Model: jackf857/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6 Source: Original Platform
682 lines
1.1 MiB
682 lines
1.1 MiB
{"epoch": 0.0, "step": 1, "batch_size": 64, "mean": -0.02287048101425171, "std": 0.42023447155952454, "min": -1.4034271240234375, "p10": -0.46674575805664065, "median": 0.04234886169433594, "p90": 0.4323463439941407, "max": 0.89263916015625, "pos_frac": 0.53125, "sample": [-0.06523895263671875, 0.436798095703125, 0.27811431884765625, -0.9194221496582031, 0.018890380859375, 0.20587158203125, 0.18878173828125, -0.3968696594238281, 0.26206207275390625, 0.2470550537109375, -0.040912628173828125, 0.4394989013671875, -0.44133758544921875, -0.39148712158203125, 0.2764854431152344, 0.89263916015625, -0.42584991455078125, -0.46125030517578125, -0.8638992309570312, -0.3508758544921875, 0.371368408203125, 0.887847900390625, -0.382904052734375, 0.36145782470703125, -0.4890003204345703, 0.052455902099609375, -0.036136627197265625, 0.23079299926757812, 0.2469482421875, 0.1643218994140625, -0.07129669189453125, 0.2790794372558594, 0.3637123107910156, -0.8916168212890625, 0.03298759460449219, -0.2790107727050781, -0.17860984802246094, 0.23892593383789062, 0.05171012878417969, -0.2564239501953125, -0.14655303955078125, 0.27777862548828125, 0.0810394287109375, -1.4034271240234375, -0.28739166259765625, -0.1489429473876953, 0.44918060302734375, 0.1693286895751953, 0.10933303833007812, -0.14766693115234375, -0.40944671630859375, -0.18532562255859375, 0.6261310577392578, -0.20856857299804688, 0.602569580078125, 0.05538177490234375, 0.1505279541015625, 0.1313800811767578, -0.006317138671875, 0.42195892333984375, -0.29936981201171875, -0.4691009521484375, 0.16705322265625, -0.5789260864257812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000001.npy"}
|
|
{"epoch": 0.0014684287812041115, "step": 2, "batch_size": 64, "mean": -0.06572240591049194, "std": 0.3523969054222107, "min": -0.9291305541992188, "p10": -0.46334152221679686, "median": -0.05502510070800781, "p90": 0.3672500610351563, "max": 1.0444793701171875, "pos_frac": 0.4375, "sample": [-0.2829437255859375, 0.3027191162109375, -0.19867706298828125, -0.3062286376953125, 0.10318756103515625, 0.20131683349609375, -0.34906005859375, 0.2802886962890625, 0.1914520263671875, -0.31072998046875, 0.08922195434570312, 0.10284614562988281, -0.03655242919921875, -0.0604095458984375, -0.06208038330078125, 0.32562255859375, -0.37982177734375, 0.2746162414550781, -0.049640655517578125, 0.3752174377441406, -0.103973388671875, 0.0699462890625, 0.36417388916015625, -0.033428192138671875, 0.37265777587890625, -0.3787078857421875, -0.6610565185546875, 0.4720420837402344, 0.47701263427734375, -0.27928924560546875, -0.44719696044921875, -0.0965118408203125, -0.7628555297851562, 0.046764373779296875, 0.06670379638671875, -0.9291305541992188, -0.7122802734375, -0.16554832458496094, 0.1485595703125, -0.07539939880371094, 0.2588920593261719, 0.039890289306640625, 0.201690673828125, 0.0623016357421875, 1.0444793701171875, -0.37696075439453125, -0.02794647216796875, -0.223297119140625, -0.35730743408203125, -0.1309051513671875, -0.3106689453125, -0.11409187316894531, -0.1669769287109375, 0.131317138671875, -0.2361297607421875, 0.4093780517578125, -0.6485977172851562, 0.36856842041015625, -0.1951904296875, -0.4702606201171875, -0.7624168395996094, 0.008928298950195312, -0.31630706787109375, 0.022550582885742188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000002.npy"}
|
|
{"epoch": 0.002936857562408223, "step": 3, "batch_size": 64, "mean": 0.009920388460159302, "std": 0.35890570282936096, "min": -1.0137100219726562, "p10": -0.3746452331542969, "median": -0.02783203125, "p90": 0.3719806671142578, "max": 0.9394073486328125, "pos_frac": 0.46875, "sample": [-0.26467132568359375, -0.356964111328125, 0.00787353515625, 0.9360504150390625, -0.2964191436767578, 0.10649681091308594, 0.4423637390136719, -0.0669403076171875, -0.032562255859375, 0.009889602661132812, -0.28409576416015625, -0.09990882873535156, 0.932281494140625, -0.023101806640625, -0.1434955596923828, 0.0839080810546875, -0.4728851318359375, -0.3758354187011719, 0.24483108520507812, -0.11363410949707031, -1.0137100219726562, 0.2391510009765625, 0.9394073486328125, 0.194793701171875, 0.2663726806640625, -0.3553047180175781, 0.18564605712890625, 0.35221099853515625, 0.10422325134277344, -0.174774169921875, 0.266937255859375, 0.6814422607421875, -0.1347942352294922, 0.134765625, -0.007045745849609375, -0.10930633544921875, -0.5578231811523438, -0.05482673645019531, -0.06429290771484375, 0.12148666381835938, -0.12294769287109375, -0.042236328125, -0.156219482421875, -0.06247711181640625, 0.02735137939453125, 0.015340805053710938, -0.04531097412109375, 0.3653450012207031, -0.15384674072265625, -0.6367835998535156, 0.1640605926513672, -0.3802165985107422, 0.1606311798095703, 0.23624420166015625, -0.1695842742919922, 0.37482452392578125, -0.3718681335449219, -0.5243682861328125, -0.140045166015625, 0.7552642822265625, 0.012342453002929688, -0.217987060546875, 0.11891937255859375, 0.18073272705078125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000003.npy"}
|
|
{"epoch": 0.004405286343612335, "step": 4, "batch_size": 64, "mean": -0.01907128095626831, "std": 0.34931251406669617, "min": -1.22222900390625, "p10": -0.4721488952636719, "median": -0.015666961669921875, "p90": 0.3651510238647461, "max": 0.9253387451171875, "pos_frac": 0.484375, "sample": [0.18079376220703125, -0.33081817626953125, -0.13544654846191406, 0.0476226806640625, 0.24208831787109375, -0.22739410400390625, 0.9253387451171875, -1.22222900390625, -0.5630836486816406, 0.0198822021484375, 0.3080005645751953, 0.6837348937988281, -0.48456573486328125, -0.03420257568359375, 0.27618408203125, -0.04583168029785156, -0.10353851318359375, 0.01659393310546875, -0.16534423828125, 0.037822723388671875, -0.1660614013671875, -0.3394012451171875, -0.1156005859375, -0.4972381591796875, 0.10091400146484375, -0.2961273193359375, 0.5940475463867188, -0.12388992309570312, 0.023321151733398438, -0.021728515625, -0.12179374694824219, -0.0110321044921875, -0.1873626708984375, -0.22484207153320312, -0.09002876281738281, 0.3572883605957031, -0.47824859619140625, 0.120086669921875, 0.06524276733398438, -0.457916259765625, 0.30242919921875, -0.17688941955566406, 0.20341873168945312, 0.0172882080078125, -0.02030181884765625, 0.19971466064453125, -0.342010498046875, 0.173736572265625, 0.011816024780273438, -0.6086044311523438, 0.295867919921875, 0.07181930541992188, 0.36852073669433594, -0.20314788818359375, 0.39978790283203125, 0.026487350463867188, 0.7148513793945312, -0.5783309936523438, 0.2876739501953125, -0.0230560302734375, 0.4710845947265625, -0.09915924072265625, -0.310577392578125, 0.041782379150390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000004.npy"}
|
|
{"epoch": 0.005873715124816446, "step": 5, "batch_size": 64, "mean": 0.041444987058639526, "std": 0.33057135343551636, "min": -0.890899658203125, "p10": -0.422476577758789, "median": 0.05960655212402344, "p90": 0.39491348266601567, "max": 0.7481155395507812, "pos_frac": 0.578125, "sample": [0.0173187255859375, 0.36215972900390625, 0.29381561279296875, -0.3009796142578125, 0.107574462890625, 0.3979644775390625, 0.3440132141113281, -0.03472709655761719, 0.21890640258789062, -0.5218505859375, -0.02246856689453125, 0.284088134765625, -0.11235809326171875, 0.4455451965332031, -0.23592376708984375, -0.3308563232421875, -0.0245513916015625, 0.08530807495117188, 0.24726104736328125, -0.0190887451171875, -0.20867538452148438, 0.6381759643554688, -0.4617424011230469, 0.22338485717773438, 0.16284751892089844, -0.13545989990234375, -0.0323638916015625, 0.500030517578125, -0.49066162109375, 0.278411865234375, 0.2608928680419922, 0.20377159118652344, 0.09161376953125, 0.4565315246582031, -0.08394241333007812, -0.0859375, 0.48076629638671875, 0.0303802490234375, 0.14144134521484375, 0.2898101806640625, 0.3505859375, 0.30921173095703125, 0.17009735107421875, -0.8291015625, 0.38779449462890625, 0.7481155395507812, -0.2441558837890625, 0.01932525634765625, -0.890899658203125, -0.2882537841796875, -0.12430381774902344, -0.14587783813476562, 0.30423736572265625, 0.06849288940429688, 0.3585700988769531, -0.0406494140625, -0.196441650390625, -0.6931991577148438, -0.46820068359375, 0.20194244384765625, 0.05072021484375, -0.0469207763671875, 0.1486968994140625, 0.042266845703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000005.npy"}
|
|
{"epoch": 0.007342143906020558, "step": 6, "batch_size": 64, "mean": -0.04951813817024231, "std": 0.41958126425743103, "min": -1.1523895263671875, "p10": -0.5275711059570313, "median": -0.05976295471191406, "p90": 0.4348419189453129, "max": 1.08441162109375, "pos_frac": 0.421875, "sample": [-0.4023895263671875, 0.0081024169921875, -0.003078460693359375, -0.4690093994140625, -0.11439895629882812, 0.97308349609375, -0.5487861633300781, -0.6870651245117188, 0.7193145751953125, -1.1042633056640625, 0.19141387939453125, 0.57684326171875, -0.502685546875, 0.2282123565673828, -0.19630050659179688, -0.5235443115234375, 0.12664031982421875, -0.06530380249023438, -0.1582489013671875, 0.2395172119140625, -0.3314018249511719, -0.5182266235351562, 0.291748046875, -0.04582786560058594, -0.30203819274902344, -0.13968658447265625, 0.248870849609375, -0.09932708740234375, -0.05422210693359375, -0.00543212890625, 0.1992034912109375, 0.28192710876464844, 0.701171875, -0.18415451049804688, -0.07865142822265625, -0.0743560791015625, 0.112548828125, -1.1523895263671875, 0.10193634033203125, -0.15077972412109375, -0.35186767578125, 0.474822998046875, -0.16613006591796875, 0.292724609375, 0.47724342346191406, 0.008802413940429688, 0.341552734375, 0.2216625213623047, 0.310577392578125, -0.4354591369628906, -0.529296875, -0.20966339111328125, -0.12174224853515625, 0.2230701446533203, 0.1937408447265625, -0.3596820831298828, 0.010143280029296875, -0.32879638671875, -0.04827880859375, 0.013824462890625, -0.6979827880859375, -0.5659332275390625, 1.08441162109375, -0.0958709716796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000006.npy"}
|
|
{"epoch": 0.00881057268722467, "step": 7, "batch_size": 64, "mean": -0.0031441152095794678, "std": 0.4035693109035492, "min": -1.224090576171875, "p10": -0.48933639526367184, "median": 0.03431892395019531, "p90": 0.4288284301757815, "max": 1.015289306640625, "pos_frac": 0.53125, "sample": [0.18746185302734375, -1.224090576171875, 0.05121421813964844, -0.18803024291992188, -0.7584762573242188, 0.4514808654785156, -0.034931182861328125, 0.6502685546875, -0.41754913330078125, 0.5682220458984375, -0.149169921875, -0.0217132568359375, -0.03990936279296875, 0.13170242309570312, -0.49320220947265625, 0.2718048095703125, 0.13867759704589844, 0.12451171875, -0.29991722106933594, -0.207061767578125, -0.2911376953125, 0.572052001953125, 0.9489173889160156, 0.08015060424804688, 0.352996826171875, -0.36980438232421875, -0.4059467315673828, -0.11310958862304688, 0.036846160888671875, -0.06490325927734375, 0.0405426025390625, -0.25136566162109375, -0.054187774658203125, -0.1462554931640625, 0.2987823486328125, 0.24590301513671875, -0.11701202392578125, 0.21442413330078125, -0.5546112060546875, -0.7730674743652344, 0.185882568359375, -0.5966796875, 0.10626983642578125, -0.7996673583984375, 0.00400543212890625, 0.2964019775390625, 0.32196044921875, 0.08468246459960938, 0.2592620849609375, 0.30267333984375, 1.015289306640625, -0.30040740966796875, -0.480316162109375, 0.2623443603515625, 0.1674652099609375, 0.24700927734375, -0.02439117431640625, 0.3759727478027344, 0.6209087371826172, 0.03179168701171875, 0.0426025390625, -0.073699951171875, -0.25350189208984375, -0.3875885009765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000007.npy"}
|
|
{"epoch": 0.010279001468428781, "step": 8, "batch_size": 64, "mean": 0.014332354068756104, "std": 0.38863858580589294, "min": -1.60589599609375, "p10": -0.3825592041015625, "median": 0.03701972961425781, "p90": 0.4730789184570315, "max": 0.8214683532714844, "pos_frac": 0.546875, "sample": [0.186248779296875, -0.050018310546875, -0.24468231201171875, 0.11299896240234375, -0.36279296875, -0.0008945465087890625, 0.2266387939453125, 0.09296417236328125, 0.41485595703125, 0.048908233642578125, 0.37328338623046875, -0.10018157958984375, -0.0004119873046875, -0.09273719787597656, 0.35573577880859375, 0.0491943359375, 0.055767059326171875, 0.0431976318359375, 0.8214683532714844, 0.28920745849609375, 0.6306610107421875, -1.60589599609375, 0.4980316162109375, 0.0870819091796875, -0.04721260070800781, -0.3194732666015625, -0.0394744873046875, 0.027246475219726562, -0.183441162109375, -0.9804153442382812, -0.04137420654296875, -0.41919708251953125, 0.03296661376953125, -0.183624267578125, -0.0474700927734375, -0.37847900390625, -0.06113433837890625, 0.05097198486328125, 0.013887405395507812, 0.13114547729492188, 0.0470428466796875, 0.40358734130859375, -0.18402099609375, -0.7345123291015625, 0.07599449157714844, -0.384307861328125, -0.38861083984375, 0.06705474853515625, 0.3489971160888672, 0.1997833251953125, 0.5926361083984375, -0.17844390869140625, 0.2194976806640625, -0.099273681640625, 0.7695770263671875, 0.29889678955078125, -0.40728759765625, -0.30939483642578125, -0.184814453125, 0.586883544921875, -0.10208892822265625, 0.6376628875732422, 0.041072845458984375, 0.2177886962890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000008.npy"}
|
|
{"epoch": 0.011747430249632892, "step": 9, "batch_size": 64, "mean": 0.013305962085723877, "std": 0.3928028345108032, "min": -0.722564697265625, "p10": -0.5825004577636719, "median": 0.025943756103515625, "p90": 0.5874847412109376, "max": 0.791961669921875, "pos_frac": 0.53125, "sample": [0.10276031494140625, -0.6230697631835938, 0.7397518157958984, 0.4614715576171875, -0.22342681884765625, -0.36392974853515625, 0.6092758178710938, 0.04821014404296875, 0.30487060546875, 0.43578338623046875, -0.40177154541015625, -0.4449462890625, 0.0558624267578125, -0.4138336181640625, -0.163421630859375, -0.63946533203125, 0.6039962768554688, -0.18720626831054688, -0.682861328125, -0.31597900390625, -0.25693511962890625, 0.07741928100585938, 0.12640380859375, -0.12660980224609375, 0.5489578247070312, 0.3689117431640625, 0.22576904296875, -0.001983642578125, -0.1429615020751953, 0.791961669921875, -0.13228607177734375, -0.07564544677734375, -0.0355987548828125, 0.09548187255859375, -0.17630577087402344, -0.5923309326171875, -0.031902313232421875, -0.037258148193359375, 0.0249786376953125, 0.6201934814453125, -0.709716796875, 0.164398193359375, -0.722564697265625, 0.02690887451171875, 0.2596282958984375, 0.0948486328125, 0.4509124755859375, 0.186065673828125, 0.2930755615234375, -0.004425048828125, -0.6156768798828125, 0.00304412841796875, -0.18581390380859375, 0.7221546173095703, 0.2699241638183594, 0.109130859375, 0.067596435546875, 0.3236541748046875, 0.219146728515625, 0.45703125, -0.5595626831054688, 0.6346282958984375, -0.52471923828125, -0.28041839599609375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000009.npy"}
|
|
{"epoch": 0.013215859030837005, "step": 10, "batch_size": 64, "mean": 0.01954329013824463, "std": 0.3827933371067047, "min": -1.011810302734375, "p10": -0.3915679931640625, "median": -0.04830169677734375, "p90": 0.5493927001953127, "max": 1.06195068359375, "pos_frac": 0.46875, "sample": [-0.11621856689453125, 0.001617431640625, -0.11187744140625, -0.19879531860351562, -0.45510292053222656, 0.30097198486328125, -0.05124664306640625, -0.1866912841796875, 0.7038497924804688, -0.615936279296875, -0.3750267028808594, 0.6574172973632812, -0.4583168029785156, 0.11907958984375, -0.194305419921875, 0.2765960693359375, -0.31919097900390625, -0.1953125, 0.08431053161621094, 0.42963218688964844, 0.6948471069335938, 0.12900161743164062, -0.0751495361328125, 0.2059612274169922, -0.14774703979492188, -0.004787445068359375, -1.011810302734375, 0.0059051513671875, 0.2818489074707031, 0.22719573974609375, -0.39410400390625, -0.14556121826171875, -0.091644287109375, 0.24053192138671875, -0.2027435302734375, 0.09796905517578125, -0.385650634765625, 0.5006904602050781, 0.3883075714111328, 0.09148406982421875, -0.30919837951660156, -0.051727294921875, -0.47834014892578125, 0.194732666015625, 0.28863525390625, 0.7072677612304688, 0.5051727294921875, -0.23343658447265625, -0.087921142578125, 0.28055572509765625, -0.3134193420410156, -0.1947307586669922, -0.242950439453125, 0.2576103210449219, -0.5859756469726562, -0.04535675048828125, 0.001220703125, -0.25905609130859375, 0.5683441162109375, 1.06195068359375, 0.8150253295898438, -0.3122062683105469, -0.2029247283935547, 0.1875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000010.npy"}
|
|
{"epoch": 0.014684287812041116, "step": 11, "batch_size": 64, "mean": -0.001968860626220703, "std": 0.34694093465805054, "min": -0.6700897216796875, "p10": -0.4605754852294922, "median": -0.056766510009765625, "p90": 0.4691604614257816, "max": 0.9034881591796875, "pos_frac": 0.4375, "sample": [0.3167915344238281, 0.9034881591796875, 0.08477592468261719, 0.0555572509765625, -0.6209030151367188, -0.05553436279296875, -0.182769775390625, -0.10219192504882812, -0.3850555419921875, 0.2967681884765625, -0.15270423889160156, -0.0579986572265625, 0.3409271240234375, 0.12384033203125, -0.6700897216796875, -0.183807373046875, 0.1179046630859375, -0.13976287841796875, 0.17529296875, 0.5274391174316406, 0.23543357849121094, 0.1297454833984375, -0.026788711547851562, -0.22332191467285156, -0.33492279052734375, 0.6355743408203125, 0.149017333984375, -0.515411376953125, -0.363616943359375, 0.3781700134277344, 0.7217254638671875, -0.5050811767578125, 0.3171195983886719, -0.029632568359375, -0.11644744873046875, 0.3865203857421875, -0.18566131591796875, 0.5665359497070312, 0.046905517578125, -0.2946510314941406, 0.16833877563476562, -0.18255996704101562, -0.4735221862792969, -0.235260009765625, 0.3545494079589844, 0.12308883666992188, -0.11865234375, -0.045490264892578125, -0.204498291015625, -0.21283721923828125, -0.3081836700439453, -0.4990234375, 0.2822265625, -0.4612312316894531, 0.2539100646972656, -0.142181396484375, -0.08632278442382812, -0.45904541015625, 0.0856475830078125, 0.50457763671875, 0.7306365966796875, -0.19318389892578125, -0.10906219482421875, -0.2611083984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000011.npy"}
|
|
{"epoch": 0.016152716593245228, "step": 12, "batch_size": 64, "mean": 0.008413076400756836, "std": 0.39482975006103516, "min": -1.2481536865234375, "p10": -0.4681068420410156, "median": 0.04549407958984375, "p90": 0.4585433959960939, "max": 0.8097076416015625, "pos_frac": 0.515625, "sample": [0.8097076416015625, -0.02996826171875, 0.0788421630859375, -1.2481536865234375, -0.47261810302734375, 0.01214599609375, -0.345062255859375, 0.7064285278320312, -0.10572052001953125, 0.373565673828125, 0.28864288330078125, -0.13037109375, -0.9254913330078125, -0.31983184814453125, -0.1049346923828125, 0.6967849731445312, 0.4721527099609375, -0.072601318359375, 0.2953643798828125, -0.157257080078125, -0.0516815185546875, -0.2646026611328125, 0.286407470703125, -0.8692169189453125, -0.1784820556640625, 0.13070297241210938, -0.23486328125, 0.264190673828125, -0.09147071838378906, 0.24249267578125, -0.9908599853515625, 0.6337738037109375, 0.1060028076171875, 0.12764739990234375, -0.2010955810546875, 0.09896469116210938, -0.04258918762207031, -0.48474884033203125, 0.2986297607421875, -0.22620773315429688, -0.45758056640625, -0.04803466796875, -0.03350067138671875, 0.4829063415527344, 0.34754180908203125, 0.28791046142578125, 0.235382080078125, 0.1548614501953125, 0.12063217163085938, 0.1895904541015625, -0.05248832702636719, 0.3434925079345703, -0.06131744384765625, 0.5325431823730469, -0.6011810302734375, 0.426788330078125, 0.0912933349609375, 0.08793258666992188, -0.06151008605957031, -0.17246055603027344, 0.21590423583984375, 0.12208938598632812, -0.13592529296875, 0.14894866943359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000012.npy"}
|
|
{"epoch": 0.01762114537444934, "step": 13, "batch_size": 64, "mean": 0.0036628246307373047, "std": 0.40722933411598206, "min": -0.751190185546875, "p10": -0.47714462280273434, "median": -0.0511016845703125, "p90": 0.46743392944335943, "max": 1.41864013671875, "pos_frac": 0.390625, "sample": [-0.70074462890625, -0.2575054168701172, -0.15354156494140625, -0.22341156005859375, -0.0339508056640625, 0.5087203979492188, -0.48380279541015625, -0.04888153076171875, 0.5263519287109375, 0.2803993225097656, -0.187347412109375, 0.809722900390625, -0.22524261474609375, -0.0987548828125, -0.007930755615234375, 1.313262939453125, -0.0579376220703125, 0.473052978515625, -0.1785869598388672, 0.45432281494140625, -0.11025810241699219, -0.0412750244140625, 0.35140228271484375, -0.08191680908203125, 0.2704963684082031, -0.46160888671875, 0.000507354736328125, 0.0062427520751953125, -0.3232746124267578, -0.5293731689453125, -0.4369659423828125, -0.23146820068359375, 0.24437332153320312, 0.21852874755859375, -0.751190185546875, -0.239044189453125, 0.072784423828125, -0.09860801696777344, -0.28020286560058594, -0.62554931640625, 0.3430633544921875, -0.013050079345703125, 0.214691162109375, 0.10726165771484375, -0.05332183837890625, -0.04659461975097656, -0.06548690795898438, -0.1615886688232422, 0.121368408203125, -0.202911376953125, -0.42778778076171875, 0.2971038818359375, -0.0759429931640625, -0.11721992492675781, 0.08463287353515625, -0.204681396484375, 0.6038932800292969, 0.27667236328125, 1.41864013671875, -0.514251708984375, -0.617279052734375, 0.246734619140625, -0.0275421142578125, 0.38622283935546875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000013.npy"}
|
|
{"epoch": 0.01908957415565345, "step": 14, "batch_size": 64, "mean": 0.005518555641174316, "std": 0.47487321496009827, "min": -1.352447509765625, "p10": -0.5095542907714843, "median": 0.041744232177734375, "p90": 0.5968893051147464, "max": 1.1178970336914062, "pos_frac": 0.515625, "sample": [-0.3084869384765625, 0.07377433776855469, -0.04345512390136719, -0.17023849487304688, 0.75482177734375, -0.2530364990234375, -1.214141845703125, 0.3788909912109375, -0.43585968017578125, 0.3391399383544922, 0.2960357666015625, -0.088897705078125, -0.305755615234375, 0.7983932495117188, 0.0971527099609375, -0.4330291748046875, -0.43476295471191406, -0.1683197021484375, 0.8887481689453125, -0.08681869506835938, 0.17547607421875, -0.3812255859375, -0.2132110595703125, 0.5271987915039062, 0.665008544921875, 0.17913436889648438, -1.147552490234375, -0.13016510009765625, 0.188873291015625, 0.4774932861328125, 0.3827362060546875, -0.1576366424560547, 0.0671234130859375, 0.22649002075195312, 0.1585521697998047, -0.12084197998046875, 0.2784080505371094, -1.352447509765625, 0.07865715026855469, 0.4608039855957031, 0.402313232421875, -0.18952178955078125, -0.55657958984375, 0.28415679931640625, 0.28514862060546875, 0.13380050659179688, 0.6267566680908203, -0.3297576904296875, -0.242462158203125, -0.22186660766601562, -0.7607955932617188, 0.01636505126953125, 1.1178970336914062, 0.23718643188476562, 0.2491302490234375, 0.8247222900390625, -0.08817291259765625, -0.000762939453125, -0.5943527221679688, -0.3585357666015625, -0.5411376953125, 0.06816482543945312, -0.15386199951171875, 0.09832382202148438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000014.npy"}
|
|
{"epoch": 0.020558002936857563, "step": 15, "batch_size": 64, "mean": 0.03296104073524475, "std": 0.3217886984348297, "min": -0.7119598388671875, "p10": -0.35047359466552735, "median": 0.0087890625, "p90": 0.4254035949707031, "max": 0.8148193359375, "pos_frac": 0.53125, "sample": [-0.020669937133789062, 0.8148193359375, 0.4349250793457031, -0.034229278564453125, 0.2262115478515625, 0.08913612365722656, -0.25541114807128906, 0.3095245361328125, 0.024806976318359375, -0.01284027099609375, 0.4239501953125, -0.1356353759765625, 0.5157699584960938, -0.5216445922851562, 0.1716747283935547, -0.2382659912109375, 0.0679779052734375, 0.159576416015625, 0.3221607208251953, -0.09491729736328125, 0.2958526611328125, -0.4338836669921875, -0.3541259765625, 0.0120391845703125, -0.5914669036865234, -0.0056304931640625, 0.418853759765625, 0.0055389404296875, -0.6413040161132812, -0.01384735107421875, -0.058013916015625, -0.7119598388671875, 0.09702301025390625, -0.1383380889892578, -0.11653900146484375, 0.547760009765625, -0.24994659423828125, 0.7084503173828125, 0.27947235107421875, -0.3419513702392578, -0.1335124969482422, 0.07052803039550781, 0.21637344360351562, 0.17806625366210938, 0.42425537109375, -0.2149505615234375, 0.07549858093261719, -0.12213516235351562, 0.2855377197265625, 0.21895599365234375, -0.0030059814453125, -0.1833038330078125, 0.0483551025390625, 0.08586883544921875, 0.15399932861328125, 0.42589569091796875, -0.33026123046875, -0.04662322998046875, 0.1272430419921875, 0.0016632080078125, -0.2795829772949219, -0.47882843017578125, -0.0970611572265625, 0.73162841796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000015.npy"}
|
|
{"epoch": 0.022026431718061675, "step": 16, "batch_size": 64, "mean": 0.0060617029666900635, "std": 0.36051100492477417, "min": -0.8533859252929688, "p10": -0.4458564758300781, "median": 0.029918670654296875, "p90": 0.45743865966796876, "max": 0.9663467407226562, "pos_frac": 0.53125, "sample": [0.0023632049560546875, 0.6439056396484375, -0.30841064453125, 0.01692962646484375, -0.12264633178710938, 0.3026771545410156, -0.5107269287109375, -0.1572437286376953, -0.04984283447265625, 0.9663467407226562, 0.458221435546875, 0.3139495849609375, -0.3464622497558594, 0.1256561279296875, -0.017889022827148438, 0.04290771484375, 0.1665496826171875, -0.5098190307617188, -0.27667999267578125, -0.3425750732421875, 0.24886322021484375, 0.2652435302734375, 0.182586669921875, 0.2778167724609375, 0.13664627075195312, -0.2859363555908203, -0.8489761352539062, -0.28528594970703125, -0.1976776123046875, -0.2404308319091797, 0.4180107116699219, 0.4865455627441406, 0.17147445678710938, -0.5732955932617188, -0.2758331298828125, 0.30838775634765625, -0.10068511962890625, -0.14837646484375, -0.13399505615234375, -0.21811676025390625, 0.0437774658203125, 0.25844573974609375, -0.10087013244628906, 0.1712646484375, 0.28681182861328125, 0.6341056823730469, 0.22898101806640625, -0.1859416961669922, -0.0448455810546875, 0.3148040771484375, 0.1797008514404297, 0.1867523193359375, -0.6900177001953125, -0.3019676208496094, 0.09296417236328125, -0.44832611083984375, -0.1976299285888672, 0.4556121826171875, 0.4639778137207031, 0.108306884765625, 0.473876953125, -0.440093994140625, 0.16746902465820312, -0.8533859252929688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000016.npy"}
|
|
{"epoch": 0.023494860499265784, "step": 17, "batch_size": 64, "mean": 0.02691650390625, "std": 0.3530113995075226, "min": -1.1724700927734375, "p10": -0.38844985961914064, "median": 0.03063201904296875, "p90": 0.4480628967285156, "max": 0.89154052734375, "pos_frac": 0.546875, "sample": [-0.03675079345703125, 0.054904937744140625, -0.024585723876953125, -0.05672454833984375, -0.7466278076171875, -0.1556072235107422, 0.0867919921875, -1.1724700927734375, 0.2760162353515625, -0.049160003662109375, -0.0365142822265625, -0.06939697265625, 0.3920440673828125, -0.205230712890625, -0.32386016845703125, 0.14067840576171875, -0.38965606689453125, 0.44664764404296875, -0.021244049072265625, 0.104034423828125, 0.2582664489746094, 0.13849639892578125, -0.17898941040039062, -0.1231689453125, -0.156280517578125, -0.1873016357421875, 0.142669677734375, 0.07014846801757812, 0.7144775390625, 0.004730224609375, -0.4593963623046875, 0.49561309814453125, 0.44866943359375, 0.23615264892578125, -0.04175758361816406, 0.0041046142578125, -0.3854331970214844, 0.10550880432128906, 0.1796131134033203, 0.28436279296875, 0.6281509399414062, 0.0134735107421875, 0.5823268890380859, 0.08005142211914062, -0.3856353759765625, 0.04779052734375, 0.20325088500976562, -0.4483928680419922, 0.071563720703125, -0.1476287841796875, 0.13573837280273438, -0.097442626953125, 0.3319244384765625, 0.23317718505859375, -0.18657684326171875, 0.725372314453125, 0.89154052734375, 0.4045219421386719, -0.4252777099609375, -0.5621719360351562, 0.149139404296875, -0.3734588623046875, -0.079437255859375, 0.16688156127929688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000017.npy"}
|
|
{"epoch": 0.024963289280469897, "step": 18, "batch_size": 64, "mean": 0.07164624333381653, "std": 0.3461942970752716, "min": -0.9116668701171875, "p10": -0.3333105087280273, "median": 0.060047149658203125, "p90": 0.5149467468261719, "max": 0.7901458740234375, "pos_frac": 0.578125, "sample": [0.0531005859375, -0.6098480224609375, 0.7901458740234375, -0.025299072265625, -0.16599273681640625, -0.00589752197265625, 0.16519927978515625, -0.9116668701171875, 0.114227294921875, 0.48645591735839844, -0.18121337890625, -0.10594558715820312, 0.1854705810546875, 0.3761253356933594, -0.0194854736328125, -0.16150665283203125, 0.13475990295410156, 0.08178329467773438, 0.325103759765625, 0.0305633544921875, -0.2684364318847656, -0.048858642578125, 0.53656005859375, -0.16498374938964844, 0.565826416015625, 0.3805389404296875, 0.3178863525390625, 0.39284515380859375, -0.19747161865234375, 0.021495819091796875, 0.5014724731445312, 0.1455249786376953, 0.06699371337890625, -0.47968292236328125, 0.2021331787109375, -0.10802459716796875, 0.37195587158203125, 0.34769630432128906, 0.2068634033203125, -0.2218475341796875, 0.3148460388183594, 0.7787094116210938, -0.497802734375, 0.5576705932617188, 0.520721435546875, 0.4287109375, 0.18225860595703125, -0.32146644592285156, -0.25830078125, 0.759033203125, 0.029933929443359375, -0.12502288818359375, 0.10504913330078125, -0.176605224609375, -0.03163909912109375, -0.139739990234375, 0.3853302001953125, -0.1422271728515625, -0.5624847412109375, 0.25562286376953125, 0.034328460693359375, -0.410491943359375, -0.33838653564453125, 0.11274528503417969], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000018.npy"}
|
|
{"epoch": 0.02643171806167401, "step": 19, "batch_size": 64, "mean": 0.11977550387382507, "std": 0.36248427629470825, "min": -0.8752899169921875, "p10": -0.30208473205566405, "median": 0.13833236694335938, "p90": 0.4846101760864258, "max": 1.2284088134765625, "pos_frac": 0.671875, "sample": [0.266876220703125, 0.009336471557617188, -0.173553466796875, 0.2487945556640625, 0.29845428466796875, -0.627044677734375, 0.21478652954101562, 0.38899993896484375, 0.139617919921875, 0.4509735107421875, 0.21559524536132812, -0.106597900390625, 0.046459197998046875, 0.35428619384765625, 0.3861236572265625, -0.36865234375, -0.4646453857421875, 0.3636627197265625, -0.8752899169921875, 0.53314208984375, -0.3269500732421875, 0.2658233642578125, 0.3367328643798828, 0.0255126953125, -0.12969970703125, 0.2603321075439453, 0.41436004638671875, 0.14402389526367188, -0.16756820678710938, 0.13704681396484375, -0.04852294921875, 0.4814739227294922, 0.021505355834960938, 0.01725006103515625, 0.43192291259765625, -0.3100395202636719, -0.2835235595703125, -0.028827667236328125, 0.48595428466796875, -0.542022705078125, 0.22374725341796875, 1.2284088134765625, -0.17754364013671875, -0.228607177734375, 0.2278766632080078, 0.18525314331054688, 0.1756439208984375, 0.54046630859375, 0.0181884765625, -0.018289566040039062, 0.7928695678710938, -0.27315521240234375, 0.050838470458984375, -0.00441741943359375, 0.7743072509765625, 0.08734512329101562, 0.27048492431640625, 0.08044815063476562, 0.1317138671875, -0.2266845703125, 0.907196044921875, 0.3613853454589844, -0.27326202392578125, 0.32530975341796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000019.npy"}
|
|
{"epoch": 0.027900146842878122, "step": 20, "batch_size": 64, "mean": 0.037224918603897095, "std": 0.31069859862327576, "min": -0.7858123779296875, "p10": -0.2971576690673828, "median": 0.01859760284423828, "p90": 0.39748725891113285, "max": 0.8305816650390625, "pos_frac": 0.546875, "sample": [-0.019475936889648438, 0.12944602966308594, -0.3740043640136719, -0.20684814453125, 0.5102005004882812, 0.23392677307128906, 0.139678955078125, 0.2696685791015625, 0.19988250732421875, 0.8305816650390625, -0.7858123779296875, -0.0273895263671875, 0.4850196838378906, 0.011697769165039062, -0.063385009765625, 0.10109710693359375, -0.3998985290527344, 0.3987312316894531, -0.5649032592773438, 0.0560302734375, 0.6360931396484375, -0.2996673583984375, 0.1725006103515625, 0.173583984375, -0.03093719482421875, -0.11600685119628906, 0.0007476806640625, 0.13135910034179688, 0.00872039794921875, -0.2887001037597656, 0.033725738525390625, -0.204193115234375, -0.2913017272949219, 0.25325584411621094, -0.12002372741699219, -0.07648468017578125, -0.120849609375, -0.71240234375, 0.11611557006835938, -0.4571075439453125, -0.16782379150390625, 0.09300994873046875, -0.0488739013671875, 0.385467529296875, -0.0462799072265625, 0.20902252197265625, 0.0254974365234375, 0.585174560546875, 0.35003662109375, -0.16139984130859375, -0.1575927734375, -0.1634063720703125, -0.09826850891113281, 0.0255584716796875, 0.3792266845703125, 0.10605621337890625, -0.224273681640625, 0.20124053955078125, 0.11553764343261719, 0.11795997619628906, 0.7605667114257812, 0.39458465576171875, -0.0146026611328125, -0.016695022583007812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000020.npy"}
|
|
{"epoch": 0.02936857562408223, "step": 21, "batch_size": 64, "mean": 0.1729956418275833, "std": 0.30118662118911743, "min": -0.4460735321044922, "p10": -0.18678016662597655, "median": 0.13497495651245117, "p90": 0.5692962646484376, "max": 0.966033935546875, "pos_frac": 0.640625, "sample": [0.2310791015625, -0.03592109680175781, 0.57745361328125, 0.966033935546875, 0.31394195556640625, 0.14787578582763672, -0.07358360290527344, -0.27984046936035156, 0.5453948974609375, -0.3088226318359375, 0.744903564453125, 0.079864501953125, 0.19910240173339844, 0.12207412719726562, 0.4605216979980469, 0.31070709228515625, 0.06065177917480469, -0.22544479370117188, -0.06545448303222656, 0.26410675048828125, 0.550262451171875, 0.0397796630859375, -0.02471923828125, -0.036083221435546875, 0.36062049865722656, 0.7228851318359375, -0.04364013671875, -0.4460735321044922, 0.30698394775390625, -0.1529693603515625, -0.01038360595703125, 0.4307708740234375, 0.4629173278808594, 0.38949012756347656, 0.0928802490234375, -0.07115936279296875, -0.19247817993164062, 0.10441207885742188, 0.0963592529296875, 0.248931884765625, -0.07373046875, 0.5321006774902344, -0.0608062744140625, 0.15498733520507812, 0.15758895874023438, -0.11738777160644531, 0.6155471801757812, 0.35201263427734375, 0.22694778442382812, -0.01081085205078125, 0.23931503295898438, 0.673553466796875, 0.7034797668457031, 0.43505859375, 0.04973793029785156, -0.09232330322265625, -0.33770751953125, -0.23223876953125, 0.4673004150390625, 0.4601860046386719, 0.19353485107421875, -0.17348480224609375, 0.07558441162109375, -0.030155181884765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000021.npy"}
|
|
{"epoch": 0.030837004405286344, "step": 22, "batch_size": 64, "mean": 0.21139344573020935, "std": 0.4874153733253479, "min": -0.899200439453125, "p10": -0.32089462280273434, "median": 0.10549068450927734, "p90": 0.9421020507812502, "max": 1.478607177734375, "pos_frac": 0.6875, "sample": [0.0284881591796875, 0.61968994140625, 0.041141510009765625, 0.3423633575439453, 0.0125274658203125, -0.899200439453125, -0.325927734375, 0.3119831085205078, 0.26621055603027344, 0.882843017578125, 1.004180908203125, 1.478607177734375, -0.045833587646484375, 0.7203521728515625, -0.16098785400390625, 0.08016204833984375, -0.20803070068359375, 0.15627288818359375, -0.30915069580078125, -0.57794189453125, 0.031681060791015625, 1.151397705078125, -0.661163330078125, -0.295135498046875, -0.05263328552246094, 0.3182353973388672, 1.0573348999023438, 0.36827850341796875, 0.41526031494140625, 0.3820343017578125, 0.12284278869628906, 0.4421844482421875, 0.010284423828125, -0.17964553833007812, 0.017765045166015625, 0.8802642822265625, 0.31865692138671875, -0.337982177734375, -0.34351348876953125, 0.8099327087402344, 0.113189697265625, 0.6470222473144531, 0.06910514831542969, 0.3584861755371094, -0.13567161560058594, 0.3280792236328125, -0.15753936767578125, 0.10584259033203125, 1.075653076171875, 0.38613128662109375, 0.09227180480957031, 0.967498779296875, -0.22530555725097656, -0.2115612030029297, -0.34503173828125, -0.210174560546875, 0.5682220458984375, 0.10513877868652344, -0.1894989013671875, 0.0386962890625, 0.189697265625, 1.3595504760742188, 0.648284912109375, 0.07726478576660156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000022.npy"}
|
|
{"epoch": 0.032305433186490456, "step": 23, "batch_size": 64, "mean": 0.24120870232582092, "std": 0.434193879365921, "min": -0.9698486328125, "p10": -0.22079410552978515, "median": 0.2136240005493164, "p90": 0.6978776931762696, "max": 1.576629638671875, "pos_frac": 0.734375, "sample": [-0.00316619873046875, 0.17895889282226562, 0.985565185546875, 0.2896232604980469, 0.5449447631835938, 0.7516517639160156, 0.3540153503417969, -0.2190704345703125, -0.5006637573242188, -0.9698486328125, 1.576629638671875, 0.67803955078125, 1.014678955078125, -0.24578094482421875, 0.6282157897949219, 0.022083282470703125, -0.05655097961425781, 0.300048828125, 0.22259521484375, 0.3983612060546875, 0.008056640625, 0.13992691040039062, 0.50079345703125, 0.07776260375976562, -0.22153282165527344, -0.4740447998046875, -0.04593658447265625, -0.14210128784179688, -0.12957763671875, 0.276611328125, 0.001544952392578125, 0.5300140380859375, -0.036834716796875, 0.6332015991210938, 0.3314247131347656, 0.1319732666015625, 0.02962493896484375, 0.10887908935546875, 0.2410411834716797, 0.4496726989746094, 0.45618438720703125, 0.21634864807128906, 0.2714996337890625, 0.21089935302734375, 0.3682670593261719, 1.39837646484375, 0.103302001953125, -0.42108154296875, 0.7032089233398438, -0.5230712890625, 0.4907855987548828, 0.3508892059326172, 0.361114501953125, -0.001056671142578125, 0.07525062561035156, -0.087066650390625, 0.6801605224609375, 0.849853515625, -0.06357383728027344, 0.123748779296875, 0.596588134765625, 0.04644012451171875, 0.18402099609375, 0.6854381561279297], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000023.npy"}
|
|
{"epoch": 0.033773861967694566, "step": 24, "batch_size": 64, "mean": 0.20983177423477173, "std": 0.4076336920261383, "min": -0.8157272338867188, "p10": -0.30777587890624997, "median": 0.16245651245117188, "p90": 0.7602710723876955, "max": 1.0697784423828125, "pos_frac": 0.6875, "sample": [0.471771240234375, 0.7869529724121094, 0.04769706726074219, -0.045986175537109375, 0.3427848815917969, 0.13445472717285156, 0.959808349609375, 0.1791534423828125, 0.0522003173828125, -0.12755584716796875, 0.32587242126464844, -0.01171112060546875, -0.05100822448730469, 1.0590171813964844, 0.6477737426757812, 0.024715423583984375, 0.14078521728515625, -0.017333984375, 0.31237030029296875, -0.019176483154296875, 0.4630165100097656, -0.532989501953125, -0.05987358093261719, 0.24164581298828125, 1.0697784423828125, 0.6002883911132812, 0.3441753387451172, -0.487396240234375, 0.5147857666015625, 0.13048553466796875, 0.994964599609375, 0.92120361328125, 0.21672439575195312, -0.8157272338867188, 0.34075927734375, -0.711181640625, 0.69244384765625, 0.0359344482421875, -0.33333587646484375, 0.14575958251953125, 0.5309600830078125, 0.13408851623535156, -0.3145599365234375, 0.18390655517578125, 0.5099868774414062, 0.7886199951171875, 0.188140869140625, 0.12054634094238281, -0.018310546875, 0.12322044372558594, 0.43019866943359375, 0.38898468017578125, -0.2919464111328125, 0.02666473388671875, 0.4509735107421875, -0.0247039794921875, 0.45961570739746094, -0.00632476806640625, 0.3179969787597656, -0.5537643432617188, 0.35443878173828125, -0.0076446533203125, -0.043914794921875, 0.6980133056640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000024.npy"}
|
|
{"epoch": 0.03524229074889868, "step": 25, "batch_size": 64, "mean": 0.27194616198539734, "std": 0.42592471837997437, "min": -0.6401596069335938, "p10": -0.19285964965820312, "median": 0.23883438110351562, "p90": 0.8800647735595709, "max": 1.4317626953125, "pos_frac": 0.71875, "sample": [0.2737998962402344, 0.09772491455078125, 0.640350341796875, 0.6315956115722656, 0.02692413330078125, -0.011217117309570312, 0.03791046142578125, 0.9381103515625, 1.242523193359375, -0.030912399291992188, -0.4272613525390625, 0.347381591796875, -0.07033729553222656, 0.29644012451171875, 1.4317626953125, -0.2891654968261719, -0.05575370788574219, 0.9561309814453125, -0.09391403198242188, 0.31171417236328125, -0.348388671875, -0.023983001708984375, 0.2682037353515625, 0.27873992919921875, -0.3926239013671875, 0.4989471435546875, -0.04483795166015625, 1.13116455078125, -0.6401596069335938, -0.025865554809570312, 0.6761703491210938, -0.18161773681640625, 0.21514129638671875, 0.301849365234375, 0.071014404296875, 0.11110687255859375, 0.18798446655273438, 1.0606536865234375, 0.33831787109375, 0.0308990478515625, 0.36948204040527344, 0.7270889282226562, 0.9826583862304688, 0.3526878356933594, 0.4261322021484375, 0.5613937377929688, 0.46431732177734375, 0.136962890625, 0.7360305786132812, 0.4580039978027344, 0.1107940673828125, 0.3115043640136719, 0.2625274658203125, 0.081695556640625, 0.7446250915527344, -0.03350257873535156, -0.3710784912109375, 0.7050437927246094, 0.05434417724609375, 0.039966583251953125, -0.12934494018554688, 0.6984176635742188, -0.1976776123046875, 0.14595794677734375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000025.npy"}
|
|
{"epoch": 0.03671071953010279, "step": 26, "batch_size": 64, "mean": 0.4091459810733795, "std": 0.5727108120918274, "min": -0.68951416015625, "p10": -0.2008462905883789, "median": 0.274200439453125, "p90": 1.2891510009765628, "max": 1.9343338012695312, "pos_frac": 0.75, "sample": [0.2655181884765625, 0.11794090270996094, 0.24736404418945312, 0.5816059112548828, 1.1949462890625, 0.30075836181640625, 0.8243331909179688, 1.331390380859375, 0.14476776123046875, -0.133941650390625, 0.989532470703125, 0.6525115966796875, 1.603118896484375, 0.23554229736328125, 0.5269927978515625, -0.496063232421875, 0.32384490966796875, 0.2828826904296875, 1.322998046875, 0.03543853759765625, 0.2635040283203125, 0.12360763549804688, 0.2259674072265625, 0.6818313598632812, -0.68951416015625, 1.5030975341796875, -0.2367095947265625, -0.43170166015625, -0.14036941528320312, 1.0065383911132812, 0.0788116455078125, -0.19031524658203125, 0.4880847930908203, 0.41301727294921875, -0.0455474853515625, -0.16357421875, 1.465728759765625, -0.0927581787109375, -0.20111465454101562, 1.6274490356445312, 0.6938705444335938, 0.8063583374023438, 1.9343338012695312, -0.21584320068359375, 0.38959503173828125, 1.1301193237304688, -0.2659034729003906, 0.3905181884765625, 0.03935813903808594, -0.05761528015136719, 0.09880256652832031, 0.11008453369140625, -0.20022010803222656, 0.2152557373046875, -0.00720977783203125, 0.29345703125, 0.58856201171875, 1.1098403930664062, 1.210174560546875, 0.21191787719726562, 0.49086761474609375, 0.026683807373046875, 0.484161376953125, 0.6706581115722656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000026.npy"}
|
|
{"epoch": 0.0381791483113069, "step": 27, "batch_size": 64, "mean": 0.5205891132354736, "std": 0.5379472970962524, "min": -0.45485687255859375, "p10": 0.0409690856933594, "median": 0.4403648376464844, "p90": 1.139288711547852, "max": 2.2773818969726562, "pos_frac": 0.90625, "sample": [0.23156356811523438, 0.5590934753417969, 0.30976104736328125, 0.46871185302734375, 0.5364799499511719, 1.4906692504882812, 0.412017822265625, 0.5542755126953125, 1.9010772705078125, -0.31926918029785156, 0.8286495208740234, 0.06208038330078125, 1.0350914001464844, 0.16236305236816406, 0.6884613037109375, 0.7024993896484375, 2.2773818969726562, 0.6060638427734375, 0.295562744140625, 0.6916122436523438, 0.51763916015625, 0.209625244140625, 0.26685333251953125, -0.019525527954101562, 0.6040420532226562, 0.2540740966796875, 0.6293411254882812, 0.08988189697265625, 0.2888069152832031, 2.09783935546875, 0.09366607666015625, 0.12372207641601562, -0.06351280212402344, 0.780670166015625, 0.30716705322265625, 0.6155548095703125, 0.6752815246582031, 1.0059280395507812, 0.10901641845703125, 0.6243209838867188, 0.10387039184570312, 0.2124786376953125, 0.30609130859375, 0.96771240234375, 0.61773681640625, 0.8827972412109375, -0.44831085205078125, 0.39795494079589844, 1.1839447021484375, 0.8003692626953125, 1.318359375, -0.45485687255859375, 0.3363533020019531, 0.2187347412109375, 0.849029541015625, 0.10149383544921875, 1.5052032470703125, -0.2876777648925781, 0.7649364471435547, 0.03192138671875, 0.107391357421875, 0.6000900268554688, 0.3124237060546875, 0.18511962890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000027.npy"}
|
|
{"epoch": 0.039647577092511016, "step": 28, "batch_size": 64, "mean": 0.2837350368499756, "std": 0.5332615971565247, "min": -1.5756072998046875, "p10": -0.2649360656738281, "median": 0.1985301971435547, "p90": 0.8296308517456056, "max": 2.064788818359375, "pos_frac": 0.765625, "sample": [0.4770164489746094, 0.3634033203125, 0.18700408935546875, 0.20623779296875, 0.7384586334228516, -0.275238037109375, 0.13275146484375, 0.3595161437988281, -0.1421794891357422, 0.15076065063476562, -0.058868408203125, 0.129302978515625, 2.064788818359375, 0.30414581298828125, 0.8355789184570312, -0.018430709838867188, 0.044162750244140625, 0.3194427490234375, 0.17934036254882812, 0.22428131103515625, 0.04412078857421875, -0.47998809814453125, 0.1356201171875, 1.09228515625, -0.38360595703125, 0.54351806640625, 0.3564796447753906, 0.6739578247070312, 1.4432830810546875, 0.21640777587890625, 0.19483566284179688, 0.5838165283203125, -0.05048370361328125, -0.2661285400390625, 0.70770263671875, 0.16875457763671875, 0.8157520294189453, 0.7850189208984375, 0.19841384887695312, -0.3521270751953125, 0.118499755859375, 0.42968177795410156, 0.08764076232910156, 0.94317626953125, 0.733673095703125, 0.41192626953125, 0.6323776245117188, 0.013925552368164062, 0.6877937316894531, 0.0460052490234375, 0.614593505859375, -0.26215362548828125, -0.1841278076171875, -0.10555267333984375, -0.10009002685546875, 0.19864654541015625, 0.633026123046875, 0.8534698486328125, 1.298492431640625, 0.11168479919433594, 0.7473602294921875, -1.5756072998046875, -0.853790283203125, 0.0292816162109375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000028.npy"}
|
|
{"epoch": 0.041116005873715125, "step": 29, "batch_size": 64, "mean": 0.5867743492126465, "std": 0.5927706956863403, "min": -0.40169525146484375, "p10": -0.0882047653198242, "median": 0.5261859893798828, "p90": 1.4358772277832035, "max": 2.1297988891601562, "pos_frac": 0.84375, "sample": [0.996002197265625, 0.008373260498046875, 0.572479248046875, 0.6986274719238281, 0.23639678955078125, 0.8006000518798828, -0.09857177734375, 1.347015380859375, 0.3001213073730469, 0.28832244873046875, 0.35750579833984375, -0.06907081604003906, 1.4739608764648438, 0.4965667724609375, 0.10760498046875, 2.0946807861328125, 1.214599609375, -0.14743423461914062, -0.039737701416015625, 1.8142242431640625, 0.318939208984375, 0.8224868774414062, 0.39388275146484375, 0.2999839782714844, 1.6984481811523438, 0.6576385498046875, -0.40169525146484375, 0.6724777221679688, 0.10858154296875, 0.267852783203125, 0.026332855224609375, 0.68377685546875, 2.1297988891601562, 0.33184814453125, 1.112823486328125, 0.2655315399169922, -0.1598663330078125, 0.025648117065429688, 1.7476272583007812, 1.1470947265625, 0.694854736328125, 0.48140716552734375, 0.57958984375, 0.39760589599609375, 0.6987800598144531, 0.031986236572265625, 0.58245849609375, -0.30426788330078125, 0.8306884765625, -0.30086326599121094, 0.12183189392089844, 0.21360015869140625, 0.543365478515625, 0.6793289184570312, 0.9544410705566406, 1.249481201171875, -0.06722259521484375, 0.9399185180664062, 0.5090065002441406, 0.5822105407714844, 1.6483306884765625, 0.8372650146484375, -0.096405029296875, 1.1446876525878906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000029.npy"}
|
|
{"epoch": 0.042584434654919234, "step": 30, "batch_size": 64, "mean": 0.6909043788909912, "std": 0.6451590061187744, "min": -0.6033401489257812, "p10": -0.01866607666015625, "median": 0.6685924530029297, "p90": 1.5041656494140627, "max": 2.41339111328125, "pos_frac": 0.859375, "sample": [0.6349334716796875, -0.2895050048828125, -0.156280517578125, 0.37926483154296875, 1.4565277099609375, 0.01749420166015625, 0.1346282958984375, 0.4449462890625, 0.010528564453125, 0.7168197631835938, 0.8579635620117188, 1.2827529907226562, 0.7022514343261719, 0.9362220764160156, 0.7821044921875, 1.1669540405273438, 0.7878189086914062, 1.5245819091796875, -0.33422088623046875, 1.2589263916015625, 0.97210693359375, 0.9930419921875, 2.41339111328125, 0.8886909484863281, 1.035400390625, 0.30246734619140625, 0.7805938720703125, 0.7811660766601562, 0.05787086486816406, 0.10077285766601562, 0.560638427734375, 0.267120361328125, -0.1403350830078125, -0.018209457397460938, -0.004138946533203125, 0.8562469482421875, 2.03131103515625, 2.0435791015625, 1.61041259765625, 0.8013801574707031, 0.19916725158691406, 1.24554443359375, -0.018861770629882812, 0.5339889526367188, -0.6033401489257812, 1.4061355590820312, 1.4170684814453125, 0.14323806762695312, 1.0247154235839844, 1.5601882934570312, 1.3166313171386719, 0.9527740478515625, 0.5899085998535156, -0.09836578369140625, 0.0384063720703125, 1.3680191040039062, 0.4562644958496094, 0.2612152099609375, 0.3317108154296875, 0.10542869567871094, 0.4416656494140625, 2.0310821533203125, 0.44869232177734375, 0.4183845520019531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000030.npy"}
|
|
{"epoch": 0.04405286343612335, "step": 31, "batch_size": 64, "mean": 0.46136197447776794, "std": 0.6517056822776794, "min": -0.7954826354980469, "p10": -0.11733169555664061, "median": 0.3666801452636719, "p90": 1.1990119934082033, "max": 3.4701766967773438, "pos_frac": 0.765625, "sample": [0.08902359008789062, 1.3113784790039062, -0.123504638671875, 0.3612937927246094, 1.881561279296875, 0.49948883056640625, 0.1800689697265625, 0.19066619873046875, 0.4117889404296875, -0.6797027587890625, 3.4701766967773438, -0.0468902587890625, 0.804168701171875, 0.3987464904785156, 0.09045791625976562, -0.010406494140625, 0.6969375610351562, 0.4691448211669922, 0.9959487915039062, 0.6613578796386719, 0.23372650146484375, -0.19630813598632812, -0.09511566162109375, 0.3253173828125, -0.18828201293945312, 0.2966728210449219, 0.5913619995117188, 0.3720664978027344, 0.6579818725585938, 0.34039306640625, 1.1414031982421875, 0.7951507568359375, 1.46514892578125, 0.8677825927734375, 0.4694328308105469, -0.050872802734375, -0.08592987060546875, 0.427215576171875, -0.0436859130859375, 0.2846527099609375, 0.47251129150390625, 1.2237014770507812, -0.7954826354980469, 0.12375068664550781, -0.10591888427734375, -0.590972900390625, 0.107421875, 0.9107437133789062, 0.0129547119140625, 0.19377517700195312, 0.6093540191650391, 1.613372802734375, 0.7314453125, 0.1213226318359375, -0.122222900390625, 0.6265029907226562, 0.27642059326171875, -0.0349273681640625, 0.8441963195800781, 0.635223388671875, 0.9043426513671875, 1.6548385620117188, 0.2110137939453125, 0.64398193359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000031.npy"}
|
|
{"epoch": 0.04552129221732746, "step": 32, "batch_size": 64, "mean": 0.7538947463035583, "std": 0.8736879825592041, "min": -0.444427490234375, "p10": -0.12421188354492188, "median": 0.5266237258911133, "p90": 1.8249977111816407, "max": 4.302703857421875, "pos_frac": 0.84375, "sample": [-0.12181854248046875, 0.36618804931640625, 1.234588623046875, 1.4384231567382812, 0.6342277526855469, 0.6451568603515625, 0.141754150390625, 0.67803955078125, 0.4739704132080078, 0.9315719604492188, 1.1781463623046875, 0.4915504455566406, -0.12452316284179688, 0.17342758178710938, 0.076416015625, 1.0290565490722656, -0.4288139343261719, 0.9993476867675781, -0.28260040283203125, 2.72491455078125, 0.3881988525390625, 1.8100814819335938, 1.831390380859375, -0.444427490234375, 0.7240753173828125, 0.3917045593261719, -0.18061065673828125, -0.0314178466796875, -0.27881622314453125, 1.2396774291992188, 0.5960845947265625, 0.2318706512451172, 0.39343833923339844, 3.0013580322265625, 1.43096923828125, 0.5043888092041016, 0.8589324951171875, 0.28345489501953125, 0.03951454162597656, 0.8176078796386719, 2.4121475219726562, 0.7389564514160156, 0.06334686279296875, 0.41424560546875, 4.302703857421875, 0.902069091796875, 0.548858642578125, 1.7025833129882812, 2.167144775390625, -0.12348556518554688, 0.16965103149414062, 0.20576095581054688, 2.1450958251953125, 0.8543167114257812, -0.18172836303710938, 0.17287254333496094, 1.1703948974609375, 1.2682952880859375, 0.2745933532714844, 1.093719482421875, 0.20573043823242188, 0.32343292236328125, 1.054473876953125, 0.4975852966308594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000032.npy"}
|
|
{"epoch": 0.04698972099853157, "step": 33, "batch_size": 64, "mean": 0.6086278557777405, "std": 0.6056771874427795, "min": -0.38796234130859375, "p10": -0.19342803955078125, "median": 0.5580692291259766, "p90": 1.374673080444336, "max": 2.2161941528320312, "pos_frac": 0.796875, "sample": [-0.27276611328125, 0.5707435607910156, 0.4403095245361328, 0.5447235107421875, 1.1209716796875, -0.19470977783203125, -0.13127899169921875, 0.904144287109375, 0.7813072204589844, 0.7610378265380859, 0.05489349365234375, 0.7480850219726562, 1.883270263671875, 1.08868408203125, 1.675262451171875, 0.14993667602539062, 0.5351791381835938, -0.38796234130859375, -0.2060089111328125, -0.33294677734375, -0.0015411376953125, 0.8742141723632812, 0.9010505676269531, 1.084991455078125, 0.5453948974609375, -0.19043731689453125, 1.7789459228515625, -0.1468963623046875, -0.08046722412109375, 1.77935791015625, 0.4052925109863281, -0.09816741943359375, 2.2161941528320312, 0.16803741455078125, 0.20547103881835938, 0.7069244384765625, 0.20196533203125, 0.4951171875, 0.8237380981445312, 0.4318733215332031, 0.08579635620117188, 1.3750038146972656, 0.5862808227539062, 0.4751472473144531, 0.9964714050292969, 1.131103515625, 0.6950531005859375, 1.5284271240234375, 1.03887939453125, 0.0097503662109375, 0.333404541015625, -0.225372314453125, 1.3739013671875, 0.7019939422607422, 0.40643310546875, 1.0461654663085938, 1.1425132751464844, -0.315582275390625, 1.0960464477539062, 0.48174285888671875, 0.8254928588867188, 0.4084014892578125, 0.8242893218994141, 1.0969047546386719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000033.npy"}
|
|
{"epoch": 0.048458149779735685, "step": 34, "batch_size": 64, "mean": 0.7293854355812073, "std": 0.866633415222168, "min": -0.960357666015625, "p10": -0.22833328247070306, "median": 0.5451021194458008, "p90": 1.9964164733886718, "max": 2.85107421875, "pos_frac": 0.84375, "sample": [2.8168487548828125, -0.3429145812988281, 0.5559234619140625, 0.16627883911132812, 2.3628997802734375, 0.39992523193359375, 1.042266845703125, 1.4329681396484375, 0.05648040771484375, -0.250885009765625, 0.359527587890625, 0.000865936279296875, 0.6592788696289062, 0.19274139404296875, 0.5954437255859375, 2.85107421875, 0.42035675048828125, 0.08471298217773438, 1.991241455078125, 0.290863037109375, 1.272918701171875, -0.3691368103027344, 0.8272228240966797, 1.4547882080078125, 1.165283203125, 1.9986343383789062, 1.9861373901367188, -0.49192047119140625, 0.314727783203125, 1.3840522766113281, -0.17571258544921875, 0.1415252685546875, 1.5545272827148438, 0.9933090209960938, 1.818145751953125, 1.1406745910644531, 0.6031208038330078, 2.138092041015625, 0.5342807769775391, -0.5153465270996094, 0.22677993774414062, 0.30027198791503906, 0.5974063873291016, 1.2015762329101562, 0.7102279663085938, 0.7723655700683594, 0.04639625549316406, -0.1212005615234375, 1.23370361328125, 0.19322967529296875, 0.3442649841308594, -0.960357666015625, 0.8747482299804688, 0.6355438232421875, 0.1399383544921875, 0.0527496337890625, 2.4610671997070312, 0.2560844421386719, -0.07916259765625, 0.4149055480957031, 1.858428955078125, 2.25408935546875, 0.12186050415039062, -0.3154716491699219], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000034.npy"}
|
|
{"epoch": 0.049926578560939794, "step": 35, "batch_size": 64, "mean": 1.1309484243392944, "std": 1.025305986404419, "min": -0.3207817077636719, "p10": 0.1719921112060547, "median": 0.8330450057983398, "p90": 2.183489990234375, "max": 5.8450927734375, "pos_frac": 0.9375, "sample": [1.3997993469238281, 0.5034675598144531, 2.5555419921875, 1.6346778869628906, 0.24982452392578125, 0.6335182189941406, 1.548309326171875, 0.3835906982421875, -0.08969879150390625, 1.6213226318359375, 2.39569091796875, 0.3183116912841797, 1.3856887817382812, 0.5842494964599609, 0.1852874755859375, 0.5997161865234375, 1.7666702270507812, 3.5867538452148438, 0.893402099609375, 0.3851776123046875, 0.8270854949951172, 1.9250717163085938, 2.2180328369140625, -0.06414031982421875, 0.13355255126953125, 0.314208984375, 0.49246978759765625, 0.8151397705078125, 0.5736083984375, 0.49147796630859375, 5.8450927734375, 1.21942138671875, 0.2694549560546875, 1.546600341796875, 0.6824569702148438, 3.3221588134765625, 0.16629409790039062, 1.6859283447265625, 2.027740478515625, 2.00537109375, 0.471649169921875, 2.011322021484375, 0.8390045166015625, 1.400390625, 0.4247016906738281, -0.000568389892578125, 2.2653656005859375, 2.1028900146484375, 0.8637237548828125, 0.09991455078125, 0.7236785888671875, 0.46138763427734375, 0.7884292602539062, 2.0242156982421875, 1.0576934814453125, 1.1018123626708984, 0.6327972412109375, -0.3207817077636719, 1.4367752075195312, 0.19466400146484375, 0.9022254943847656, 1.9299163818359375, 1.6359710693359375, 0.29518890380859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000035.npy"}
|
|
{"epoch": 0.0513950073421439, "step": 36, "batch_size": 64, "mean": 1.0601813793182373, "std": 1.1890830993652344, "min": -2.0792388916015625, "p10": -0.2772954940795897, "median": 0.9595260620117188, "p90": 2.853773498535156, "max": 3.7945785522460938, "pos_frac": 0.796875, "sample": [1.0662803649902344, -0.1429920196533203, 1.0922985076904297, 1.1394424438476562, 1.3089447021484375, 0.5316925048828125, 0.886932373046875, -0.3854827880859375, -0.1251544952392578, 2.231689453125, 2.367706298828125, -0.05944061279296875, -0.3814201354980469, 0.4625568389892578, -0.06915855407714844, 1.5374298095703125, 0.026056289672851562, 1.6954193115234375, 1.5261688232421875, 0.554290771484375, 0.097137451171875, -0.5346107482910156, 0.28741455078125, 0.19686317443847656, 3.7945785522460938, 2.8245391845703125, 1.7146644592285156, -0.39742279052734375, 3.020599365234375, 0.6469573974609375, 0.38060760498046875, 1.3733978271484375, -0.3348541259765625, 3.2718276977539062, -0.09897613525390625, -0.022253036499023438, 3.201873779296875, 0.6687774658203125, 2.4071578979492188, 0.6309661865234375, 2.1274642944335938, 1.0867595672607422, 2.2349777221679688, 1.0321197509765625, 2.866302490234375, 0.5541610717773438, 0.3470420837402344, 3.601898193359375, 1.1574115753173828, -2.0792388916015625, 1.9746742248535156, 0.25240325927734375, 1.5601348876953125, 0.16569137573242188, 2.4169082641601562, 1.1583976745605469, -0.4279937744140625, 0.4962024688720703, 2.501983642578125, 0.4522533416748047, 1.4362106323242188, 2.94189453125, 1.1665191650390625, 0.4349250793457031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000036.npy"}
|
|
{"epoch": 0.05286343612334802, "step": 37, "batch_size": 64, "mean": 0.9678912162780762, "std": 1.364072561264038, "min": -2.879150390625, "p10": -0.18969726562499994, "median": 0.7306289672851562, "p90": 2.7184593200683596, "max": 6.10504150390625, "pos_frac": 0.8125, "sample": [1.6753692626953125, 0.041046142578125, 0.7909317016601562, 0.9531021118164062, 0.3415946960449219, 0.18933868408203125, -0.5883560180664062, 0.7643661499023438, 1.5055618286132812, -0.029596328735351562, -0.215057373046875, 2.79876708984375, 0.3791961669921875, -0.70391845703125, -0.408233642578125, 0.15491485595703125, 0.6773567199707031, 0.1251373291015625, 1.9030628204345703, 6.10504150390625, 1.1669235229492188, 4.613861083984375, 1.00579833984375, 0.6968917846679688, 0.7850418090820312, 1.5400009155273438, 0.21048736572265625, 0.47495269775390625, 1.6071548461914062, 1.0992450714111328, 0.8758640289306641, -0.130523681640625, 0.4801349639892578, -0.7280044555664062, 0.9123916625976562, 0.02953338623046875, 0.2745800018310547, 2.1871795654296875, 2.3310012817382812, 2.7956390380859375, -0.023403167724609375, 2.671783447265625, 0.2758636474609375, 1.3469696044921875, 0.23397445678710938, 0.7804355621337891, -0.0033416748046875, -0.333526611328125, 0.985076904296875, 1.6232681274414062, 1.9604263305664062, 0.06048583984375, 2.698822021484375, 2.7268753051757812, 0.2810077667236328, 3.0892486572265625, 1.4393272399902344, -2.879150390625, -0.025884628295898438, 0.25817108154296875, 2.2119178771972656, 3.551116943359375, 0.11692619323730469, 0.21086883544921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000037.npy"}
|
|
{"epoch": 0.05433186490455213, "step": 38, "batch_size": 64, "mean": 1.3253748416900635, "std": 1.5792897939682007, "min": -0.9647064208984375, "p10": -0.1167657852172851, "median": 0.9349708557128906, "p90": 3.260762023925783, "max": 7.71075439453125, "pos_frac": 0.875, "sample": [0.19437026977539062, 7.71075439453125, 4.201934814453125, 2.1335601806640625, 0.6677360534667969, 0.16122055053710938, 0.9906692504882812, 0.7597808837890625, 1.4974365234375, 1.494659423828125, 1.90997314453125, 0.6602325439453125, 0.074920654296875, 2.8141937255859375, 0.9964065551757812, -0.2563037872314453, 0.37500762939453125, 0.06166839599609375, -0.21075439453125, 1.8349761962890625, 0.8847198486328125, 1.66595458984375, 2.3867263793945312, 0.05987739562988281, 3.4521484375, 0.12061882019042969, 2.1564559936523438, 2.5733489990234375, 1.7771682739257812, 0.24701309204101562, 0.5942974090576172, 2.68914794921875, -0.9647064208984375, 1.24566650390625, 0.4818687438964844, 0.4105072021484375, 0.4798126220703125, 5.6009674072265625, 0.520263671875, 0.5887069702148438, 3.5533218383789062, 4.36419677734375, 1.9725189208984375, 0.2510185241699219, 0.8693218231201172, -0.1409912109375, 0.1762542724609375, 0.36545562744140625, 0.9532089233398438, 1.3313674926757812, -0.7050857543945312, 0.39453887939453125, -0.06023979187011719, 1.2873497009277344, 0.9852237701416016, 1.9910202026367188, 1.7528228759765625, 5.442481994628906, 0.9687881469726562, -0.44625091552734375, 2.1048049926757812, -0.321319580078125, 0.9167327880859375, 1.774444580078125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000038.npy"}
|
|
{"epoch": 0.055800293685756244, "step": 39, "batch_size": 64, "mean": 1.5791467428207397, "std": 1.3257520198822021, "min": -2.053924560546875, "p10": 0.23936023712158205, "median": 1.347611427307129, "p90": 3.5636543273925794, "max": 4.9185791015625, "pos_frac": 0.953125, "sample": [2.1199569702148438, 1.8076629638671875, 4.3089447021484375, 2.411346435546875, 1.4335098266601562, -0.6573944091796875, 2.1593246459960938, 2.6083946228027344, 1.3378772735595703, 2.2914066314697266, 0.5130996704101562, 3.1808700561523438, 1.3573455810546875, 1.0174732208251953, 3.6832199096679688, 0.7993545532226562, 0.6335182189941406, 0.5364227294921875, 0.07598876953125, 1.2024612426757812, 0.34316253662109375, 1.46240234375, 2.3541717529296875, 0.6625041961669922, 1.0807018280029297, 1.84844970703125, 0.07222747802734375, 1.9467144012451172, 4.9185791015625, 0.21588897705078125, 0.39839935302734375, 0.26201629638671875, 3.28466796875, 2.175933837890625, -2.053924560546875, 1.6504783630371094, 0.22965049743652344, 1.7323627471923828, 3.719310760498047, 4.3253326416015625, 1.0449028015136719, 2.0917205810546875, 3.8849258422851562, 1.19342041015625, 1.1518745422363281, 0.491607666015625, 0.4756011962890625, 0.5257568359375, 1.9302978515625, 1.012969970703125, 1.0070877075195312, 3.019012451171875, 3.081512451171875, 2.669830322265625, 0.4351158142089844, 3.9604034423828125, 2.8770294189453125, 1.0717830657958984, 2.1307315826416016, 0.3244132995605469, 0.8157596588134766, -0.05703163146972656, 0.5470752716064453, 1.9297752380371094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000039.npy"}
|
|
