, 272.1676330566406, -129.9127197265625, 64.77218627929688, 668.5455322265625, -82.58116149902344, 271.22576904296875, -85.28912353515625, 170.9968719482422, 205.9327392578125, 534.79443359375, 675.3955078125, 244.94729614257812, 263.3096923828125, 241.59375, 13.35614013671875, -62.86285400390625, -11.484405517578125, 557.5845336914062, 70.10917663574219, 187.45745849609375, 341.359375, 599.263671875, -22.209213256835938, 339.6231689453125, 59.4073486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000441.npy"}
{"epoch": 0.9235602094240838, "step": 442, "batch_size": 128, "mean": 96.5578842163086, "std": 222.5062713623047, "min": -512.4790649414062, "p10": -188.41709899902344, "median": 83.79436492919922, "p90": 372.7318908691406, "max": 769.5205078125, "pos_frac": 0.671875, "sample": [359.05023193359375, 116.60751342773438, 14.668045043945312, -156.22413635253906, 375.5509033203125, 97.47906494140625, -14.449277877807617, -224.62557983398438, 182.84210205078125, 6.015777587890625, 220.04197692871094, 169.47386169433594, -146.65676879882812, 0.4195709228515625, 466.1044616699219, -187.45437622070312, 260.2205810546875, 89.17132568359375, 247.70391845703125, -46.76970672607422, 35.66436767578125, -6.934440612792969, 60.8831787109375, -8.834232330322266, 476.6358642578125, -44.232330322265625, 122.47988891601562, -201.16748046875, 338.9325256347656, 317.9704284667969, -140.27716064453125, -16.3707275390625, 38.6077880859375, 325.1379699707031, 182.14669799804688, 119.70843505859375, -140.81932067871094, -190.6634521484375, -162.8802490234375, 113.98871612548828, 53.47938537597656, -71.33354187011719, 78.72019958496094, 174.7565460205078, 162.72283935546875, -512.4790649414062, 38.480743408203125, 240.173095703125, 77.65055084228516, -195.4974365234375, 122.21089172363281, 182.44467163085938, 187.21853637695312, 344.16754150390625, 197.17730712890625, 16.303466796875, 32.442779541015625, 0.0, -59.05793762207031, -4.1015472412109375, -272.106689453125, -91.79640197753906, -197.15769958496094, 150.04208374023438, -372.98175048828125, 254.96798706054688, 307.96136474609375, 84.711181640625, 42.31878662109375, 170.7239532470703, -32.36289978027344, -197.1341552734375, 425.711669921875, 168.27908325195312, 286.0482482910156, 255.98825073242188, 399.73565673828125, 167.4580078125, -113.8355712890625, 1.088897705078125, 69.30595397949219, 54.5029296875, -27.760486602783203, 769.5205078125, 234.37274169921875, 625.7177124023438, -261.77471923828125, -66.26882934570312, -157.3255157470703, 99.26676940917969, 432.823486328125, 25.783401489257812, -378.1188659667969, 641.2033081054688, -13.385879516601562, 131.09982299804688, 693.30322265625, 182.37213134765625, -169.7195281982422, 371.52374267578125, 379.3438415527344, 179.4366455078125, 293.8616027832031, 332.787353515625, 82.87754821777344, 611.2974243164062, -84.30712890625, -145.03009033203125, -313.28948974609375, 252.86376953125, 203.45501708984375, 169.29183959960938, -39.48846435546875, 260.3789978027344, 235.94363403320312, 56.86885070800781, 7.342376708984375, -17.4193115234375, 76.8809814453125, -209.4735107421875, 153.76019287109375, 180.15518188476562, 394.88385009765625, 106.00872802734375, -47.147003173828125, 39.801025390625, 120.89952087402344, 268.699462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000442.npy"}
{"epoch": 0.9256544502617801, "step": 443, "batch_size": 128, "mean": 108.38372802734375, "std": 202.13128662109375, "min": -596.646728515625, "p10": -116.56532287597655, "median": 88.5433578491211, "p90": 362.2712341308594, "max": 636.5155029296875, "pos_frac": 0.7265625, "sample": [129.3084716796875, 28.20001220703125, 449.0177001953125, 351.16021728515625, -6.380615234375, 215.0377197265625, 380.292236328125, -33.87042236328125, 411.5150146484375, -332.5407409667969, -9.31817626953125, 129.5577392578125, -70.1185073852539, 363.63555908203125, -216.85336303710938, -3.8720855712890625, -114.29879760742188, 85.06116485595703, 60.20310974121094, 37.26356506347656, 219.27423095703125, 181.35232543945312, 98.92683410644531, 242.152099609375, 2.2257080078125, 200.11058044433594, 331.47076416015625, 228.49732971191406, 427.324462890625, -39.872596740722656, -596.646728515625, -4.2835693359375, -341.267578125, 26.881370544433594, -50.16351318359375, 70.98880004882812, 284.908447265625, 57.540740966796875, 119.63021850585938, 68.51628112792969, 232.21868896484375, -200.3023681640625, 83.48902893066406, -0.46905517578125, 126.88557434082031, 181.7522735595703, 343.2362365722656, -157.92384338378906, 184.34915161132812, -47.030517578125, 157.87164306640625, -231.77877807617188, -14.804031372070312, 214.69873046875, 101.54165649414062, 361.6865234375, 33.80949401855469, 636.5155029296875, 103.02833557128906, 263.64422607421875, 14.667999267578125, 319.6830139160156, 90.5040283203125, 221.0053253173828, 463.85589599609375, 133.09414672851562, 21.185867309570312, -134.67373657226562, 186.91236877441406, 376.8653564453125, 28.856765747070312, -4.9033966064453125, 358.79046630859375, 136.47215270996094, 622.240966796875, 31.97406005859375, -73.71017456054688, 165.6297607421875, -53.8013916015625, -37.3802490234375, 170.4976043701172, 210.3277587890625, 161.52313232421875, 6.032497406005859, 153.40280151367188, -231.92825317382812, 153.79637145996094, -58.966651916503906, 170.81878662109375, 312.2510070800781, 228.0390625, 309.8223571777344, 86.58268737792969, 494.7953186035156, 79.80764770507812, -52.07164001464844, -29.47857666015625, 539.383056640625, 148.2040252685547, 258.4181213378906, 218.60572814941406, 64.29074096679688, -256.3655700683594, 150.8457794189453, 99.00189208984375, -121.8538818359375, 81.23941040039062, 83.55398559570312, 58.877777099609375, 38.23655700683594, 307.9104309082031, 165.8593292236328, -104.83572387695312, 63.12836456298828, 40.18284606933594, -73.14838409423828, 64.91461181640625, 435.722412109375, 86.3385009765625, -406.3600158691406, 300.138427734375, 356.9713134765625, 493.6461486816406, -170.31497192382812, 71.80776977539062, -58.836029052734375, 30.448379516601562, 121.599609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000443.npy"}
{"epoch": 0.9277486910994764, "step": 444, "batch_size": 128, "mean": 121.7252197265625, "std": 237.4546356201172, "min": -547.7760009765625, "p10": -142.28927307128907, "median": 93.37278366088867, "p90": 381.2238494873047, "max": 1145.28369140625, "pos_frac": 0.6953125, "sample": [-27.771080017089844, 193.651123046875, 231.78265380859375, -5.159423828125, 291.0748291015625, 151.84449768066406, -547.7760009765625, 59.2745361328125, 139.23452758789062, 147.31417846679688, -14.776397705078125, 133.24908447265625, -174.48388671875, 553.3042602539062, 43.578033447265625, 141.3023681640625, 226.35659790039062, 174.1328582763672, 0.58612060546875, 94.1043930053711, 283.25775146484375, 91.01089477539062, 199.76898193359375, -14.14801025390625, 4.05548095703125, -144.73187255859375, 301.2859802246094, 50.59235763549805, 31.9278564453125, 43.876129150390625, 177.28759765625, 213.72119140625, 383.7757263183594, -107.88360595703125, 641.8654174804688, -366.5152282714844, -243.84378051757812, 236.38616943359375, 292.82867431640625, 543.0963134765625, 143.479248046875, 327.59869384765625, -66.32720947265625, -31.198528289794922, 136.16690063476562, 269.6787109375, 92.64117431640625, 18.24828338623047, -45.1318359375, -142.01193237304688, 326.7650146484375, 121.02532958984375, -64.27296447753906, 418.4571228027344, -16.728893280029297, 378.071044921875, 380.13018798828125, 553.92236328125, 45.380615234375, 63.57666015625, 5.919572830200195, 800.2996215820312, 78.98426818847656, 12.980194091796875, 234.20751953125, -152.96240234375, 206.296875, 241.16647338867188, 206.0473175048828, 594.9519653320312, 308.1719055175781, 126.5577621459961, -149.1021728515625, 268.60675048828125, -39.44476318359375, -32.81658935546875, 462.8089294433594, 186.399169921875, 305.2027893066406, 121.875244140625, 1145.28369140625, -173.55300903320312, 12.505172729492188, 691.37841796875, 75.5302505493164, 405.4644775390625, 72.43321990966797, -46.629425048828125, 268.76171875, -172.126220703125, -249.82846069335938, 241.842041015625, -47.113014221191406, 116.63456726074219, 85.0555419921875, 23.994720458984375, -94.86367797851562, 272.9624328613281, 360.0057067871094, -142.9364013671875, -70.834228515625, 96.0983657836914, 492.5629577636719, 54.256126403808594, -71.63803100585938, 253.3984375, 17.724340438842773, 212.1042938232422, -78.59811401367188, 379.4290466308594, -22.316570281982422, 42.0181884765625, -89.12557983398438, 200.43759155273438, 137.76626586914062, -77.18370819091797, 284.32501220703125, 351.26556396484375, -76.19303894042969, -17.548439025878906, -8.896238327026367, 86.83824157714844, 0.301025390625, -84.10272216796875, -308.0065612792969, 99.642822265625, 226.9690704345703, -418.72650146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000444.npy"}
{"epoch": 0.9298429319371728, "step": 445, "batch_size": 128, "mean": 111.25466918945312, "std": 234.02490234375, "min": -662.5330810546875, "p10": -119.03953628540039, "median": 106.00527954101562, "p90": 389.15179748535155, "max": 1239.1796875, "pos_frac": 0.703125, "sample": [157.68447875976562, -329.09820556640625, 391.64215087890625, -65.80145263671875, 158.78802490234375, 762.4583129882812, -58.34489440917969, 98.80437469482422, 307.3232421875, -321.3648986816406, 132.40243530273438, 477.377685546875, 76.48790740966797, -130.1322021484375, 310.3798522949219, 12.650726318359375, 48.90753173828125, 180.1715850830078, 269.475341796875, 53.600555419921875, 1239.1796875, 2.152587890625, 94.6611328125, -119.83563995361328, 4.48712158203125, 215.4472198486328, 324.7751159667969, -5.311809539794922, 323.6043701171875, 160.16221618652344, 234.91827392578125, 173.6876220703125, 4.66510009765625, 102.61505126953125, -662.5330810546875, -118.69834899902344, 21.263774871826172, 29.011512756347656, 413.0308837890625, 143.62173461914062, 4.5289306640625, 228.21595764160156, 458.7999572753906, 11.11767578125, 201.67474365234375, 29.782352447509766, 75.38670349121094, 298.55926513671875, 26.46442413330078, -80.52890014648438, -18.83609390258789, -107.83828735351562, 154.323974609375, -20.641952514648438, 299.6154479980469, -16.409027099609375, 152.70562744140625, 173.0433349609375, -58.17132568359375, 593.1211547851562, 144.3035888671875, 234.49826049804688, 436.99951171875, 172.52352905273438, 162.45208740234375, 388.0845031738281, 127.69270324707031, -529.3507080078125, 432.0436706542969, 369.18988037109375, 130.069091796875, 172.3131103515625, -69.45565795898438, 159.6982421875, 74.19828796386719, 171.4521026611328, 11.692626953125, -78.52337646484375, 219.23553466796875, 141.41233825683594, 109.3955078125, -3.3465423583984375, -240.30624389648438, 14.34051513671875, 56.77579116821289, -120.5909423828125, 142.55531311035156, -6.360809326171875, 222.99420166015625, -26.687393188476562, -51.019195556640625, 130.81329345703125, 129.30938720703125, 72.49691772460938, 124.28018188476562, 157.38494873046875, -12.702293395996094, 95.3092041015625, 258.21099853515625, -0.927001953125, 14.169975280761719, -7.6107177734375, 375.0286865234375, 198.4004669189453, -70.64337158203125, 111.9681396484375, 11.152938842773438, 131.01852416992188, 544.31103515625, -41.16705322265625, -160.75439453125, 117.52555847167969, 0.0, -17.068397521972656, 141.51040649414062, -225.68984985351562, 163.0787353515625, 189.46270751953125, 55.6551513671875, 456.8674011230469, -177.10675048828125, -187.059326171875, -215.50927734375, -49.14314270019531, 224.95269775390625, 475.12945556640625, -34.77565002441406, 775.2074584960938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000445.npy"}
{"epoch": 0.9319371727748691, "step": 446, "batch_size": 128, "mean": 121.28301239013672, "std": 225.9171142578125, "min": -411.86328125, "p10": -109.53175811767578, "median": 112.75277709960938, "p90": 379.1654541015625, "max": 1327.970703125, "pos_frac": 0.7109375, "sample": [413.7630310058594, 342.2391662597656, 182.15478515625, -35.302978515625, 34.77752685546875, -22.785293579101562, 176.2314453125, -55.426658630371094, 109.13288879394531, -411.86328125, -49.576812744140625, 134.50204467773438, 92.39009094238281, -16.427642822265625, 111.34991455078125, 27.319961547851562, 257.88995361328125, 6.957649230957031, -188.11968994140625, 17.49120330810547, -92.06912231445312, -21.472900390625, 178.3739013671875, 490.6097412109375, 97.5293960571289, -111.00694274902344, 402.88360595703125, -18.96672821044922, -40.60169982910156, 146.93551635742188, -6.733428955078125, 23.361358642578125, -233.78515625, 73.77617645263672, -357.76214599609375, 111.25179290771484, 425.1893310546875, 117.36177062988281, 252.10302734375, 80.04561614990234, 283.2474365234375, 283.92877197265625, 118.04222106933594, -14.2777099609375, 33.32586669921875, 251.99542236328125, 102.01239013671875, 481.9573059082031, 378.9583435058594, -379.4884338378906, 133.70877075195312, 115.64462280273438, 344.7305908203125, 13.051841735839844, 258.8369445800781, 268.40264892578125, 126.39781188964844, 515.988525390625, -354.01171875, -53.37446594238281, 473.12445068359375, -77.0086669921875, 38.727264404296875, 0.0, 329.67535400390625, 202.60845947265625, 100.94320678710938, 127.78231811523438, 1327.970703125, 198.392822265625, -250.66070556640625, 483.74951171875, 147.06106567382812, 338.103271484375, 146.958740234375, 298.1378173828125, 320.6555480957031, 158.76370239257812, -4.64862060546875, 0.0, 213.99520874023438, 26.157913208007812, 379.6487121582031, 6.609222412109375, 190.77243041992188, 55.72027587890625, -11.28680419921875, 279.05352783203125, 297.561767578125, 223.8436737060547, -184.09661865234375, 835.8916015625, 170.48919677734375, 200.43360900878906, 390.96923828125, 123.80471801757812, 94.300048828125, 135.2613525390625, 115.3753890991211, 255.35888671875, 114.1556396484375, -14.16815185546875, 327.4847412109375, 375.79150390625, -279.3288879394531, -171.20758056640625, -56.76512908935547, 229.64913940429688, 12.169830322265625, -21.58984375, -4.860893249511719, 0.0, 0.553985595703125, 107.51608276367188, 171.38973999023438, -133.38613891601562, 131.6769256591797, -108.8995361328125, 503.1995544433594, 130.20579528808594, 37.87200927734375, 101.76144409179688, 202.33753967285156, -115.15756225585938, 146.20840454101562, 5.166435241699219, 169.2044677734375, -49.751861572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000446.npy"}
{"epoch": 0.9340314136125655, "step": 447, "batch_size": 128, "mean": 137.65089416503906, "std": 219.3703155517578, "min": -500.4207763671875, "p10": -127.12655944824218, "median": 127.21032333374023, "p90": 409.0862518310546, "max": 1118.279541015625, "pos_frac": 0.734375, "sample": [-26.797348022460938, 21.83394432067871, 295.1744689941406, 281.8878173828125, 21.1932373046875, 52.60280227661133, 536.836669921875, 265.5583190917969, -80.9791259765625, 542.0801391601562, 120.22564697265625, 98.52455139160156, 57.34783935546875, 579.1261596679688, 225.0946502685547, 161.86801147460938, 0.0, -123.81393432617188, 280.8853759765625, 66.44772338867188, 62.457550048828125, 189.7664794921875, 333.45068359375, 168.96759033203125, 280.7850036621094, 34.365753173828125, 64.87567138671875, 182.26052856445312, -179.508056640625, 604.1873779296875, -144.52175903320312, 107.878662109375, 216.58892822265625, 224.07008361816406, 208.67648315429688, 274.18524169921875, 354.163818359375, 371.7232666015625, 120.58341217041016, 535.77490234375, -35.84754943847656, 16.92791748046875, -231.35464477539062, 203.39178466796875, 105.71270751953125, 51.838287353515625, 148.79476928710938, 302.77496337890625, 40.15532684326172, 0.0, 210.0335693359375, 274.6585998535156, 140.28720092773438, 104.1778793334961, -166.65139770507812, -230.7867431640625, -192.55133056640625, 298.60968017578125, -8.872169494628906, 400.5628662109375, 156.28579711914062, 126.5420913696289, -234.51495361328125, -53.048919677734375, -159.12973022460938, 289.0761413574219, 466.9719543457031, 122.33000183105469, -31.676513671875, 26.47885513305664, 254.5897216796875, -44.22003173828125, -500.4207763671875, 117.06610107421875, 191.03970336914062, 6.3128814697265625, -161.818115234375, 205.81800842285156, 27.779296875, 264.865966796875, 328.0671691894531, -17.10150146484375, -56.09382629394531, 186.9761962890625, -277.51153564453125, 0.0, -37.210838317871094, 222.0584259033203, -134.85601806640625, 377.35406494140625, 107.11968994140625, 134.46142578125, 52.96099853515625, 131.3280792236328, 310.3739013671875, 151.13446044921875, 471.22906494140625, 597.7589111328125, 0.0, 277.328369140625, 133.0030059814453, 194.78692626953125, 36.3116455078125, 170.85379028320312, 53.300140380859375, 264.854248046875, 22.856597900390625, -276.8531494140625, 283.5845031738281, 114.3660888671875, -7.587188720703125, -104.06132507324219, 467.61578369140625, 522.2799682617188, 1118.279541015625, 153.62106323242188, 91.99732208251953, -42.79010009765625, -81.31147766113281, -48.43055725097656, 233.489013671875, 520.6048583984375, 186.2458038330078, 127.87855529785156, 185.23260498046875, 428.9741516113281, 188.30316162109375, -31.4849853515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000447.npy"}
{"epoch": 0.9361256544502617, "step": 448, "batch_size": 128, "mean": 129.2357177734375, "std": 204.60609436035156, "min": -356.3482971191406, "p10": -103.10514526367187, "median": 116.94071960449219, "p90": 365.2323944091797, "max": 892.6845092773438, "pos_frac": 0.75, "sample": [58.7894287109375, 122.20274353027344, 60.011260986328125, 44.77410888671875, 187.73834228515625, 222.67864990234375, 279.30181884765625, -187.44140625, 98.00105285644531, 125.72140502929688, -102.17498779296875, -57.11456298828125, 110.19903564453125, 113.78268432617188, 50.60137939453125, 132.067626953125, 170.97695922851562, 221.05902099609375, -35.2974853515625, -5.76695442199707, 250.91961669921875, 442.5273132324219, 542.111572265625, -59.429447174072266, -7.676246643066406, 74.97264099121094, 192.71896362304688, -5.337238311767578, 120.0987548828125, -32.146240234375, -24.308876037597656, 665.958740234375, 503.8019104003906, 137.19119262695312, 311.2093505859375, -229.6785888671875, 262.792724609375, 892.6845092773438, 60.970367431640625, 271.88726806640625, 138.49679565429688, 173.81529235839844, 180.51016235351562, 56.163780212402344, 14.94097900390625, 200.73594665527344, 271.6878967285156, -81.79242706298828, 141.9202423095703, 551.3443603515625, 80.68416595458984, 161.57464599609375, 13.176025390625, 67.59391784667969, 227.07028198242188, -45.63970947265625, 53.104888916015625, 204.44003295898438, 6.6953125, 296.1900634765625, 143.52037048339844, 163.05111694335938, 193.48931884765625, 37.57909393310547, 315.9189453125, 510.6783447265625, 121.94859313964844, 129.58326721191406, 594.2735595703125, 480.121337890625, 510.3880615234375, 273.02423095703125, 113.59336853027344, 93.61126708984375, 43.673606872558594, 124.69439697265625, 334.0975646972656, 57.760406494140625, 132.6253662109375, 196.06906127929688, -193.00814819335938, 368.2803039550781, -20.530555725097656, 315.4474182128906, 31.23614501953125, -173.38204956054688, -356.3482971191406, -316.73089599609375, 363.9261474609375, 70.12178039550781, -94.55294799804688, -91.1109619140625, 43.845306396484375, 336.0075988769531, 254.32606506347656, -57.00799560546875, -105.2755126953125, -176.15708923339844, 153.54833984375, 266.2309875488281, 328.8515930175781, 34.61919021606445, 195.3047332763672, -5.627799987792969, 427.54144287109375, -107.71728515625, -139.6222381591797, -146.33384704589844, 256.28668212890625, -234.62339782714844, 295.8970642089844, 43.8814697265625, -57.54402160644531, 155.903564453125, 19.358245849609375, 6.3226318359375, 106.2691650390625, -10.787200927734375, 267.501220703125, 59.11812210083008, 247.0045623779297, -57.6439208984375, 1.4486408233642578, 240.86607360839844, 663.4472045898438, -147.78744506835938, 158.68975830078125, 58.88993835449219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000448.npy"}
{"epoch": 0.9382198952879581, "step": 449, "batch_size": 128, "mean": 117.74109649658203, "std": 218.0215606689453, "min": -637.3128662109375, "p10": -126.68909301757812, "median": 141.0052947998047, "p90": 356.50755615234374, "max": 720.9700927734375, "pos_frac": 0.75, "sample": [251.62332153320312, -40.12669372558594, -18.172775268554688, 117.91062927246094, 186.12811279296875, -27.3798828125, -11.743194580078125, 353.65411376953125, 24.37976837158203, -51.47918701171875, 29.336563110351562, 106.2545166015625, 205.73338317871094, 86.80033874511719, 70.54725646972656, 492.9679260253906, 293.9748840332031, 252.7403564453125, 26.6353759765625, 36.52481460571289, -158.71429443359375, 259.2361755371094, 410.79443359375, 126.68511962890625, 284.9639892578125, 96.833251953125, -236.39483642578125, 141.03640747070312, 9.246837615966797, 173.22991943359375, 432.38775634765625, 610.494140625, 341.8914794921875, 259.6460876464844, -46.2838134765625, 195.4500732421875, 170.59817504882812, 69.55445861816406, 204.54330444335938, 119.40878295898438, 301.92333984375, 167.94287109375, 213.90985107421875, 408.2353515625, 363.3306884765625, 371.1580810546875, 63.609771728515625, 150.638427734375, 388.39971923828125, 142.34759521484375, -332.31341552734375, 62.00712585449219, -69.53231811523438, -52.68855285644531, 41.06001663208008, 192.71389770507812, 158.1216278076172, -27.462844848632812, 56.0970458984375, -353.2178649902344, 122.86615753173828, 218.27549743652344, 220.7408447265625, 108.89251708984375, 172.8621368408203, 206.1593017578125, 24.235748291015625, 231.32920837402344, 114.21795654296875, 249.05917358398438, -80.87332153320312, 300.9010009765625, 129.27496337890625, 271.0603332519531, -188.43524169921875, -33.9593391418457, 33.63953399658203, 145.02651977539062, 250.27613830566406, 265.4128112792969, 538.8128051757812, 94.3604965209961, 297.1635437011719, -61.573974609375, 326.8674621582031, -22.03631591796875, 363.16558837890625, -1.393157958984375, 247.55282592773438, -85.22454833984375, 35.06156921386719, 196.09552001953125, -505.07525634765625, 287.651611328125, 277.3018798828125, 296.36749267578125, -41.311859130859375, -116.9241714477539, 226.39578247070312, 140.97418212890625, -518.5689086914062, 274.9966735839844, 31.390838623046875, -637.3128662109375, -334.9152526855469, 221.6026611328125, -338.5860595703125, 2.9296340942382812, 187.697021484375, 189.14193725585938, 57.00231170654297, 108.1732177734375, -92.8065414428711, 15.171119689941406, 525.3795166015625, 41.887184143066406, 235.36734008789062, 143.3675079345703, -125.5478515625, 632.9111328125, 219.487060546875, -152.3705596923828, 720.9700927734375, 287.4755859375, 173.66854858398438, -162.4674835205078, -129.35198974609375, 141.77699279785156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000449.npy"}
{"epoch": 0.9403141361256544, "step": 450, "batch_size": 128, "mean": 106.791259765625, "std": 204.13636779785156, "min": -437.647216796875, "p10": -138.7618225097656, "median": 95.74844360351562, "p90": 328.1114074707031, "max": 871.8405151367188, "pos_frac": 0.6953125, "sample": [0.0, 10.383087158203125, 325.5223083496094, -151.3577880859375, 1.7105712890625, 36.80633544921875, 0.16566848754882812, -51.1165771484375, 259.7041015625, 51.04998779296875, -52.73223876953125, 871.8405151367188, 59.859031677246094, 20.8658447265625, 373.16021728515625, 85.81361389160156, -63.58155059814453, 170.55093383789062, 291.581787109375, -46.01646423339844, 111.17583465576172, 162.75473022460938, 153.71109008789062, 307.8392333984375, 187.16061401367188, 149.01840209960938, -7.1248321533203125, 95.4327392578125, -283.31683349609375, 62.528228759765625, -72.27484130859375, -15.513015747070312, -54.2342529296875, -17.578262329101562, 129.43609619140625, 35.6378173828125, 250.2215576171875, -94.93431091308594, 102.99334716796875, 59.48443603515625, 114.14137268066406, 317.8655700683594, 149.26170349121094, 268.27655029296875, -5.041351318359375, 125.94929504394531, 9.5816650390625, 69.18984985351562, 288.21014404296875, 227.58963012695312, 121.95992279052734, -20.002792358398438, 24.539012908935547, 60.86885070800781, 492.2973327636719, 166.79644775390625, 214.16653442382812, 303.4428405761719, 265.1045227050781, -178.5609130859375, 281.5774841308594, -197.24636840820312, 326.45904541015625, -142.322021484375, -77.22491455078125, 592.5166625976562, 331.9669189453125, 97.35858154296875, 263.13818359375, 303.29443359375, -148.2967987060547, 134.60610961914062, -37.880584716796875, 334.56365966796875, 166.26292419433594, 101.26781463623047, 289.1195068359375, 435.21624755859375, -234.96311950683594, 365.6461181640625, -229.21600341796875, 221.40582275390625, -193.35247802734375, -123.6746826171875, 290.8587646484375, 130.1246337890625, 23.913503646850586, 94.76171875, -119.57872772216797, -29.5560302734375, 43.110137939453125, 35.66413116455078, 744.8901977539062, 387.65179443359375, -143.06936645507812, 481.7416687011719, -137.23602294921875, -167.81207275390625, 24.751495361328125, -437.647216796875, 106.90695190429688, 297.8885498046875, 223.2274169921875, 163.11721801757812, 539.3576049804688, 250.68624877929688, -337.956298828125, 159.47471618652344, 181.9473876953125, 96.06414794921875, 254.44451904296875, 21.09166717529297, -41.79638671875, -17.783416748046875, 283.60601806640625, -13.577571868896484, -91.32379150390625, 464.03955078125, 243.77352905273438, 144.14971923828125, 0.0, 11.337158203125, -41.616302490234375, 115.55976867675781, 13.899169921875, 185.1850128173828, -110.25131225585938, 12.775650024414062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000450.npy"}