{"epoch": 0.05726872246696035, "step": 40, "batch_size": 64, "mean": 1.445512294769287, "std": 1.5635247230529785, "min": -0.8258209228515625, "p10": -0.2501495361328125, "median": 1.1048221588134766, "p90": 3.5082733154296877, "max": 7.0177764892578125, "pos_frac": 0.828125, "sample": [-0.40358734130859375, -0.0729217529296875, 0.5618896484375, 1.0310096740722656, 1.5743331909179688, 0.437530517578125, 0.662841796875, 0.9710159301757812, 0.166351318359375, 2.161439895629883, 6.1761322021484375, 3.175994873046875, 1.0067901611328125, 3.6075668334960938, 1.6649169921875, -0.4024696350097656, 2.1148204803466797, 1.7138671875, 0.34905052185058594, -0.8258209228515625, 2.7070159912109375, -0.06346511840820312, 0.413299560546875, -0.2539215087890625, 2.2696685791015625, 1.5271224975585938, 0.3231697082519531, 0.9549884796142578, 0.87225341796875, 0.03570556640625, 3.3166656494140625, -0.09536170959472656, 3.5546417236328125, -0.2413482666015625, 0.357818603515625, 4.006172180175781, 0.13328933715820312, 3.5081024169921875, 0.9412155151367188, 1.3564205169677734, 4.705047607421875, 0.9531822204589844, 0.7610969543457031, 7.0177764892578125, -0.3349456787109375, 1.6455001831054688, -0.44315338134765625, 2.0613479614257812, 2.5960617065429688, 1.854583740234375, 1.477081298828125, 1.9545974731445312, 3.2003250122070312, 1.1786346435546875, 1.3138618469238281, 1.495025634765625, 3.5083465576171875, -0.26025390625, 1.7816696166992188, 0.40515899658203125, 0.3886260986328125, 1.5239334106445312, 2.12603759765625, 0.3090400695800781], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000040.npy"}
|
|
{"epoch": 0.05873715124816446, "step": 41, "batch_size": 64, "mean": 1.6995975971221924, "std": 1.826016902923584, "min": -2.6802215576171875, "p10": -0.03212127685546867, "median": 1.4204292297363281, "p90": 4.268289184570313, "max": 7.877388000488281, "pos_frac": 0.890625, "sample": [0.4853363037109375, 0.5708198547363281, 3.6865692138671875, 1.4206657409667969, 0.5752029418945312, 1.940185546875, 2.90679931640625, 2.1384811401367188, -0.0648651123046875, 3.231536865234375, 4.3856658935546875, 0.9189300537109375, 1.7175140380859375, 7.877388000488281, 0.25724029541015625, 0.044281005859375, 0.8276519775390625, 1.4934272766113281, 0.11614227294921875, -1.13677978515625, 1.5323486328125, 3.3555870056152344, 4.04571533203125, 0.8267765045166016, 5.936065673828125, 1.8742179870605469, 3.926116943359375, 4.363677978515625, 0.5613250732421875, 0.201751708984375, 2.0440597534179688, 0.613983154296875, 5.38494873046875, 0.4578227996826172, 0.9621639251708984, 3.645355224609375, 1.550079345703125, 2.6028289794921875, -0.4813232421875, 2.4977188110351562, 4.460258483886719, 1.4201927185058594, -2.6802215576171875, 3.04833984375, 0.4020843505859375, 0.78631591796875, 1.6951217651367188, 1.9593276977539062, 2.3049182891845703, 0.4854717254638672, 1.0085334777832031, 1.0026702880859375, 0.6740570068359375, 1.471221923828125, 0.7742595672607422, 0.9687938690185547, -0.5175628662109375, 3.276081085205078, 0.6180267333984375, 0.33048248291015625, -0.6029586791992188, -0.2750110626220703, 1.9896469116210938, 4.8807830810546875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000041.npy"}
|
|
{"epoch": 0.06020558002936858, "step": 42, "batch_size": 64, "mean": 2.4108238220214844, "std": 2.2141408920288086, "min": -1.492279052734375, "p10": 0.38587036132812513, "median": 1.7838611602783203, "p90": 5.446164703369141, "max": 10.79144287109375, "pos_frac": 0.9375, "sample": [1.3207550048828125, -0.0418243408203125, 2.1790618896484375, 9.457183837890625, 10.79144287109375, 2.9225692749023438, 0.03680419921875, 5.473541259765625, 2.3687667846679688, 2.451446533203125, 1.6239852905273438, 6.030250549316406, 2.5776901245117188, 5.742340087890625, 0.6376628875732422, 1.8265457153320312, 2.5239486694335938, -0.2454681396484375, 2.5538330078125, 3.5929946899414062, 1.7411766052246094, 1.2640342712402344, 3.5496826171875, 0.7019729614257812, 1.2349395751953125, 2.729328155517578, 0.533050537109375, 2.4339447021484375, 1.8304939270019531, 0.540130615234375, 1.9346656799316406, 5.5415802001953125, 0.337310791015625, 5.60443115234375, 2.1598281860351562, 1.491912841796875, 1.0949230194091797, 1.595651626586914, 1.526947021484375, -0.4642791748046875, 3.8060760498046875, 0.499176025390625, 1.6192169189453125, 4.6932525634765625, 1.4001655578613281, 3.8199310302734375, 1.1510753631591797, 0.5620689392089844, 1.2230567932128906, 1.14141845703125, 0.9096450805664062, 4.8458251953125, -1.492279052734375, 0.2615833282470703, 1.6200637817382812, 3.5409698486328125, 0.6031684875488281, 0.9312591552734375, 5.115028381347656, 1.9814910888671875, 4.22509765625, 5.382286071777344, 1.0068283081054688, 4.241058349609375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000042.npy"}
|
|
{"epoch": 0.06167400881057269, "step": 43, "batch_size": 64, "mean": 2.3211941719055176, "std": 1.808465600013733, "min": -0.95770263671875, "p10": 0.31635684967041017, "median": 2.269068717956543, "p90": 4.323015594482422, "max": 9.28509521484375, "pos_frac": 0.953125, "sample": [0.3411827087402344, 0.3026695251464844, 0.3060932159423828, 3.1602096557617188, 3.7779388427734375, 0.973785400390625, 3.8362655639648438, 2.7203521728515625, 2.2975597381591797, 6.168342590332031, 0.7189102172851562, 2.093914031982422, 0.6708106994628906, 2.014232635498047, 2.837574005126953, 0.8039169311523438, 2.496826171875, 5.262237548828125, 1.7059783935546875, 0.2875957489013672, 5.768890380859375, 3.0014877319335938, 4.9055633544921875, 0.8292465209960938, 3.168048858642578, 2.57171630859375, 0.365966796875, 2.3178653717041016, -0.1934070587158203, 4.39556884765625, 0.9892044067382812, 2.930622100830078, 3.349395751953125, 2.995758056640625, 2.5082015991210938, 3.229248046875, 1.0649642944335938, 2.0078392028808594, 1.6110115051269531, 0.55914306640625, 1.6513214111328125, 1.9151229858398438, 6.263359069824219, 3.9974212646484375, 1.23760986328125, -0.95770263671875, 1.1081829071044922, 3.1249160766601562, 1.431854248046875, 1.760284423828125, 2.428997039794922, 3.0492019653320312, 2.2405776977539062, 0.26213836669921875, 2.4448165893554688, 9.28509521484375, 3.661773681640625, 1.5514640808105469, -0.18532943725585938, 0.5490360260009766, 2.6433792114257812, 4.153724670410156, 3.4461441040039062, 0.3403053283691406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000043.npy"}
|
|
{"epoch": 0.0631424375917768, "step": 44, "batch_size": 64, "mean": 2.597970962524414, "std": 2.0015811920166016, "min": -0.7966213226318359, "p10": 0.40288391113281263, "median": 2.172618865966797, "p90": 5.087345886230469, "max": 9.1015625, "pos_frac": 0.96875, "sample": [3.9325637817382812, 2.06976318359375, 2.06243896484375, 0.5298881530761719, 6.518348693847656, 2.2653961181640625, 2.5346221923828125, 1.2232742309570312, 1.6504592895507812, 5.092559814453125, 5.484931945800781, 0.7672080993652344, 4.4273223876953125, 5.009765625, 4.811981201171875, 3.2746963500976562, 0.9362831115722656, 2.07025146484375, 1.9400081634521484, 4.753631591796875, 0.35716819763183594, 2.700775146484375, 2.896352767944336, 6.675323486328125, 2.145721435546875, 3.1401596069335938, 0.347015380859375, 1.7362594604492188, 5.0751800537109375, 1.9388313293457031, 2.598175048828125, 0.03692626953125, 1.7014026641845703, 1.1053390502929688, 1.6490001678466797, 3.11810302734375, 0.5112991333007812, 1.4887542724609375, 4.992431640625, 2.707489013671875, 1.0029144287109375, 1.8878860473632812, 3.70770263671875, 3.341899871826172, 1.3427467346191406, 3.1637420654296875, 2.333251953125, 2.1995162963867188, 0.6019287109375, 0.5095539093017578, 5.8659820556640625, 4.0857696533203125, 2.4396820068359375, 3.4205322265625, 0.8339481353759766, 9.1015625, 0.3485908508300781, -0.76263427734375, 2.9228744506835938, 1.578399658203125, -0.7966213226318359, 0.226593017578125, 1.2067070007324219, 7.430519104003906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000044.npy"}
|
|
{"epoch": 0.06461086637298091, "step": 45, "batch_size": 64, "mean": 2.0074782371520996, "std": 2.375420570373535, "min": -1.3950843811035156, "p10": -0.12087020874023435, "median": 1.4381332397460938, "p90": 5.391035461425782, "max": 11.65838623046875, "pos_frac": 0.84375, "sample": [3.6381301879882812, 1.5404510498046875, 0.24584197998046875, 2.3409996032714844, 0.24299240112304688, 1.7833099365234375, 2.96905517578125, 5.8848419189453125, 4.75762939453125, 2.5244598388671875, 0.1355438232421875, 6.869842529296875, 3.408447265625, 0.04119873046875, 0.0066165924072265625, 1.439788818359375, -0.09464263916015625, 1.4997634887695312, 0.08892631530761719, 4.717067718505859, -0.00604248046875, 0.1614818572998047, 1.8778839111328125, 0.6865081787109375, 0.8597202301025391, 4.904420852661133, 5.965995788574219, 4.02886962890625, 0.3773822784423828, 0.5330429077148438, 4.524169921875, 1.85546875, 5.2776336669921875, -0.07001495361328125, -0.5677337646484375, 1.4364776611328125, 5.7989959716796875, 2.7552490234375, -0.21860504150390625, 1.5745010375976562, 1.4075489044189453, 0.6356658935546875, -1.3950843811035156, -1.3260574340820312, 1.3759765625, 11.65838623046875, 0.7166290283203125, 6.3665313720703125, 2.5577163696289062, -0.6718864440917969, 2.3751697540283203, 1.3546829223632812, 1.6490898132324219, 0.30486297607421875, 1.1370220184326172, 5.43963623046875, 0.6122894287109375, 2.528228759765625, 2.005207061767578, 1.0707550048828125, -0.132110595703125, 3.4035720825195312, 0.5414543151855469, -0.9323539733886719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000045.npy"}
|
|
{"epoch": 0.06607929515418502, "step": 46, "batch_size": 64, "mean": 2.6675515174865723, "std": 2.8319804668426514, "min": -1.9260787963867188, "p10": -0.08981781005859364, "median": 1.8766021728515625, "p90": 5.97862777709961, "max": 12.632720947265625, "pos_frac": 0.890625, "sample": [0.45467185974121094, -0.13469696044921875, 4.90716552734375, 6.9805908203125, 3.87738037109375, -1.9260787963867188, 1.4156608581542969, 1.5706729888916016, 5.9972381591796875, 3.2934913635253906, 5.707698822021484, 1.1002159118652344, 0.39734649658203125, 2.283475875854492, 4.42840576171875, 0.15087127685546875, -0.493377685546875, 0.82000732421875, 4.001182556152344, 10.5860595703125, -0.3787040710449219, 0.095672607421875, 0.07874298095703125, 3.7307662963867188, 0.15582656860351562, 4.14703369140625, 4.374786376953125, 1.2557926177978516, 3.7203750610351562, 2.423917770385742, 3.9345474243164062, 0.7043285369873047, -1.765228271484375, 1.0534629821777344, 2.509124755859375, 5.395732879638672, 5.0917205810546875, 0.35302734375, 7.654426574707031, 0.04619598388671875, 0.3036842346191406, 2.16241455078125, 4.8326416015625, 0.01490020751953125, 12.632720947265625, 0.32806396484375, 5.6691131591796875, 5.5601959228515625, 0.7443408966064453, 2.9455642700195312, 1.5495872497558594, 7.04876708984375, 2.943645477294922, -0.22873687744140625, 1.5231742858886719, 6.5179443359375, 1.0695457458496094, 2.319997787475586, 3.46734619140625, 5.935203552246094, 1.590789794921875, 1.0235271453857422, 1.5047149658203125, -0.7353706359863281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000046.npy"}
|
|
{"epoch": 0.06754772393538913, "step": 47, "batch_size": 64, "mean": 2.6299524307250977, "std": 2.4420642852783203, "min": -2.0868759155273438, "p10": -0.04467315673828117, "median": 1.8658437728881836, "p90": 6.009722137451173, "max": 9.6715087890625, "pos_frac": 0.890625, "sample": [1.846832275390625, 2.7824325561523438, 0.5620956420898438, 1.4771556854248047, 4.679805755615234, 1.6355018615722656, 0.4455413818359375, 1.7811470031738281, 6.170921325683594, 2.848081588745117, 0.180023193359375, 5.6335906982421875, 5.5108489990234375, 1.155029296875, 3.3740463256835938, 2.472686767578125, 0.03218841552734375, 1.6248397827148438, 1.0895309448242188, 5.1947479248046875, 6.3907318115234375, 1.0599594116210938, 0.8876323699951172, 3.066669464111328, 2.5795669555664062, 1.1547698974609375, -0.4713897705078125, 5.4713897705078125, 6.927001953125, 2.3017234802246094, -0.07761383056640625, 3.0104007720947266, 5.629154205322266, 3.325653076171875, -2.0868759155273438, 0.2597942352294922, -0.14472198486328125, 1.0602645874023438, 0.9448394775390625, 9.6715087890625, 4.7952117919921875, 2.3307647705078125, 0.9227809906005859, 1.779693603515625, 8.028823852539062, 5.4846038818359375, -0.7599105834960938, 3.083353042602539, 8.585296630859375, -0.6158218383789062, 1.1754837036132812, 1.3167877197265625, -0.2612152099609375, 1.1000804901123047, 3.0424728393554688, 1.26702880859375, 6.3317108154296875, 1.4851913452148438, 4.79779052734375, 1.679443359375, 3.615966796875, 3.7407684326171875, 2.0483016967773438, 1.8848552703857422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000047.npy"}
|
|
{"epoch": 0.06901615271659324, "step": 48, "batch_size": 64, "mean": 2.5794191360473633, "std": 2.559356451034546, "min": -5.94854736328125, "p10": 0.03637981414794922, "median": 1.9281444549560547, "p90": 6.237547302246096, "max": 9.9652099609375, "pos_frac": 0.90625, "sample": [3.1425933837890625, 0.60107421875, 6.4668731689453125, 6.849189758300781, 5.70245361328125, 2.7736053466796875, -0.35699462890625, 2.0582427978515625, 4.8364105224609375, 3.7630081176757812, 1.1584930419921875, 4.5001983642578125, 1.6286087036132812, 0.03619384765625, -0.3102874755859375, 0.9383506774902344, 4.9483642578125, 0.03681373596191406, 1.0874176025390625, -0.4021415710449219, 0.6708259582519531, 7.4296112060546875, -0.29195594787597656, 1.8703689575195312, 5.626365661621094, 1.4118423461914062, 2.9394454956054688, 3.4265995025634766, 0.6101112365722656, 3.1019668579101562, 1.7646369934082031, 1.7285232543945312, 3.1805572509765625, 1.9319381713867188, 7.2244720458984375, 3.521177291870117, 4.4445343017578125, 7.731971740722656, 3.5375823974609375, 1.2201766967773438, 0.6001930236816406, 1.3484764099121094, 0.523895263671875, 2.6871871948242188, 0.97821044921875, 3.5307884216308594, 3.0236968994140625, 1.6805343627929688, -1.2775897979736328, 6.982673645019531, 1.9607677459716797, 2.937265396118164, 1.768646240234375, 9.9652099609375, 3.8579788208007812, -5.94854736328125, 4.968296051025391, 1.3855342864990234, 0.4534282684326172, 0.930267333984375, 4.8805389404296875, 1.71026611328125, 1.9243507385253906, 1.6715316772460938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000048.npy"}
|
|
{"epoch": 0.07048458149779736, "step": 49, "batch_size": 64, "mean": 2.9200520515441895, "std": 2.956599473953247, "min": -2.7773208618164062, "p10": 0.0032308578491212145, "median": 2.452362060546875, "p90": 6.387650299072266, "max": 12.682449340820312, "pos_frac": 0.890625, "sample": [2.40411376953125, 0.8148269653320312, 0.3390655517578125, 2.7896728515625, 2.0951576232910156, 4.673503875732422, 0.7888031005859375, 2.2081851959228516, -0.9376068115234375, 0.6171035766601562, 0.12497329711914062, 1.5488014221191406, 1.6376113891601562, 6.853424072265625, 6.1166534423828125, 3.0048294067382812, 3.2650413513183594, 1.9469738006591797, 2.7012863159179688, 1.2118072509765625, 1.192880630493164, -2.5296878814697266, 2.158111572265625, -0.12030029296875, 10.2830810546875, 7.765655517578125, 0.3341827392578125, 5.5875091552734375, 3.6777191162109375, 0.5766315460205078, 5.8151397705078125, 12.682449340820312, 3.7484893798828125, 0.9606781005859375, -1.18817138671875, -2.7773208618164062, 3.3698272705078125, 3.3311996459960938, 3.8749656677246094, 4.317741394042969, 2.3005313873291016, -0.04894447326660156, 0.735870361328125, 3.4823455810546875, 3.7178726196289062, 10.186248779296875, 2.0641098022460938, 2.5006103515625, 6.503791809082031, 4.61810302734375, 4.799720764160156, 2.8595237731933594, 3.7545013427734375, 4.142723083496094, 10.517745971679688, 0.1410961151123047, -0.6495361328125, 2.1579971313476562, 1.3275413513183594, 3.03033447265625, 3.56146240234375, 1.7636566162109375, 4.484394073486328, 1.6666259765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000049.npy"}
|
|
{"epoch": 0.07195301027900147, "step": 50, "batch_size": 64, "mean": 3.2729973793029785, "std": 3.1046013832092285, "min": -2.7686500549316406, "p10": -0.39459457397460934, "median": 2.8819656372070312, "p90": 7.300894165039063, "max": 12.069198608398438, "pos_frac": 0.84375, "sample": [5.811248779296875, 0.42797088623046875, -0.514923095703125, 8.302509307861328, 1.3165206909179688, 2.4547958374023438, 6.166709899902344, 1.6234054565429688, 3.710010528564453, 1.6635808944702148, -0.3495521545410156, 5.6554107666015625, 3.0072479248046875, 11.602615356445312, -0.02850341796875, 2.854156494140625, 6.23541259765625, 2.9485225677490234, 7.5700225830078125, 2.9097747802734375, 0.7902774810791016, 4.514862060546875, 7.2445831298828125, 2.2746658325195312, 1.917938232421875, 4.055585861206055, 0.11411285400390625, -0.4138984680175781, 2.3822708129882812, 7.3250274658203125, 5.4925079345703125, 4.450340270996094, 7.767414093017578, 12.069198608398438, -2.7686500549316406, 0.504241943359375, 0.974700927734375, 1.9132080078125, 0.11565399169921875, 3.0811080932617188, 1.9159889221191406, 1.6587028503417969, 5.323173522949219, 3.1777496337890625, 7.006206512451172, 5.001983642578125, -1.2581939697265625, 1.6661720275878906, 4.568580627441406, -1.5262527465820312, 5.3196563720703125, 2.3939971923828125, 3.6607933044433594, 4.678337097167969, 2.6030120849609375, 5.108375549316406, 5.781944274902344, 2.5083999633789062, 8.683418273925781, -0.8125, 5.450103759765625, 1.5341567993164062, -2.0541000366210938, -0.08995437622070312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000050.npy"}
|
|
{"epoch": 0.07342143906020558, "step": 51, "batch_size": 64, "mean": 3.2261362075805664, "std": 4.030710697174072, "min": -4.2945404052734375, "p10": -0.5188758850097656, "median": 2.244091033935547, "p90": 8.388035583496094, "max": 16.0135498046875, "pos_frac": 0.84375, "sample": [8.239578247070312, 3.43310546875, 0.648956298828125, -0.4603271484375, 0.8626861572265625, 0.1839447021484375, 6.944877624511719, -0.6628360748291016, 2.2884292602539062, -0.3075408935546875, 8.45166015625, 6.574588775634766, 1.507303237915039, 4.073905944824219, 2.033802032470703, 4.1526947021484375, 3.061573028564453, -0.21026229858398438, 8.638259887695312, 5.124385833740234, 3.878316879272461, -0.767333984375, 0.05010986328125, 3.526966094970703, 1.0917510986328125, 6.277767181396484, 0.41106605529785156, 5.280487060546875, 0.6103935241699219, 3.8952560424804688, 16.0135498046875, 2.1928958892822266, 0.5267372131347656, 4.4945526123046875, 2.840822219848633, 11.214859008789062, 1.4366378784179688, 13.228561401367188, 4.8514251708984375, 1.8742179870605469, 2.197998046875, 1.0527534484863281, 3.178741455078125, 8.012718200683594, 1.2953147888183594, 0.08416748046875, 3.8017807006835938, -3.4174652099609375, -0.6745109558105469, 14.612823486328125, 11.065170288085938, 5.601104736328125, 5.806304931640625, -0.5439682006835938, 0.7683372497558594, 4.8066253662109375, 0.0006999969482421875, 0.04930877685546875, 2.1997528076171875, -4.2945404052734375, 2.8415451049804688, -1.8161773681640625, 2.3304672241210938, 0.0059356689453125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000051.npy"}
|
|
{"epoch": 0.07488986784140969, "step": 52, "batch_size": 64, "mean": 5.0320281982421875, "std": 4.021976947784424, "min": -1.9964981079101562, "p10": 0.45050430297851585, "median": 4.218349456787109, "p90": 10.667947006225587, "max": 17.400558471679688, "pos_frac": 0.9375, "sample": [1.6774425506591797, 10.967758178710938, 3.852691650390625, 10.510963439941406, 6.652000427246094, 7.925201416015625, 2.6860504150390625, 6.324012756347656, 7.817474365234375, 10.883224487304688, 0.369354248046875, 10.735225677490234, 4.3272247314453125, 4.900913238525391, 7.45789909362793, 7.035425186157227, 2.330272674560547, 4.392435073852539, 14.542648315429688, 7.7637786865234375, 3.362560272216797, 0.29558563232421875, 9.188796997070312, 2.181936264038086, 0.05766105651855469, 5.961948394775391, 2.402130126953125, 1.724344253540039, 8.000471115112305, 17.400558471679688, 7.945098876953125, 1.6363868713378906, 8.838905334472656, 4.111198425292969, 3.4918212890625, 11.049827575683594, 3.6221389770507812, -0.426025390625, 3.9787673950195312, -0.8688888549804688, -1.9964981079101562, 1.8243560791015625, 6.280517578125, 4.9897613525390625, 1.3357505798339844, 7.6866302490234375, 4.104024887084961, 9.248748779296875, 5.986656188964844, 1.620077133178711, 0.6398544311523438, 2.791900634765625, 2.973339080810547, 4.32550048828125, 3.0603675842285156, 8.68685531616211, 4.9933624267578125, 1.07061767578125, 14.6085205078125, -1.2595062255859375, 4.830665588378906, 3.394062042236328, 2.2487258911132812, 1.4982986450195312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000052.npy"}
|
|
{"epoch": 0.0763582966226138, "step": 53, "batch_size": 64, "mean": 5.014308452606201, "std": 5.417908191680908, "min": -2.8591461181640625, "p10": 0.246147918701172, "median": 3.453550338745117, "p90": 12.612397766113283, "max": 24.999008178710938, "pos_frac": 0.921875, "sample": [-1.5367927551269531, 3.058685302734375, 2.8260116577148438, 3.595306396484375, 0.14252853393554688, 3.1728172302246094, 10.964851379394531, 0.3705940246582031, 5.9650726318359375, -1.6594390869140625, 1.5186767578125, 22.22711181640625, -2.164703369140625, 1.2735099792480469, 5.311408996582031, 5.905139923095703, 2.160888671875, 9.526161193847656, 2.1469192504882812, 10.654838562011719, 3.9847869873046875, 4.9901123046875, 11.213325500488281, 6.124473571777344, 12.698379516601562, 0.62786865234375, 4.784610748291016, 0.4035186767578125, 4.558078765869141, 1.9225635528564453, 14.739471435546875, 7.24969482421875, 6.080234527587891, 7.11956787109375, 3.530181884765625, 0.19281387329101562, 0.6098518371582031, 3.4646263122558594, 1.0299568176269531, 3.766305923461914, 2.9641647338867188, 6.405096054077148, 1.3650989532470703, 12.201675415039062, -2.8591461181640625, 14.213790893554688, 14.496368408203125, 2.54693603515625, 4.516632080078125, 2.5529136657714844, 1.9114341735839844, 2.962465286254883, 2.4877471923828125, 1.918121337890625, 2.6461715698242188, 3.7013778686523438, 2.625041961669922, 24.999008178710938, 1.9076042175292969, 12.411773681640625, 3.6268882751464844, 3.442474365234375, 13.579277038574219, -0.2572002410888672], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000053.npy"}
|
|
{"epoch": 0.07782672540381791, "step": 54, "batch_size": 64, "mean": 4.548583984375, "std": 4.16992712020874, "min": -0.8868618011474609, "p10": 0.3700267791748048, "median": 3.5031280517578125, "p90": 10.257807922363282, "max": 19.075927734375, "pos_frac": 0.96875, "sample": [6.346351623535156, 4.5608978271484375, 7.905780792236328, 0.8741416931152344, 0.705108642578125, 2.40155029296875, 0.6946334838867188, 2.3242435455322266, 4.63385009765625, 7.706691741943359, 0.766571044921875, 4.271003723144531, 10.441879272460938, 1.8933753967285156, 3.5894622802734375, 9.464187622070312, -0.11425209045410156, 4.309913635253906, 4.769752502441406, 5.5554656982421875, 0.01688385009765625, 1.3036346435546875, 13.418838500976562, 0.5107269287109375, 6.0109100341796875, 0.5618667602539062, 4.223552703857422, 4.122505187988281, 4.953399658203125, 5.624488830566406, 1.7913074493408203, 0.2832908630371094, 6.5037994384765625, 0.017856597900390625, 0.3097267150878906, 3.159191131591797, 10.767753601074219, 2.7935714721679688, 5.472900390625, 11.499374389648438, 1.9999065399169922, 2.4939727783203125, 19.075927734375, 9.193367004394531, -0.8868618011474609, 3.345752716064453, 0.05803108215332031, 11.126983642578125, 0.8629550933837891, 3.4167938232421875, 1.6704940795898438, 3.8689002990722656, 9.214309692382812, 1.2333450317382812, 2.22515869140625, 15.481048583984375, 1.1163215637207031, 3.967996597290039, 3.3377609252929688, 9.66729736328125, 2.2759761810302734, 9.82830810546875, 1.0109500885009766, 9.078506469726562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000054.npy"}
|
|
{"epoch": 0.07929515418502203, "step": 55, "batch_size": 64, "mean": 5.350862503051758, "std": 5.858836650848389, "min": -7.040985107421875, "p10": -0.8934982299804687, "median": 4.865802764892578, "p90": 13.433475494384767, "max": 23.186172485351562, "pos_frac": 0.796875, "sample": [7.502723693847656, 8.786544799804688, 4.8881988525390625, -2.815216064453125, 20.644805908203125, -1.2167129516601562, 5.652599334716797, 3.6884117126464844, 13.602462768554688, -0.354949951171875, 5.246185302734375, 4.851470947265625, 1.9796981811523438, 9.347122192382812, 16.27972412109375, 8.425933837890625, 3.6320343017578125, 7.730865478515625, 23.186172485351562, 3.33453369140625, 10.114013671875, -1.2467041015625, 3.217041015625, 7.100456237792969, 19.170822143554688, -0.04525566101074219, 13.010910034179688, -0.0179290771484375, 0.7657203674316406, 6.659488677978516, 3.7367172241210938, -0.901275634765625, 5.001422882080078, 0.058300018310546875, 9.36907958984375, 3.4722518920898438, -1.0190887451171875, 0.7039108276367188, 0.3239727020263672, -3.8628158569335938, 8.158916473388672, 2.470956802368164, 1.6902847290039062, 13.0850830078125, 5.9423370361328125, -0.6543769836425781, 0.9813461303710938, 8.623046875, 4.880134582519531, -7.040985107421875, 2.308055877685547, 13.582786560058594, 5.943473815917969, 3.546215057373047, 6.338493347167969, 4.773984909057617, -0.8753509521484375, 7.512199401855469, 6.7327880859375, 11.469169616699219, 5.073585510253906, -0.384765625, 4.46734619140625, 13.82684326171875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000055.npy"}
|
|
{"epoch": 0.08076358296622614, "step": 56, "batch_size": 64, "mean": 5.061400413513184, "std": 5.83642053604126, "min": -7.288730621337891, "p10": -1.1002866744995117, "median": 4.823518753051758, "p90": 12.325593566894536, "max": 21.908493041992188, "pos_frac": 0.796875, "sample": [-1.1060562133789062, 6.234767913818359, -2.562225341796875, 7.2322540283203125, -0.5967273712158203, -7.288730621337891, 8.952339172363281, -1.0868244171142578, 3.129913330078125, -3.7030410766601562, 3.0009536743164062, 2.1665000915527344, 6.166542053222656, 13.673065185546875, 1.9005069732666016, -0.050933837890625, -0.20329856872558594, 9.305412292480469, 8.827564239501953, 7.389015197753906, -1.5047454833984375, 8.319961547851562, 2.0219573974609375, 17.69622802734375, 7.92242431640625, 12.777740478515625, 0.9749565124511719, 0.04743385314941406, 8.089452743530273, 2.1120834350585938, 6.200294494628906, 7.0296630859375, 8.07305908203125, 1.4058475494384766, 9.212699890136719, 10.006916046142578, 0.15934371948242188, 15.3076171875, 6.398284912109375, 5.896903991699219, 5.049415588378906, 3.8815078735351562, 2.8888797760009766, 17.333984375, 6.3845062255859375, 10.763900756835938, 5.558368682861328, 8.805572509765625, 0.5656051635742188, 21.908493041992188, 1.7565784454345703, -0.13352012634277344, 3.526702880859375, 11.270584106445312, -5.606048583984375, 2.1110095977783203, -0.2984771728515625, 18.837142944335938, 4.597621917724609, -2.71026611328125, 2.2768402099609375, 1.8552017211914062, 5.297126770019531, 8.479804992675781], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000056.npy"}
|
|
{"epoch": 0.08223201174743025, "step": 57, "batch_size": 64, "mean": 5.646149635314941, "std": 5.056145668029785, "min": -7.62908935546875, "p10": 0.12002334594726569, "median": 5.634683609008789, "p90": 13.07173652648926, "max": 18.977325439453125, "pos_frac": 0.90625, "sample": [7.060329437255859, 2.807466506958008, 0.4285011291503906, -1.1676025390625, 8.41015625, -7.62908935546875, 9.384231567382812, 0.18004608154296875, 11.15087890625, 5.756061553955078, 13.921585083007812, 3.2712459564208984, 18.977325439453125, 2.8867340087890625, 6.1533966064453125, 5.5897216796875, -0.7053642272949219, 2.7582836151123047, 5.694183349609375, 16.49267578125, 9.918144226074219, 13.422943115234375, 9.83856201171875, 3.674396514892578, 5.679645538330078, 1.2696380615234375, 13.385143280029297, 4.652402877807617, 3.576448440551758, 2.4244766235351562, 0.921875, 4.139476776123047, 3.2387237548828125, 0.09429931640625, 6.75360107421875, 6.014678955078125, -0.098541259765625, 1.5895042419433594, 2.1288299560546875, 7.355030059814453, 8.628890991210938, 1.7315349578857422, 8.587005615234375, 5.0192413330078125, 8.101455688476562, -2.1202125549316406, 7.557464599609375, 6.36651611328125, 7.56781005859375, 2.636281967163086, 2.172821044921875, 14.340316772460938, 5.083099365234375, 2.7079200744628906, 7.7384796142578125, 3.6462631225585938, 0.86346435546875, 10.641571044921875, 10.039085388183594, -4.06597900390625, 12.3404541015625, 16.315963745117188, 6.437906265258789, 7.616180419921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000057.npy"}
|
|
{"epoch": 0.08370044052863436, "step": 58, "batch_size": 64, "mean": 6.553309440612793, "std": 6.680759906768799, "min": -7.563301086425781, "p10": 0.16150760650634785, "median": 4.615547180175781, "p90": 15.689215850830085, "max": 24.065948486328125, "pos_frac": 0.90625, "sample": [14.342071533203125, -0.9767169952392578, 0.4939727783203125, 4.319572448730469, 3.633026123046875, 3.773172378540039, 16.724571228027344, 2.2548828125, 12.351692199707031, 2.7618484497070312, 4.790031433105469, 6.634773254394531, 6.6880340576171875, 3.51751708984375, 2.3309326171875, 12.155303955078125, 9.9588623046875, 7.781377792358398, 4.441062927246094, 2.821573257446289, 6.573326110839844, 12.796966552734375, 12.17779541015625, -1.63067626953125, 10.041778564453125, 2.1719207763671875, 9.053409576416016, 6.315671920776367, 19.993881225585938, 3.4947147369384766, 24.065948486328125, 7.934574127197266, 6.393135070800781, 0.08004570007324219, 13.08642578125, 1.0245361328125, 11.356689453125, 3.3505020141601562, 2.287109375, -1.5031890869140625, 1.8212242126464844, 1.8287582397460938, 20.411590576171875, -1.2758712768554688, 9.031845092773438, 0.35158538818359375, 1.0994586944580078, 6.5649871826171875, -3.963531494140625, 22.35870361328125, 1.977884292602539, 6.362705230712891, 1.180044174194336, 13.484783172607422, 9.801349639892578, -7.563301086425781, 21.522903442382812, 1.98583984375, 2.7453994750976562, 6.8895721435546875, 14.291908264160156, 1.8057117462158203, 0.5695648193359375, 16.266563415527344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000058.npy"}
|
|
{"epoch": 0.08516886930983847, "step": 59, "batch_size": 64, "mean": 6.788931846618652, "std": 6.866553783416748, "min": -5.3889007568359375, "p10": -0.9095188140869136, "median": 5.9434814453125, "p90": 17.524111938476562, "max": 28.617416381835938, "pos_frac": 0.875, "sample": [8.033515930175781, 6.764762878417969, 13.960784912109375, 18.025390625, 14.624740600585938, 4.086601257324219, 28.617416381835938, 17.626724243164062, 6.995859146118164, 14.19586181640625, 1.0374641418457031, 8.471908569335938, -3.384246826171875, 2.2765884399414062, 18.450408935546875, -0.4119453430175781, 2.1178770065307617, 11.558784484863281, 5.887748718261719, 17.648338317871094, 1.2349853515625, 2.0327606201171875, 0.9271621704101562, 0.8726844787597656, 3.1764297485351562, 20.76800537109375, 2.4362545013427734, 2.0024051666259766, 6.9484710693359375, -5.3889007568359375, 8.197532653808594, -1.6626472473144531, 7.292686462402344, 7.53387451171875, 8.1361083984375, 15.801811218261719, 2.792633056640625, 11.274993896484375, 6.148017883300781, 3.3804473876953125, -4.8641510009765625, -3.6774673461914062, 5.120426177978516, -1.1227645874023438, 3.7808990478515625, 9.269538879394531, 7.1640472412109375, 2.8653717041015625, 6.7534637451171875, 2.431812286376953, 2.5213356018066406, 5.507957458496094, 2.6643543243408203, 8.287776947021484, 4.656791687011719, 12.83416748046875, 8.108585357666016, 2.8592147827148438, 2.192474365234375, -1.4270858764648438, 17.284683227539062, 18.011016845703125, 5.999214172363281, 16.779693603515625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000059.npy"}
|
|
{"epoch": 0.08663729809104258, "step": 60, "batch_size": 64, "mean": 5.555472373962402, "std": 6.117406845092773, "min": -11.987640380859375, "p10": -1.2378425598144531, "median": 6.1671142578125, "p90": 11.378736114501953, "max": 24.243812561035156, "pos_frac": 0.828125, "sample": [6.915321350097656, -1.4255828857421875, 1.455434799194336, 11.168388366699219, 0.605712890625, 2.778654098510742, 10.693878173828125, 7.288196563720703, 1.8692398071289062, -11.987640380859375, 6.025819778442383, -1.57989501953125, 1.7948417663574219, 3.525928497314453, 10.001102447509766, 5.6266021728515625, 3.8435211181640625, 9.331710815429688, 2.92620849609375, 8.882896423339844, 11.297767639160156, 11.413436889648438, 16.35943603515625, 8.294303894042969, 8.615188598632812, 6.851276397705078, -0.5663223266601562, 6.244758605957031, 8.43463134765625, 0.052745819091796875, 8.087482452392578, 17.230667114257812, 3.4386520385742188, 6.089469909667969, 7.283817291259766, -1.2791671752929688, 9.090446472167969, 8.692489624023438, 1.7888870239257812, -4.446563720703125, 10.583412170410156, 2.8456649780273438, 4.7649993896484375, 7.8198089599609375, 3.175018310546875, -0.9104995727539062, -9.24185562133789, 14.283126831054688, 18.105438232421875, 10.302253723144531, 12.411033630371094, 2.3526611328125, -1.555999755859375, -1.14141845703125, 10.121320724487305, 6.3044586181640625, 24.243812561035156, 7.553009033203125, 1.01904296875, 8.008115768432617, -0.6501235961914062, 6.6816253662109375, 4.749988555908203, 1.011587142944336], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000060.npy"}
|
|
{"epoch": 0.0881057268722467, "step": 61, "batch_size": 64, "mean": 6.724519729614258, "std": 8.864770889282227, "min": -8.653182983398438, "p10": -0.7846603393554685, "median": 4.983585357666016, "p90": 15.90578384399414, "max": 50.384796142578125, "pos_frac": 0.828125, "sample": [2.5400772094726562, 1.808258056640625, -0.6127471923828125, 1.9117660522460938, 12.313186645507812, -0.2069110870361328, 0.44742393493652344, 6.6736907958984375, -0.5762176513671875, 23.302291870117188, 18.26342010498047, 17.294708251953125, 24.836448669433594, 10.498435974121094, 6.0999298095703125, 8.702590942382812, -0.9232063293457031, 1.7200031280517578, 11.16299819946289, 10.849250793457031, 3.2558059692382812, 5.742092132568359, 13.949691772460938, 7.628475189208984, 5.324638366699219, 6.1854095458984375, 4.827857971191406, 5.139312744140625, 7.543354034423828, 7.523681640625, 0.00852203369140625, 0.639434814453125, 12.342063903808594, 15.340164184570312, 14.583831787109375, 19.07379150390625, 15.880325317382812, 1.336090087890625, -0.85833740234375, 13.301483154296875, 2.8485183715820312, 0.1224517822265625, 10.074462890625, 2.3852577209472656, 1.1733627319335938, 15.916694641113281, 1.1289596557617188, 2.7767677307128906, 50.384796142578125, 10.918731689453125, 11.013145446777344, -0.9230575561523438, -8.653182983398438, 10.645782470703125, 2.4236392974853516, -4.803009033203125, 4.55853271484375, 4.594860076904297, -5.719505310058594, 1.994384765625, 6.914825439453125, -4.933586120605469, -0.578460693359375, 1.2318172454833984], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000061.npy"}
|
|
{"epoch": 0.08957415565345081, "step": 62, "batch_size": 64, "mean": 5.107054233551025, "std": 6.349539279937744, "min": -7.791961669921875, "p10": -2.4232673645019527, "median": 4.747087478637695, "p90": 12.360660171508789, "max": 26.341567993164062, "pos_frac": 0.875, "sample": [7.679466247558594, 1.2732467651367188, 0.7390375137329102, 9.65655517578125, 5.2754058837890625, 10.401840209960938, 6.567234039306641, 0.8761348724365234, 9.013893127441406, -4.9440765380859375, 12.065414428710938, 5.4888153076171875, 1.76873779296875, 0.09379768371582031, 8.576484680175781, 3.0985107421875, 8.069610595703125, 4.72802734375, 0.06386375427246094, 5.0797576904296875, 5.709964752197266, 7.2478485107421875, 1.1256790161132812, 4.8522491455078125, -3.6167030334472656, 7.9092254638671875, 3.8380298614501953, 6.885101318359375, 0.03193473815917969, 26.341567993164062, 16.497337341308594, 6.977500915527344, 5.258323669433594, 2.4421310424804688, 2.2727737426757812, 4.766147613525391, 1.44677734375, 15.4561767578125, -7.791961669921875, -2.6425018310546875, -5.3862762451171875, 0.6613960266113281, 1.3395347595214844, 0.9083251953125, 1.433807373046875, 7.807254791259766, 15.005802154541016, 8.390911102294922, -3.035409927368164, 12.387481689453125, 7.640266418457031, 4.682178497314453, -5.0109100341796875, 3.526386260986328, 12.298076629638672, 3.165740966796875, 2.878772735595703, 4.2064361572265625, 23.162933349609375, 17.871963500976562, 6.629291534423828, 1.5043182373046875, -1.9117202758789062, 6.115570068359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000062.npy"}
|
|
{"epoch": 0.09104258443465492, "step": 63, "batch_size": 64, "mean": 6.8794989585876465, "std": 6.991913795471191, "min": -10.127243041992188, "p10": -0.6253108978271477, "median": 6.210033416748047, "p90": 14.093233489990237, "max": 28.49627685546875, "pos_frac": 0.890625, "sample": [22.348526000976562, 7.130790710449219, 4.349485397338867, 7.766548156738281, 2.617218017578125, 1.3765335083007812, 4.269401550292969, -5.793304443359375, 6.059295654296875, 5.6075439453125, -10.127243041992188, 13.602333068847656, 8.281341552734375, 6.15264892578125, 2.4485721588134766, -1.244802474975586, 14.303619384765625, 4.149799346923828, 10.610122680664062, 19.01557159423828, 11.053390502929688, 12.3438720703125, 6.171501159667969, 0.20476913452148438, -0.9420547485351562, 11.567411422729492, -3.3673248291015625, 0.11375808715820312, 4.5042877197265625, 3.2188282012939453, 10.614273071289062, 7.309806823730469, 6.853523254394531, 8.29166030883789, 9.469169616699219, 10.651115417480469, 11.8726806640625, 11.826774597167969, 4.7266845703125, 3.1301498413085938, 0.5028839111328125, 11.849563598632812, 2.099102020263672, 5.4260406494140625, -8.897918701171875, 12.591070175170898, 6.248565673828125, 1.83721923828125, 2.644378662109375, 5.5997467041015625, 2.66656494140625, 12.649665832519531, 15.912704467773438, 7.623443603515625, 16.302268981933594, 0.9136810302734375, 8.107076644897461, 28.49627685546875, -2.8201828002929688, 13.393367767333984, 4.4151153564453125, 23.37506103515625, 6.944496154785156, 9.869438171386719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000063.npy"}
|
|
{"epoch": 0.09251101321585903, "step": 64, "batch_size": 64, "mean": 8.005149841308594, "std": 8.873629570007324, "min": -14.090255737304688, "p10": -0.46825847625732403, "median": 6.314287185668945, "p90": 19.06625213623047, "max": 42.3349609375, "pos_frac": 0.828125, "sample": [1.3651199340820312, 15.627349853515625, 24.330245971679688, 6.840217590332031, 5.537422180175781, 11.379249572753906, 2.8135929107666016, 13.677421569824219, -0.5477771759033203, 5.812982559204102, 1.8771133422851562, 9.792503356933594, 1.0472259521484375, 4.581840515136719, 10.83721923828125, 27.251113891601562, 4.153957366943359, -1.1391677856445312, 15.659866333007812, 6.740821838378906, 9.873893737792969, -3.7884902954101562, 25.731658935546875, 6.328239440917969, 0.10792160034179688, 5.770172119140625, -0.023645401000976562, 5.893123626708984, -14.090255737304688, -0.07877159118652344, 6.638397216796875, -0.6943855285644531, 42.3349609375, 5.273277282714844, 2.3336219787597656, 8.985816955566406, 14.138236999511719, 18.421463012695312, 9.333572387695312, 5.055789947509766, 5.223567962646484, 20.296249389648438, 10.986358642578125, 7.797210693359375, -0.28271484375, -7.472099304199219, 21.413497924804688, 3.2563095092773438, 19.34259033203125, 1.2436981201171875, 13.512916564941406, 11.92779541015625, 11.190711975097656, -1.6681671142578125, 11.632575988769531, -0.21915626525878906, 4.544708251953125, 6.300334930419922, 11.667762756347656, 12.876144409179688, 2.114086151123047, 10.154853820800781, 17.210357666015625, 4.097087860107422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000064.npy"}
|
|
{"epoch": 0.09397944199706314, "step": 65, "batch_size": 64, "mean": 7.948797225952148, "std": 8.555886268615723, "min": -19.34033966064453, "p10": -0.2859138488769531, "median": 6.794950485229492, "p90": 19.739297485351564, "max": 26.3668212890625, "pos_frac": 0.84375, "sample": [26.3668212890625, 22.9339599609375, 14.642807006835938, 11.678504943847656, 5.067792892456055, 8.048500061035156, 0.6193771362304688, 12.44937515258789, 3.3433990478515625, 18.57360076904297, 14.886970520019531, 1.3236846923828125, -0.25627899169921875, 19.3475341796875, 24.49859619140625, 0.8577766418457031, 8.675773620605469, 2.3439903259277344, -3.3154754638671875, 4.2317657470703125, 0.9289073944091797, 15.121292114257812, 5.675506591796875, -19.34033966064453, 14.06341552734375, 7.729156494140625, -3.927947998046875, 6.722530364990234, 11.705238342285156, 5.287075042724609, 5.437995910644531, 6.1036224365234375, -1.7381858825683594, 4.554422378540039, 25.383132934570312, 6.390079498291016, 19.907196044921875, 13.338027954101562, 22.77265167236328, 7.059980392456055, -0.298614501953125, -8.143386840820312, 8.125961303710938, 2.107637405395508, 3.394012451171875, 15.427238464355469, 20.78241729736328, -0.11517333984375, 5.8265380859375, 17.375083923339844, 11.183826446533203, 2.5359954833984375, 17.571426391601562, -0.028444290161132812, 10.5736083984375, 6.86737060546875, 4.7751312255859375, 1.7056121826171875, 14.257781982421875, 10.73089599609375, 0.07053375244140625, 7.9334564208984375, -0.3956794738769531, 6.967569351196289], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000065.npy"}
|
|
{"epoch": 0.09544787077826726, "step": 66, "batch_size": 64, "mean": 7.422572612762451, "std": 8.746773719787598, "min": -7.0030670166015625, "p10": -0.7030525207519528, "median": 6.207252502441406, "p90": 17.09927291870117, "max": 42.38328552246094, "pos_frac": 0.84375, "sample": [6.487007141113281, 42.38328552246094, 1.6155624389648438, 13.530067443847656, 12.202987670898438, 10.70513916015625, 1.9554557800292969, -0.044464111328125, 9.347723007202148, 6.482208251953125, -7.0030670166015625, 2.140514373779297, 5.729509353637695, -0.8469161987304688, 11.87112808227539, 3.0345001220703125, 3.72430419921875, 0.11871719360351562, 2.334667205810547, 16.70703125, 6.258674621582031, 9.790454864501953, 0.6398983001708984, 9.519203186035156, 15.577713012695312, 2.9207916259765625, 3.2673797607421875, -2.58660888671875, -3.4729080200195312, -5.8927459716796875, 7.900577545166016, 16.887962341308594, 17.926422119140625, 6.8358917236328125, 2.8815841674804688, 10.036087036132812, 6.624553680419922, -0.1932964324951172, 1.6640987396240234, 13.319221496582031, 9.642677307128906, 3.4820938110351562, -0.36737060546875, 6.155830383300781, 5.230508804321289, -6.606964111328125, 8.452827453613281, 1.2237796783447266, -1.76678466796875, 11.738426208496094, 3.3531341552734375, 17.189834594726562, 24.181320190429688, 18.97222137451172, 3.2327117919921875, 14.423126220703125, 2.0694618225097656, 7.169353485107422, 13.497329711914062, 2.1741409301757812, 29.6243896484375, 1.0804443359375, 26.831634521484375, 11.680198669433594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000066.npy"}
|
|
{"epoch": 0.09691629955947137, "step": 67, "batch_size": 64, "mean": 7.723735809326172, "std": 8.613935470581055, "min": -8.62335205078125, "p10": -0.7855262756347652, "median": 5.357308387756348, "p90": 18.277169036865235, "max": 40.480804443359375, "pos_frac": 0.859375, "sample": [19.502174377441406, 3.4477005004882812, 19.533531188964844, 1.2085342407226562, 1.8540916442871094, -0.22718048095703125, 13.983085632324219, 4.678457260131836, -5.3149871826171875, 2.185131072998047, 15.42413330078125, 9.140155792236328, 3.8083953857421875, 13.7972412109375, -0.3575592041015625, -4.544290542602539, 10.5843505859375, 8.913864135742188, 0.9199142456054688, -8.62335205078125, 18.2999267578125, 6.915489196777344, 0.9555625915527344, 0.12390899658203125, 4.30938720703125, 1.9927444458007812, 18.22406768798828, 5.329065322875977, -0.9689407348632812, 7.2259521484375, 4.1270904541015625, 12.332332611083984, 5.5767059326171875, 2.745248794555664, 1.84881591796875, 8.938636779785156, 13.55729866027832, 5.018503189086914, 1.9285545349121094, -4.54888916015625, 5.108497619628906, 16.91290283203125, 3.966888427734375, 3.3297672271728516, 23.747207641601562, 13.057762145996094, 6.132089614868164, 15.380889892578125, 12.661174774169922, 12.393295288085938, 11.436363220214844, 5.385551452636719, 11.248208999633789, 28.811355590820312, 5.2219696044921875, 1.0891971588134766, 40.480804443359375, 3.5802459716796875, -2.28326416015625, 22.277816772460938, 17.8114013671875, 6.193447113037109, -1.3826751708984375, 7.913356781005859], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000067.npy"}
|
|
{"epoch": 0.09838472834067548, "step": 68, "batch_size": 64, "mean": 7.507518291473389, "std": 7.371875286102295, "min": -5.4189300537109375, "p10": 0.3213464736938478, "median": 5.600500106811523, "p90": 18.66374969482422, "max": 32.47615051269531, "pos_frac": 0.90625, "sample": [-4.030723571777344, 3.8059539794921875, 21.401519775390625, 6.861320495605469, 5.384792327880859, -4.6241302490234375, 7.690948486328125, 20.055343627929688, 0.2767333984375, 0.4254436492919922, 12.502487182617188, 13.906051635742188, 15.322479248046875, 11.160125732421875, 10.048805236816406, 0.48193359375, 6.078815460205078, 11.144058227539062, 2.3607635498046875, 6.70050048828125, -0.10822296142578125, 4.275102615356445, 4.705099105834961, 2.8587646484375, 6.073652267456055, 12.433574676513672, 17.799118041992188, 9.243827819824219, 8.951297760009766, 18.440887451171875, 18.759262084960938, 20.709613800048828, 2.4402084350585938, 5.3064422607421875, 8.734756469726562, 1.5824604034423828, -0.4436378479003906, 4.936187744140625, 2.7077865600585938, 5.158226013183594, 14.500370025634766, 2.3796119689941406, 9.760721206665039, -5.4189300537109375, 5.984283447265625, 5.8162078857421875, 3.601909637451172, 19.548385620117188, 2.673858642578125, 1.316671371459961, 1.7567729949951172, 32.47615051269531, 21.96302032470703, 16.191452026367188, 11.521560668945312, 1.0517044067382812, 2.2266082763671875, 4.589942932128906, 11.237709045410156, -2.434326171875, 3.9592971801757812, 2.4743080139160156, 5.191791534423828, 6.594451904296875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000068.npy"}
|
|
{"epoch": 0.09985315712187959, "step": 69, "batch_size": 64, "mean": 8.629631042480469, "std": 8.320178985595703, "min": -3.0742835998535156, "p10": -0.07507781982421857, "median": 6.388641357421875, "p90": 17.219404602050783, "max": 38.892242431640625, "pos_frac": 0.890625, "sample": [4.3885498046875, 6.1136474609375, 26.549148559570312, 6.8189544677734375, 10.882820129394531, 13.696182250976562, -1.812042236328125, -0.5830841064453125, -0.23190689086914062, 2.6017913818359375, 4.816497802734375, 7.902469635009766, -1.8759765625, 7.469490051269531, 7.486339569091797, 13.714332580566406, 21.247772216796875, 6.369659423828125, 15.744537353515625, 13.650032043457031, 0.8198471069335938, 20.878036499023438, 12.19968032836914, 10.589244842529297, 16.584121704101562, 11.72705078125, 4.914176940917969, 13.384689331054688, 11.246238708496094, 4.411333084106445, 0.5016517639160156, 0.606414794921875, 16.377525329589844, 38.892242431640625, 3.7844886779785156, 16.24578857421875, 5.0316925048828125, 15.109619140625, 14.050689697265625, 0.5688934326171875, 2.0714569091796875, 14.552169799804688, 10.029129028320312, 0.9943428039550781, -0.31464576721191406, 12.765514373779297, 6.353492736816406, 33.54937744140625, 3.7940444946289062, 6.5375823974609375, 6.24699592590332, -3.0742835998535156, 4.579620361328125, 3.4223556518554688, 2.0652809143066406, 6.407623291015625, 17.491668701171875, 20.182586669921875, 0.106597900390625, 14.91302490234375, 1.6041107177734375, 1.2745132446289062, 4.024116516113281, -0.1529388427734375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000069.npy"}
|
|
{"epoch": 0.1013215859030837, "step": 70, "batch_size": 64, "mean": 9.449182510375977, "std": 9.565919876098633, "min": -6.245552062988281, "p10": -0.32259101867675716, "median": 9.31822395324707, "p90": 23.378153991699225, "max": 33.842620849609375, "pos_frac": 0.890625, "sample": [5.793956756591797, 0.3393707275390625, 9.81295394897461, 18.22564697265625, 21.950820922851562, 11.122987747192383, -6.245552062988281, 1.35748291015625, 11.775527954101562, 3.5415515899658203, -0.808319091796875, 11.105934143066406, -1.7468318939208984, 7.244537353515625, 33.842620849609375, 1.9364147186279297, 10.4310302734375, 23.9898681640625, 9.953668594360352, 33.770416259765625, 9.487899780273438, 9.34572982788086, 1.545654296875, 26.105377197265625, 0.3308525085449219, 15.093551635742188, 16.052711486816406, 12.373735427856445, -0.8077640533447266, 6.296726226806641, 4.79473876953125, 25.850051879882812, 1.8108558654785156, 1.01318359375, 2.2322540283203125, 0.3233222961425781, 11.677921295166016, 2.4390640258789062, 10.115516662597656, -1.1764259338378906, 2.9561843872070312, 3.206714630126953, 0.7447967529296875, 1.835968017578125, 9.706584930419922, 2.001434326171875, 17.091094970703125, 20.890647888183594, -5.806060791015625, 4.9568023681640625, 14.696319580078125, 1.9095611572265625, 1.4999160766601562, 28.84168243408203, 11.627864837646484, -0.5994110107421875, 29.863174438476562, 9.290718078613281, 19.843963623046875, 14.832014083862305, 3.2211246490478516, 20.243125915527344, 11.300128936767578, 18.29425048828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000070.npy"}
|
|
{"epoch": 0.1027900146842878, "step": 71, "batch_size": 64, "mean": 10.571918487548828, "std": 10.121339797973633, "min": -8.611181259155273, "p10": 1.0113800048828128, "median": 8.905685424804688, "p90": 19.753301239013673, "max": 40.8831787109375, "pos_frac": 0.90625, "sample": [-4.954505920410156, 33.96942138671875, 6.929737091064453, 1.220367431640625, 5.587223052978516, 10.596336364746094, 6.285835266113281, 10.813709259033203, 16.36117172241211, 4.819896697998047, 18.262619018554688, 5.268911361694336, 14.408843994140625, 4.2363433837890625, 2.7412109375, 7.0994415283203125, 10.025760650634766, 13.594589233398438, -0.09008026123046875, 1.800313949584961, 4.2741851806640625, 36.70323181152344, 5.993316650390625, 5.426429748535156, 14.501838684082031, 15.617847442626953, 5.354831695556641, 10.737430572509766, 11.249029159545898, -7.010353088378906, 5.129905700683594, 14.4090576171875, 8.864517211914062, -0.02960205078125, 4.507938385009766, 15.980300903320312, 15.72039794921875, 0.92181396484375, 18.66754150390625, 12.367164611816406, 5.142299652099609, 11.426338195800781, 7.035888671875, 19.03851318359375, 14.070404052734375, 13.798088073730469, 4.323329925537109, 30.977142333984375, 40.419464111328125, 40.8831787109375, 1.67083740234375, 20.05963897705078, -8.611181259155273, -0.8982582092285156, 10.592277526855469, 4.217761993408203, 6.783184051513672, 10.548683166503906, 18.05051040649414, 14.345306396484375, 27.609161376953125, 2.8543167114257812, 4.955045700073242, 8.946853637695312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000071.npy"}
|
|
{"epoch": 0.10425844346549193, "step": 72, "batch_size": 64, "mean": 11.129253387451172, "std": 10.438271522521973, "min": -9.232162475585938, "p10": -0.8368692398071277, "median": 8.9052152633667, "p90": 26.976441955566415, "max": 43.15673828125, "pos_frac": 0.890625, "sample": [15.952766418457031, 13.453653335571289, 6.084968566894531, 12.240409851074219, 7.799613952636719, 1.9638690948486328, 6.749794006347656, 22.934768676757812, -9.232162475585938, 22.1834716796875, 12.672771453857422, 11.706048965454102, 7.587614059448242, 0.355194091796875, 15.259815216064453, 8.336734771728516, 10.1622314453125, 19.984268188476562, 11.23480224609375, 28.707359313964844, 33.14863586425781, -1.3477535247802734, 5.359504699707031, 8.989656448364258, 13.956550598144531, 4.2169036865234375, 16.555217742919922, 7.493431091308594, 7.537376403808594, 14.133003234863281, 2.77862548828125, 6.0195159912109375, 8.785537719726562, -9.228225708007812, 28.57861328125, 23.868194580078125, 16.785385131835938, 6.349699020385742, 20.278457641601562, -1.920867919921875, 7.113019943237305, 32.89715576171875, 43.15673828125, -6.404289245605469, 8.82077407836914, 24.56695556640625, 8.00408935546875, 6.527866363525391, 0.8874073028564453, 28.009078979492188, -1.406036376953125, 3.1303863525390625, 3.006275177001953, 16.568084716796875, 11.42818832397461, 17.238327026367188, 9.308647155761719, 14.193046569824219, 29.01458740234375, 10.240930557250977, 5.647041320800781, 2.7958221435546875, -2.31671142578125, 1.3694114685058594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000072.npy"}
|
|
{"epoch": 0.10572687224669604, "step": 73, "batch_size": 64, "mean": 10.851455688476562, "std": 14.591455459594727, "min": -18.808639526367188, "p10": -5.655471801757811, "median": 7.589433670043945, "p90": 28.255183029174805, "max": 52.79838562011719, "pos_frac": 0.796875, "sample": [27.72125244140625, -9.02655029296875, 5.8314666748046875, 12.7828369140625, 5.438056945800781, 28.483673095703125, -3.2680130004882812, 23.332229614257812, 5.525390625, -1.038970947265625, 21.444107055664062, 28.458065032958984, 19.19396209716797, -0.6576957702636719, 2.0578765869140625, -11.584716796875, 45.930267333984375, -1.3126449584960938, 10.028776168823242, 15.036895751953125, 14.323219299316406, 6.168327331542969, 7.652473449707031, 0.9123897552490234, 1.7000236511230469, 7.526393890380859, -12.749141693115234, 9.841373443603516, 24.760833740234375, 13.015300750732422, 11.142585754394531, 2.074953079223633, 44.7373046875, -8.044570922851562, 5.505023956298828, 24.427711486816406, 24.23558807373047, -7.627895355224609, 3.329498291015625, 3.205659866333008, 0.4208488464355469, -3.93890380859375, -6.391143798828125, 7.182222366333008, 14.90118408203125, 14.099395751953125, 2.204620361328125, 4.783998489379883, 19.21656036376953, 34.95445251464844, 10.737640380859375, 5.590534210205078, 16.537384033203125, 5.147651672363281, 10.915924072265625, -2.671417236328125, 12.838897705078125, 45.63023376464844, 19.513595581054688, 52.79838562011719, 2.932279586791992, 27.78179168701172, 17.602340698242188, -18.808639526367188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000073.npy"}
|
|
{"epoch": 0.10719530102790015, "step": 74, "batch_size": 64, "mean": 11.434505462646484, "std": 15.839223861694336, "min": -30.411766052246094, "p10": -1.7023494720458976, "median": 9.248804092407227, "p90": 26.3915641784668, "max": 89.76190185546875, "pos_frac": 0.796875, "sample": [25.36627960205078, 8.349388122558594, 9.81719970703125, -0.2592601776123047, 21.33477783203125, 4.884246826171875, 7.976041793823242, 45.549171447753906, 2.5801753997802734, 9.512954711914062, 10.934486389160156, 9.894491195678711, 16.22811508178711, 3.174835205078125, 14.91302490234375, 89.76190185546875, 7.385833740234375, 25.592971801757812, 12.877521514892578, 2.2019577026367188, 1.6341571807861328, 39.33158874511719, 2.1859588623046875, 26.73381805419922, 23.60308074951172, 8.73004150390625, -2.961740493774414, 2.4563140869140625, -0.2175006866455078, -14.152809143066406, 12.522567749023438, -0.5329513549804688, -0.6054763793945312, 7.855316162109375, -30.411766052246094, 17.827491760253906, -0.6267471313476562, 21.04867172241211, 4.612329483032227, -2.0294456481933594, -7.311576843261719, 9.200813293457031, 14.0797119140625, 22.182174682617188, -0.9391250610351562, 32.947967529296875, 11.374015808105469, 5.603584289550781, 5.7768402099609375, 7.202816009521484, 9.296794891357422, 30.584930419921875, 15.596687316894531, 18.892730712890625, 30.527748107910156, 16.631980895996094, 4.837869644165039, 12.057693481445312, 19.914047241210938, 22.279342651367188, 1.8926811218261719, -3.7315521240234375, -3.1451263427734375, 10.976356506347656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000074.npy"}
|
|
{"epoch": 0.10866372980910426, "step": 75, "batch_size": 64, "mean": 16.138355255126953, "std": 15.486407279968262, "min": -8.260162353515625, "p10": -2.0665700912475566, "median": 13.358339309692383, "p90": 39.16917953491211, "max": 59.20985412597656, "pos_frac": 0.875, "sample": [-0.3141040802001953, 17.574493408203125, 13.74624252319336, 13.337570190429688, 16.482009887695312, 11.421306610107422, 9.549285888671875, 6.241523742675781, 35.60017395019531, 48.92730712890625, 26.65145492553711, 12.695671081542969, -2.817626953125, 21.690628051757812, 59.20985412597656, 11.74411392211914, 31.78033447265625, 50.08845520019531, 7.905910491943359, 0.7426929473876953, 34.93083953857422, 18.359115600585938, 13.144866943359375, 10.776008605957031, 19.050201416015625, 38.903587341308594, 39.28300476074219, 3.015674591064453, 6.869026184082031, 26.804035186767578, 15.657241821289062, 13.379108428955078, 5.3353271484375, 52.04508972167969, -3.0183639526367188, -3.2821731567382812, 20.330490112304688, -8.260162353515625, 16.683815002441406, 16.867961883544922, 4.051921844482422, 10.671676635742188, -5.5938720703125, 5.863655090332031, -7.557046890258789, 21.498565673828125, 2.143890380859375, 8.622882843017578, 30.3779296875, 20.2923583984375, 3.792095184326172, -5.62408447265625, 6.823139190673828, 4.740478515625, 12.851936340332031, 4.214962005615234, 7.4387054443359375, 17.137420654296875, 15.060501098632812, 13.982093811035156, 28.655601501464844, 41.826080322265625, 48.371002197265625, 14.080825805664062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000075.npy"}
|
|
{"epoch": 0.11013215859030837, "step": 76, "batch_size": 64, "mean": 11.078810691833496, "std": 12.268765449523926, "min": -31.567703247070312, "p10": -0.7141248703002926, "median": 10.611437797546387, "p90": 24.56656379699707, "max": 53.492828369140625, "pos_frac": 0.859375, "sample": [12.20390510559082, 8.67184829711914, 2.199583053588867, 20.796707153320312, 2.1301956176757812, 17.135543823242188, 13.490180969238281, 22.027603149414062, 24.566272735595703, 10.421424865722656, 11.067344665527344, 15.877029418945312, 7.886213302612305, 27.615570068359375, 3.7844161987304688, 4.577428817749023, 8.077613830566406, 53.492828369140625, 26.81281280517578, 3.9594039916992188, -2.62860107421875, 10.801450729370117, 0.4756889343261719, 9.995353698730469, -11.764213562011719, 6.886465072631836, 4.41270637512207, 11.762649536132812, 4.9275970458984375, 8.62326431274414, 24.46269989013672, 17.153114318847656, 18.0687255859375, 7.661380767822266, 24.566688537597656, 3.4535770416259766, -1.3481025695800781, 20.925338745117188, 3.7780532836914062, -6.729560852050781, 6.86863899230957, 0.4344367980957031, -4.185916900634766, -0.8612060546875, 14.011497497558594, 23.0572509765625, 41.43896484375, 22.57495880126953, 8.325542449951172, 4.242828369140625, 10.953239440917969, -0.06041908264160156, 14.815177917480469, 12.489166259765625, 13.432975769042969, 11.589717864990234, -0.37093544006347656, 9.698539733886719, 29.064002990722656, 17.600982666015625, 17.231918334960938, 24.619781494140625, 11.362236022949219, -31.567703247070312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000076.npy"}
|
|
{"epoch": 0.11160058737151249, "step": 77, "batch_size": 64, "mean": 17.389745712280273, "std": 16.946420669555664, "min": -26.308624267578125, "p10": 0.7801502227783212, "median": 17.50349998474121, "p90": 42.17462768554688, "max": 57.676971435546875, "pos_frac": 0.90625, "sample": [12.436470031738281, 38.051002502441406, 26.069538116455078, 32.707271575927734, 34.365196228027344, 1.8279228210449219, 47.93342590332031, 19.082164764404297, 7.995716094970703, 10.002449035644531, 4.928442001342773, 28.938858032226562, 31.127601623535156, 6.031955718994141, 7.6132659912109375, 19.830894470214844, -26.308624267578125, 15.876914978027344, 27.270408630371094, 26.51849365234375, 12.718547821044922, -2.226696014404297, 8.186737060546875, 0.3835792541503906, 1.7054824829101562, 41.53224182128906, 2.680377960205078, 20.54962158203125, 8.2542724609375, -4.472131729125977, 26.842334747314453, 23.85369873046875, 12.254631042480469, 52.12599182128906, 27.672889709472656, 8.425161361694336, 44.411529541015625, 57.676971435546875, 17.896644592285156, 8.928695678710938, -24.819381713867188, 6.102428436279297, -8.581146240234375, 3.9324989318847656, 9.88237190246582, 25.271095275878906, 26.976058959960938, 20.86207389831543, 21.281097412109375, 22.269996643066406, 6.102569580078125, -1.017242431640625, 4.541559219360352, 20.6265869140625, 22.188743591308594, 19.850265502929688, 2.9695205688476562, 43.58070373535156, 55.26487731933594, 9.033489227294922, 18.538944244384766, 42.44993591308594, 17.110355377197266, 6.82635498046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000077.npy"}
|
|
{"epoch": 0.1130690161527166, "step": 78, "batch_size": 64, "mean": 13.888814926147461, "std": 14.680152893066406, "min": -14.801681518554688, "p10": -1.0611577987670895, "median": 13.972806930541992, "p90": 34.431660461425786, "max": 63.556671142578125, "pos_frac": 0.84375, "sample": [7.109733581542969, 1.6136951446533203, 33.271697998046875, 13.315681457519531, -0.48842620849609375, 37.96894836425781, 15.431583404541016, 63.556671142578125, 22.52362060546875, 19.841670989990234, 17.089698791503906, 7.093639373779297, -4.025775909423828, 14.125785827636719, 19.575347900390625, 18.363590240478516, -0.3478546142578125, 6.757396697998047, -6.5580596923828125, 1.1961669921875, 7.456647872924805, 15.82078742980957, 21.347896575927734, 17.09954071044922, 15.900089263916016, 14.10982894897461, 14.720218658447266, 13.835784912109375, 17.233642578125, 5.301898956298828, 34.92878723144531, 14.306497573852539, 5.155109405517578, -7.019138336181641, 21.420608520507812, -6.821990966796875, 40.291168212890625, 17.302459716796875, 36.931663513183594, 3.98114013671875, 32.4060173034668, 4.186546325683594, 41.874351501464844, 29.157062530517578, 22.63837432861328, 1.7456741333007812, 9.511157989501953, 3.042745590209961, 11.42441177368164, 7.335212707519531, 4.758037567138672, -0.7015247344970703, 13.72866439819336, -1.2152862548828125, 25.198348999023438, -14.801681518554688, 44.618988037109375, 3.550273895263672, -12.590217590332031, 23.268539428710938, 2.705718994140625, 24.333816528320312, 1.482147216796875, 20.509368896484375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000078.npy"}
|
|
{"epoch": 0.1145374449339207, "step": 79, "batch_size": 64, "mean": 15.608173370361328, "std": 18.716690063476562, "min": -14.1917724609375, "p10": -4.196015930175781, "median": 11.265382766723633, "p90": 43.66345825195315, "max": 63.68940734863281, "pos_frac": 0.734375, "sample": [-1.8884735107421875, 46.93230438232422, 6.132659912109375, 56.43409729003906, 6.748573303222656, -0.4984130859375, -1.3502483367919922, -8.149093627929688, 37.26499938964844, 35.24244689941406, 17.051193237304688, -14.1917724609375, 22.117210388183594, 57.2392578125, 9.103858947753906, 21.616500854492188, 63.68940734863281, 46.40565490722656, 15.76812744140625, -4.698699951171875, 61.614471435546875, 15.322860717773438, -1.6065826416015625, -8.86517333984375, 23.36389923095703, 1.4834938049316406, 3.2133941650390625, 14.316680908203125, 11.61453628540039, 50.32227325439453, 3.3119583129882812, 8.038543701171875, 10.916229248046875, 25.00714111328125, -6.8265380859375, 9.169204711914062, -1.7826004028320312, 6.449993133544922, 36.512603759765625, 32.88563537597656, 15.69369888305664, 28.13068389892578, 28.046463012695312, 17.13446807861328, 23.325294494628906, -3.597972869873047, 2.7562408447265625, 25.19159698486328, 30.322662353515625, 9.937759399414062, 12.32666015625, -4.452320098876953, -1.1877899169921875, 32.621856689453125, -1.564971923828125, 16.595123291015625, 5.436214447021484, 5.33399772644043, -0.3459320068359375, -5.7681884765625, 8.116161346435547, 36.01385498046875, -3.33221435546875, 16.75812530517578], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000079.npy"}
|
|
{"epoch": 0.11600587371512482, "step": 80, "batch_size": 64, "mean": 13.985532760620117, "std": 18.749555587768555, "min": -16.097335815429688, "p10": -5.440498542785644, "median": 11.523453712463379, "p90": 34.42599563598634, "max": 72.47317504882812, "pos_frac": 0.765625, "sample": [27.141807556152344, 7.832115173339844, -4.4853668212890625, 21.21674346923828, 4.544443130493164, 3.0087509155273438, -1.7684364318847656, -7.121330261230469, 21.252182006835938, 7.421348571777344, 3.3516712188720703, 14.385906219482422, -13.309059143066406, -16.097335815429688, -2.0863189697265625, 19.34911346435547, 72.47317504882812, 30.785125732421875, 69.24630737304688, 41.27332305908203, 18.77978515625, 6.468009948730469, 35.42509460449219, 32.094764709472656, 29.65435028076172, 12.83648681640625, -0.7698917388916016, 24.278594970703125, 12.09361457824707, 12.858718872070312, 1.3006591796875, -12.795997619628906, 5.851222991943359, -5.804405212402344, 21.066207885742188, -12.912269592285156, 49.904266357421875, 25.45813751220703, -6.188198089599609, 67.8583984375, 9.380075454711914, 1.2537689208984375, 3.3696556091308594, -0.1341705322265625, 37.481117248535156, 10.060722351074219, 19.952430725097656, 21.607589721679688, 15.256298065185547, 10.953292846679688, -1.624551773071289, -2.848234176635742, 0.48616600036621094, 3.8905868530273438, 21.02751350402832, 18.281723022460938, 5.221195220947266, 3.5015411376953125, 27.741783142089844, 20.115325927734375, 15.054298400878906, 29.928390502929688, -4.59138298034668, 13.837318420410156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000080.npy"}