{"epoch": 0.9424083769633508, "step": 451, "batch_size": 128, "mean": 89.88134765625, "std": 203.0156707763672, "min": -564.5939331054688, "p10": -81.68745956420898, "median": 61.21512222290039, "p90": 317.39662780761716, "max": 662.40625, "pos_frac": 0.640625, "sample": [192.6658935546875, 169.2316436767578, 39.98443603515625, -29.32170867919922, 415.00537109375, 115.4884033203125, 315.446533203125, 26.775039672851562, -310.9403991699219, -53.40460205078125, 26.701446533203125, 61.37958526611328, 291.7612609863281, 22.93902587890625, 662.40625, 61.0506591796875, -29.421661376953125, -298.1996765136719, 372.0102844238281, -119.4637451171875, 479.2007141113281, -21.978607177734375, 16.315673828125, -88.86209106445312, 158.91616821289062, 38.15850830078125, 644.5504760742188, -5.2401580810546875, -65.94183349609375, -34.79693603515625, -333.84857177734375, 72.66256713867188, -384.6357421875, 319.8786926269531, 80.62210083007812, -49.656951904296875, 139.96286010742188, 42.742095947265625, 253.6221923828125, 232.0792236328125, 205.79676818847656, 392.8807067871094, 275.02294921875, 166.9986572265625, 173.6204833984375, 209.90298461914062, -67.9224624633789, -25.553386688232422, 487.8597412109375, -46.39794921875, -13.32550048828125, 310.73150634765625, -51.89741516113281, 589.581787109375, -3.470306396484375, -72.73858642578125, -6.040199279785156, 381.286376953125, -361.2863464355469, 201.80563354492188, 174.24359130859375, 146.2742462158203, 88.16641235351562, 176.65573120117188, -10.120094299316406, 22.059173583984375, -58.38371276855469, -67.06700897216797, 462.2141418457031, 316.3328857421875, -51.640869140625, 14.041046142578125, -38.73193359375, 130.00942993164062, 17.57611083984375, -252.3148193359375, 263.2808837890625, -186.77703857421875, 120.7320556640625, 196.43212890625, 270.8800048828125, -35.81194305419922, 3.365478515625, 251.60494995117188, 292.4076843261719, 73.01024627685547, 75.74484252929688, 292.82861328125, -20.889041900634766, 34.161949157714844, 213.22039794921875, 69.33511352539062, -564.5939331054688, 201.779052734375, 8.077560424804688, -149.75848388671875, -21.608932495117188, -42.74037170410156, -30.851486206054688, 234.10647583007812, 232.5758056640625, 210.66329956054688, 294.4013671875, 497.72393798828125, -110.15228271484375, 105.90782165527344, -70.43942260742188, 61.92631530761719, 165.9097900390625, 18.291275024414062, 171.92510986328125, -11.147628784179688, -28.303955078125, 240.14944458007812, -9.270179748535156, -15.944904327392578, 195.89434814453125, 44.581878662109375, -78.61261749267578, 398.9583740234375, 8.019287109375, -29.46112060546875, -308.51446533203125, 111.01705932617188, 92.54142761230469, 247.19076538085938, 42.8599853515625, 236.1717987060547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000451.npy"}
{"epoch": 0.9445026178010472, "step": 452, "batch_size": 128, "mean": 146.87001037597656, "std": 285.5296325683594, "min": -429.2840881347656, "p10": -125.1316909790039, "median": 114.00779724121094, "p90": 424.85863342285154, "max": 1953.498291015625, "pos_frac": 0.7421875, "sample": [-111.851318359375, 245.37295532226562, 575.2027587890625, 1953.498291015625, -329.86602783203125, 262.7886657714844, 71.79437255859375, 385.8359375, -296.027099609375, -82.56243133544922, 555.601318359375, 113.64947509765625, 68.25308227539062, 152.46151733398438, -111.28074645996094, 44.32958984375, 273.6246643066406, -59.01922607421875, -56.07086181640625, 319.1261291503906, 7.7455902099609375, -178.615966796875, -273.42279052734375, 284.53167724609375, 150.76513671875, 50.3436279296875, 72.00018310546875, 29.750778198242188, 66.24288940429688, 13.951400756835938, 82.96188354492188, 288.4313659667969, -43.13578414916992, 320.7353515625, 260.90185546875, -156.59185791015625, 951.70947265625, 229.8406982421875, 6.1110992431640625, -22.153732299804688, 143.6236572265625, -52.128662109375, 9.92449951171875, 182.0623779296875, 220.59060668945312, -6.570598602294922, 346.01446533203125, 114.36611938476562, 110.20835876464844, 102.827392578125, 229.589599609375, 112.9288330078125, 33.8370361328125, 152.16966247558594, 99.199462890625, -315.9342346191406, 125.63905334472656, -56.284027099609375, 283.2459716796875, 727.6650390625, 150.15301513671875, 142.64468383789062, 25.2225341796875, 144.9251708984375, -129.72967529296875, 578.1792602539062, 133.2429656982422, 286.22235107421875, 256.903076171875, 18.4466552734375, 214.38897705078125, 302.83294677734375, 152.82928466796875, 29.40252685546875, 266.48345947265625, 552.3499755859375, 352.757568359375, -169.61715698242188, 119.48681640625, 231.65663146972656, -429.2840881347656, 490.56964111328125, 482.4544372558594, 65.01669311523438, 427.8296813964844, 205.55059814453125, -63.83885955810547, 444.36090087890625, 423.5853271484375, -46.954345703125, 183.24029541015625, 914.7442626953125, 6.172943115234375, 23.252029418945312, -60.97520446777344, -96.20755004882812, 36.809349060058594, 234.56541442871094, 167.700439453125, -100.2552490234375, 345.8600769042969, 71.865234375, 134.54196166992188, 247.72650146484375, 123.81871795654297, 22.985414505004883, -92.2218017578125, 250.94692993164062, 313.3793029785156, 831.3911743164062, 312.1208190917969, 201.76791381835938, -129.4372100830078, 106.43951416015625, 404.8692321777344, 0.0, 379.6515197753906, 7.908489227294922, -216.1466064453125, -247.71783447265625, 125.27035522460938, 102.85543823242188, -276.49591064453125, -23.80816650390625, 16.368255615234375, 269.029296875, -123.28646850585938, -7.375545501708984], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000452.npy"}
{"epoch": 0.9465968586387434, "step": 453, "batch_size": 128, "mean": 123.36492156982422, "std": 223.7774200439453, "min": -514.8670654296875, "p10": -126.49278259277344, "median": 102.51293563842773, "p90": 384.35792846679686, "max": 934.6131591796875, "pos_frac": 0.7265625, "sample": [118.02295684814453, 116.98872375488281, -124.59375, 54.18463134765625, 934.6131591796875, 31.4686279296875, 113.12178039550781, -22.321739196777344, 142.647705078125, 212.07720947265625, 105.09246826171875, -107.55484008789062, 380.535400390625, 190.78369140625, 44.00543212890625, 19.901412963867188, 262.0201721191406, -265.79351806640625, -79.05792236328125, 393.27716064453125, -130.92385864257812, 360.07147216796875, 272.3578186035156, 154.2114715576172, 320.88623046875, 93.214599609375, -183.68438720703125, 632.31494140625, 97.30673217773438, 415.89923095703125, 341.4111328125, 243.15371704101562, 142.64598083496094, 257.007080078125, 185.8128204345703, 396.99041748046875, 92.75335693359375, -19.43255615234375, 347.3478088378906, 196.92059326171875, 119.28388977050781, 605.0410766601562, 235.0494384765625, -67.69134521484375, 0.34674835205078125, -157.86453247070312, 47.53240966796875, 178.22998046875, 119.79608154296875, -103.89834594726562, -14.13653564453125, 263.8006896972656, 16.793663024902344, -29.79492950439453, -46.00128173828125, 43.23590087890625, -138.36715698242188, -284.1902770996094, 477.877685546875, 21.109481811523438, 122.75244140625, 50.972434997558594, 0.0, 44.945098876953125, 301.60687255859375, 17.067527770996094, 54.07875061035156, 316.0155029296875, 45.820892333984375, -12.463050842285156, 31.2725830078125, 206.30670166015625, 173.11288452148438, -280.6535949707031, 45.782684326171875, 103.34725189208984, 227.47698974609375, 205.149658203125, 14.981191635131836, 81.9859619140625, 235.28237915039062, 615.973388671875, 529.5242309570312, -160.5390625, 204.568359375, -63.6583251953125, 222.39669799804688, -51.37298583984375, 210.8565673828125, 728.7584228515625, -200.11117553710938, 273.38201904296875, -56.721710205078125, -95.88909912109375, 146.33111572265625, 45.503936767578125, -158.15814208984375, -10.313552856445312, 92.07012939453125, 755.2699584960938, 62.8214111328125, 0.981201171875, -71.22586822509766, 101.67861938476562, 94.33773803710938, 234.9700927734375, -177.10801696777344, 5.0010223388671875, 476.01953125, -421.00238037109375, -0.4202423095703125, 170.3081817626953, 48.91680908203125, 374.2102355957031, 251.51156616210938, 155.30599975585938, 284.5270080566406, 440.4775085449219, -43.987548828125, 162.33790588378906, 348.98602294921875, -1.4170608520507812, 220.85794067382812, -88.65243530273438, 188.41448974609375, -514.8670654296875, 137.70223999023438, 315.47601318359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000453.npy"}
{"epoch": 0.9486910994764398, "step": 454, "batch_size": 128, "mean": 145.03086853027344, "std": 242.02085876464844, "min": -787.657958984375, "p10": -95.44400634765623, "median": 118.91107940673828, "p90": 437.36799316406245, "max": 1159.535888671875, "pos_frac": 0.7578125, "sample": [-7.63861083984375, 282.9053649902344, -199.640869140625, 47.70314025878906, -86.79385375976562, 150.02862548828125, 244.2681884765625, 243.1021728515625, 126.87380981445312, 299.5305480957031, 148.75746154785156, -25.7523193359375, -20.002559661865234, 172.03167724609375, 210.16839599609375, 163.58969116210938, -258.22003173828125, -17.87117576599121, 8.434158325195312, -7.042594909667969, 327.23486328125, 8.794807434082031, 418.42523193359375, 458.5723876953125, 444.416015625, 336.2034912109375, -25.69805908203125, 415.48828125, -420.874755859375, 118.56817626953125, 27.92237091064453, 219.7852783203125, 100.51676177978516, 142.94390869140625, 499.655517578125, 66.80035400390625, 264.4808349609375, -787.657958984375, 137.267333984375, 121.51443481445312, 254.9114990234375, 416.7854919433594, 299.31646728515625, 83.93148803710938, 162.11834716796875, 364.1943054199219, 307.3647766113281, 34.08172607421875, 250.9780731201172, 85.5860595703125, 60.16180419921875, 1159.535888671875, 546.2073364257812, -222.57736206054688, 167.26211547851562, -87.1636962890625, 77.18539428710938, 300.34661865234375, 598.405029296875, 486.56939697265625, 138.3592071533203, 252.96817016601562, 434.347412109375, -61.143707275390625, 861.0784912109375, 382.16583251953125, 300.7131652832031, 134.83544921875, -33.229248046875, -133.32029724121094, 264.506591796875, 217.614013671875, 12.4212646484375, 62.284332275390625, 34.786407470703125, 223.05130004882812, -129.05885314941406, 91.65864562988281, -112.86419677734375, -79.51641845703125, 10.266876220703125, -43.376731872558594, 669.721435546875, -87.97821044921875, 92.53341674804688, 277.24151611328125, 222.0791778564453, 142.94677734375, 27.827880859375, 236.61538696289062, 119.25398254394531, 78.44058227539062, -14.314453125, 591.6568603515625, 370.77117919921875, -54.42779541015625, 261.66876220703125, 111.82083129882812, 549.8568115234375, 37.79120635986328, -37.33106231689453, 45.45166015625, 647.9219970703125, 17.189149856567383, 24.45404052734375, -45.10382080078125, 552.9541015625, 179.71409606933594, -205.772705078125, 228.27474975585938, -114.56704711914062, 223.88253784179688, 112.12921142578125, -161.71905517578125, 57.862213134765625, 140.665771484375, 92.024169921875, 126.53762817382812, -218.57080078125, 116.48871612548828, 64.64822387695312, 112.51353454589844, 30.5828857421875, 309.6751708984375, 25.773651123046875, -7.51202392578125, 207.76707458496094, -118.02706909179688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000454.npy"}