|
|
{"epoch": 0.11747430249632893, "step": 81, "batch_size": 64, "mean": 19.705984115600586, "std": 22.37936019897461, "min": -19.717529296875, "p10": -6.808140563964841, "median": 16.761493682861328, "p90": 46.64240264892579, "max": 93.48944091796875, "pos_frac": 0.875, "sample": [34.602088928222656, 26.5123291015625, -19.717529296875, 34.152191162109375, 20.36719512939453, 44.81245422363281, 15.80417251586914, -10.090030670166016, 6.290412902832031, 29.178752899169922, 25.61676025390625, -13.25616455078125, 2.0575790405273438, 18.765304565429688, 68.46353149414062, 6.868501663208008, 18.54949951171875, 27.96173095703125, 47.426666259765625, -10.922805786132812, 1.1171550750732422, 11.801284790039062, 19.249629974365234, 58.81636047363281, 20.933456420898438, 11.524696350097656, 12.288177490234375, 5.453815460205078, 5.277395248413086, -4.051239013671875, 21.325469970703125, 9.37176513671875, 4.484975814819336, 10.895538330078125, 27.614994049072266, 92.93377685546875, 36.30112075805664, 16.10346221923828, 2.5944976806640625, 48.416168212890625, 34.743621826171875, 39.9678955078125, 11.799514770507812, 1.7067794799804688, 5.083320617675781, 0.2553558349609375, -13.792400360107422, 12.893556594848633, 19.33415985107422, 24.604400634765625, 14.398090362548828, 13.647706985473633, 17.419525146484375, -9.647468566894531, 34.53318786621094, 38.577484130859375, 93.48944091796875, 52.25999450683594, 7.488319396972656, 31.32324981689453, 3.7926254272460938, 22.83099365234375, -7.9896697998046875, 26.568206787109375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000081.npy"}
|
|
{"epoch": 0.11894273127753303, "step": 82, "batch_size": 64, "mean": 15.803937911987305, "std": 24.256502151489258, "min": -47.51829528808594, "p10": -8.055263900756835, "median": 14.922784805297852, "p90": 42.38107376098633, "max": 97.15487670898438, "pos_frac": 0.796875, "sample": [-3.4057846069335938, 25.678482055664062, -13.322952270507812, 16.491355895996094, 18.992496490478516, 17.35978889465332, -19.924888610839844, 13.035072326660156, 21.599395751953125, 49.551124572753906, 56.40773010253906, 11.951194763183594, -8.298786163330078, 6.85089111328125, 13.878232955932617, 87.70201110839844, 10.640220642089844, -44.71629333496094, -6.745429992675781, 28.468063354492188, 1.7757949829101562, 29.269088745117188, 42.723533630371094, 3.880491256713867, -14.722457885742188, -3.7060470581054688, 3.2348403930664062, 18.642234802246094, 18.27710723876953, 11.499275207519531, 15.86981201171875, 31.578140258789062, 29.723731994628906, -0.7382297515869141, 60.6732177734375, 17.083858489990234, 41.582000732421875, 14.122615814208984, 9.12813949584961, 2.1035842895507812, -0.9564132690429688, 24.36102294921875, 46.77458190917969, -13.2176513671875, 25.321853637695312, 17.31414031982422, 6.494070053100586, 41.24199295043945, 25.02680206298828, -7.4870452880859375, 6.7437286376953125, 6.424591064453125, -47.51829528808594, 5.931938171386719, 27.714462280273438, 19.518367767333984, 7.612342834472656, 22.719406127929688, 15.722953796386719, 97.15487670898438, 12.505058288574219, 17.403148651123047, 0.117462158203125, 40.33595275878906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000082.npy"}
|
|
{"epoch": 0.12041116005873716, "step": 83, "batch_size": 64, "mean": 18.518720626831055, "std": 21.40931510925293, "min": -22.627098083496094, "p10": -5.940245056152342, "median": 15.556842803955078, "p90": 51.75247802734377, "max": 73.0406494140625, "pos_frac": 0.8125, "sample": [17.386795043945312, 19.720165252685547, 14.234439849853516, 60.872100830078125, 37.324432373046875, 11.261219024658203, -4.923431396484375, 54.24871826171875, 45.92791748046875, 7.5723114013671875, 8.031272888183594, 15.683738708496094, 5.9061737060546875, -2.5592803955078125, 16.774017333984375, 58.87477111816406, -11.838897705078125, 73.0406494140625, 8.809921264648438, 58.472084045410156, 32.51780700683594, 9.684440612792969, 44.8612060546875, -15.106147766113281, -22.627098083496094, -6.3760223388671875, 45.762176513671875, -15.313583374023438, 16.340822219848633, 54.53126525878906, 15.429946899414062, 12.137161254882812, 4.456205368041992, 3.2544174194335938, 18.838027954101562, 12.137123107910156, 22.977882385253906, -0.5535068511962891, 32.70941162109375, 9.185115814208984, 16.697250366210938, 12.511295318603516, 36.06221008300781, 61.61383056640625, 18.119239807128906, 40.33622741699219, 5.638233184814453, -15.184257507324219, -1.4941215515136719, 1.5582389831542969, -12.060226440429688, 38.73249816894531, 8.725143432617188, 36.68492126464844, 20.801345825195312, 19.4459228515625, -2.2068862915039062, 6.904331207275391, 16.508167266845703, 26.45423126220703, 37.217506408691406, 15.296340942382812, 23.550827026367188, 3.620086669921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000083.npy"}
|
|
{"epoch": 0.12187958883994127, "step": 84, "batch_size": 64, "mean": 16.74671173095703, "std": 16.72605323791504, "min": -32.87890625, "p10": -1.0438888549804688, "median": 14.477331161499023, "p90": 36.8553955078125, "max": 67.62715148925781, "pos_frac": 0.859375, "sample": [7.983428955078125, 11.10418701171875, 3.6457901000976562, 4.292390823364258, 37.06842041015625, 5.33001708984375, 4.330909729003906, 19.116905212402344, 13.033121109008789, 10.194259643554688, 25.90671157836914, 29.05121612548828, -14.40380859375, 15.010009765625, 21.23989486694336, 29.18608856201172, 8.742116928100586, 67.62715148925781, 25.874156951904297, 2.2014312744140625, 7.090496063232422, 27.660438537597656, 9.681602478027344, 6.358034133911133, -1.1107025146484375, 36.35833740234375, 21.270957946777344, 30.01068115234375, -32.87890625, -1.0673503875732422, 46.22279357910156, 4.050636291503906, 32.083736419677734, 14.463603973388672, 14.441162109375, 2.8726158142089844, 40.29912567138672, 35.54351806640625, 25.283340454101562, 21.300399780273438, 4.031883239746094, 28.53235626220703, -2.3978347778320312, 14.491058349609375, 25.345001220703125, 25.631458282470703, -0.9891452789306641, -2.0407962799072266, 28.666099548339844, 29.214920043945312, 2.132965087890625, 37.43214416503906, 45.514068603515625, 18.760589599609375, 50.228179931640625, 13.516067504882812, 26.603347778320312, 6.073326110839844, -10.775482177734375, 27.875015258789062, 11.526363372802734, 16.833511352539062, 9.389200210571289, -0.2736492156982422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000084.npy"}
|
|
{"epoch": 0.12334801762114538, "step": 85, "batch_size": 64, "mean": 16.87063217163086, "std": 20.186023712158203, "min": -23.274864196777344, "p10": -3.1044647216796872, "median": 14.781204223632812, "p90": 47.97838745117188, "max": 61.40452575683594, "pos_frac": 0.75, "sample": [36.108482360839844, 10.595710754394531, 27.12236785888672, 3.031848907470703, 2.466094970703125, 20.286495208740234, 19.559226989746094, 3.6063880920410156, 52.65227508544922, 7.4774017333984375, -2.4868412017822266, 5.4001007080078125, 51.11663818359375, 20.820693969726562, 19.62419319152832, 18.436790466308594, -3.1666259765625, -13.072044372558594, 17.811141967773438, 0.22602081298828125, -23.274864196777344, 57.11262512207031, 14.7039794921875, -1.9414119720458984, -0.6116180419921875, 39.76036071777344, 7.958580017089844, -3.14111328125, 15.567298889160156, 9.764808654785156, 39.14947509765625, -3.018951416015625, -11.390281677246094, -2.342832565307617, 12.50152587890625, 4.961801528930664, 58.18934631347656, 15.596233367919922, 31.6099853515625, 0.6104450225830078, -2.5425548553466797, 61.40452575683594, -1.350728988647461, 37.43141174316406, 46.571807861328125, 12.68878173828125, 45.66480255126953, 51.27772521972656, 7.948699951171875, 5.046758651733398, 25.322181701660156, 42.86089324951172, -1.0431671142578125, -4.838039398193359, 14.858428955078125, 15.786478042602539, 48.581207275390625, 17.673866271972656, 26.793869018554688, 32.068965911865234, -16.324493408203125, 17.342418670654297, -0.4628181457519531, 37.577674865722656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000085.npy"}
|
|
{"epoch": 0.12481644640234948, "step": 86, "batch_size": 64, "mean": 18.096515655517578, "std": 29.81615447998047, "min": -41.666778564453125, "p10": -11.002169799804687, "median": 11.881586074829102, "p90": 53.87670745849612, "max": 112.893310546875, "pos_frac": 0.765625, "sample": [-9.513526916503906, 14.035266876220703, 56.88267517089844, -14.189456939697266, 32.15135192871094, 24.247451782226562, 24.10491371154785, -3.054168701171875, 83.802978515625, 9.7279052734375, 5.738363265991211, -0.28404998779296875, 1.8453598022460938, 17.718719482421875, -4.986663818359375, 16.87200164794922, 41.943450927734375, 18.458396911621094, -2.4226646423339844, -34.370819091796875, 47.060882568359375, 16.601871490478516, 26.93288803100586, 9.097841262817383, 9.293289184570312, -13.185253143310547, 0.4285125732421875, 56.79777526855469, -5.028438568115234, 6.34037971496582, 31.84762954711914, 17.64954376220703, 6.647939682006836, -41.666778564453125, 36.944854736328125, 45.5943603515625, 23.16461181640625, -27.76386260986328, 59.19776916503906, 3.525209426879883, 112.893310546875, 16.99435806274414, 24.050933837890625, 9.68648910522461, -2.3660888671875, 108.02398681640625, 6.306280136108398, 97.97412109375, 38.463653564453125, 0.395751953125, -25.470436096191406, 34.692138671875, 8.595680236816406, 30.88561248779297, 24.027626037597656, 32.12345504760742, 33.969215393066406, 3.80059814453125, 4.383598327636719, 5.421382904052734, -2.9069576263427734, -11.640159606933594, 3.2655487060546875, 16.41836166381836], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000086.npy"}
|
|
{"epoch": 0.1262848751835536, "step": 87, "batch_size": 64, "mean": 20.468475341796875, "std": 26.248069763183594, "min": -52.390533447265625, "p10": -4.003399658203124, "median": 18.04893684387207, "p90": 53.062477111816406, "max": 106.63444519042969, "pos_frac": 0.8125, "sample": [19.96561622619629, 28.346519470214844, 0.3490924835205078, 8.488004684448242, 32.272125244140625, -2.2653732299804688, 21.676422119140625, -0.5678672790527344, 1.679168701171875, -2.046173095703125, -14.763717651367188, 33.684478759765625, -4.644493103027344, 53.79588317871094, 24.031097412109375, 45.62835693359375, 45.932823181152344, 9.142885208129883, 77.56118774414062, 44.314666748046875, 19.316444396972656, 16.749778747558594, 6.620841979980469, 34.790733337402344, 31.90108871459961, 2.195566177368164, 31.189712524414062, 106.63444519042969, 13.32659912109375, 0.8606147766113281, 13.678728103637695, 59.15423583984375, 52.417205810546875, 7.351814270019531, 48.08238220214844, 21.35875701904297, 12.778093338012695, 33.454830169677734, 16.834381103515625, 53.33592224121094, 0.6828727722167969, 65.18934631347656, 17.163070678710938, 18.517898559570312, -52.390533447265625, 8.61323356628418, 32.86719512939453, -29.211992263793945, 32.005615234375, 57.09928894042969, -11.114398956298828, -10.444976806640625, 7.191719055175781, 17.579975128173828, -1.987823486328125, 52.4244384765625, 28.041854858398438, 28.044967651367188, -28.5657958984375, 18.552200317382812, 15.068572998046875, 1.9510002136230469, 40.59925842285156, -2.5075149536132812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000087.npy"}
|
|
{"epoch": 0.1277533039647577, "step": 88, "batch_size": 64, "mean": 15.591540336608887, "std": 24.60508918762207, "min": -40.677581787109375, "p10": -4.452506637573241, "median": 11.89677906036377, "p90": 46.393656921386736, "max": 97.76141357421875, "pos_frac": 0.78125, "sample": [9.71435546875, 11.355690002441406, 55.506439208984375, 48.41633605957031, 29.595626831054688, 37.73382568359375, 2.6389541625976562, 51.519500732421875, -0.849365234375, 11.467386245727539, 25.386016845703125, 5.895381927490234, 3.7806854248046875, 36.338134765625, -18.461532592773438, 24.990554809570312, -16.700546264648438, 15.104629516601562, -35.73109436035156, -40.677581787109375, -3.032510757446289, 58.81060791015625, 28.76013946533203, -3.576953887939453, 38.176231384277344, 4.256153106689453, 22.534271240234375, -4.8277435302734375, 18.899381637573242, 20.391754150390625, 34.18159484863281, 6.584022521972656, -0.7292137145996094, 24.57067108154297, 1.4822769165039062, 73.0606689453125, 13.755943298339844, 41.674072265625, 4.134349822998047, 97.76141357421875, 4.08929443359375, 35.061973571777344, 4.215976715087891, 12.566202163696289, 12.326171875, 19.04228973388672, 3.201894760131836, -1.0903434753417969, 26.604446411132812, 63.973876953125, 28.56157684326172, 13.005348205566406, 1.9707221984863281, 7.0715484619140625, -38.0364990234375, -1.524322509765625, 9.008712768554688, -1.3854236602783203, 9.177831649780273, 14.329498291015625, 16.911863327026367, -17.52599334716797, 8.959396362304688, 33.45196533203125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000088.npy"}
|
|
{"epoch": 0.12922173274596183, "step": 89, "batch_size": 64, "mean": 17.231029510498047, "std": 21.109743118286133, "min": -43.97027587890625, "p10": -1.3126770019531238, "median": 14.717820167541504, "p90": 39.61615219116211, "max": 94.75932312011719, "pos_frac": 0.890625, "sample": [2.9037704467773438, 1.5571231842041016, 4.095867156982422, 17.928617477416992, -6.866680145263672, 39.80543518066406, 29.006362915039062, 14.521659851074219, -3.2023468017578125, 49.344276428222656, 46.957313537597656, 20.573646545410156, 18.479408264160156, 36.49646759033203, 27.24793243408203, 27.756786346435547, 12.585678100585938, 8.897415161132812, -1.9381465911865234, 15.220829010009766, 12.217536926269531, 1.6230831146240234, 14.673698425292969, 9.382282257080078, -7.262725830078125, 2.9112987518310547, 0.10345840454101562, 8.914905548095703, 55.562164306640625, 16.52979278564453, 19.931053161621094, 71.73916625976562, 14.131641387939453, 39.17449188232422, 17.71676254272461, 4.536376953125, 13.383703231811523, 94.75932312011719, 2.5613746643066406, 0.02447509765625, 24.874134063720703, 2.2827491760253906, 17.10286521911621, 25.75728988647461, 14.761941909790039, 5.541181564331055, 9.341753005981445, 5.545269012451172, 5.37750244140625, 28.950485229492188, 17.00152587890625, 30.73223876953125, -1.8857421875, 12.681236267089844, 33.65496063232422, -43.97027587890625, -16.116455078125, 22.794471740722656, 15.8310546875, 15.336872100830078, 22.947479248046875, 21.586387634277344, 73.92231750488281, 4.749275207519531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000089.npy"}
|
|
{"epoch": 0.13069016152716592, "step": 90, "batch_size": 64, "mean": 22.610496520996094, "std": 31.92949104309082, "min": -36.97068786621094, "p10": -5.530064582824705, "median": 15.108715057373047, "p90": 70.7471290588379, "max": 114.47027587890625, "pos_frac": 0.828125, "sample": [50.595123291015625, -34.662132263183594, 27.505111694335938, 70.36190795898438, 20.25806427001953, 29.917877197265625, 1.1355819702148438, -2.0442466735839844, 48.672088623046875, -0.6311931610107422, 38.376869201660156, 15.578598022460938, 0.8684921264648438, 11.003677368164062, 29.544708251953125, 70.91222381591797, -2.684459686279297, 114.47027587890625, 40.56620788574219, 1.4180221557617188, 3.2675647735595703, 7.254249572753906, 14.638832092285156, -6.3369903564453125, 90.961181640625, 15.843770980834961, 22.3203125, 91.7923583984375, 4.932899475097656, 20.661869049072266, 111.0770263671875, 15.751998901367188, -13.012451171875, 8.657613754272461, 2.0638504028320312, -36.97068786621094, 23.58443832397461, 6.2947998046875, 50.036895751953125, 4.895423889160156, 9.885475158691406, 10.040618896484375, -17.501846313476562, -3.647237777709961, 13.819305419921875, 64.49676513671875, 77.427978515625, -29.252159118652344, -18.18280029296875, 9.683792114257812, 28.07172393798828, 1.8833236694335938, 16.62318229675293, 25.316879272460938, 50.887298583984375, 27.039878845214844, 9.867431640625, 8.215364456176758, 40.233551025390625, 11.427326202392578, 17.45380401611328, 12.59358024597168, 77.99099731445312, 33.819671630859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000090.npy"}
|
|
{"epoch": 0.13215859030837004, "step": 91, "batch_size": 64, "mean": 20.096027374267578, "std": 26.49957847595215, "min": -24.979217529296875, "p10": -10.978360748291015, "median": 18.1312255859375, "p90": 54.3477851867676, "max": 91.18373107910156, "pos_frac": 0.75, "sample": [15.539474487304688, 44.90776062011719, -22.530609130859375, 7.010990142822266, 56.68470764160156, 0.021951675415039062, -7.990516662597656, 37.61727523803711, 4.643333435058594, 16.9134521484375, 35.47160339355469, 31.450714111328125, 34.088966369628906, 11.744178771972656, -24.979217529296875, 91.18373107910156, -9.99111557006836, 2.611114501953125, 6.154388427734375, 67.64533996582031, 10.514358520507812, 27.217437744140625, 88.97467041015625, -10.302078247070312, 25.556480407714844, 47.596656799316406, -11.268196105957031, 1.7152347564697266, 18.832019805908203, -3.513275146484375, -8.988716125488281, 18.8508243560791, 47.57691955566406, -14.127334594726562, 17.71160125732422, 17.598236083984375, 46.547332763671875, -9.654243469238281, 18.55084991455078, -2.7443695068359375, 7.0824432373046875, 31.40045166015625, 48.89496612548828, 3.4204025268554688, -12.07061767578125, 29.622024536132812, 20.656715393066406, 65.37802124023438, 28.27203369140625, 31.65441131591797, 13.786430358886719, -18.92888641357422, 0.2161846160888672, 19.48480224609375, -1.6265430450439453, 38.98841094970703, -0.2514381408691406, 68.69851684570312, 36.37266540527344, 18.762451171875, 41.13127136230469, -12.6326904296875, 37.71095275878906, 65.28092193603516], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000091.npy"}
|
|
{"epoch": 0.13362701908957417, "step": 92, "batch_size": 64, "mean": 19.803434371948242, "std": 26.34551239013672, "min": -35.50672912597656, "p10": -7.9789184570312495, "median": 17.13759422302246, "p90": 47.492624282836914, "max": 98.1591796875, "pos_frac": 0.75, "sample": [-2.369993209838867, 4.9323272705078125, 9.123619079589844, 38.050086975097656, -7.833343505859375, 5.6715545654296875, -35.50672912597656, -12.85081672668457, 36.218902587890625, 20.165512084960938, 53.20328903198242, 0.5896625518798828, 33.7237548828125, 2.1810836791992188, -1.0770111083984375, 89.42616271972656, 46.83818817138672, 98.1591796875, -26.084259033203125, -5.3493804931640625, 21.728466033935547, -8.555641174316406, 25.054428100585938, -5.652923583984375, -11.030838012695312, 44.43672180175781, -7.880096435546875, 61.897979736328125, 26.71471405029297, 6.594121932983398, 62.34076690673828, -1.897073745727539, 30.22478485107422, 1.2111949920654297, 44.382171630859375, 5.53425407409668, 24.100936889648438, 78.96421813964844, 9.000247955322266, 45.96211242675781, 36.987693786621094, 45.6162109375, -1.2293472290039062, -1.2851104736328125, 6.450416564941406, 18.22763442993164, 0.1395721435546875, 33.50555419921875, 30.241558074951172, 8.595132827758789, 0.2909965515136719, -9.904468536376953, 47.749629974365234, 12.864385604858398, 16.04755401611328, 20.10132598876953, 31.267127990722656, 38.87968444824219, 34.8944091796875, 19.816509246826172, 9.935964584350586, -8.021270751953125, 46.8929443359375, 29.01335906982422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000092.npy"}
|
|
{"epoch": 0.13509544787077826, "step": 93, "batch_size": 64, "mean": 20.0338077545166, "std": 23.45270347595215, "min": -28.685585021972656, "p10": -2.1178302764892574, "median": 13.184707641601562, "p90": 49.83616561889649, "max": 122.54043579101562, "pos_frac": 0.84375, "sample": [18.34893035888672, 7.58745002746582, 20.69994354248047, 52.485755920410156, 50.22766876220703, 41.933815002441406, 5.1890716552734375, 18.37908935546875, 35.173057556152344, -7.459739685058594, 57.271820068359375, 10.754131317138672, 41.219783782958984, 43.40937805175781, 122.54043579101562, 9.86838150024414, 10.551811218261719, 7.0926055908203125, 6.075462341308594, 71.7012710571289, -6.262142181396484, 11.525297164916992, 18.300106048583984, -0.4457550048828125, 27.211517333984375, 11.956672668457031, 18.50403594970703, 6.822975158691406, -2.330432891845703, 12.77008056640625, -0.42577552795410156, 13.748504638671875, 8.185367584228516, 20.20648193359375, 30.357810974121094, 39.46507263183594, 52.24462890625, -14.320831298828125, -3.632863998413086, 37.62353515625, 35.207618713378906, 32.090755462646484, 22.26172637939453, 5.472738265991211, 9.315872192382812, 7.352272033691406, 23.474761962890625, 2.422100067138672, 49.158409118652344, 3.0966224670410156, -1.6217575073242188, 50.12663269042969, 10.961936950683594, 12.146926879882812, 4.872489929199219, 5.22430419921875, -13.874412536621094, 31.819778442382812, 36.91459655761719, 30.816299438476562, 34.733428955078125, 13.599334716796875, -28.685585021972656, 0.7223834991455078], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000093.npy"}
|
|
{"epoch": 0.13656387665198239, "step": 94, "batch_size": 64, "mean": 22.42888641357422, "std": 24.066194534301758, "min": -16.024642944335938, "p10": -5.482787513732909, "median": 20.203731536865234, "p90": 53.35195655822754, "max": 86.09991455078125, "pos_frac": 0.796875, "sample": [21.793800354003906, -11.643081665039062, 37.48335266113281, 15.492401123046875, 7.184059143066406, -9.777206420898438, 7.434135437011719, 52.598270416259766, 4.832685470581055, -12.89654541015625, 21.71909523010254, 12.949264526367188, 20.53404998779297, 4.462608337402344, 9.389625549316406, 0.6905345916748047, 47.132415771484375, 11.51070785522461, -5.908359527587891, 86.09991455078125, 6.2247772216796875, 20.935222625732422, 24.944168090820312, 45.10395050048828, 62.821990966796875, 19.8734130859375, 12.943801879882812, 23.797142028808594, 17.247852325439453, -2.8869781494140625, -1.2464656829833984, 14.983449935913086, 30.600927352905273, 42.45068359375, 37.252681732177734, 15.736225128173828, 30.425491333007812, 53.674964904785156, 30.616416931152344, 21.434385299682617, 4.666728973388672, 39.13469696044922, 1.9366226196289062, -10.262809753417969, 57.45970153808594, 61.01939392089844, -4.489786148071289, 52.423126220703125, 81.54843139648438, -3.8719940185546875, -16.024642944335938, 75.24969482421875, 22.827346801757812, 43.640838623046875, -3.1736602783203125, 15.613372802734375, -9.847381591796875, -2.078838348388672, 30.25597381591797, 10.297409057617188, 30.874691009521484, 43.98127746582031, 50.5157470703125, 35.737030029296875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000094.npy"}
|
|
{"epoch": 0.13803230543318648, "step": 95, "batch_size": 64, "mean": 18.47789764404297, "std": 21.90042495727539, "min": -49.39756774902344, "p10": -2.5551736831665033, "median": 14.963945388793945, "p90": 47.15420989990234, "max": 69.49911499023438, "pos_frac": 0.84375, "sample": [11.214523315429688, 32.996429443359375, 11.582466125488281, 19.147056579589844, 16.40754508972168, 53.829833984375, 9.202892303466797, 5.870710372924805, -4.557186126708984, 16.789466857910156, 18.414518356323242, 4.006305694580078, 28.108753204345703, 62.74334716796875, 43.37785339355469, 0.6395111083984375, 30.66675567626953, 4.329423904418945, 11.25105094909668, 33.91021728515625, 13.572357177734375, 30.565963745117188, 16.563982009887695, 12.505136489868164, 9.077219009399414, -6.401420593261719, 8.233757019042969, 37.950782775878906, 17.52777671813965, 23.98729705810547, -4.601524353027344, -1.8485260009765625, 47.21875, 69.49911499023438, 64.81298065185547, 45.608428955078125, 24.949432373046875, 2.239126205444336, 1.8670425415039062, 47.00361633300781, 37.190425872802734, 61.1951904296875, 58.49287414550781, 3.8100738525390625, 13.628721237182617, 8.520120620727539, -0.9375228881835938, 26.173730850219727, 18.378944396972656, -49.39756774902344, 1.4782257080078125, 1.553375244140625, -17.42132568359375, 34.02293395996094, 36.538246154785156, 13.1588134765625, 34.631591796875, 11.096988677978516, 14.666370391845703, 15.261520385742188, -2.858022689819336, -0.7559375762939453, 15.890045166015625, -21.995201110839844], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000095.npy"}
|
|
{"epoch": 0.1395007342143906, "step": 96, "batch_size": 64, "mean": 24.822622299194336, "std": 27.602855682373047, "min": -39.830718994140625, "p10": 0.5880634307861329, "median": 18.398839950561523, "p90": 65.52639389038089, "max": 122.05363464355469, "pos_frac": 0.921875, "sample": [46.87702941894531, 19.361907958984375, 30.907577514648438, 24.616687774658203, 23.24932861328125, 50.726165771484375, 6.3170928955078125, 1.960275650024414, 0.6894435882568359, 27.426307678222656, 88.94416809082031, 12.382667541503906, 13.504112243652344, -0.6242485046386719, 0.4895782470703125, 18.836986541748047, 25.14459991455078, 38.66126251220703, 74.21436309814453, 69.20391845703125, 8.898017883300781, 35.85065460205078, 29.23169708251953, 122.05363464355469, 56.94550323486328, 15.145179748535156, 20.80142593383789, 14.545394897460938, 14.878555297851562, 14.764537811279297, 8.362899780273438, 102.28732299804688, 28.2634334564209, 24.844802856445312, 3.5061187744140625, 70.40219116210938, 41.97314453125, 5.3510894775390625, 29.477325439453125, 2.166950225830078, 35.82306671142578, 26.814359664916992, 16.40149688720703, -4.709175109863281, 12.958778381347656, -0.6824359893798828, 4.348663330078125, 31.90355682373047, 1.0843868255615234, 17.960693359375, 10.483734130859375, 42.809349060058594, -10.933773040771484, 1.4267349243164062, 45.095489501953125, 6.326438903808594, 14.559488296508789, 43.49317932128906, 74.53105163574219, 7.929756164550781, 0.5446147918701172, 24.39173126220703, -39.830718994140625, 3.278289794921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000096.npy"}
|
|
{"epoch": 0.14096916299559473, "step": 97, "batch_size": 64, "mean": 19.596725463867188, "std": 21.54547119140625, "min": -42.99547576904297, "p10": -4.234297943115233, "median": 17.719078063964844, "p90": 43.995846557617185, "max": 84.80709838867188, "pos_frac": 0.828125, "sample": [14.414527893066406, 42.75238037109375, 12.66314697265625, 3.9281082153320312, -2.8031578063964844, 2.695758819580078, 15.139202117919922, 23.452299118041992, 1.337860107421875, 37.32341003417969, 17.563190460205078, 34.090843200683594, -0.6842441558837891, 16.856807708740234, 67.06588745117188, 45.53523254394531, -11.687156677246094, 19.36700439453125, 1.3021774291992188, 11.50633430480957, 23.098861694335938, 6.568748474121094, 13.332490921020508, -12.7215576171875, 55.48504638671875, 17.954561233520508, 21.478858947753906, 24.677154541015625, 34.21637725830078, 13.600627899169922, 22.024768829345703, -2.8685550689697266, 13.066497802734375, -7.539789199829102, 37.05400848388672, 28.559661865234375, 42.77574157714844, 24.106830596923828, -42.99547576904297, -4.819616317749023, 40.340789794921875, 30.51092529296875, -6.4810791015625, 5.069690704345703, 43.86271667480469, 28.453685760498047, 2.1962528228759766, 17.87496566772461, 18.57271385192871, 13.611015319824219, -2.7689266204833984, 43.79441833496094, -6.797019958496094, 42.57809066772461, 4.206073760986328, 48.09605407714844, 1.1703758239746094, 44.05290222167969, 41.021568298339844, 52.89801025390625, 27.660308837890625, 4.955768585205078, 84.80709838867188, 15.629150390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000097.npy"}
|
|
{"epoch": 0.14243759177679882, "step": 98, "batch_size": 64, "mean": 22.12537384033203, "std": 27.538694381713867, "min": -70.3204345703125, "p10": -9.212904357910155, "median": 22.232389450073242, "p90": 54.628932189941416, "max": 91.09271240234375, "pos_frac": 0.8125, "sample": [28.430343627929688, 6.894113540649414, 59.126922607421875, 32.0848388671875, 5.1128692626953125, 16.880746841430664, 19.09294891357422, 22.31352996826172, 21.431900024414062, 5.231452941894531, 25.26066017150879, 4.52020263671875, 18.07117462158203, 15.939353942871094, 55.49072265625, 40.93769073486328, -10.029529571533203, 7.974029541015625, 2.155670166015625, -17.849075317382812, 35.48684310913086, 30.575164794921875, -18.41607666015625, -18.014007568359375, 47.733253479003906, 43.11484146118164, 29.847671508789062, 57.49407196044922, 52.61808776855469, 23.629480361938477, 30.844451904296875, -1.1982154846191406, 49.87427520751953, 91.09271240234375, -70.3204345703125, 64.87165832519531, 49.69001770019531, 23.14495849609375, 33.22064208984375, 11.462333679199219, -7.307445526123047, 34.286102294921875, -33.10736083984375, 19.959625244140625, 30.145599365234375, 2.4951705932617188, 34.53912353515625, 69.95388793945312, 48.78790283203125, 32.228267669677734, 11.757675170898438, 2.5000381469726562, 52.4219970703125, 80.05519104003906, 30.06017303466797, 22.151248931884766, 19.00273895263672, 10.156026840209961, -0.22948837280273438, -5.232048034667969, 46.749794006347656, -1.8512306213378906, 3.3935680389404297, -12.714927673339844], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000098.npy"}
|
|
{"epoch": 0.14390602055800295, "step": 99, "batch_size": 64, "mean": 26.147567749023438, "std": 30.607072830200195, "min": -20.89226531982422, "p10": -2.8760292053222654, "median": 17.738847732543945, "p90": 59.72794799804688, "max": 129.6669464111328, "pos_frac": 0.859375, "sample": [0.5383663177490234, -2.9816970825195312, 5.171222686767578, 18.479389190673828, 4.708202362060547, 36.23849868774414, 17.81466293334961, 20.674468994140625, 83.554931640625, 17.09437370300293, 17.66303253173828, 91.21810913085938, 12.057720184326172, -20.605880737304688, 59.44517517089844, 1.962005615234375, 16.92449951171875, 31.146697998046875, -0.2730216979980469, 12.930252075195312, 48.9976806640625, 11.653076171875, 8.369489669799805, 126.3035888671875, 6.169069290161133, 42.3963623046875, 5.1386566162109375, -13.635269165039062, 10.728532791137695, -20.89226531982422, -3.8834266662597656, 46.34918975830078, 30.892845153808594, 19.995906829833984, 48.38688659667969, 129.6669464111328, 59.84913635253906, 26.365718841552734, 15.78704833984375, 15.006195068359375, 3.6376609802246094, 23.48452377319336, 29.35132598876953, 58.7165412902832, 84.7645492553711, 2.8535537719726562, 11.751884460449219, 54.61138916015625, -3.1796836853027344, 3.0179080963134766, 27.38922119140625, 64.83694458007812, 24.41818618774414, 9.882862091064453, 43.2762451171875, -2.6294708251953125, 34.45549392700195, 13.878534317016602, 51.47857666015625, 49.41743469238281, 19.728012084960938, 39.11296081542969, 1.9853935241699219, -10.202133178710938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000099.npy"}
|
|
{"epoch": 0.14537444933920704, "step": 100, "batch_size": 64, "mean": 16.52963638305664, "std": 26.838048934936523, "min": -47.70014953613281, "p10": -12.762859535217281, "median": 16.262008666992188, "p90": 52.61933021545412, "max": 109.21725463867188, "pos_frac": 0.796875, "sample": [44.952056884765625, 27.46226692199707, -5.530550003051758, 13.574607849121094, 17.829158782958984, -0.1621551513671875, 28.56648063659668, 16.924354553222656, 8.599052429199219, 20.909912109375, 10.628097534179688, 0.8303680419921875, 29.259872436523438, 18.016281127929688, 2.852884292602539, 64.03085327148438, 16.539093017578125, -34.3779296875, -14.19732666015625, -32.795623779296875, -9.458654403686523, -6.442951202392578, 1.9720420837402344, -43.83759689331055, 47.5880241394043, 19.0555419921875, 42.018150329589844, 21.492919921875, 1.0244312286376953, 5.856100082397461, 34.62825012207031, 0.36504364013671875, 38.4987907409668, 2.8512535095214844, 55.14817810058594, 27.03765106201172, 7.639326095581055, 11.035163879394531, 46.36145782470703, 17.66571044921875, 32.171329498291016, 12.705001831054688, 9.933578491210938, 0.6940441131591797, 24.254005432128906, 28.68719482421875, 1.0218582153320312, -20.055084228515625, -4.9653778076171875, 56.3441162109375, 28.72821807861328, 57.54088592529297, 21.21344566345215, -47.70014953613281, 109.21725463867188, -5.458192825317383, 15.98492431640625, 61.43077087402344, -14.178947448730469, 54.775604248046875, 33.31813049316406, 6.240020751953125, 28.014328002929688, 13.569110870361328], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000100.npy"}
|
|
{"epoch": 0.14684287812041116, "step": 101, "batch_size": 64, "mean": 19.492538452148438, "std": 29.761550903320312, "min": -35.32715606689453, "p10": -13.373750686645506, "median": 16.496055603027344, "p90": 63.78419876098638, "max": 113.79049682617188, "pos_frac": 0.703125, "sample": [68.73689270019531, 113.79049682617188, 25.222686767578125, 23.279006958007812, -7.25927734375, 76.96788787841797, 84.56063842773438, 6.021369934082031, 71.08871459960938, 82.27969360351562, -19.2364501953125, 21.488616943359375, -22.928787231445312, 38.81975555419922, 11.756216049194336, -4.552909851074219, -11.420928955078125, 11.006559371948242, 44.260032653808594, 24.91600799560547, -5.89141845703125, 18.143264770507812, 40.63935852050781, 45.680381774902344, 17.447952270507812, 15.247137069702148, 7.0724029541015625, 52.22791290283203, 81.77682495117188, 17.971969604492188, 1.316680908203125, -5.285858154296875, -10.808998107910156, 39.06220626831055, 16.5666446685791, 30.02490234375, 49.72575378417969, 33.348968505859375, -4.86480712890625, 7.321750640869141, -1.8024978637695312, -14.210674285888672, 16.425466537475586, -2.2147598266601562, 7.753929138183594, -9.10261344909668, 23.843467712402344, 25.45029067993164, 7.34904670715332, 19.80121421813965, -35.32715606689453, 31.828048706054688, 10.067153930664062, -19.805240631103516, 22.646434783935547, 15.737228393554688, -6.249248504638672, 31.059078216552734, 5.383026123046875, -1.4041194915771484, 43.45863342285156, -20.608295440673828, 29.996421813964844, -18.071578979492188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000101.npy"}
|
|
{"epoch": 0.14831130690161526, "step": 102, "batch_size": 64, "mean": 14.546875, "std": 27.602758407592773, "min": -47.592010498046875, "p10": -15.616078186035155, "median": 13.903221130371094, "p90": 40.51064529418946, "max": 105.48599243164062, "pos_frac": 0.640625, "sample": [2.2018890380859375, 36.23106384277344, 39.802757263183594, 6.426719665527344, 13.052566528320312, 4.923492431640625, -18.239517211914062, -31.97979736328125, 19.71955680847168, -45.729644775390625, 24.634765625, 63.07801818847656, -3.7571640014648438, 31.819793701171875, 21.072349548339844, -5.968372344970703, 11.722938537597656, 27.69827651977539, 36.943511962890625, -0.8930854797363281, -4.059967041015625, -3.1580448150634766, 30.852294921875, -2.5217456817626953, 40.81402587890625, 17.073352813720703, 29.981313705444336, 105.48599243164062, -1.766387939453125, -0.2853736877441406, 82.3952407836914, -0.34956932067871094, 31.490909576416016, 35.458717346191406, 3.1597938537597656, -10.696380615234375, 22.47776985168457, -47.592010498046875, -0.386810302734375, -19.66616439819336, 0.3649864196777344, -26.292922973632812, -16.116546630859375, 1.3961257934570312, 39.68415069580078, 17.20062255859375, 29.857826232910156, 63.049652099609375, 34.67816162109375, 14.753875732421875, 53.14855194091797, -14.448318481445312, -3.3160171508789062, 9.21323013305664, 44.41322326660156, -9.736099243164062, 30.381431579589844, 25.246376037597656, 18.98101234436035, 32.61797332763672, -9.374198913574219, -11.758964538574219, 30.574264526367188, 35.01456832885742], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000102.npy"}
|
|
{"epoch": 0.14977973568281938, "step": 103, "batch_size": 64, "mean": 27.926231384277344, "std": 29.527536392211914, "min": -17.285446166992188, "p10": -5.126941680908202, "median": 24.435474395751953, "p90": 63.84285812377931, "max": 124.48117065429688, "pos_frac": 0.8125, "sample": [-1.5926322937011719, 5.9008941650390625, 12.38984489440918, 24.341217041015625, -4.5525360107421875, -15.744888305664062, 30.752124786376953, 32.70379638671875, 10.71551513671875, 18.094432830810547, 23.373558044433594, 35.25251007080078, 11.776290893554688, 22.039939880371094, 0.46738433837890625, 53.42401123046875, 84.7528076171875, 29.03125762939453, 39.128868103027344, 42.56739044189453, -12.372482299804688, -1.85882568359375, 33.73854064941406, 31.90692138671875, -17.285446166992188, 57.023658752441406, 28.121780395507812, -12.707199096679688, 8.413742065429688, 19.110328674316406, 24.535812377929688, 111.06781005859375, 18.95703887939453, 20.983272552490234, 52.84625244140625, 43.77904510498047, -5.373115539550781, -11.44036865234375, 18.195907592773438, 16.80732536315918, -2.4103431701660156, 29.986007690429688, 43.76250457763672, -7.56365966796875, 17.084762573242188, 82.95977020263672, 61.56011962890625, 24.52973175048828, 124.48117065429688, 76.09379577636719, 81.92215728759766, 58.22529602050781, 12.428543090820312, 28.423721313476562, 55.27147674560547, -1.1514225006103516, 28.494232177734375, 2.0223865509033203, 4.650447845458984, 43.64282989501953, 64.82117462158203, 15.89602279663086, 25.462478637695312, 37.41380310058594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000103.npy"}
|
|
{"epoch": 0.1512481644640235, "step": 104, "batch_size": 64, "mean": 26.393295288085938, "std": 34.32978820800781, "min": -44.71409606933594, "p10": -6.998972702026367, "median": 18.070026397705078, "p90": 70.65101776123048, "max": 128.39605712890625, "pos_frac": 0.796875, "sample": [65.34075927734375, -2.2772216796875, 10.843076705932617, 9.826839447021484, 48.702178955078125, -0.3012542724609375, -15.66583251953125, 2.2314319610595703, 4.9713287353515625, 31.35083770751953, -9.486581802368164, 23.356441497802734, 113.58087158203125, 44.59678649902344, 62.557090759277344, -0.19113922119140625, 72.06259155273438, 29.107454299926758, -9.297515869140625, 88.13241577148438, 23.720809936523438, 10.075798034667969, 19.65804672241211, 7.2718963623046875, 12.757043838500977, 2.6565704345703125, 128.39605712890625, -29.00426483154297, 44.02843475341797, 99.95811462402344, 103.24853515625, 21.86382293701172, 9.649166107177734, 47.339332580566406, 9.412555694580078, 9.218414306640625, -6.493511199951172, -13.440231323242188, 78.49008178710938, 12.607398986816406, 42.10100173950195, 2.7289276123046875, 53.25428771972656, 5.2832183837890625, 67.35734558105469, 35.79212188720703, 27.400856018066406, 37.48879623413086, 21.831130981445312, 30.499160766601562, 2.751331329345703, 66.36549377441406, 9.5799560546875, 20.01477813720703, 55.31965637207031, 52.898216247558594, -3.018552780151367, 25.310197830200195, -0.05771636962890625, 16.482006072998047, -7.215599060058594, 2.304891586303711, -44.71409606933594, 8.558815002441406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000104.npy"}
|
|
{"epoch": 0.1527165932452276, "step": 105, "batch_size": 64, "mean": 37.05156707763672, "std": 30.721956253051758, "min": -29.48284912109375, "p10": 3.0505264282226583, "median": 33.948604583740234, "p90": 76.67654876708987, "max": 140.37950134277344, "pos_frac": 0.921875, "sample": [31.759368896484375, 25.681182861328125, 40.168785095214844, 27.544952392578125, 36.78385925292969, 31.359092712402344, 58.52796936035156, 87.28290557861328, 11.782022476196289, -29.48284912109375, -1.0956954956054688, 65.83538818359375, 59.51606750488281, 12.483541488647461, 32.93946838378906, 37.763004302978516, -23.160995483398438, 5.188636779785156, -0.669677734375, 83.27788543701172, 25.260055541992188, 84.63410949707031, 69.82539367675781, 14.7100830078125, 10.12823486328125, 65.1971664428711, 48.79979705810547, 32.16679382324219, 7.538774490356445, 34.957740783691406, 47.478759765625, 114.43936157226562, 15.046707153320312, 41.79032897949219, 12.955490112304688, 16.87378692626953, 52.73490905761719, 71.88264465332031, 35.49073791503906, 14.78744888305664, 42.07130432128906, 80.28564453125, 27.34337043762207, 29.805612564086914, 24.203353881835938, 42.102561950683594, 0.14090347290039062, 53.555747985839844, 55.617008209228516, 71.04571533203125, 25.507469177246094, 21.654861450195312, -8.8360595703125, 6.323642730712891, 47.843017578125, 46.714019775390625, 42.383033752441406, 78.7310791015625, 44.91029357910156, 140.37950134277344, 15.679244995117188, 2.1341934204101562, 11.308273315429688, 60.21321105957031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000105.npy"}
|
|
{"epoch": 0.15418502202643172, "step": 106, "batch_size": 64, "mean": 23.31535530090332, "std": 26.32763671875, "min": -38.53949737548828, "p10": -6.568291473388672, "median": 23.77244758605957, "p90": 55.385108184814456, "max": 109.26697540283203, "pos_frac": 0.828125, "sample": [22.7406005859375, 3.7238426208496094, 4.807104110717773, -1.5818233489990234, 16.87904930114746, -0.4911327362060547, 40.49732971191406, -6.012214660644531, 26.12590789794922, 53.83999252319336, 31.62590789794922, 30.12560272216797, 3.5708961486816406, 61.963096618652344, 20.07721710205078, 64.73243713378906, 2.404691696166992, 12.696016311645508, 53.003143310546875, -0.1522846221923828, 52.30769348144531, 28.93233871459961, 4.174568176269531, 63.945220947265625, 18.34724235534668, 32.247894287109375, 11.136512756347656, -11.950103759765625, 35.47998809814453, -18.462600708007812, 34.570640563964844, 14.459369659423828, 29.59572982788086, 34.30564498901367, 0.3192424774169922, 7.910259246826172, 19.145774841308594, 5.908546447753906, 65.941650390625, 55.39496612548828, 43.13250732421875, 22.354557037353516, -38.53949737548828, 1.441253662109375, 109.26697540283203, 25.423110961914062, 24.80429458618164, 31.430763244628906, 18.9024658203125, 5.548286437988281, 1.7325267791748047, 41.108673095703125, 79.50687408447266, 29.471433639526367, 40.819244384765625, 43.08647155761719, -12.575363159179688, 25.50794219970703, -6.806610107421875, 26.687110900878906, 55.36210632324219, -23.519577026367188, -12.26019287109375, 36.01145935058594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000106.npy"}
|
|
{"epoch": 0.15565345080763582, "step": 107, "batch_size": 64, "mean": 26.43893814086914, "std": 30.21693992614746, "min": -53.694549560546875, "p10": -7.760647583007806, "median": 17.467987060546875, "p90": 65.87166213989258, "max": 100.77372741699219, "pos_frac": 0.875, "sample": [60.0655517578125, 79.48480224609375, 37.70359802246094, 54.714149475097656, 18.85796356201172, 23.66997528076172, 39.773529052734375, -1.7610015869140625, -53.694549560546875, 5.377765655517578, 13.125223159790039, 58.6483154296875, -19.837646484375, 26.888099670410156, 9.51860237121582, 8.917060852050781, 12.057245254516602, 70.04047393798828, 1.9859466552734375, -17.287368774414062, 73.45476531982422, 53.98504638671875, 62.47833251953125, 5.535379409790039, -10.331924438476562, 4.126873016357422, 3.4732894897460938, 5.169464111328125, 100.77372741699219, 52.385154724121094, 20.251670837402344, 16.07801055908203, 65.15599822998047, 31.59033966064453, 50.35530090332031, 72.1595230102539, 14.120098114013672, 23.35198974609375, 11.08216667175293, -13.736747741699219, 48.55683898925781, -31.4923095703125, 48.89942932128906, 63.05057144165039, 11.181739807128906, 66.17837524414062, 3.2010650634765625, 66.52366638183594, 45.03895568847656, 40.91655731201172, 12.125839233398438, 15.109659194946289, 10.3306884765625, -16.23969268798828, 5.985084533691406, 5.7791290283203125, 11.12240219116211, 54.94969940185547, 14.735031127929688, 23.446758270263672, 46.646514892578125, 61.20806884765625, 3.7528228759765625, 11.378799438476562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000107.npy"}
|
|
{"epoch": 0.15712187958883994, "step": 108, "batch_size": 64, "mean": 29.75396728515625, "std": 40.20465087890625, "min": -36.56855773925781, "p10": -15.4573766708374, "median": 22.806105613708496, "p90": 88.65200042724611, "max": 143.1185302734375, "pos_frac": 0.75, "sample": [27.689353942871094, 96.23283386230469, 32.349639892578125, -28.5113525390625, -36.56855773925781, 71.7237548828125, 25.761308670043945, 6.633903503417969, 143.1185302734375, -3.7176589965820312, -2.623992919921875, 90.65535736083984, 5.836647033691406, 2.2955322265625, -2.2744197845458984, 51.143043518066406, -4.199708938598633, 68.35542297363281, 110.27583312988281, 83.97750091552734, 83.051513671875, 61.56282043457031, -0.3664207458496094, -16.335718154907227, 34.883567810058594, 9.5228271484375, 22.71272087097168, 13.022235870361328, 59.455169677734375, 5.739900588989258, 15.839702606201172, -0.0099334716796875, 59.128929138183594, -13.407913208007812, 127.18173217773438, -19.374786376953125, 61.37574768066406, -21.643604278564453, -9.346778869628906, 4.086969375610352, 24.04108428955078, 7.8218841552734375, 35.46070861816406, 104.08615112304688, 25.316505432128906, 7.31591796875, 64.88147735595703, 52.01513671875, 21.453874588012695, -34.27787780761719, 9.231094360351562, 42.906494140625, 91.60379028320312, 44.043800354003906, 2.8893203735351562, -1.6118946075439453, 36.15977478027344, 14.192407608032227, 57.8016357421875, -36.12855911254883, 22.899490356445312, 39.89104080200195, 40.24517822265625, 16.783782958984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000108.npy"}
|
|
{"epoch": 0.15859030837004406, "step": 109, "batch_size": 64, "mean": 32.8599853515625, "std": 34.17582321166992, "min": -33.98883056640625, "p10": -6.744778060913084, "median": 29.71529197692871, "p90": 79.11140747070313, "max": 128.77032470703125, "pos_frac": 0.828125, "sample": [41.2657470703125, 5.284992218017578, 87.35496520996094, 63.84827423095703, 18.980331420898438, 43.50907516479492, -28.34716796875, 13.506195068359375, 13.682815551757812, 41.4747314453125, 93.86862182617188, 31.575916290283203, 53.23332977294922, 32.37236022949219, 45.75787353515625, 6.402641296386719, 19.192474365234375, 31.51155662536621, -7.689502716064453, -4.5404205322265625, -3.2097549438476562, 128.77032470703125, -0.2080554962158203, -7.74041748046875, 24.55671501159668, 3.4313888549804688, 95.04429626464844, 27.209095001220703, 19.847715377807617, -22.635517120361328, 43.193878173828125, 51.66448974609375, 6.377195358276367, 42.40521240234375, 46.41925048828125, 30.64618492126465, 69.49809265136719, 55.842674255371094, 78.0649642944336, 79.55988311767578, 4.634429931640625, 9.482772827148438, -33.98883056640625, 23.11688232421875, 2.3238525390625, 58.078102111816406, 31.1964111328125, 29.21718978881836, 18.702760696411133, 66.34225463867188, -27.1107177734375, 29.37017822265625, 30.060405731201172, 80.97319030761719, 94.45835876464844, 20.63311767578125, 66.92554473876953, 25.085250854492188, 63.82499694824219, 76.24796295166016, -3.690643310546875, -9.3648681640625, 5.321329116821289, 70.2166976928711], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000109.npy"}
|
|
{"epoch": 0.16005873715124816, "step": 110, "batch_size": 64, "mean": 29.92901611328125, "std": 37.43989181518555, "min": -65.90080261230469, "p10": -3.64809455871582, "median": 23.66917133331299, "p90": 73.65596008300781, "max": 132.37808227539062, "pos_frac": 0.859375, "sample": [71.56845092773438, 34.53541564941406, 22.593990325927734, 43.98158264160156, 42.78667449951172, 29.02179718017578, 37.654720306396484, 12.763629913330078, -65.90080261230469, 33.49076843261719, 48.6492919921875, 14.222663879394531, -30.0640869140625, 14.401878356933594, 13.992473602294922, 0.07699394226074219, 24.11789894104004, 127.9505615234375, 82.57939147949219, 21.700828552246094, 4.183132171630859, -2.9738121032714844, 132.37808227539062, 94.55050659179688, 74.33578491210938, 70.3553466796875, 11.129959106445312, 1.7982635498046875, -3.93707275390625, 51.81110382080078, -24.846450805664062, -12.507415771484375, 7.829891204833984, 16.413890838623047, 5.509954452514648, 115.98980712890625, 39.584075927734375, 68.75430297851562, 72.0697021484375, 2.7229557037353516, 55.369956970214844, -2.9659910202026367, 23.220443725585938, 33.028968811035156, 13.407001495361328, 20.575271606445312, 58.469154357910156, -44.11383056640625, 52.368255615234375, -14.1390380859375, 27.507896423339844, 2.1126976013183594, 39.036712646484375, 24.820514678955078, 41.3803825378418, 31.63384246826172, 18.332082748413086, 21.527664184570312, 15.384899139404297, 37.53522491455078, 50.05931091308594, 1.7809314727783203, 1.9744987487792969, 101.87405395507812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000110.npy"}
|
|
{"epoch": 0.16152716593245228, "step": 111, "batch_size": 64, "mean": 32.667232513427734, "std": 40.49457550048828, "min": -58.318084716796875, "p10": -14.30110473632812, "median": 26.99739170074463, "p90": 84.91012802124024, "max": 147.88262939453125, "pos_frac": 0.78125, "sample": [147.88262939453125, -27.489501953125, -58.318084716796875, 75.40301513671875, -6.2352142333984375, 39.104888916015625, 7.40380859375, 25.087539672851562, 18.646087646484375, 102.06559753417969, 25.737747192382812, 60.94233703613281, 32.94841766357422, 38.85693359375, 22.78247833251953, 66.57707214355469, 85.74300384521484, -11.020111083984375, 72.12979125976562, 8.651796340942383, 55.588783264160156, -15.707244873046875, -10.867313385009766, -2.8751220703125, 46.51164245605469, 82.96675109863281, 1.9230003356933594, 77.27027893066406, -20.5015869140625, 9.267024993896484, -8.908660888671875, 26.738462448120117, 87.69320678710938, 17.379127502441406, 104.29368591308594, 36.970726013183594, 22.879135131835938, 40.454402923583984, 48.98120880126953, 102.13121795654297, 18.892776489257812, -22.152847290039062, 16.0186767578125, 20.97867202758789, 105.54620361328125, 43.04447937011719, 25.567413330078125, 35.74183654785156, 76.69146728515625, 2.280719757080078, 79.12300109863281, 31.14301300048828, 66.21630096435547, 58.46730041503906, 36.303253173828125, 9.092336654663086, -46.35301208496094, 27.25632095336914, 19.802637100219727, 34.04945373535156, -4.4518585205078125, 67.00254821777344, -32.421058654785156, -6.225746154785156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000111.npy"}
|
|
{"epoch": 0.16299559471365638, "step": 112, "batch_size": 64, "mean": 21.566303253173828, "std": 29.46368408203125, "min": -45.55120849609375, "p10": -14.254753112792967, "median": 18.499897956848145, "p90": 67.49326705932619, "max": 90.80036926269531, "pos_frac": 0.765625, "sample": [28.91965103149414, 90.80036926269531, 57.241302490234375, -4.635610580444336, -27.539722442626953, 22.01454734802246, 27.994140625, 19.45635986328125, 14.834403991699219, -6.988494873046875, 77.55838012695312, 62.95465850830078, 1.6677360534667969, -0.7024154663085938, 69.43838500976562, 49.284358978271484, 31.854442596435547, 45.86444854736328, 33.03947830200195, 26.454071044921875, 38.083648681640625, 33.88288879394531, 52.39849853515625, -4.948272705078125, 21.001976013183594, 21.500951766967773, 24.688743591308594, 18.055667877197266, -16.603012084960938, 15.930377960205078, 3.477029800415039, -22.61294937133789, 20.63397979736328, 78.8602523803711, 83.84041595458984, -20.296279907226562, 31.193138122558594, 12.905403137207031, 50.697235107421875, 3.570331573486328, -45.55120849609375, 73.45623016357422, 16.518051147460938, 9.12921142578125, -3.872833251953125, 14.77334213256836, 42.93504333496094, 2.6350231170654297, 14.318260192871094, 72.80572509765625, 13.730491638183594, 6.241970062255859, 35.171905517578125, -20.5755615234375, 7.190523147583008, -14.648521423339844, -13.335960388183594, -3.8826065063476562, 52.92376708984375, 1.2348270416259766, 18.944128036499023, 6.3132171630859375, -4.7599029541015625, 32.77766418457031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000112.npy"}
|
|
{"epoch": 0.1644640234948605, "step": 113, "batch_size": 64, "mean": 24.083799362182617, "std": 27.024009704589844, "min": -21.376007080078125, "p10": -8.81769561767578, "median": 26.15747833251953, "p90": 61.47525978088379, "max": 85.32772064208984, "pos_frac": 0.78125, "sample": [-21.376007080078125, 52.87348937988281, -11.090753555297852, 61.94061279296875, 11.8369140625, 26.177841186523438, 16.973506927490234, 32.92707824707031, 8.346887588500977, -1.1114273071289062, 5.969566345214844, 56.17778015136719, 44.135292053222656, 76.525390625, 27.942995071411133, -13.531198501586914, 1.1305465698242188, -3.7245559692382812, 11.82965087890625, 60.454856872558594, -13.795486450195312, 33.89664077758789, 57.7281494140625, -9.24481201171875, 39.460533142089844, 3.7848739624023438, 7.345987319946289, 85.32772064208984, 32.89293670654297, -21.059280395507812, 11.85092544555664, 62.29603576660156, 27.21352767944336, -7.8210906982421875, 29.586593627929688, 43.65016174316406, 9.588577270507812, 61.56340789794922, 27.815940856933594, -0.2824249267578125, 41.18840789794922, 55.42634582519531, -7.28856086730957, 2.5378856658935547, -20.138275146484375, 36.78098678588867, 26.137115478515625, 31.29261016845703, -7.298439025878906, 9.085016250610352, 79.28445434570312, 2.037059783935547, 28.207138061523438, 35.44934844970703, 69.58368682861328, 61.26958084106445, 29.309158325195312, 52.24104309082031, -3.383180618286133, 9.36669921875, 12.914093017578125, 12.64975357055664, 21.711040496826172, 36.792755126953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000113.npy"}
|
|
{"epoch": 0.16593245227606462, "step": 114, "batch_size": 64, "mean": 36.693424224853516, "std": 45.44995880126953, "min": -23.691909790039062, "p10": -12.279164886474609, "median": 36.21592712402344, "p90": 78.90257186889652, "max": 242.1492919921875, "pos_frac": 0.765625, "sample": [122.8294677734375, 70.71393585205078, 44.00462341308594, 49.91986846923828, 19.82036590576172, 93.41943359375, 13.280401229858398, 35.114463806152344, -12.333984375, 46.359779357910156, -23.56264305114746, 46.88983154296875, 41.694847106933594, 9.275535583496094, -23.691909790039062, 60.31486511230469, 10.213794708251953, -15.34088134765625, 28.637901306152344, 82.4119873046875, 14.845037460327148, 38.270355224609375, -8.800445556640625, 83.85143280029297, 40.148582458496094, 51.30402374267578, 23.80040740966797, 21.56085205078125, 23.702316284179688, 48.92599868774414, 50.72938537597656, -2.001068115234375, -5.109161376953125, 47.425537109375, -12.151252746582031, 11.669790267944336, 32.36915588378906, 45.169952392578125, 46.387908935546875, 67.34336853027344, 47.37582778930664, 65.7122573852539, 3.8146400451660156, 36.27702331542969, 27.942317962646484, -4.30035400390625, 176.7985382080078, 4.536403656005859, 36.46689987182617, 111.32404327392578, -2.162202835083008, 57.594268798828125, -18.639564514160156, -14.091354370117188, 242.1492919921875, 50.661739349365234, -1.840768814086914, 56.3907470703125, 36.15483093261719, -22.59549331665039, 18.171485900878906, 69.01870727539062, -6.616512298583984, 58.822540283203125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000114.npy"}
|
|
{"epoch": 0.16740088105726872, "step": 115, "batch_size": 64, "mean": 26.151029586791992, "std": 44.99739074707031, "min": -92.85504150390625, "p10": -13.134564208984374, "median": 17.17689323425293, "p90": 94.45272293090824, "max": 172.81192016601562, "pos_frac": 0.765625, "sample": [4.559593200683594, -3.3614959716796875, 98.79402160644531, 38.786956787109375, -24.831207275390625, 1.999114990234375, -22.354400634765625, 10.189186096191406, 7.223957061767578, 16.89236831665039, 17.46141815185547, -7.857147216796875, 36.60679244995117, 44.512237548828125, 21.590044021606445, 120.04145812988281, 21.270036697387695, -1.5388031005859375, 172.81192016601562, -8.994379043579102, 51.08814239501953, 3.9509735107421875, 98.0260009765625, 52.11549377441406, 119.16610717773438, -11.622970581054688, 23.922805786132812, 28.496463775634766, 5.222545623779297, 66.35310363769531, 121.70832824707031, 67.72917175292969, 7.40032958984375, -82.64513397216797, 7.280170440673828, 31.20940399169922, 42.115806579589844, 15.24578857421875, 13.455940246582031, 33.25676727294922, 22.9820556640625, 30.281570434570312, 25.118988037109375, 19.72325897216797, 106.41888427734375, -13.314208984375, 15.450546264648438, 10.474624633789062, -4.349510192871094, 8.359268188476562, 16.865097045898438, 57.548431396484375, 0.8649673461914062, 21.944133758544922, -14.583803176879883, 86.11507415771484, 53.40400695800781, -7.438224792480469, 49.90885925292969, -92.85504150390625, 71.12026977539062, -20.91162872314453, 5.976739883422852, -12.71539306640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000115.npy"}
|
|
{"epoch": 0.16886930983847284, "step": 116, "batch_size": 64, "mean": 27.65015983581543, "std": 31.748279571533203, "min": -19.698490142822266, "p10": -6.732036018371581, "median": 22.17237377166748, "p90": 70.9504135131836, "max": 161.19212341308594, "pos_frac": 0.8125, "sample": [7.559019088745117, 1.6665077209472656, -19.698490142822266, -12.369728088378906, 21.651418685913086, 33.125267028808594, 36.67047119140625, 19.476215362548828, 39.23371887207031, 45.666107177734375, 19.02387237548828, -0.6609630584716797, 30.83226776123047, 23.41327667236328, 4.086112976074219, -4.909856796264648, 38.804283142089844, -15.716377258300781, 59.49446105957031, 68.54364013671875, 28.152191162109375, 1.1795883178710938, 17.125816345214844, 10.335685729980469, 36.19903564453125, 19.173057556152344, 45.8211669921875, 78.99253845214844, 19.34156036376953, -18.066978454589844, 86.01287841796875, 30.8509521484375, 18.515762329101562, 28.92730712890625, -7.241327285766602, -15.484573364257812, 12.805057525634766, 12.525177001953125, 4.684942245483398, 73.76698303222656, 4.870136260986328, 84.20487976074219, 22.693328857421875, 60.28245544433594, 51.54304504394531, 19.68062973022461, 48.43280792236328, -11.223930358886719, -5.543689727783203, 57.23302459716797, 24.469345092773438, -5.274988174438477, 161.19212341308594, 47.42481994628906, 14.047698974609375, 35.996559143066406, 43.68513488769531, 42.202022552490234, -3.840024948120117, 11.0625, 1.3761520385742188, 39.30064392089844, 74.3055419921875, 71.98188781738281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000116.npy"}
|
|
{"epoch": 0.17033773861967694, "step": 117, "batch_size": 64, "mean": 24.427927017211914, "std": 34.99918746948242, "min": -60.828125, "p10": -5.438888549804687, "median": 21.476882934570312, "p90": 66.16867980957034, "max": 139.63336181640625, "pos_frac": 0.78125, "sample": [14.266109466552734, 41.11394500732422, 37.90632629394531, -1.9453392028808594, 16.977575302124023, 3.5109176635742188, 20.14349937438965, 23.151138305664062, 15.951568603515625, 29.675046920776367, 37.748714447021484, 50.95458984375, 84.599609375, 100.48320007324219, 21.615196228027344, 30.837295532226562, -33.91648864746094, 57.328704833984375, 26.158584594726562, -59.952056884765625, -16.3116455078125, 7.49688720703125, 14.712028503417969, 46.301204681396484, 38.9857063293457, -0.7766571044921875, -5.672187805175781, 2.780000686645508, 25.210105895996094, -4.894523620605469, 28.842437744140625, 92.00047302246094, -4.214042663574219, 18.007720947265625, 98.2000961303711, 14.170127868652344, 139.63336181640625, 13.422027587890625, 18.635784149169922, 46.11224365234375, 8.435680389404297, 1.9002723693847656, 6.875494003295898, -2.431365966796875, -60.828125, 21.33856964111328, 58.775421142578125, 27.870193481445312, 72.50067138671875, 29.343107223510742, -27.325607299804688, -4.261623382568359, 5.330150604248047, 42.431522369384766, 69.33721923828125, 57.84880065917969, 7.181821823120117, 38.83098602294922, 46.263145446777344, 26.823665618896484, 26.675247192382812, -1.0829925537109375, -21.783157348632812, 44.08891296386719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000117.npy"}
|
|
{"epoch": 0.17180616740088106, "step": 118, "batch_size": 64, "mean": 27.263755798339844, "std": 32.080867767333984, "min": -27.532699584960938, "p10": -4.479463577270508, "median": 19.252948760986328, "p90": 70.97304229736329, "max": 116.22744750976562, "pos_frac": 0.828125, "sample": [-4.787147521972656, 16.3245849609375, 3.929128646850586, -16.701263427734375, 17.423242568969727, 15.152240753173828, 50.82162094116211, 38.040313720703125, 40.398231506347656, 72.58335876464844, 63.33354187011719, 18.3519287109375, 89.176513671875, 24.745162963867188, -3.831348419189453, 38.06718444824219, -4.354587554931641, -3.63531494140625, 24.879520416259766, 116.22744750976562, 81.74856567382812, -5.974821090698242, -27.532699584960938, 4.075721740722656, -26.468338012695312, 17.707170486450195, 3.2002906799316406, 23.544633865356445, 4.329524993896484, 6.7200775146484375, 10.740726470947266, 47.801612854003906, -4.532981872558594, 57.58769226074219, 19.240219116210938, 34.381629943847656, 19.26567840576172, 29.1900634765625, 103.29849243164062, 0.5291900634765625, 18.206153869628906, -10.54824447631836, 1.7068099975585938, 29.629241943359375, 97.89959716796875, 56.941993713378906, 92.85646057128906, 36.93519592285156, 63.11492156982422, 12.258718490600586, 2.4927444458007812, 45.28385925292969, 67.21563720703125, 49.61492919921875, 37.43281555175781, 22.273330688476562, 28.656301498413086, 12.997016906738281, 1.7577533721923828, 22.328575134277344, 0.01282501220703125, 61.93815612792969, 0.7659072875976562, -1.8871536254882812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000118.npy"}
|
|
{"epoch": 0.17327459618208516, "step": 119, "batch_size": 64, "mean": 27.355575561523438, "std": 38.056827545166016, "min": -41.52613830566406, "p10": -18.726582336425775, "median": 23.227401733398438, "p90": 72.085270690918, "max": 143.33758544921875, "pos_frac": 0.75, "sample": [18.108980178833008, 33.099822998046875, 22.357112884521484, 100.21035766601562, 37.6258544921875, -11.693374633789062, -4.63470458984375, 38.4144287109375, -5.534229278564453, 46.20429992675781, -0.4136962890625, -2.0459136962890625, 143.33758544921875, 39.90983581542969, 8.74432373046875, 89.67135620117188, 20.734256744384766, 9.092369079589844, 137.66697692871094, 38.06925964355469, 54.81524658203125, 55.60047912597656, 53.177513122558594, 1.6549568176269531, 22.000938415527344, 30.181854248046875, -23.7677001953125, 11.469398498535156, 43.33290100097656, 42.953887939453125, 4.3888092041015625, -6.010589599609375, 75.50128173828125, 24.09769058227539, 113.19906616210938, -8.408660888671875, 45.93181610107422, 43.91236877441406, 3.047811508178711, 18.301071166992188, -22.261016845703125, 30.131160736083984, -31.302032470703125, 91.90840911865234, 33.571929931640625, 45.09324645996094, -4.7352447509765625, -21.740814208984375, 61.233909606933594, 33.71428298950195, -22.754737854003906, -41.52613830566406, 6.461433410644531, 51.04042053222656, 21.6378173828125, 40.97873306274414, 37.97919464111328, 8.231075286865234, 64.11457824707031, -3.684896469116211, -31.93378448486328, 5.0343780517578125, 27.021514892578125, 8.238288879394531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000119.npy"}
|
|
{"epoch": 0.17474302496328928, "step": 120, "batch_size": 64, "mean": 36.529258728027344, "std": 40.864158630371094, "min": -44.6563720703125, "p10": -11.009513473510735, "median": 31.433149337768555, "p90": 96.37615509033203, "max": 135.44747924804688, "pos_frac": 0.875, "sample": [111.66294860839844, 56.777923583984375, 27.934341430664062, 84.6749267578125, 30.489498138427734, 12.602991104125977, -35.33380126953125, 80.27867126464844, 17.5461368560791, 53.31853485107422, 48.51483917236328, 117.80770874023438, 59.30835723876953, 16.454208374023438, 12.485845565795898, 11.733245849609375, 91.62802124023438, 35.45989990234375, 48.172401428222656, -26.68878936767578, -31.557327270507812, 97.18072509765625, 30.018600463867188, 32.636329650878906, 32.665618896484375, 24.500015258789062, 32.376800537109375, 7.952060699462891, 106.02899169921875, -44.6563720703125, -14.177154541015625, 39.77806091308594, 25.95568084716797, 3.4935302734375, 73.51029968261719, 4.785249710083008, 135.44747924804688, 44.36650466918945, 50.003517150878906, 2.2574634552001953, 5.431148529052734, 1.5150947570800781, 46.71855926513672, 113.2803955078125, 25.91534423828125, 35.466827392578125, 94.49882507324219, 133.77719116210938, 25.532379150390625, 34.780303955078125, 0.48383331298828125, 60.043861389160156, -22.778945922851562, 19.11517333984375, 43.27851104736328, 18.407928466796875, 2.279521942138672, 74.66537475585938, 48.16539001464844, -3.6183509826660156, 57.408180236816406, -14.423843383789062, 27.758865356445312, 2.777130126953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000120.npy"}
|
|