{"epoch": 0.9507853403141361, "step": 455, "batch_size": 128, "mean": 88.60137939453125, "std": 220.6866912841797, "min": -447.33587646484375, "p10": -136.92631225585936, "median": 53.00345230102539, "p90": 340.9591125488281, "max": 1100.98046875, "pos_frac": 0.6875, "sample": [584.2152099609375, -100.9993667602539, -35.24174499511719, 21.829193115234375, 227.30819702148438, -53.32342529296875, -88.54100036621094, 554.489013671875, 14.644882202148438, -19.4564208984375, -52.583221435546875, 142.7735595703125, 17.94122314453125, 21.35877227783203, 15.86684799194336, 178.31356811523438, 161.70001220703125, 111.38217163085938, 413.109619140625, 272.99853515625, 32.497894287109375, 91.89779663085938, 204.97909545898438, 207.56497192382812, 872.410888671875, 233.6905975341797, 357.3662414550781, 93.5089111328125, -0.4899787902832031, 1.2991943359375, -276.1209716796875, -378.91912841796875, -4.486789703369141, 76.56343078613281, 252.50247192382812, -134.67221069335938, -36.871307373046875, 14.3284912109375, 61.574798583984375, 70.57327270507812, 287.4463195800781, 25.323486328125, 17.891399383544922, 346.069580078125, -76.8834228515625, 406.97900390625, 339.1715087890625, 71.91244506835938, 103.03221130371094, 8.97454833984375, -69.19386291503906, 36.45672607421875, 58.27838134765625, -11.105476379394531, -3.69482421875, -16.40509033203125, 60.46484375, -169.51016235351562, -37.96983337402344, 62.74230194091797, 71.39183044433594, 45.03358459472656, 32.21864318847656, -243.11959838867188, -165.56329345703125, -30.871763229370117, 127.62799072265625, -226.2156982421875, 311.47906494140625, 146.4825897216797, -76.44821166992188, 104.1329345703125, 238.4656982421875, 278.74554443359375, 279.9786376953125, -186.6634521484375, 5.63330078125, 116.46221923828125, 24.13897705078125, 390.07891845703125, 16.9471435546875, 3.5818099975585938, 55.84571075439453, -235.094482421875, 345.13018798828125, -146.3177947998047, -363.53668212890625, 94.830078125, -41.722930908203125, 236.99740600585938, 72.64955139160156, -55.05816650390625, -104.2523193359375, 12.92718505859375, -142.18588256835938, 182.94570922851562, 259.68035888671875, 1100.98046875, -20.0693359375, -307.8917236328125, 403.2745361328125, 19.26318359375, 159.11798095703125, 282.5460205078125, 0.656829833984375, -39.5543098449707, 248.15975952148438, 22.2835693359375, 132.5391845703125, 186.54876708984375, -29.264312744140625, 127.5191650390625, 67.693115234375, -24.44232177734375, 209.2596435546875, -12.345514297485352, -447.33587646484375, 50.16119384765625, 99.02462768554688, 264.5471496582031, 133.11050415039062, 70.97079467773438, 835.0352783203125, 142.2354736328125, 254.5685577392578, 24.18389892578125, 400.0286865234375, -15.1474609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000455.npy"}
{"epoch": 0.9528795811518325, "step": 456, "batch_size": 128, "mean": 129.9503936767578, "std": 265.9058532714844, "min": -457.1153564453125, "p10": -127.28134765624998, "median": 88.44943237304688, "p90": 405.077490234375, "max": 1521.73974609375, "pos_frac": 0.703125, "sample": [50.325531005859375, -2.3198394775390625, -457.1153564453125, 240.33079528808594, 186.8955841064453, -144.31146240234375, 162.94140625, -47.25830078125, 259.3167724609375, 50.978851318359375, 193.2230224609375, -329.0525817871094, 116.33750915527344, -27.71478271484375, 119.19818115234375, 241.80010986328125, 86.95089721679688, -117.98123168945312, -119.98272705078125, 81.81979370117188, 488.6776123046875, 30.79443359375, -160.97589111328125, 280.6757507324219, -7.614480972290039, 51.58234405517578, 104.0018310546875, -64.18809509277344, 215.98336791992188, 91.03045654296875, 228.653076171875, -74.87606811523438, 180.58349609375, -257.8616943359375, 278.79046630859375, 413.42767333984375, 56.403656005859375, 451.80438232421875, 219.40406799316406, 89.94796752929688, 562.3168334960938, 33.3961181640625, 376.70562744140625, 5.87567138671875, 232.02908325195312, 261.24169921875, 544.16650390625, 48.689178466796875, 99.43228149414062, 164.68060302734375, 71.35562133789062, 84.77880859375, 23.66278076171875, 92.74190521240234, -348.8271484375, 29.331329345703125, -167.37884521484375, -89.71394348144531, 44.35986328125, 382.4226989746094, 26.74609375, -23.048091888427734, 53.269073486328125, -34.001739501953125, 62.34979248046875, 596.39697265625, 133.81692504882812, -247.60064697265625, 134.31625366210938, 13.80902099609375, 38.118751525878906, 130.14303588867188, -55.624908447265625, 180.99099731445312, 80.33634948730469, 363.21728515625, 385.93988037109375, 220.6541748046875, -48.033966064453125, 165.35498046875, 204.79574584960938, 1249.11328125, -185.5543212890625, 305.542724609375, -3.499664306640625, 217.21224975585938, 29.0059814453125, 667.641845703125, 163.50238037109375, 336.56884765625, 51.161163330078125, 356.4482727050781, -62.026702880859375, 143.64419555664062, 0.0, -13.646041870117188, 77.58246612548828, 366.8091125488281, 208.78140258789062, -39.04478454589844, 401.49884033203125, 99.74343872070312, 155.884521484375, 32.24821472167969, 252.33529663085938, -310.2450256347656, 479.7354736328125, 182.5312957763672, 55.893798828125, -186.30157470703125, -203.40072631835938, -54.76081085205078, -54.748504638671875, -116.91763305664062, 215.17230224609375, -5.824909210205078, 1521.73974609375, 171.4478759765625, 716.4234008789062, 158.81298828125, 303.54803466796875, -106.75326538085938, -12.67291259765625, -197.05191040039062, 460.0660705566406, 149.8367919921875, -87.86851501464844, 750.197021484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000456.npy"}
{"epoch": 0.9549738219895288, "step": 457, "batch_size": 128, "mean": 135.99757385253906, "std": 255.5071563720703, "min": -1193.224365234375, "p10": -117.67619628906249, "median": 154.40110778808594, "p90": 397.4163879394531, "max": 964.770263671875, "pos_frac": 0.75, "sample": [297.49609375, 134.015380859375, -205.26126098632812, -34.15917205810547, 471.9580078125, 401.0111083984375, 230.29922485351562, 191.1478271484375, 264.1544494628906, -114.90432739257812, 264.3354797363281, 11.811019897460938, 290.766357421875, 17.206771850585938, 78.19781494140625, 242.13287353515625, 395.14947509765625, 206.47100830078125, -12.588653564453125, 311.8227233886719, -8.582443237304688, 235.62222290039062, 46.855812072753906, 417.7167053222656, 325.51904296875, 259.84637451171875, 283.4143981933594, -6.2711639404296875, 383.474853515625, 17.951187133789062, 305.62847900390625, 22.99481201171875, 21.670806884765625, -38.89996337890625, 248.6186981201172, 193.98171997070312, 38.52970504760742, 339.2139892578125, -142.10882568359375, 296.89581298828125, 235.50711059570312, 356.03802490234375, -13.855720520019531, 17.6180419921875, 106.0971908569336, -343.7474060058594, 258.62200927734375, 20.40673828125, 291.9095458984375, 162.9486541748047, -13.10162353515625, 170.80731201171875, 47.08479309082031, -1193.224365234375, 274.70147705078125, 33.686927795410156, -92.4759521484375, 468.57879638671875, 184.12542724609375, 45.90716552734375, -9.9263916015625, -128.21893310546875, 300.42608642578125, -211.0762481689453, 289.9591064453125, 964.770263671875, -21.99658203125, 15.087471008300781, 291.04791259765625, -32.270782470703125, -57.58787536621094, 127.43792724609375, 389.1822814941406, 240.637939453125, 645.6753540039062, -127.36578369140625, -785.0970458984375, 194.09201049804688, 58.805870056152344, 130.5421600341797, 140.56396484375, 238.98590087890625, 142.54229736328125, 280.81689453125, 152.94158935546875, -348.9124755859375, 160.32745361328125, 155.86062622070312, 99.298828125, 534.4979248046875, -18.75847053527832, -54.0372314453125, 38.4677734375, -124.14389038085938, -127.09051513671875, 470.85595703125, -215.0465087890625, 395.87579345703125, 99.56036376953125, 573.3048706054688, 224.944580078125, 48.7354736328125, 282.76336669921875, 269.2004699707031, -436.1097412109375, 247.880859375, 221.337646484375, 185.49099731445312, 320.715576171875, 462.32049560546875, 600.675048828125, 228.18215942382812, 51.75726318359375, 249.48118591308594, 457.54791259765625, 193.23374938964844, 145.15806579589844, 4.8560333251953125, -42.31805419921875, 597.2537841796875, 255.67428588867188, 9.603906631469727, 0.0, 223.75750732421875, 37.25397491455078, -111.05598449707031, 122.30699157714844, -13.762413024902344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000457.npy"}
{"epoch": 0.9570680628272251, "step": 458, "batch_size": 128, "mean": 110.84642028808594, "std": 239.9043731689453, "min": -626.7691650390625, "p10": -151.23709716796876, "median": 84.19349670410156, "p90": 339.64692382812495, "max": 894.7635498046875, "pos_frac": 0.6953125, "sample": [53.63020324707031, -131.15606689453125, 116.54214477539062, -250.6263885498047, 241.10545349121094, 58.68031311035156, -97.36508178710938, 300.4991760253906, 268.1693115234375, -183.70614624023438, 91.44219970703125, -82.01841735839844, -626.7691650390625, 4.397001266479492, 10.511966705322266, 293.944091796875, 144.55078125, -257.3497314453125, -61.725341796875, 683.1586303710938, 323.8076171875, 20.48422622680664, 267.6153564453125, -158.9916534423828, 120.32925415039062, 0.8822460174560547, 243.62445068359375, 716.3327026367188, -21.383941650390625, 199.07647705078125, 348.5328674316406, 16.533966064453125, 307.878662109375, 291.8876037597656, 287.44427490234375, 133.35772705078125, 145.7802734375, 336.9578552246094, 229.18270874023438, -80.20790100097656, 333.7586364746094, -12.856222152709961, -114.19790649414062, -126.3023681640625, 17.18756103515625, -160.92758178710938, -51.70062255859375, -69.34102630615234, 21.33733367919922, -65.31907653808594, 83.05311584472656, -322.12908935546875, -416.71368408203125, 241.4176025390625, 261.79156494140625, -214.7882080078125, 33.4588623046875, -41.629425048828125, 73.2209243774414, 299.96185302734375, -28.3751220703125, 4.998912811279297, 220.73822021484375, 127.30404663085938, -5.587802886962891, -150.3695068359375, 109.34463500976562, -120.90567016601562, -5.52740478515625, 293.32952880859375, 308.3440856933594, -114.86253356933594, 64.77899169921875, 894.7635498046875, 79.37530517578125, 184.2942657470703, 44.0531005859375, -208.17982482910156, -28.080703735351562, 182.12002563476562, 394.4005126953125, -153.261474609375, 47.886383056640625, 145.7511749267578, 345.9214172363281, 6.9923858642578125, 666.8201293945312, 58.6968994140625, 124.52108764648438, 273.5810546875, -25.79376220703125, 324.8274230957031, -49.35826110839844, 177.99229431152344, 294.65087890625, -41.62236785888672, 123.40179443359375, 26.224365234375, 16.673181533813477, 182.9686279296875, 67.05438232421875, 15.70977783203125, 203.47918701171875, 366.94708251953125, 44.384735107421875, -125.78677368164062, 151.58309936523438, 161.13003540039062, 715.655029296875, 270.00225830078125, 162.41571044921875, 875.0538330078125, 220.55201721191406, 283.01751708984375, 715.8956909179688, 419.21270751953125, 187.7947998046875, 286.1493835449219, -190.61114501953125, -307.020263671875, 386.97821044921875, 309.96533203125, -68.11290740966797, 16.5938720703125, -144.98931884765625, 123.00543212890625, 89.79638671875, 85.33387756347656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000458.npy"}
{"epoch": 0.9591623036649215, "step": 459, "batch_size": 128, "mean": 98.65508270263672, "std": 216.96096801757812, "min": -461.7247314453125, "p10": -122.9703010559082, "median": 62.94158935546875, "p90": 346.000732421875, "max": 1073.1817626953125, "pos_frac": 0.625, "sample": [130.56268310546875, -36.32568359375, 31.506446838378906, 208.85531616210938, -461.7247314453125, 120.00723266601562, -167.72933959960938, 123.787353515625, 147.93374633789062, 287.7891845703125, 295.2452087402344, -149.1584014892578, 81.46749114990234, -87.84893798828125, 138.94540405273438, -124.2691879272461, 266.111572265625, 344.6842041015625, -48.385658264160156, -33.35382080078125, 107.02032470703125, 188.24822998046875, -112.16505432128906, 245.80410766601562, 27.29205322265625, -142.1399688720703, 25.437896728515625, -13.114524841308594, 175.92758178710938, 98.35850524902344, -22.515668869018555, 0.0, 44.921478271484375, -62.99505615234375, 189.81396484375, 34.59222412109375, 396.267333984375, 3.791402816772461, -214.46188354492188, 110.908203125, 575.4326171875, 63.2078857421875, 46.912109375, 254.70770263671875, 31.142822265625, 2.0382080078125, -272.43505859375, 296.3538818359375, -131.73635864257812, 258.917724609375, -32.020965576171875, 25.045673370361328, 9.622440338134766, 563.9606323242188, 207.46142578125, -74.07303619384766, 526.44482421875, -101.00482177734375, 440.89263916015625, 419.9669189453125, 262.2888488769531, -112.23672485351562, 394.5054931640625, 262.3809509277344, 498.0603942871094, 338.3604431152344, 324.9210205078125, -127.511474609375, -87.97372436523438, -44.226776123046875, 62.67529296875, 94.955322265625, 186.06016540527344, 256.94390869140625, -84.28007507324219, -173.57696533203125, 119.7896728515625, 168.54849243164062, 57.21661376953125, -94.18511962890625, -265.66192626953125, -20.7613525390625, -17.606689453125, -92.25587463378906, 281.58331298828125, 105.59417724609375, 58.0035400390625, 182.78836059570312, 0.0, 51.015098571777344, -8.726943969726562, -122.41363525390625, 118.7130126953125, 351.6884765625, -87.7506103515625, 236.05081176757812, -104.81501770019531, 207.72677612304688, 574.3685913085938, -20.038589477539062, 156.21884155273438, 264.4109802246094, 117.26190185546875, -41.294036865234375, -102.41612243652344, -111.19465637207031, 291.22900390625, -52.697509765625, -115.44850158691406, 105.87831115722656, -349.7809143066406, -44.76751708984375, 1073.1817626953125, -95.8338623046875, 329.06634521484375, 349.0726318359375, -54.24491882324219, 94.1293716430664, 302.34185791015625, 151.69384765625, 33.365386962890625, 648.956298828125, 343.61663818359375, -124.99177551269531, 194.02972412109375, -40.34527587890625, 246.71609497070312, 69.55125427246094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000459.npy"}
{"epoch": 0.9612565445026178, "step": 460, "batch_size": 128, "mean": 97.72859191894531, "std": 203.4007568359375, "min": -408.1522216796875, "p10": -124.11708679199216, "median": 75.96998596191406, "p90": 368.1994415283203, "max": 578.32421875, "pos_frac": 0.671875, "sample": [-6.49615478515625, 73.69500732421875, 10.180084228515625, 244.18316650390625, 353.44873046875, 305.97454833984375, -213.66766357421875, 14.79104232788086, -301.54681396484375, 325.78424072265625, -118.90756225585938, 268.6727600097656, 12.099609375, -227.9487762451172, -53.89830017089844, -136.27264404296875, 39.3631591796875, 97.52734375, 363.32342529296875, 418.8005676269531, -408.1522216796875, 28.646240234375, 148.1302490234375, 103.10162353515625, 219.9679412841797, 100.23421478271484, 188.50973510742188, -44.01762390136719, 423.6724853515625, -160.15228271484375, -65.22225952148438, 217.75946044921875, -41.108551025390625, 496.2244873046875, 276.88409423828125, -382.17889404296875, 184.07029724121094, -66.9146728515625, 519.0428466796875, -26.758758544921875, -378.570556640625, 443.0527038574219, 105.81927490234375, 146.10227966308594, 62.73541259765625, 190.7855224609375, -97.96112060546875, 578.32421875, 379.5768127441406, -25.066177368164062, 337.28448486328125, -101.93659973144531, 19.090383529663086, 34.05674743652344, -35.56558609008789, 10.706634521484375, -4.249298095703125, -48.723388671875, -195.12391662597656, -45.22137451171875, 207.25823974609375, -26.072052001953125, 428.08770751953125, 87.00663757324219, 2.9803619384765625, -20.097137451171875, -15.118144989013672, 269.9250793457031, 125.79437255859375, 195.6545867919922, 329.2919921875, -57.2972412109375, 50.763824462890625, 84.39776611328125, 570.5240478515625, -30.22955894470215, 95.09219360351562, 307.0704650878906, 171.56951904296875, -101.52426147460938, 322.6166076660156, 104.76541137695312, -38.290313720703125, 426.6723327636719, 354.68023681640625, 493.74139404296875, 414.4196472167969, 280.46319580078125, -63.8524169921875, 254.93675231933594, 437.30712890625, 94.14981079101562, 49.47381591796875, 112.5197525024414, -102.74320983886719, 328.3100280761719, 237.3588409423828, -244.63433837890625, 177.28277587890625, -263.6366882324219, 183.0673370361328, -289.66943359375, 95.3284912109375, -14.681243896484375, 184.61474609375, 41.84770202636719, 17.796714782714844, -48.57977294921875, 103.33111572265625, 65.58511352539062, 68.17431640625, 88.79721069335938, 72.15325927734375, 71.24468994140625, -161.29107666015625, 23.61587142944336, 51.052635192871094, 269.640625, -95.57177734375, 307.609619140625, 78.24496459960938, 44.577842712402344, 319.5359802246094, -35.962890625, 122.97227478027344, 197.9088134765625, 202.32839965820312, -60.990814208984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000460.npy"}
{"epoch": 0.9633507853403142, "step": 461, "batch_size": 128, "mean": 123.66246795654297, "std": 233.16534423828125, "min": -517.3217163085938, "p10": -134.2118225097656, "median": 104.8855972290039, "p90": 380.11721496582027, "max": 1076.357421875, "pos_frac": 0.671875, "sample": [147.3981475830078, -196.14303588867188, 301.5008544921875, 59.05080032348633, 160.8546600341797, 399.19439697265625, 189.12574768066406, -3.96484375, -57.491485595703125, 460.8848571777344, 472.439208984375, 0.0, 67.27558898925781, 883.49560546875, 125.02490234375, 225.8181915283203, 291.344482421875, 150.67233276367188, 324.8828430175781, 196.4281768798828, 210.67117309570312, 182.46051025390625, 86.53379821777344, -48.89306640625, 238.333740234375, -76.74462890625, 132.440185546875, 1076.357421875, -246.43028259277344, 329.68731689453125, -54.462158203125, 298.32135009765625, 199.185546875, 216.8893280029297, -42.536163330078125, 170.18357849121094, 363.9963073730469, 374.6426696777344, -517.3217163085938, -117.13113403320312, 485.409912109375, -184.1763916015625, -75.5671615600586, 81.66775512695312, 249.82464599609375, 149.1129150390625, 30.618850708007812, 262.400634765625, 281.6051940917969, 250.62139892578125, 284.69158935546875, -32.093318939208984, 72.33836364746094, -68.19857788085938, 53.26996994018555, -0.3106689453125, 186.4442138671875, -170.16912841796875, -76.82794189453125, 153.59982299804688, 376.19744873046875, 92.16169738769531, 31.717575073242188, 337.2141418457031, 508.7738952636719, 164.6728515625, -230.71478271484375, 76.1971435546875, 264.3617858886719, 48.17823791503906, -3.7783279418945312, 389.2633361816406, -41.60974884033203, 306.1090087890625, 88.82415771484375, 221.95013427734375, 831.65234375, 56.26031494140625, -168.61474609375, -145.97711181640625, 248.59356689453125, 75.18692016601562, -129.1695556640625, 181.51800537109375, 156.46514892578125, 136.61599731445312, -27.865341186523438, 78.61758422851562, -61.37174987792969, -2.918914794921875, -189.53817749023438, 92.25865936279297, 9.6309814453125, 216.79380798339844, -38.8369140625, 681.7855224609375, -65.61465454101562, 450.4760437011719, 77.66253662109375, -63.93754196166992, 145.5653533935547, -9.377494812011719, 528.6236572265625, -10.498870849609375, -232.75509643554688, -63.5689697265625, 212.81968688964844, 258.68341064453125, 79.1725845336914, -159.384033203125, -91.32484436035156, 153.42515563964844, -4.977636337280273, 197.1937255859375, 19.570098876953125, 180.38673400878906, 117.51253509521484, 244.1990203857422, -251.46881103515625, -118.48829650878906, 45.184051513671875, 119.600830078125, -406.95758056640625, 479.52978515625, -53.71833038330078, 19.867095947265625, 216.41033935546875, 276.14337158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000461.npy"}
{"epoch": 0.9654450261780104, "step": 462, "batch_size": 128, "mean": 136.45535278320312, "std": 209.66116333007812, "min": -321.26123046875, "p10": -107.89997634887695, "median": 131.01673889160156, "p90": 378.29044799804683, "max": 1171.088623046875, "pos_frac": 0.765625, "sample": [325.11236572265625, 358.54547119140625, 328.6272277832031, 62.7509765625, 491.21173095703125, 84.93310546875, 407.07440185546875, 171.9503173828125, 126.466796875, 181.35052490234375, -266.24237060546875, 317.482177734375, 597.2426147460938, 106.12053680419922, 303.3549499511719, -149.4781036376953, 244.10430908203125, 280.41680908203125, 184.0819091796875, 24.77008056640625, 132.79315185546875, 216.29183959960938, -18.434539794921875, 544.02685546875, -62.95899200439453, 21.7789306640625, 137.293701171875, -15.973735809326172, 256.075927734375, -107.8117904663086, -209.78604125976562, 261.7674560546875, 233.3072509765625, 162.68389892578125, 75.943603515625, 139.51153564453125, 342.55877685546875, -51.749359130859375, -128.31094360351562, 49.613037109375, 363.879638671875, 33.8233642578125, 258.3346862792969, 13.520606994628906, 316.8804931640625, 24.98086929321289, 426.2610168457031, 417.34906005859375, 367.5838623046875, 131.02487182617188, 337.391357421875, 181.18051147460938, 174.545654296875, -119.8584213256836, 21.749069213867188, -73.46736907958984, 254.99490356445312, 15.67034912109375, 175.865966796875, 10.71612548828125, 202.42413330078125, 375.0255126953125, -108.10574340820312, 528.7594604492188, 23.781463623046875, -1.3595962524414062, -300.90380859375, 178.65673828125, 385.90863037109375, -144.1661834716797, 92.921142578125, 420.93707275390625, 183.74417114257812, 78.381103515625, 306.99920654296875, 98.51042175292969, -1.5380859375, 85.96710205078125, -29.204530715942383, 131.00860595703125, 89.58401489257812, 7.554534912109375, -97.14990234375, 45.05323791503906, -89.42962646484375, 4.8184967041015625, 156.93743896484375, 312.25738525390625, -43.25569152832031, -5.2961883544921875, -27.062278747558594, 356.06903076171875, 177.7277374267578, 6.4660186767578125, 134.17706298828125, 1171.088623046875, -321.26123046875, 98.9945068359375, 86.8809814453125, 581.76904296875, 66.29399108886719, 137.17868041992188, 475.45721435546875, 387.845703125, 32.24847412109375, 47.275177001953125, 246.66812133789062, -76.37823486328125, 142.54693603515625, 27.514892578125, 219.00830078125, 49.13995361328125, 194.1182861328125, 313.5664978027344, -283.4747314453125, 241.36671447753906, -155.3963623046875, 186.74945068359375, 148.12818908691406, -15.08782958984375, 301.8883056640625, 178.51113891601562, -124.0247802734375, -12.124198913574219, 79.41323852539062, 177.09967041015625, 84.64517211914062, -248.48577880859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000462.npy"}
{"epoch": 0.9675392670157068, "step": 463, "batch_size": 128, "mean": 102.5419692993164, "std": 249.15994262695312, "min": -814.6707153320312, "p10": -123.54250793457031, "median": 77.65730285644531, "p90": 314.71156311035156, "max": 1182.447265625, "pos_frac": 0.6875, "sample": [119.25335693359375, -128.37835693359375, 130.56651306152344, -59.85414123535156, 157.04757690429688, 39.85900115966797, 235.0777587890625, -105.3837890625, -466.4061584472656, -701.99853515625, 202.94850158691406, 13.663116455078125, 80.93545532226562, -121.47000122070312, 149.08181762695312, 346.4612121582031, -30.143592834472656, 305.040283203125, -6.247589111328125, 10.525627136230469, 13.967971801757812, -360.08087158203125, -86.9022216796875, 95.12977600097656, -57.671470642089844, 67.05611419677734, -63.32891845703125, 296.73626708984375, 478.82025146484375, 302.25146484375, -28.954025268554688, 317.92626953125, 64.27098083496094, 37.06109619140625, -97.70962524414062, 303.03436279296875, 341.9156494140625, 475.11968994140625, 991.9164428710938, 103.11656188964844, 35.196563720703125, 147.99395751953125, -43.4017333984375, 271.771728515625, 43.966087341308594, 0.30584716796875, 656.0361938476562, 46.27021026611328, 93.00628662109375, 101.63922882080078, -211.37158203125, 96.73280334472656, 297.6184387207031, 441.5301208496094, 101.51449584960938, 243.99740600585938, 209.16244506835938, 223.61984252929688, 184.779541015625, 174.39169311523438, 5.638824462890625, 280.35723876953125, -8.17877197265625, -35.223297119140625, -169.58924865722656, 36.61997985839844, 180.26373291015625, 185.5764923095703, -80.39395141601562, 101.4393310546875, 320.39013671875, -150.76345825195312, 156.3359832763672, 286.308837890625, 226.14804077148438, -13.36749267578125, -77.93728637695312, -83.97615051269531, -156.92745971679688, 236.1940155029297, 264.32275390625, 169.9676513671875, 235.47164916992188, 759.032470703125, 1182.447265625, 67.12332153320312, 25.087093353271484, 306.723876953125, 701.5166015625, -38.25319290161133, -4.016082763671875, -38.4842529296875, -101.35060119628906, 0.0, -146.6932373046875, 35.52197265625, 50.7494010925293, 178.5108642578125, 81.72695922851562, -197.1179656982422, -138.5919189453125, 122.32856750488281, -73.64422607421875, 4.538869857788086, 71.64469909667969, 70.15579223632812, -22.32354736328125, 171.50428771972656, 219.3187255859375, -19.118846893310547, 142.42239379882812, 86.36575317382812, 74.379150390625, 161.2276153564453, -73.96343994140625, 45.626163482666016, 255.11917114257812, 313.3338317871094, 483.5076904296875, 288.4918212890625, 179.44985961914062, 53.316986083984375, 2.9548797607421875, 305.44476318359375, -141.46002197265625, 119.81524658203125, -814.6707153320312, -12.017547607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000463.npy"}