{"epoch": 0.1762114537444934, "step": 121, "batch_size": 64, "mean": 40.623958587646484, "std": 48.14680480957031, "min": -74.8006591796875, "p10": -6.331312561035156, "median": 36.92974090576172, "p90": 110.80593261718754, "max": 153.86972045898438, "pos_frac": 0.8125, "sample": [4.964334487915039, 0.20426177978515625, -0.5322113037109375, 150.51747131347656, -5.227697372436523, 7.526664733886719, 65.79855346679688, -8.182292938232422, 68.16585540771484, 51.48523712158203, 5.58880615234375, 48.719215393066406, 101.06495666503906, 87.08069610595703, 52.413726806640625, -5.878837585449219, 58.798675537109375, 52.2636604309082, 48.29570770263672, -22.542020797729492, 24.903711318969727, 94.44229125976562, -32.450950622558594, 85.50993347167969, 121.61770629882812, 44.22174072265625, 24.287948608398438, 9.247394561767578, 84.96882629394531, 4.549694061279297, 52.072418212890625, -13.685356140136719, 40.89404296875, 51.82014465332031, 3.7104434967041016, 133.98507690429688, 4.239105224609375, 125.49237060546875, 32.155433654785156, 152.0960693359375, 4.785194396972656, 96.5372543334961, -6.525230407714844, 9.539596557617188, -12.70458984375, 23.770103454589844, 63.32992935180664, 13.040931701660156, 4.518043518066406, 39.26177978515625, 6.281742095947266, 13.028377532958984, 51.06849670410156, 89.30364990234375, 114.98063659667969, 60.05988311767578, -74.8006591796875, 41.6373291015625, -5.1721343994140625, 60.55035400390625, 153.86972045898438, 17.72106170654297, -3.3486461639404297, 34.59770202636719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000121.npy"}
|
|
{"epoch": 0.1776798825256975, "step": 122, "batch_size": 64, "mean": 50.65019989013672, "std": 45.004573822021484, "min": -46.6185302734375, "p10": 4.127088737487797, "median": 42.979461669921875, "p90": 107.13658142089844, "max": 162.86000061035156, "pos_frac": 0.921875, "sample": [100.99713134765625, 13.514892578125, 45.25859069824219, 63.26971435546875, 16.550134658813477, 24.634109497070312, 52.51081848144531, 73.39038848876953, 20.76427459716797, 81.38835144042969, 13.065698623657227, 88.58585357666016, 88.64537811279297, 162.27536010742188, 81.09806060791016, 30.905662536621094, 13.73828125, 52.89214324951172, 107.49491882324219, 42.213714599609375, 11.304140090942383, 106.30046081542969, 24.73873519897461, 1.3807907104492188, 46.39360046386719, 20.295562744140625, 43.594627380371094, 102.6556396484375, 58.923988342285156, 52.04596710205078, 58.181678771972656, -16.094947814941406, 129.46551513671875, -4.617280960083008, 41.453102111816406, 104.85479736328125, 51.921485900878906, 30.912071228027344, 104.84847259521484, 116.98686218261719, 8.004796981811523, 45.473663330078125, -22.70093536376953, 162.86000061035156, 22.856483459472656, 33.47819900512695, 24.72486114501953, 24.753772735595703, 30.82347869873047, -46.6185302734375, -4.256614685058594, 22.843338012695312, 18.082422256469727, 64.3122329711914, 151.585205078125, 8.80642318725586, 84.75053405761719, 42.364295959472656, 2.4652137756347656, 9.315017700195312, 116.69245910644531, 56.44703674316406, 20.878154754638672, 104.93260192871094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000122.npy"}
|
|
{"epoch": 0.17914831130690162, "step": 123, "batch_size": 64, "mean": 33.586708068847656, "std": 51.60758972167969, "min": -106.8572006225586, "p10": -16.021846580505372, "median": 23.47213363647461, "p90": 100.56116333007813, "max": 176.22506713867188, "pos_frac": 0.71875, "sample": [51.339012145996094, 61.363006591796875, -9.601402282714844, 7.110939025878906, 54.53938293457031, 51.15771484375, 78.16004943847656, 73.58993530273438, 58.163368225097656, 39.31499481201172, 0.885986328125, -0.8324508666992188, 11.278186798095703, -19.838600158691406, 122.59986877441406, 59.9151611328125, 15.68387222290039, -5.164581298828125, -4.9562835693359375, 43.81022644042969, -2.186492919921875, 99.87466430664062, 42.172454833984375, -53.49678039550781, 22.069366455078125, 65.21505737304688, 176.22506713867188, 17.99941062927246, 100.85537719726562, 110.347412109375, -16.341089248657227, 60.106956481933594, 18.45785140991211, 7.044715881347656, 68.36481475830078, 90.33540344238281, 24.874900817871094, 41.30628967285156, 133.89402770996094, -15.276947021484375, 61.122589111328125, 55.038612365722656, -30.91556167602539, -33.85906219482422, 57.26726531982422, -3.6162109375, -28.532588958740234, -2.888700485229492, -14.128120422363281, 168.90811157226562, -14.526634216308594, 42.60972595214844, 33.387176513671875, 126.74507904052734, 1.314117431640625, 19.92182731628418, 90.69967651367188, 29.91221809387207, -106.8572006225586, 8.410957336425781, 6.486091613769531, -4.343578338623047, 1.0970706939697266, 5.935739517211914], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000123.npy"}
|
|
{"epoch": 0.18061674008810572, "step": 124, "batch_size": 64, "mean": 47.546592712402344, "std": 65.2303237915039, "min": -66.68653869628906, "p10": -17.689897155761717, "median": 31.999963760375977, "p90": 123.68824157714846, "max": 274.8267517089844, "pos_frac": 0.828125, "sample": [99.34542846679688, -15.825469970703125, 35.566192626953125, 28.689712524414062, 11.287559509277344, 110.65046691894531, -66.68653869628906, -7.453037261962891, 99.3077392578125, 143.79373168945312, 2.281951904296875, 25.249225616455078, 4.140485763549805, 47.972190856933594, 46.85120391845703, 184.38717651367188, 107.4697265625, 36.23362350463867, -3.4929351806640625, 49.86421203613281, -63.50071716308594, -27.78204345703125, 101.89453125, 2.8061790466308594, 147.85934448242188, 100.258056640625, 51.82255554199219, 4.277305603027344, 41.236183166503906, 85.71357727050781, -18.488937377929688, 28.992374420166016, 50.20213317871094, 23.647157669067383, 7.4323577880859375, 238.02545166015625, 77.66476440429688, 126.03349304199219, 18.148338317871094, 114.55967712402344, 72.47193145751953, -24.28813934326172, 22.634227752685547, -14.707870483398438, 32.97064971923828, 82.99124145507812, -55.993499755859375, 51.76837158203125, 18.092084884643555, 24.977142333984375, 31.029277801513672, 45.942832946777344, -35.901031494140625, 162.64364624023438, 20.06085205078125, 274.8267517089844, 21.940399169921875, 118.21598815917969, 1.1009502410888672, 25.116111755371094, 40.93994140625, 50.08910369873047, 11.557079315185547, 14.069477081298828], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000124.npy"}
|
|
{"epoch": 0.18208516886930984, "step": 125, "batch_size": 64, "mean": 43.85457229614258, "std": 50.878387451171875, "min": -44.866966247558594, "p10": -10.956294250488279, "median": 26.3569393157959, "p90": 114.35088653564456, "max": 175.77279663085938, "pos_frac": 0.828125, "sample": [-2.0755653381347656, 91.79434204101562, 55.026702880859375, 13.606674194335938, 0.45804405212402344, 9.00772476196289, 77.49339294433594, -12.646629333496094, 108.24296569824219, 18.93256950378418, 4.569347381591797, -21.14986801147461, 22.014902114868164, 11.921829223632812, 75.9034194946289, 69.58895874023438, 55.09600830078125, 45.29009246826172, 10.223823547363281, 9.011932373046875, 94.53402709960938, 15.876312255859375, -21.949111938476562, 9.07171630859375, 70.39311981201172, 175.77279663085938, -27.77337646484375, 27.465635299682617, 66.25753784179688, 20.463546752929688, 6.164360046386719, -44.866966247558594, 27.361587524414062, 13.620780944824219, 24.467300415039062, 146.80235290527344, 3.235553741455078, 146.2445068359375, 90.57376098632812, -7.699798583984375, 9.771064758300781, 84.7908935546875, 123.90496826171875, 73.90397644042969, 149.91220092773438, 66.96482849121094, 50.867759704589844, 34.333213806152344, 116.96856689453125, 45.119625091552734, 12.484375, 89.4901123046875, 28.967819213867188, -4.113286972045898, 81.82707214355469, 24.944503784179688, -7.9582977294921875, 94.95033264160156, 25.352291107177734, -31.4696044921875, 152.5458526611328, 98.68128204345703, -12.24114990234375, 18.367952346801758], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000125.npy"}
|
|
{"epoch": 0.18355359765051396, "step": 126, "batch_size": 64, "mean": 40.362632751464844, "std": 47.943973541259766, "min": -60.36212921142578, "p10": -10.353904724121094, "median": 31.06861114501953, "p90": 120.33871307373049, "max": 160.2379150390625, "pos_frac": 0.859375, "sample": [81.0439453125, 49.90484619140625, 10.813457489013672, 72.11341094970703, 11.873924255371094, 31.941017150878906, -24.61309051513672, 2.209941864013672, 40.584320068359375, 122.93252563476562, -16.879425048828125, 12.429401397705078, 134.365234375, 25.153602600097656, 64.05008697509766, 0.1916961669921875, 96.42700958251953, 31.935882568359375, 85.972900390625, 46.07528305053711, 37.588714599609375, 39.78670883178711, -3.4688949584960938, 26.437084197998047, 46.14484405517578, 125.29229736328125, -10.390228271484375, 114.28648376464844, 29.688854217529297, 38.25234603881836, 17.356021881103516, 19.268314361572266, 44.81217956542969, 12.944732666015625, 125.64498901367188, 38.463714599609375, 49.845489501953125, -48.051246643066406, 89.41958618164062, 2.8701133728027344, -25.811492919921875, -10.269149780273438, 160.2379150390625, 4.305654525756836, 19.71405601501465, 8.123523712158203, 4.201862335205078, 15.070270538330078, 90.39178466796875, 78.05085754394531, -60.36212921142578, 151.84237670898438, 59.24651336669922, 17.904708862304688, 24.945602416992188, 63.93119812011719, 30.201339721679688, 54.00172424316406, 16.064727783203125, -23.15496063232422, 9.487815856933594, 72.17388916015625, 11.334266662597656, 136.85824584960938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000126.npy"}
|
|
{"epoch": 0.18502202643171806, "step": 127, "batch_size": 64, "mean": 31.902957916259766, "std": 41.71892166137695, "min": -62.86869812011719, "p10": -19.415142822265626, "median": 32.408119201660156, "p90": 83.10592346191407, "max": 124.76454162597656, "pos_frac": 0.78125, "sample": [42.8092155456543, 61.384071350097656, -3.2246170043945312, 37.297203063964844, 3.9119796752929688, -50.816680908203125, 111.41856384277344, 11.391311645507812, 81.10877227783203, -62.790008544921875, 8.067892074584961, 42.17249298095703, 60.58821105957031, -23.850515365600586, 18.05460548400879, 27.54339027404785, 63.840858459472656, 10.740312576293945, -62.86869812011719, 60.417022705078125, 23.542144775390625, 60.1180419921875, 90.2510986328125, 18.508399963378906, 54.7288703918457, 116.93061065673828, 23.630535125732422, 56.21173095703125, -7.3518218994140625, 20.377967834472656, 101.15930938720703, 42.645606994628906, 77.10882568359375, 18.92969512939453, 31.3662109375, 105.123779296875, -2.063558578491211, -11.04095458984375, 22.31264877319336, -1.338623046875, 13.278011322021484, 83.96184539794922, 7.597772598266602, 48.88291931152344, 34.43492126464844, 51.55992889404297, 38.695098876953125, 78.1009521484375, 73.96680450439453, 44.615325927734375, -31.562973022460938, 34.69725036621094, 17.56307601928711, 124.76454162597656, -40.94007873535156, 39.603233337402344, 24.53771209716797, 80.02497100830078, 38.72991943359375, 9.789676666259766, 33.45002746582031, -3.7349319458007812, -18.964248657226562, -19.608383178710938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000127.npy"}
|
|
{"epoch": 0.18649045521292218, "step": 128, "batch_size": 64, "mean": 32.72411346435547, "std": 50.408626556396484, "min": -116.041748046875, "p10": -20.254205513000485, "median": 26.569278717041016, "p90": 107.21317977905275, "max": 182.1339874267578, "pos_frac": 0.75, "sample": [32.502445220947266, 33.71904373168945, 100.103759765625, -14.26080322265625, 78.16751098632812, -28.380863189697266, 84.8551025390625, 111.19312286376953, 20.86357879638672, 28.40180206298828, 115.29306030273438, 35.993431091308594, 39.96124267578125, 25.734825134277344, -42.550331115722656, 79.51287841796875, 95.99052429199219, 5.378273010253906, 3.463380813598633, 61.124210357666016, -47.711265563964844, -3.0101699829101562, 182.1339874267578, -116.041748046875, 109.0504150390625, 27.403732299804688, 89.65008544921875, 108.90584564208984, 33.75817108154297, 40.23944091796875, -5.105525970458984, -27.529685974121094, 85.21548461914062, 4.032167434692383, -6.311077117919922, 19.15875244140625, -43.51043701171875, 1.3670616149902344, 25.6072998046875, 41.6729850769043, 55.915428161621094, -22.127395629882812, 49.512451171875, -15.883428573608398, -3.109100341796875, 109.85630798339844, 18.750843048095703, 37.23701477050781, 19.318157196044922, 35.33439254760742, 17.932273864746094, -0.6492443084716797, 64.14434814453125, -1.2435169219970703, 14.036748886108398, 6.13232421875, 3.9027175903320312, 46.88780975341797, 67.20703125, 6.143177032470703, -14.867462158203125, 104.36773681640625, 108.43265533447266, 1.0703411102294922], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000128.npy"}
|
|
{"epoch": 0.18795888399412627, "step": 129, "batch_size": 64, "mean": 34.58758544921875, "std": 55.88490676879883, "min": -92.53314208984375, "p10": -19.492017364501955, "median": 26.867488861083984, "p90": 109.41941299438477, "max": 234.23797607421875, "pos_frac": 0.671875, "sample": [-2.0083141326904297, 28.189849853515625, -28.978694915771484, 82.74485778808594, -26.215009689331055, 69.35525512695312, 11.861770629882812, -17.8712158203125, 124.0074462890625, 25.545127868652344, 9.28546142578125, 73.91267395019531, 22.181163787841797, 33.0908203125, 41.815364837646484, 53.87199401855469, 108.83782958984375, -12.623092651367188, -5.612945556640625, 52.183326721191406, 143.931884765625, 70.23472595214844, 139.46942138671875, -58.833885192871094, 49.26515197753906, -51.0252685546875, -15.569252014160156, 33.296485900878906, 234.23797607421875, 109.66866302490234, 25.440982818603516, 17.312255859375, -7.45355224609375, 58.47197723388672, -5.5484771728515625, -1.3111591339111328, 15.6558837890625, 89.60964965820312, 16.73102569580078, -11.371490478515625, -41.034820556640625, 2.538820266723633, 31.27923583984375, 64.5738525390625, 125.37297058105469, -92.53314208984375, 14.942625045776367, 16.365812301635742, -5.360170364379883, -4.770206451416016, 93.26448059082031, 127.94648742675781, 34.87157440185547, 37.60265350341797, 56.49854278564453, -11.360336303710938, 38.089439392089844, 40.72223663330078, -19.1348876953125, 61.80389404296875, 95.87754821777344, -19.64507293701172, 69.93921661376953, -0.03191947937011719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000129.npy"}
|
|
{"epoch": 0.1894273127753304, "step": 130, "batch_size": 64, "mean": 51.21843338012695, "std": 58.9157829284668, "min": -81.62429809570312, "p10": -9.964857482910157, "median": 37.27092361450195, "p90": 135.87802429199218, "max": 216.26663208007812, "pos_frac": 0.796875, "sample": [56.689640045166016, 39.22898864746094, 20.247543334960938, 141.60427856445312, -10.052665710449219, 18.491302490234375, 129.46420288085938, 73.73055267333984, -21.13800811767578, -2.2355175018310547, 15.25892448425293, 37.758548736572266, 30.21828842163086, 119.995849609375, 158.34927368164062, 14.75640869140625, 75.909423828125, -6.0221405029296875, 32.42732238769531, 61.377357482910156, -0.7361412048339844, 23.816497802734375, 1.8105087280273438, 66.02824401855469, 77.6822509765625, 10.077369689941406, 34.05070495605469, 77.60052490234375, 47.74248504638672, 135.191162109375, 111.38963317871094, 145.8121337890625, -18.601852416992188, 18.447891235351562, 102.94061279296875, -81.62429809570312, 28.324466705322266, 23.584312438964844, 26.296749114990234, -15.08148193359375, 52.19532775878906, 42.15280532836914, 62.84892272949219, 174.22752380371094, 125.98375701904297, -7.093767166137695, 216.26663208007812, -21.17620086669922, -3.34625244140625, 63.70598602294922, 33.820533752441406, 34.92978286743164, 141.21926879882812, -9.759971618652344, 112.02134704589844, -77.24066925048828, 100.52281188964844, 3.5576934814453125, 73.75619506835938, 99.24557495117188, 136.17239379882812, 52.946128845214844, 36.78329849243164, 33.42932891845703], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000130.npy"}
|
|
{"epoch": 0.19089574155653452, "step": 131, "batch_size": 64, "mean": 49.05486297607422, "std": 49.355010986328125, "min": -79.37567138671875, "p10": 2.718857002258301, "median": 37.21183395385742, "p90": 110.7612777709961, "max": 170.93478393554688, "pos_frac": 0.921875, "sample": [42.4655876159668, 35.59346008300781, 40.47368621826172, 27.96194839477539, 170.93478393554688, -79.37567138671875, 32.77595138549805, 25.96045684814453, 33.310035705566406, 71.12030792236328, 92.95529174804688, 18.588668823242188, 31.488357543945312, 10.63187026977539, 61.35484313964844, 26.31781768798828, 29.992393493652344, -2.0176429748535156, 96.55230712890625, 50.862491607666016, 37.07196807861328, 114.20303344726562, 0.8199920654296875, -8.509668350219727, 37.35169982910156, 24.719871520996094, 72.57905578613281, 7.632301330566406, 135.13113403320312, 33.533905029296875, 33.934776306152344, -51.8218994140625, 107.10315704345703, 6.337764739990234, 163.03802490234375, 2.690969467163086, 61.121437072753906, 17.506351470947266, 108.36119079589844, 39.5527229309082, 87.36943054199219, 26.875778198242188, 150.1861572265625, 22.554080963134766, -64.09019470214844, 87.09918212890625, 9.471168518066406, 85.11966705322266, 4.20744514465332, 26.08745574951172, 2.7839279174804688, 89.1888427734375, 13.199607849121094, 134.22828674316406, 98.98786926269531, 111.78988647460938, 66.75384521484375, 51.395416259765625, 24.02552032470703, 63.76322937011719, 78.4978256225586, 81.71016693115234, 83.49568176269531, 44.50599670410156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000131.npy"}
|
|
{"epoch": 0.19236417033773862, "step": 132, "batch_size": 64, "mean": 42.35327911376953, "std": 58.60791015625, "min": -93.38394165039062, "p10": -6.129800224304199, "median": 36.67895698547363, "p90": 104.82260055541992, "max": 249.4310302734375, "pos_frac": 0.84375, "sample": [13.679574966430664, 10.845966339111328, 113.43705749511719, -40.47967529296875, 47.23905944824219, 36.53062438964844, 53.21212387084961, 209.10540771484375, 211.40423583984375, 7.1391448974609375, 44.919189453125, 76.74578094482422, 10.925920486450195, 40.22486877441406, -93.38394165039062, 13.14453125, 135.15679931640625, -18.31366729736328, 42.949195861816406, -6.41203498840332, 3.997098922729492, -5.47125244140625, 87.20246124267578, 62.22175598144531, 11.725128173828125, 10.005653381347656, 4.384941101074219, 78.41759490966797, 3.47454833984375, 8.327011108398438, -0.6985244750976562, 6.188232421875, 15.146150588989258, 69.69571685791016, 69.1832275390625, 104.82506561279297, 31.94959259033203, 44.04682922363281, 62.64176559448242, 44.81022262573242, 37.78767395019531, 33.21236801147461, 36.82728958129883, 104.81684875488281, -62.72602844238281, 37.55460739135742, 93.95502471923828, -31.04802703857422, 9.847896575927734, 73.61695861816406, -13.295053482055664, 49.737937927246094, 87.37068176269531, 10.08404541015625, 41.43247985839844, -0.403900146484375, 20.80652618408203, 76.63070678710938, 24.83245849609375, 11.933486938476562, 138.47003173828125, 18.15521240234375, 41.43614196777344, 249.4310302734375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000132.npy"}
|
|
{"epoch": 0.19383259911894274, "step": 133, "batch_size": 64, "mean": 37.795230865478516, "std": 53.30853271484375, "min": -61.9285888671875, "p10": -9.979099273681639, "median": 27.690040588378906, "p90": 102.34151916503909, "max": 189.98309326171875, "pos_frac": 0.796875, "sample": [15.614486694335938, 65.96232604980469, 3.6501922607421875, 72.96139526367188, 10.137496948242188, -19.704994201660156, -0.001251220703125, 20.723419189453125, -49.427459716796875, 168.18775939941406, 39.74322509765625, 60.807411193847656, -1.8689346313476562, 89.4514389038086, -48.717864990234375, 8.822311401367188, 54.50677490234375, -4.01739501953125, 152.71388244628906, 9.101198196411133, 62.177734375, 44.19709777832031, -8.309822082519531, 97.24368286132812, 2.4856948852539062, 80.8128662109375, 36.486663818359375, 60.432437896728516, 62.698341369628906, 34.201202392578125, 15.966156005859375, 14.15500259399414, 19.870298385620117, 180.75469970703125, 2.3222808837890625, 45.79570770263672, 52.06659698486328, -34.77626037597656, 149.90447998046875, 27.420875549316406, 41.88130187988281, 27.959205627441406, 36.45390319824219, 189.98309326171875, -61.9285888671875, -5.403839111328125, 53.85017395019531, 130.68020629882812, 7.532318115234375, 35.76424026489258, 7.246543884277344, 80.85771179199219, -34.370216369628906, 104.52630615234375, 68.09600067138672, -10.694503784179688, 65.63973999023438, 36.97100067138672, 4.974769592285156, 18.002986907958984, 6.398773193359375, -0.041736602783203125, 8.081380844116211, 11.882835388183594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000133.npy"}
|
|
{"epoch": 0.19530102790014683, "step": 134, "batch_size": 64, "mean": 41.66035079956055, "std": 47.5550651550293, "min": -29.599281311035156, "p10": -11.820767211914061, "median": 31.112594604492188, "p90": 105.71014633178712, "max": 220.076171875, "pos_frac": 0.828125, "sample": [10.33074951171875, -29.599281311035156, -14.434013366699219, 98.09913635253906, 156.97454833984375, 24.036029815673828, 22.939517974853516, 27.57193374633789, -6.323890686035156, 36.96375274658203, 88.18418884277344, 17.430049896240234, 10.206718444824219, 7.490318298339844, 44.744117736816406, 31.171035766601562, -15.292886734008789, 33.058868408203125, 59.29344940185547, 12.772140502929688, 5.599103927612305, 7.772422790527344, 64.44420623779297, 26.74811553955078, 134.64044189453125, 27.16241455078125, 81.27178955078125, 83.31829833984375, 39.4178581237793, 14.470596313476562, 31.054153442382812, 105.10543823242188, 4.942678451538086, 33.240787506103516, 114.01220703125, 94.7009048461914, 105.96930694580078, -16.707101821899414, 64.61160278320312, -2.7051620483398438, 32.5150146484375, -11.518341064453125, 43.15310287475586, 5.473077774047852, 220.076171875, 18.078205108642578, 67.51585388183594, 31.593551635742188, 54.195892333984375, -1.3740482330322266, 44.12672424316406, -11.95037841796875, -18.202560424804688, 137.3861083984375, 59.204689025878906, 20.56018829345703, 49.7503662109375, 28.14126968383789, 30.575836181640625, 57.08177947998047, -22.3619384765625, 59.40595626831055, 22.505008697509766, 115.64425659179688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000134.npy"}
|
|
{"epoch": 0.19676945668135096, "step": 135, "batch_size": 64, "mean": 53.88473892211914, "std": 63.78515625, "min": -58.85286331176758, "p10": -15.700802230834956, "median": 37.02612495422363, "p90": 149.14140319824222, "max": 245.80218505859375, "pos_frac": 0.828125, "sample": [29.225685119628906, 83.3973388671875, 22.693477630615234, 158.03427124023438, 73.05657958984375, 57.67363739013672, 185.9994354248047, 35.727352142333984, 172.007568359375, 152.7064666748047, 105.28909301757812, 42.49092102050781, 136.2611083984375, 100.02496337890625, 189.861083984375, 85.3514404296875, 27.421974182128906, 134.63223266601562, 11.099601745605469, -17.80473518371582, 4.2722625732421875, 95.38125610351562, 106.04647827148438, 65.07194519042969, 27.245481491088867, 80.31948852539062, 11.152389526367188, 245.80218505859375, -17.382205963134766, 87.03196716308594, -32.53074645996094, 150.67550659179688, 145.56182861328125, -58.85286331176758, 30.064510345458984, -9.303287506103516, 23.22955894470215, 45.244873046875, 78.55435180664062, 3.4416351318359375, -40.380210876464844, 26.625213623046875, 64.16429138183594, 30.850074768066406, 4.728115081787109, 39.967018127441406, 11.089771270751953, -5.850372314453125, 38.32489776611328, 15.11782455444336, 23.628475189208984, 72.95045471191406, 68.3668441772461, 63.31275939941406, 4.0097808837890625, 66.57785034179688, 34.35675048828125, 135.07901000976562, 14.2886962890625, -52.86747741699219, -10.203887939453125, -11.77752685546875, -18.500381469726562, 8.589164733886719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000135.npy"}
|
|
{"epoch": 0.19823788546255505, "step": 136, "batch_size": 64, "mean": 37.198089599609375, "std": 62.01396560668945, "min": -44.68341064453125, "p10": -27.820067596435546, "median": 20.154842376708984, "p90": 93.66239624023439, "max": 281.50225830078125, "pos_frac": 0.734375, "sample": [55.76215744018555, 44.10832214355469, 66.96319580078125, 68.30899047851562, -1.7955093383789062, -33.56247329711914, 213.3014373779297, 100.13565063476562, -16.956695556640625, 47.599246978759766, 72.29413604736328, 47.319889068603516, 11.630584716796875, 11.438056945800781, 62.63514709472656, 36.91908264160156, 57.718536376953125, 23.204574584960938, 92.25553894042969, 213.65396118164062, -5.0251922607421875, 64.31694030761719, 281.50225830078125, 11.419181823730469, 80.42202758789062, -0.4309234619140625, 153.80914306640625, 3.2552833557128906, 166.020751953125, 16.729331970214844, 13.210823059082031, 6.621707916259766, 47.77240753173828, -44.68341064453125, 31.162399291992188, -28.494003295898438, 17.10511016845703, -12.947113037109375, 57.129417419433594, 62.13426208496094, 14.506052017211914, -13.033153533935547, -9.68170166015625, 52.394622802734375, 1.108682632446289, 1.206552505493164, -9.280101776123047, 31.565860748291016, 70.1947021484375, 83.16613006591797, -26.24755096435547, -28.768516540527344, 30.757415771484375, 44.666114807128906, -31.493865966796875, 6.612213134765625, 34.24934387207031, -11.61959457397461, -33.804927825927734, 14.621938705444336, 4.546459197998047, 94.26533508300781, 6.631187438964844, -39.84979248046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000136.npy"}
|
|
{"epoch": 0.19970631424375918, "step": 137, "batch_size": 64, "mean": 57.65531921386719, "std": 78.2221450805664, "min": -113.58782958984375, "p10": -12.307447624206542, "median": 34.73969650268555, "p90": 175.4115570068361, "max": 280.78289794921875, "pos_frac": 0.84375, "sample": [-5.223388671875, 17.13900375366211, 32.294525146484375, 85.42241668701172, 113.44129180908203, 20.014427185058594, 32.69233703613281, 60.07087707519531, -12.634395599365234, 88.8448486328125, 4.971782684326172, 77.96563720703125, 280.78289794921875, 13.183490753173828, 31.229843139648438, -68.989501953125, 81.21841430664062, 16.664905548095703, 27.807336807250977, 122.10287475585938, 133.2637939453125, 60.716041564941406, 3.5113964080810547, 253.7716064453125, 82.452880859375, 229.3591766357422, 135.824951171875, 73.65653991699219, 13.350334167480469, 137.61807250976562, -58.84172058105469, 21.3341064453125, 54.76288604736328, 11.26043701171875, 106.28076171875, 5.003440856933594, 104.67396545410156, 47.59349822998047, -1.1717910766601562, 36.76319122314453, 50.7064208984375, 32.71620178222656, 12.182052612304688, -113.58782958984375, -28.381683349609375, -44.41582489013672, 191.6087646484375, -13.253894805908203, 29.597679138183594, 243.15289306640625, 39.253387451171875, 92.43017578125, 4.453716278076172, 78.01844024658203, -11.54456901550293, 14.836431503295898, 38.53327941894531, 7.343624114990234, 2.519449234008789, 78.86254119873047, 216.0696563720703, 5.8024139404296875, 97.37323760986328, 195.48068237304688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000137.npy"}
|
|
{"epoch": 0.2011747430249633, "step": 138, "batch_size": 64, "mean": 43.667884826660156, "std": 51.51856231689453, "min": -85.04052734375, "p10": -7.946293640136718, "median": 39.095632553100586, "p90": 105.1514877319336, "max": 200.60647583007812, "pos_frac": 0.828125, "sample": [118.50894165039062, -8.38446044921875, 62.97629928588867, 103.68597412109375, 16.03829574584961, 76.05501556396484, -4.1968536376953125, -1.0650177001953125, 1.8401870727539062, 116.80906677246094, 27.070056915283203, 17.69000244140625, 44.60200500488281, 9.309370040893555, 21.73407554626465, -85.04052734375, 14.455467224121094, -14.139713287353516, 0.6907958984375, 0.8315353393554688, 134.71044921875, 130.94412231445312, 11.335979461669922, 5.874885559082031, 100.4830322265625, 100.38713073730469, 74.01464080810547, 5.664131164550781, 72.8062515258789, 82.98200988769531, 47.877593994140625, 19.572185516357422, 88.35476684570312, 12.149124145507812, 105.3105697631836, 101.81022644042969, 44.847076416015625, 41.61359786987305, -50.59001922607422, 31.72209930419922, 51.35584259033203, 24.766334533691406, 45.753692626953125, 4.10455322265625, -1.4453659057617188, -33.632835388183594, 81.2462387084961, 104.7802963256836, 15.656902313232422, 36.577667236328125, 83.06044006347656, 26.21697425842285, 97.26547241210938, -31.81311798095703, -23.229108810424805, 48.06290817260742, -6.9239044189453125, 48.553863525390625, 12.454643249511719, 200.60647583007812, 81.68506622314453, 45.561126708984375, 82.12833404541016, 120.61181640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000138.npy"}
|
|
{"epoch": 0.2026431718061674, "step": 139, "batch_size": 64, "mean": 54.313804626464844, "std": 60.00843811035156, "min": -62.979888916015625, "p10": -14.653192138671875, "median": 44.75266075134277, "p90": 115.0948097229004, "max": 255.01480102539062, "pos_frac": 0.84375, "sample": [-18.86811065673828, 105.5107192993164, 80.92823028564453, 123.78712463378906, -7.400886535644531, 56.436004638671875, 97.6020736694336, -3.384784698486328, 54.59928512573242, 112.73535919189453, -62.979888916015625, 43.17823791503906, -27.157917022705078, 193.14306640625, 110.3518295288086, 41.92760467529297, 76.79566955566406, 43.1888427734375, 29.706024169921875, 46.31647872924805, 110.54755401611328, 55.89435577392578, -39.99707794189453, 106.1050033569336, 35.44569396972656, 1.1570987701416016, 116.10600280761719, 64.45637512207031, 10.16920280456543, 37.020484924316406, 109.99859619140625, 12.935152053833008, 84.06576538085938, 77.11937713623047, 38.820884704589844, 62.20660400390625, 3.354095458984375, 0.06637001037597656, 4.662090301513672, 11.732269287109375, 48.97467803955078, 1.2013320922851562, 11.37796401977539, 111.99813842773438, 22.4764404296875, 177.99771118164062, 84.45503997802734, -16.349014282226562, 50.40027618408203, -14.8375244140625, 65.74848175048828, -14.22308349609375, 26.384336471557617, 66.08914184570312, 255.01480102539062, 34.19276809692383, 198.9528350830078, 27.421615600585938, 47.68921661376953, -19.820236206054688, 136.7776641845703, 27.694122314453125, 36.88016128540039, 111.30577087402344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000139.npy"}
|
|
{"epoch": 0.20411160058737152, "step": 140, "batch_size": 64, "mean": 50.007423400878906, "std": 65.07683563232422, "min": -70.89320373535156, "p10": -23.295787811279297, "median": 44.9897575378418, "p90": 131.59839172363283, "max": 251.22198486328125, "pos_frac": 0.75, "sample": [-13.61715316772461, 102.82363891601562, 251.22198486328125, -44.94188690185547, -4.514305114746094, 94.05009460449219, 72.49632263183594, 130.34628295898438, 51.39764404296875, 127.15229797363281, 86.88129425048828, 50.662010192871094, 93.33711242675781, 29.302410125732422, 20.29638671875, 58.863441467285156, 208.50613403320312, 8.77663803100586, -0.6342277526855469, 53.56627655029297, 16.608016967773438, 56.63023376464844, 73.43119812011719, 88.7074966430664, 0.7677040100097656, -25.458236694335938, 183.48878479003906, -23.598312377929688, 29.791793823242188, 69.86427307128906, 21.083524703979492, 25.101524353027344, 8.944280624389648, 86.93707275390625, -16.546737670898438, 52.22279357910156, 78.09144592285156, 37.504539489746094, 1.2022705078125, 117.2841567993164, 106.4237289428711, -22.58989715576172, 132.135009765625, -69.00471496582031, -5.7394561767578125, 52.802894592285156, 38.15594482421875, 71.97909545898438, 145.14968872070312, 15.0369873046875, -10.783853530883789, 123.85899353027344, -10.549217224121094, 25.702537536621094, -70.89320373535156, -19.225849151611328, -41.22045135498047, 32.42474365234375, 151.0369415283203, 132.47637939453125, -29.406272888183594, 56.419212341308594, 39.3175048828125, 98.93804168701172], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000140.npy"}
|
|
{"epoch": 0.2055800293685756, "step": 141, "batch_size": 64, "mean": 48.09896469116211, "std": 62.61170196533203, "min": -152.97146606445312, "p10": -17.287796401977538, "median": 43.67482376098633, "p90": 130.09526519775392, "max": 186.16073608398438, "pos_frac": 0.828125, "sample": [77.46420288085938, 31.020034790039062, 1.0921516418457031, 83.11209106445312, 92.68244934082031, -16.271556854248047, 64.93017578125, 45.631385803222656, 8.277046203613281, -3.744476318359375, 53.28969955444336, 163.99993896484375, -152.97146606445312, 36.360748291015625, -5.169807434082031, 38.84092712402344, 1.197103500366211, -25.0377197265625, 177.0116729736328, -47.969696044921875, 10.043510437011719, 130.4991912841797, 41.10263442993164, 68.21817016601562, 118.1748046875, 62.98565673828125, 69.58366394042969, 9.572685241699219, 4.119354248046875, 57.222835540771484, 92.91452026367188, 42.36599349975586, 22.753564834594727, 124.75117492675781, 186.16073608398438, 29.779878616333008, 82.33512878417969, 19.591615676879883, -20.0500545501709, 16.374427795410156, -11.980743408203125, 43.085296630859375, -17.72332763671875, 148.99038696289062, 10.816410064697266, 106.34908294677734, 132.15634155273438, 60.20783996582031, 11.267389297485352, -85.76437377929688, 18.155609130859375, 77.74054718017578, 44.26435089111328, 52.293190002441406, 74.25115966796875, 99.24254608154297, 185.9039306640625, 93.54607391357422, 1.1581554412841797, 129.15277099609375, 51.27622985839844, 72.25970458984375, -30.1715087890625, 19.61236572265625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000141.npy"}
|
|
{"epoch": 0.20704845814977973, "step": 142, "batch_size": 64, "mean": 55.20783233642578, "std": 76.08611297607422, "min": -142.8724365234375, "p10": -19.112730026245114, "median": 33.68441200256348, "p90": 173.36289825439454, "max": 205.86346435546875, "pos_frac": 0.765625, "sample": [-10.907119750976562, 169.47232055664062, 15.385210037231445, 30.69922637939453, -13.538536071777344, 175.03028869628906, -19.925518035888672, -15.072072982788086, -54.677703857421875, 28.892559051513672, 89.25458526611328, 91.72283935546875, 126.40673065185547, 24.146299362182617, 38.29180908203125, 156.6289520263672, 8.570781707763672, -142.8724365234375, -67.02812194824219, 53.27831268310547, 34.7286376953125, 35.166969299316406, 16.633224487304688, 205.86346435546875, 130.0956268310547, 33.56534194946289, 107.02978515625, 26.18886375427246, 103.80496978759766, -16.354808807373047, 68.386962890625, 100.36782836914062, 141.8653106689453, 69.325439453125, -1.5377311706542969, 155.79351806640625, 175.5381622314453, 19.71517562866211, 14.155988693237305, 26.770431518554688, -33.161590576171875, -32.697566986083984, 6.555591583251953, 184.01817321777344, 51.48149108886719, 179.3765106201172, 12.422218322753906, 4.610694885253906, 47.28472900390625, -17.216224670410156, 50.51688766479492, -9.277572631835938, 19.26500129699707, -55.169769287109375, 162.7437286376953, 203.6805419921875, 29.71506118774414, 158.43353271484375, 14.662010192871094, -0.9793338775634766, 101.45417022705078, 187.46218872070312, 33.80348205566406, 103.45572662353516], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000142.npy"}
|
|
{"epoch": 0.20851688693098386, "step": 143, "batch_size": 64, "mean": 35.0554313659668, "std": 65.20579528808594, "min": -106.28990173339844, "p10": -37.219903564453126, "median": 25.504597663879395, "p90": 110.6291275024414, "max": 251.18142700195312, "pos_frac": 0.65625, "sample": [63.141082763671875, -1.7510528564453125, 44.696632385253906, 29.234268188476562, 185.36868286132812, 53.80786895751953, -38.046409606933594, 80.46170043945312, 89.53898620605469, 21.273916244506836, 96.48519897460938, -71.06149291992188, -39.10553741455078, 91.96977233886719, 68.33084106445312, -79.15863800048828, -32.84032440185547, -0.9060554504394531, 79.39743041992188, -9.244461059570312, 22.224258422851562, 170.43185424804688, -46.020896911621094, 78.55531311035156, 20.348716735839844, -13.755317687988281, 45.251930236816406, 110.38933563232422, 25.904884338378906, 20.708087921142578, -11.995330810546875, -35.29138946533203, -24.674301147460938, 7.970283508300781, 93.77772521972656, 18.056106567382812, 29.180938720703125, 48.21894836425781, 22.797393798828125, 251.18142700195312, -12.771003723144531, -7.417669296264648, 33.1861572265625, 77.2938461303711, -0.608245849609375, 53.41197967529297, 110.73189544677734, 115.1699447631836, -5.347536087036133, -13.68603515625, -47.54608154296875, 187.1068115234375, -17.330543518066406, 59.679039001464844, 46.34193420410156, 25.104310989379883, 78.15394592285156, 28.762489318847656, 128.36123657226562, -106.28990173339844, 0.12610626220703125, 31.487077713012695, -2.02703857421875, 16.80242919921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000143.npy"}
|
|
{"epoch": 0.20998531571218795, "step": 144, "batch_size": 64, "mean": 45.4619255065918, "std": 54.7241096496582, "min": -75.53864288330078, "p10": -17.917892837524413, "median": 37.475013732910156, "p90": 114.76944427490234, "max": 184.49191284179688, "pos_frac": 0.796875, "sample": [-18.595260620117188, 81.69962310791016, 6.000749588012695, 127.15902709960938, -75.53864288330078, -3.5089187622070312, 60.5435791015625, 95.4587631225586, 37.047325134277344, 24.551036834716797, 58.46839141845703, -16.33736801147461, 51.248626708984375, 69.24214172363281, 52.66778564453125, 113.55510711669922, 45.022377014160156, 124.0796127319336, 89.08948516845703, 110.92589569091797, 106.74130249023438, 140.98068237304688, -14.192899703979492, -53.301361083984375, 79.99422454833984, -6.0198822021484375, 44.915931701660156, 7.154563903808594, 63.40027618408203, 16.28626823425293, -30.41990852355957, 72.49179077148438, 53.82862091064453, 24.162818908691406, -26.14594268798828, 114.71743774414062, 37.90270233154297, 21.79421615600586, 21.925016403198242, 67.7930679321289, 27.102005004882812, 50.734825134277344, 184.49191284179688, 32.066001892089844, 23.40241241455078, 63.96525573730469, -26.787944793701172, 23.037696838378906, 29.196029663085938, 87.3098373413086, 91.66698455810547, 174.0635528564453, 2.7205047607421875, -7.500152587890625, -2.0325469970703125, 0.9844741821289062, 169.91648864746094, 16.422344207763672, 20.984663009643555, 15.596076965332031, 41.51451110839844, 30.974960327148438, 114.79173278808594, -31.84658432006836], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000144.npy"}
|
|
{"epoch": 0.21145374449339208, "step": 145, "batch_size": 64, "mean": 47.55194854736328, "std": 58.76460266113281, "min": -69.14280700683594, "p10": -19.099438858032222, "median": 37.865631103515625, "p90": 117.08267669677734, "max": 216.6170654296875, "pos_frac": 0.78125, "sample": [20.530731201171875, -15.76959228515625, -51.478363037109375, 216.6170654296875, 113.47785949707031, 94.78783416748047, 160.3695068359375, 4.144763946533203, -69.14280700683594, 48.301326751708984, 8.742828369140625, 48.05448913574219, 112.12832641601562, -27.853923797607422, 13.85238265991211, 88.1717300415039, 54.17566680908203, 32.10503387451172, 78.67835998535156, -30.8881893157959, -6.05726432800293, 29.51849365234375, 33.997276306152344, 116.97806549072266, 71.46965026855469, -50.85893249511719, 10.487167358398438, 14.03091049194336, -10.659255981445312, 132.79336547851562, 26.648595809936523, 14.427337646484375, 91.29690551757812, -20.52651596069336, -31.179107666015625, 91.59019470214844, 9.4423828125, 15.908815383911133, 59.874725341796875, -1.1268463134765625, 56.81862258911133, 3.894023895263672, 109.93749237060547, -8.81698989868164, 94.32826232910156, 143.62216186523438, 76.22117614746094, -14.377525329589844, 39.95852279663086, 117.12751007080078, 20.600650787353516, 35.77273941040039, 105.76765441894531, 122.34972381591797, 55.554866790771484, 15.999113082885742, -0.41355133056640625, 51.824974060058594, 79.51081848144531, 95.19164276123047, 178.93653869628906, 47.40531539916992, 11.507911682128906, 107.5421142578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000145.npy"}
|
|
{"epoch": 0.21292217327459617, "step": 146, "batch_size": 64, "mean": 37.79475021362305, "std": 63.57891082763672, "min": -155.14340209960938, "p10": -48.13290710449218, "median": 39.419342041015625, "p90": 111.66568145751954, "max": 186.3214569091797, "pos_frac": 0.734375, "sample": [36.538818359375, -14.817451477050781, 99.12272644042969, -49.390480041503906, 15.59383773803711, 7.939361572265625, 73.20195770263672, 51.374755859375, 56.664222717285156, 42.81048583984375, 157.03646850585938, 56.53527069091797, 28.825456619262695, 4.6399993896484375, 24.17730712890625, 7.194740295410156, 71.17057037353516, -0.3218994140625, 15.544639587402344, -8.974382400512695, -1.121652603149414, 27.922393798828125, 58.13470458984375, 106.64883422851562, -63.65760803222656, 58.97870635986328, 110.10417175292969, -155.14340209960938, -53.538719177246094, 140.80557250976562, -55.76557922363281, 54.60005187988281, 1.4852123260498047, 61.636962890625, 162.70082092285156, -53.742881774902344, 41.377220153808594, 37.461463928222656, 92.17752838134766, 186.3214569091797, 15.740179061889648, 31.801895141601562, 63.74472427368164, 57.73423767089844, 44.39825439453125, 77.83123016357422, 91.50647735595703, -62.72758483886719, 36.40608215332031, -45.198570251464844, 72.48204040527344, 105.21353912353516, 112.33489990234375, 49.588783264160156, -0.7657699584960938, 45.823692321777344, 21.638628005981445, -18.365066528320312, 65.10987854003906, -35.487831115722656, 170.2367401123047, -21.15314292907715, 135.57066345214844, -26.85162353515625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000146.npy"}
|
|
{"epoch": 0.2143906020558003, "step": 147, "batch_size": 64, "mean": 51.21372985839844, "std": 67.43608093261719, "min": -93.89433288574219, "p10": -16.012209320068358, "median": 39.32014465332031, "p90": 148.29690399169925, "max": 222.5404052734375, "pos_frac": 0.765625, "sample": [109.8499755859375, -46.08815383911133, 44.37956237792969, 179.00360107421875, -13.375526428222656, -4.085929870605469, 74.78115844726562, 82.37289428710938, 30.88214874267578, -20.74547576904297, 21.882471084594727, 64.44759368896484, 48.24763107299805, 100.80450439453125, 27.06515121459961, 160.61923217773438, 91.36907958984375, 20.91642189025879, -1.08685302734375, 142.26919555664062, 143.30039978027344, 1.0668411254882812, 130.20208740234375, 22.677988052368164, 212.93212890625, 38.431907653808594, 173.66183471679688, 81.13824462890625, -1.4473590850830078, 40.5413818359375, 24.347919464111328, -93.89433288574219, 6.963186264038086, 97.03256225585938, 83.19706726074219, -19.106689453125, 53.14739990234375, 222.5404052734375, 57.52660369873047, 20.260753631591797, 131.60653686523438, 13.581939697265625, -4.659765243530273, 14.115325927734375, 24.666534423828125, 49.94092559814453, -0.1107330322265625, 58.41230392456055, 55.777374267578125, -16.581344604492188, 24.135009765625, 39.91536331176758, 19.769126892089844, 120.46757507324219, -14.684226989746094, 44.506465911865234, 187.94784545898438, -10.34079360961914, 22.65424346923828, -88.44256591796875, 150.43826293945312, -39.70489501953125, 47.5145149230957, 38.72492599487305], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000147.npy"}
|
|
{"epoch": 0.21585903083700442, "step": 148, "batch_size": 64, "mean": 41.63152313232422, "std": 59.94535827636719, "min": -72.06304931640625, "p10": -16.936394119262694, "median": 31.151267051696777, "p90": 111.62704315185547, "max": 244.0223388671875, "pos_frac": 0.828125, "sample": [0.3323020935058594, 39.828826904296875, 5.511138916015625, 81.61181640625, -29.002220153808594, 68.25225067138672, 79.43922424316406, 34.07734298706055, -22.274864196777344, 155.00660705566406, 14.48126220703125, 111.4769515991211, 20.93824577331543, 6.594032287597656, 112.99726867675781, 176.20452880859375, 1.8487396240234375, 2.3203125, -24.34357452392578, -43.052154541015625, 110.04731750488281, 99.8116455078125, -3.7753467559814453, 66.5487060546875, 85.37029266357422, 36.90849304199219, 54.21320343017578, 32.547760009765625, 88.37162780761719, -1.1413002014160156, 111.69136810302734, 25.048095703125, 31.157625198364258, 45.6156005859375, -17.872547149658203, 31.872791290283203, 20.452835083007812, 19.975051879882812, 87.99048614501953, 32.27348709106445, 105.01775360107422, 31.149635314941406, -72.06304931640625, 28.23883056640625, 31.15289878845215, 20.52307891845703, 122.21142578125, 32.24921417236328, 9.324409484863281, -14.752037048339844, -66.80763244628906, 33.77105712890625, 10.629831314086914, 8.87806510925293, 31.968795776367188, 6.284509658813477, 25.412818908691406, 244.0223388671875, 229.7880859375, 7.129426956176758, -2.034952163696289, 10.472419738769531, 77.27119445800781, 5.224264144897461], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000148.npy"}
|
|
{"epoch": 0.2173274596182085, "step": 149, "batch_size": 64, "mean": 52.64115905761719, "std": 59.961753845214844, "min": -120.08415222167969, "p10": -0.4945817947387692, "median": 42.17257118225098, "p90": 122.7820640563965, "max": 237.5161895751953, "pos_frac": 0.875, "sample": [90.77681732177734, 53.32789611816406, 164.38812255859375, 41.89479446411133, 84.92626190185547, 42.450347900390625, 53.06189727783203, -11.798269271850586, -19.7987060546875, 5.803325653076172, 16.99593162536621, 4.976234436035156, 54.21151351928711, 37.55500030517578, 39.99003601074219, -38.448028564453125, 87.90797424316406, 72.64569091796875, 34.528533935546875, 97.16182708740234, 72.77406311035156, 134.5307159423828, 82.87377166748047, 23.300249099731445, 13.837301254272461, 72.06758880615234, 36.47123718261719, 100.45020294189453, 67.31698608398438, 98.01336669921875, 54.04618835449219, -0.6367969512939453, 237.5161895751953, 29.86394691467285, 121.58390045166016, 23.54656982421875, 173.76193237304688, 49.382179260253906, 23.42010498046875, 87.06584930419922, 25.657520294189453, 71.20571899414062, 13.44230842590332, 40.76148986816406, -2.9778823852539062, 48.2733154296875, 17.712017059326172, 54.85063934326172, 2.266632080078125, -120.08415222167969, 58.0714111328125, 23.341712951660156, 5.865386962890625, 123.29556274414062, 16.265905380249023, 178.15628051757812, 13.504219055175781, 37.02545928955078, 117.17329406738281, 207.8824462890625, 12.557426452636719, -0.16274642944335938, 69.68251037597656, -58.47510528564453], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000149.npy"}
|
|
{"epoch": 0.21879588839941264, "step": 150, "batch_size": 64, "mean": 49.994728088378906, "std": 69.78427124023438, "min": -64.0477523803711, "p10": -11.983818626403805, "median": 32.85163497924805, "p90": 132.88994598388672, "max": 305.1092529296875, "pos_frac": 0.828125, "sample": [212.7515869140625, 243.39854431152344, 11.583869934082031, 21.143814086914062, 53.028343200683594, -14.248283386230469, 131.21359252929688, 32.034751892089844, -64.0477523803711, 37.75286102294922, 305.1092529296875, -13.809442520141602, -2.5466690063476562, 13.812591552734375, 4.430213928222656, 91.03543090820312, 22.682830810546875, 45.792701721191406, 43.95509338378906, 52.40721893310547, 31.701332092285156, 55.316856384277344, 130.06680297851562, 26.307090759277344, 91.13519287109375, 72.75843811035156, 148.0994110107422, 100.6773681640625, 16.27550506591797, 62.75020980834961, 8.254486083984375, -37.51091003417969, 61.788917541503906, 133.60838317871094, 33.66851806640625, 6.627170562744141, 135.56439208984375, 18.471189498901367, 104.85762786865234, -4.926170349121094, 70.67353820800781, -14.324592590332031, -21.335350036621094, 0.49022674560546875, 99.4807357788086, 3.603759765625, 16.692726135253906, 53.01128387451172, 16.66857147216797, 31.136199951171875, 35.90580749511719, 51.236881256103516, 52.508766174316406, 13.660186767578125, 37.14424133300781, 260.681884765625, 14.397214889526367, 13.849403381347656, 7.309175491333008, 39.07355499267578, 60.53858184814453, -7.724029541015625, -57.294776916503906, -0.6938667297363281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000150.npy"}
|
|
{"epoch": 0.22026431718061673, "step": 151, "batch_size": 64, "mean": 52.55384063720703, "std": 58.179969787597656, "min": -89.37619018554688, "p10": -16.96456642150879, "median": 43.36668014526367, "p90": 140.8277069091797, "max": 189.3645477294922, "pos_frac": 0.875, "sample": [-89.37619018554688, 189.3645477294922, 44.78529357910156, 2.0066909790039062, 26.411270141601562, 39.70916748046875, 56.73944854736328, 86.07624816894531, 67.86911010742188, 29.96142578125, 136.39654541015625, -26.531137466430664, -17.439346313476562, 88.93827819824219, 39.840065002441406, -24.201854705810547, 1.7558555603027344, 32.53020477294922, 42.43925476074219, 38.90569305419922, 171.38653564453125, 53.59128952026367, 95.58969116210938, 46.129112243652344, -36.11090850830078, 151.71743774414062, 19.780540466308594, 14.664825439453125, -15.856746673583984, 33.217529296875, 42.1151123046875, 24.93781089782715, 19.716815948486328, 14.517728805541992, 69.03253173828125, 142.72677612304688, -53.79833984375, 62.209129333496094, 75.52075958251953, 76.30206298828125, 10.072736740112305, 53.045166015625, 66.26898193359375, 85.08596801757812, 18.016082763671875, 159.5531005859375, 89.67063903808594, 44.294105529785156, 100.48989868164062, 13.301864624023438, 123.08656311035156, 11.016311645507812, 3.180543899536133, 109.00183868408203, 20.829681396484375, 12.809268951416016, 158.581787109375, 90.50074005126953, 180.12127685546875, 101.02114868164062, -35.42975616455078, 63.79681396484375, 17.417694091796875, 94.14306640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000151.npy"}
|
|
{"epoch": 0.22173274596182085, "step": 152, "batch_size": 64, "mean": 60.06842803955078, "std": 69.07414245605469, "min": -56.738250732421875, "p10": -16.94092636108398, "median": 49.68106269836426, "p90": 158.9708053588868, "max": 241.45765686035156, "pos_frac": 0.828125, "sample": [27.9302978515625, 33.06324005126953, 130.21157836914062, 139.05523681640625, 56.52688980102539, 142.19915771484375, 51.207881927490234, 89.86944580078125, 221.02879333496094, 27.52098846435547, 123.23966217041016, 102.49420166015625, 26.970386505126953, -38.00935363769531, 8.156974792480469, 175.08184814453125, 104.4108657836914, 32.428009033203125, 56.51884841918945, 38.22572326660156, 51.682464599609375, 40.85688781738281, -41.17228698730469, 57.5108642578125, 7.73333740234375, 241.45765686035156, 96.28653717041016, 174.85707092285156, 34.991424560546875, 14.470649719238281, 61.71565246582031, -55.00897216796875, 28.941463470458984, -2.85797119140625, 3.4177303314208984, 51.981361389160156, 26.190872192382812, 142.8719482421875, 32.266151428222656, 34.41571807861328, 222.66452026367188, -7.22308349609375, 48.15424346923828, 117.84890747070312, 51.7586669921875, -12.767745971679688, 193.81455993652344, 27.082151412963867, -46.21826934814453, 87.30003356933594, 17.33384132385254, 81.58160400390625, 51.22357177734375, 34.68943786621094, 165.8703155517578, -31.44239044189453, 110.65963745117188, -56.738250732421875, 2.1467056274414062, 92.87619018554688, -17.463241577148438, -15.722190856933594, 100.44112396240234, 75.76995086669922], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000152.npy"}
|
|
{"epoch": 0.22320117474302498, "step": 153, "batch_size": 64, "mean": 51.24233627319336, "std": 70.5565414428711, "min": -109.33355712890625, "p10": -23.538407135009766, "median": 37.899452209472656, "p90": 148.54944305419926, "max": 302.01080322265625, "pos_frac": 0.796875, "sample": [71.86271667480469, 17.943161010742188, 13.07012939453125, 1.5029067993164062, 14.445659637451172, 10.149650573730469, 25.739013671875, 97.86102294921875, -23.84991455078125, 152.21681213378906, 69.61198425292969, 21.889190673828125, -16.68532943725586, 67.80841064453125, 185.05113220214844, -54.015907287597656, 75.62789916992188, 12.553510665893555, 47.08301544189453, 86.84724426269531, 42.724761962890625, 302.01080322265625, 33.031715393066406, 55.853248596191406, 62.972442626953125, 139.99224853515625, -8.579872131347656, -23.8824462890625, 158.1267852783203, 117.59921264648438, 115.27848815917969, 20.547996520996094, 153.338623046875, 7.192914962768555, 1.203216552734375, 4.444854736328125, 103.2431411743164, 66.15318298339844, -24.39883804321289, -23.843135833740234, 83.39561462402344, 33.07414245605469, 9.110488891601562, 173.65557861328125, 20.86932373046875, -10.631465911865234, -47.71727752685547, 48.96476745605469, 18.28900909423828, -109.33355712890625, 71.39869689941406, 248.90882873535156, 62.01416015625, -22.827373504638672, 51.72503662109375, -1.4358253479003906, 73.07577514648438, 32.667510986328125, 92.7583999633789, 113.2536849975586, 46.95028305053711, -13.858482360839844, 31.756683349609375, 93.72390747070312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000153.npy"}
|
|
{"epoch": 0.22466960352422907, "step": 154, "batch_size": 64, "mean": 50.88260269165039, "std": 73.4244613647461, "min": -117.58816528320312, "p10": -34.427805709838864, "median": 39.6772575378418, "p90": 141.52395172119142, "max": 216.68695068359375, "pos_frac": 0.703125, "sample": [-5.696508407592773, 31.269065856933594, -117.58816528320312, 15.13414192199707, 213.75198364257812, 48.713287353515625, -4.923341751098633, 89.85836791992188, 121.81169128417969, -28.385234832763672, 56.31006622314453, 173.23812866210938, -46.79620361328125, 88.14215087890625, 97.55116271972656, 54.85141372680664, 0.09795570373535156, -0.7470550537109375, 31.28249740600586, -1.5306663513183594, 78.81253051757812, 63.52390670776367, 3.197887420654297, 48.072017669677734, -8.887908935546875, 130.50807189941406, -16.883792877197266, 142.01788330078125, 112.41167449951172, -43.67597961425781, -8.291267395019531, 142.51321411132812, -44.17949676513672, 135.29916381835938, -37.017478942871094, 157.3344268798828, 216.68695068359375, 132.54684448242188, 4.5484771728515625, 136.64797973632812, 140.37144470214844, 96.10052490234375, 17.750690460205078, 60.045135498046875, 135.65049743652344, -0.13771820068359375, 2.728271484375, 7.422178268432617, -63.32598114013672, 89.58943176269531, 108.13862609863281, 5.472526550292969, -5.019350051879883, 27.5263671875, 5.708984375, -61.67952346801758, 89.15910339355469, 101.35952758789062, 86.18376922607422, -27.648223876953125, -1.9176483154296875, 13.149991989135742, 62.52647399902344, 205.8016357421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000154.npy"}
|
|
{"epoch": 0.2261380323054332, "step": 155, "batch_size": 64, "mean": 52.38172912597656, "std": 67.4118881225586, "min": -148.2384796142578, "p10": -9.917257499694824, "median": 49.47332572937012, "p90": 126.22151031494147, "max": 238.03672790527344, "pos_frac": 0.796875, "sample": [5.172813415527344, 60.365257263183594, 30.36726951599121, -5.02642822265625, 70.50135803222656, 10.99073600769043, 71.77667236328125, -11.169082641601562, 64.18450164794922, 16.531213760375977, -9.125152587890625, 48.825077056884766, 62.663612365722656, -29.983154296875, 67.9283218383789, 62.793487548828125, -1.7828330993652344, 44.082061767578125, 51.2459716796875, 15.659881591796875, 57.945098876953125, 50.12157440185547, 47.354156494140625, 238.03672790527344, 166.29525756835938, 8.710456848144531, 99.03460693359375, 86.28704071044922, 19.325408935546875, -38.22703552246094, 15.62293815612793, 2.4130859375, 22.642589569091797, 65.43612670898438, 152.29510498046875, -148.2384796142578, 203.6746063232422, 34.80231475830078, 108.0521469116211, -3.7205047607421875, 8.943485260009766, 75.5227279663086, 111.19978332519531, 34.40126037597656, 231.64016723632812, 70.607666015625, 218.02459716796875, 110.19027709960938, -45.696136474609375, 56.27325439453125, -3.47222900390625, 132.65939331054688, 12.482555389404297, 87.80469512939453, 70.70429992675781, -31.17156219482422, 55.364524841308594, 106.12516021728516, 96.31014251708984, 6.161762237548828, 100.73873901367188, -0.28359031677246094, -10.256731033325195, 44.29186248779297], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000155.npy"}
|
|
{"epoch": 0.2276064610866373, "step": 156, "batch_size": 64, "mean": 32.21952819824219, "std": 76.53553771972656, "min": -203.2432861328125, "p10": -57.03666687011718, "median": 30.56932258605957, "p90": 109.9540382385254, "max": 199.29132080078125, "pos_frac": 0.71875, "sample": [104.49882507324219, 43.35917663574219, 191.3549346923828, 195.08067321777344, 122.05894470214844, 73.62069702148438, -135.98867797851562, -0.3490447998046875, -35.2696533203125, 14.257377624511719, -114.92411804199219, 108.18731689453125, 67.54816436767578, -13.377891540527344, -36.16306686401367, -105.58495330810547, 43.34394073486328, 83.07406616210938, 188.22569274902344, 46.05503845214844, 12.584701538085938, 89.94377136230469, 25.68072509765625, 31.23666763305664, 6.070222854614258, -1.9372406005859375, 14.985950469970703, 33.24890899658203, 199.29132080078125, -7.405433654785156, 176.436767578125, 110.7112045288086, 102.992431640625, 45.43421173095703, -3.3012619018554688, 3.710397720336914, 48.89508056640625, -72.37661743164062, -203.2432861328125, -49.53379821777344, 60.286598205566406, -27.712356567382812, 19.81524658203125, 93.36720275878906, 19.487136840820312, -60.25218200683594, 64.54851531982422, 99.1811752319336, 56.41642761230469, 12.662263870239258, -1.4051361083984375, 66.67205810546875, 40.24004364013672, 49.31292724609375, -88.99591827392578, -18.881393432617188, 29.9019775390625, 38.96385192871094, 20.311080932617188, 5.7808380126953125, 25.283416748046875, 57.67888641357422, 6.702404022216797, 90.25259399414062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000156.npy"}
|
|
{"epoch": 0.2290748898678414, "step": 157, "batch_size": 64, "mean": 63.14281463623047, "std": 71.9948959350586, "min": -154.55377197265625, "p10": -15.881114959716786, "median": 56.49549865722656, "p90": 150.59973602294923, "max": 272.12005615234375, "pos_frac": 0.84375, "sample": [117.27203369140625, 124.90734100341797, 70.167724609375, 146.61119079589844, 89.16930389404297, 9.985862731933594, 56.21720886230469, 139.178466796875, 10.558937072753906, 105.42987060546875, 48.0206413269043, 155.421142578125, 42.00225830078125, 61.315765380859375, 142.889892578125, 139.1353302001953, 98.19332122802734, 74.47190856933594, 34.74638366699219, 127.1994857788086, 39.46215057373047, 17.3416748046875, 56.77378845214844, -21.271102905273438, 86.69180297851562, 155.23451232910156, -2.6868133544921875, 29.029495239257812, 4.457916259765625, 97.00187683105469, -154.55377197265625, -5.710535049438477, 19.68963623046875, 30.26370620727539, 90.61585998535156, 71.16841888427734, 165.84320068359375, 92.96369934082031, 86.6471176147461, -20.23993492126465, 15.892322540283203, 236.1195068359375, 27.772062301635742, 29.001670837402344, 5.687923431396484, 154.93255615234375, 87.83137512207031, -22.68794059753418, 152.30911254882812, 83.5230712890625, 131.09567260742188, 27.139968872070312, -37.38069534301758, 19.068031311035156, 128.6153564453125, 102.02694702148438, 272.12005615234375, 52.203712463378906, 35.47943115234375, -93.83560943603516, 23.864227294921875, -28.484283447265625, 8.762901306152344, -1.5340576171875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000157.npy"}
|
|
{"epoch": 0.2305433186490455, "step": 158, "batch_size": 64, "mean": 65.51558685302734, "std": 92.3259506225586, "min": -77.06822967529297, "p10": -20.903793716430663, "median": 43.50282669067383, "p90": 197.4324310302735, "max": 362.1778869628906, "pos_frac": 0.734375, "sample": [13.92892837524414, 10.145942687988281, 48.84039306640625, 41.857460021972656, 42.030487060546875, -16.827308654785156, 22.748783111572266, 49.355018615722656, 35.26613235473633, -20.17813491821289, 150.24673461914062, -75.29959106445312, 181.90745544433594, 69.40367889404297, -3.45989990234375, -29.57909393310547, -10.015130996704102, 77.65583801269531, -1.0841655731201172, 9.08599853515625, 182.87045288085938, -21.21479034423828, 180.8212890625, -67.86863708496094, 215.21865844726562, 44.97516632080078, 101.08837890625, 17.455007553100586, 18.474044799804688, 15.527286529541016, 56.3991813659668, 60.63154602050781, -3.1113433837890625, 116.667724609375, -26.864131927490234, -12.752670288085938, 255.43321228027344, 93.39520263671875, -77.06822967529297, 242.39060974121094, 296.00439453125, 13.911026000976562, 45.27204132080078, 362.1778869628906, 14.472564697265625, 14.94619369506836, -27.969146728515625, 83.85678100585938, 127.53372192382812, 62.31560516357422, 277.91986083984375, 136.72726440429688, 84.82847595214844, 116.87913513183594, -14.954046249389648, 123.2279052734375, 31.908676147460938, -7.108152389526367, 33.59333038330078, 72.67705535888672, 203.67327880859375, 52.100746154785156, 108.92608642578125, -8.420492172241211], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000158.npy"}
|
|
{"epoch": 0.23201174743024963, "step": 159, "batch_size": 64, "mean": 47.499595642089844, "std": 73.97724151611328, "min": -140.80502319335938, "p10": -34.35602111816406, "median": 45.32338333129883, "p90": 147.01845550537112, "max": 251.89425659179688, "pos_frac": 0.765625, "sample": [9.733230590820312, 32.961097717285156, 88.97560119628906, -36.54597473144531, 169.1827392578125, 96.12173461914062, -140.80502319335938, -26.046489715576172, 87.1041488647461, 100.4891357421875, 67.00837707519531, -36.7840576171875, 14.829935073852539, -6.759979248046875, 55.6485595703125, -28.429214477539062, -41.184295654296875, 52.54732131958008, 136.8135223388672, 110.77520751953125, 251.89425659179688, 7.098545074462891, -31.709808349609375, 14.19964599609375, 16.970870971679688, -29.625869750976562, 14.904808044433594, -35.4901123046875, 45.6282958984375, 44.106353759765625, -2.578369140625, 151.39199829101562, 48.813560485839844, 9.218212127685547, 74.57842254638672, -18.148134231567383, 51.880210876464844, -38.549072265625, 3.5967044830322266, 185.35641479492188, 101.72012329101562, 161.05047607421875, 49.68870544433594, 60.951698303222656, 6.681236267089844, 216.44168090820312, 10.699850082397461, 0.5232009887695312, 53.476112365722656, 49.19886779785156, 105.0095443725586, 24.068016052246094, -24.966346740722656, -95.36911010742188, 55.42329406738281, 64.26083374023438, 85.8230209350586, 85.87568664550781, 15.713785171508789, 225.36004638671875, 134.26351928710938, 45.018470764160156, 33.946929931640625, 105.94194030761719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000159.npy"}
|
|
{"epoch": 0.23348017621145375, "step": 160, "batch_size": 64, "mean": 60.404296875, "std": 88.49148559570312, "min": -88.56803894042969, "p10": -47.44531173706054, "median": 38.79554748535156, "p90": 171.5130340576172, "max": 331.0348815917969, "pos_frac": 0.71875, "sample": [69.61607360839844, 5.81120491027832, 128.95736694335938, 120.39747619628906, -0.7331886291503906, -39.906951904296875, 92.52886199951172, 136.20741271972656, 131.87802124023438, 3.7060394287109375, -56.73197937011719, 6.0506591796875, 119.68821716308594, 176.58648681640625, 90.33435821533203, -49.963623046875, 24.97365951538086, 38.18663024902344, -88.56803894042969, 159.11634826660156, 229.84523010253906, -62.411033630371094, 144.3981475830078, -41.569252014160156, 103.36903381347656, 25.689590454101562, 10.611732482910156, -1.7845916748046875, 331.0348815917969, 54.425537109375, 14.879583358764648, 52.21307373046875, -3.5426788330078125, 271.758056640625, -6.0313262939453125, 29.874595642089844, 39.40446472167969, 135.92384338378906, 253.232177734375, 230.27061462402344, 33.98717498779297, 149.68792724609375, -52.0345458984375, 164.41513061523438, 80.97628784179688, 20.576980590820312, -23.342599868774414, 24.6763858795166, 60.810096740722656, 86.89990234375, -6.578727722167969, 29.914146423339844, 57.01139831542969, -50.63819122314453, 46.821533203125, 70.64474487304688, -3.395975112915039, -0.645263671875, -66.56490325927734, 102.90135192871094, 16.642372131347656, 174.55499267578125, -15.095977783203125, 83.92393493652344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000160.npy"}
|
|
{"epoch": 0.23494860499265785, "step": 161, "batch_size": 64, "mean": 54.21880340576172, "std": 74.31616973876953, "min": -74.84642791748047, "p10": -35.3088493347168, "median": 37.71525955200195, "p90": 159.29404602050784, "max": 239.50698852539062, "pos_frac": 0.765625, "sample": [63.39885711669922, 154.49911499023438, -33.69239044189453, 128.53269958496094, 205.19430541992188, -74.84642791748047, -36.001617431640625, -39.618072509765625, 61.78619384765625, -31.32978057861328, 46.95066833496094, 93.58952331542969, 66.66671752929688, -53.35602569580078, -2.499401092529297, 83.29985046386719, 217.3448028564453, 61.23878479003906, 103.17110443115234, 193.1900177001953, 27.4224853515625, 43.65022659301758, 12.500564575195312, -23.106704711914062, 40.22254943847656, -6.286640167236328, 89.34207916259766, 24.174232482910156, 217.565673828125, 239.50698852539062, 87.3043212890625, -8.186250686645508, 14.528495788574219, -21.07622528076172, 156.95703125, 92.37936401367188, 7.1690673828125, -41.03184509277344, 16.350112915039062, 72.02447509765625, 15.77386474609375, 12.212028503417969, -11.552970886230469, 33.456214904785156, 35.207969665527344, 29.004302978515625, 160.29562377929688, 64.96785736083984, 10.674537658691406, 84.43428802490234, 145.722412109375, -62.121490478515625, 8.920181274414062, 146.32339477539062, 27.411102294921875, 27.968612670898438, 25.60416030883789, 68.89791870117188, 78.35269165039062, -44.27970886230469, 30.246219635009766, 186.372802734375, 90.44020080566406, 56.74231719970703], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000161.npy"}
|
|
{"epoch": 0.23641703377386197, "step": 162, "batch_size": 64, "mean": 55.859214782714844, "std": 79.09968566894531, "min": -97.84980773925781, "p10": -19.25245590209961, "median": 44.612545013427734, "p90": 145.21487884521486, "max": 375.6427917480469, "pos_frac": 0.765625, "sample": [-15.872795104980469, 39.27811813354492, 70.90740966796875, 171.92579650878906, 18.966323852539062, 173.54107666015625, 15.049568176269531, 103.64431762695312, 28.4946231842041, -13.446081161499023, -15.377189636230469, 145.24398803710938, 48.37232208251953, -49.416481018066406, -19.484466552734375, -9.244640350341797, 120.22918701171875, 114.5528793334961, 6.4593963623046875, 90.59088134765625, 59.742286682128906, 22.363304138183594, 46.440765380859375, 12.300148010253906, 10.357393264770508, 138.29486083984375, 99.14404296875, -5.4775848388671875, -97.84980773925781, 99.387451171875, 86.3544692993164, -63.46784973144531, -6.154535293579102, 16.058385848999023, 120.99808502197266, 50.2995719909668, 157.97756958007812, 375.6427917480469, 61.63214874267578, 76.96883392333984, 70.85424041748047, 64.76287078857422, 123.84434509277344, 130.05506896972656, 117.74395751953125, 8.79901123046875, -2.9834442138671875, 11.696861267089844, 85.83084106445312, 27.983261108398438, -20.966812133789062, 145.14695739746094, 1.865966796875, -42.330543518066406, 265.63385009765625, 36.67565155029297, -78.16329956054688, 14.150314331054688, 48.70088195800781, 71.04320526123047, 42.784324645996094, -18.711097717285156, 32.227455139160156, 152.9193115234375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000162.npy"}
|
|
{"epoch": 0.23788546255506607, "step": 163, "batch_size": 64, "mean": 53.46131134033203, "std": 71.76974487304688, "min": -76.07685089111328, "p10": -20.073341369628903, "median": 43.07544708251953, "p90": 157.14022674560547, "max": 235.04966735839844, "pos_frac": 0.75, "sample": [2.9458160400390625, -5.563148498535156, 201.1818389892578, 220.76780700683594, -20.93238067626953, -61.0606689453125, 51.71623992919922, -7.9594268798828125, -3.6007537841796875, 59.96076965332031, 90.56621551513672, -6.123588562011719, 92.5697250366211, 235.04966735839844, 74.38372039794922, 51.59349822998047, 90.04129791259766, 37.56361389160156, 15.918655395507812, -18.06891632080078, 35.361061096191406, 95.31014251708984, 77.60891723632812, 21.32129669189453, 65.89479064941406, 3.0341148376464844, 106.11573028564453, -76.07685089111328, -33.084251403808594, 69.51972198486328, 153.0605926513672, 8.334403991699219, 97.05577850341797, 17.452306747436523, 53.626991271972656, -8.434286117553711, -34.72021484375, 210.1117706298828, 98.05298614501953, 30.934219360351562, -37.71577453613281, 34.74816131591797, -7.06182861328125, 158.88864135742188, 45.31573486328125, 82.93061065673828, 2.4975357055664062, 202.2544708251953, 88.69877624511719, 15.81656265258789, 38.21501159667969, 74.47454833984375, 40.83515930175781, 149.3154296875, 69.95500183105469, -7.72869873046875, 64.64533233642578, 3.0378952026367188, -63.204994201660156, 0.2058391571044922, 195.22958374023438, 62.237030029296875, 119.33353424072266, -2.8287601470947266], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000163.npy"}
|
|
{"epoch": 0.2393538913362702, "step": 164, "batch_size": 64, "mean": 63.41717529296875, "std": 73.10796356201172, "min": -99.57868957519531, "p10": -10.321340751647945, "median": 56.971384048461914, "p90": 136.0906921386719, "max": 299.3934326171875, "pos_frac": 0.859375, "sample": [-99.57868957519531, 110.97305297851562, 123.4307632446289, 110.96207427978516, 158.69830322265625, 43.44055938720703, 63.99647521972656, 95.15878295898438, 64.50152587890625, 68.19229125976562, 181.44256591796875, 28.23779296875, 1.581125259399414, 47.974220275878906, -13.87652587890625, 74.63328552246094, 53.78223419189453, 55.00834655761719, -13.420663833618164, 46.45533752441406, -6.823579788208008, 87.53279876708984, 58.93442153930664, 121.792724609375, 35.2509765625, 87.89910888671875, 0.10139846801757812, -53.32463073730469, 100.18602752685547, 18.097816467285156, 32.195892333984375, -11.820381164550781, 104.1259765625, 133.16397094726562, 100.92313385009766, 91.66236877441406, 24.13941192626953, 88.38764953613281, 25.172996520996094, 122.00328063964844, 26.642105102539062, 60.80915451049805, 47.718849182128906, 39.622802734375, 40.32750701904297, 2.5823516845703125, 66.0030517578125, 9.525352478027344, -89.47683715820312, 96.67588806152344, 58.9776611328125, 45.90740966796875, 249.98007202148438, 10.363037109375, 276.8250732421875, 178.2222900390625, 6.88836669921875, 137.34500122070312, 79.8092041015625, -25.892921447753906, -0.9850234985351562, 79.39056396484375, 299.3934326171875, 0.8505325317382812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000164.npy"}
|
|
{"epoch": 0.24082232011747431, "step": 165, "batch_size": 64, "mean": 41.572532653808594, "std": 65.63346862792969, "min": -109.90958404541016, "p10": -40.92247085571289, "median": 34.609375, "p90": 134.2801483154297, "max": 176.59507751464844, "pos_frac": 0.734375, "sample": [62.371803283691406, 64.5324935913086, -42.29115295410156, 15.692268371582031, -40.4095458984375, -0.5294132232666016, 62.232398986816406, 75.4207763671875, -1.9855804443359375, -50.37171173095703, 74.1404800415039, -109.90958404541016, -31.6317138671875, 36.90007019042969, 30.080169677734375, -63.44275665283203, -9.234668731689453, -16.151397705078125, 107.7802505493164, 116.47332763671875, 71.55569458007812, 29.098880767822266, 15.734580993652344, 172.96282958984375, -75.66484069824219, 60.37300109863281, -3.1671199798583984, -38.18882751464844, 13.217109680175781, 109.05838012695312, -41.142295837402344, 19.280073165893555, 142.2520751953125, 130.7658233642578, 25.560272216796875, -20.79737091064453, 65.8319091796875, 23.227859497070312, 52.41095733642578, 6.059717178344727, 32.31867980957031, 40.778839111328125, 86.35298156738281, 60.86414337158203, 171.9066162109375, 1.192840576171875, 76.17321014404297, 55.629638671875, 20.02104949951172, -48.632904052734375, 18.413726806640625, 135.755126953125, 37.524749755859375, -24.675987243652344, 41.92979049682617, 173.6261444091797, 22.426101684570312, 176.59507751464844, 166.872802734375, 109.10533142089844, 74.57478332519531, 8.817512512207031, 54.138031005859375, 130.83853149414062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000165.npy"}
|
|
{"epoch": 0.2422907488986784, "step": 166, "batch_size": 64, "mean": 74.51507568359375, "std": 96.98413848876953, "min": -44.368080139160156, "p10": -17.2642951965332, "median": 52.4578971862793, "p90": 200.08306427001963, "max": 401.06549072265625, "pos_frac": 0.84375, "sample": [335.35040283203125, 217.6140594482422, 15.844810485839844, 347.56170654296875, -22.510154724121094, 22.28683090209961, 16.305496215820312, 59.36676025390625, 149.09735107421875, 3.833587646484375, 79.33412170410156, -5.679008483886719, 8.63827896118164, 113.32118225097656, -17.56214141845703, 53.09862518310547, 109.86346435546875, 34.653282165527344, 27.79349136352539, 120.27933502197266, 10.416543960571289, 401.06549072265625, 27.834026336669922, -28.109050750732422, -44.368080139160156, -35.52885055541992, 29.76342010498047, 90.9100112915039, -43.857666015625, 167.21121215820312, 212.14198303222656, 316.0727233886719, 89.15223693847656, 68.51829528808594, 19.44952392578125, 267.53289794921875, 12.0322265625, 130.34152221679688, 51.817169189453125, 91.2203369140625, -5.146747589111328, 64.43717956542969, 14.176338195800781, 23.833263397216797, 155.64474487304688, 94.83497619628906, 171.94558715820312, 9.777362823486328, 64.05979919433594, 105.06008911132812, 26.225574493408203, 78.4060287475586, -38.80644226074219, 45.24103927612305, 5.373085021972656, -16.569320678710938, 96.31977081298828, 8.715957641601562, 75.86051940917969, 8.342559814453125, 77.92813110351562, 12.732015609741211, 134.78436279296875, 53.68175506591797], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000166.npy"}
|
|
{"epoch": 0.24375917767988253, "step": 167, "batch_size": 64, "mean": 79.16014862060547, "std": 79.81779479980469, "min": -92.4237060546875, "p10": -5.82389678955078, "median": 65.50175476074219, "p90": 184.79736175537118, "max": 277.51971435546875, "pos_frac": 0.84375, "sample": [100.49021911621094, 59.74409103393555, 106.09675598144531, 162.1293182373047, 89.3175048828125, 113.72393798828125, 48.406280517578125, 51.144004821777344, 88.28911590576172, 48.0545768737793, 277.51971435546875, 146.19796752929688, 59.031402587890625, -4.6772003173828125, 15.245697021484375, 48.10129928588867, 62.72368621826172, -29.782272338867188, 96.87809753417969, 141.3601531982422, 24.582101821899414, 32.062110900878906, 52.44805908203125, 39.797203063964844, 85.38822937011719, -92.4237060546875, 213.1094207763672, 161.74502563476562, -2.7801513671875, 148.2267608642578, -10.103385925292969, 59.403045654296875, 76.40194702148438, 32.500450134277344, 257.44287109375, 118.6085205078125, 214.22705078125, 68.27982330322266, 145.47312927246094, -83.97822570800781, 151.5529327392578, 194.51223754882812, 235.095458984375, 111.46175384521484, 265.0234375, -1.3138427734375, 129.18023681640625, 74.54008483886719, 154.87936401367188, 140.96499633789062, 75.93040466308594, 24.31304931640625, 16.530778884887695, 26.281578063964844, 26.947265625, 107.94144439697266, 73.39078521728516, 56.326133728027344, 15.003433227539062, -55.88825225830078, 34.7173957824707, 8.918050765991211, -14.148513793945312, -6.315338134765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000167.npy"}
|
|
{"epoch": 0.24522760646108663, "step": 168, "batch_size": 64, "mean": 60.49474334716797, "std": 72.946044921875, "min": -59.59507751464844, "p10": -21.953336715698235, "median": 41.66514587402344, "p90": 141.47415618896486, "max": 293.667724609375, "pos_frac": 0.8125, "sample": [20.108291625976562, 82.92406463623047, 26.61328887939453, 21.32150650024414, 8.752269744873047, 31.439594268798828, 113.13695526123047, -16.82645034790039, 28.734195709228516, 8.818214416503906, -58.09751892089844, 130.3550567626953, 293.667724609375, 69.15611267089844, -24.15057373046875, 141.73974609375, 131.5006103515625, 160.15879821777344, 140.25978088378906, 41.00904083251953, 166.44589233398438, 204.5843505859375, 140.8544464111328, 117.0980453491211, -15.617637634277344, -38.91447448730469, 1.7252464294433594, 100.81477355957031, 112.33993530273438, 88.33161163330078, 57.487030029296875, 27.41461944580078, -41.05702209472656, 69.00447082519531, 32.53357696533203, 100.00277709960938, -34.15167236328125, -5.929821014404297, -6.705951690673828, 42.04838562011719, 25.322580337524414, 73.50769805908203, 41.887298583984375, 68.60279846191406, 68.79905700683594, 61.377662658691406, 20.680830001831055, -32.33098602294922, 172.72689819335938, 39.840675354003906, 112.81895446777344, 18.94423484802246, 31.480440139770508, 18.17333984375, 98.23688507080078, 41.4429931640625, 28.998075485229492, -59.59507751464844, 52.3841667175293, -3.5723114013671875, 291.2916564941406, 20.44709014892578, 90.44371032714844, 120.82553100585938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000168.npy"}
|
|
{"epoch": 0.24669603524229075, "step": 169, "batch_size": 64, "mean": 53.629798889160156, "std": 93.36234283447266, "min": -157.9364776611328, "p10": -41.759467315673824, "median": 33.3319034576416, "p90": 168.51304931640627, "max": 406.5721740722656, "pos_frac": 0.765625, "sample": [49.400062561035156, 66.10311889648438, -4.401813507080078, 16.967742919921875, 142.7707977294922, 96.29277801513672, 18.678512573242188, 96.26631164550781, 64.97408294677734, 22.761547088623047, -11.921627044677734, 50.11511993408203, -157.9364776611328, -1.1663742065429688, 35.780826568603516, -13.286056518554688, 20.686195373535156, -44.80198669433594, 65.0323486328125, 114.78939056396484, 101.888427734375, -34.660255432128906, 42.736785888671875, -48.104740142822266, 159.40316772460938, 170.30435180664062, 48.1073112487793, 18.227951049804688, 13.256210327148438, 24.951906204223633, 164.33334350585938, 7.260688781738281, 14.483480453491211, 51.78050994873047, 73.48020935058594, 101.31541442871094, 172.75604248046875, 15.12446403503418, 65.40361022949219, 159.89125061035156, 9.042470932006836, 331.5035400390625, -66.65314483642578, 32.687255859375, 27.62256622314453, 45.45326232910156, 24.050735473632812, -48.098854064941406, 260.4544677734375, -3.7947540283203125, 406.5721740722656, -0.425750732421875, 82.58879852294922, 22.91553497314453, 199.91366577148438, 0.1417694091796875, 64.70115661621094, 4.561187744140625, 33.9765510559082, -10.081245422363281, -85.63109588623047, 77.05894470214844, 178.06307983398438, -103.35999298095703], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000169.npy"}
|
|
{"epoch": 0.24816446402349487, "step": 170, "batch_size": 64, "mean": 65.30517578125, "std": 76.70462036132812, "min": -138.02093505859375, "p10": -16.969087982177733, "median": 55.20012092590332, "p90": 155.0759490966797, "max": 235.94140625, "pos_frac": 0.796875, "sample": [146.83120727539062, -93.84688568115234, 81.43060302734375, 67.28208923339844, 155.20770263671875, 56.8459358215332, 116.66036987304688, 13.665435791015625, 119.08920288085938, -4.4025115966796875, 72.829345703125, -1.0636100769042969, 60.91325378417969, -37.10649108886719, 6.20451545715332, -21.887039184570312, 81.75560760498047, 218.76382446289062, 25.965408325195312, 196.71609497070312, 4.2288818359375, -1.723012924194336, 118.97491455078125, 40.389957427978516, 1.1798858642578125, 36.517860412597656, -138.02093505859375, -18.085044860839844, 56.98858642578125, 153.1063690185547, 88.50578308105469, 104.21621704101562, 50.765438079833984, 74.5281982421875, 149.232421875, -16.223495483398438, 235.94140625, 5.390113830566406, 1.0452632904052734, 235.52291870117188, 71.70465087890625, -38.572357177734375, 50.6751594543457, 116.9710693359375, 147.0614013671875, -17.28862762451172, 34.888465881347656, -0.39757347106933594, 25.533714294433594, 45.47452163696289, 78.66194915771484, 53.55430603027344, 154.76852416992188, 36.14486312866211, 128.2415771484375, 49.637786865234375, 137.9043426513672, 168.73126220703125, -6.571132659912109, 154.25717163085938, 190.20608520507812, 20.081125259399414, 118.24592590332031, 15.281585693359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000170.npy"}
|
|
{"epoch": 0.24963289280469897, "step": 171, "batch_size": 64, "mean": 53.23592758178711, "std": 92.11030578613281, "min": -161.5408172607422, "p10": -54.479141235351555, "median": 50.28714942932129, "p90": 175.1548217773438, "max": 419.755126953125, "pos_frac": 0.71875, "sample": [-45.96180725097656, 183.51327514648438, 41.176490783691406, 56.62627410888672, -56.276771545410156, -17.782806396484375, 9.47784423828125, 73.09002685546875, -50.64802551269531, 107.76690673828125, 10.366104125976562, 118.07417297363281, 75.75546264648438, 80.59239196777344, 44.716888427734375, -13.939292907714844, 74.33016967773438, -56.12104797363281, 82.76356506347656, -3.1251296997070312, 160.11990356445312, 48.456478118896484, 64.57774353027344, 63.746612548828125, 250.09915161132812, 57.2233772277832, -66.47913360595703, 419.755126953125, -73.7999496459961, 44.887176513671875, 77.68960571289062, -161.5408172607422, 270.49078369140625, 93.44383239746094, 62.39662170410156, -57.148895263671875, 181.59835815429688, 83.93003845214844, 21.001739501953125, 215.15789794921875, 28.235443115234375, 52.117820739746094, 140.1002197265625, -3.336484909057617, -72.43070983886719, 20.778717041015625, 65.47283172607422, 54.98332214355469, -41.09853744506836, 55.712623596191406, -11.016639709472656, -22.919403076171875, 107.6780776977539, 39.194984436035156, 58.25677490234375, -0.7320270538330078, 24.962982177734375, 131.40966796875, -9.403434753417969, 8.306777954101562, 65.0562515258789, 13.987071990966797, 190.29470825195312, 41.48799514770508], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000171.npy"}
|
|
{"epoch": 0.2511013215859031, "step": 172, "batch_size": 64, "mean": 55.54555892944336, "std": 81.6054458618164, "min": -99.04751586914062, "p10": -36.09282989501953, "median": 63.321189880371094, "p90": 147.03509521484375, "max": 336.41278076171875, "pos_frac": 0.6875, "sample": [-16.08576202392578, 95.82132720947266, -99.04751586914062, 73.02053833007812, 20.5548095703125, -24.108001708984375, 112.75315856933594, 74.77873229980469, 51.67108917236328, -51.45073699951172, -91.25582885742188, 41.26442337036133, 20.229286193847656, -34.95121765136719, 124.84759521484375, 70.5875244140625, 18.691814422607422, 43.15864562988281, -41.760562896728516, 65.17510223388672, -17.523712158203125, 79.92890930175781, 123.71056365966797, 120.4136962890625, 12.276908874511719, 60.84864044189453, 160.93789672851562, 148.52969360351562, -36.58209228515625, 206.89959716796875, -20.14783477783203, 83.89606475830078, 101.3377914428711, 69.0545654296875, 94.58883666992188, 236.33273315429688, 336.41278076171875, 46.219215393066406, 85.85236358642578, 29.38410186767578, -17.12238311767578, -31.719627380371094, 62.22235107421875, 66.38395690917969, 143.54769897460938, -10.584716796875, -1.7673206329345703, 31.784696578979492, -15.340484619140625, 134.1163330078125, 121.81443786621094, -2.8160324096679688, -21.473388671875, 87.06697082519531, 83.82474517822266, -94.9532470703125, 69.43380737304688, 84.77731323242188, 189.76907348632812, 64.42002868652344, -10.25177001953125, -57.413490295410156, 171.66256713867188, 131.26901245117188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000172.npy"}
|
|
{"epoch": 0.2525697503671072, "step": 173, "batch_size": 64, "mean": 55.74905014038086, "std": 69.81259155273438, "min": -125.27297973632812, "p10": -10.879204177856439, "median": 44.68349647521973, "p90": 156.1718444824219, "max": 266.56463623046875, "pos_frac": 0.84375, "sample": [36.450469970703125, 61.097740173339844, 40.139244079589844, 7.883323669433594, -3.3578643798828125, 63.05711364746094, 5.609439849853516, 118.33169555664062, 98.3795166015625, 5.336786270141602, 14.4727783203125, -23.878334045410156, 14.566780090332031, 14.068794250488281, 58.68707275390625, 197.06349182128906, 51.602203369140625, 47.862056732177734, 54.60784149169922, 148.14239501953125, 84.61106872558594, 54.01568603515625, 77.89700317382812, -39.835472106933594, 41.96998596191406, 30.76317596435547, 266.56463623046875, 21.59330940246582, 134.22879028320312, 159.613037109375, 49.594825744628906, 19.065675735473633, -35.5982666015625, 107.21863555908203, 93.26927185058594, -2.1844024658203125, 91.71288299560547, 47.47161865234375, -4.128013610839844, 185.96942138671875, 68.83543395996094, 125.00065612792969, -13.772571563720703, 29.92267608642578, 7.119789123535156, 15.366092681884766, 185.76747131347656, 96.24795532226562, -35.222877502441406, 12.043838500976562, 73.84783172607422, 43.0202751159668, 98.35237121582031, -48.0252685546875, -125.27297973632812, 238.85223388671875, 33.171119689941406, 3.53436279296875, 64.02894592285156, 16.773212432861328, 170.63043212890625, 46.346717834472656, 38.936283111572266, 28.499786376953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000173.npy"}
|
|
{"epoch": 0.2540381791483113, "step": 174, "batch_size": 64, "mean": 62.442710876464844, "std": 85.28379821777344, "min": -86.16241455078125, "p10": -31.24297332763671, "median": 47.85796928405762, "p90": 192.2117645263673, "max": 329.3406677246094, "pos_frac": 0.734375, "sample": [16.101272583007812, -0.0626678466796875, -34.48027801513672, -40.44834899902344, -3.1457977294921875, 27.50768280029297, -2.4202232360839844, 24.07576560974121, 88.65218353271484, 144.20066833496094, 170.2779541015625, -6.708259582519531, 48.51453399658203, 13.534915924072266, 54.58758544921875, -23.68926239013672, 100.18338012695312, 116.77128601074219, 117.87467956542969, 201.61196899414062, 51.95524597167969, 47.2014045715332, 40.33155059814453, -7.947822570800781, 221.03668212890625, -77.95368957519531, 100.66271209716797, 216.9324951171875, 11.352447509765625, 78.16580200195312, 15.51966667175293, -11.01888656616211, 39.936912536621094, -51.16373062133789, 89.1632080078125, -23.376251220703125, 100.82418823242188, 81.916259765625, -82.27938842773438, 205.1984405517578, 226.47149658203125, 283.63177490234375, 18.41266632080078, 77.78298950195312, -7.177778244018555, -86.16241455078125, -0.1008148193359375, 92.65060424804688, 88.004638671875, 94.3139877319336, 69.70967102050781, 42.672569274902344, 26.08954620361328, 111.78546905517578, -44.82581329345703, 152.77134704589844, 127.59881591796875, 82.08718872070312, 37.00257873535156, 14.89348030090332, 329.3406677246094, 95.32804107666016, 76.17451477050781, 28.48187828063965], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000174.npy"}
|
|
{"epoch": 0.2555066079295154, "step": 175, "batch_size": 64, "mean": 86.63121032714844, "std": 82.90579223632812, "min": -48.96678924560547, "p10": -1.7081892013549789, "median": 70.55760955810547, "p90": 199.93163452148443, "max": 328.7877197265625, "pos_frac": 0.875, "sample": [12.402120590209961, 29.84955596923828, 46.930030822753906, -40.11927795410156, 75.80865478515625, 99.30557250976562, 180.1953125, 65.00931549072266, 21.676721572875977, 127.88725280761719, 242.38352966308594, -42.38154220581055, -2.417816162109375, 240.5345458984375, 69.7398681640625, 28.33245849609375, 184.269775390625, 151.0569610595703, 152.10946655273438, 140.17230224609375, 133.560791015625, -30.46704864501953, 171.0220489501953, 328.7877197265625, 26.443954467773438, 90.92949676513672, 274.76861572265625, 44.03727722167969, 117.309814453125, 63.60743713378906, -2.9609222412109375, 41.227783203125, 231.11007690429688, 22.32046127319336, 12.795448303222656, 32.551353454589844, 0.2528266906738281, 206.64385986328125, 7.098775863647461, 56.231178283691406, 99.13796997070312, 96.66102600097656, 106.2833480834961, 168.01385498046875, -0.05239295959472656, 152.99673461914062, 175.25286865234375, 157.0886688232422, -23.663578033447266, 97.27019500732422, 136.3621826171875, 72.51217651367188, 71.37535095214844, 113.35547637939453, 34.33843231201172, 47.998775482177734, 103.87901306152344, 2.1482677459716797, 43.33873748779297, 216.2421875, -48.96678924560547, 55.19768524169922, 19.493499755859375, 38.11805725097656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000175.npy"}
|
|
{"epoch": 0.25697503671071953, "step": 176, "batch_size": 64, "mean": 71.29586791992188, "std": 87.8052978515625, "min": -45.61896514892578, "p10": -17.487634277343744, "median": 52.326541900634766, "p90": 185.63299560546878, "max": 377.07537841796875, "pos_frac": 0.75, "sample": [20.381927490234375, 20.2996826171875, 71.18109130859375, 137.51756286621094, 57.306583404541016, 15.334217071533203, 15.340103149414062, 222.21273803710938, 15.12725830078125, -4.17791748046875, -12.128173828125, 323.1188659667969, -19.81361961364746, -27.166305541992188, 187.31683349609375, 59.899261474609375, -6.647216796875, 193.3240203857422, 128.7637939453125, -1.0535354614257812, -1.122467041015625, 68.64417266845703, 106.08428955078125, -40.32203674316406, 181.70404052734375, -4.668306350708008, 21.897907257080078, 151.62103271484375, 130.25115966796875, -0.16622161865234375, 88.28605651855469, -39.570068359375, 100.96760559082031, -45.61896514892578, 30.873077392578125, 117.53421783447266, -20.447376251220703, 49.011375427246094, 117.83594512939453, 112.18582153320312, 133.73001098632812, 0.9993305206298828, -19.7845458984375, -0.007137298583984375, 130.0651397705078, 103.56954956054688, 111.31414794921875, 55.64170837402344, 118.82362365722656, 236.68264770507812, 41.73710632324219, -11.970672607421875, 28.003753662109375, 13.694433212280273, 110.39788818359375, 57.928863525390625, 0.41497039794921875, 23.134090423583984, 117.31078338623047, 6.523860931396484, 144.0869140625, 377.07537841796875, 22.201866149902344, 240.24314880371094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000176.npy"}
|
|
{"epoch": 0.25844346549192365, "step": 177, "batch_size": 64, "mean": 72.19395446777344, "std": 88.57938385009766, "min": -56.1005859375, "p10": -17.831476020812985, "median": 44.425018310546875, "p90": 185.6975524902344, "max": 432.2127380371094, "pos_frac": 0.765625, "sample": [114.11404418945312, 39.753814697265625, 153.4247283935547, 82.68901824951172, 51.839454650878906, 105.0191650390625, -38.76881408691406, 65.09440612792969, 164.68145751953125, 128.77291870117188, 188.9617156982422, 33.86644744873047, 77.69674682617188, 183.00393676757812, 98.45237731933594, 63.500972747802734, 20.09105682373047, 263.3757629394531, 16.499252319335938, 149.75836181640625, 26.532583236694336, 135.64588928222656, -53.288002014160156, 126.89633178710938, -2.6161880493164062, 33.22113037109375, 186.85195922851562, -19.2996826171875, -4.5605010986328125, 17.085006713867188, 44.91058349609375, 224.36610412597656, 103.65605163574219, -3.0090789794921875, 26.5450439453125, 43.939453125, 0.400299072265625, 210.042236328125, -1.4950542449951172, -22.28047752380371, -56.1005859375, 104.99472045898438, 0.9639625549316406, 41.48389434814453, -43.73529052734375, 118.99296569824219, 119.66395568847656, 7.304012298583984, -2.2019119262695312, -3.7044830322265625, 145.15879821777344, 22.92100715637207, 83.33157348632812, -15.431198120117188, 137.67669677734375, 27.651477813720703, 29.594263076782227, -18.860166549682617, 133.70559692382812, 20.510047912597656, 432.2127380371094, 230.04856872558594, 75.7345199584961, -6.872711181640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000177.npy"}
|
|
{"epoch": 0.2599118942731278, "step": 178, "batch_size": 64, "mean": 91.52853393554688, "std": 108.96728515625, "min": -122.26632690429688, "p10": -19.68857421874999, "median": 77.102294921875, "p90": 233.00981903076175, "max": 464.6838684082031, "pos_frac": 0.8125, "sample": [55.46549606323242, 5.934562683105469, 128.28176879882812, -0.33026123046875, 106.53311920166016, 72.6195297241211, 99.29718017578125, 29.71142578125, 21.748737335205078, 224.4149627685547, 4.614952087402344, 108.36192321777344, 194.36529541015625, 87.43763732910156, 44.11875915527344, 125.40147399902344, 323.0637512207031, 73.81942749023438, 464.6838684082031, -24.129257202148438, 166.55845642089844, 49.30278015136719, 6.660614013671875, 55.683616638183594, 20.730947494506836, 148.93746948242188, 249.35281372070312, -31.58130645751953, 50.7862548828125, -0.42398834228515625, -38.53770446777344, 82.20608520507812, 183.38497924804688, 23.359825134277344, 99.70703887939453, 168.10772705078125, 80.38516235351562, -50.95390319824219, 54.0885009765625, 193.61361694335938, 24.819625854492188, 96.97676086425781, -122.26632690429688, 347.0511779785156, -6.245424270629883, 126.90496826171875, 89.86512756347656, 126.5951156616211, -37.06023406982422, 360.4010925292969, 31.46225929260254, 106.70037078857422, 236.69332885742188, 3.649442672729492, 33.66961669921875, 149.2405242919922, 82.12931823730469, -9.326980590820312, 28.511337280273438, 175.0843963623047, 262.1465759277344, -61.465850830078125, 160.58340454101562, -5.0766143798828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000178.npy"}
|
|
{"epoch": 0.26138032305433184, "step": 179, "batch_size": 64, "mean": 68.92756652832031, "std": 101.30906677246094, "min": -185.9692840576172, "p10": -47.25114288330078, "median": 69.4088363647461, "p90": 216.00533447265627, "max": 279.19903564453125, "pos_frac": 0.6875, "sample": [87.69646453857422, 131.219970703125, 139.6746826171875, 279.19903564453125, 49.61219024658203, -60.77631378173828, 155.83876037597656, 31.755229949951172, -2.7975234985351562, 3.625265121459961, 87.22882080078125, 60.47462463378906, 30.37319564819336, 73.24886322021484, -8.930122375488281, 95.96211242675781, -38.85498046875, 251.21676635742188, -15.826335906982422, -23.256683349609375, 64.06510162353516, -52.89460372924805, 244.98776245117188, 21.26609992980957, 181.2942352294922, 111.60195922851562, 131.90682983398438, 232.81825256347656, 150.079833984375, 47.091163635253906, 72.8066177368164, 262.0574035644531, 252.36256408691406, 80.23133850097656, 80.22742462158203, 91.62620544433594, -24.885589599609375, -7.562030792236328, -0.41881561279296875, 71.95892333984375, 69.05713653564453, -27.57386016845703, 69.76053619384766, -48.76930236816406, 118.59490966796875, 209.73477172851562, -71.14705657958984, 161.17320251464844, 99.93950653076172, -49.294715881347656, -27.76734161376953, 184.0412139892578, 26.831192016601562, 175.55972290039062, 218.69271850585938, 188.66648864746094, -126.58380126953125, 137.44439697265625, -40.22093200683594, -43.708770751953125, -185.9692840576172, 47.847015380859375, 23.27875328063965, -35.52722930908203], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000179.npy"}
|
|
{"epoch": 0.26284875183553597, "step": 180, "batch_size": 64, "mean": 82.76273345947266, "std": 91.8753433227539, "min": -136.93368530273438, "p10": -12.653247451782224, "median": 79.66606140136719, "p90": 200.37100677490236, "max": 354.91796875, "pos_frac": 0.828125, "sample": [78.46734619140625, 136.99270629882812, -13.371932983398438, -3.8640403747558594, -62.81591033935547, 354.91796875, -10.976314544677734, 42.317657470703125, 89.61477661132812, 58.09126281738281, 3.3189563751220703, 173.22344970703125, -136.93368530273438, 128.67892456054688, 122.22837829589844, 82.7691650390625, 92.22798156738281, 80.86477661132812, 89.85294342041016, 37.97975158691406, 140.71688842773438, -78.24002075195312, 30.675098419189453, 223.35629272460938, 60.44702911376953, 195.66378784179688, 124.03423309326172, 16.562294006347656, 25.456926345825195, 38.66635513305664, 205.79676818847656, 4.758188247680664, 188.37960815429688, 130.09979248046875, 84.5536880493164, -5.596242904663086, 95.0674057006836, 163.84689331054688, 0.1335601806640625, 146.64443969726562, 66.63541412353516, 0.26882171630859375, 210.02511596679688, 49.93974304199219, -80.36642456054688, 58.92148971557617, 15.4044189453125, 187.4874725341797, 199.61265563964844, 52.836692810058594, 102.75540161132812, 200.69601440429688, -13.46760368347168, 265.51202392578125, 56.805015563964844, -4.65167236328125, 140.61395263671875, 93.25181579589844, 25.34374237060547, -47.31884002685547, 223.281494140625, 182.2887725830078, 21.124977111816406, 155.20742797851562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000180.npy"}
|
|
{"epoch": 0.2643171806167401, "step": 181, "batch_size": 64, "mean": 52.12580871582031, "std": 82.5654067993164, "min": -136.45596313476562, "p10": -40.52664260864257, "median": 39.141326904296875, "p90": 170.47627716064454, "max": 273.5484619140625, "pos_frac": 0.78125, "sample": [86.88591003417969, -30.47064208984375, 15.567964553833008, 53.11537551879883, 94.4778823852539, -5.431098937988281, 18.043941497802734, 36.29637145996094, 62.166900634765625, 87.57783508300781, 18.301288604736328, 27.770036697387695, 4.579229354858398, 10.392662048339844, -100.34475708007812, 166.96218872070312, 70.57321166992188, -8.739418029785156, 61.887054443359375, -1.0453910827636719, 35.11545944213867, -45.08690643310547, -11.152618408203125, 171.98231506347656, -44.83635711669922, 67.4507827758789, 84.50924682617188, 217.3856658935547, 69.3483657836914, 35.11753845214844, 103.36629486083984, 62.030052185058594, -120.75166320800781, 41.98628234863281, 24.262466430664062, 49.82143783569336, 10.67741584777832, 118.96434020996094, 125.24568176269531, -58.91914367675781, 25.933181762695312, -136.45596313476562, 182.47879028320312, -1.0588798522949219, 226.56053161621094, 129.2627410888672, -106.45753479003906, 55.96086120605469, -10.918983459472656, 102.65021514892578, 176.47793579101562, 273.5484619140625, 103.04435729980469, 26.521556854248047, 107.18533325195312, 2.5756263732910156, 3.5733070373535156, 3.364837646484375, 246.06675720214844, 51.3448600769043, 29.144813537597656, 114.63003540039062, 89.9857406616211, 35.54988098144531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000181.npy"}
|
|
{"epoch": 0.2657856093979442, "step": 182, "batch_size": 64, "mean": 86.55928802490234, "std": 97.99131774902344, "min": -170.52554321289062, "p10": -32.13698959350585, "median": 72.76244354248047, "p90": 213.38663024902345, "max": 282.853271484375, "pos_frac": 0.828125, "sample": [222.6387939453125, 61.935791015625, 235.94052124023438, 125.95516967773438, 68.26803588867188, 158.81756591796875, 179.9407958984375, 208.20645141601562, 282.853271484375, 69.70693969726562, 75.81794738769531, 42.56295394897461, 25.689210891723633, -12.27389144897461, 187.50436401367188, -170.52554321289062, 183.45306396484375, 38.935630798339844, 5.08673095703125, 31.332656860351562, 278.787109375, -2.8735294342041016, 210.2777099609375, 82.87339782714844, 108.32360076904297, -35.96461486816406, 193.53207397460938, 59.894535064697266, 14.036975860595703, 61.41941833496094, 125.0128173828125, 189.4351348876953, -50.90569305419922, -86.36341094970703, 115.48303985595703, 173.99911499023438, 46.376312255859375, 83.34331512451172, 149.15423583984375, 179.78106689453125, -37.58446502685547, 55.46673583984375, 33.975982666015625, 114.18177795410156, 100.12535095214844, 132.15951538085938, 129.70492553710938, 108.06051635742188, -64.560546875, 3.130523681640625, 275.8042907714844, 113.80821990966797, 52.444068908691406, 5.414529800415039, -23.20586395263672, 10.779518127441406, -96.40251922607422, 43.81324005126953, 214.71902465820312, 24.93187713623047, 252.64190673828125, -16.07486343383789, 55.873207092285156, 133.118408203125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000182.npy"}
|
|
{"epoch": 0.26725403817914833, "step": 183, "batch_size": 64, "mean": 50.97412109375, "std": 76.53619384765625, "min": -94.15888214111328, "p10": -40.842343139648435, "median": 45.492835998535156, "p90": 139.20699462890624, "max": 240.4326934814453, "pos_frac": 0.796875, "sample": [46.35789489746094, 63.49220275878906, 3.0813541412353516, 43.047607421875, 1.9301338195800781, 139.11505126953125, 113.94013977050781, 112.93502044677734, 47.149688720703125, 44.627777099609375, -29.971939086914062, 95.0557861328125, 44.60173416137695, 15.648721694946289, 11.845794677734375, 64.1092529296875, 240.4326934814453, 198.36056518554688, 118.15001678466797, 30.995996475219727, 48.35266876220703, 79.08375549316406, 211.68873596191406, 85.62863159179688, 1.6414108276367188, 2.8264389038085938, 59.016361236572266, 35.45490646362305, 13.263099670410156, -89.09263610839844, 133.46539306640625, 66.9129638671875, -35.92827606201172, 9.541938781738281, 111.30687713623047, 19.666336059570312, 41.95572280883789, 100.58386993408203, 117.66863250732422, 202.46929931640625, -70.25201416015625, -15.988929748535156, 236.00750732421875, -42.94837188720703, 36.96381378173828, -26.87188720703125, 79.73983764648438, -67.4822006225586, -67.3944320678711, -71.1562728881836, 139.24639892578125, 71.86048889160156, 148.1183624267578, 134.572265625, 48.0842399597168, 50.181800842285156, 19.345932006835938, -30.45507049560547, 88.3542709350586, 70.11595153808594, -94.15888214111328, -22.426288604736328, 16.565261840820312, 11.91058349609375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000183.npy"}
|
|
{"epoch": 0.2687224669603524, "step": 184, "batch_size": 64, "mean": 51.08043670654297, "std": 78.0625991821289, "min": -87.8362808227539, "p10": -22.12866821289062, "median": 27.04665756225586, "p90": 157.82004699707034, "max": 258.89874267578125, "pos_frac": 0.75, "sample": [14.268585205078125, -65.5882797241211, 51.00434112548828, 66.61640930175781, 151.2262725830078, 111.36643981933594, -7.110504150390625, 9.676002502441406, 112.10831451416016, -8.82159423828125, 249.96707153320312, 229.33856201171875, 46.03089141845703, 136.3343505859375, -13.559288024902344, 135.59475708007812, 25.96856689453125, 139.99942016601562, 81.48216247558594, 181.77769470214844, -7.331869125366211, 219.45706176757812, -45.879364013671875, 11.539459228515625, 22.219676971435547, 36.905494689941406, 34.94933319091797, 8.887779235839844, 81.86871337890625, 6.626577377319336, 41.71125793457031, 56.72918701171875, 110.72554016113281, 149.9904022216797, -61.91232681274414, 161.96487426757812, 11.199592590332031, -23.717918395996094, -12.95511245727539, 24.507335662841797, 75.15219116210938, -10.897655487060547, 10.071918487548828, 4.072793960571289, 28.12474822998047, 160.6459503173828, 45.82392883300781, 48.1825065612793, -18.42041778564453, -72.9840087890625, -39.51266098022461, 37.89448547363281, 6.622814178466797, 9.149589538574219, -87.8362808227539, 21.729598999023438, -1.2823638916015625, 141.2529296875, 23.784713745117188, 59.52691650390625, 16.836875915527344, -1.7103157043457031, 78.85519409179688, 258.89874267578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000184.npy"}
|
|
{"epoch": 0.2701908957415565, "step": 185, "batch_size": 64, "mean": 41.93297576904297, "std": 82.44960021972656, "min": -132.06195068359375, "p10": -38.63670959472656, "median": 23.66563892364502, "p90": 162.10013122558595, "max": 315.4991149902344, "pos_frac": 0.71875, "sample": [-117.93054962158203, -53.94796371459961, -3.197183609008789, 78.98204040527344, 59.6906852722168, 21.756181716918945, -5.715696334838867, 45.16448974609375, 253.27474975585938, 17.228256225585938, 12.63735580444336, 84.36040496826172, 81.28315734863281, 6.181514739990234, 174.44546508789062, 48.36927795410156, -9.730403900146484, 26.463302612304688, 90.18519592285156, -62.640869140625, -5.211889266967773, 37.02140808105469, 17.84640121459961, 52.70936584472656, 53.2169303894043, 14.040504455566406, -59.84504699707031, 25.575096130371094, 56.32923126220703, -69.32121276855469, 68.12155151367188, 7.799327850341797, 89.51038360595703, 73.84181213378906, 9.05126953125, 253.74497985839844, -39.20867156982422, 51.228614807128906, 154.6572265625, -15.28860092163086, 11.510242462158203, 13.388883590698242, -12.866401672363281, -8.356521606445312, -8.983213424682617, 9.630878448486328, 202.2547607421875, 121.04058837890625, 119.7974853515625, 7.112913131713867, 315.4991149902344, 1.678863525390625, 30.343830108642578, 46.92766189575195, -7.91534423828125, 165.28994750976562, 28.506591796875, 97.66639709472656, -132.06195068359375, 174.855712890625, -37.30213165283203, 48.067665100097656, 5.3820037841796875, -30.435718536376953], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000185.npy"}
|
|
{"epoch": 0.27165932452276065, "step": 186, "batch_size": 64, "mean": 76.79740905761719, "std": 83.27879333496094, "min": -121.26593017578125, "p10": -4.673223686218261, "median": 62.49678611755371, "p90": 185.64752502441408, "max": 322.41119384765625, "pos_frac": 0.84375, "sample": [-121.26593017578125, 111.58203125, 222.87945556640625, 112.51480102539062, 110.06476593017578, 4.112335205078125, 59.600677490234375, 32.56303024291992, 179.329345703125, 132.5513916015625, 42.0501708984375, 171.290771484375, 6.961174011230469, 57.60691833496094, 38.35923385620117, -0.9540214538574219, 71.04873657226562, -71.95376586914062, 35.28509521484375, 19.435609817504883, 198.41680908203125, 34.4853401184082, 12.829872131347656, 125.82173156738281, 32.891746520996094, 61.40082931518555, 84.36308288574219, 118.3687973022461, 220.63894653320312, 64.13420104980469, 154.62750244140625, -3.4865970611572266, 123.91273498535156, 48.55743408203125, 33.122039794921875, 104.33778381347656, 288.45709228515625, 188.35531616210938, 56.29227828979492, -1.8747901916503906, 118.85497283935547, 23.493227005004883, 63.592742919921875, 46.892669677734375, -11.821502685546875, 304.24310302734375, -5.1817779541015625, 71.92536163330078, -12.866424560546875, -16.398971557617188, 322.41119384765625, 130.44094848632812, 56.62788391113281, 121.17375183105469, 17.796554565429688, 114.57410430908203, 68.169189453125, -25.86309051513672, 14.691299438476562, 84.26837158203125, 102.89695739746094, 87.04728698730469, 14.601287841796875, 64.75128173828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000186.npy"}
|
|
{"epoch": 0.27312775330396477, "step": 187, "batch_size": 64, "mean": 62.11200714111328, "std": 79.40628814697266, "min": -110.83611297607422, "p10": -24.421350669860836, "median": 51.17785453796387, "p90": 160.25939636230473, "max": 291.80841064453125, "pos_frac": 0.796875, "sample": [-110.83611297607422, 181.82192993164062, -10.374897003173828, 50.32295608520508, -62.84299850463867, 142.00613403320312, 67.94466400146484, 95.0712890625, 41.983909606933594, 49.98826217651367, 64.139404296875, 58.8836669921875, 242.9005584716797, 49.76886749267578, 76.45515441894531, 146.721435546875, 130.52879333496094, 33.58088684082031, 60.67314147949219, -35.44008255004883, 52.032752990722656, 22.261077880859375, 291.80841064453125, -11.6854248046875, 119.00126647949219, 47.914459228515625, 194.02120971679688, -27.122089385986328, 146.12533569335938, 27.388277053833008, -26.479217529296875, 80.93535614013672, 17.318626403808594, -42.85328674316406, -19.619661331176758, 144.81893920898438, -8.778007507324219, 10.047224044799805, -11.719276428222656, 151.19479370117188, 126.34297180175781, 267.7359619140625, 68.95279693603516, 22.815628051757812, 81.37957763671875, 29.537857055664062, 28.47354507446289, 31.124229431152344, -95.0468521118164, 42.08209991455078, 58.3531494140625, 0.44658660888671875, 97.78819274902344, 77.56312561035156, 70.82512664794922, 114.20362854003906, 68.74872589111328, 8.386283874511719, 52.8447265625, 189.14828491210938, 164.14422607421875, 30.825044631958008, -5.916341781616211, 14.502105712890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000187.npy"}
|
|
{"epoch": 0.2745961820851689, "step": 188, "batch_size": 64, "mean": 78.13198852539062, "std": 90.39881134033203, "min": -119.4709243774414, "p10": -9.102280998229979, "median": 54.61555099487305, "p90": 213.14136962890626, "max": 355.385009765625, "pos_frac": 0.84375, "sample": [231.81591796875, 91.24674224853516, 186.93199157714844, 37.666744232177734, 68.51946258544922, 15.654403686523438, -22.8712158203125, -69.4633560180664, 213.4539031982422, 80.49698638916016, 45.67259979248047, -16.99968719482422, 89.65030670166016, 68.10440063476562, 212.41212463378906, 179.6839141845703, 66.57665252685547, -13.824005126953125, 54.043052673339844, 6.280906677246094, 17.300918579101562, 17.086631774902344, 60.05964660644531, 113.37471771240234, 40.62598419189453, 32.99695587158203, 22.734893798828125, 3.377330780029297, 247.49356079101562, 27.49645233154297, 293.3074951171875, -119.4709243774414, 215.71539306640625, 96.70405578613281, 146.44937133789062, 95.36000061035156, 27.84619140625, 162.84814453125, -7.862705230712891, 156.18341064453125, 104.34158325195312, 31.231048583984375, -62.41143798828125, -6.371862411499023, 158.9492950439453, 37.13948059082031, 181.95228576660156, 34.13979721069336, 28.724502563476562, 248.22735595703125, 49.987403869628906, 137.59283447265625, 46.92606735229492, -9.633527755737305, 137.7943878173828, 56.40266418457031, 90.12895965576172, 355.385009765625, 80.64627075195312, 52.460601806640625, 55.18804931640625, -6.3385772705078125, 9.453834533691406, 13.852184295654297], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000188.npy"}
|
|
{"epoch": 0.27606461086637296, "step": 189, "batch_size": 64, "mean": 66.65364074707031, "std": 99.3444595336914, "min": -136.14895629882812, "p10": -32.911557579040526, "median": 43.85004997253418, "p90": 163.55956115722657, "max": 436.06396484375, "pos_frac": 0.78125, "sample": [29.39556884765625, 146.62535095214844, -83.40022277832031, 251.1668701171875, -31.415542602539062, -40.368309020996094, 39.41705322265625, -1.3473739624023438, -13.132211685180664, 52.38067626953125, 122.28338623046875, 131.3330078125, 117.09747314453125, 30.9361572265625, 132.84011840820312, 44.53956604003906, 130.87063598632812, 30.719436645507812, -92.66670227050781, 8.704339981079102, -31.710474014282227, 40.261268615722656, 67.42787170410156, 163.96524047851562, 129.6488494873047, 51.17889404296875, -37.20188903808594, 43.1605339050293, 25.844562530517578, 162.00827026367188, 4.557613372802734, 24.377836227416992, 88.15573120117188, 350.4051513671875, 26.481552124023438, -4.7744293212890625, 15.068496704101562, 209.0216827392578, 69.74479675292969, 23.24726104736328, -101.45689392089844, 91.19559478759766, -18.173355102539062, 72.14122772216797, 39.3291015625, 275.0819091796875, 164.0310821533203, 436.06396484375, -9.450241088867188, 102.19172668457031, 150.394775390625, 25.518686294555664, 59.75481033325195, 119.52345275878906, 66.74108123779297, 130.41067504882812, 162.61297607421875, 76.0438232421875, 11.863304138183594, -136.14895629882812, 24.524673461914062, -33.426307678222656, 11.533493041992188, 118.68437194824219], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000189.npy"}
|
|
{"epoch": 0.2775330396475771, "step": 190, "batch_size": 64, "mean": 75.96780395507812, "std": 105.48849487304688, "min": -149.4563751220703, "p10": -34.29169845581054, "median": 60.300045013427734, "p90": 225.05569000244142, "max": 326.07623291015625, "pos_frac": 0.796875, "sample": [-2.0547637939453125, 92.51237487792969, 127.01296997070312, 82.66265106201172, 10.555442810058594, -62.476776123046875, 183.01385498046875, 34.59027862548828, 10.754627227783203, 247.64630126953125, 56.5521240234375, 43.804847717285156, 96.47945404052734, 29.541976928710938, 306.1495361328125, 107.74778747558594, 79.60470581054688, 81.84101867675781, 159.3736572265625, 35.79541015625, 2.1903133392333984, 225.6307373046875, -149.4563751220703, 64.04796600341797, 218.8153076171875, -11.628631591796875, 41.65283203125, 245.69322204589844, 163.45411682128906, 64.04959106445312, -15.471115112304688, 223.7139129638672, 13.689682006835938, -6.535547256469727, 96.0926742553711, 75.95649719238281, -56.71184539794922, -26.257308959960938, -129.0155029296875, 25.237930297851562, 104.50174713134766, -37.735008239746094, 90.42887115478516, 38.90669250488281, -21.820165634155273, -60.589630126953125, 107.06161499023438, 209.76513671875, 11.425018310546875, 52.84491729736328, -141.84120178222656, 295.6387939453125, 83.6993408203125, 29.907052993774414, 168.59817504882812, 101.75170135498047, 54.752113342285156, 326.07623291015625, 3.1056346893310547, 296.8667297363281, 111.98069763183594, 182.84568786621094, 30.456722259521484, 37.056610107421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000190.npy"}
|
|
{"epoch": 0.2790014684287812, "step": 191, "batch_size": 64, "mean": 110.20065307617188, "std": 118.65868377685547, "min": -94.22189331054688, "p10": 3.1875776290893567, "median": 92.03886413574219, "p90": 275.81493530273445, "max": 453.7021179199219, "pos_frac": 0.921875, "sample": [324.7238464355469, 79.69929504394531, 453.0314636230469, 21.54443359375, 215.00115966796875, 72.30970764160156, 129.6874542236328, 214.27322387695312, 7.264341354370117, 298.7929992675781, 90.31924438476562, 110.43767547607422, 17.458091735839844, 204.94866943359375, 232.9820556640625, 10.784431457519531, 30.414995193481445, 67.24903106689453, 21.320755004882812, 67.60340118408203, 190.12669372558594, 51.67529296875, 35.415618896484375, 228.8314208984375, 100.93956756591797, 20.629531860351562, 4.681161880493164, 439.55029296875, 97.73258972167969, 7.140861511230469, 171.09548950195312, -65.48226928710938, 16.11480712890625, -30.42300033569336, 102.62948608398438, 138.78436279296875, 123.84210205078125, 115.22434997558594, 286.19036865234375, 25.412397384643555, 1.3585758209228516, 177.46563720703125, 188.58181762695312, -94.22189331054688, 36.50788879394531, 453.7021179199219, 50.492034912109375, -38.35810852050781, 93.75848388671875, 105.46479797363281, 135.08203125, 251.6055908203125, 30.046546936035156, 5.063518524169922, 106.83316040039062, 109.76715087890625, -16.92713165283203, 118.20481872558594, 2.5474700927734375, 86.18964385986328, 304.77984619140625, 49.334495544433594, 35.4527587890625, 130.1533203125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000191.npy"}
|
|
{"epoch": 0.28046989720998533, "step": 192, "batch_size": 64, "mean": 75.54448699951172, "std": 109.02069091796875, "min": -161.25686645507812, "p10": -30.57740097045898, "median": 51.553314208984375, "p90": 245.43007965087895, "max": 407.94964599609375, "pos_frac": 0.765625, "sample": [-6.686439514160156, -88.57595825195312, 49.72821044921875, 144.37937927246094, 129.1644744873047, -2.1293563842773438, -31.845619201660156, -6.538076400756836, 27.435501098632812, -161.25686645507812, 17.0518798828125, 71.76140594482422, -27.61822509765625, 173.70310974121094, 27.319198608398438, 29.778961181640625, -20.12420654296875, 79.62782287597656, -21.727874755859375, 125.11038208007812, -79.88297271728516, 318.7815856933594, -48.15019607543945, 65.55205535888672, -6.9608917236328125, 265.02532958984375, 275.24822998046875, 308.5657958984375, 81.36434936523438, 257.96771240234375, 72.83871459960938, 64.58880615234375, 407.94964599609375, 78.61239624023438, 133.95213317871094, 23.084718704223633, 109.25680541992188, 183.31211853027344, 193.97579956054688, 35.43436813354492, 50.60487365722656, 237.5514373779297, 19.15007972717285, 35.48582458496094, 113.58718872070312, 51.053443908691406, 52.053184509277344, -117.40855407714844, 142.75308227539062, 34.281593322753906, 73.91220092773438, 5.11517333984375, 8.955360412597656, -46.20378875732422, 13.337997436523438, 190.79019165039062, 71.16972351074219, -4.877891540527344, 13.770111083984375, 158.45626831054688, 121.38957977294922, 98.107666015625, 13.931289672851562, 248.806640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000192.npy"}
|
|
{"epoch": 0.28193832599118945, "step": 193, "batch_size": 64, "mean": 43.941375732421875, "std": 99.14176177978516, "min": -282.9678955078125, "p10": -50.02880477905273, "median": 47.98051452636719, "p90": 153.74323730468757, "max": 294.8065185546875, "pos_frac": 0.703125, "sample": [86.32079315185547, -19.319725036621094, 28.36817169189453, 119.76533508300781, -270.6556396484375, 117.9683837890625, 61.61705780029297, 55.20592498779297, -282.9678955078125, 225.5634765625, 62.74074935913086, -20.918493270874023, 136.93020629882812, -19.521495819091797, 34.89019775390625, 80.94691467285156, 69.0544204711914, 85.82568359375, -2.0825271606445312, 120.22265625, 2.8642311096191406, -1.9067764282226562, 22.987689971923828, 97.6526870727539, 23.886545181274414, 101.41181945800781, -44.60440444946289, 0.41469573974609375, 2.097320556640625, -32.846649169921875, -21.50525665283203, 163.0737762451172, -51.68132019042969, -46.172935485839844, 160.94882202148438, -86.38224029541016, 121.08428192138672, -41.67694854736328, 8.434114456176758, 55.99095916748047, 7.672431945800781, 98.4684066772461, 176.96331787109375, 103.68376922607422, 86.35281372070312, -112.30424499511719, 240.7741241455078, 86.08563995361328, 90.23495483398438, 104.63875579833984, -60.50975799560547, 80.35787963867188, 5.814441680908203, -37.214500427246094, -0.21124267578125, 113.39069366455078, 294.8065185546875, 110.20848083496094, -59.89906692504883, 18.33922576904297, 211.52801513671875, 40.755104064941406, 22.616744995117188, 85.67093658447266], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000193.npy"}
|
|
{"epoch": 0.2834067547723935, "step": 194, "batch_size": 64, "mean": 71.57523345947266, "std": 75.69204711914062, "min": -81.24632263183594, "p10": -20.677723693847657, "median": 56.83050537109375, "p90": 190.40935668945312, "max": 230.72364807128906, "pos_frac": 0.84375, "sample": [107.15608978271484, 5.765167236328125, 22.462575912475586, 22.677886962890625, 84.41931915283203, 163.92083740234375, 56.64524841308594, 160.57864379882812, 2.9554214477539062, 4.329833984375, 25.90850257873535, 203.07818603515625, -35.21464538574219, -21.013031005859375, 35.594093322753906, 32.02137756347656, 163.962646484375, 43.51116180419922, 62.81290054321289, 111.6451187133789, 30.734256744384766, 74.58463287353516, 7.375232696533203, 6.553531646728516, 93.0856704711914, -81.24632263183594, -2.496072769165039, 224.54080200195312, 54.53411865234375, 79.98782348632812, 209.16258239746094, 112.98994445800781, 190.61781311035156, 121.94910430908203, 18.214004516601562, 230.72364807128906, 57.01576232910156, 24.417173385620117, 109.87794494628906, 38.56617736816406, 155.44786071777344, -29.976531982421875, 31.14896583557129, -2.6807632446289062, 89.0740966796875, 129.40859985351562, 14.78414535522461, -39.167694091796875, 27.656888961791992, -29.3155517578125, 101.66433715820312, -36.41448211669922, 48.405799865722656, 189.92295837402344, 51.93307876586914, 226.94088745117188, 104.7593994140625, 75.42703247070312, 206.29652404785156, 83.51526641845703, 146.6697998046875, 121.8531723022461, 78.95118713378906, -19.895339965820312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000194.npy"}
|
|
{"epoch": 0.28487518355359764, "step": 195, "batch_size": 64, "mean": 74.96640014648438, "std": 83.07173919677734, "min": -85.59515380859375, "p10": -28.965104675292963, "median": 68.36989212036133, "p90": 175.5665985107422, "max": 309.372314453125, "pos_frac": 0.828125, "sample": [70.72725677490234, 28.857666015625, 171.70602416992188, 55.650447845458984, 69.59803771972656, 297.4441833496094, 36.596168518066406, 274.3753662109375, 127.12164306640625, -1.245626449584961, 21.007965087890625, 37.40260314941406, 1.6062545776367188, -85.59515380859375, 142.9180908203125, 81.58009338378906, 67.75125122070312, 189.2960662841797, -40.07103729248047, 245.04493713378906, -37.05540084838867, 100.07606506347656, -73.28765869140625, 19.707489013671875, 94.79974365234375, 80.24292755126953, 127.10116577148438, 198.3367156982422, 117.67344665527344, 131.36721801757812, 100.3795394897461, -51.617156982421875, 60.146461486816406, 143.32809448242188, 37.630950927734375, 112.98330688476562, -10.728044509887695, 20.87364959716797, 120.06570434570312, 158.5051727294922, -24.095054626464844, 177.22113037109375, 38.896488189697266, 35.02562713623047, 309.372314453125, 109.17137908935547, 68.98853302001953, 59.93257141113281, -31.052268981933594, 23.95662498474121, 110.86212158203125, -13.067138671875, 141.98114013671875, 85.72122192382812, 55.52497863769531, 32.193603515625, 25.77112579345703, 45.51715850830078, 17.84266471862793, 79.17511749267578, -34.40283966064453, 48.979427337646484, 111.41795349121094, 80.6142578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000195.npy"}
|
|
{"epoch": 0.28634361233480177, "step": 196, "batch_size": 64, "mean": 84.77229309082031, "std": 100.75812530517578, "min": -147.9295654296875, "p10": -26.991283798217772, "median": 64.45485305786133, "p90": 219.14822082519532, "max": 317.94586181640625, "pos_frac": 0.796875, "sample": [166.25643920898438, 162.95103454589844, 189.56503295898438, 162.11996459960938, -51.141258239746094, 188.7883758544922, 215.82484436035156, 306.45892333984375, 175.017578125, 59.30735778808594, 5.281711578369141, 52.78040313720703, 215.10281372070312, 184.52227783203125, 46.12554931640625, 14.913164138793945, -16.869991302490234, 65.12672424316406, 4.311943054199219, 66.4130859375, 293.36260986328125, -17.66793441772461, 55.13490295410156, 32.4705924987793, 97.35914611816406, 88.54596710205078, 104.85379791259766, 126.9942398071289, 31.530502319335938, -3.2855167388916016, 126.68011474609375, 68.31471252441406, 313.18505859375, 317.94586181640625, -13.952545166015625, 122.05865478515625, 71.01394653320312, 55.942474365234375, 75.52719116210938, 47.26731872558594, 48.4140625, 63.782981872558594, 194.4017333984375, -1.842041015625, 98.40279388427734, 48.11232376098633, 54.702003479003906, -94.33114624023438, 56.168914794921875, -147.9295654296875, -45.093536376953125, 119.07769012451172, 66.16372680664062, -27.93710708618164, 220.57252502441406, -40.402427673339844, -44.39280700683594, 54.119606018066406, 22.135513305664062, 115.1335220336914, 227.44784545898438, 251.17262268066406, -24.78436279296875, 6.194520950317383], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000196.npy"}
|
|
{"epoch": 0.2878120411160059, "step": 197, "batch_size": 64, "mean": 66.4241714477539, "std": 80.31172180175781, "min": -81.98101806640625, "p10": -20.995037078857404, "median": 49.99422836303711, "p90": 178.02629089355474, "max": 290.3178405761719, "pos_frac": 0.859375, "sample": [34.53388977050781, 270.7517395019531, 105.79513549804688, 7.3017730712890625, 114.06346130371094, 5.075595855712891, 31.64605712890625, -39.96249771118164, 52.119895935058594, 290.3178405761719, 279.121826171875, 139.39044189453125, 62.785247802734375, 76.65185546875, -61.866600036621094, -3.4580001831054688, 68.7210693359375, 226.47682189941406, 31.91149139404297, 84.84577941894531, 92.31179809570312, 148.274169921875, 47.868560791015625, 8.707916259765625, 54.716575622558594, 8.695322036743164, -64.38702392578125, 91.14797973632812, 85.50879669189453, 95.87513732910156, 205.25775146484375, 100.23220825195312, 38.044776916503906, 185.39797973632812, 1.905975341796875, 119.14167785644531, 33.36480712890625, 2.009033203125, -29.86669921875, -81.98101806640625, 24.581350326538086, 100.66484832763672, -0.10817146301269531, 21.42462921142578, 1.937795639038086, 46.44956970214844, 186.5347137451172, 128.96395874023438, 45.8892822265625, 149.75279235839844, 0.32250022888183594, 47.69834518432617, 89.85128021240234, 19.750343322753906, 13.236015319824219, 80.19095611572266, 112.26731872558594, 63.972686767578125, 160.82568359375, -28.510910034179688, -35.29032897949219, 23.88127899169922, 18.985809326171875, 59.42668151855469], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000197.npy"}
|
|
{"epoch": 0.28928046989721, "step": 198, "batch_size": 64, "mean": 79.5688705444336, "std": 102.76666259765625, "min": -108.0997314453125, "p10": -6.34443302154541, "median": 67.69969177246094, "p90": 166.41469573974612, "max": 458.6191711425781, "pos_frac": 0.859375, "sample": [-101.9778060913086, 42.01445007324219, -6.422460556030273, 387.3268127441406, 92.4678726196289, 153.55810546875, 217.893798828125, 40.52861404418945, 87.26774597167969, 66.5975570678711, 118.04556274414062, 109.93496704101562, -93.01992797851562, 34.688148498535156, 15.12567138671875, 159.2953338623047, 102.8614273071289, 77.84243774414062, 68.89346313476562, 74.49903869628906, -13.323095321655273, -6.1623687744140625, 33.21202087402344, 83.68177795410156, 11.207223892211914, 49.48143768310547, -45.309120178222656, 169.46585083007812, 83.00556945800781, 88.00386047363281, 83.30828094482422, 68.80182647705078, 65.82896423339844, 42.59577941894531, 85.71061706542969, 2.6441917419433594, 50.67881393432617, 130.75030517578125, 101.27325439453125, 156.2779083251953, -108.0997314453125, 389.7027587890625, 52.24536895751953, 51.888755798339844, -3.292224884033203, 4.4761199951171875, 195.93630981445312, -15.730096817016602, 5.749300003051758, 54.321746826171875, 14.984615325927734, 61.6173095703125, 69.3623046875, 108.73481750488281, 13.868154525756836, 128.04217529296875, 111.88088989257812, 146.10198974609375, 7.485004425048828, 312.341064453125, 3.2241744995117188, 458.6191711425781, 40.28443908691406, 100.10928344726562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000198.npy"}
|
|
{"epoch": 0.2907488986784141, "step": 199, "batch_size": 64, "mean": 76.12931060791016, "std": 96.7540512084961, "min": -122.57451629638672, "p10": -42.48816070556641, "median": 62.19293975830078, "p90": 201.64264373779298, "max": 304.80023193359375, "pos_frac": 0.796875, "sample": [136.97512817382812, 131.10992431640625, -73.87814331054688, -2.4686813354492188, 200.78549194335938, 58.07542419433594, -41.780609130859375, 166.5910186767578, -122.57451629638672, 106.30756378173828, 7.426856994628906, -42.79139709472656, -103.28761291503906, 154.6805419921875, 170.0687255859375, 73.03304290771484, 9.292259216308594, 106.25772857666016, 258.98040771484375, 32.349327087402344, 53.416015625, 136.39710998535156, 270.1945495605469, 202.00999450683594, -66.98330688476562, 22.90740966796875, -10.957099914550781, 77.91515350341797, 164.54586791992188, 32.05621337890625, 26.18536376953125, 194.5310821533203, 276.29815673828125, 60.932861328125, 44.220611572265625, 176.55697631835938, -19.186141967773438, 91.8314208984375, 13.019744873046875, 93.48870849609375, 34.17994689941406, 161.323974609375, 110.12285614013672, 22.617496490478516, 212.11080932617188, 73.84026336669922, 46.59787368774414, -43.2452392578125, -21.33951187133789, -107.97465515136719, 168.11392211914062, 51.93049621582031, 63.45301818847656, 85.65216064453125, 240.6841278076172, 103.29072570800781, 7.076929092407227, 73.35751342773438, 32.334503173828125, 41.940895080566406, 34.296630859375, 122.96129608154297, 304.80023193359375, -10.383407592773438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000199.npy"}
|
|
{"epoch": 0.2922173274596182, "step": 200, "batch_size": 64, "mean": 80.00682830810547, "std": 106.65370178222656, "min": -194.26239013671875, "p10": -20.1780704498291, "median": 55.3380126953125, "p90": 234.39905090332041, "max": 414.55615234375, "pos_frac": 0.8125, "sample": [185.76150512695312, 109.2723388671875, 54.034603118896484, 243.81234741210938, 75.53738403320312, 116.49768829345703, 12.4400634765625, 4.7357025146484375, 298.5606689453125, -57.84077835083008, 251.27523803710938, 172.559814453125, 80.84503173828125, -8.658931732177734, 46.2607536315918, 212.4346923828125, 166.57391357421875, 260.00390625, 15.397964477539062, 182.76754760742188, 51.67040252685547, 15.935552597045898, 123.33343505859375, 160.1331329345703, -12.508686065673828, 66.91179656982422, 140.7334747314453, -76.01141357421875, 414.55615234375, 42.578208923339844, 83.03987121582031, -18.203571319580078, 5.408180236816406, 56.00947570800781, 43.24627685546875, -3.9321422576904297, -194.26239013671875, 0.99163818359375, 97.68565368652344, 277.2297058105469, 13.600187301635742, 76.72905731201172, 39.080413818359375, 31.12201690673828, 1.2819938659667969, -99.5633544921875, 58.96058654785156, -21.02428436279297, 197.0968017578125, -14.830068588256836, 11.987258911132812, 274.04217529296875, 45.84747314453125, 20.78020477294922, 52.828369140625, 173.11744689941406, 131.22802734375, 138.48275756835938, 54.66654968261719, 103.5107421875, -80.05853271484375, 121.94880676269531, -22.02190399169922, 114.8080062866211], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000200.npy"}
|
|
{"epoch": 0.2936857562408223, "step": 201, "batch_size": 64, "mean": 80.59831237792969, "std": 103.9575424194336, "min": -167.04449462890625, "p10": -31.192222595214837, "median": 69.65281677246094, "p90": 201.92131500244142, "max": 382.98333740234375, "pos_frac": 0.828125, "sample": [194.05364990234375, 39.11051940917969, 157.9179229736328, -146.687255859375, 40.60408020019531, -34.519317626953125, 18.848918914794922, 12.270116806030273, -12.827804565429688, 50.320068359375, 0.645721435546875, 30.413299560546875, 24.699066162109375, 3.2744140625, 308.7262878417969, 87.04345703125, 52.573577880859375, 82.94401550292969, 25.717323303222656, 202.16928100585938, 274.3872985839844, 119.07860565185547, 16.5714111328125, 69.98773193359375, 124.43173217773438, 250.482421875, -52.80940628051758, 143.2958221435547, 131.87738037109375, -5.818670272827148, 74.73007202148438, 104.26889038085938, 107.68025207519531, 110.93702697753906, -23.429000854492188, -51.34621810913086, 163.72250366210938, 41.245262145996094, 44.091529846191406, 382.98333740234375, 79.1553955078125, -8.699607849121094, 76.00379180908203, 55.04601287841797, 34.72337341308594, 136.54238891601562, 83.44742584228516, 168.83668518066406, 287.13970947265625, 69.31790161132812, 79.60327911376953, -41.06712341308594, 94.01344299316406, 68.71760559082031, -167.04449462890625, 201.3427276611328, 330.5429992675781, 109.6539077758789, 20.47269630432129, -50.07261657714844, 183.81015014648438, 25.04143524169922, 20.057907104492188, 138.0421142578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000201.npy"}
|
|
{"epoch": 0.29515418502202645, "step": 202, "batch_size": 64, "mean": 56.86698532104492, "std": 99.45360565185547, "min": -147.0736541748047, "p10": -60.16089172363281, "median": 52.003196716308594, "p90": 199.64144287109377, "max": 347.9344482421875, "pos_frac": 0.703125, "sample": [347.9344482421875, -28.90436553955078, 81.93904876708984, 209.14479064941406, 3.9002857208251953, -5.576316833496094, 153.00094604492188, -21.592348098754883, 104.22114562988281, 68.46875, 33.42852020263672, -130.4746856689453, 122.45084381103516, 195.44601440429688, 140.4920654296875, 49.91454315185547, -147.0736541748047, 30.1923828125, 82.25247192382812, 243.99327087402344, 62.771522521972656, 211.51930236816406, -81.45439147949219, 129.36764526367188, 19.345970153808594, 24.170310974121094, 289.32025146484375, -5.558677673339844, 78.30593872070312, 130.52732849121094, 63.099334716796875, 118.5703125, 200.767578125, -9.575210571289062, -21.701065063476562, -62.85687255859375, 61.12907409667969, 51.20603942871094, 35.18321990966797, 170.54501342773438, 86.07356262207031, -13.263473510742188, 18.47265625, 209.66571044921875, 86.69589233398438, 92.63954162597656, 67.15695190429688, -134.87733459472656, 54.782623291015625, 39.139190673828125, 11.666309356689453, 33.149925231933594, -53.870269775390625, -45.183998107910156, 20.894180297851562, -71.64851379394531, 61.67463684082031, -45.543128967285156, -32.19206619262695, 52.80035400390625, 197.0137939453125, -68.07072448730469, 78.56005096435547, -4.089641571044922], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000202.npy"}