{"epoch": 0.9696335078534032, "step": 464, "batch_size": 128, "mean": 110.7613525390625, "std": 204.6195831298828, "min": -554.1276245117188, "p10": -136.1015838623047, "median": 111.0276870727539, "p90": 333.55597534179685, "max": 823.2061767578125, "pos_frac": 0.6953125, "sample": [0.0, 202.58465576171875, -15.295036315917969, 351.25927734375, -188.4732666015625, 268.7242431640625, 165.73727416992188, -222.84774780273438, 175.30422973632812, -14.10365104675293, -67.11700439453125, -1.235015869140625, 190.63516235351562, 123.95449829101562, 71.13613891601562, -7.878025054931641, 294.8999938964844, 123.10470581054688, 317.89837646484375, -66.20730590820312, 554.768310546875, 159.2042236328125, 41.619300842285156, -247.48745727539062, 124.47726440429688, 233.08346557617188, -16.888999938964844, -28.434967041015625, 332.02545166015625, -194.1546630859375, -238.81195068359375, 56.62739562988281, 30.566650390625, 212.3053741455078, 280.23382568359375, 370.9805908203125, 173.75912475585938, 161.36981201171875, -20.60272216796875, 25.55944061279297, 153.27264404296875, 38.6466064453125, 250.96121215820312, 823.2061767578125, 145.64785766601562, 1.6312713623046875, -132.32647705078125, -42.1842041015625, -33.535858154296875, 290.688720703125, -8.757049560546875, 346.03973388671875, -68.68878936767578, 298.7977600097656, -52.80637741088867, 179.84283447265625, 183.4078369140625, 228.79071044921875, -315.57257080078125, 321.6537780761719, -134.7950439453125, 323.31378173828125, 6.8311614990234375, 245.58132934570312, 620.2945556640625, 28.58465576171875, 145.41580200195312, 94.34988403320312, 150.63031005859375, 263.80059814453125, 9.67681884765625, -286.7623291015625, 15.906631469726562, -28.390106201171875, 210.56765747070312, 15.947433471679688, -31.77587890625, 28.809295654296875, 560.4572143554688, 149.00946044921875, 149.08477783203125, 320.4818115234375, 134.78515625, 417.7076110839844, -202.10333251953125, 149.820068359375, -9.579769134521484, -139.15017700195312, 25.9320068359375, 12.844970703125, 294.9717102050781, -226.58059692382812, 204.84808349609375, -6.00927734375, 9.73223876953125, 149.98695373535156, 75.56689453125, -147.6141357421875, 41.67401123046875, -62.623294830322266, 59.50505065917969, 670.51171875, 52.91279602050781, 275.74371337890625, 450.0936279296875, 57.80889892578125, 386.05584716796875, -90.22396850585938, 290.00469970703125, -49.885650634765625, 204.20172119140625, -29.109554290771484, 98.95066833496094, 244.12326049804688, 338.10589599609375, 175.9183349609375, 337.127197265625, 205.46864318847656, 235.73394775390625, -554.1276245117188, 262.3514099121094, -24.2281494140625, 78.41799926757812, 241.58485412597656, 75.54119873046875, -199.70806884765625, 305.3638916015625, 176.98500061035156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000464.npy"}
{"epoch": 0.9717277486910995, "step": 465, "batch_size": 128, "mean": 140.13165283203125, "std": 223.9285888671875, "min": -303.9912109375, "p10": -77.17540740966797, "median": 101.93342971801758, "p90": 408.62666015624995, "max": 1038.537109375, "pos_frac": 0.734375, "sample": [-76.23298645019531, -60.98480224609375, 557.2395629882812, 55.9288330078125, 582.5118408203125, -21.736074447631836, 22.53167724609375, 358.21636962890625, 165.16116333007812, 75.22891235351562, 6.2313232421875, 104.71826171875, -168.50497436523438, -19.12469482421875, -166.5634765625, -51.838134765625, 162.02078247070312, 395.9652099609375, 362.4535827636719, 1038.537109375, 353.9893798828125, -303.9912109375, 143.61302185058594, 238.50283813476562, 53.39422607421875, 208.50341796875, 75.16842651367188, 306.2649230957031, 288.6390380859375, 25.319747924804688, 129.9453125, 221.3774871826172, 67.9083251953125, 131.72952270507812, 136.7373046875, -227.56631469726562, -57.1546630859375, 301.9199523925781, 73.85763549804688, 420.5254821777344, 10.083984375, 295.59185791015625, 33.86016845703125, -125.7763671875, 64.61415100097656, 221.62936401367188, -44.98683166503906, 15.902664184570312, -89.5111083984375, 20.037933349609375, 117.70965576171875, 313.4201354980469, 212.2694854736328, 120.78974914550781, 131.65528869628906, -233.94345092773438, 357.74072265625, 226.30679321289062, -75.14474487304688, -13.242218017578125, 230.21469116210938, 277.79034423828125, 155.24435424804688, 125.0380859375, 697.2941284179688, 329.1259460449219, 283.432373046875, 518.95458984375, 71.95016479492188, 99.71302032470703, 188.25128173828125, 289.09912109375, 153.79331970214844, 172.30429077148438, 67.72978973388672, -269.44580078125, -20.80535888671875, 439.4869384765625, -9.060028076171875, 82.08842468261719, 18.281524658203125, 45.113861083984375, 27.7784423828125, 92.98745727539062, 31.483993530273438, 373.6958923339844, 219.73922729492188, -79.3743896484375, 182.91903686523438, 89.28823852539062, 933.0773315429688, 403.7559814453125, 419.9915771484375, 209.8346710205078, 48.858428955078125, 104.15383911132812, 603.1656494140625, 612.968994140625, 339.7078857421875, -38.980613708496094, -16.1314697265625, 270.3188781738281, 279.2471923828125, 88.32110595703125, -35.55357360839844, 162.04803466796875, -70.2962646484375, 294.5208740234375, -188.3450927734375, 24.88818359375, -155.7867431640625, -69.381103515625, 367.1109924316406, 183.19580078125, 121.02852630615234, 115.28192138671875, -173.45953369140625, 485.8454895019531, -13.897186279296875, -25.415855407714844, -25.83629608154297, 76.85183715820312, -29.76611328125, 16.483505249023438, 487.8192443847656, -279.2528991699219, -12.6781005859375, 69.5938720703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000465.npy"}
{"epoch": 0.9738219895287958, "step": 466, "batch_size": 128, "mean": 140.89503479003906, "std": 220.63790893554688, "min": -540.9541015625, "p10": -93.58323211669921, "median": 108.85020446777344, "p90": 385.30561523437484, "max": 953.8428955078125, "pos_frac": 0.7265625, "sample": [187.44017028808594, 114.32085418701172, 174.8109130859375, 719.8045654296875, 104.68006896972656, 574.2977905273438, -34.07000732421875, 0.0, 50.790069580078125, 48.31181335449219, 234.84109497070312, -62.55853271484375, 70.00250244140625, -0.221099853515625, 248.755126953125, 315.12127685546875, -85.82614135742188, -7.034515380859375, 242.73486328125, 129.22796630859375, 81.96902465820312, 198.55096435546875, 95.66342163085938, 165.78854370117188, 9.16693115234375, 119.7392578125, 141.09356689453125, 2.094085693359375, 272.634033203125, 72.57780456542969, -63.8450927734375, 186.48721313476562, -6.325218200683594, 166.99951171875, 329.5740966796875, -85.36198425292969, 318.3449401855469, 111.48410034179688, 485.7864685058594, -122.281005859375, 768.75, -104.87310791015625, 232.53692626953125, 161.05606079101562, 336.51361083984375, 337.55267333984375, 126.07621765136719, 432.5155029296875, 83.02879333496094, 8.040283203125, -128.95187377929688, 73.8883056640625, 3.46234130859375, -42.727806091308594, 335.98345947265625, 249.57073974609375, -0.2859344482421875, -130.67059326171875, 114.75921630859375, 203.26388549804688, 278.7507019042969, 325.851318359375, 20.4967041015625, 263.5183410644531, 221.50222778320312, 74.44427490234375, 353.6915283203125, 179.97850036621094, 669.180419921875, -141.8682861328125, 0.0, 210.0623321533203, 336.2288818359375, 550.02099609375, -53.615333557128906, 239.44163513183594, 151.02879333496094, -173.62168884277344, 162.66220092773438, 334.5491943359375, 80.37267303466797, 15.3485107421875, 901.0088500976562, -8.38885498046875, 418.8924560546875, -162.6558074951172, 25.24615478515625, -34.088897705078125, 31.20367431640625, 329.5077209472656, 126.17018127441406, -92.29417419433594, 106.21630859375, -166.82078552246094, 509.80108642578125, 34.91620635986328, 83.23135375976562, 953.8428955078125, -162.60421752929688, -540.9541015625, 97.51741790771484, 27.095794677734375, -31.770004272460938, 457.0339660644531, 70.58759307861328, 0.0, 189.8974609375, 91.2056884765625, -139.9319610595703, 357.37713623046875, 263.37481689453125, 244.49172973632812, -77.82456970214844, 86.5421371459961, 189.33154296875, -69.95278930664062, 171.529052734375, 168.0894012451172, -25.581893920898438, 113.4114990234375, -8.3372802734375, 62.192359924316406, 370.9112548828125, 171.0019073486328, -96.59103393554688, -131.99893188476562, 47.96966552734375, 421.6798095703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000466.npy"}
{"epoch": 0.9759162303664921, "step": 467, "batch_size": 128, "mean": 115.45263671875, "std": 238.17979431152344, "min": -470.3927001953125, "p10": -115.68435440063476, "median": 102.36065673828125, "p90": 339.2083984375, "max": 1202.38671875, "pos_frac": 0.6953125, "sample": [339.109375, 114.24034118652344, 330.23944091796875, 132.68251037597656, 218.03070068359375, -109.63751983642578, 77.37109375, 43.35423278808594, 446.7901611328125, 67.03704833984375, -78.84112548828125, 7.6668701171875, 207.18707275390625, -32.91815185546875, 4.26190185546875, 196.18402099609375, 142.943115234375, 350.0508117675781, -90.04022216796875, 158.6302490234375, -7.9692840576171875, 265.3328552246094, -155.8430633544922, 164.2040557861328, 311.4232482910156, 104.44181823730469, 216.1453399658203, 104.49232482910156, 80.79745483398438, -53.07330322265625, 97.60638427734375, 256.5360412597656, 458.13873291015625, 1202.38671875, 356.78857421875, 333.96575927734375, 213.41049194335938, 313.65679931640625, 49.30146789550781, -415.021240234375, 291.464111328125, 162.60391235351562, -11.204607009887695, 248.85195922851562, 213.46771240234375, 339.439453125, -14.386564254760742, -71.07766723632812, 265.3069152832031, 74.23616027832031, 54.846343994140625, -240.08152770996094, 4.9712982177734375, 115.43871307373047, -210.81192016601562, 214.24774169921875, -28.52470588684082, -465.21728515625, 252.8953399658203, 556.885986328125, -0.21929931640625, 8.097993850708008, -0.5128707885742188, -41.537139892578125, 196.65615844726562, 78.95794677734375, 42.474754333496094, -72.83536529541016, -56.12139892578125, 12.393068313598633, 115.38336181640625, -171.1611785888672, 197.44912719726562, 0.0, 69.90777587890625, -65.62387084960938, -112.49246215820312, 311.12420654296875, 22.52886962890625, -15.758056640625, 0.0, 196.81201171875, 326.8248291015625, 11.7064208984375, 181.48419189453125, 41.60688781738281, 222.3177490234375, -332.41864013671875, 212.89739990234375, 252.755615234375, -53.129241943359375, 446.8377685546875, 135.75433349609375, 123.70994567871094, 171.48458862304688, 137.33334350585938, -123.1321029663086, 193.821533203125, 19.545654296875, -42.532958984375, -189.33984375, 158.8935089111328, 280.8370361328125, -19.98236083984375, 101.5997314453125, 100.68524169921875, -470.3927001953125, -432.82574462890625, -20.112060546875, 328.2008056640625, 103.12158203125, 602.909912109375, -38.017059326171875, 145.16619873046875, 53.65100860595703, -56.38165283203125, -169.33682250976562, 168.6580810546875, 829.160888671875, 163.2645721435547, 229.4849395751953, 96.6025619506836, 380.1544189453125, 290.25201416015625, 439.0693664550781, 1004.387451171875, 46.3294677734375, -164.90899658203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000467.npy"}
{"epoch": 0.9780104712041885, "step": 468, "batch_size": 128, "mean": 84.46461486816406, "std": 232.58059692382812, "min": -1235.954833984375, "p10": -122.64704208374023, "median": 66.85630798339844, "p90": 327.23965148925777, "max": 715.1458740234375, "pos_frac": 0.6640625, "sample": [118.11328125, 256.87060546875, 254.76876831054688, -33.99329376220703, 280.73907470703125, 89.07510375976562, -77.05613708496094, 576.1841430664062, 8.192903518676758, 165.09019470214844, 95.37982177734375, -403.7249755859375, 67.55172729492188, 377.564453125, 101.6365966796875, -69.95684051513672, 59.2406005859375, 451.2442321777344, 42.26611328125, -153.748291015625, 141.33935546875, 111.16349792480469, -14.320449829101562, 36.05059814453125, -109.383544921875, 470.47296142578125, 110.02728271484375, 390.6221923828125, 55.85888671875, -7.649745941162109, 1.197998046875, -105.59573364257812, 301.28802490234375, 30.259286880493164, 261.8201599121094, -88.53450012207031, 5.2913665771484375, 602.369140625, -1235.954833984375, -36.582489013671875, 58.45982360839844, 145.99850463867188, -3.204376220703125, 172.00543212890625, -94.17190551757812, -120.32685089111328, 294.9300537109375, -77.6583251953125, 312.63360595703125, 81.2718276977539, 55.475433349609375, -7.050283432006836, -398.64373779296875, 66.160888671875, 72.76194763183594, 23.4638671875, 49.455169677734375, 152.66148376464844, -19.68920135498047, 28.042556762695312, 715.1458740234375, -37.39056396484375, -19.29218292236328, -2.4571685791015625, 621.9173583984375, -82.46487426757812, -69.482177734375, 240.61309814453125, -109.46307373046875, 253.21865844726562, 304.83343505859375, 164.869873046875, -97.16836547851562, 194.998779296875, 321.95684814453125, 96.73828125, 55.213409423828125, -114.35775756835938, -156.52786254882812, 319.6083068847656, 30.45502471923828, 426.19970703125, 120.17362213134766, 263.2568664550781, 160.00982666015625, 213.22662353515625, 10.47406005859375, -72.1493911743164, -130.56248474121094, 46.325469970703125, 167.41964721679688, 71.90779113769531, -8.94140625, 567.7180786132812, -27.2662353515625, 140.76626586914062, 116.65151977539062, 43.14927673339844, 9.569091796875, 43.1436767578125, 149.28480529785156, 318.1590576171875, 153.73904418945312, 73.40509033203125, -451.8388671875, -161.73452758789062, -183.86514282226562, 145.1605224609375, 125.890869140625, 407.0882568359375, -53.221954345703125, -128.06082153320312, 119.55792236328125, 339.5661926269531, -95.22787475585938, 217.3494873046875, 228.92120361328125, -51.846534729003906, -208.59454345703125, -173.50076293945312, 197.9306182861328, 236.25616455078125, 94.93389892578125, 283.2899169921875, -112.21945190429688, 309.4892578125, -130.60946655273438, 452.37884521484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000468.npy"}