|
|
{"epoch": 0.2966226138032305, "step": 203, "batch_size": 64, "mean": 69.2591781616211, "std": 85.10139465332031, "min": -87.51575469970703, "p10": -21.879294013977052, "median": 52.31894111633301, "p90": 178.83851470947272, "max": 334.39154052734375, "pos_frac": 0.796875, "sample": [4.867160797119141, 100.42706298828125, 132.92620849609375, 43.35369110107422, -21.975383758544922, 230.81344604492188, -43.551387786865234, 84.16937255859375, 30.033477783203125, 84.2650146484375, 3.0789642333984375, 91.7554702758789, 89.1944808959961, 99.44440460205078, -67.49411010742188, 132.8845672607422, 61.38389587402344, -6.3670501708984375, 138.2591552734375, 89.97669982910156, 220.93438720703125, 282.364990234375, -3.692768096923828, 76.02041625976562, 107.19762420654297, 41.94291687011719, 334.39154052734375, 14.685550689697266, -63.76576232910156, 97.24258422851562, -10.836875915527344, 129.40231323242188, 111.54313659667969, 207.33740234375, 31.7764892578125, 131.39218139648438, 79.10477447509766, 82.30812072753906, 42.277828216552734, 63.840293884277344, 31.618209838867188, 52.228172302246094, 38.43632507324219, 162.96624755859375, -18.32476043701172, -87.51575469970703, -29.0257568359375, 52.40970993041992, -21.65508460998535, 12.667278289794922, 134.96978759765625, 15.312522888183594, 24.447681427001953, 39.689918518066406, -16.511444091796875, -43.39276123046875, 110.9719467163086, 44.80924987792969, 162.04763793945312, 251.4090576171875, 2.8579769134521484, 185.6409149169922, 48.34850311279297, 25.269561767578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000203.npy"}
|
|
{"epoch": 0.29809104258443464, "step": 204, "batch_size": 64, "mean": 66.98109436035156, "std": 101.72553253173828, "min": -104.65696716308594, "p10": -35.51360549926757, "median": 54.96611213684082, "p90": 193.3239959716797, "max": 380.55169677734375, "pos_frac": 0.75, "sample": [125.24943542480469, 65.29560852050781, 28.411373138427734, 38.25733184814453, 80.0708236694336, 15.424751281738281, 153.624267578125, -75.38372039794922, 77.25106811523438, 22.396982192993164, -30.339801788330078, 23.30039405822754, 10.686233520507812, 69.59159851074219, 54.013763427734375, -14.708755493164062, -77.97787475585938, 113.54633331298828, 38.232513427734375, 16.4855899810791, 193.5969696044922, -80.43236541748047, -17.963912963867188, -15.705501556396484, 58.040191650390625, -7.676790237426758, 56.74128723144531, 127.45588684082031, 144.66119384765625, -30.823143005371094, 340.6988830566406, 7.293601989746094, 29.688262939453125, 24.83095359802246, -20.674453735351562, 120.24168395996094, 50.16883850097656, 105.0005874633789, 328.67645263671875, -104.65696716308594, 192.6870574951172, 135.5436248779297, 99.36857604980469, 55.918460845947266, 66.04660034179688, 2.0742931365966797, 36.0169677734375, 154.3821563720703, 244.904296875, -7.0766754150390625, 66.9384765625, 135.71212768554688, -78.29620361328125, 67.3648910522461, 380.55169677734375, 255.6309814453125, -17.92938804626465, 114.38570404052734, 28.198877334594727, 249.51730346679688, -71.93353271484375, 67.73703002929688, 103.98077392578125, -37.5238037109375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000204.npy"}
|
|
{"epoch": 0.29955947136563876, "step": 205, "batch_size": 64, "mean": 86.33795166015625, "std": 104.87971496582031, "min": -163.7601776123047, "p10": -46.39562034606933, "median": 90.74582290649414, "p90": 217.6935684204102, "max": 309.9691467285156, "pos_frac": 0.796875, "sample": [30.26715087890625, 95.52018737792969, 134.52969360351562, 309.9691467285156, 77.5387954711914, 18.563684463500977, -3.7482128143310547, 5.407073974609375, 28.869651794433594, 172.29388427734375, 190.77125549316406, -77.6507568359375, -70.45433044433594, 86.74210357666016, -8.827629089355469, 48.02833557128906, 14.018959045410156, 282.932861328125, 150.11207580566406, 28.723133087158203, 98.39368438720703, 61.48426818847656, -88.49786376953125, 127.86787414550781, 221.82901000976562, 224.55453491210938, 157.69100952148438, 10.164630889892578, -48.71869659423828, 191.19448852539062, -37.90314865112305, 296.1468811035156, 80.53668212890625, 59.634979248046875, 32.35565948486328, 93.64405822753906, 269.3705139160156, 182.03826904296875, 94.37857055664062, 93.68852233886719, -149.77044677734375, 126.69338989257812, 119.71849822998047, 173.21731567382812, 108.51947021484375, -40.9751091003418, 113.12904357910156, 69.66759490966797, 200.32958984375, 123.94824981689453, 208.04420471191406, 171.15335083007812, 144.59915161132812, 64.23574829101562, 123.24996948242188, 259.5190124511719, 87.84758758544922, -10.745658874511719, 6.6924285888671875, -163.7601776123047, -56.83418273925781, 185.75010681152344, -25.192535400390625, 53.1312255859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000205.npy"}
|
|
{"epoch": 0.3010279001468429, "step": 206, "batch_size": 64, "mean": 93.97215270996094, "std": 99.19384765625, "min": -155.0047607421875, "p10": -16.829483032226562, "median": 84.80393981933594, "p90": 223.23864135742198, "max": 332.30706787109375, "pos_frac": 0.828125, "sample": [64.44551086425781, 144.3482666015625, 65.87379455566406, 86.37611389160156, 332.30706787109375, 248.93099975585938, 84.07754516601562, 37.788108825683594, 186.16697692871094, 191.4549102783203, 51.37324523925781, 69.13731384277344, -1.2931041717529297, 67.62891387939453, -155.0047607421875, 84.80337524414062, 69.85415649414062, 99.65684509277344, 53.02544403076172, -16.084684371948242, 148.99993896484375, -17.14325714111328, 187.63809204101562, 280.7300720214844, 107.68545532226562, 158.35040283203125, 26.13823699951172, 165.59803771972656, -87.7674560546875, -85.05286407470703, 171.9891357421875, 233.5469207763672, 157.6165771484375, 155.9166717529297, 21.912797927856445, 190.166015625, 50.2926025390625, -57.75214767456055, 148.54666137695312, -76.32391357421875, 30.383155822753906, 176.79934692382812, 11.465030670166016, 121.29847717285156, 59.59095001220703, -10.906539916992188, 84.80450439453125, 68.4552230834961, 145.6523895263672, 72.05023193359375, -41.09002685546875, -16.09734344482422, 93.34449005126953, 136.48876953125, 311.7593994140625, 199.1859893798828, 3.021566390991211, 112.05615234375, 97.53402709960938, 249.99539184570312, 48.84528350830078, 17.13237953186035, 242.415283203125, 154.0797119140625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000206.npy"}
|
|
{"epoch": 0.302496328928047, "step": 207, "batch_size": 64, "mean": 107.59861755371094, "std": 117.97628021240234, "min": -70.26038360595703, "p10": -7.911489105224598, "median": 69.1389389038086, "p90": 239.91858673095703, "max": 451.5445556640625, "pos_frac": 0.890625, "sample": [12.163677215576172, -70.26038360595703, 43.76353454589844, 55.30088424682617, 126.15930938720703, 46.947174072265625, 240.7493133544922, 3.2133102416992188, 14.5198974609375, 11.433563232421875, 69.67977905273438, 60.264991760253906, 215.34225463867188, -12.67926025390625, 86.16204071044922, 127.96733856201172, 40.48150634765625, 117.765625, 30.697643280029297, 22.197813034057617, 34.50153732299805, 135.98910522460938, 28.34136199951172, 203.93202209472656, 18.344345092773438, -43.3638916015625, 237.980224609375, 48.249603271484375, 19.598194122314453, -24.131000518798828, 234.44229125976562, 209.8968048095703, -31.912277221679688, 3.291351318359375, 37.45771026611328, 205.06280517578125, 30.724960327148438, 193.5742950439453, 83.37576293945312, 157.8445587158203, 208.61334228515625, 110.01911926269531, 79.37406921386719, 135.44151306152344, 180.81475830078125, 180.0093994140625, 64.64909362792969, 309.46728515625, 395.4091491699219, 201.48780822753906, 41.399261474609375, 352.82666015625, -22.122404098510742, 58.65699768066406, 451.5445556640625, 41.332275390625, 68.59809875488281, 74.36650085449219, 71.51406860351562, 420.64019775390625, -70.07343292236328, 182.5491943359375, 320.8539733886719, 3.870391845703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000207.npy"}
|
|
{"epoch": 0.3039647577092511, "step": 208, "batch_size": 64, "mean": 82.68034362792969, "std": 111.11237335205078, "min": -218.97442626953125, "p10": -26.07698669433593, "median": 66.01898193359375, "p90": 226.1193374633789, "max": 360.92364501953125, "pos_frac": 0.84375, "sample": [-132.97561645507812, 98.42030334472656, -94.83345031738281, 195.6678924560547, 25.040374755859375, 1.3878593444824219, 258.07672119140625, 11.36262321472168, 91.05989074707031, 184.83059692382812, 174.72337341308594, 50.476768493652344, 76.6263427734375, 142.34341430664062, 149.76840209960938, 63.186683654785156, 360.92364501953125, 29.4437255859375, 321.4640808105469, 106.32437133789062, 25.729766845703125, 17.316917419433594, 175.30941772460938, 126.54261016845703, 71.05397033691406, 30.58525848388672, 35.279937744140625, 20.231109619140625, -114.92707824707031, 200.83294677734375, -103.05845642089844, 21.198589324951172, 35.79117202758789, 72.0353012084961, 110.9337387084961, 223.87184143066406, 179.81881713867188, 39.2437858581543, -2.2554855346679688, 166.41212463378906, -5.738067626953125, 32.492095947265625, 17.35601806640625, 227.08255004882812, 68.85128021240234, 274.3774108886719, 76.04969024658203, -218.97442626953125, -30.70954132080078, 94.936279296875, 8.299957275390625, 58.57707214355469, 253.76705932617188, 81.70576477050781, 53.51385498046875, -29.578887939453125, 19.738006591796875, 297.7713623046875, 180.9210205078125, 195.10292053222656, 156.61322021484375, 38.44375228881836, 13.584831237792969, -17.9058837890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000208.npy"}
|
|
{"epoch": 0.3054331864904552, "step": 209, "batch_size": 64, "mean": 73.58404541015625, "std": 102.48482513427734, "min": -205.5150146484375, "p10": -41.610701370239255, "median": 59.84882354736328, "p90": 228.0842056274414, "max": 293.804443359375, "pos_frac": 0.75, "sample": [273.22576904296875, 66.0372314453125, 123.57133483886719, -3.7227554321289062, 285.95635986328125, 279.5630798339844, 239.78961181640625, 60.81333923339844, 99.20565032958984, 58.884307861328125, -38.6479606628418, 70.85381317138672, 136.63674926757812, 55.33081817626953, 137.48146057128906, 84.90343475341797, 142.19224548339844, -98.1209716796875, -21.273223876953125, 91.99185943603516, 164.31512451171875, 166.77484130859375, 6.485614776611328, -76.4066162109375, 56.98292541503906, -28.870140075683594, -9.296884536743164, 42.78052520751953, 54.091644287109375, 136.36424255371094, 84.33045959472656, 70.67809295654297, -86.88062286376953, 1.5576744079589844, 55.17504119873047, 43.55230712890625, -57.0130500793457, -34.909969329833984, 293.804443359375, 186.89454650878906, 34.81065368652344, 251.005859375, 17.250160217285156, -6.523948669433594, -19.185012817382812, 29.74024200439453, 76.70109558105469, 56.64168167114258, 143.32870483398438, 105.87156677246094, 104.95523071289062, -12.04155158996582, -42.88044738769531, -205.5150146484375, 142.97723388671875, 57.0797119140625, 41.47269058227539, 42.315147399902344, 93.34713745117188, -86.70590209960938, 229.19288635253906, 225.49728393554688, 162.52825927734375, 152.43270874023438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000209.npy"}
|
|
{"epoch": 0.3069016152716593, "step": 210, "batch_size": 64, "mean": 80.08528137207031, "std": 97.43148803710938, "min": -144.17626953125, "p10": -10.954869079589841, "median": 56.49946212768555, "p90": 236.2073059082032, "max": 329.18280029296875, "pos_frac": 0.78125, "sample": [-15.170166015625, 12.381433486938477, 109.20852661132812, 55.657745361328125, 7.5875091552734375, -6.249301910400391, 147.22012329101562, -0.05903053283691406, 95.47952270507812, -23.322471618652344, 136.66290283203125, -1.1913833618164062, 34.152706146240234, 43.076499938964844, 216.63441467285156, 119.3987808227539, -4.233253479003906, 61.06672668457031, 105.57711029052734, 7.076679229736328, 208.29591369628906, 84.01995086669922, 33.03167724609375, 146.89891052246094, 128.3479461669922, 2.600130081176758, -7.791015625, 195.75741577148438, 148.38827514648438, 244.59568786621094, 78.5679702758789, 179.0525360107422, 89.91980743408203, 262.95513916015625, -63.30738830566406, -12.310806274414062, -36.61629867553711, 16.326461791992188, 24.74412727355957, 14.56466293334961, 116.03145599365234, 115.47537231445312, 264.76409912109375, 54.820865631103516, 11.877313613891602, -5.95745849609375, 267.4512939453125, 1.8082962036132812, 83.05941772460938, 94.32598114013672, 208.3152618408203, 329.18280029296875, 198.2041015625, 62.05107879638672, -16.52899169921875, 57.34117889404297, 37.783470153808594, 13.792274475097656, 15.83233642578125, 262.142578125, -0.14422607421875, 5.627143859863281, -144.17626953125, 253.38218688964844], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000210.npy"}
|
|
{"epoch": 0.30837004405286345, "step": 211, "batch_size": 64, "mean": 70.13721466064453, "std": 103.44092559814453, "min": -128.70648193359375, "p10": -25.1540283203125, "median": 55.406070709228516, "p90": 183.01339721679688, "max": 388.8066711425781, "pos_frac": 0.765625, "sample": [77.28218078613281, -18.152250289916992, 181.74383544921875, 68.6293716430664, 72.186279296875, 103.67356872558594, -2.1647796630859375, 127.77831268310547, 360.70697021484375, -128.70648193359375, 3.7828445434570312, 21.38066291809082, 23.811565399169922, -8.77520751953125, 83.41714477539062, 178.7905731201172, 123.61494445800781, 131.75482177734375, 80.92819213867188, 9.523544311523438, 14.855667114257812, -4.1408843994140625, 127.73916625976562, 75.93318939208984, -18.53314971923828, 0.2827911376953125, -25.38463592529297, 307.7034606933594, 29.22451400756836, 388.8066711425781, 284.012939453125, -24.615943908691406, -69.81028747558594, 86.9525146484375, 178.4481658935547, 174.98663330078125, -65.302001953125, 111.80073547363281, 28.019683837890625, 29.930145263671875, 88.66609191894531, 222.6399383544922, -18.14960479736328, 41.18682861328125, 26.329816818237305, 50.086204528808594, 54.02745056152344, -62.148193359375, -8.851207733154297, 172.63711547851562, -93.02857971191406, 68.88557434082031, 139.66270446777344, 64.76146697998047, 104.32867431640625, 200.85955810546875, 56.784690856933594, 183.5574951171875, 7.7181243896484375, 104.21873474121094, 47.27260971069336, 22.1922607421875, 12.616561889648438, -119.58816528320312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000211.npy"}
|
|
{"epoch": 0.30983847283406757, "step": 212, "batch_size": 64, "mean": 100.10625457763672, "std": 118.42173767089844, "min": -177.6555633544922, "p10": -29.29695854187011, "median": 107.26559448242188, "p90": 235.43085327148444, "max": 406.6883850097656, "pos_frac": 0.78125, "sample": [-177.6555633544922, -76.74031066894531, 283.6854553222656, 208.3536376953125, 51.99517059326172, 187.05722045898438, 59.329437255859375, -21.339771270751953, 3.274934768676758, 3.76495361328125, 167.8643341064453, 115.02716827392578, -110.02146911621094, -12.28057861328125, -10.802249908447266, 16.329551696777344, 216.92471313476562, 196.34365844726562, 327.10479736328125, 172.3400115966797, 88.08444213867188, -88.8213882446289, 36.81894302368164, 81.64419555664062, 118.44082641601562, 116.59801483154297, 190.34188842773438, -32.37529373168945, 211.51864624023438, -5.486780166625977, 89.92806243896484, 132.6423797607422, 406.6883850097656, 116.67817687988281, 35.18467712402344, -70.96382141113281, 210.6111602783203, -23.614578247070312, 109.02783966064453, 398.43878173828125, 220.64010620117188, 106.40948486328125, 44.04413604736328, 155.31494140625, 20.2843017578125, 45.57037353515625, -24.59744644165039, 13.856708526611328, 209.7098388671875, 38.43461990356445, 108.1217041015625, 255.3248291015625, 267.00433349609375, 170.47537231445312, 241.76974487304688, -13.52138900756836, 152.62498474121094, 48.831119537353516, 144.76904296875, 120.2181396484375, 37.21928405761719, 215.2816162109375, -31.31103515625, 138.385498046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000212.npy"}
|
|
{"epoch": 0.31130690161527164, "step": 213, "batch_size": 64, "mean": 98.24685668945312, "std": 109.84493255615234, "min": -155.27743530273438, "p10": -25.642171859741207, "median": 83.47657775878906, "p90": 254.2536071777344, "max": 328.83721923828125, "pos_frac": 0.8125, "sample": [249.26922607421875, -46.7828369140625, 33.170570373535156, -76.33698272705078, 145.5290069580078, 11.01104736328125, 46.47735595703125, 105.68513488769531, -155.27743530273438, -61.824310302734375, 7.122322082519531, 130.41339111328125, 231.78909301757812, -23.055727005004883, 67.4059066772461, 137.4385223388672, 7.324945449829102, 57.58204650878906, 142.8108367919922, 217.98643493652344, 328.83721923828125, 219.36810302734375, 36.79216003417969, 299.580078125, 121.24864196777344, 229.08177185058594, 275.9911804199219, -23.085224151611328, 7.060249328613281, 16.164623260498047, 133.27474975585938, 79.5990982055664, 130.73849487304688, 86.14410400390625, -13.630279541015625, 80.80905151367188, 120.1843490600586, 218.59860229492188, 241.922119140625, 128.97303771972656, 9.955463409423828, 123.4375, 312.8780212402344, -20.986602783203125, 150.8154296875, 40.89178466796875, 122.4739761352539, 317.4010314941406, -44.273406982421875, 176.32955932617188, 190.53793334960938, 256.3897705078125, 264.4635009765625, 14.674518585205078, -26.738006591796875, -1.196502685546875, -45.61273956298828, 69.17927551269531, 65.75634765625, 42.649024963378906, 49.445228576660156, 30.272308349609375, 119.95999145507812, 123.70475769042969], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000213.npy"}
|
|
{"epoch": 0.31277533039647576, "step": 214, "batch_size": 64, "mean": 83.96929931640625, "std": 108.2212142944336, "min": -138.57516479492188, "p10": -26.087282943725583, "median": 60.12006378173828, "p90": 235.09481506347657, "max": 399.91510009765625, "pos_frac": 0.78125, "sample": [54.631263732910156, 301.0823059082031, 9.38369369506836, 125.12736511230469, 8.842546463012695, -12.47433090209961, -37.64632034301758, -138.57516479492188, 211.2420654296875, -23.149188995361328, 133.02035522460938, 43.848731994628906, 68.91431427001953, -0.9735927581787109, 41.73541259765625, 76.87335205078125, 135.2164306640625, -8.432764053344727, 4.5340576171875, 130.11163330078125, 59.84278106689453, 52.46593475341797, 91.72953033447266, 197.94952392578125, 181.24929809570312, 42.321502685546875, 31.79180145263672, -10.429567337036133, 94.05265808105469, -33.679595947265625, 44.10921859741211, 60.39734649658203, 8.280769348144531, 267.7554931640625, -80.54401397705078, 352.5069885253906, 237.56295776367188, 18.506317138671875, 14.579818725585938, 271.5422058105469, 149.9320068359375, 130.854248046875, 213.36708068847656, 229.3358154296875, -121.43079376220703, 399.91510009765625, 28.649625778198242, 127.14728546142578, 173.71975708007812, 84.88057708740234, -13.254478454589844, 108.13258361816406, -27.346466064453125, 84.37052917480469, -8.939468383789062, 168.58419799804688, 13.70608901977539, 243.26495361328125, 50.602081298828125, 88.93978881835938, 66.80300903320312, 168.22349548339844, -36.42002868652344, 25.695026397705078], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000214.npy"}
|
|
{"epoch": 0.3142437591776799, "step": 215, "batch_size": 64, "mean": 90.1728515625, "std": 100.8108901977539, "min": -107.79747009277344, "p10": -29.29387664794922, "median": 74.86587142944336, "p90": 209.6424606323243, "max": 415.2935791015625, "pos_frac": 0.8125, "sample": [52.76609420776367, 241.9359130859375, 100.26611328125, 143.29147338867188, -36.88871765136719, -49.91546630859375, 13.489723205566406, 111.68456268310547, 120.26597595214844, 71.17593383789062, 93.71852111816406, 37.16548156738281, 18.50994300842285, -3.199277877807617, 315.2042236328125, 157.54556274414062, 181.88006591796875, 215.60304260253906, 415.2935791015625, 105.64920806884766, 58.18678283691406, 246.00518798828125, 99.30106353759766, -28.650283813476562, 45.44917297363281, 131.66748046875, -44.04865646362305, 66.14820861816406, 166.55819702148438, 195.73443603515625, 17.245603561401367, 45.31926727294922, 64.46923828125, -11.736875534057617, 145.2654266357422, 74.35330200195312, -85.59610748291016, 128.5170135498047, 146.9730224609375, 181.35198974609375, 245.94723510742188, 104.4924087524414, 143.48056030273438, -10.133712768554688, 138.33628845214844, 37.77622985839844, 63.68379211425781, -29.5697021484375, 46.78690719604492, 195.5652313232422, -107.79747009277344, 85.98465728759766, 340.7894592285156, 24.009078979492188, 56.595001220703125, 70.05487060546875, 191.64654541015625, -6.7904815673828125, 98.57980346679688, 16.25245475769043, 93.35579681396484, 10.984331130981445, -62.30105972290039, 75.3784408569336], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000215.npy"}
|
|
{"epoch": 0.315712187958884, "step": 216, "batch_size": 64, "mean": 90.40126037597656, "std": 114.3624038696289, "min": -168.7686767578125, "p10": -41.74292373657226, "median": 60.61299514770508, "p90": 245.32876892089845, "max": 373.194580078125, "pos_frac": 0.796875, "sample": [167.91415405273438, 265.184814453125, -37.582481384277344, 122.82858276367188, 138.78326416015625, 31.409788131713867, -110.44207763671875, 132.09027099609375, 168.89666748046875, 15.899513244628906, 373.194580078125, 55.896728515625, 108.19760131835938, -43.525970458984375, -2.573261260986328, -9.106475830078125, -76.6209487915039, 172.2877960205078, 17.189868927001953, -145.37881469726562, 12.59625244140625, 237.12139892578125, -8.185020446777344, 37.190826416015625, 100.70709228515625, 113.03868103027344, 24.7353515625, 142.32928466796875, 113.53965759277344, 36.392486572265625, 80.25658416748047, 130.68319702148438, 53.621307373046875, -20.010652542114258, 288.2918395996094, 236.52098083496094, -57.85612487792969, 237.64895629882812, 39.22455596923828, 64.7565689086914, 28.644567489624023, 178.85061645507812, 22.67845344543457, -168.7686767578125, 38.17472839355469, 129.32777404785156, 150.178466796875, 294.30120849609375, 39.72062683105469, -16.804237365722656, 211.88095092773438, 48.82991027832031, 56.46942138671875, 110.45023345947266, 28.984506607055664, 246.49147033691406, 156.3316650390625, 242.6157989501953, 284.10821533203125, 282.89166259765625, 22.788299560546875, -44.20133972167969, 25.78160858154297, 208.80807495117188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000216.npy"}
|
|
{"epoch": 0.31718061674008813, "step": 217, "batch_size": 64, "mean": 104.92652893066406, "std": 106.1994857788086, "min": -119.08464050292969, "p10": -10.241486358642574, "median": 80.52801895141602, "p90": 273.83560943603527, "max": 375.4555358886719, "pos_frac": 0.875, "sample": [106.59466552734375, 107.29460144042969, 73.70520782470703, 81.32801055908203, 49.00215148925781, 35.28663635253906, 57.107757568359375, 165.3752899169922, 110.56449127197266, 57.024169921875, 17.066207885742188, 285.7091064453125, 127.52222442626953, 242.3146209716797, -77.85327911376953, 4.27960205078125, 39.612098693847656, -20.451461791992188, 137.6372528076172, 355.8285827636719, 0.3901710510253906, 70.73023223876953, 154.03054809570312, 299.5015563964844, 109.39214324951172, 177.52761840820312, -12.368949890136719, 40.4017448425293, -119.08464050292969, 88.88835144042969, 375.4555358886719, 156.04132080078125, 17.43653678894043, 38.223365783691406, 169.42083740234375, 21.528106689453125, 169.494873046875, 198.57174682617188, -5.27740478515625, 324.5213623046875, 288.3719482421875, -53.187705993652344, 246.1307830810547, 199.4364013671875, 123.05250549316406, 286.4618835449219, 29.848922729492188, 59.09454345703125, 61.721824645996094, 54.08583068847656, 79.72802734375, 151.09768676757812, 243.5474853515625, 27.217058181762695, 130.3301544189453, 49.55833435058594, -12.719688415527344, -18.165321350097656, 47.31214141845703, 35.40403747558594, 112.45573425292969, 162.9520721435547, 170.46859741210938, 11.321344375610352], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000217.npy"}
|
|
{"epoch": 0.3186490455212922, "step": 218, "batch_size": 64, "mean": 84.04899597167969, "std": 109.21852111816406, "min": -107.45062255859375, "p10": -40.688913345336914, "median": 71.08927154541016, "p90": 232.67899017333986, "max": 416.15478515625, "pos_frac": 0.765625, "sample": [73.20675659179688, 41.47126007080078, 136.85330200195312, 132.40492248535156, 120.55413055419922, 23.07766342163086, 260.4210510253906, 29.719375610351562, 14.191669464111328, 63.14075469970703, 134.5287628173828, 74.890625, -0.6585350036621094, 125.32872009277344, 323.938720703125, 97.19869995117188, 96.64762878417969, -16.03955841064453, 63.81855010986328, 143.67771911621094, 32.946197509765625, 39.71294403076172, 17.727947235107422, 164.03875732421875, -62.90547180175781, -107.45062255859375, 235.24200439453125, 87.52566528320312, 135.85214233398438, 142.5863037109375, 27.663185119628906, 276.075439453125, 237.78118896484375, 200.05398559570312, 140.10943603515625, 63.08295440673828, -9.920469284057617, -16.007532119750977, 95.38004302978516, 22.88119125366211, -38.18280792236328, -58.07891845703125, 107.11150360107422, -41.76295852661133, -35.779640197753906, -38.061134338378906, 416.15478515625, 222.7250518798828, 167.3329620361328, -22.47472381591797, -78.63307189941406, 183.02552795410156, 349.4744873046875, 8.569734573364258, 31.138240814208984, 94.36736297607422, 137.89402770996094, 85.6258544921875, 12.958559036254883, -52.55803680419922, 63.96466827392578, -92.09321594238281, 226.69862365722656, 68.97178649902344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000218.npy"}
|
|
{"epoch": 0.3201174743024963, "step": 219, "batch_size": 64, "mean": 71.45269012451172, "std": 111.0450210571289, "min": -238.55520629882812, "p10": -38.68991165161132, "median": 59.704952239990234, "p90": 234.64386291503908, "max": 363.466064453125, "pos_frac": 0.828125, "sample": [317.73651123046875, 95.56981658935547, 11.104520797729492, 237.75238037109375, 9.940668106079102, 300.70404052734375, -238.55520629882812, 6.97783088684082, 21.346282958984375, 174.28057861328125, 1.187103271484375, 71.63705444335938, 159.3995819091797, 301.23822021484375, 363.466064453125, 88.22879028320312, 34.37773513793945, 227.39065551757812, 117.71249389648438, 131.39093017578125, 163.42977905273438, 14.304000854492188, -71.15128326416016, 39.297119140625, 20.74811553955078, -24.71623992919922, 160.890380859375, 2.2479171752929688, 254.49807739257812, 241.48593139648438, 61.27513885498047, 188.15426635742188, 23.514694213867188, 30.24592399597168, 35.346466064453125, 129.9033660888672, 91.84782409667969, 10.632286071777344, 72.85696411132812, 18.0135498046875, 191.22994995117188, 108.44415283203125, 86.78764343261719, -97.04885864257812, 40.78059387207031, 103.72459411621094, -107.95072937011719, 8.083433151245117, 16.411365509033203, 92.27375030517578, 21.01507568359375, -57.845481872558594, 1.3617935180664062, -23.000713348388672, -131.7449188232422, 101.72940826416016, 106.37619018554688, 66.03594970703125, -19.072193145751953, -29.735031127929688, 91.16928100585938, 58.134765625, 92.62986755371094, -42.52771759033203], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000219.npy"}
|
|
{"epoch": 0.32158590308370044, "step": 220, "batch_size": 64, "mean": 60.78836441040039, "std": 79.92025756835938, "min": -126.96530151367188, "p10": -33.76715316772461, "median": 61.9359016418457, "p90": 145.77715911865235, "max": 247.83724975585938, "pos_frac": 0.78125, "sample": [171.09457397460938, 30.09507942199707, 65.04607391357422, -3.3000411987304688, 41.58317565917969, 247.83724975585938, 237.64788818359375, 10.12479019165039, 103.41717529296875, 116.84353637695312, 88.12230682373047, 137.65728759765625, 116.5683822631836, 103.69281768798828, -126.96530151367188, 62.297943115234375, 152.9725341796875, 49.581024169921875, 134.51931762695312, 144.40444946289062, 81.50990295410156, 14.574892044067383, 13.462806701660156, -108.379150390625, -34.188636779785156, 146.36546325683594, -9.531869888305664, 41.253082275390625, 182.9204864501953, 45.614864349365234, -32.78369140625, 111.40646362304688, 94.49317932128906, -14.829360961914062, 111.16911315917969, 61.57386016845703, 233.26202392578125, 137.35450744628906, 40.84046173095703, 72.15906524658203, 5.214277267456055, 27.748661041259766, -2.247701644897461, 61.43135070800781, -39.99986267089844, 47.712921142578125, 66.93197631835938, 95.26466369628906, -106.97679138183594, 70.7265625, 36.10026550292969, 20.356483459472656, 143.36764526367188, -24.135286331176758, 27.534076690673828, 122.18418884277344, 112.43896484375, -13.515769958496094, 105.6138916015625, 95.178955078125, 5.692531585693359, 109.69098663330078, -61.126129150390625, -86.21920013427734], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000220.npy"}
|
|
{"epoch": 0.32305433186490456, "step": 221, "batch_size": 64, "mean": 121.70396423339844, "std": 110.1458969116211, "min": -125.88165283203125, "p10": -1.009260559082024, "median": 124.32421493530273, "p90": 281.9357543945313, "max": 394.0113220214844, "pos_frac": 0.890625, "sample": [63.31959533691406, 51.7186279296875, 91.14057922363281, 6.159095764160156, 72.8910140991211, 394.0113220214844, 11.661445617675781, 295.0455322265625, 177.10400390625, 204.63339233398438, 221.45889282226562, 7.015903472900391, 232.7156524658203, 87.0511474609375, 94.7540512084961, 184.17578125, 10.380889892578125, 286.3445129394531, -11.332986831665039, 130.76690673828125, 184.02496337890625, 139.4233856201172, 271.6486511230469, 334.54876708984375, 298.455810546875, 178.72259521484375, 139.7836456298828, -125.88165283203125, 44.45289611816406, 44.599998474121094, 137.01605224609375, 152.21231079101562, -5.8176727294921875, 68.78829193115234, 51.6416015625, 115.30126190185547, 74.29788970947266, -94.65269470214844, 157.68020629882812, 267.739501953125, -4.081413269042969, 181.66925048828125, 168.24412536621094, 197.71920776367188, 82.12397766113281, 166.10772705078125, 28.791061401367188, 288.9753112792969, 219.71994018554688, 173.71624755859375, 10.114765167236328, -53.716041564941406, 36.45555877685547, 180.469482421875, 23.169326782226562, -52.379913330078125, 14.485208511352539, 117.88152313232422, 41.2678108215332, 143.98284912109375, 250.92642211914062, 61.52265167236328, 311.75714111328125, 155.1298065185547], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000221.npy"}
|
|
{"epoch": 0.3245227606461087, "step": 222, "batch_size": 64, "mean": 95.16747283935547, "std": 110.0402603149414, "min": -145.16644287109375, "p10": -24.63846912384033, "median": 107.16362762451172, "p90": 246.07276611328126, "max": 361.8472900390625, "pos_frac": 0.828125, "sample": [-19.84319305419922, 308.3854064941406, 7.734405517578125, -49.890899658203125, 26.872459411621094, -25.12858009338379, -3.19586181640625, 38.354736328125, 1.13592529296875, 108.29278564453125, 85.62340545654297, 117.8451156616211, 319.5919494628906, 43.829010009765625, 112.73148345947266, 106.7667465209961, -145.16644287109375, 129.72918701171875, 191.97921752929688, 122.42898559570312, 126.58928680419922, 85.20785522460938, -107.1905517578125, 248.0797119140625, -76.60283660888672, 177.86241149902344, 361.8472900390625, 65.41075134277344, 24.917518615722656, 12.001075744628906, 107.56050872802734, 21.469635009765625, 167.79898071289062, -58.538909912109375, 158.1543731689453, 142.5518035888672, 36.38315963745117, 148.22003173828125, -23.494876861572266, 164.27340698242188, -21.89855194091797, 123.88121032714844, 34.394691467285156, 21.880905151367188, 241.389892578125, 174.09597778320312, 114.6378402709961, 165.31439208984375, 111.3805923461914, 42.33490753173828, 50.60420227050781, 109.83573150634766, 300.019287109375, 48.5650749206543, 227.04837036132812, 135.71102905273438, 129.47618103027344, -97.41435241699219, 156.50888061523438, 26.670005798339844, 345.4385681152344, 42.6256103515625, 280.02008056640625, 67.62132263183594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000222.npy"}
|
|
{"epoch": 0.32599118942731276, "step": 223, "batch_size": 64, "mean": 80.98518371582031, "std": 123.77351379394531, "min": -210.06546020507812, "p10": -77.82265968322753, "median": 84.76811981201172, "p90": 236.19184570312507, "max": 410.53582763671875, "pos_frac": 0.734375, "sample": [79.91813659667969, -59.895057678222656, 160.16146850585938, 131.7403564453125, 80.82047271728516, 254.7161865234375, 155.27215576171875, 71.16141510009766, 168.83583068847656, 112.50016784667969, -62.55220031738281, 26.272186279296875, 37.14006042480469, 197.484130859375, 211.0045623779297, 201.75430297851562, 86.74364471435547, 69.833251953125, 346.3521728515625, -26.667240142822266, 114.8155288696289, -25.85115623474121, 88.8917007446289, -62.833187103271484, -21.21923065185547, 217.49765014648438, -94.41978454589844, 259.5062255859375, 88.62162780761719, -111.36713409423828, 13.603561401367188, -15.990310668945312, 196.5657501220703, -201.31515502929688, 94.26202392578125, 149.993408203125, -84.24671936035156, 117.4498062133789, 58.82596969604492, 410.53582763671875, 211.908935546875, -95.9771499633789, 125.50438690185547, 195.961181640625, -114.31173706054688, -15.032367706298828, 193.87741088867188, 244.20364379882812, 114.76374816894531, -14.285926818847656, 45.44103240966797, 23.895065307617188, 18.447662353515625, 268.25408935546875, 23.04932403564453, 21.48912811279297, -12.014471054077148, 17.134550094604492, 127.88400268554688, 117.13885498046875, 82.79259490966797, 126.56441497802734, -210.06546020507812, 250.50643920898438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000223.npy"}
|
|
{"epoch": 0.3274596182085169, "step": 224, "batch_size": 64, "mean": 96.40702056884766, "std": 127.55052947998047, "min": -213.3570556640625, "p10": -36.07624130249023, "median": 68.7740707397461, "p90": 281.58894958496114, "max": 476.7486572265625, "pos_frac": 0.78125, "sample": [153.60369873046875, 3.2479400634765625, 108.80223083496094, 30.30309295654297, -36.302040100097656, 361.113037109375, 188.17701721191406, 67.75146484375, 86.66631317138672, -213.3570556640625, 51.521759033203125, 135.8430633544922, 140.57217407226562, 107.44284057617188, 328.48187255859375, 69.79667663574219, 85.6701889038086, 175.63511657714844, 162.94692993164062, 476.7486572265625, 231.72421264648438, 78.22297668457031, 200.6983642578125, 193.54489135742188, -60.62931823730469, -52.23857498168945, 223.585693359375, 63.91251754760742, 165.5377197265625, -23.8450927734375, 40.61061096191406, 43.26347351074219, 16.42210578918457, 402.8365173339844, 189.3865966796875, 186.4626007080078, 89.1873550415039, -40.879554748535156, 18.372434616088867, 42.24466323852539, 174.53753662109375, -0.43804931640625, 30.18402862548828, -24.97177505493164, 64.54360961914062, -63.030982971191406, -27.650299072265625, -15.420654296875, -93.06204223632812, 14.297863006591797, 156.60850524902344, 129.265869140625, 301.4261474609375, 13.02775764465332, 13.628105163574219, 44.13498306274414, -0.3333740234375, 235.30215454101562, 38.692535400390625, -35.54937744140625, 71.4281005859375, 309.9716491699219, 314.4717102050781, 25.90017318725586], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000224.npy"}
|
|
{"epoch": 0.328928046989721, "step": 225, "batch_size": 64, "mean": 118.33515930175781, "std": 133.5341796875, "min": -167.36172485351562, "p10": -48.985718917846675, "median": 114.5133171081543, "p90": 305.96421813964844, "max": 531.2093505859375, "pos_frac": 0.78125, "sample": [210.80984497070312, 117.86679077148438, 307.6123046875, 316.1448669433594, 29.54854393005371, 85.09822845458984, 26.38323974609375, -50.70709228515625, -58.37371826171875, 338.7298583984375, 313.5141296386719, 211.82366943359375, -81.71379089355469, -32.087303161621094, 224.0993194580078, -1.2259483337402344, -88.76277160644531, 1.8887386322021484, -21.173656463623047, 40.03770446777344, 132.05426025390625, 99.73675537109375, 156.18133544921875, 139.45364379882812, -3.428590774536133, 120.25788116455078, 47.793670654296875, 136.25204467773438, 18.782535552978516, 187.65884399414062, 263.5020751953125, 131.82688903808594, 169.0432586669922, 85.04246520996094, -15.243244171142578, 345.97247314453125, 351.01446533203125, 2.6798629760742188, 204.9071044921875, 302.1186828613281, 75.25364685058594, 227.34642028808594, 102.43949890136719, -58.12506866455078, 137.84934997558594, 101.0537338256836, 531.2093505859375, 210.29901123046875, 163.71157836914062, 75.7159423828125, 269.44512939453125, 16.84075927734375, 188.43963623046875, -63.8948974609375, -44.969181060791016, 291.63531494140625, 215.64295959472656, 133.21829223632812, 162.66905212402344, -167.36172485351562, 111.15984344482422, 110.4671630859375, -5.592643737792969, 23.878036499023438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000225.npy"}
|
|
{"epoch": 0.3303964757709251, "step": 226, "batch_size": 64, "mean": 91.46955108642578, "std": 131.2726593017578, "min": -195.21095275878906, "p10": -32.38587188720703, "median": 68.094970703125, "p90": 257.8060821533204, "max": 473.58868408203125, "pos_frac": 0.765625, "sample": [473.58868408203125, -108.93925476074219, -33.26316452026367, 31.936508178710938, -14.181407928466797, 153.59938049316406, -0.14581298828125, 68.61293029785156, 133.80795288085938, -30.338855743408203, 70.67891693115234, 21.218795776367188, -21.505996704101562, 40.204689025878906, -119.29559326171875, 67.76268005371094, -54.0950813293457, 232.61871337890625, 240.07833862304688, 57.289390563964844, 265.4036865234375, 313.78314208984375, 59.55084991455078, 95.00410461425781, 124.35682678222656, 218.3285675048828, 33.11766815185547, -3.811737060546875, 93.87936401367188, 355.7191467285156, 210.1376495361328, -50.88572692871094, 147.2212371826172, 68.42726135253906, 26.484506607055664, 391.70550537109375, 199.35247802734375, 394.7063293457031, 206.78079223632812, -4.019767761230469, 3.598724365234375, 82.4852523803711, -195.21095275878906, 120.94497680664062, 33.699459075927734, 71.91734313964844, -11.961204528808594, 19.932769775390625, 25.15001106262207, 68.80412292480469, 1.995758056640625, 223.47181701660156, 87.25344848632812, 153.18551635742188, 35.18622589111328, 102.72285461425781, -58.844512939453125, 7.469577789306641, 378.58746337890625, 40.29787063598633, 24.742164611816406, 111.61198425292969, -16.131988525390625, 188.269287109375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000226.npy"}
|
|
{"epoch": 0.33186490455212925, "step": 227, "batch_size": 64, "mean": 117.9362564086914, "std": 131.2845458984375, "min": -301.91778564453125, "p10": -23.199317359924308, "median": 104.95598983764648, "p90": 306.71429138183595, "max": 453.27374267578125, "pos_frac": 0.859375, "sample": [253.15975952148438, 78.72378540039062, 129.11972045898438, 317.2996826171875, 161.99786376953125, 46.92718505859375, 80.04798889160156, 102.65856170654297, 245.40098571777344, -13.373016357421875, 98.69669342041016, 128.5374298095703, 162.06332397460938, 12.377904891967773, -42.7946662902832, 20.690750122070312, 139.51797485351562, 92.37831115722656, 169.1123504638672, 453.27374267578125, 101.34919738769531, 64.94778442382812, 247.39413452148438, 163.45651245117188, 79.95674133300781, 218.5051727294922, 137.318115234375, 191.52740478515625, 90.84320068359375, 90.88787841796875, 139.3547821044922, 78.54643249511719, 261.8587646484375, 4.983451843261719, -48.04655456542969, -93.45970153808594, 138.74700927734375, 184.9261932373047, 340.308837890625, 107.25341796875, 144.81321716308594, 308.4805908203125, -90.82958221435547, 429.95977783203125, 29.828428268432617, 27.36418914794922, 147.86192321777344, 26.682960510253906, 170.83062744140625, 29.794761657714844, 115.91011047363281, 341.45794677734375, 147.81884765625, 28.773239135742188, 1.897317886352539, -33.34825134277344, 302.5929260253906, -301.91778564453125, -4.263540267944336, 77.51156616210938, -27.41058921813965, 36.621070861816406, 119.09291076660156, 381.9228820800781], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000227.npy"}
|
|
{"epoch": 0.3333333333333333, "step": 228, "batch_size": 64, "mean": 81.7197494506836, "std": 119.18263244628906, "min": -187.48562622070312, "p10": -68.54853744506835, "median": 85.45028305053711, "p90": 244.0647689819336, "max": 402.2614440917969, "pos_frac": 0.75, "sample": [-114.8799819946289, -130.07907104492188, -187.48562622070312, 121.10531616210938, 105.34912872314453, 246.5939178466797, -150.5238037109375, 135.87771606445312, 138.8255157470703, -104.26593017578125, 13.80438232421875, 199.75552368164062, 307.81634521484375, 210.71078491210938, 294.2049560546875, 259.52667236328125, 277.65338134765625, 136.8809356689453, 88.30632019042969, -23.74176025390625, -30.25921630859375, 238.16342163085938, 72.9848403930664, 248.4245147705078, 123.42948150634766, 113.02064514160156, 96.50761413574219, 16.834766387939453, -75.3687515258789, 104.87837219238281, 18.270065307617188, 127.14427185058594, 7.14935302734375, 176.11813354492188, 171.8759002685547, 60.60781478881836, 79.73011016845703, 71.2738037109375, 38.90385437011719, 28.3231201171875, 85.80598449707031, 174.03515625, -52.63470458984375, -21.72271728515625, 92.25108337402344, 88.01570892333984, 402.2614440917969, 14.787162780761719, 84.58445739746094, 41.75653839111328, 230.2173309326172, -19.573657989501953, -26.313541412353516, 162.17013549804688, 173.5552215576172, 135.35989379882812, 154.23008728027344, 85.0945816040039, -48.630104064941406, 19.459259033203125, -10.631660461425781, 50.537384033203125, -95.89360046386719, -2.1043167114257812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000228.npy"}
|
|
{"epoch": 0.33480176211453744, "step": 229, "batch_size": 64, "mean": 92.12015533447266, "std": 121.35041809082031, "min": -257.49798583984375, "p10": -35.241252136230464, "median": 87.65231704711914, "p90": 251.6521759033203, "max": 429.98089599609375, "pos_frac": 0.765625, "sample": [3.7383880615234375, 12.129749298095703, -34.01397705078125, 309.64739990234375, 30.001384735107422, -132.63983154296875, 96.70694732666016, 134.7765350341797, 67.10946655273438, 168.33575439453125, 233.65762329101562, 251.14022827148438, 61.466766357421875, 332.1693420410156, 105.30665588378906, 92.3896255493164, 170.40750122070312, 88.66785430908203, 31.65768051147461, 253.05520629882812, -35.76722717285156, 72.98814392089844, 2.3763084411621094, 181.0214080810547, 111.10935974121094, 30.649259567260742, 429.98089599609375, -72.75262451171875, 22.093902587890625, 172.19873046875, 151.09226989746094, 208.33175659179688, 158.8014678955078, 263.628662109375, 251.87158203125, 50.110618591308594, 216.5702362060547, 23.068206787109375, 86.63677978515625, -30.709014892578125, 81.93917083740234, 101.89985656738281, -27.43964385986328, 157.03497314453125, -40.37482452392578, 95.519287109375, 157.21551513671875, 180.4213409423828, -17.338333129882812, -101.4755859375, 90.75177764892578, 357.33026123046875, 22.545785903930664, 75.51741027832031, 81.24845123291016, 120.88404846191406, -4.026252746582031, 145.43392944335938, -6.0797576904296875, 163.89337158203125, -257.49798583984375, -0.4219226837158203, -37.28697204589844, -13.014686584472656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000229.npy"}
|
|
{"epoch": 0.33627019089574156, "step": 230, "batch_size": 64, "mean": 126.29830932617188, "std": 123.33448791503906, "min": -214.65310668945312, "p10": -21.893614196777335, "median": 115.63056564331055, "p90": 306.0173004150391, "max": 377.0670166015625, "pos_frac": 0.84375, "sample": [84.06732177734375, 295.6848449707031, -9.578733444213867, 243.2391815185547, 163.64755249023438, 66.75865173339844, -4.749269485473633, 198.5630645751953, 91.46353149414062, 116.26895141601562, 154.2669677734375, 146.41275024414062, -37.411537170410156, 175.769287109375, 4.1854095458984375, 69.52395629882812, -28.4168701171875, 99.88720703125, 44.686798095703125, 89.65220642089844, 46.26300048828125, -84.92723083496094, -24.820655822753906, 139.9306640625, 114.99217987060547, 139.15614318847656, 176.58963012695312, 285.66876220703125, 206.97772216796875, 152.34844970703125, 68.72393035888672, 18.1416015625, 204.91822814941406, 365.3026428222656, 136.2781219482422, -57.38042449951172, 104.66466522216797, 151.2965087890625, 186.17105102539062, 377.0670166015625, -29.966476440429688, 8.298294067382812, -214.65310668945312, -15.063850402832031, 25.389816284179688, 25.921463012695312, 306.6327819824219, 86.01971435546875, 325.5335388183594, 340.5416259765625, 159.58352661132812, 73.2274169921875, 76.70158386230469, 29.63274574279785, 300.439697265625, 228.83387756347656, 351.2854309082031, 101.40608215332031, 304.5811767578125, 147.23492431640625, 187.08773803710938, 105.30411529541016, 131.0366973876953, 356.7996826171875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000230.npy"}
|
|
{"epoch": 0.3377386196769457, "step": 231, "batch_size": 64, "mean": 89.06995391845703, "std": 108.75059509277344, "min": -77.86627960205078, "p10": -29.282697296142576, "median": 77.51030731201172, "p90": 239.55731811523444, "max": 428.5423889160156, "pos_frac": 0.8125, "sample": [367.55938720703125, 86.46513366699219, -28.09484100341797, 194.54840087890625, 10.964256286621094, 69.51036071777344, 245.758056640625, 24.544464111328125, 56.04034423828125, 99.10050201416016, 50.071693420410156, 225.08892822265625, -45.109683990478516, 75.16651916503906, 340.4707336425781, 92.623046875, 176.01046752929688, 103.06645202636719, 116.8686294555664, 172.73599243164062, 251.60818481445312, 89.63059997558594, 79.9074478149414, 57.31077575683594, 38.49137878417969, -44.504051208496094, 262.27642822265625, -41.549747467041016, 161.22860717773438, 428.5423889160156, 376.9499816894531, 173.50042724609375, 91.189208984375, 13.169042587280273, -72.36237335205078, 53.0969123840332, 79.85409545898438, 155.86282348632812, 69.9654541015625, -12.350128173828125, -18.384231567382812, 35.02025604248047, 143.8380889892578, 44.694305419921875, -7.559928894042969, 89.18245697021484, -29.791778564453125, -77.86627960205078, 33.082340240478516, 108.42526245117188, 189.78822326660156, 106.14912414550781, 57.82616424560547, 23.780834197998047, -14.288326263427734, 97.68377685546875, -69.95646667480469, 83.03927612304688, 21.904449462890625, 6.50274658203125, 16.191726684570312, 107.01229858398438, 86.19392395019531, 22.80267333984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000231.npy"}
|
|
{"epoch": 0.3392070484581498, "step": 232, "batch_size": 64, "mean": 73.85276794433594, "std": 119.61105346679688, "min": -118.27458190917969, "p10": -53.97903518676757, "median": 44.42803764343262, "p90": 238.5003295898439, "max": 432.221435546875, "pos_frac": 0.765625, "sample": [116.47705078125, 73.23554229736328, 1.3030872344970703, 41.897117614746094, 10.196212768554688, -73.66493225097656, 296.0931091308594, 164.72003173828125, -14.212295532226562, 194.86776733398438, 40.88976287841797, 297.61981201171875, 47.531402587890625, -45.362693786621094, -17.299972534179688, 34.668739318847656, -8.96246337890625, 150.6910400390625, -80.9083480834961, 67.98871612548828, 95.07736206054688, 413.772705078125, 25.784561157226562, 70.53848266601562, 100.08670806884766, -15.197101593017578, 432.221435546875, 5.949249267578125, 12.2884521484375, 204.32974243164062, 116.35028839111328, 168.8931884765625, -57.6717529296875, 50.46990966796875, -73.82575988769531, 76.8204345703125, -86.72542572021484, 23.529434204101562, -13.396743774414062, 398.9243469238281, 8.26776123046875, -22.396751403808594, 21.12038803100586, -118.27458190917969, 41.06386184692383, 54.231300354003906, 53.24273681640625, -104.96192932128906, 176.18641662597656, 78.64755249023438, 272.33319091796875, 19.051280975341797, 20.936767578125, 253.14486694335938, 188.23751831054688, 10.890958786010742, 107.4998550415039, 177.06387329101562, -18.07350730895996, 44.96472930908203, 28.20645523071289, 43.8913459777832, 66.95358276367188, 78.36084747314453], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000232.npy"}
|
|
{"epoch": 0.3406754772393539, "step": 233, "batch_size": 64, "mean": 112.51909637451172, "std": 150.56126403808594, "min": -130.57992553710938, "p10": -47.44851188659668, "median": 74.97991943359375, "p90": 313.6578796386719, "max": 516.827880859375, "pos_frac": 0.75, "sample": [19.66019630432129, 14.250324249267578, 386.5438232421875, -6.1944580078125, 127.41738891601562, 162.63662719726562, -110.75001525878906, -66.1925277709961, 9.667953491210938, -126.76141357421875, 76.68659973144531, 41.35926818847656, 107.4671630859375, 206.81185913085938, 219.46322631835938, 516.827880859375, 117.2857437133789, 57.556129455566406, 379.86541748046875, 237.7625274658203, 34.73832702636719, 129.3479766845703, 69.068603515625, 378.8031311035156, 11.718391418457031, 189.68678283691406, 18.173110961914062, -38.0775146484375, -120.00635528564453, -12.67642593383789, 151.1331024169922, 301.5658874511719, 157.50335693359375, 135.24148559570312, 240.97535705566406, 318.4184265136719, -17.5098819732666, 313.30877685546875, -18.021072387695312, 40.21971130371094, 114.98950958251953, 299.6722412109375, -1.5188484191894531, 202.62985229492188, 44.75111389160156, 251.51812744140625, -24.777523040771484, -130.57992553710938, -90.02342224121094, 2.6072998046875, 36.862648010253906, 19.52051544189453, 262.9146728515625, 73.27323913574219, -44.04522705078125, 164.1525421142578, -23.980396270751953, 313.8074951171875, -48.90706253051758, 473.2958984375, 289.2521057128906, 32.27922058105469, 182.39065551757812, 146.16259765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000233.npy"}
|
|
{"epoch": 0.342143906020558, "step": 234, "batch_size": 64, "mean": 111.787109375, "std": 116.0333480834961, "min": -132.5538787841797, "p10": -26.698401641845695, "median": 109.95230865478516, "p90": 277.28901367187507, "max": 440.6780090332031, "pos_frac": 0.8125, "sample": [-58.611228942871094, 197.91551208496094, 172.32861328125, 127.00311279296875, 220.27626037597656, 140.1761474609375, 125.88241577148438, 145.08177185058594, 317.79693603515625, -6.966352462768555, 190.1357879638672, 101.49659729003906, -132.5538787841797, 43.48599624633789, 45.922813415527344, 184.16989135742188, 47.48529052734375, -56.72528839111328, 53.97979736328125, 191.4246826171875, 440.6780090332031, 114.9455337524414, 13.492483139038086, -16.63233184814453, 48.51874542236328, -16.626953125, 156.26611328125, 227.79736328125, 14.626480102539062, 282.01251220703125, -4.747602462768555, 128.57058715820312, 266.26751708984375, 115.31172943115234, 325.19683837890625, 80.70597839355469, 50.97010040283203, 151.74932861328125, 81.76914978027344, 165.22694396972656, 169.29617309570312, 125.50398254394531, 207.8839111328125, 45.75652313232422, 19.147674560546875, 64.63600158691406, -29.678497314453125, -20.005821228027344, 173.6610870361328, 98.27558898925781, 67.77173614501953, 301.4407958984375, 407.5821838378906, -54.445228576660156, 180.41363525390625, 104.9590835571289, -29.566650390625, 5.353658676147461, 127.03565979003906, 38.10954284667969, 94.794677734375, -58.9720458984375, 289.3995361328125, 150.21832275390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000234.npy"}
|
|
{"epoch": 0.3436123348017621, "step": 235, "batch_size": 64, "mean": 72.9931640625, "std": 115.5566635131836, "min": -218.55905151367188, "p10": -38.56837577819824, "median": 54.212188720703125, "p90": 208.0507766723633, "max": 370.4739990234375, "pos_frac": 0.75, "sample": [28.423309326171875, 172.50735473632812, -218.55905151367188, 19.48846435546875, 157.15939331054688, 370.4739990234375, 92.4507827758789, 243.8714599609375, -37.46042251586914, 113.4182357788086, -119.92755126953125, -53.1236572265625, 40.80706787109375, 209.18516540527344, 48.4403076171875, -6.0369873046875, -24.674880981445312, 25.11296844482422, 104.4561767578125, -151.08380126953125, 14.195514678955078, 18.62322998046875, -5.580757141113281, 17.259010314941406, 266.1138916015625, 159.78196716308594, 5.346611022949219, -25.113311767578125, 135.474853515625, 205.40386962890625, 106.52645874023438, 58.35438537597656, 196.9833984375, -27.04327392578125, 117.48530578613281, 151.00428771972656, 24.190176010131836, 153.18496704101562, 50.06999206542969, 40.066131591796875, 20.99850082397461, 9.495086669921875, 337.2679748535156, 113.19908142089844, 136.5787811279297, 149.5354766845703, 131.16146850585938, -26.30426788330078, -144.83230590820312, 68.31779479980469, 350.6437683105469, 188.03204345703125, -53.703636169433594, 5.0923919677734375, 92.68281555175781, -21.799291610717773, -39.043212890625, 110.86846923828125, 93.21426391601562, 15.107667922973633, 138.93450927734375, 86.23225402832031, 251.81704711914062, -19.189453125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000235.npy"}
|
|
{"epoch": 0.34508076358296624, "step": 236, "batch_size": 64, "mean": 110.42259216308594, "std": 99.87767791748047, "min": -58.3642578125, "p10": -0.917276763916014, "median": 90.47358703613281, "p90": 247.05466766357424, "max": 363.7062683105469, "pos_frac": 0.890625, "sample": [269.7436828613281, 52.171775817871094, -4.8651580810546875, 105.38170623779297, 148.13046264648438, 109.08201599121094, 264.8888244628906, 28.999317169189453, 62.28823471069336, 89.92924499511719, 119.10784912109375, 245.60842895507812, 86.71864318847656, 104.12508392333984, 179.55892944335938, 212.90426635742188, 40.14918518066406, 197.6072540283203, 8.850791931152344, 247.6744842529297, 95.65936279296875, -9.654342651367188, 46.280311584472656, 209.42587280273438, 192.94662475585938, 25.955364227294922, 77.8742904663086, 238.04150390625, 178.06228637695312, 31.908042907714844, 363.7062683105469, 71.4083480834961, -33.68882751464844, 173.17300415039062, 22.990543365478516, 191.92234802246094, 16.721839904785156, 327.57379150390625, 67.7385025024414, 6.833560943603516, 93.47638702392578, 52.704872131347656, 9.183263778686523, 22.102073669433594, 0.7341575622558594, 185.7379913330078, 82.26376342773438, 50.923858642578125, -58.3642578125, -30.39147186279297, 124.9832763671875, 329.40850830078125, 252.5348663330078, 58.454551696777344, 161.15579223632812, 91.01792907714844, 224.91476440429688, 77.98200225830078, 4.527473449707031, 135.09593200683594, -36.81591796875, 231.92811584472656, -1.6250343322753906, 142.17930603027344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000236.npy"}
|
|
{"epoch": 0.3465491923641703, "step": 237, "batch_size": 64, "mean": 92.47233581542969, "std": 120.62464141845703, "min": -218.98397827148438, "p10": -39.81077995300292, "median": 85.16849899291992, "p90": 263.8693817138673, "max": 469.41357421875, "pos_frac": 0.75, "sample": [135.65208435058594, 159.42425537109375, 213.4593505859375, 139.74325561523438, -3.7024002075195312, 99.2850112915039, 89.80998992919922, 127.47154235839844, 3.4395713806152344, 86.33695983886719, 56.835914611816406, 34.497772216796875, 335.010498046875, 84.00003814697266, 192.55702209472656, -22.588546752929688, -42.49370193481445, 124.11393737792969, -51.335121154785156, 99.38321685791016, 103.47213745117188, -45.517974853515625, 279.4675598144531, 469.41357421875, -65.96492004394531, 279.53759765625, 188.927978515625, 19.670730590820312, 51.09858703613281, 0.6338310241699219, 353.25762939453125, 73.48454284667969, 149.1031036376953, 181.24554443359375, 20.29676055908203, 164.5994110107422, -9.963645935058594, 61.09807586669922, 194.726318359375, -13.686729431152344, 192.42495727539062, 27.69969940185547, 141.2659912109375, 159.47915649414062, 286.99700927734375, 26.490753173828125, -11.760009765625, 81.24649810791016, -218.98397827148438, 232.57858276367188, -12.811368942260742, 49.28020477294922, -86.41020202636719, -33.550628662109375, 152.08908081054688, 166.17388916015625, -20.213111877441406, 277.27972412109375, 11.098869323730469, -67.63703918457031, -25.16351318359375, 102.92150115966797, 36.90484619140625, 135.02732849121094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000237.npy"}
|
|
{"epoch": 0.34801762114537443, "step": 238, "batch_size": 64, "mean": 89.06282043457031, "std": 117.49359893798828, "min": -120.11820220947266, "p10": -28.920471954345697, "median": 67.67056655883789, "p90": 209.65640411376955, "max": 518.6874389648438, "pos_frac": 0.8125, "sample": [29.429733276367188, 2.610137939453125, 5.214080810546875, 113.27705383300781, 23.160770416259766, 3.4366798400878906, 96.75529479980469, 178.282470703125, -58.49781799316406, 246.2247314453125, 62.25019073486328, 137.3792266845703, 81.99574279785156, 42.31308364868164, 16.158071517944336, 104.62745666503906, 202.9759521484375, -40.3499755859375, 34.243614196777344, 51.40327072143555, 5.643842697143555, 454.5730895996094, 28.086898803710938, 172.40065002441406, -35.91020202636719, 380.6938171386719, 73.0909423828125, 11.403358459472656, 168.64926147460938, -50.43467712402344, 118.800537109375, 55.19682312011719, 88.94915771484375, 117.45662689208984, 241.164794921875, 10.581012725830078, 33.379878997802734, 94.14778137207031, 518.6874389648438, 97.86286163330078, -13.980609893798828, -6.993011474609375, 29.561817169189453, -88.0085220336914, 8.621124267578125, 146.18508911132812, 127.97933959960938, 148.50592041015625, 61.85791015625, 299.2565002441406, 184.3024444580078, -6.2342376708984375, 152.96783447265625, 186.72848510742188, -120.11820220947266, 212.5194549560547, 152.4468994140625, 32.28515625, -32.06884002685547, -21.57427978515625, 116.8918228149414, 123.36347961425781, -3.07598876953125, 91.2872085571289], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000238.npy"}
|
|
{"epoch": 0.34948604992657856, "step": 239, "batch_size": 64, "mean": 102.07293701171875, "std": 114.96412658691406, "min": -128.99298095703125, "p10": -33.76628227233886, "median": 95.51105880737305, "p90": 278.7167602539063, "max": 353.438232421875, "pos_frac": 0.828125, "sample": [327.09967041015625, 174.11619567871094, -52.48387145996094, 275.63751220703125, -55.7874755859375, 98.16778564453125, 85.5159912109375, 110.77275085449219, -34.25281524658203, -94.247314453125, 37.95352554321289, 31.822952270507812, 42.9228630065918, 186.9356689453125, 49.746421813964844, 4.02606201171875, 322.6430969238281, 105.04705810546875, 141.06890869140625, 118.2654037475586, 163.99285888671875, 120.68916320800781, 183.44786071777344, -51.2039680480957, 31.402000427246094, 22.604963302612305, 143.43417358398438, 297.2738952636719, 149.0503387451172, 78.14820861816406, -4.846837997436523, 32.47796630859375, 292.8717041015625, 280.03643798828125, -113.78208923339844, 62.913002014160156, 11.569313049316406, 275.23834228515625, -28.603164672851562, 263.39813232421875, 39.88077163696289, 250.01417541503906, 174.73202514648438, 136.30770874023438, 26.456466674804688, 50.48389434814453, -128.99298095703125, 56.84745788574219, -32.631038665771484, 124.46408081054688, 110.28702545166016, 130.22604370117188, 218.27662658691406, 51.6199836730957, 353.438232421875, 157.2160186767578, 305.95111083984375, 60.22915267944336, 92.87850952148438, 98.14360809326172, 42.10699462890625, -29.583213806152344, 134.23536682128906, 24.997365951538086], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000239.npy"}
|
|
{"epoch": 0.3509544787077827, "step": 240, "batch_size": 64, "mean": 90.84451293945312, "std": 126.16971588134766, "min": -145.53817749023438, "p10": -104.042520904541, "median": 100.27936172485352, "p90": 272.6774047851562, "max": 381.86749267578125, "pos_frac": 0.78125, "sample": [133.5618133544922, 151.0087890625, 130.72616577148438, 57.46392059326172, 112.99986267089844, 272.6279296875, 74.70668029785156, 276.6914367675781, -67.26082611083984, -41.156822204589844, 173.7196502685547, 94.38313293457031, -137.06350708007812, 154.373779296875, 272.6986083984375, 29.114614486694336, -108.35736083984375, 34.50422668457031, 22.93938446044922, 178.7659912109375, 136.73892211914062, 263.49163818359375, 113.1982421875, 114.85930633544922, 53.26896667480469, 16.369579315185547, -145.53817749023438, 71.63558197021484, 80.50733947753906, 343.02374267578125, 189.16297912597656, 98.72869110107422, 107.06864929199219, 41.691505432128906, -116.40737915039062, 284.42108154296875, 157.6831512451172, 148.1262664794922, -121.60301971435547, 73.99103546142578, -29.774646759033203, 208.51866149902344, -107.94355773925781, 45.27769088745117, 381.86749267578125, -94.94010162353516, 200.93441772460938, -29.29924964904785, 211.0626220703125, 128.39064025878906, 83.67677307128906, 306.70379638671875, 15.33172607421875, 86.51961517333984, -65.35682678222656, 111.70052337646484, 104.64762115478516, -36.91429138183594, 104.69010925292969, 143.7200164794922, 40.07991027832031, 101.83003234863281, 320.6472473144531, -144.18678283691406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000240.npy"}
|
|
{"epoch": 0.3524229074889868, "step": 241, "batch_size": 64, "mean": 133.21487426757812, "std": 149.75619506835938, "min": -180.54515075683594, "p10": -14.60877075195312, "median": 112.27143859863281, "p90": 312.3628143310547, "max": 639.2767333984375, "pos_frac": 0.84375, "sample": [70.3748779296875, 85.79249572753906, -97.89417266845703, -19.519495010375977, 1.6631927490234375, 168.05999755859375, 183.12564086914062, 320.71331787109375, 80.74517822265625, 122.8234634399414, 109.75509643554688, -8.644287109375, -1.9437751770019531, 401.58074951171875, -17.16497802734375, 139.01535034179688, 0.7775001525878906, 254.4178466796875, 152.97268676757812, 114.78778076171875, 197.09523010253906, 103.00933837890625, 71.77523040771484, 81.59779357910156, 639.2767333984375, 285.2353820800781, 115.91552734375, 128.39187622070312, 474.9226379394531, 23.54889678955078, 571.301025390625, 308.57098388671875, 257.8821105957031, 59.68275451660156, 55.597259521484375, 81.43464660644531, 158.48455810546875, 92.4701156616211, -0.013580322265625, 120.16899108886719, 100.77642059326172, 58.86494445800781, 78.58514404296875, 7.755889892578125, 188.42864990234375, 47.56599426269531, 186.73641967773438, -32.77641677856445, 188.2106170654297, 161.23779296875, 57.71986389160156, 48.731971740722656, 173.8380126953125, 395.0065612792969, 184.97137451171875, 120.86267852783203, -146.99200439453125, 216.15769958496094, 227.12149047851562, 238.7320556640625, -180.54515075683594, 313.9878845214844, 62.9296760559082, -59.941925048828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000241.npy"}
|
|
{"epoch": 0.35389133627019087, "step": 242, "batch_size": 64, "mean": 95.71859741210938, "std": 122.99861145019531, "min": -166.3212890625, "p10": -29.428987121582026, "median": 71.21233749389648, "p90": 251.22441253662123, "max": 516.8119506835938, "pos_frac": 0.84375, "sample": [-38.36592102050781, 37.025386810302734, 104.43115997314453, 168.60243225097656, 39.69784164428711, -166.3212890625, 20.244155883789062, -0.29405975341796875, 218.85728454589844, 70.10818481445312, 13.833150863647461, 163.64898681640625, 429.43365478515625, 7.507612228393555, 25.33075714111328, 54.959556579589844, 64.89974212646484, 85.7252197265625, 74.89743041992188, 10.638753890991211, 185.37193298339844, 24.397216796875, 185.29713439941406, 163.2010498046875, 206.5791015625, -32.495079040527344, 121.37505340576172, 72.31649017333984, 91.20685577392578, -40.732757568359375, 130.55743408203125, 150.14886474609375, 66.1795425415039, -94.3349609375, 199.76632690429688, 74.26981353759766, 265.0960388183594, 336.61712646484375, 45.765342712402344, 24.2188720703125, 53.075653076171875, 140.51498413085938, 18.37353515625, 516.8119506835938, 38.574745178222656, 55.632469177246094, 344.6526794433594, 75.34266662597656, 25.747817993164062, 19.871095657348633, -66.80037689208984, 350.9869384765625, 110.85404968261719, -2.7464370727539062, 77.30496215820312, 61.18798828125, 114.15669250488281, -87.77105712890625, 177.47923278808594, 118.72708129882812, 7.546930313110352, 291.56988525390625, 147.51039123535156, -22.27477264404297], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000242.npy"}
|
|