{"epoch": 0.9801047120418848, "step": 469, "batch_size": 128, "mean": 132.7445068359375, "std": 228.1553192138672, "min": -487.4931640625, "p10": -134.061360168457, "median": 129.12811279296875, "p90": 397.63171997070305, "max": 1074.157470703125, "pos_frac": 0.7421875, "sample": [250.0780792236328, 411.4888916015625, 0.0, 81.51918029785156, 72.43023681640625, 236.67347717285156, -94.14388275146484, 124.2703857421875, -80.05474853515625, 139.21131896972656, 136.5912322998047, 530.9373779296875, 244.18267822265625, -125.81858825683594, 155.41552734375, -71.57891845703125, 545.6690673828125, 239.87594604492188, 44.114288330078125, 165.82106018066406, -33.69842529296875, 43.587677001953125, 74.73297882080078, -2.0686912536621094, -153.29449462890625, 133.98583984375, 28.67266845703125, 172.6099853515625, 243.07412719726562, 96.4053955078125, 165.17520141601562, 179.3922119140625, 421.56463623046875, 320.298095703125, 110.34590148925781, 71.13844299316406, 259.76116943359375, 34.078369140625, 390.35345458984375, 242.0032958984375, 55.99830627441406, 119.85931396484375, 12.044921875, 55.0181884765625, 288.77081298828125, -42.25787353515625, 56.34388732910156, 220.8643798828125, 534.9666137695312, 106.7933349609375, -36.15605926513672, 253.31964111328125, 267.94500732421875, 331.5863342285156, -172.30535888671875, 21.852020263671875, 71.58734130859375, 329.3487243652344, 168.0648193359375, -206.26699829101562, 25.24561309814453, -394.542236328125, 375.4423828125, -178.575439453125, -326.7585144042969, 391.69293212890625, 220.25025939941406, 107.14044189453125, -189.12442016601562, 197.7384033203125, -158.37158203125, 191.03599548339844, 723.96435546875, 333.60589599609375, -404.9781188964844, 20.3622989654541, 147.60704040527344, 280.53057861328125, 14.124298095703125, 443.3212890625, 303.3239440917969, 154.77149963378906, 109.45779418945312, 225.78245544433594, 335.5997009277344, -85.30018615722656, -487.4931640625, -26.57412338256836, 189.38380432128906, -168.7662353515625, 69.94741821289062, 199.20474243164062, -78.30795288085938, 253.7686004638672, -249.1143798828125, -300.97064208984375, -9.5751953125, -0.2965240478515625, 56.48409652709961, 300.73626708984375, -99.94245910644531, -110.06270599365234, 63.971527099609375, 1074.157470703125, -15.549789428710938, 159.96902465820312, 238.35687255859375, 96.08403778076172, 316.83935546875, 516.8062133789062, 545.7259521484375, -10.098876953125, 432.4529724121094, 172.37255859375, 298.09600830078125, 94.277587890625, 24.853378295898438, 541.9942626953125, 236.51962280273438, 227.79168701171875, 215.97769165039062, 192.03900146484375, 113.42627716064453, -114.67999267578125, 147.5425567626953, 199.30450439453125, 634.0052490234375, -54.883880615234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000469.npy"}
{"epoch": 0.9821989528795811, "step": 470, "batch_size": 128, "mean": 136.0166778564453, "std": 225.01171875, "min": -498.9411926269531, "p10": -85.27392425537109, "median": 121.513671875, "p90": 399.5470947265624, "max": 1002.9129638671875, "pos_frac": 0.78125, "sample": [-113.615478515625, 73.984130859375, 210.97271728515625, 203.81439208984375, 43.83087158203125, 16.33001708984375, 22.470130920410156, 136.35154724121094, 182.97650146484375, 196.14749145507812, -71.55548095703125, 19.0169677734375, 176.47760009765625, 207.18499755859375, 80.51422119140625, 121.78359985351562, 444.8084411621094, 45.31396484375, -120.03938293457031, 126.72958374023438, 111.70350646972656, -166.20166015625, 21.61322021484375, 59.296875, 656.781494140625, 31.170684814453125, 133.83584594726562, 358.70709228515625, 108.876953125, 179.16082763671875, 418.2421875, 95.80598449707031, -70.76309204101562, 8.22869873046875, 132.40199279785156, -498.9411926269531, 365.52569580078125, 317.4938659667969, 151.41461181640625, 16.510360717773438, 294.4400939941406, 585.0887451171875, 379.03070068359375, -47.31829833984375, 16.370391845703125, 356.4129638671875, 198.80166625976562, 106.99854278564453, 33.68939208984375, -65.59783935546875, 241.99037170410156, 309.3536682128906, 43.648712158203125, 234.84571838378906, 18.40106201171875, 252.8670654296875, 46.81044006347656, -31.372039794921875, 391.534912109375, -70.94424438476562, 281.8291320800781, 17.39098358154297, 162.79415893554688, -48.85670471191406, 276.6419982910156, -5.588356018066406, -75.25088500976562, -135.24090576171875, 56.77081298828125, 126.88787841796875, 28.08831787109375, 98.50389862060547, 225.58245849609375, 44.74602508544922, 159.1142578125, 65.05974578857422, 56.117919921875, 190.81561279296875, 284.13726806640625, -81.3431396484375, 120.23452758789062, -166.01815795898438, 171.92916870117188, 272.15509033203125, -42.08447265625, -498.401123046875, -94.44575500488281, 627.9676513671875, 277.337890625, 600.16650390625, -211.05384826660156, 1002.9129638671875, 153.08763122558594, 182.50306701660156, -16.210845947265625, -8.126708984375, 193.98361206054688, 455.46527099609375, 192.874755859375, -273.1345520019531, 531.3026123046875, 828.4462280273438, 83.7864761352539, 542.4497680664062, 224.66488647460938, 255.4462890625, 20.25971794128418, 212.6892547607422, 112.20068359375, -421.82427978515625, 16.662261962890625, -174.846923828125, 147.3125, -26.05902099609375, 223.6735382080078, 359.4134521484375, 77.63336181640625, 273.6736755371094, 143.38006591796875, 121.24374389648438, 177.5318603515625, -138.03591918945312, 149.14810180664062, 70.19195556640625, 196.39752197265625, 498.264892578125, 509.9509582519531, -3.57916259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000470.npy"}
{"epoch": 0.9842931937172775, "step": 471, "batch_size": 128, "mean": 135.33375549316406, "std": 283.0873718261719, "min": -441.99786376953125, "p10": -131.72126464843748, "median": 105.392578125, "p90": 422.85576171875, "max": 2158.806396484375, "pos_frac": 0.734375, "sample": [95.63331604003906, 421.4803466796875, 13.11273193359375, 299.4472961425781, 136.92642211914062, 336.0257568359375, 195.746826171875, 171.8434600830078, 9.229917526245117, 57.967559814453125, -422.99169921875, -45.9097900390625, 231.84716796875, 152.43177795410156, -22.648658752441406, 129.4200439453125, 209.88458251953125, 289.2174377441406, -67.65882110595703, 193.13726806640625, 12.26629638671875, 87.65814208984375, -9.709625244140625, -55.42596435546875, -110.64341735839844, 37.4454345703125, 857.5844116210938, 6.998565673828125, -441.99786376953125, 215.285400390625, 192.956787109375, 165.72756958007812, 184.95700073242188, 18.654693603515625, -249.19976806640625, 129.85061645507812, -79.05581665039062, 111.9660873413086, 97.24008178710938, -16.711181640625, 46.638214111328125, 35.50439453125, 125.73519897460938, 388.85260009765625, 6.421014785766602, 209.4308624267578, 104.25111389160156, 451.9211730957031, 153.1273193359375, 5.434013366699219, 238.68679809570312, 71.916748046875, 343.95062255859375, 199.345947265625, -334.1839599609375, 204.9971923828125, 171.10533142089844, -34.12226104736328, 32.28771209716797, 467.9928283691406, 295.0885314941406, 2158.806396484375, 3.421009063720703, -152.69097900390625, -70.0697021484375, -8.27391242980957, 28.353134155273438, 35.645477294921875, -214.32705688476562, -126.1990966796875, 197.24746704101562, 89.45167541503906, 394.34539794921875, 45.47969055175781, -204.82510375976562, 156.1178741455078, 451.2066345214844, 196.05838012695312, 268.654541015625, 65.766357421875, 99.71649169921875, 10.8330078125, -337.57720947265625, 64.74110412597656, 143.58740234375, 241.6477508544922, 106.53404235839844, 76.22640991210938, 199.29217529296875, -61.48822021484375, 486.4122314453125, 206.01055908203125, 108.61111450195312, 140.67694091796875, 35.29547119140625, 188.82791137695312, 838.7364501953125, -37.657676696777344, -186.45240783691406, -157.54110717773438, 221.59580993652344, -29.6298828125, 456.54742431640625, -12.434326171875, 258.7317810058594, 126.43769073486328, 263.5714111328125, 339.6136779785156, 167.61074829101562, 83.40640258789062, 54.107818603515625, 403.93914794921875, 292.019287109375, -234.25364685058594, 342.9351806640625, -62.80310821533203, 564.4993896484375, 0.0, 259.5274353027344, -65.0425796508789, 496.6170654296875, 590.100830078125, 426.0650634765625, -45.57635498046875, 609.0601806640625, -82.34426879882812, -161.94677734375, -144.6063232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000471.npy"}
{"epoch": 0.9863874345549738, "step": 472, "batch_size": 128, "mean": 145.8914337158203, "std": 229.65628051757812, "min": -337.16131591796875, "p10": -85.94692077636718, "median": 127.55207824707031, "p90": 473.9580535888671, "max": 922.6153564453125, "pos_frac": 0.7421875, "sample": [800.8455200195312, 150.44357299804688, -2.2542457580566406, -26.185562133789062, 17.734106063842773, 18.644134521484375, 96.60953521728516, -85.29910278320312, -14.63812255859375, 47.7645263671875, 4.223724365234375, -18.005905151367188, 463.9052429199219, 170.39404296875, -118.18467712402344, 181.9032745361328, -238.06295776367188, 130.1705322265625, -69.4349365234375, 497.41461181640625, 78.70391845703125, 139.99444580078125, 49.96836853027344, 122.33709716796875, -46.063072204589844, 334.64202880859375, 157.79025268554688, -192.00857543945312, -57.85334014892578, 312.08642578125, 396.1729736328125, 67.48839569091797, 126.29547119140625, 543.4599609375, -81.82475280761719, -46.5189208984375, 352.1776123046875, 187.06689453125, -249.06170654296875, 14.853675842285156, 188.89447021484375, 4.074462890625, 151.01254272460938, -3.72900390625, 125.39913177490234, -87.45849609375, 40.3873291015625, 134.31698608398438, 214.92291259765625, 130.53253173828125, 261.42138671875, 546.771484375, 668.396484375, -206.11868286132812, 70.47447204589844, 507.302734375, 922.6153564453125, 140.72006225585938, 76.39447021484375, 31.646699905395508, 252.82220458984375, 332.28009033203125, -63.18450927734375, -26.74342155456543, 62.29608154296875, 250.72549438476562, 192.87332153320312, 141.99713134765625, 1.0022048950195312, 47.74981689453125, 80.87418365478516, 24.626834869384766, 55.63885498046875, 546.1715087890625, 102.968017578125, 161.8419189453125, 140.0537567138672, -27.712934494018555, 317.5382995605469, 128.80868530273438, -331.7331237792969, 370.1387939453125, 510.8788757324219, 437.4346923828125, 95.55667114257812, 213.9029541015625, 280.4228820800781, 416.73089599609375, 110.30644226074219, -243.3929443359375, 162.33163452148438, -101.93084716796875, 133.4322967529297, -8.14593505859375, 324.35797119140625, 295.6376037597656, 772.5379638671875, 222.47274780273438, -29.08636474609375, 223.55601501464844, 394.5796813964844, -99.16907501220703, 272.1982727050781, 180.27560424804688, -26.387161254882812, 184.98504638671875, 4.45257568359375, 94.60452270507812, -39.5626220703125, 111.59378051757812, 523.25146484375, 264.32513427734375, 2.5903873443603516, 225.64450073242188, 410.5938415527344, -337.16131591796875, -91.27500915527344, 788.416748046875, -48.161468505859375, 177.64306640625, 520.0010986328125, 130.98944091796875, -283.0460510253906, 164.6732177734375, 155.74725341796875, 294.400146484375, -36.65507507324219, 20.847312927246094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000472.npy"}
{"epoch": 0.9884816753926702, "step": 473, "batch_size": 128, "mean": 133.216552734375, "std": 194.7994384765625, "min": -521.1180419921875, "p10": -76.30262756347656, "median": 120.00057983398438, "p90": 347.49500732421876, "max": 848.4385986328125, "pos_frac": 0.7421875, "sample": [26.06000518798828, 313.4712829589844, 103.28901672363281, -9.48444938659668, 192.505859375, -76.02288818359375, -17.3626708984375, 58.2032470703125, 138.3043212890625, -109.75665283203125, 254.82757568359375, 2.1039505004882812, -261.74053955078125, 9.304006576538086, -62.65540313720703, 52.22674560546875, -53.71728515625, 94.82762908935547, 330.99456787109375, 256.7606506347656, 15.096282958984375, 158.63699340820312, 296.9256896972656, 347.03314208984375, 86.05438232421875, 76.3201904296875, -31.240020751953125, 72.36251831054688, 103.54286193847656, 54.26387023925781, 471.03564453125, 193.4185791015625, 206.25460815429688, 228.78985595703125, -61.64545440673828, -121.11685180664062, 118.95903015136719, -22.933517456054688, 348.57269287109375, 149.16575622558594, 210.44927978515625, 145.4884033203125, 157.13156127929688, 21.51201820373535, -33.048583984375, -95.53781127929688, -188.74862670898438, 848.4385986328125, 211.99032592773438, 41.17340087890625, 5.661094665527344, 284.5674743652344, -17.069746017456055, 409.1564636230469, -80.49261474609375, 680.768310546875, 91.17398071289062, 19.128204345703125, 596.5860595703125, -172.74191284179688, 147.013916015625, 444.8569641113281, 209.39202880859375, -12.351119995117188, -20.57647705078125, 264.66839599609375, 221.62957763671875, 310.6661376953125, -42.8265380859375, 151.79119873046875, 212.96539306640625, 324.07415771484375, 83.078857421875, 141.2457275390625, -4.70683479309082, 282.755859375, 446.486083984375, -22.9774169921875, 336.81378173828125, 455.1735534667969, 98.75695037841797, -61.73493957519531, 496.28179931640625, 288.5726623535156, 158.18173217773438, 314.29693603515625, -521.1180419921875, 46.53614807128906, 330.4811096191406, -76.95535278320312, 39.67767333984375, 330.8648681640625, 18.918357849121094, 183.69627380371094, 4.6456451416015625, -111.62543487548828, 106.08466339111328, 183.22454833984375, -19.168495178222656, -203.8600311279297, 270.9501953125, 192.49612426757812, 190.49703979492188, 0.0, 244.8330078125, -14.219635009765625, -171.67422485351562, -38.8955078125, 133.70803833007812, 41.54264831542969, 223.13162231445312, 116.58865356445312, 183.18124389648438, 354.09759521484375, 212.18194580078125, 14.651947021484375, 205.59146118164062, -252.34310913085938, 119.423828125, 279.23724365234375, 424.882568359375, 448.53546142578125, 99.51913452148438, 188.92498779296875, 341.56622314453125, 261.56378173828125, 259.017333984375, 120.57733154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000473.npy"}
{"epoch": 0.9905759162303664, "step": 474, "batch_size": 128, "mean": 121.66482543945312, "std": 212.3841094970703, "min": -461.0216064453125, "p10": -113.51194610595702, "median": 93.17112731933594, "p90": 351.79901733398435, "max": 942.290771484375, "pos_frac": 0.75, "sample": [64.43893432617188, -34.6175537109375, 942.290771484375, -45.48736572265625, -18.937252044677734, -0.40179443359375, 43.743865966796875, 495.06396484375, 71.37004089355469, 73.95028686523438, 257.8262939453125, 93.03860473632812, -172.8597412109375, -63.148895263671875, 30.650482177734375, -111.67304992675781, 128.4328155517578, 23.86071014404297, 82.97067260742188, 330.1615905761719, 140.83978271484375, 157.3777618408203, 78.55111694335938, -18.667022705078125, -4.3148956298828125, -187.19833374023438, 152.5252685546875, 47.3280029296875, -91.71142578125, 279.32427978515625, 330.0614013671875, 147.91104125976562, 97.18909454345703, 215.66842651367188, 42.592193603515625, 72.64312744140625, 387.72735595703125, 38.4140625, -65.07828521728516, 233.21343994140625, 93.30364990234375, 511.0399475097656, 271.2848205566406, 869.9299926757812, 257.9429016113281, 48.39928436279297, 0.0, 26.67151641845703, 139.6202392578125, 87.9595718383789, 45.603668212890625, 27.839935302734375, -208.0675811767578, 275.52349853515625, 351.88006591796875, 33.875450134277344, -43.468421936035156, -181.6275634765625, 521.4915771484375, 233.212890625, 40.96745300292969, -20.747314453125, -104.78138732910156, 363.71240234375, 257.18438720703125, 17.320480346679688, 323.6971740722656, 269.3514404296875, -185.32818603515625, 195.53414916992188, 485.39410400390625, 116.73539733886719, -136.09466552734375, 179.80795288085938, 22.672698974609375, -244.16213989257812, 327.82281494140625, 49.46375274658203, 208.24649047851562, -32.99275207519531, -117.80270385742188, 35.70597839355469, 76.5280532836914, 234.00767517089844, 23.748403549194336, -461.0216064453125, -299.86322021484375, 351.7642822265625, -205.21824645996094, 139.79904174804688, 234.37425231933594, -344.078369140625, 169.57696533203125, -29.0498046875, 150.2500762939453, 63.779964447021484, 159.06793212890625, -8.14031982421875, 130.60116577148438, 113.3856201171875, 753.923828125, 299.9587097167969, 288.6048583984375, 168.30831909179688, 235.33721923828125, 250.60879516601562, 299.78533935546875, 429.84136962890625, 490.4607238769531, 87.80699157714844, 136.2550048828125, 317.8199157714844, -30.27227783203125, 294.48834228515625, 65.15887451171875, 189.166015625, 124.6949462890625, 425.1126708984375, 110.343017578125, 148.46434020996094, -44.21867370605469, 174.83425903320312, 49.603485107421875, -232.83346557617188, 25.278961181640625, 137.03884887695312, 57.77703857421875, 161.04507446289062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000474.npy"}
{"epoch": 0.9926701570680628, "step": 475, "batch_size": 128, "mean": 107.21458435058594, "std": 203.9332733154297, "min": -303.3690185546875, "p10": -93.44215393066403, "median": 86.36373138427734, "p90": 328.75499572753904, "max": 948.6132202148438, "pos_frac": 0.734375, "sample": [335.17718505859375, 714.578369140625, 104.52268981933594, -24.857215881347656, 209.96441650390625, -142.78463745117188, -303.3690185546875, 337.18017578125, -233.611083984375, -194.89794921875, -219.96998596191406, -54.490997314453125, 44.48876953125, 90.2775650024414, 36.784881591796875, 112.69773864746094, 286.6253356933594, 241.35964965820312, 58.552459716796875, -17.378341674804688, 235.70989990234375, -127.06884765625, 133.57374572753906, 283.3384704589844, 201.3839111328125, 314.5516357421875, 265.96466064453125, -23.40283203125, 323.854248046875, -7.011802673339844, 77.74056243896484, 85.54698181152344, 11.003021240234375, 266.8448181152344, 42.031982421875, 159.89414978027344, -164.4818115234375, 155.9189453125, -21.323715209960938, 282.330322265625, -156.9088134765625, 160.2020263671875, 223.5357666015625, 87.18048095703125, 948.6132202148438, -22.696136474609375, 81.12750244140625, -27.67242431640625, 112.18936157226562, -77.42379760742188, 29.256072998046875, 179.02188110351562, 412.42974853515625, 142.5411376953125, -1.794952392578125, 137.884765625, 207.79656982421875, -38.67216873168945, 89.15187072753906, 47.927001953125, 41.089508056640625, -13.070854187011719, 119.8735122680664, -179.0346221923828, 90.60504150390625, 108.64990234375, -69.36219787597656, 295.7875671386719, -82.51359558105469, 95.04766845703125, 20.3385009765625, 326.76702880859375, 34.463401794433594, 159.285400390625, 130.209716796875, -29.85064697265625, 10.390388488769531, 497.1050109863281, 3.417694091796875, 47.25304412841797, 45.09486389160156, 398.91650390625, 145.427978515625, -87.50692749023438, 230.41932678222656, -41.34733581542969, 146.1158905029297, 17.54571533203125, 53.725250244140625, 176.02249145507812, 36.28034210205078, 166.6765899658203, 103.68832397460938, 88.30233764648438, 67.6351318359375, 452.9322814941406, 196.49407958984375, 333.3935852050781, 730.97119140625, -10.372482299804688, -65.72496032714844, -34.8935546875, 238.50503540039062, 133.115966796875, 89.6239013671875, 259.51654052734375, -212.20928955078125, 56.893524169921875, 16.576568603515625, 23.152996063232422, 134.337646484375, 3.4533843994140625, 54.497798919677734, -289.4139709472656, 100.95181274414062, -269.1393127441406, 175.419921875, 141.83656311035156, -107.291015625, 8.50726318359375, 933.8572998046875, 435.85382080078125, 59.262821197509766, 401.40350341796875, 14.028923034667969, 117.21054077148438, -14.367948532104492, 50.697113037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000475.npy"}
{"epoch": 0.9947643979057592, "step": 476, "batch_size": 128, "mean": 113.36674499511719, "std": 231.09825134277344, "min": -360.9239196777344, "p10": -123.43398590087888, "median": 79.57023620605469, "p90": 358.06591796875, "max": 1223.39990234375, "pos_frac": 0.703125, "sample": [115.382568359375, -238.66558837890625, -63.94160461425781, -136.77911376953125, 190.73202514648438, 170.997802734375, 77.9621353149414, 52.03771209716797, 79.98678588867188, -117.71464538574219, 169.321044921875, 40.8258056640625, 489.95440673828125, 328.80035400390625, 299.92950439453125, 783.305908203125, 305.0921325683594, 431.9653015136719, 70.5465087890625, 26.34485626220703, -16.43134307861328, 20.531829833984375, 123.7266845703125, 154.53079223632812, 174.31431579589844, 115.22478485107422, 785.3675537109375, 38.75908660888672, -355.70697021484375, 452.47393798828125, 59.65216064453125, -101.99591064453125, -98.25787353515625, 252.72509765625, 363.9232482910156, 79.1536865234375, 253.63214111328125, -5.000099182128906, 39.9306640625, -59.57537841796875, 14.33331298828125, -216.45965576171875, 98.69050598144531, 84.06292724609375, 451.021484375, 233.26666259765625, 181.18919372558594, 254.21649169921875, 128.72186279296875, -207.09503173828125, 1.717376708984375, 213.509765625, -87.745361328125, -95.46640014648438, -19.56576919555664, 345.9533996582031, 287.1525573730469, -35.7044677734375, 180.0089111328125, 5.628173828125, 267.949951171875, 58.89854431152344, -41.065460205078125, 531.1544799804688, 213.60565185546875, 179.97882080078125, 204.24072265625, 72.3492431640625, 55.31440734863281, 125.308837890625, 72.671875, -235.11412048339844, 508.5924072265625, -20.482330322265625, 191.27401733398438, -24.258602142333984, 272.95672607421875, -285.46307373046875, 112.34188842773438, 271.400634765625, 63.88801574707031, 192.09194946289062, 89.77314758300781, 55.79547119140625, 87.75267028808594, 100.87291717529297, -13.238067626953125, 219.2199249267578, -224.52667236328125, 176.0931396484375, -23.109405517578125, 28.56005859375, 165.13690185546875, 14.473709106445312, 355.5556335449219, -360.9239196777344, -77.33282470703125, 224.787109375, 53.67803955078125, -27.23883819580078, -140.447265625, 57.686065673828125, 7.56585693359375, 375.0077819824219, 89.61865234375, 227.9209442138672, -224.04885864257812, 233.5009765625, -81.77774047851562, 1223.39990234375, 88.5301513671875, 312.9561462402344, 200.07208251953125, -96.2901611328125, -93.91465759277344, -84.79666137695312, 806.9027099609375, -172.10940551757812, 183.54457092285156, -22.863372802734375, 288.53179931640625, 26.305374145507812, -198.582763671875, 530.9482421875, -27.50665283203125, 194.45556640625, 75.55363464355469, -72.68157196044922], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000476.npy"}
{"epoch": 0.9968586387434555, "step": 477, "batch_size": 128, "mean": 113.29786682128906, "std": 256.9198303222656, "min": -639.30224609375, "p10": -164.83968505859374, "median": 102.07744598388672, "p90": 450.5130828857421, "max": 763.45263671875, "pos_frac": 0.6953125, "sample": [-216.74371337890625, -6.65240478515625, 176.4195556640625, -280.47705078125, 313.5500793457031, 81.48544311523438, -136.16290283203125, 21.649307250976562, -431.6302490234375, 58.70826721191406, 8.027183532714844, 670.7979736328125, -121.27987670898438, 326.0362548828125, -299.1719665527344, 34.4776611328125, 335.71295166015625, 172.99346923828125, -448.4620361328125, 302.1262512207031, -73.03790283203125, -58.07049560546875, -59.109458923339844, 443.0061340332031, 621.0785522460938, 186.1282958984375, 338.29669189453125, -7.813220977783203, 66.26341247558594, 763.45263671875, 229.13800048828125, 258.19140625, -92.62533569335938, 750.249267578125, 75.87057495117188, 26.458465576171875, 342.2313232421875, 213.43142700195312, -73.38909912109375, 0.0, 260.2475280761719, 115.48822784423828, -186.12261962890625, -73.7926025390625, 276.5787658691406, -221.40892028808594, 109.99807739257812, 468.029296875, 292.44683837890625, -456.24066162109375, -155.32571411132812, 11.235702514648438, -639.30224609375, 226.9522705078125, 57.46160888671875, 21.139984130859375, 250.28085327148438, 334.63189697265625, 190.92596435546875, 380.0352783203125, 116.53253173828125, 203.97607421875, 590.7542114257812, -340.04638671875, 177.6202392578125, 34.17932891845703, 426.69976806640625, 4.916351318359375, 99.61911010742188, -164.1043701171875, 50.54307556152344, 474.96502685546875, 2.5717926025390625, 264.65460205078125, 711.576416015625, 376.8953857421875, -131.87460327148438, -1.4194488525390625, -15.390586853027344, 104.53578186035156, -74.10920715332031, 170.41563415527344, 48.14263916015625, 109.86318969726562, -56.965362548828125, -42.44219207763672, 549.0918579101562, 130.02899169921875, 382.10699462890625, 135.8587646484375, -22.324831008911133, 176.79885864257812, 96.3453369140625, 478.37567138671875, 43.524993896484375, 226.680419921875, 11.870941162109375, 43.17657470703125, 27.65301513671875, 142.9224090576172, -543.4974365234375, 553.2166137695312, -144.59425354003906, -166.555419921875, 348.14306640625, 18.860015869140625, -144.61282348632812, -60.823333740234375, 470.75628662109375, 105.27902221679688, 104.8905029296875, 537.1707153320312, 367.3701171875, 427.294677734375, -53.28425598144531, 230.63143920898438, 258.0877990722656, -248.68484497070312, 132.74755859375, 169.33071899414062, 289.11907958984375, 198.48312377929688, 149.1591796875, 62.74892807006836, -17.953025817871094, 95.9840316772461, 167.31033325195312, -143.08450317382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000477.npy"}