{"epoch": 0.355359765051395, "step": 243, "batch_size": 64, "mean": 120.15345764160156, "std": 125.64413452148438, "min": -163.09097290039062, "p10": -17.95867156982422, "median": 100.0732307434082, "p90": 286.3038726806641, "max": 537.669189453125, "pos_frac": 0.859375, "sample": [249.9483184814453, 1.6043357849121094, 64.0915756225586, 313.829345703125, 159.14889526367188, 92.32808685302734, 90.74998474121094, 155.12841796875, -76.03218078613281, -63.73633575439453, -65.69406127929688, 63.0787353515625, 187.19448852539062, 40.389404296875, -44.65718078613281, 53.029144287109375, -9.954574584960938, 243.157470703125, 115.02885437011719, 104.58311462402344, -163.09097290039062, 23.01323699951172, 135.91677856445312, 36.172569274902344, 11.20695686340332, 236.2519073486328, 40.406349182128906, 118.41033935546875, 93.88788604736328, 252.75927734375, 210.45550537109375, 245.758544921875, -23.554296493530273, 126.4775390625, -17.26092529296875, 275.5357971191406, 184.50125122070312, 352.0760803222656, 216.47999572753906, 177.96194458007812, 250.3684844970703, 81.49427032470703, 328.4390869140625, 537.669189453125, 290.91876220703125, 53.16859436035156, 95.56334686279297, 19.498554229736328, 298.42083740234375, -18.257705688476562, 136.21109008789062, 144.55123901367188, 35.186279296875, 119.7845458984375, 372.6483459472656, 29.651527404785156, 113.16089630126953, 70.17475891113281, 2.2991714477539062, 185.1125946044922, 42.45603942871094, 59.17362976074219, 83.26799011230469, 152.27841186523438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000243.npy"}
|
|
{"epoch": 0.3568281938325991, "step": 244, "batch_size": 64, "mean": 117.43812561035156, "std": 112.46257019042969, "min": -82.00071716308594, "p10": -8.2418212890625, "median": 102.73597717285156, "p90": 288.3338195800782, "max": 408.78106689453125, "pos_frac": 0.859375, "sample": [113.85554504394531, 41.384132385253906, 74.78231811523438, 104.5335922241211, 23.199527740478516, 41.187374114990234, 330.3035888671875, 141.29246520996094, -40.888641357421875, -1.370880126953125, 125.29310607910156, 156.86502075195312, -8.172393798828125, 218.35704040527344, 35.32275390625, 111.54219055175781, 0.7920417785644531, 345.3159484863281, 18.766620635986328, 96.81421661376953, -82.00071716308594, 98.88076782226562, 210.43785095214844, -44.209625244140625, 170.3316650390625, 23.704980850219727, 83.14794158935547, 174.84439086914062, 86.35041809082031, 8.531196594238281, 192.22019958496094, 100.93836212158203, 259.41937255859375, 230.89410400390625, 330.3204345703125, 328.7984619140625, 130.26063537597656, 326.8593444824219, 44.37669372558594, 408.78106689453125, 164.36717224121094, -10.63880729675293, 257.70355224609375, 135.29843139648438, 130.8682403564453, -48.29352569580078, 72.84827423095703, 2.8141002655029297, -8.271575927734375, 142.48471069335938, 39.17097473144531, 15.545478820800781, 116.89907836914062, 56.62574768066406, 21.93133544921875, 75.62130737304688, 263.202392578125, 165.30160522460938, -13.042346954345703, 126.77682495117188, 86.46943664550781, 299.10443115234375, 151.1818389892578, 260.00823974609375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000244.npy"}
|
|
{"epoch": 0.35829662261380324, "step": 245, "batch_size": 64, "mean": 101.28717803955078, "std": 143.47035217285156, "min": -311.3232116699219, "p10": -26.476494789123528, "median": 74.00566864013672, "p90": 277.1503326416016, "max": 512.2391967773438, "pos_frac": 0.78125, "sample": [62.007286071777344, 338.0640869140625, 129.6826171875, 108.89376831054688, 133.10537719726562, 117.70936584472656, 48.414390563964844, -16.785430908203125, 168.302001953125, 95.89332580566406, 57.894378662109375, 97.2408447265625, 147.5284881591797, 36.361331939697266, 67.18348693847656, 157.6602783203125, 187.74978637695312, 218.5364532470703, 72.72341918945312, 10.690265655517578, 442.0550842285156, 113.65963745117188, -17.71666145324707, 287.52191162109375, 31.613344192504883, -5.29962158203125, 116.25218200683594, -30.230709075927734, -131.35647583007812, -3.0297927856445312, 214.6633758544922, 130.36849975585938, 512.2391967773438, -89.15473937988281, 154.6156005859375, 6.820716857910156, -10.529884338378906, 70.13801574707031, 241.25262451171875, 86.51551818847656, -6.279243469238281, 129.24468994140625, 151.0452423095703, -44.172454833984375, 46.384788513183594, -311.3232116699219, 229.9787139892578, 260.6895446777344, 29.170265197753906, 156.5673370361328, 284.2049560546875, 11.887847900390625, 476.04571533203125, -30.371185302734375, -1.7584800720214844, 472.92236328125, 75.28791809082031, 110.9212417602539, 12.4334716796875, -78.13446044921875, 55.59487533569336, 35.82423400878906, 43.997650146484375, 12.964147567749023], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000245.npy"}
|
|
{"epoch": 0.35976505139500736, "step": 246, "batch_size": 64, "mean": 98.57553100585938, "std": 138.6250762939453, "min": -150.05284118652344, "p10": -41.956742858886706, "median": 51.383989334106445, "p90": 304.5341033935547, "max": 552.3124389648438, "pos_frac": 0.828125, "sample": [148.1510009765625, 313.2065734863281, 101.80978393554688, 34.876075744628906, 307.01422119140625, 136.11399841308594, 552.3124389648438, -122.28881072998047, 14.81185531616211, -150.05284118652344, 49.785301208496094, 47.061546325683594, 51.40629959106445, 7.975107192993164, 145.92506408691406, 306.94921875, 292.89056396484375, 22.96210479736328, 33.1549072265625, 50.138916015625, 37.34425354003906, 317.6830139160156, 139.117431640625, -53.09477996826172, 97.50663757324219, -47.4556884765625, 165.15391540527344, -26.7718505859375, 272.12310791015625, -56.87443542480469, 5.2143402099609375, 10.722816467285156, 83.419677734375, 59.98008728027344, 228.64700317382812, 13.381645202636719, 34.04430389404297, 22.44559097290039, 14.048698425292969, 151.7408447265625, 298.8988342285156, 93.41327667236328, 30.271766662597656, 175.3545379638672, 51.36167907714844, 102.5006103515625, 53.382633209228516, -29.125869750976562, 5.265613555908203, 144.7616729736328, -110.57289123535156, 63.082061767578125, 216.54873657226562, 109.72129821777344, 31.552337646484375, 39.019752502441406, 12.017557144165039, 493.7989501953125, 246.36868286132812, 367.2239990234375, -20.09833526611328, 184.48052978515625, -54.42817687988281, -8.54543685913086], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000246.npy"}
|
|
{"epoch": 0.36123348017621143, "step": 247, "batch_size": 64, "mean": 144.7928009033203, "std": 154.8876495361328, "min": -242.39065551757812, "p10": -0.9169128417968715, "median": 114.97848510742188, "p90": 352.0866424560548, "max": 588.8934326171875, "pos_frac": 0.890625, "sample": [-161.2572021484375, 54.52565002441406, -2.3974761962890625, 362.00506591796875, 81.83940124511719, 422.8876953125, 130.27462768554688, 299.4764404296875, -26.1613826751709, 251.1787109375, 306.0345458984375, 412.3410339355469, 145.689697265625, 263.5716552734375, 49.68782043457031, 277.327880859375, 431.92816162109375, 27.632530212402344, 299.28302001953125, 311.49395751953125, 111.63514709472656, 35.10205078125, 16.286956787109375, 58.10376739501953, 10.959335327148438, 186.7692413330078, 18.157936096191406, 66.40193176269531, 96.1412582397461, 88.04736328125, 187.5612335205078, 248.4298858642578, 56.20923614501953, 94.96725463867188, 56.18519592285156, 318.37225341796875, 332.3610534667969, 238.1439208984375, 121.62367248535156, -75.5958480834961, 8.110820770263672, 150.5458526611328, 2.5377349853515625, 127.28994750976562, 41.267967224121094, 17.304412841796875, 118.32182312011719, 272.6584167480469, 447.5623779296875, 80.69324493408203, 30.31079864501953, -67.15017700195312, 98.20079040527344, 588.8934326171875, 164.31991577148438, -4.211311340332031, 150.63571166992188, 236.9685516357422, 165.2766876220703, -242.39065551757812, 360.54046630859375, 8.43316650390625, 79.91056823730469, 227.48382568359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000247.npy"}
|
|
{"epoch": 0.36270190895741555, "step": 248, "batch_size": 64, "mean": 119.71734619140625, "std": 106.87957000732422, "min": -77.65962219238281, "p10": 1.2081577301025401, "median": 99.8012580871582, "p90": 279.9550842285157, "max": 402.13189697265625, "pos_frac": 0.90625, "sample": [253.70687866210938, 318.7702941894531, 99.5858383178711, 147.90011596679688, 142.96778869628906, 0.7232875823974609, 87.891357421875, 297.8073425292969, 92.04032897949219, 38.954376220703125, 30.880157470703125, 233.2294921875, 152.40057373046875, 2.3395214080810547, 158.97398376464844, 17.867103576660156, 23.88536262512207, -2.6632556915283203, 97.4223403930664, 21.291240692138672, 124.96428680419922, 140.70687866210938, 193.51077270507812, 300.3772277832031, 33.598846435546875, -17.983551025390625, 135.89889526367188, 249.19630432128906, 79.09208679199219, 151.78939819335938, 117.27824401855469, -77.65962219238281, 110.10163116455078, 247.59173583984375, 87.98927307128906, 138.56887817382812, 41.239044189453125, 25.864898681640625, 15.506149291992188, 47.1186408996582, 160.96669006347656, 209.74635314941406, 165.55850219726562, 307.46942138671875, 83.5330810546875, 49.2696533203125, 402.13189697265625, 241.88449096679688, 4.4372100830078125, 327.61102294921875, 17.263160705566406, 61.54175567626953, 252.70285034179688, 169.6387176513672, 196.47186279296875, 291.2043151855469, -4.0313568115234375, 53.72993469238281, 100.01667785644531, 201.25332641601562, 52.910926818847656, 28.410816192626953, -13.22945785522461, -59.30569076538086], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000248.npy"}
|
|
{"epoch": 0.3641703377386197, "step": 249, "batch_size": 64, "mean": 70.76123046875, "std": 135.01190185546875, "min": -236.7727508544922, "p10": -81.0691146850586, "median": 56.00648880004883, "p90": 249.5760940551758, "max": 517.5519409179688, "pos_frac": 0.6875, "sample": [-45.763763427734375, 219.61305236816406, -23.559829711914062, -44.67784118652344, 83.23746490478516, -67.89812469482422, 43.753570556640625, 100.39347839355469, 63.253387451171875, 284.5131530761719, 14.196746826171875, 43.711944580078125, 46.624916076660156, 46.68463897705078, 172.1223602294922, -236.7727508544922, -66.71452331542969, 248.55853271484375, 182.6624755859375, -57.00809097290039, 106.90725708007812, 132.690185546875, -89.80139923095703, 76.66365051269531, -12.961318969726562, 264.3741455078125, -81.73069763183594, 248.99571228027344, 139.45407104492188, 114.66098022460938, -15.41622543334961, 29.724796295166016, 161.1600341796875, 26.292831420898438, 80.86458587646484, 2.5235061645507812, 135.14727783203125, -211.7584686279297, 111.59188842773438, 253.426025390625, 132.49386596679688, -10.908447265625, 517.5519409179688, -61.77397155761719, -166.737548828125, -95.50761413574219, 48.75959014892578, -8.104015350341797, 124.7115478515625, 135.8041534423828, 103.1269302368164, 180.2432861328125, -112.5511245727539, 194.85873413085938, 249.8248291015625, 211.38180541992188, 46.16253662109375, -79.52542114257812, -45.886558532714844, 88.708984375, 26.34217071533203, 292.29144287109375, 272.25872802734375, 5.452674865722656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000249.npy"}
|
|
{"epoch": 0.3656387665198238, "step": 250, "batch_size": 64, "mean": 123.74378967285156, "std": 110.35969543457031, "min": -125.26909637451172, "p10": -24.913870429992674, "median": 126.85346984863281, "p90": 275.0840118408203, "max": 375.754638671875, "pos_frac": 0.859375, "sample": [234.2977294921875, 52.854461669921875, 36.92362976074219, -23.12110710144043, -66.80694580078125, 151.01368713378906, 100.62836456298828, 319.56219482421875, 118.42678833007812, -26.465560913085938, 62.507164001464844, 350.2441101074219, 144.11129760742188, 166.89051818847656, 144.1243438720703, 126.00643920898438, 19.14794158935547, 139.28579711914062, 271.27239990234375, 236.785400390625, 127.70050048828125, -39.38970184326172, -29.43299102783203, 97.49459838867188, 69.96340942382812, 154.19284057617188, 67.64664459228516, 9.179361343383789, 67.06896209716797, 75.01123046875, 19.95400047302246, 5.223604202270508, 117.77079772949219, 230.86306762695312, -25.68219757080078, 375.754638671875, 192.64422607421875, 153.39752197265625, -42.31488800048828, 6.716514587402344, 139.4989013671875, 286.6640625, 222.1885528564453, 127.75038146972656, 133.260009765625, 60.005043029785156, 210.42938232421875, 97.13394927978516, 180.93104553222656, 100.34170532226562, 55.74650192260742, -2.5316925048828125, 142.39041137695312, 218.27163696289062, -125.26909637451172, 185.08682250976562, 276.7175598144531, 197.6208038330078, 120.52698516845703, 358.634765625, 188.5002899169922, 35.594200134277344, 306.8080139160156, 211.8513946533203], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000250.npy"}
|
|
{"epoch": 0.3671071953010279, "step": 251, "batch_size": 64, "mean": 90.27076721191406, "std": 137.9052734375, "min": -278.05010986328125, "p10": -54.07984237670898, "median": 72.65323638916016, "p90": 266.7992309570313, "max": 478.7767639160156, "pos_frac": 0.75, "sample": [341.4139099121094, 7.634899139404297, 300.47576904296875, 31.54590606689453, 176.6434326171875, -91.2606201171875, -278.05010986328125, 478.7767639160156, 31.48017120361328, 157.6262969970703, 187.40585327148438, -87.06499481201172, 240.43553161621094, 46.14904022216797, 58.52226638793945, -50.918739318847656, -213.55111694335938, 349.6830139160156, -24.663137435913086, 140.9662628173828, 37.47317123413086, 189.07188415527344, 1.1370620727539062, -15.572364807128906, -5.071075439453125, 94.04493713378906, -92.25039672851562, -55.434600830078125, 88.21023559570312, 248.720458984375, 43.561279296875, -67.14842224121094, 257.49774169921875, -9.976310729980469, 62.06272888183594, 170.47955322265625, 40.522911071777344, 93.4802017211914, 114.82220458984375, 114.22232055664062, -6.8184051513671875, 145.77267456054688, -2.6291751861572266, 256.4522705078125, 0.7371444702148438, 280.1313781738281, 125.18006896972656, 178.0084686279297, 66.36334991455078, 78.94312286376953, -9.422384262084961, 21.113853454589844, 428.88336181640625, 99.49617004394531, 270.78558349609375, 174.62353515625, -13.727489471435547, 174.07208251953125, 79.65084838867188, 1.1484184265136719, 35.40032196044922, 147.30435180664062, 37.421043395996094, 95.33460998535156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000251.npy"}
|
|
{"epoch": 0.368575624082232, "step": 252, "batch_size": 64, "mean": 117.80094909667969, "std": 104.40287017822266, "min": -320.0909423828125, "p10": 2.0731681823730477, "median": 126.83620071411133, "p90": 249.30050201416017, "max": 318.37078857421875, "pos_frac": 0.90625, "sample": [220.16574096679688, 213.81472778320312, 61.09083938598633, 184.6436004638672, 207.84945678710938, 47.013763427734375, 137.92459106445312, 219.2667236328125, 74.37130737304688, 146.55905151367188, 147.21803283691406, 38.881011962890625, 153.7050323486328, 30.647018432617188, 96.96900939941406, 137.674560546875, 79.21693420410156, 72.85780334472656, 160.43856811523438, 1.7870864868164062, 231.4840545654297, 154.4890594482422, 21.151439666748047, 259.4754333496094, 192.13290405273438, -67.73368835449219, -14.45631217956543, 79.48678588867188, 2.740692138671875, 119.74844360351562, -26.292224884033203, 113.60540771484375, 84.83955383300781, 163.76931762695312, 81.48919677734375, 96.02001953125, 129.54708862304688, 300.7276306152344, 297.3231201171875, 251.85501098632812, 57.79975891113281, 122.73577880859375, 131.9452667236328, 15.599334716796875, 149.71258544921875, 261.78790283203125, 258.5281982421875, 165.79873657226562, 148.70822143554688, 169.2145233154297, 15.222488403320312, -320.0909423828125, 243.33998107910156, 224.2480926513672, 64.13873291015625, 82.84440612792969, -6.983522415161133, 124.12531280517578, 318.37078857421875, -22.227840423583984, 66.66801452636719, 46.247215270996094, 180.39385986328125, 137.63629150390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000252.npy"}
|
|
{"epoch": 0.3700440528634361, "step": 253, "batch_size": 64, "mean": 109.68413543701172, "std": 132.4770965576172, "min": -282.2687072753906, "p10": -33.70382118225098, "median": 97.9322738647461, "p90": 296.36021423339844, "max": 453.0985107421875, "pos_frac": 0.796875, "sample": [-30.94382667541504, 10.320167541503906, 224.57537841796875, 21.902252197265625, 12.168472290039062, 97.49738311767578, 183.74563598632812, 79.32061004638672, 203.29591369628906, 87.23460388183594, 82.05340576171875, 453.0985107421875, 188.8490753173828, 46.038658142089844, 145.83245849609375, 113.5599365234375, 294.169189453125, 77.06147766113281, -44.73854064941406, 88.62226104736328, -15.287687301635742, -282.2687072753906, 151.61215209960938, 302.7252502441406, 297.2992248535156, 8.171276092529297, 362.6534423828125, -12.972160339355469, -8.534095764160156, 37.18118667602539, 14.393531799316406, -11.84187126159668, 85.52792358398438, 177.85870361328125, 255.87228393554688, 178.82467651367188, 170.28436279296875, -103.86966705322266, 275.580078125, 98.3671646118164, 199.40142822265625, 263.53826904296875, 118.67687225341797, 224.27313232421875, 308.0777587890625, 61.36585998535156, 138.11390686035156, 297.786376953125, 80.54561614990234, -172.89697265625, 191.26980590820312, 132.3924560546875, -65.54501342773438, 55.34587860107422, -36.97264099121094, 59.20915985107422, 115.68869018554688, 159.33985900878906, 186.81727600097656, -34.874786376953125, -30.971569061279297, 25.115020751953125, 330.1145324707031, 98.73410034179688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000253.npy"}
|
|
{"epoch": 0.37151248164464024, "step": 254, "batch_size": 64, "mean": 99.6583251953125, "std": 151.4174041748047, "min": -160.5799102783203, "p10": -64.59278907775878, "median": 71.80940246582031, "p90": 286.84734497070315, "max": 599.7950439453125, "pos_frac": 0.75, "sample": [73.34539794921875, -0.2987060546875, 135.29556274414062, 63.92327880859375, 93.75958251953125, 109.50625610351562, -3.4098052978515625, -51.03087615966797, 84.23927307128906, 34.468536376953125, 130.4986572265625, 155.42112731933594, 186.04293823242188, 70.27340698242188, -59.14504623413086, -160.5799102783203, 153.19320678710938, -10.804607391357422, 106.01849365234375, 135.98046875, -19.786266326904297, 184.98727416992188, 387.1507873535156, 256.0256042480469, 288.7999267578125, -124.49591064453125, 282.29132080078125, 257.2024841308594, -44.823646545410156, 4.835321426391602, 53.131080627441406, 183.1795654296875, 63.75751495361328, -146.21707153320312, 24.409626007080078, 166.07821655273438, -28.48180389404297, 357.2659912109375, -68.51921081542969, 55.018192291259766, -16.519447326660156, -147.955322265625, 39.205406188964844, 343.63775634765625, 14.060800552368164, 189.59800720214844, 3.118753433227539, 281.87860107421875, 41.809715270996094, 136.18304443359375, 599.7950439453125, 111.04246520996094, 60.271087646484375, 3.4020233154296875, -125.2144546508789, 448.39215087890625, 129.69268798828125, 188.64407348632812, 139.34426879882812, 67.83088684082031, -66.92753601074219, 443.3472595214844, 95.98944091796875, 18.999725341796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000254.npy"}
|
|
{"epoch": 0.37298091042584436, "step": 255, "batch_size": 64, "mean": 106.40800476074219, "std": 143.49526977539062, "min": -322.44232177734375, "p10": -32.858056640624994, "median": 84.5599479675293, "p90": 291.2938995361328, "max": 453.493408203125, "pos_frac": 0.796875, "sample": [453.493408203125, -48.48480224609375, 18.48968505859375, 192.41664123535156, 134.39959716796875, -14.017578125, -33.97199249267578, -4.9423065185546875, -322.44232177734375, 21.582515716552734, 249.9751739501953, 79.96763610839844, 2.74700927734375, 27.566539764404297, 181.6930389404297, 155.94100952148438, -124.00434112548828, -204.85134887695312, 127.4508056640625, 224.13943481445312, 217.61907958984375, 47.96318054199219, 54.83568572998047, 26.76932144165039, 88.26280212402344, -0.8612594604492188, -69.82012176513672, 317.13323974609375, 5.826038360595703, 158.6312255859375, 92.07127380371094, 157.9481658935547, 93.06617736816406, 293.4363708496094, 299.5010681152344, 251.6529541015625, 229.41818237304688, 301.32470703125, 286.2947998046875, 450.49029541015625, 16.996925354003906, -16.294160842895508, 230.28311157226562, 124.29747009277344, 25.128475189208984, 149.30369567871094, 57.68473815917969, 172.532958984375, -106.17512512207031, 272.2137451171875, 351.1187744140625, 58.709716796875, 121.88968658447266, 29.95696258544922, -8.349967956542969, 75.8228759765625, 51.852088928222656, 80.85709381103516, 22.892662048339844, 156.55224609375, -30.258872985839844, 56.6496467590332, 243.53286743164062, 254.20359802246094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000255.npy"}
|
|
{"epoch": 0.3744493392070485, "step": 256, "batch_size": 64, "mean": 113.09561157226562, "std": 134.13714599609375, "min": -166.339111328125, "p10": -38.17844467163086, "median": 103.7409896850586, "p90": 278.43906250000003, "max": 466.61370849609375, "pos_frac": 0.78125, "sample": [98.45584106445312, 454.3909912109375, 84.19879913330078, 200.5933074951172, 153.295654296875, 219.54464721679688, 32.58674621582031, 149.72796630859375, -32.04806137084961, 24.02252197265625, 227.3880157470703, 194.4761962890625, 151.79629516601562, 250.72216796875, 212.28857421875, -93.87893676757812, 60.69332504272461, 168.1107635498047, -166.339111328125, 30.86993408203125, 168.14845275878906, 82.13520812988281, -119.38954162597656, 221.4608612060547, 54.54489517211914, 209.57965087890625, -39.709190368652344, -46.71026611328125, -42.32328796386719, 139.6821746826172, 313.4648132324219, -1.2162017822265625, 168.49310302734375, 90.14596557617188, 23.244720458984375, 113.82239532470703, 283.8321228027344, -14.9635009765625, 59.78190612792969, 41.0196533203125, -142.12510681152344, 23.667388916015625, 11.882835388183594, 29.606781005859375, 90.83930969238281, -8.436695098876953, 148.77084350585938, 114.627685546875, -14.605575561523438, 210.43609619140625, -34.60670471191406, 109.02613830566406, 25.47905731201172, 182.07464599609375, 286.8046875, 301.6854553222656, 126.8857421875, 58.540531158447266, 466.61370849609375, 420.34033203125, 179.63900756835938, 261.6673278808594, -2.4890193939208984, 265.8552551269531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000256.npy"}
|
|
{"epoch": 0.37591776798825255, "step": 257, "batch_size": 64, "mean": 120.23841857910156, "std": 141.38995361328125, "min": -162.28683471679688, "p10": -26.402582550048827, "median": 97.55659103393555, "p90": 343.3350982666016, "max": 491.0191345214844, "pos_frac": 0.8125, "sample": [361.861572265625, 1.0213661193847656, 30.392959594726562, 94.90245056152344, 285.5790100097656, 0.838653564453125, 178.16238403320312, 330.6195068359375, 197.3489990234375, 131.85427856445312, -3.2940196990966797, -34.944374084472656, 71.96086883544922, 258.42034912109375, 192.795166015625, 29.92987060546875, 173.8142547607422, 171.29124450683594, 24.631282806396484, 348.7846374511719, 50.98796081542969, 357.7177429199219, 242.67059326171875, -98.77455139160156, 87.31539154052734, 99.14857482910156, -21.952373504638672, 1.1239395141601562, -24.55651092529297, 355.3833923339844, 177.0216522216797, 194.06602478027344, 223.80551147460938, 491.0191345214844, 81.02227020263672, -27.193756103515625, 96.76626586914062, 5.009603500366211, 122.92908477783203, -162.28683471679688, 65.23055267333984, 326.04461669921875, 28.852405548095703, 220.53955078125, 108.44511413574219, 160.01800537109375, 62.93700408935547, 1.6468582153320312, 395.229248046875, 173.51544189453125, 165.26953125, 254.9291534423828, -38.924903869628906, 98.34691619873047, 43.47639465332031, 421.9351501464844, -65.11866760253906, -11.270635604858398, 106.3082275390625, -111.17628479003906, 30.438400268554688, 7.4220733642578125, -16.54266357421875, 170.513671875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000257.npy"}
|
|
{"epoch": 0.37738619676945667, "step": 258, "batch_size": 64, "mean": 108.45994567871094, "std": 145.51258850097656, "min": -333.8253479003906, "p10": -45.792788314819326, "median": 112.99049758911133, "p90": 268.59180297851566, "max": 498.4493103027344, "pos_frac": 0.78125, "sample": [217.5057373046875, 11.599931716918945, -38.288021087646484, 224.41748046875, 57.90455627441406, 111.87210845947266, 209.16213989257812, 88.27880859375, 161.08602905273438, 6.268402099609375, -29.403535842895508, -49.13726043701172, -26.38313865661621, 56.74957275390625, 68.77566528320312, -18.09954833984375, -50.33025360107422, 93.91415405273438, 169.5711212158203, -2.7807998657226562, 25.349777221679688, 228.51416015625, 425.54278564453125, 169.39588928222656, 324.7362976074219, 212.1038818359375, 117.6510009765625, 24.690345764160156, 94.7113037109375, 263.51470947265625, 114.10888671875, 181.71713256835938, 244.2343292236328, 127.27314758300781, -159.35885620117188, 78.48654174804688, 230.40380859375, -278.52581787109375, 36.39179229736328, 95.36270904541016, 302.41046142578125, -5.221065521240234, -49.009117126464844, 270.7677001953125, 26.07430648803711, -333.8253479003906, 498.4493103027344, 138.1800537109375, -92.96851348876953, 128.199462890625, 141.536376953125, -29.007827758789062, 218.63856506347656, 225.36709594726562, 330.4582214355469, 47.723609924316406, 228.18479919433594, 297.9459533691406, 141.25497436523438, 200.66824340820312, 50.76405334472656, 75.94031524658203, 187.8968505859375, 122.0207290649414], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000258.npy"}
|
|
{"epoch": 0.3788546255506608, "step": 259, "batch_size": 64, "mean": 98.13949584960938, "std": 134.16647338867188, "min": -170.46339416503906, "p10": -64.2553253173828, "median": 63.631202697753906, "p90": 276.52532043457035, "max": 372.7879638671875, "pos_frac": 0.765625, "sample": [187.04946899414062, -14.98625373840332, 156.43502807617188, 145.45811462402344, 19.192745208740234, -128.51206970214844, 241.14883422851562, 347.75701904296875, 332.9944763183594, 197.52394104003906, 34.556068420410156, -15.944969177246094, 127.27168273925781, -31.296913146972656, 18.463348388671875, -71.59977722167969, 233.49618530273438, -29.857711791992188, 368.18548583984375, 159.41580200195312, 99.28443908691406, 76.12030029296875, 46.333770751953125, 44.03706741333008, 229.0909881591797, 12.021854400634766, 53.10057067871094, 63.65281677246094, 372.7879638671875, -115.67976379394531, 27.63671875, 40.045440673828125, 370.07525634765625, 150.61021423339844, 44.233238220214844, -47.18324279785156, 106.73021697998047, 286.952392578125, -81.7504653930664, 277.55950927734375, -100.07864379882812, 22.736770629882812, 251.17518615722656, -15.463573455810547, 63.609588623046875, 31.729095458984375, 274.1122131347656, 233.00665283203125, -170.46339416503906, 96.30003356933594, -27.271032333374023, 261.82183837890625, 237.59210205078125, 18.8934326171875, 251.20172119140625, -68.46892547607422, 82.11845397949219, 213.74142456054688, 112.4013900756836, -54.42359161376953, 54.45225143432617, 27.313091278076172, 108.58914947509766, 43.89228057861328], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000259.npy"}
|
|
{"epoch": 0.3803230543318649, "step": 260, "batch_size": 64, "mean": 117.58509826660156, "std": 130.2021942138672, "min": -117.71946716308594, "p10": -6.117125701904295, "median": 96.08465576171875, "p90": 293.7619201660157, "max": 582.6774291992188, "pos_frac": 0.84375, "sample": [211.23291015625, -81.45406341552734, -16.448875427246094, -64.79084014892578, 44.915245056152344, 123.99496459960938, 96.45112609863281, 103.20112609863281, 216.1190185546875, 46.18316650390625, 31.721576690673828, 381.2221984863281, 88.43936157226562, 67.55731201171875, 112.62751770019531, 582.6774291992188, 152.25051879882812, 139.12796020507812, 13.89849853515625, 369.75946044921875, -117.71946716308594, 99.22986602783203, 22.427492141723633, 152.0037841796875, -81.73381805419922, -36.35956573486328, 141.81309509277344, -4.376861572265625, 181.88690185546875, 94.97636413574219, 83.4742660522461, 270.2418518066406, -6.862953186035156, 169.16859436035156, 99.52116394042969, 235.09033203125, 313.2704772949219, 107.85497283935547, 95.71818542480469, 53.261077880859375, 43.77537155151367, 54.495208740234375, 259.8070068359375, 63.81855010986328, 123.6236801147461, 51.64223861694336, -0.7264251708984375, 73.88775634765625, 132.70187377929688, 162.8086395263672, 137.912841796875, 123.71155548095703, 44.96076202392578, 337.8518371582031, 47.59557342529297, -3.8765029907226562, 31.29083251953125, 257.59442138671875, 13.526748657226562, 461.95068359375, 56.706207275390625, 303.8419494628906, 30.787628173828125, 224.18646240234375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000260.npy"}
|
|
{"epoch": 0.38179148311306904, "step": 261, "batch_size": 64, "mean": 121.63386535644531, "std": 141.7743682861328, "min": -227.25067138671875, "p10": -27.099716186523434, "median": 103.71771621704102, "p90": 291.3938659667969, "max": 637.8239135742188, "pos_frac": 0.8125, "sample": [283.9543151855469, 248.88003540039062, 23.75360870361328, -119.61109924316406, 93.7601318359375, 637.8239135742188, 45.33392333984375, 340.59033203125, -14.612110137939453, 145.68759155273438, 20.914045333862305, 66.39026641845703, -28.137786865234375, 292.974853515625, 54.49932861328125, 231.91802978515625, 185.71353149414062, 232.84445190429688, 266.136474609375, -29.681808471679688, 80.24385070800781, 79.302978515625, 112.91216278076172, 164.04412841796875, 161.2682342529297, 287.70489501953125, 304.5867614746094, 52.117156982421875, 270.2794494628906, 159.7828826904297, 16.975189208984375, 62.32776641845703, 9.36868667602539, 93.70649719238281, 132.2357940673828, 33.76543426513672, 235.78543090820312, 280.29345703125, 181.5423126220703, 66.57569122314453, 217.57369995117188, -9.681026458740234, 302.6790466308594, 305.9419860839844, 41.776206970214844, 257.05303955078125, -119.16935729980469, 160.1126251220703, 17.66046905517578, 161.5694580078125, 109.8391342163086, 352.9912109375, -1.2698135375976562, -227.25067138671875, -136.57125854492188, 97.59629821777344, -30.797119140625, -8.642295837402344, 129.93829345703125, 167.6947479248047, 54.795738220214844, 85.33577728271484, 116.11831665039062, -24.67755126953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000261.npy"}
|
|
{"epoch": 0.3832599118942731, "step": 262, "batch_size": 64, "mean": 98.66909790039062, "std": 114.92432403564453, "min": -198.33935546875, "p10": -39.41488227844238, "median": 107.39225769042969, "p90": 203.98177337646487, "max": 470.78277587890625, "pos_frac": 0.8125, "sample": [67.96005249023438, 240.57809448242188, 91.42424774169922, 173.53445434570312, -124.41500854492188, 206.4549560546875, 370.5975341796875, 102.23902893066406, 109.596435546875, 172.365234375, -118.5356674194336, -198.33935546875, 87.66999816894531, 470.78277587890625, 66.61834716796875, -107.78234100341797, 159.56210327148438, 168.3785400390625, -12.499776840209961, 130.2738037109375, 119.58761596679688, 59.44532012939453, -61.61765670776367, 179.97360229492188, 112.17799377441406, 6.4162750244140625, -0.048664093017578125, 311.2720947265625, 141.451904296875, 229.50286865234375, 44.61781311035156, 64.4111557006836, 68.83587646484375, -33.91746139526367, 87.36686706542969, 90.29642486572266, 105.18807983398438, 20.762771606445312, 157.19744873046875, 180.0716552734375, 139.1975860595703, 134.9218292236328, 73.3757553100586, 175.66558837890625, 180.93629455566406, 164.79550170898438, 197.32818603515625, 72.58513641357422, -29.94469451904297, -5.534921646118164, 15.269424438476562, 88.80059814453125, 6.5357818603515625, 207.96749877929688, -100.07376098632812, 8.232917785644531, 124.43841552734375, 151.5426025390625, 198.2110137939453, 195.32269287109375, 150.98577880859375, -41.77091979980469, 156.70497131347656, 109.87334442138672], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000262.npy"}
|
|
{"epoch": 0.38472834067547723, "step": 263, "batch_size": 64, "mean": 139.3936309814453, "std": 137.85968017578125, "min": -188.83106994628906, "p10": -29.940345573425283, "median": 134.72117614746094, "p90": 319.4784790039063, "max": 454.12493896484375, "pos_frac": 0.84375, "sample": [241.605712890625, -188.83106994628906, 128.83233642578125, 95.79545593261719, 261.81964111328125, 39.544639587402344, 261.92108154296875, 70.98587036132812, 37.77702331542969, 149.52850341796875, -148.3489990234375, 143.94390869140625, 74.03820037841797, 215.0917205810547, 234.35702514648438, 150.7113494873047, 16.652502059936523, 228.85121154785156, 87.48995208740234, 135.51138305664062, -20.043100357055664, -37.360321044921875, 65.12158966064453, 294.466552734375, -44.716312408447266, 2.675689697265625, 204.4107666015625, 96.93585205078125, -4.640892028808594, 59.20524978637695, 68.506103515625, 15.119119644165039, 304.3609924316406, 55.85688781738281, 12.386260986328125, 310.3267822265625, 238.4703369140625, 77.63497924804688, 338.9106750488281, 211.8944091796875, 56.63822937011719, 323.400634765625, 124.84269714355469, 234.5660400390625, 306.4592590332031, 133.93096923828125, 395.33367919921875, -10.402236938476562, 190.46485900878906, 233.99429321289062, 231.76312255859375, 337.8854675292969, 454.12493896484375, -95.73094177246094, -34.18202209472656, 354.982421875, 40.18498992919922, -45.43494415283203, 205.0132293701172, 171.23583984375, 136.92942810058594, 416.7774353027344, 133.7730712890625, 137.84304809570312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000263.npy"}
|
|
{"epoch": 0.38619676945668135, "step": 264, "batch_size": 64, "mean": 115.5482177734375, "std": 130.22593688964844, "min": -183.602294921875, "p10": -12.874418640136716, "median": 88.26299285888672, "p90": 313.7861663818359, "max": 459.8367919921875, "pos_frac": 0.84375, "sample": [190.6607666015625, 245.62667846679688, 46.097049713134766, 341.0743408203125, 90.10087585449219, -125.34515380859375, -9.002479553222656, 324.1669921875, 70.43783569335938, 313.8387451171875, -6.082206726074219, 45.324302673339844, 99.75927734375, -21.372642517089844, 118.91175079345703, 120.12227630615234, 225.27236938476562, 238.61004638671875, 235.11114501953125, 160.1299591064453, 101.89279174804688, 86.42510986328125, 128.80633544921875, 146.6249542236328, 46.77068328857422, 49.96669006347656, 45.672950744628906, 43.14790344238281, 31.22376251220703, 109.09321594238281, -46.81169128417969, 26.12885284423828, 36.11573028564453, 359.9453125, 3.2365646362304688, 377.5730895996094, 100.03974151611328, 150.206298828125, 83.42597961425781, 190.96969604492188, 76.45126342773438, 125.00440979003906, 86.32247924804688, 37.87763977050781, 28.662757873535156, 459.8367919921875, 239.77081298828125, -183.602294921875, 52.282936096191406, -79.00626373291016, 176.58897399902344, 17.439868927001953, 146.75946044921875, 366.8279724121094, -6.7625732421875, 64.17674255371094, 81.7902603149414, 150.40255737304688, -116.37083435058594, 313.6634826660156, 284.9992370605469, 271.9269104003906, -14.533821105957031, 40.680999755859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000264.npy"}
|
|
{"epoch": 0.3876651982378855, "step": 265, "batch_size": 64, "mean": 112.95767211914062, "std": 163.32020568847656, "min": -317.3308410644531, "p10": -36.32170867919922, "median": 92.50582885742188, "p90": 344.6746948242188, "max": 541.427490234375, "pos_frac": 0.765625, "sample": [101.30110931396484, 332.6750183105469, 284.96759033203125, 177.69505310058594, 70.80243682861328, 107.60777282714844, 205.6976318359375, 7.454154968261719, 487.1551208496094, 53.9892578125, 213.39096069335938, -148.1002197265625, -1.0168113708496094, -4.901645660400391, -20.743072509765625, 42.54254913330078, 157.82373046875, -91.950927734375, 87.99332427978516, -31.80279541015625, 211.75772094726562, 541.427490234375, -317.3308410644531, 384.20794677734375, -60.1112060546875, 135.81918334960938, 78.05996704101562, 384.4452209472656, -30.445476531982422, 532.0728759765625, -204.59986877441406, 136.7940673828125, 50.658782958984375, -33.47251510620117, 24.329437255859375, -35.027313232421875, 13.830780029296875, 39.60858154296875, 253.12203979492188, 20.564224243164062, 349.8174133300781, 82.04107666015625, 73.49971771240234, 249.33953857421875, -36.87644958496094, 108.72505187988281, 6.810844421386719, 129.16217041015625, -6.1653289794921875, 149.85682678222656, 44.43834686279297, 90.20068359375, 183.43032836914062, 296.64849853515625, 107.93783569335938, 123.61909484863281, 175.06040954589844, 95.42950439453125, 94.81097412109375, -123.73399353027344, 352.2480163574219, 84.018310546875, 231.9708709716797, 208.70973205566406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000265.npy"}
|
|
{"epoch": 0.3891336270190896, "step": 266, "batch_size": 64, "mean": 109.93594360351562, "std": 133.7215576171875, "min": -200.8842315673828, "p10": -41.30707015991211, "median": 80.65837478637695, "p90": 307.2008117675782, "max": 430.2483825683594, "pos_frac": 0.796875, "sample": [37.65129089355469, 394.5120849609375, 64.96794891357422, 70.33490753173828, 182.23780822753906, 141.2751007080078, 61.58067321777344, 255.54092407226562, 95.94989013671875, 330.25054931640625, 164.16854858398438, 255.35902404785156, -33.47004699707031, 251.14697265625, -45.01872253417969, 55.164344787597656, 257.698486328125, 78.94277954101562, 153.5074462890625, 27.328636169433594, 89.88188171386719, -200.8842315673828, -51.105201721191406, 351.787353515625, 103.8259506225586, 263.1880798339844, 69.89918518066406, 30.994361877441406, 28.95293426513672, 317.10906982421875, 373.36053466796875, 187.2783203125, 430.2483825683594, 284.08154296875, 180.17575073242188, -13.638092041015625, 133.2618865966797, 205.6973876953125, 38.127113342285156, 84.53731536865234, 72.73213195800781, 100.95545959472656, 82.37397003173828, 33.34162139892578, 149.30215454101562, 20.52255630493164, -112.68666076660156, -114.3458251953125, 218.3496856689453, 94.47716522216797, -42.07794189453125, -39.50836944580078, 42.60005187988281, 76.30853271484375, -6.895149230957031, -12.955497741699219, -3.869943618774414, 18.21009063720703, -65.84302520751953, 342.385498046875, 223.83270263671875, 181.89801025390625, 5.303674697875977, 65.58108520507812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000266.npy"}
|
|
{"epoch": 0.39060205580029367, "step": 267, "batch_size": 64, "mean": 119.896484375, "std": 136.2825164794922, "min": -321.8373107910156, "p10": -14.187271118164062, "median": 117.33602523803711, "p90": 280.7080078125, "max": 472.19488525390625, "pos_frac": 0.828125, "sample": [90.34198760986328, 176.291015625, 197.38673400878906, -35.35420608520508, 129.4561767578125, -19.14442253112793, 206.603515625, 159.7593231201172, -20.12066650390625, 61.81389236450195, 14.899444580078125, -5.357269287109375, -9.08001708984375, 235.03207397460938, 127.78362274169922, 118.60786437988281, 33.838233947753906, 218.07168579101562, 362.064453125, 472.19488525390625, 79.13265991210938, 98.72614288330078, 116.0641860961914, -9.8599853515625, -34.17472839355469, 79.86976623535156, 75.91452026367188, 62.90886688232422, 243.8482208251953, 169.9617462158203, 311.959716796875, 281.2262878417969, 446.748046875, 63.761932373046875, 21.476207733154297, 185.14859008789062, 8.975994110107422, -13.461219787597656, 27.017704010009766, 125.81476593017578, 279.4986877441406, 10.4815673828125, 0.6599235534667969, 451.4453430175781, -321.8373107910156, -14.498435974121094, -149.9738006591797, 151.50033569335938, 141.62109375, 110.66505432128906, 154.36346435546875, 173.45626831054688, 41.286861419677734, 262.3197937011719, 122.40778350830078, 142.63841247558594, 18.679031372070312, 285.7418518066406, 67.27307891845703, 137.3898162841797, 233.29635620117188, 149.48831176757812, 256.8497009277344, 112.4743881225586], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000267.npy"}
|
|
{"epoch": 0.3920704845814978, "step": 268, "batch_size": 64, "mean": 114.72981262207031, "std": 146.72447204589844, "min": -242.46710205078125, "p10": -43.31502933502196, "median": 116.90542602539062, "p90": 257.0099243164063, "max": 585.5128784179688, "pos_frac": 0.796875, "sample": [12.568336486816406, 322.60870361328125, 102.72235870361328, -79.5425796508789, 84.45603942871094, 74.86835479736328, 139.48854064941406, 102.12408447265625, 239.73663330078125, 143.2774658203125, 130.61550903320312, -152.13565063476562, 147.00137329101562, 51.20872497558594, 248.0130615234375, -111.39569091796875, 76.54439544677734, 31.861543655395508, -151.65277099609375, 135.00120544433594, 137.98306274414062, 585.5128784179688, 65.27873229980469, 155.92454528808594, 46.21906661987305, 199.3128204345703, 265.31805419921875, 391.92974853515625, -26.27996826171875, 260.86572265625, -72.1148910522461, 34.12480163574219, 33.433074951171875, 29.426891326904297, 418.52618408203125, 87.08329772949219, 153.86318969726562, 123.8951416015625, 170.1998291015625, 244.7000274658203, 157.30502319335938, 212.98577880859375, 9.461448669433594, -49.323516845703125, 200.31422424316406, 46.16387939453125, 190.41099548339844, 237.68582153320312, -2.415740966796875, 199.65673828125, 146.26393127441406, 130.92173767089844, 528.8848266601562, 190.78350830078125, 66.13232421875, -12.986495971679688, 66.57136535644531, -7.764091491699219, 220.04281616210938, -29.295225143432617, 109.91571044921875, -242.46710205078125, -8.005847930908203, 128.8940887451172], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000268.npy"}
|
|
{"epoch": 0.3935389133627019, "step": 269, "batch_size": 64, "mean": 146.7906494140625, "std": 168.16656494140625, "min": -280.02667236328125, "p10": -40.34431915283203, "median": 147.35305786132812, "p90": 377.7395080566407, "max": 515.0316162109375, "pos_frac": 0.78125, "sample": [126.94898986816406, 286.75811767578125, 174.7407684326172, 197.73809814453125, -43.016624450683594, 213.49810791015625, 285.1072692871094, 158.67041015625, -7.7694244384765625, -34.10894012451172, 229.3377227783203, -13.662429809570312, 352.5043029785156, 57.14106750488281, 315.298095703125, 172.55584716796875, 19.003257751464844, 125.79786682128906, 16.245864868164062, -280.02667236328125, 228.62991333007812, 16.488758087158203, 188.50714111328125, -140.38153076171875, -44.44124221801758, 324.36688232421875, -3.9895858764648438, 442.6051940917969, 416.876708984375, 69.76605224609375, 151.14923095703125, -25.3392333984375, 348.85565185546875, 266.2098083496094, 35.16346740722656, 264.21258544921875, 291.5331115722656, 112.16881561279297, 268.4013671875, 31.2265625, -14.847541809082031, 123.22636413574219, -113.29966735839844, 12.117034912109375, 130.98695373535156, 428.8995361328125, 202.37109375, 342.88458251953125, 304.69561767578125, 17.899917602539062, 239.63943481445312, 143.556884765625, 47.4935302734375, 515.0316162109375, -62.00498962402344, 417.9316101074219, 23.985931396484375, 29.96105194091797, -10.993480682373047, 162.82489013671875, 396.3125, 388.5545959472656, -151.10650634765625, 223.70928955078125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000269.npy"}
|
|
{"epoch": 0.39500734214390604, "step": 270, "batch_size": 64, "mean": 105.11456298828125, "std": 146.1101837158203, "min": -270.2066345214844, "p10": -53.37800750732422, "median": 92.57122039794922, "p90": 305.77003173828126, "max": 526.9364624023438, "pos_frac": 0.765625, "sample": [215.00035095214844, 262.367919921875, 303.8736572265625, -188.84315490722656, -270.2066345214844, -5.0073089599609375, 34.05388641357422, 186.77352905273438, 98.04141998291016, 210.04928588867188, 326.71795654296875, -21.411468505859375, 84.05840301513672, 166.83935546875, 8.761199951171875, 20.66986083984375, 86.19915008544922, 115.93931579589844, -61.69629669189453, 128.33462524414062, 264.03961181640625, 114.72575378417969, -53.71660614013672, 196.47311401367188, 317.04730224609375, 176.27703857421875, -5.97722053527832, 69.95072174072266, 142.52914428710938, 87.88309478759766, -149.69688415527344, 335.650634765625, 526.9364624023438, -77.91777038574219, -49.233367919921875, 193.13970947265625, 284.7657470703125, 45.90818786621094, 88.65394592285156, 21.331966400146484, 443.2873840332031, -87.61752319335938, 128.23306274414062, 8.16079330444336, 113.76576232910156, 118.84061431884766, 5.0692291259765625, 177.91998291015625, -33.26762390136719, 40.18034744262695, 179.27279663085938, 216.43951416015625, 72.04338073730469, -3.5003833770751953, 31.469070434570312, 89.97787475585938, -28.060455322265625, 15.284088134765625, -52.58794403076172, 323.0417175292969, 147.29055786132812, 191.05661010742188, 95.16456604003906, 306.582763671875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000270.npy"}
|
|
{"epoch": 0.3964757709251101, "step": 271, "batch_size": 64, "mean": 127.55348205566406, "std": 126.66629028320312, "min": -79.04463195800781, "p10": -5.353232765197753, "median": 96.00717163085938, "p90": 309.91431579589846, "max": 479.94598388671875, "pos_frac": 0.859375, "sample": [479.94598388671875, 251.03695678710938, 94.4100341796875, 0.6010322570800781, 167.36355590820312, -4.118762969970703, 220.765380859375, 280.6538391113281, 83.7914810180664, 42.47490692138672, 199.21148681640625, -79.04463195800781, 376.30194091796875, -38.24726104736328, 90.69092559814453, 14.003044128417969, 147.5643310546875, 31.256813049316406, 118.89225769042969, -5.722625732421875, -43.88673400878906, 34.847259521484375, 311.42877197265625, 273.012939453125, 372.78466796875, 257.58782958984375, 52.877159118652344, -4.491315841674805, 77.84993743896484, 140.46774291992188, 1.2625007629394531, 97.95838928222656, 107.0914306640625, -11.261848449707031, -31.027740478515625, 107.40461730957031, 36.49031066894531, 36.57891082763672, 229.49322509765625, 96.83407592773438, 6.489479064941406, 364.3565673828125, 68.62972259521484, 15.867109298706055, 110.02793884277344, 39.93918228149414, 81.25643920898438, 69.16227722167969, 266.87786865234375, 146.08358764648438, 407.64300537109375, 229.27761840820312, 85.82254791259766, 29.284957885742188, 323.7991943359375, -26.812828063964844, 163.11090087890625, 306.3805847167969, 95.18026733398438, 224.6460723876953, 58.537330627441406, 246.85098266601562, 115.22500610351562, 120.65457153320312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000271.npy"}
|
|
{"epoch": 0.39794419970631423, "step": 272, "batch_size": 64, "mean": 148.55596923828125, "std": 142.92333984375, "min": -179.35552978515625, "p10": -33.18047943115234, "median": 155.87381744384766, "p90": 338.71860351562503, "max": 479.2953796386719, "pos_frac": 0.828125, "sample": [186.0271759033203, 167.00778198242188, 7.880155563354492, -23.799118041992188, -15.690628051757812, 287.82965087890625, 158.4634552001953, 365.18292236328125, 341.66473388671875, 215.8472137451172, 269.3921813964844, 281.25531005859375, -79.24555969238281, 112.82069396972656, -179.35552978515625, 72.85125732421875, 20.383277893066406, 154.99388122558594, 148.23690795898438, 93.80826568603516, -35.240386962890625, 243.8427734375, 224.70040893554688, 140.5067138671875, 187.28271484375, 39.81525421142578, 11.481220245361328, 258.36041259765625, 85.55884552001953, 451.5966796875, 121.76216125488281, 136.9532470703125, 166.3127899169922, 120.55093383789062, 56.00904846191406, 266.9037780761719, 388.26251220703125, -154.4831085205078, 159.87103271484375, 156.75375366210938, 479.2953796386719, 55.024566650390625, 396.4521484375, 138.9271697998047, 165.82952880859375, -48.239654541015625, 231.403564453125, 52.92710494995117, 199.63255310058594, 331.6419982910156, -29.385604858398438, 174.5356903076172, -24.157867431640625, -52.19587707519531, -34.806854248046875, 218.78489685058594, 192.71429443359375, 55.69245147705078, 331.84429931640625, 111.81942749023438, 159.0501708984375, 281.5169372558594, 399.87640380859375, 107.04476165771484], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000272.npy"}
|
|
{"epoch": 0.39941262848751835, "step": 273, "batch_size": 64, "mean": 151.05178833007812, "std": 148.19105529785156, "min": -158.71055603027344, "p10": -20.698742675781236, "median": 139.5584259033203, "p90": 327.73037109375, "max": 551.5562133789062, "pos_frac": 0.875, "sample": [190.32052612304688, 239.413818359375, 36.741912841796875, 114.382080078125, 448.17608642578125, 425.644287109375, 139.79788208007812, -37.85321807861328, 434.87628173828125, -46.50895690917969, 214.56500244140625, 161.766357421875, 43.610107421875, 46.209068298339844, 6.609376907348633, 139.3189697265625, 1.4533958435058594, 162.57986450195312, 166.49270629882812, 303.7613525390625, 231.2694854736328, 39.269561767578125, 55.6890869140625, 138.2826690673828, 79.06947326660156, 96.80323028564453, 271.2880859375, 107.197265625, 140.01522827148438, 150.4666748046875, 186.8735809326172, 4.073371887207031, 231.86978149414062, 333.1690979003906, 9.88221549987793, 436.2734680175781, 451.33966064453125, 150.61839294433594, 121.38609313964844, 223.8415985107422, -141.937255859375, -31.141769409179688, 213.73338317871094, -57.61937713623047, 213.14239501953125, 274.0750732421875, 287.026611328125, 41.04985046386719, 80.82737731933594, -158.71055603027344, 551.5562133789062, 141.98416137695312, 110.92289733886719, -26.121292114257812, 23.743194580078125, 281.7465515136719, 65.18275451660156, 87.70785522460938, 135.94107055664062, -8.046127319335938, 271.4381408691406, 44.73670196533203, 301.0013732910156, 315.0400085449219], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000273.npy"}
|
|
{"epoch": 0.4008810572687225, "step": 274, "batch_size": 64, "mean": 121.4140625, "std": 139.25538635253906, "min": -156.78224182128906, "p10": -41.76100540161132, "median": 104.82731246948242, "p90": 335.2276519775391, "max": 404.9438171386719, "pos_frac": 0.765625, "sample": [72.60968780517578, 77.32023620605469, 30.532196044921875, -95.13087463378906, 206.71585083007812, 133.72235107421875, 114.87283325195312, -21.90833282470703, 67.41667938232422, -0.5302276611328125, -91.46670532226562, 352.6714172363281, 266.11767578125, 184.39166259765625, 223.0526123046875, 127.16239929199219, 83.21028900146484, 390.62554931640625, -156.78224182128906, 245.293212890625, -27.465805053710938, 158.9475555419922, -37.44915771484375, 95.62049865722656, -15.670402526855469, 233.82440185546875, 356.7424011230469, -33.3829345703125, 82.00965118408203, 390.88922119140625, -29.803279876708984, 91.37960815429688, 210.698974609375, 106.10797882080078, 12.599481582641602, 190.7992706298828, 384.09356689453125, -14.22747802734375, 205.78871154785156, 73.91706848144531, 241.86224365234375, 58.05567932128906, 127.11183166503906, 197.03659057617188, 290.7628173828125, -43.60894012451172, 95.34599304199219, 21.0354061126709, 129.00747680664062, 404.9438171386719, 250.58029174804688, 103.54664611816406, 4.489654541015625, 148.13265991210938, -91.4524154663086, 221.5043182373047, -79.63461303710938, -87.85556030273438, 337.7480163574219, 147.23818969726562, 91.95263671875, 329.3468017578125, 10.927337646484375, 217.10733032226562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000274.npy"}
|
|
{"epoch": 0.4023494860499266, "step": 275, "batch_size": 64, "mean": 122.2330551147461, "std": 149.51480102539062, "min": -185.86553955078125, "p10": -49.756709289550784, "median": 85.41697311401367, "p90": 325.1905456542969, "max": 439.575927734375, "pos_frac": 0.8125, "sample": [185.3188934326172, 79.43853759765625, 226.60423278808594, 20.087482452392578, 19.4503231048584, -51.993019104003906, 312.7799987792969, 75.38267517089844, 375.49273681640625, -85.74970245361328, 224.33139038085938, 324.3616027832031, 435.6319580078125, 75.1350326538086, 146.94528198242188, 170.4359588623047, -36.83501052856445, 27.3487548828125, 80.7668228149414, 142.1328887939453, -49.41484832763672, 194.94937133789062, 381.2182922363281, -47.939598083496094, 56.91913604736328, -142.99374389648438, 213.7694091796875, 171.14166259765625, 190.8551025390625, 49.51494598388672, -62.86426544189453, 43.72528076171875, 214.79441833496094, 193.88955688476562, 45.719970703125, 3.120208740234375, -167.63485717773438, 34.679046630859375, -41.97785186767578, 25.334442138671875, 44.787353515625, 225.0301513671875, -185.86553955078125, 60.45689010620117, 192.3940887451172, 315.38824462890625, 171.677734375, 439.575927734375, 219.58334350585938, 382.39874267578125, 325.5458068847656, -49.903221130371094, 219.31991577148438, 326.14263916015625, 13.360664367675781, 250.45556640625, 90.06712341308594, 292.00396728515625, -40.128639221191406, 170.2175750732422, 37.35681915283203, 253.59841918945312, 7.32470703125, 8.254791259765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000275.npy"}
|
|
{"epoch": 0.40381791483113066, "step": 276, "batch_size": 64, "mean": 136.44924926757812, "std": 161.78880310058594, "min": -170.3204345703125, "p10": -33.54083404541015, "median": 119.72754669189453, "p90": 373.1957641601563, "max": 604.9081420898438, "pos_frac": 0.8125, "sample": [105.5130386352539, 604.9081420898438, 112.64952087402344, 417.7744140625, 375.1993408203125, 56.533973693847656, -1.355926513671875, 458.5818786621094, 22.711469650268555, 176.29763793945312, 129.5640869140625, 38.58637237548828, -30.448394775390625, 241.39453125, 286.7174072265625, 255.17788696289062, 153.42681884765625, -168.50357055664062, 91.15617370605469, 27.527057647705078, 29.688732147216797, 196.6940460205078, 153.45089721679688, 295.07470703125, -18.056106567382812, -170.3204345703125, 24.909160614013672, 92.82701110839844, 155.10470581054688, 281.7732849121094, -34.86616516113281, 138.56874084472656, 90.35991668701172, 140.01248168945312, 259.691162109375, 239.628173828125, -71.03961181640625, -126.10304260253906, 5.718278884887695, 19.42135238647461, -85.9100341796875, -20.02093505859375, 52.4771842956543, 429.3145446777344, 97.73126983642578, 95.79051971435547, 160.7206268310547, 189.34214782714844, 257.2059020996094, 31.73419952392578, 368.520751953125, -11.647735595703125, 163.93170166015625, 212.78909301757812, 46.855072021484375, 213.75523376464844, 126.80557250976562, 131.9444580078125, 305.21295166015625, 74.81939697265625, -94.07330322265625, 482.35467529296875, 33.66023254394531, 413.48895263671875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000276.npy"}
|
|
{"epoch": 0.4052863436123348, "step": 277, "batch_size": 64, "mean": 134.67898559570312, "std": 155.97427368164062, "min": -188.1819305419922, "p10": -49.63037338256835, "median": 98.12105178833008, "p90": 365.9699371337891, "max": 481.0352783203125, "pos_frac": 0.828125, "sample": [18.011089324951172, 267.0709533691406, -97.58616638183594, 71.02960205078125, 47.02322006225586, 204.40643310546875, 267.8139343261719, 143.4583740234375, -41.70875549316406, 171.18161010742188, 137.63723754882812, 13.074914932250977, 274.48748779296875, -39.72655487060547, 247.7537078857422, 288.94696044921875, 72.79573822021484, 84.2237548828125, 174.53492736816406, 216.85755920410156, -53.025352478027344, 279.7414245605469, 61.38520812988281, 70.99353790283203, -108.07007598876953, -34.373924255371094, 460.24371337890625, 80.31582641601562, 5.610687255859375, 103.9967041015625, -188.1819305419922, 445.03546142578125, 87.06426239013672, 270.1173095703125, -64.04390716552734, -128.84027099609375, 481.0352783203125, 238.12667846679688, 138.820556640625, 13.7645263671875, 187.5250244140625, 50.89241027832031, 409.72613525390625, 52.62825393676758, 9.092880249023438, 76.1246566772461, 121.72545623779297, 350.81500244140625, 368.49639892578125, 416.37744140625, -90.76815795898438, 238.23098754882812, 32.983985900878906, 72.6614761352539, 12.181869506835938, -2.580669403076172, 208.6820526123047, 92.24539947509766, 360.0748596191406, 369.6704406738281, 114.31564331054688, 159.51763916015625, 74.73825073242188, 253.09588623046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000277.npy"}
|
|
{"epoch": 0.4067547723935389, "step": 278, "batch_size": 64, "mean": 150.7223358154297, "std": 165.11805725097656, "min": -119.78927612304688, "p10": -30.205466079711915, "median": 124.44777297973633, "p90": 417.80332336425784, "max": 564.7399291992188, "pos_frac": 0.796875, "sample": [-56.02031707763672, 259.4488830566406, -72.61444854736328, 59.32765579223633, 229.05905151367188, 254.2505340576172, 165.41360473632812, 72.10447692871094, 156.4916534423828, 98.85765838623047, 127.05870056152344, 162.3905029296875, 40.82691192626953, -29.872879028320312, 100.20724487304688, 216.6630096435547, -20.52984619140625, 275.468505859375, 422.5508728027344, 19.59099578857422, 15.58387565612793, 514.1488647460938, 334.86553955078125, 174.412841796875, 238.79498291015625, 261.97650146484375, 100.27637481689453, -20.782798767089844, 130.14215087890625, 453.9554443359375, 397.3514099121094, -31.410507202148438, 112.154296875, -45.370758056640625, 462.48883056640625, 435.52972412109375, 15.440818786621094, 24.006134033203125, 48.31169891357422, 177.55728149414062, 92.26856231689453, 387.002685546875, 127.54844665527344, 39.52899169921875, 37.830596923828125, 437.9466857910156, 121.83684539794922, 187.40394592285156, -13.24301528930664, 406.7257080078125, 145.89083862304688, 128.2086944580078, 9.61846923828125, 50.232704162597656, 239.34683227539062, 273.4671630859375, -20.652620315551758, -6.520009994506836, 564.7399291992188, -102.23341369628906, 288.4473876953125, 120.86627197265625, -119.78927612304688, -30.348003387451172], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000278.npy"}
|
|
{"epoch": 0.40822320117474303, "step": 279, "batch_size": 64, "mean": 159.4352569580078, "std": 194.2080535888672, "min": -295.0009765625, "p10": -53.79413681030273, "median": 130.95172119140625, "p90": 411.88001403808596, "max": 732.602783203125, "pos_frac": 0.828125, "sample": [69.51795196533203, 67.9254150390625, 47.54814529418945, 40.5751953125, 22.494140625, -54.03783416748047, 399.19232177734375, 363.7461242675781, 10.797294616699219, 0.3566856384277344, 36.232994079589844, 246.79090881347656, 126.485595703125, 39.625274658203125, 512.94140625, 515.8619384765625, 41.50124740600586, 131.65048217773438, 232.67198181152344, 202.84747314453125, 558.7166137695312, 117.64379119873047, -6.239435195922852, -203.7001953125, 35.28147888183594, 75.78536224365234, 45.6063232421875, 192.97796630859375, 66.2401123046875, 491.6807861328125, 109.74502563476562, 89.57170867919922, 217.05258178710938, 2.0061283111572266, 195.85653686523438, 417.3175964355469, 226.06805419921875, 732.602783203125, 439.0465087890625, 269.623779296875, -53.22550964355469, -44.014434814453125, -136.58847045898438, -295.0009765625, 147.0385284423828, 383.8085632324219, 115.65521240234375, 216.27366638183594, 309.4512939453125, 140.5018310546875, 330.11761474609375, 293.28057861328125, 130.25296020507812, -80.59136962890625, 356.4983825683594, 157.00204467773438, 295.3622131347656, 180.50306701660156, -77.26702880859375, -124.93177795410156, 351.516845703125, 289.7334899902344, -12.998722076416016, 203.86996459960938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000279.npy"}
|
|
{"epoch": 0.40969162995594716, "step": 280, "batch_size": 64, "mean": 155.0647430419922, "std": 146.4903564453125, "min": -164.67031860351562, "p10": -31.018798065185546, "median": 147.13382720947266, "p90": 346.1157684326172, "max": 436.307373046875, "pos_frac": 0.828125, "sample": [144.88438415527344, 250.7538299560547, 112.25202178955078, 180.78343200683594, 13.351272583007812, -31.86743927001953, 274.994384765625, 388.4130859375, 367.52288818359375, 58.96246337890625, 235.0559844970703, -9.829582214355469, -34.18819046020508, 222.95623779296875, 347.5339660644531, 342.806640625, 100.29871368408203, 240.38978576660156, 150.02828979492188, 229.67855834960938, 65.61112976074219, 436.307373046875, 420.5440673828125, 182.69540405273438, 208.83099365234375, 107.2996826171875, 311.0091552734375, 289.88433837890625, 5.92291259765625, 264.30706787109375, 137.99307250976562, 139.56106567382812, 402.62103271484375, 112.22616577148438, -29.03863525390625, 155.80181884765625, -130.822509765625, 325.28961181640625, 159.5722198486328, 86.02417755126953, 196.9273681640625, 98.2486801147461, -119.17402648925781, 416.14031982421875, -23.45541763305664, -164.67031860351562, 117.07747650146484, 269.6973571777344, 149.38327026367188, 123.40544891357422, 157.9791717529297, 322.85308837890625, 10.26923942565918, 126.52227020263672, -15.097808837890625, -58.74706268310547, 106.91812896728516, 91.33741760253906, 103.15542602539062, 301.9246826171875, 239.6441650390625, -122.8652114868164, 59.63984680175781, 300.6090087890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000280.npy"}
|
|
{"epoch": 0.4111600587371512, "step": 281, "batch_size": 64, "mean": 143.2918701171875, "std": 150.60755920410156, "min": -475.5090026855469, "p10": -13.249736976623527, "median": 149.1793670654297, "p90": 342.68790893554694, "max": 500.066162109375, "pos_frac": 0.875, "sample": [78.07233428955078, 256.7258605957031, 114.74407958984375, 227.3983917236328, 87.41252136230469, 89.26615905761719, 86.26194763183594, 170.4395751953125, 154.48353576660156, 47.32659149169922, 159.5579833984375, 314.9371337890625, 210.1339874267578, 42.422447204589844, 205.88973999023438, -72.06643676757812, 148.72695922851562, 412.7979736328125, 76.5802993774414, 149.63177490234375, 10.709747314453125, 234.1033477783203, 221.4375762939453, 422.6300354003906, 191.73297119140625, 200.04605102539062, 157.56549072265625, 241.50929260253906, -4.598133087158203, 500.066162109375, 316.78338623046875, 97.11328125, 156.8720245361328, 352.70855712890625, 183.709716796875, -475.5090026855469, 73.65504455566406, 106.34886169433594, 136.78433227539062, 53.406585693359375, 434.7891845703125, 55.59568786621094, 72.45320129394531, 221.76112365722656, 163.7274169921875, 242.10836791992188, 67.37140655517578, 171.6677703857422, 350.13226318359375, 92.25916290283203, -81.5038070678711, 112.01909637451172, -61.27839660644531, 61.305633544921875, -68.04510498046875, 6.499542236328125, -35.02033996582031, 325.3177490234375, 175.9571990966797, 36.327613830566406, 115.0490951538086, 240.80836486816406, -16.95756721496582, 350.5162658691406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000281.npy"}
|
|
{"epoch": 0.41262848751835535, "step": 282, "batch_size": 64, "mean": 109.85005187988281, "std": 158.4193572998047, "min": -126.29586791992188, "p10": -74.98539123535156, "median": 95.4391098022461, "p90": 283.57225341796885, "max": 852.4067993164062, "pos_frac": 0.765625, "sample": [327.7107238769531, 97.88377380371094, 53.6705436706543, -90.6553726196289, 188.34791564941406, 189.18930053710938, 87.98555755615234, 55.283966064453125, 116.784912109375, 260.852783203125, 258.3487548828125, 155.08477783203125, 262.5116271972656, 83.05036163330078, -16.113815307617188, 160.10552978515625, 852.4067993164062, 126.99764251708984, 121.2615966796875, 85.54684448242188, 80.65289306640625, 200.0897216796875, 21.54531478881836, -80.6278076171875, 414.5416259765625, 309.4072570800781, 125.09129333496094, 117.12751770019531, 138.32846069335938, 387.21258544921875, 222.28660583496094, 124.42655944824219, 250.30657958984375, -38.334991455078125, 74.97554016113281, 106.67788696289062, -111.87620544433594, -55.159629821777344, 49.830474853515625, -87.53329467773438, -59.968017578125, 49.00538635253906, 102.8788070678711, 160.29193115234375, 240.69227600097656, 8.752983093261719, -118.69168090820312, 292.5982360839844, 176.28121948242188, 92.99444580078125, -49.626792907714844, -27.416885375976562, 39.34333801269531, 49.45566940307617, 108.4542236328125, -72.30123901367188, 222.9764404296875, 39.16627502441406, -45.683372497558594, 307.90106201171875, -126.29586791992188, 33.515010833740234, -76.1357421875, 56.99275207519531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000282.npy"}
|
|
{"epoch": 0.41409691629955947, "step": 283, "batch_size": 64, "mean": 161.82949829101562, "std": 150.88572692871094, "min": -251.9842529296875, "p10": -24.67976131439209, "median": 142.71009063720703, "p90": 354.309326171875, "max": 517.4010009765625, "pos_frac": 0.84375, "sample": [17.732967376708984, -22.01323699951172, 144.1009979248047, 38.157928466796875, 274.14453125, 292.826171875, 26.57867431640625, 208.1521759033203, 314.1198425292969, 57.47943878173828, -9.701576232910156, 202.83485412597656, -28.19350814819336, 74.01411437988281, 81.14478302001953, 106.35256958007812, -251.9842529296875, 82.19308471679688, 272.9017639160156, -19.66350746154785, 415.6413879394531, 358.6027526855469, 112.16029357910156, 260.1318359375, 96.80888366699219, 91.30218505859375, -44.72624206542969, 357.0994873046875, 246.8084716796875, 445.8070068359375, 24.936614990234375, 244.29409790039062, -61.69520568847656, 152.66986083984375, 287.0616760253906, 72.93937683105469, 58.632843017578125, -50.6082649230957, 85.62259674072266, -25.82255744934082, 318.32147216796875, 196.3572540283203, 274.0583190917969, 109.7752685546875, 264.9139099121094, 217.92263793945312, 73.92244720458984, 141.31918334960938, 327.18865966796875, 189.78460693359375, 463.8446960449219, 103.61678314208984, 313.99017333984375, 105.08648681640625, 46.43103790283203, 273.62115478515625, 382.9273986816406, 172.3621368408203, 291.17108154296875, 517.4010009765625, 347.7989501953125, 100.95714569091797, 168.08172607421875, -32.60969543457031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000283.npy"}
|
|
{"epoch": 0.4155653450807636, "step": 284, "batch_size": 64, "mean": 154.01272583007812, "std": 151.04161071777344, "min": -166.01930236816406, "p10": 1.9163242340087976, "median": 131.50997924804688, "p90": 355.0551635742188, "max": 622.386962890625, "pos_frac": 0.890625, "sample": [-52.00396728515625, 77.45755004882812, 257.39691162109375, 37.024513244628906, 111.2321548461914, 46.351654052734375, 289.3848876953125, 24.052947998046875, 243.79122924804688, 269.1953125, 339.447265625, -13.055057525634766, 373.3997497558594, 34.135398864746094, 216.14645385742188, 76.25614929199219, 26.02611541748047, 229.23696899414062, 216.64669799804688, 409.6349792480469, -1.741851806640625, 515.0283203125, 161.70628356933594, 232.89215087890625, 150.02117919921875, 31.401107788085938, 125.62031555175781, 44.59011459350586, 231.0869903564453, -44.31996154785156, 51.691162109375, 539.4649047851562, 160.65084838867188, -44.53498077392578, 112.7978744506836, 104.54290771484375, 15.907249450683594, 72.75936126708984, 191.38047790527344, 622.386962890625, 127.80287170410156, 168.4085693359375, 135.242431640625, 45.48780822753906, -14.39208984375, -166.01930236816406, 160.46389770507812, 303.71124267578125, 84.94837951660156, 234.05499267578125, 448.67645263671875, 135.2170867919922, 181.1300048828125, 154.3412322998047, 72.39576721191406, 10.452068328857422, 13.409004211425781, 36.26000213623047, 272.16949462890625, 235.7925567626953, 68.2439956665039, 361.7442626953125, 77.23848724365234, 224.94595336914062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000284.npy"}
|
|
{"epoch": 0.4170337738619677, "step": 285, "batch_size": 64, "mean": 134.28878784179688, "std": 140.67820739746094, "min": -321.5919189453125, "p10": -29.166984939575176, "median": 146.46817016601562, "p90": 304.5568786621094, "max": 475.3704833984375, "pos_frac": 0.84375, "sample": [279.9034423828125, 222.22799682617188, 40.87963104248047, 304.7174072265625, 59.400001525878906, 97.86363220214844, 146.40219116210938, 43.56006622314453, 159.6724853515625, 332.8688659667969, 217.23777770996094, 164.97686767578125, 266.5646667480469, 123.3917007446289, 122.56265258789062, 465.80224609375, 8.5887451171875, 25.274383544921875, 177.14993286132812, 172.66867065429688, 322.7481384277344, -147.28109741210938, 63.98833465576172, 201.57492065429688, 327.2809143066406, 41.98938751220703, -11.416263580322266, 99.35501098632812, 55.179290771484375, 23.596107482910156, 201.34906005859375, 85.69563293457031, -68.0441665649414, 39.87136459350586, 207.62600708007812, 304.18231201171875, 75.5994644165039, -321.5919189453125, 207.39288330078125, 68.12509155273438, 163.46632385253906, -4.32464599609375, 475.3704833984375, 196.79879760742188, -46.72063446044922, 268.60931396484375, 174.18722534179688, 146.53414916992188, 246.14462280273438, 278.1625671386719, 91.1905517578125, -73.45024108886719, 221.80111694335938, 152.55307006835938, -106.41084289550781, -10.61740493774414, 320.02825927734375, 64.77424621582031, 99.40547943115234, 148.3590850830078, -36.774436950683594, 282.85760498046875, 204.65533447265625, 128.94955444335938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000285.npy"}
|
|
{"epoch": 0.4185022026431718, "step": 286, "batch_size": 64, "mean": 152.447265625, "std": 163.04327392578125, "min": -222.06903076171875, "p10": -54.4760124206543, "median": 155.9247589111328, "p90": 397.01622924804695, "max": 471.1394958496094, "pos_frac": 0.78125, "sample": [55.66883087158203, -13.963432312011719, 453.7057800292969, 139.3841552734375, 147.87948608398438, 151.89898681640625, 66.98192596435547, 144.67588806152344, 260.4888000488281, -213.42636108398438, 218.67160034179688, 312.1426696777344, 28.183364868164062, -33.555335998535156, 256.5856628417969, 192.3006591796875, 428.8406066894531, -39.76055908203125, 139.88214111328125, 158.2868194580078, 402.825927734375, 99.17010498046875, -222.06903076171875, 378.275146484375, 223.8147735595703, -33.6717529296875, -37.02106857299805, 312.8153076171875, 275.2071533203125, 409.632080078125, 403.991455078125, -68.4312744140625, 223.62350463867188, 96.91818237304688, 47.62376403808594, 72.18270874023438, 218.8638458251953, 153.5626983642578, 45.316009521484375, -54.22160339355469, 228.11492919921875, -8.498817443847656, 192.54293823242188, -64.10794830322266, 47.088623046875, 293.51092529296875, 383.46026611328125, 170.6813507080078, -106.25401306152344, 426.1437683105469, -54.585044860839844, 194.8662109375, 62.446807861328125, 304.17779541015625, 167.67791748046875, 298.8394775390625, 128.7110595703125, 187.33120727539062, 170.74505615234375, -60.6558837890625, 471.1394958496094, 286.1061706542969, 168.5606689453125, 65.3024673461914], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000286.npy"}
|
|
{"epoch": 0.4199706314243759, "step": 287, "batch_size": 64, "mean": 129.17588806152344, "std": 156.53341674804688, "min": -219.6273193359375, "p10": -43.58566665649411, "median": 113.41763687133789, "p90": 361.361328125, "max": 628.490234375, "pos_frac": 0.84375, "sample": [92.26829528808594, 87.63607788085938, 12.404302597045898, 90.88603973388672, 111.28118133544922, -10.758087158203125, 235.4835205078125, 498.77484130859375, 161.66822814941406, 135.64154052734375, 436.35479736328125, 99.89923095703125, 151.69529724121094, 424.79632568359375, -55.68231201171875, -76.70436096191406, 82.77767181396484, 373.1551513671875, 173.58993530273438, 259.06146240234375, 136.28515625, 5.5142974853515625, 153.2655029296875, -219.6273193359375, 628.490234375, 83.5102767944336, 51.964111328125, 5.34747314453125, 45.62193298339844, -13.252906799316406, 115.55409240722656, 14.719657897949219, -54.980377197265625, 141.58949279785156, 32.46577835083008, 231.0967254638672, 44.518218994140625, 214.33399963378906, 83.43208312988281, 123.36744689941406, 40.904815673828125, 275.4918212890625, 132.72384643554688, 126.87649536132812, 18.720779418945312, 101.94023132324219, 74.47854614257812, 35.33528137207031, 305.2637939453125, 165.462158203125, -129.39430236816406, 57.83379364013672, 180.7482452392578, 141.8585968017578, 363.90777587890625, -132.14524841308594, -141.27102661132812, -16.998008728027344, 188.36746215820312, 245.59213256835938, 381.48779296875, 355.41961669921875, 177.11233520507812, 210.09503173828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000287.npy"}
|
|
{"epoch": 0.42143906020558003, "step": 288, "batch_size": 64, "mean": 156.4290771484375, "std": 146.03602600097656, "min": -249.23382568359375, "p10": -4.1346231460571135, "median": 146.53201293945312, "p90": 362.77530822753926, "max": 550.724365234375, "pos_frac": 0.890625, "sample": [159.41162109375, 131.11489868164062, -18.620540618896484, 226.07089233398438, 229.26478576660156, 274.96600341796875, 31.85401725769043, 550.724365234375, 77.45951843261719, 318.7418518066406, 182.3865966796875, 109.76858520507812, -48.065528869628906, 18.14935302734375, 47.33600997924805, 128.46942138671875, 123.38198852539062, 310.8143005371094, 137.02593994140625, 265.2406921386719, 166.3021697998047, 224.49261474609375, 227.1876220703125, 28.1029052734375, 111.79612731933594, 197.19818115234375, 176.68511962890625, 10.697717666625977, 155.08834838867188, 56.83100891113281, 110.29663848876953, 156.33279418945312, 461.9342956542969, 161.44393920898438, 388.02008056640625, 258.52374267578125, 95.22795104980469, 144.17007446289062, 166.1932373046875, 90.6021957397461, -249.23382568359375, 444.8639831542969, 73.72998046875, 188.1559295654297, 63.68516540527344, 148.89395141601562, 102.10680389404297, 54.82554626464844, 442.161376953125, 96.13457489013672, 256.5380859375, 381.64678955078125, 163.97549438476562, -154.52122497558594, 127.52755737304688, 407.84490966796875, 203.589599609375, 65.50160217285156, 74.23263549804688, 297.5541076660156, -10.491340637207031, 304.56634521484375, -69.43550109863281, -45.01329040527344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000288.npy"}
|
|
{"epoch": 0.42290748898678415, "step": 289, "batch_size": 64, "mean": 175.6532745361328, "std": 178.9029541015625, "min": -139.64669799804688, "p10": -24.522603225708, "median": 128.21361541748047, "p90": 463.82048645019535, "max": 563.8390502929688, "pos_frac": 0.828125, "sample": [172.36264038085938, -10.885833740234375, 500.1783447265625, 475.1864318847656, -139.64669799804688, 488.5121154785156, 258.4041748046875, 212.38211059570312, 73.28961181640625, 458.4064025878906, 396.5390319824219, 302.17535400390625, 239.81187438964844, 466.14080810546875, 304.36505126953125, 264.433349609375, 41.36717224121094, 563.8390502929688, -5.396598815917969, 249.9422607421875, 106.69577026367188, 7.731748580932617, 65.32479858398438, -15.874691009521484, -43.982269287109375, 104.99554443359375, 282.4303894042969, 293.0121154785156, 49.244293212890625, 133.99256896972656, 341.8807067871094, 553.360107421875, 276.52569580078125, -76.55293273925781, 81.43549346923828, 68.83613586425781, 500.50860595703125, 87.96766662597656, -64.9398193359375, 25.17963218688965, 85.74429321289062, 360.00396728515625, -28.228851318359375, 223.9954833984375, 134.20631408691406, 306.5203857421875, 396.87939453125, -0.7237529754638672, 121.34178161621094, -94.35481262207031, 67.22938537597656, 166.884033203125, 97.60272216796875, 40.37309265136719, 412.2661437988281, 147.3583984375, 59.57483673095703, 72.43407440185547, 54.329132080078125, 191.49050903320312, 122.43466186523438, -130.09396362304688, 252.26931762695312, 93.0952377319336], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000289.npy"}
|
|
{"epoch": 0.4243759177679883, "step": 290, "batch_size": 64, "mean": 175.46475219726562, "std": 198.45616149902344, "min": -223.6421356201172, "p10": -63.30973052978514, "median": 154.2686004638672, "p90": 445.5837615966798, "max": 708.317138671875, "pos_frac": 0.84375, "sample": [417.87567138671875, 157.45010375976562, 129.8782958984375, 130.71722412109375, -193.57473754882812, 247.42056274414062, 365.04693603515625, -72.09927368164062, -34.81446838378906, 114.89559936523438, -95.66980743408203, 309.3421936035156, 119.3172607421875, -24.940338134765625, 208.620361328125, 212.27374267578125, 425.354736328125, 26.995147705078125, 199.73089599609375, 57.434879302978516, 24.001174926757812, 65.19779968261719, -68.19967651367188, 151.08709716796875, 121.53289794921875, -223.6421356201172, 120.398681640625, 191.86383056640625, 649.7843017578125, 599.2193603515625, 57.57057189941406, 203.54859924316406, 253.42283630371094, 527.8863525390625, 224.55718994140625, 41.84660339355469, 279.87530517578125, 101.38157653808594, 218.521240234375, 134.348388671875, 17.11627197265625, 208.23207092285156, 185.8117218017578, 265.92156982421875, 33.96900939941406, 315.28106689453125, 5.525367736816406, -51.89985656738281, 266.19305419921875, -101.5810546875, 537.102783203125, 380.88116455078125, 208.43038940429688, 453.8385314941406, 38.868587493896484, 426.3226318359375, 455.00579833984375, 262.66204833984375, 19.311687469482422, 36.009735107421875, 29.864688873291016, 708.317138671875, 246.11257934570312, -93.010009765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000290.npy"}
|
|
{"epoch": 0.42584434654919234, "step": 291, "batch_size": 64, "mean": 163.03326416015625, "std": 167.61721801757812, "min": -317.3730163574219, "p10": -25.611536407470695, "median": 185.15804290771484, "p90": 380.8241577148438, "max": 544.8604736328125, "pos_frac": 0.796875, "sample": [139.17918395996094, 210.64886474609375, 21.701705932617188, 69.34125518798828, -9.494781494140625, 266.116455078125, 439.8672790527344, 198.2154541015625, 289.88671875, 222.31597900390625, 46.10139465332031, 321.70806884765625, 251.32472229003906, 43.20270538330078, 544.8604736328125, -28.59539794921875, 387.0753479003906, 195.10130310058594, -8.887948989868164, 193.0348663330078, 367.27618408203125, 29.883834838867188, 229.44654846191406, 67.63186645507812, 208.92149353027344, 308.7112731933594, 177.28121948242188, 269.4776611328125, 275.1854553222656, 2.6208152770996094, 239.1563720703125, 128.17178344726562, 302.3460693359375, -61.27732849121094, 76.85140991210938, -77.30874633789062, -18.649192810058594, 77.17876434326172, -317.3730163574219, 416.8015441894531, 305.329833984375, 350.1961669921875, 315.7667541503906, 163.76806640625, -105.50740051269531, -18.487548828125, -53.52385711669922, 278.29949951171875, 30.65089225769043, 112.49776458740234, 54.17015075683594, 105.09156799316406, 338.3684387207031, 222.126708984375, -2.6220703125, 197.17086791992188, 386.63043212890625, 85.81510925292969, 445.68353271484375, -6.446861267089844, -45.43077850341797, 344.92230224609375, 430.43450927734375, 4.187273025512695], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000291.npy"}
|
|
{"epoch": 0.42731277533039647, "step": 292, "batch_size": 64, "mean": 163.12124633789062, "std": 200.2244873046875, "min": -502.37042236328125, "p10": -40.06289405822753, "median": 172.20995330810547, "p90": 432.0457641601563, "max": 599.430419921875, "pos_frac": 0.828125, "sample": [190.67596435546875, 51.84054946899414, 5.6742706298828125, 155.31024169921875, 250.31100463867188, 180.6881103515625, 113.958251953125, 115.75737762451172, 180.79415893554688, -502.37042236328125, 554.540771484375, 27.69463348388672, 459.5115966796875, 81.28781127929688, 203.126220703125, -44.254852294921875, -217.41619873046875, 444.84112548828125, 55.14917755126953, 314.6447448730469, 5.791175842285156, 577.6860961914062, 269.2553405761719, 204.25411987304688, 343.5333557128906, 437.0198059082031, 255.53683471679688, 337.72650146484375, 141.58651733398438, 74.17428588867188, -82.97540283203125, 165.73907470703125, 7.03611946105957, 350.78729248046875, -15.302574157714844, 80.58026885986328, -63.81007385253906, 170.53843688964844, 162.1331024169922, 92.44334411621094, 50.60682678222656, 420.4396667480469, 221.16567993164062, 218.3444366455078, 184.63697814941406, 215.03811645507812, 266.9219055175781, 311.9744873046875, 599.430419921875, 23.197233200073242, 126.91889190673828, 194.95440673828125, 250.5842742919922, -201.02061462402344, 216.05442810058594, -188.010498046875, 557.1246337890625, -30.281658172607422, -11.430339813232422, 396.9477233886719, 229.3284454345703, 173.8814697265625, 107.21955871582031, -29.76426887512207], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000292.npy"}
|
|
{"epoch": 0.4287812041116006, "step": 293, "batch_size": 64, "mean": 131.80682373046875, "std": 186.2793731689453, "min": -193.10292053222656, "p10": -93.95841293334959, "median": 116.31324768066406, "p90": 370.7536163330078, "max": 603.4658203125, "pos_frac": 0.71875, "sample": [34.881919860839844, 7.799346923828125, 35.96088409423828, 48.03307342529297, 214.31887817382812, 196.91673278808594, 413.2166442871094, 140.4118194580078, 387.488037109375, 27.35845375061035, 367.9106140136719, -33.77348327636719, 205.15216064453125, -38.32784652709961, 603.4658203125, -137.91513061523438, 166.78921508789062, 202.796630859375, 49.618133544921875, 186.32672119140625, 300.5898742675781, -143.16226196289062, 200.38626098632812, -71.43932342529297, 32.307350158691406, 41.689353942871094, -56.44213104248047, 127.48222351074219, 7.4804534912109375, 451.0313720703125, 371.9720458984375, 291.69720458984375, 262.79486083984375, 283.77557373046875, 243.46055603027344, -103.60945129394531, 64.34259033203125, 154.60638427734375, 149.40835571289062, -52.38848114013672, -39.08623504638672, 75.93024444580078, -0.88641357421875, 210.05502319335938, 321.591064453125, 327.6053466796875, 189.2769775390625, 285.98944091796875, 105.14427185058594, -170.89105224609375, -3.7118988037109375, -193.10292053222656, -61.78239440917969, 363.98834228515625, 590.6944580078125, 66.84078216552734, -110.2152099609375, 182.7294464111328, 76.07106018066406, -29.921157836914062, -125.7537612915039, 351.87005615234375, 458.90325927734375, -70.11268615722656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000293.npy"}
|
|
{"epoch": 0.4302496328928047, "step": 294, "batch_size": 64, "mean": 147.9698486328125, "std": 186.2820281982422, "min": -179.26986694335938, "p10": -65.70995292663574, "median": 92.11553192138672, "p90": 423.2611145019532, "max": 649.6482543945312, "pos_frac": 0.859375, "sample": [52.791786193847656, 57.5660400390625, 273.1177978515625, 389.92901611328125, 69.98898315429688, 68.59742736816406, -101.14447784423828, 111.16602325439453, 6.426719665527344, 115.69416046142578, 11.702957153320312, 15.023750305175781, -53.5951042175293, 476.0009765625, 29.29204559326172, 265.51666259765625, 149.75975036621094, 378.0388488769531, 61.47978591918945, 140.6489715576172, 573.269775390625, -114.22453308105469, 230.74288940429688, 96.97108459472656, 501.46319580078125, 306.09393310546875, 236.3742218017578, 101.22003936767578, 151.11215209960938, -70.90203094482422, 78.6880111694336, 270.23291015625, 274.21771240234375, 649.6482543945312, -179.26986694335938, -125.58065795898438, 535.7598266601562, 118.403076171875, 284.70855712890625, 49.69770050048828, -152.0067138671875, -100.48235321044922, 234.65036010742188, 267.5611877441406, 492.68450927734375, 92.62408447265625, 44.47832107543945, 116.23347473144531, 435.375244140625, 4.54405403137207, 300.0235290527344, 319.4789733886719, -24.425025939941406, 91.60697937011719, 71.54374694824219, 33.94122314453125, 34.21773910522461, 69.3070297241211, 12.693119049072266, 81.60794067382812, 394.99481201171875, 61.183441162109375, 49.101226806640625, 52.505470275878906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000294.npy"}
|
|
{"epoch": 0.43171806167400884, "step": 295, "batch_size": 64, "mean": 142.3737335205078, "std": 175.1631317138672, "min": -215.8462677001953, "p10": -62.948388671874994, "median": 140.2094497680664, "p90": 388.5173767089845, "max": 589.224853515625, "pos_frac": 0.78125, "sample": [-167.10696411132812, 24.68101692199707, 399.39404296875, 5.815925598144531, -215.8462677001953, 198.74343872070312, -52.33579635620117, 271.4131774902344, 151.645263671875, -179.17904663085938, 253.14144897460938, 11.43402099609375, 124.72949981689453, 310.109130859375, 332.5873718261719, 244.50448608398438, 236.84974670410156, 194.1123046875, -37.30908966064453, 78.5164566040039, 293.86431884765625, -11.211141586303711, 32.512115478515625, 273.3424987792969, -17.563804626464844, 72.1890869140625, 408.8367614746094, -101.90602111816406, 461.3807373046875, 142.19686889648438, 254.89370727539062, 7.9697265625, -33.388187408447266, -112.1016845703125, 363.13848876953125, 486.70751953125, 26.45305633544922, -59.55591583251953, -71.38008117675781, 119.20814514160156, 138.22203063964844, 259.1190185546875, 314.5869140625, 52.57362365722656, -22.870201110839844, 55.71992492675781, 187.8274688720703, 60.36066436767578, 209.93917846679688, 423.9520568847656, 160.14028930664062, 157.00335693359375, 19.482879638671875, 589.224853515625, 328.59185791015625, 344.13323974609375, 127.14012145996094, 151.64614868164062, 147.20529174804688, 165.89230346679688, 82.13069915771484, 87.18843078613281, 415.6249694824219, -64.40230560302734], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000295.npy"}
|
|
{"epoch": 0.4331864904552129, "step": 296, "batch_size": 64, "mean": 110.34952545166016, "std": 189.4136199951172, "min": -381.4075622558594, "p10": -98.51003036499023, "median": 108.94740295410156, "p90": 344.8096343994142, "max": 551.3016967773438, "pos_frac": 0.71875, "sample": [122.1807632446289, 93.44498443603516, 42.9976806640625, 551.3016967773438, -86.68634033203125, 303.2404479980469, -99.08626556396484, -48.93467712402344, -36.24119567871094, -339.83074951171875, 120.94305419921875, 106.15715789794922, -226.2281036376953, 239.22418212890625, 120.18827819824219, -175.6446533203125, 114.15902709960938, 230.2335968017578, 445.2210388183594, 288.117431640625, 189.41824340820312, 174.48553466796875, -42.102027893066406, -25.210540771484375, 53.66583251953125, 298.44342041015625, 26.9620361328125, -97.16548156738281, -16.757965087890625, -101.18946838378906, 435.1859130859375, 51.80311584472656, 224.9013214111328, 63.18785858154297, 61.188995361328125, 482.1174621582031, 194.2497100830078, 53.493648529052734, -381.4075622558594, 220.94229125976562, -196.0220947265625, 265.6465148925781, -31.180580139160156, -27.330232620239258, 407.0042724609375, 91.26294708251953, 122.1020736694336, 174.37127685546875, 111.48348236083984, 45.956443786621094, 96.11994171142578, 58.25513458251953, -11.229713439941406, 550.5123901367188, 212.46690368652344, 225.78445434570312, 106.41132354736328, 158.98968505859375, 362.625, -71.49044799804688, 201.36595153808594, 135.47579956054688, 171.55381774902344, 271.2652587890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000296.npy"}
|
|
{"epoch": 0.434654919236417, "step": 297, "batch_size": 64, "mean": 128.75506591796875, "std": 145.5901336669922, "min": -224.7506103515625, "p10": -8.962481689453117, "median": 98.93846130371094, "p90": 355.72581787109385, "max": 517.7348022460938, "pos_frac": 0.875, "sample": [76.49432373046875, 135.67222595214844, 285.5772705078125, 406.0452575683594, 81.36433410644531, 2.0596866607666016, 329.53631591796875, 170.85775756835938, -95.70270538330078, -58.45119094848633, 70.72053527832031, 107.00946044921875, 220.17205810546875, 231.0805206298828, 378.0858154296875, 9.655675888061523, 120.67169189453125, 46.348663330078125, 17.214874267578125, -12.633888244628906, 62.25578308105469, -117.48634338378906, 1.8215770721435547, 62.18830871582031, 366.94989013671875, 46.370086669921875, 282.77691650390625, 54.46556091308594, 192.09124755859375, 232.63040161132812, 166.93658447265625, 90.86746215820312, 280.81243896484375, 157.3320770263672, 257.34246826171875, 37.87730026245117, 134.25111389160156, 77.5877456665039, 395.460693359375, 197.98330688476562, 21.001312255859375, 171.40475463867188, -170.05145263671875, 196.67237854003906, 46.17144775390625, -30.10344696044922, 159.58935546875, 67.5230712890625, 517.7348022460938, 1.8485107421875, 132.15573120117188, 368.1244201660156, 198.65731811523438, 1.905029296875, -224.7506103515625, 184.05364990234375, -0.39586639404296875, 52.52522277832031, 185.06069946289062, 373.7395935058594, 85.19204711914062, 291.2183837890625, 22.8367862701416, 85.91708374023438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000297.npy"}
|
|
{"epoch": 0.43612334801762115, "step": 298, "batch_size": 64, "mean": 165.982666015625, "std": 204.1497344970703, "min": -253.9370574951172, "p10": -58.42209396362304, "median": 131.08927154541016, "p90": 466.9826934814454, "max": 882.7343139648438, "pos_frac": 0.828125, "sample": [90.4698257446289, -61.950042724609375, 32.69944381713867, -253.9370574951172, 84.65328979492188, -100.62158203125, 6.72191047668457, 528.31787109375, 144.406005859375, 75.37451171875, 136.77523803710938, 390.6395263671875, 5.9231414794921875, -3.3929290771484375, -60.02122497558594, 6.351493835449219, 165.9559326171875, 136.70416259765625, 115.22037506103516, 141.0894012451172, 575.3414916992188, -9.695343017578125, 251.06092834472656, -130.36865234375, 471.746337890625, 266.12152099609375, -47.96397399902344, 372.76593017578125, 455.8675231933594, 512.7897338867188, -71.51079559326172, -104.48443603515625, -54.69078826904297, 9.195243835449219, 9.153968811035156, 501.6968994140625, 498.523681640625, 143.46385192871094, 319.64923095703125, 236.31710815429688, 38.32679748535156, 4.896881103515625, 882.7343139648438, 244.98684692382812, 411.25531005859375, 175.50692749023438, 275.85711669921875, 320.66998291015625, 73.43402099609375, 92.34701538085938, 65.63949584960938, 295.9250183105469, 264.9762268066406, 110.27005767822266, 329.64111328125, 156.37120056152344, 49.42523193359375, 263.0901184082031, 22.205581665039062, 179.92739868164062, 107.32443237304688, 125.47438049316406, 53.918731689453125, 292.3270263671875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000298.npy"}
|
|
{"epoch": 0.43759177679882527, "step": 299, "batch_size": 64, "mean": 155.0985565185547, "std": 187.90306091308594, "min": -243.77667236328125, "p10": -48.57968978881836, "median": 123.49248886108398, "p90": 469.01091003417974, "max": 652.4439086914062, "pos_frac": 0.796875, "sample": [-50.93720245361328, -243.77667236328125, -107.62754821777344, 232.13693237304688, -60.8858642578125, -32.30935287475586, 311.93505859375, 91.94586181640625, 84.79528045654297, 90.1222915649414, 496.75689697265625, 496.603271484375, 572.9730224609375, 15.999490737915039, 163.6156463623047, 619.1432495117188, -135.70623779296875, 293.88983154296875, 128.10659790039062, 128.5625762939453, -34.62901306152344, 257.30621337890625, 68.56172943115234, 177.06634521484375, 258.0298156738281, 84.57232666015625, 652.4439086914062, 62.21821975708008, 109.24828338623047, -36.614566802978516, -3.510965347290039, 107.02415466308594, 34.269744873046875, 129.58204650878906, 198.8872833251953, 452.0728454589844, 85.83280944824219, 196.82896423339844, 20.369384765625, 143.0439453125, 118.87837982177734, 262.05352783203125, 211.04383850097656, -26.455829620361328, 198.36636352539062, 62.36821746826172, -164.62083435058594, 356.2090148925781, 107.14886474609375, 485.0065612792969, -52.830421447753906, 33.95166778564453, 182.95391845703125, 246.24801635742188, 185.16067504882812, 272.78631591796875, 114.37467956542969, 476.27008056640625, 155.74636840820312, 99.53254699707031, -43.078826904296875, 315.6813659667969, 24.88587188720703, 246.68026733398438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000299.npy"}
|
|
{"epoch": 0.4390602055800294, "step": 300, "batch_size": 64, "mean": 174.70237731933594, "std": 215.3582763671875, "min": -288.01324462890625, "p10": -130.5445526123047, "median": 200.09738159179688, "p90": 409.8004486083984, "max": 622.7669677734375, "pos_frac": 0.78125, "sample": [223.79376220703125, 243.25872802734375, 148.02883911132812, 256.10504150390625, 399.80181884765625, -288.01324462890625, -66.19774627685547, 403.87396240234375, 325.64996337890625, -15.86734390258789, -96.92407989501953, 155.18539428710938, 232.23622131347656, 411.3999938964844, 200.1646728515625, 330.5771484375, 105.04414367675781, 406.06817626953125, 32.83659362792969, 88.2277603149414, -49.4945068359375, 210.59515380859375, 218.0283203125, 40.788414001464844, 341.0867614746094, 347.00933837890625, -143.760986328125, 78.12312316894531, 216.9889678955078, 178.14646911621094, 84.79869079589844, 547.4139404296875, 219.02337646484375, -219.98556518554688, 403.01751708984375, -195.14852905273438, 424.447998046875, 200.03009033203125, 366.708251953125, 177.3391876220703, 112.0492935180664, 365.12249755859375, 470.10162353515625, 239.21096801757812, 10.09866714477539, 273.7735900878906, 22.399486541748047, -131.42471313476562, 597.4432373046875, 400.23712158203125, 330.57537841796875, -245.1200408935547, 154.59527587890625, -201.56346130371094, -24.544139862060547, 145.66287231445312, 386.1705627441406, 60.282318115234375, 532.498291015625, 622.7669677734375, -128.4908447265625, 55.2678108215332, 232.9451904296875, -39.51188659667969], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000300.npy"}
|
|
{"epoch": 0.44052863436123346, "step": 301, "batch_size": 64, "mean": 176.1392059326172, "std": 193.44532775878906, "min": -133.37530517578125, "p10": -34.15375385284423, "median": 156.9959487915039, "p90": 417.51270141601566, "max": 812.8997802734375, "pos_frac": 0.84375, "sample": [154.13792419433594, 263.1689758300781, -46.74146270751953, 372.4145812988281, 316.0447692871094, 307.6274108886719, 583.890625, 105.13734436035156, 380.303955078125, -30.88624382019043, 583.3389892578125, 325.78399658203125, 110.81534576416016, 240.6866912841797, -75.10749816894531, 164.8992156982422, 84.05536651611328, 424.00604248046875, 57.61656188964844, 224.66502380371094, 636.4244995117188, 39.91057205200195, -35.554115295410156, 41.80247497558594, 326.14849853515625, 118.0242919921875, 60.48442840576172, 24.836990356445312, 221.83770751953125, 254.78518676757812, 278.446044921875, 180.2190704345703, 75.64242553710938, 114.95439910888672, 51.38153839111328, 84.4176025390625, 196.52989196777344, 3.6120452880859375, 10.005674362182617, 22.541961669921875, 104.45803833007812, 9.644887924194336, 176.73651123046875, -13.806266784667969, 18.047103881835938, 812.8997802734375, 217.44683837890625, 159.85397338867188, 402.361572265625, 164.41485595703125, 451.2275695800781, 168.49688720703125, 256.744873046875, 356.39422607421875, 315.04180908203125, 50.99330520629883, -102.19290161132812, -115.01826477050781, 469.6275329589844, -133.37530517578125, 72.79310607910156, -67.35122680664062, -26.00436019897461, 271.1651306152344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000301.npy"}
|
|
{"epoch": 0.4419970631424376, "step": 302, "batch_size": 64, "mean": 139.933349609375, "std": 176.7194366455078, "min": -183.6825714111328, "p10": -60.30549774169919, "median": 124.78667449951172, "p90": 382.3047149658204, "max": 671.868408203125, "pos_frac": 0.765625, "sample": [63.73854064941406, 329.8935546875, 460.27362060546875, -21.92757797241211, 171.45724487304688, -5.930418014526367, 218.04115295410156, 16.279571533203125, 117.83316040039062, 51.861419677734375, 276.8902893066406, -1.7378387451171875, 104.12621307373047, 67.1991958618164, 142.1995849609375, 5.331153869628906, 285.79437255859375, -3.596057891845703, 304.0595703125, 268.37060546875, 172.35357666015625, 132.40139770507812, 219.05160522460938, 201.02755737304688, -94.27241516113281, -70.74323272705078, 276.63189697265625, 406.1230773925781, 76.48076629638672, -171.7836456298828, 135.34002685546875, -35.950782775878906, 187.5702362060547, -183.6825714111328, -95.45460510253906, 441.6659851074219, 671.868408203125, 361.7659912109375, 131.7401885986328, -29.052703857421875, 11.726470947265625, 346.39227294921875, -172.25051879882812, 187.47940063476562, 302.76129150390625, 76.14861297607422, 38.72767639160156, 482.04718017578125, 167.0614776611328, 204.90023803710938, 6.606088638305664, 175.91676330566406, -142.58624267578125, 187.09410095214844, 69.08482360839844, 213.485107421875, -0.1852569580078125, 391.1070251464844, 83.24302673339844, 495.25823974609375, -26.370819091796875, 65.87364196777344, 103.6852798461914, 105.29095458984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000302.npy"}
|
|
{"epoch": 0.4434654919236417, "step": 303, "batch_size": 64, "mean": 133.57591247558594, "std": 184.2747039794922, "min": -255.99453735351562, "p10": -68.9985137939453, "median": 107.12435150146484, "p90": 413.642855834961, "max": 606.7400512695312, "pos_frac": 0.71875, "sample": [208.27801513671875, -1.8058662414550781, 45.77030944824219, 461.9517822265625, -101.40103912353516, 258.6346130371094, 352.4024658203125, 529.4505615234375, -1.7217121124267578, -50.881256103515625, 260.12982177734375, 183.20233154296875, 303.4625549316406, -79.96786499023438, 15.907585144042969, 122.80632019042969, 70.02897644042969, 67.42776489257812, -71.0169906616211, 308.90203857421875, -255.99453735351562, -39.62568283081055, -34.79692840576172, 249.57803344726562, 380.73602294921875, 477.40826416015625, -64.28873443603516, 233.73915100097656, -163.83880615234375, 134.70323181152344, -14.131927490234375, 215.79434204101562, 442.82305908203125, 164.82386779785156, 334.43206787109375, 73.78521728515625, -144.86962890625, 91.4423828125, 72.22779083251953, 606.7400512695312, 437.2297668457031, 65.1776123046875, 173.07894897460938, 417.9250793457031, 87.12216186523438, 124.58819580078125, 195.89495849609375, 279.828857421875, 127.26424407958984, -13.961126327514648, 127.34716796875, 4.326408386230469, 177.05291748046875, 43.93404006958008, 38.21197509765625, 403.6510009765625, -50.100154876708984, -49.308197021484375, 31.05972671508789, -29.936199188232422, 174.2909698486328, 26.054840087890625, -74.57050323486328, 190.44796752929688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000303.npy"}
|
|
{"epoch": 0.44493392070484583, "step": 304, "batch_size": 64, "mean": 157.3292694091797, "std": 131.16189575195312, "min": -187.43797302246094, "p10": 11.490444946289069, "median": 159.87114715576172, "p90": 317.2415283203125, "max": 481.32012939453125, "pos_frac": 0.90625, "sample": [35.74479675292969, 196.3828887939453, 271.34747314453125, 123.83061981201172, -4.0494384765625, -53.83229064941406, 24.380767822265625, 28.664138793945312, 102.89385986328125, 147.1903076171875, 244.79104614257812, 273.1712646484375, 134.18917846679688, 261.4649353027344, 89.2236328125, 246.62631225585938, 242.2325439453125, 172.54953002929688, 47.43486785888672, 57.235111236572266, 180.35214233398438, 318.4971923828125, 78.02598571777344, 53.551116943359375, 227.27474975585938, 117.20623779296875, 217.0110626220703, 31.653413772583008, 314.3116455078125, 163.4002227783203, 118.26719665527344, 143.68508911132812, 481.32012939453125, 100.97110748291016, -84.56314849853516, -187.43797302246094, 171.98910522460938, 42.40838623046875, 48.80961608886719, 215.88365173339844, -26.665679931640625, 403.132568359375, 37.39988708496094, 49.43926239013672, 379.9024353027344, 146.36334228515625, 8.953353881835938, 85.15462493896484, 236.28436279296875, 228.77713012695312, 275.171875, -57.45381546020508, 190.83775329589844, 430.6637268066406, 340.187255859375, 17.410324096679688, 300.85797119140625, 216.69281005859375, 242.08592224121094, 330.5467224121094, 265.0926818847656, 156.34207153320312, 244.02716064453125, 173.78097534179688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000304.npy"}
|
|
{"epoch": 0.44640234948604995, "step": 305, "batch_size": 64, "mean": 145.38482666015625, "std": 184.05686950683594, "min": -183.5518798828125, "p10": -36.786102294921875, "median": 115.3179817199707, "p90": 373.3438781738283, "max": 871.9810791015625, "pos_frac": 0.8125, "sample": [330.7234802246094, 227.16265869140625, 137.6048583984375, 171.1226806640625, 207.8662872314453, 608.8095092773438, 229.08517456054688, 18.939817428588867, 220.0716094970703, 433.1597900390625, 176.1455841064453, 70.32991790771484, 88.38066864013672, 33.8074951171875, -104.98516845703125, 115.15618133544922, 229.40963745117188, 332.6595153808594, 64.72171783447266, 90.3691635131836, 871.9810791015625, 259.272216796875, 42.63387680053711, 236.5533905029297, 297.41107177734375, -0.3905467987060547, 37.046669006347656, 230.57913208007812, 246.58094787597656, -125.11349487304688, -21.96508026123047, 194.16641235351562, -37.55610656738281, 283.9815673828125, 148.521240234375, 95.80418395996094, 391.4514465332031, 242.6766815185547, 38.45246887207031, 105.54393768310547, 79.94902801513672, 50.25782012939453, 115.47978210449219, 63.656272888183594, 418.3939208984375, 433.90338134765625, 192.10337829589844, -33.679595947265625, -154.17092895507812, 148.68017578125, -177.88558959960938, 386.35247802734375, 342.990478515625, 76.1493148803711, -34.98942565917969, 22.060546875, -62.729774475097656, 26.35240936279297, 191.67977905273438, -183.5518798828125, 37.201454162597656, 161.08499145507812, -28.367568969726562, 15.537212371826172], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000305.npy"}
|
|
{"epoch": 0.447870778267254, "step": 306, "batch_size": 64, "mean": 141.0995635986328, "std": 164.1031494140625, "min": -252.19747924804688, "p10": -23.175714874267577, "median": 105.43740844726562, "p90": 357.08369445800787, "max": 782.18896484375, "pos_frac": 0.828125, "sample": [28.450305938720703, 8.864490509033203, 73.69796752929688, 18.835853576660156, -23.71759033203125, -30.447219848632812, 294.2040100097656, 59.931617736816406, 61.6099853515625, 7.156013488769531, 188.1041717529297, 372.4459228515625, 28.246505737304688, 296.54132080078125, 58.54705810546875, 114.87203216552734, 19.414798736572266, 396.3785095214844, 77.38646697998047, 46.57582092285156, 49.56957244873047, -21.911338806152344, -10.824981689453125, 345.2924499511719, 201.57211303710938, 315.6787109375, 8.14974594116211, -5.403507232666016, 335.1242980957031, 178.5652313232422, 31.947998046875, 144.203857421875, 234.51687622070312, 156.89996337890625, 73.24700927734375, 216.4071807861328, 58.951568603515625, -33.248069763183594, 55.888607025146484, 239.8762664794922, -252.19747924804688, 264.4317626953125, 96.0027847290039, -77.70268249511719, -18.954633712768555, 251.07757568359375, -25.09943389892578, 140.31716918945312, 448.3674011230469, 404.01568603515625, 188.55227661132812, 272.30828857421875, 228.8783416748047, -42.053932189941406, 362.1370849609375, 17.919536590576172, 115.66441345214844, 39.346763610839844, 201.2578582763672, 126.14498901367188, 782.18896484375, 234.2364501953125, 219.0442657470703, 382.8868103027344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000306.npy"}
|
|
{"epoch": 0.44933920704845814, "step": 307, "batch_size": 64, "mean": 133.7398681640625, "std": 167.43890380859375, "min": -400.1659240722656, "p10": -44.62353591918945, "median": 94.4310188293457, "p90": 351.13260498046884, "max": 579.4926147460938, "pos_frac": 0.8125, "sample": [238.73098754882812, 99.24500274658203, 83.24715423583984, 159.7213134765625, 160.30355834960938, 137.11863708496094, -73.2158203125, 47.700706481933594, -44.52610778808594, 436.38262939453125, 41.029754638671875, 36.19452667236328, 103.5333251953125, -100.03929138183594, 304.626708984375, -400.1659240722656, 458.5625915527344, -91.62602233886719, 236.855712890625, -44.66529083251953, 169.87841796875, 208.62371826171875, 74.12981414794922, 173.73194885253906, 452.34368896484375, 177.01132202148438, 80.07307434082031, 85.02621459960938, 67.8387451171875, 314.2158203125, 87.72152709960938, 372.7228698730469, 67.13651275634766, 68.24494934082031, 6.0108489990234375, 89.61703491210938, 62.753662109375, 74.14179992675781, 331.4383544921875, 323.346923828125, -18.325706481933594, 252.93862915039062, 9.529197692871094, 208.013671875, 196.9309539794922, 33.18266296386719, 188.6115264892578, 237.92979431152344, -18.07489585876465, 72.90303039550781, 165.11129760742188, 10.133598327636719, 359.572998046875, 497.28350830078125, 308.0022888183594, 579.4926147460938, 123.40911102294922, -48.796085357666016, 227.22389221191406, 53.132415771484375, -11.304367065429688, 175.73385620117188, -20.345901489257812, -97.95833587646484], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000307.npy"}
|
|
{"epoch": 0.45080763582966227, "step": 308, "batch_size": 64, "mean": 158.48333740234375, "std": 151.64788818359375, "min": -193.94088745117188, "p10": -35.81675434112548, "median": 164.01773071289062, "p90": 374.9291015625, "max": 447.242919921875, "pos_frac": 0.875, "sample": [159.0962677001953, 345.832275390625, -131.22171020507812, 181.61224365234375, 24.53661346435547, 392.7684326171875, -193.94088745117188, 16.387210845947266, 91.34588623046875, 85.63525390625, 376.29034423828125, 15.23782730102539, 174.46810913085938, 94.90509033203125, 292.95672607421875, 388.7867126464844, -91.44895935058594, 169.45535278320312, 122.59905242919922, 17.711647033691406, 180.8932342529297, 441.04095458984375, 287.3865966796875, 371.75286865234375, -39.57365417480469, 223.15464782714844, 102.90451049804688, 111.40760803222656, 111.9998779296875, 25.807832717895508, 225.71762084960938, 108.06952667236328, 182.68533325195312, 244.42628479003906, 195.9082489013672, 361.9531555175781, 418.5626525878906, 63.063995361328125, -39.26490783691406, 42.53443908691406, 30.448057174682617, 270.0020751953125, 227.84457397460938, 168.93919372558594, 225.83253479003906, 395.6829528808594, 242.44912719726562, 112.26484680175781, 31.358543395996094, 214.78033447265625, 447.242919921875, 256.5406799316406, 143.88278198242188, 31.031038284301758, -27.77106285095215, 59.097930908203125, 366.00152587890625, 328.7072448730469, 231.68016052246094, 22.01154327392578, 290.3990173339844, 40.123191833496094, -75.50688171386719, -43.552406311035156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000308.npy"}
|
|
{"epoch": 0.4522760646108664, "step": 309, "batch_size": 64, "mean": 157.86776733398438, "std": 169.70144653320312, "min": -342.90753173828125, "p10": -21.77799377441406, "median": 136.22009658813477, "p90": 390.16607055664065, "max": 558.20458984375, "pos_frac": 0.84375, "sample": [92.47061157226562, 62.18501281738281, 45.94025421142578, -56.1243896484375, 195.67965698242188, 144.70989990234375, 105.70362091064453, -12.310916900634766, 422.34881591796875, 29.04888916015625, 296.907958984375, 309.71356201171875, 83.6269302368164, 438.5242614746094, 127.73029327392578, 53.50762176513672, -45.34762954711914, 260.7948303222656, 331.3020935058594, 270.3267822265625, 217.58709716796875, -342.90753173828125, 210.99148559570312, -22.221145629882812, 104.30410766601562, 118.13614654541016, 236.501708984375, 237.8682403564453, 29.595626831054688, 100.26718139648438, -116.90570831298828, 544.3455810546875, 50.6412353515625, 40.74635314941406, 108.17755126953125, 400.7799072265625, 558.20458984375, 305.82696533203125, 98.79503631591797, -20.743972778320312, 28.038379669189453, 61.091888427734375, 103.45936584472656, 354.26251220703125, 208.447998046875, 244.8343505859375, 382.4915771484375, 224.7524871826172, 252.98545837402344, 19.806575775146484, 332.6380310058594, -173.04351806640625, 183.40835571289062, 268.2442321777344, -87.38204956054688, 47.93693923950195, 418.86419677734375, 393.45513916015625, -17.746932983398438, 152.28762817382812, 170.43304443359375, 76.60676574707031, 282.9849548339844, 157.9508819580078], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000309.npy"}
|
|
{"epoch": 0.45374449339207046, "step": 310, "batch_size": 64, "mean": 139.72027587890625, "std": 134.702880859375, "min": -142.68408203125, "p10": -46.63193664550781, "median": 146.856689453125, "p90": 327.72092590332045, "max": 441.7794189453125, "pos_frac": 0.84375, "sample": [-89.73280334472656, 119.3542251586914, 269.5343322753906, -25.765769958496094, 97.42918395996094, 244.9861297607422, 86.28984069824219, -48.737022399902344, 50.51261901855469, 185.65200805664062, -102.37635040283203, 234.57420349121094, 353.8284606933594, 43.1955680847168, 52.125587463378906, 182.177001953125, 192.3699493408203, 195.2691192626953, 95.893310546875, 201.39634704589844, 45.9450798034668, 55.97221374511719, 250.115966796875, 58.1422119140625, 106.80935668945312, 392.42822265625, 246.8728485107422, 11.24411392211914, 27.847185134887695, 142.96212768554688, -142.68408203125, 377.2894592285156, 27.903697967529297, 128.24896240234375, 441.7794189453125, 173.57833862304688, 246.7039794921875, 3.7864551544189453, -51.989112854003906, -41.720069885253906, 62.963043212890625, 157.01358032226562, 296.90435791015625, 202.5869598388672, 114.23456573486328, 281.74041748046875, 395.515380859375, -48.797767639160156, 339.08233642578125, 150.75125122070312, 197.9190673828125, -74.5933609008789, 185.39794921875, 185.75950622558594, -9.439834594726562, 46.59149932861328, 90.00335693359375, 342.9851379394531, 301.2109680175781, 184.89610290527344, 69.86651611328125, 263.1475830078125, 207.03472900390625, 160.1121063232422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000310.npy"}
|
|
{"epoch": 0.4552129221732746, "step": 311, "batch_size": 64, "mean": 170.3639373779297, "std": 193.82199096679688, "min": -157.18690490722656, "p10": -53.70741882324218, "median": 167.67792510986328, "p90": 423.68232421875, "max": 690.6173095703125, "pos_frac": 0.765625, "sample": [316.2568359375, 313.14923095703125, 328.5816345214844, 24.22193145751953, -119.4195327758789, 349.10247802734375, -42.101016998291016, 252.14112854003906, 180.4232635498047, 466.7137145996094, 210.748779296875, 57.90130615234375, 314.4328308105469, 650.2058715820312, 227.38507080078125, 89.72891998291016, 363.75244140625, 100.6934585571289, 9.906883239746094, 420.2241516113281, 192.3075408935547, 225.67904663085938, 425.1643981933594, 217.07688903808594, 65.7920150756836, 216.25013732910156, 15.337600708007812, -126.87277221679688, 150.64031982421875, 130.07330322265625, -64.92800903320312, 251.83758544921875, -83.82140350341797, 221.7931671142578, 288.83660888671875, -105.33563232421875, -0.8139495849609375, 690.6173095703125, -14.2047119140625, 252.34320068359375, 51.15821075439453, 77.09015655517578, 215.712158203125, 43.43397521972656, 479.26385498046875, 198.74441528320312, 4.508411407470703, -19.829191207885742, 59.58993911743164, 101.58065795898438, -10.11520004272461, 495.929931640625, 136.00384521484375, -26.72968292236328, -157.18690490722656, 403.75384521484375, 173.8888702392578, 161.46697998046875, -18.850624084472656, 196.98020935058594, 370.24835205078125, -58.68159103393555, -22.197731018066406, 585.7069091796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000311.npy"}
|
|
{"epoch": 0.4566813509544787, "step": 312, "batch_size": 64, "mean": 156.37979125976562, "std": 176.18411254882812, "min": -185.68624877929688, "p10": -8.555122947692867, "median": 144.9092788696289, "p90": 370.79314575195315, "max": 663.0660400390625, "pos_frac": 0.875, "sample": [112.17257690429688, 650.3129272460938, 417.61181640625, 119.578125, 88.40449523925781, 74.29053497314453, 4.270286560058594, 61.96446228027344, 142.1551513671875, 363.803955078125, -185.68624877929688, 383.0986633300781, 663.0660400390625, 219.8181610107422, 186.89422607421875, 177.78787231445312, -165.92877197265625, 167.67645263671875, 297.9735107421875, 191.22451782226562, 338.1081237792969, 12.422252655029297, 101.79806518554688, 273.8491516113281, 111.67526245117188, 3.404237747192383, 31.851097106933594, 481.1888122558594, 268.8940124511719, 335.5409851074219, 111.91342163085938, 290.8135681152344, 153.26535034179688, 147.6634063720703, -71.81460571289062, 108.68104553222656, 184.09791564941406, 373.78851318359375, 48.07246398925781, 90.55264282226562, 259.59552001953125, 105.10123443603516, -10.008035659790039, 272.330322265625, 159.06103515625, 45.0335693359375, -17.60186767578125, 579.7923583984375, -5.1649932861328125, 48.282142639160156, 202.28521728515625, 159.7671661376953, -165.3818359375, 39.91316223144531, 4.474983215332031, 178.8800811767578, 158.5003204345703, 61.24757385253906, -175.39483642578125, 261.2132263183594, 222.50497436523438, 11.729736328125, 40.00667190551758, 205.88482666015625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000312.npy"}
|
|
{"epoch": 0.4581497797356828, "step": 313, "batch_size": 64, "mean": 145.47706604003906, "std": 182.49468994140625, "min": -301.64813232421875, "p10": -62.4614608764648, "median": 141.07061004638672, "p90": 350.485629272461, "max": 602.5475463867188, "pos_frac": 0.8125, "sample": [86.56573486328125, 250.68197631835938, 32.877655029296875, 282.58880615234375, -110.63714599609375, 59.31150817871094, 357.3419494628906, 153.33419799804688, 319.32952880859375, 15.97641372680664, 29.752151489257812, -79.48820495605469, 263.21893310546875, 137.27281188964844, 17.381284713745117, 98.84434509277344, 261.66143798828125, 169.39134216308594, 80.32759094238281, 156.96078491210938, 40.25623321533203, 321.2814025878906, -261.1817932128906, -118.15522766113281, 150.1482696533203, 403.9139099121094, 246.3174591064453, 86.00640869140625, 221.2159423828125, 559.2996826171875, 298.28546142578125, 162.82269287109375, 13.381532669067383, 90.62223815917969, -219.779541015625, 33.21049499511719, -22.302499771118164, -301.64813232421875, 334.487548828125, 186.1316680908203, -14.255958557128906, 201.04632568359375, 34.448246002197266, 233.959716796875, 304.6926574707031, 144.868408203125, 41.777870178222656, 299.53607177734375, 260.02349853515625, -140.92308044433594, 293.7458801269531, 124.37033081054688, 102.89000701904297, -0.20211029052734375, 130.9628143310547, 475.36456298828125, 473.2933044433594, -3.8484649658203125, -22.732391357421875, 64.80117797851562, 225.02667236328125, 286.5377197265625, 385.5948486328125, 602.5475463867188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000313.npy"}
|
|
{"epoch": 0.45961820851688695, "step": 314, "batch_size": 64, "mean": 137.9976806640625, "std": 187.06932067871094, "min": -306.57574462890625, "p10": -76.94926605224609, "median": 99.4889907836914, "p90": 341.99182739257816, "max": 625.9620361328125, "pos_frac": 0.75, "sample": [532.77685546875, -190.60693359375, 125.11272430419922, 57.23241424560547, 40.36871337890625, 237.96963500976562, 80.82920837402344, 224.257080078125, 57.960262298583984, -66.12110900878906, 235.05438232421875, -1.6900482177734375, 221.083740234375, 119.75810241699219, 163.4699249267578, -306.57574462890625, -111.62644958496094, 82.61605834960938, 516.2376708984375, 532.8966674804688, 22.18865966796875, -29.2575626373291, -76.50222778320312, 76.59059143066406, 79.24908447265625, -109.02323913574219, 35.25917053222656, -107.57160949707031, 11.171455383300781, 329.7173767089844, 344.0736083984375, 36.380859375, 243.919677734375, 337.13433837890625, 267.3460998535156, 102.21122741699219, -37.096168518066406, 135.05943298339844, 89.2479248046875, 487.85723876953125, 320.5577697753906, 324.5603942871094, 625.9620361328125, 297.07012939453125, -6.177097320556641, 234.56378173828125, -110.98963165283203, 347.32464599609375, -12.817337036132812, 153.39776611328125, 202.40611267089844, 36.708221435546875, 265.6004638671875, 288.78570556640625, 308.6706237792969, 281.67291259765625, 85.05558776855469, 96.76675415039062, 150.79112243652344, 25.60442352294922, -42.00700378417969, -27.504261016845703, 274.0307312011719, -77.14085388183594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000314.npy"}
|
|
{"epoch": 0.461086637298091, "step": 315, "batch_size": 64, "mean": 133.40902709960938, "std": 158.90463256835938, "min": -135.39093017578125, "p10": -33.13536739349364, "median": 103.53678512573242, "p90": 343.2328735351563, "max": 595.054443359375, "pos_frac": 0.8125, "sample": [358.4840087890625, 9.560495376586914, 166.2742156982422, 39.02935791015625, -47.115413665771484, 81.43461608886719, 45.76189422607422, 168.20204162597656, 188.67515563964844, -19.326278686523438, 238.15309143066406, -12.716032028198242, -99.99385070800781, 157.57760620117188, 62.247642517089844, 406.2615966796875, 276.2301025390625, 142.14764404296875, -114.95563507080078, 296.4870910644531, 126.1790771484375, 432.1195068359375, 334.39984130859375, 36.16400146484375, -37.45098876953125, 129.52804565429688, 310.14825439453125, 114.91104888916016, 346.9339294433594, 125.21083068847656, 389.5862121582031, 151.35792541503906, 51.148773193359375, 92.16252136230469, 175.01869201660156, 266.7917785644531, 25.4527587890625, 23.814666748046875, 189.86302185058594, 19.238006591796875, 30.035511016845703, 41.525238037109375, 595.054443359375, 62.67340850830078, 215.68817138671875, 17.563987731933594, -135.21786499023438, -10.891324996948242, -58.92304992675781, 334.5970764160156, 56.0765380859375, 160.38290405273438, 555.1455078125, 86.19696807861328, 167.63047790527344, -23.065584182739258, 277.5861511230469, 28.704776763916016, 271.9015197753906, 259.54833984375, -135.39093017578125, -20.90162467956543, 58.29738998413086, 58.962425231933594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000315.npy"}
|
|
{"epoch": 0.46255506607929514, "step": 316, "batch_size": 64, "mean": 178.52488708496094, "std": 187.04168701171875, "min": -171.85369873046875, "p10": -4.9105407714843725, "median": 144.7793426513672, "p90": 503.9295593261719, "max": 686.0717163085938, "pos_frac": 0.875, "sample": [260.92950439453125, 97.35820007324219, 420.17950439453125, 54.10382080078125, 104.6977767944336, 598.5847778320312, 134.3873291015625, 526.1558837890625, 336.632568359375, 503.751953125, 279.7850646972656, 265.9111328125, 16.317306518554688, -12.34994125366211, 1.5492782592773438, 197.5396270751953, 114.18537902832031, 130.37527465820312, 2.2761497497558594, 274.36126708984375, 70.04131317138672, 47.42142105102539, 548.7164916992188, 308.59423828125, 212.51226806640625, 74.83345794677734, 68.49398803710938, 504.00567626953125, 145.32090759277344, 391.0758972167969, 26.720117568969727, 93.3695297241211, 34.98529052734375, 181.74794006347656, 273.7682800292969, 255.6739044189453, 302.5625, 45.680023193359375, 174.29052734375, 11.645248413085938, 506.90374755859375, 207.83572387695312, 25.032135009765625, 289.1320495605469, 161.09950256347656, 131.40774536132812, -157.6385955810547, 69.52006530761719, 686.0717163085938, 144.23777770996094, 517.8582763671875, -171.85369873046875, 195.54620361328125, 155.3347625732422, 258.91156005859375, -5.881462097167969, -44.121299743652344, -50.90010070800781, -2.6450576782226562, 23.734031677246094, 223.95468139648438, -121.90962982177734, 86.65531921386719, 219.11648559570312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000316.npy"}
|
|
{"epoch": 0.46402349486049926, "step": 317, "batch_size": 64, "mean": 207.04443359375, "std": 178.2103729248047, "min": -121.43312072753906, "p10": -31.304811286926263, "median": 189.26093292236328, "p90": 457.6822174072266, "max": 568.2066650390625, "pos_frac": 0.859375, "sample": [300.01177978515625, 75.71845245361328, 3.6114578247070312, 374.1099853515625, 430.949951171875, 67.71725463867188, 285.8331604003906, 312.60015869140625, -121.43312072753906, 431.671142578125, 568.2066650390625, 120.23301696777344, 319.483642578125, 32.4359130859375, 142.48544311523438, 307.0623779296875, 177.1127166748047, 22.732643127441406, 146.4261474609375, 311.8573913574219, 33.888397216796875, 278.20623779296875, 308.2975158691406, 469.165283203125, 45.17072677612305, 417.4470520019531, -76.77362060546875, 97.93671417236328, 560.2766723632812, 273.4031982421875, 461.27325439453125, 389.3671875, 107.03282165527344, 475.3792419433594, 322.60400390625, -12.247879028320312, 173.69537353515625, 142.7847442626953, 275.58349609375, 128.35372924804688, 118.48455810546875, 449.3031311035156, -23.91041374206543, -36.14390563964844, 475.67681884765625, 230.39068603515625, 372.3589172363281, 234.20053100585938, 223.43746948242188, -96.04500579833984, 138.00526428222656, 146.13565063476562, 201.40914916992188, 34.26325225830078, -34.473838806152344, 153.56753540039062, 111.81039428710938, -76.43077850341797, 525.6512451171875, 312.74951171875, 76.72391510009766, 329.90484619140625, -65.36656188964844, 269.4712219238281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000317.npy"}
|
|
{"epoch": 0.4654919236417034, "step": 318, "batch_size": 64, "mean": 165.05731201171875, "std": 170.49114990234375, "min": -212.17262268066406, "p10": -33.27899932861326, "median": 147.05197143554688, "p90": 380.36542358398447, "max": 620.90283203125, "pos_frac": 0.84375, "sample": [77.6765365600586, -15.266250610351562, 602.8320922851562, 146.6500244140625, 224.44131469726562, 105.51246643066406, -3.9650821685791016, 162.1326446533203, 108.11528778076172, 620.90283203125, 25.33707618713379, 117.47119140625, 219.98703002929688, 321.42877197265625, 468.93597412109375, 300.38043212890625, 11.744110107421875, 84.30894470214844, -41.617027282714844, -10.025613784790039, 397.13018798828125, 48.496864318847656, -93.66587829589844, 423.15386962890625, 508.6109924316406, 274.0701904296875, -61.788692474365234, 2.770465850830078, 207.9312286376953, 321.4529113769531, 50.875030517578125, 310.07293701171875, 49.25714874267578, 339.6813049316406, 183.0118865966797, 127.20355987548828, 387.6026306152344, 0.5028228759765625, -40.998748779296875, 363.4786071777344, 363.025634765625, 196.97000122070312, 38.1578254699707, 86.8044662475586, 250.86375427246094, 169.9307861328125, 147.45391845703125, 230.96966552734375, 59.78852844238281, -44.20658874511719, -212.17262268066406, 94.57477569580078, 212.66506958007812, 198.26678466796875, 308.67047119140625, 292.42999267578125, 183.7242431640625, 141.82815551757812, 43.90193176269531, 200.01657104492188, 97.6489028930664, 260.4954528808594, -119.44112396240234, 35.47032928466797], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000318.npy"}
|
|
{"epoch": 0.4669603524229075, "step": 319, "batch_size": 64, "mean": 175.37954711914062, "std": 184.80685424804688, "min": -220.764404296875, "p10": -70.11929931640624, "median": 181.1000518798828, "p90": 394.20151672363284, "max": 691.3362426757812, "pos_frac": 0.796875, "sample": [230.41845703125, 141.95423889160156, 244.75326538085938, 219.28057861328125, 138.7898406982422, 259.09942626953125, 277.7824401855469, 327.52874755859375, 229.58837890625, 61.755958557128906, -57.42536926269531, -192.58392333984375, 239.02554321289062, 131.05982971191406, 181.52468872070312, 151.7714385986328, -0.8042831420898438, 198.46336364746094, 299.4998474121094, 147.87069702148438, 121.68641662597656, 294.47882080078125, 149.20541381835938, 287.3709716796875, 243.6634979248047, 27.563140869140625, 317.35443115234375, 623.4603271484375, 261.74798583984375, 327.35711669921875, -95.66488647460938, 120.41548156738281, 117.1129150390625, 391.50048828125, -220.764404296875, 223.28762817382812, -209.22109985351562, -77.9447021484375, -47.476806640625, 308.1301574707031, 447.9089050292969, 395.3591003417969, 437.1882019042969, -20.644529342651367, 172.89695739746094, 8.501113891601562, 438.28948974609375, 439.5237731933594, 237.91004943847656, 381.3487548828125, 691.3362426757812, -55.02607727050781, 136.38311767578125, 320.733154296875, -75.55955505371094, 31.92278289794922, -93.4290542602539, -54.213462829589844, 217.11776733398438, 134.54513549804688, 125.8646240234375, 181.02981567382812, 181.1702880859375, 151.5189666748047], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000319.npy"}
|
|
{"epoch": 0.4684287812041116, "step": 320, "batch_size": 64, "mean": 201.1544189453125, "std": 195.6520538330078, "min": -152.15512084960938, "p10": -17.785831451416016, "median": 185.25377655029297, "p90": 456.5531982421876, "max": 826.6787109375, "pos_frac": 0.84375, "sample": [154.05906677246094, 192.73297119140625, 651.4869384765625, 108.32466125488281, 274.5611572265625, -14.028770446777344, 192.67910766601562, -152.15512084960938, 671.25048828125, 151.23402404785156, 372.13330078125, 249.01205444335938, -37.943084716796875, 274.5697021484375, 174.51095581054688, 466.26715087890625, 430.9317626953125, 245.32925415039062, 285.8480224609375, -77.87427520751953, 277.7027587890625, 297.75048828125, 40.34661102294922, 131.54962158203125, 68.951416015625, 77.36663055419922, 285.46185302734375, 230.01971435546875, -13.858047485351562, 463.41485595703125, -35.76130676269531, 247.23916625976562, 41.235069274902344, 204.72866821289062, 341.1214599609375, 116.006591796875, 99.1837158203125, 156.85350036621094, 826.6787109375, 180.65048217773438, 395.4102478027344, 66.71533966064453, 124.679443359375, -18.06146240234375, 548.9840087890625, 42.08721160888672, 276.366943359375, 169.08876037597656, 5.192171096801758, 23.593788146972656, 238.69314575195312, 276.60125732421875, 189.85707092285156, 339.7649841308594, 374.1623840332031, 507.45880126953125, 440.54266357421875, 9.118391036987305, -68.68545532226562, -17.14269256591797, 32.45733642578125, 165.58457946777344, 196.78192138671875, -94.93997192382812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000320.npy"}
|
|
{"epoch": 0.4698972099853157, "step": 321, "batch_size": 64, "mean": 173.90859985351562, "std": 212.14559936523438, "min": -305.88330078125, "p10": -53.57699470520017, "median": 168.83988189697266, "p90": 444.51282043457036, "max": 743.3228759765625, "pos_frac": 0.796875, "sample": [585.6917114257812, 127.9723129272461, 264.80303955078125, 236.3909912109375, -9.320276260375977, 295.8612365722656, 273.3714904785156, 193.14385986328125, 122.05718994140625, 431.76885986328125, 339.75390625, 298.58148193359375, -2.534027099609375, -30.228633880615234, -184.4424591064453, 246.75779724121094, 400.50543212890625, 257.1640625, 168.23068237304688, 402.5921630859375, 607.5233154296875, 120.1677017211914, -305.88330078125, 235.72946166992188, 169.44908142089844, 276.9023132324219, 461.7969970703125, -287.8324890136719, -82.99263000488281, 161.18760681152344, 97.65629577636719, 449.9745178222656, -15.25714111328125, 310.8377685546875, 159.16995239257812, 366.8624267578125, 184.17543029785156, -180.5, 743.3228759765625, 92.4034194946289, -63.58343505859375, 67.5286636352539, 29.734140396118164, 124.59283447265625, 15.194404602050781, 177.12045288085938, 27.26519775390625, 4.676219940185547, 197.10903930664062, 233.19102478027344, 586.1510620117188, 517.380859375, -5.664287567138672, -244.8311767578125, 184.5731201171875, 150.95059204101562, 127.9197769165039, 174.91592407226562, 10.251893997192383, 370.47576904296875, 112.143310546875, 230.29400634765625, 140.8497314453125, -20.902923583984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000321.npy"}
|
|
{"epoch": 0.4713656387665198, "step": 322, "batch_size": 64, "mean": 187.88018798828125, "std": 192.41372680664062, "min": -213.794189453125, "p10": -43.10749359130858, "median": 197.9693603515625, "p90": 454.509423828125, "max": 636.150390625, "pos_frac": 0.78125, "sample": [-98.58417510986328, 636.150390625, 215.71853637695312, 628.4307861328125, 65.0060043334961, -129.38087463378906, 467.7529296875, 315.3819885253906, 265.2144470214844, -89.12509155273438, -34.20976257324219, -4.261558532714844, 303.54998779296875, 234.07571411132812, 271.97174072265625, 314.2703857421875, 22.16339874267578, 189.76434326171875, -10.508644104003906, -46.920806884765625, 486.1455383300781, 458.0875244140625, 204.75140380859375, 482.04144287109375, -8.216644287109375, 191.18731689453125, 262.53204345703125, 446.1605224609375, -63.90270233154297, 428.11993408203125, -5.978986740112305, 165.35491943359375, 54.63850784301758, 341.2890625, -32.417205810546875, 20.00619125366211, 243.1505889892578, 157.95433044433594, 12.982257843017578, 190.35845947265625, 325.36065673828125, 404.90008544921875, 393.60675048828125, -75.67321014404297, 344.724365234375, 149.48057556152344, 157.6465301513672, -23.288982391357422, 226.7890625, 48.40071105957031, 130.135009765625, 530.0419311523438, 221.71578979492188, 260.0188903808594, 285.6330871582031, 281.79376220703125, 74.91328430175781, 104.6150131225586, 20.393495559692383, -213.794189453125, 22.674957275390625, 239.9476776123047, 346.29547119140625, 217.29725646972656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000322.npy"}
|
|
{"epoch": 0.47283406754772395, "step": 323, "batch_size": 64, "mean": 159.92843627929688, "std": 196.71099853515625, "min": -366.0095520019531, "p10": -37.99214706420897, "median": 171.97760772705078, "p90": 454.1925964355469, "max": 653.9987182617188, "pos_frac": 0.8125, "sample": [372.5679016113281, 240.800537109375, 460.84735107421875, -366.0095520019531, 507.2806091308594, -186.81509399414062, 450.8760986328125, 653.9987182617188, -12.988323211669922, 93.07323455810547, 258.0725402832031, 30.62457275390625, 119.56973266601562, 132.39364624023438, 8.416786193847656, 182.9906463623047, 87.5835189819336, 219.95526123046875, 5.849876403808594, 531.4373168945312, -13.685981750488281, 44.00419616699219, 475.999755859375, -25.1514835357666, -71.65101623535156, 309.2489318847656, 79.45764923095703, 146.6501922607422, 205.32752990722656, 87.65847778320312, 288.1695556640625, 195.7989501953125, 242.7338104248047, -42.158050537109375, 267.601318359375, 15.27557373046875, 59.65031433105469, 108.7053451538086, 267.89154052734375, 253.64352416992188, 251.88015747070312, 51.2818603515625, 181.13687133789062, 436.9420166015625, -272.9084777832031, 168.48121643066406, 285.1706848144531, 455.61395263671875, 298.8202209472656, -28.271705627441406, 212.96499633789062, 473.33746337890625, -84.06951141357422, 199.1821746826172, -211.9909210205078, 264.5938720703125, 102.50738525390625, 200.381103515625, 49.313987731933594, 247.66517639160156, 43.947410583496094, -11.613605499267578, 175.4739990234375, 59.884063720703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000323.npy"}
|
|
{"epoch": 0.47430249632892807, "step": 324, "batch_size": 64, "mean": 170.25628662109375, "std": 225.35804748535156, "min": -312.08734130859375, "p10": -109.52237243652344, "median": 155.81209564208984, "p90": 435.4815582275391, "max": 807.33984375, "pos_frac": 0.765625, "sample": [-110.3428955078125, 540.1260986328125, -68.19744873046875, 148.43881225585938, 107.65477752685547, 225.76251220703125, 204.512939453125, 253.37289428710938, 208.48362731933594, 187.9404296875, 694.213623046875, -46.723175048828125, 226.33151245117188, 86.07984924316406, 6.977519989013672, 608.8978271484375, -156.44371032714844, -220.1188201904297, 258.6930847167969, 207.49227905273438, 26.819000244140625, 153.53204345703125, -131.77914428710938, 179.64773559570312, 251.6592559814453, 742.1600952148438, 335.91180419921875, -155.8714141845703, 109.60519409179688, 371.6752014160156, 430.4968566894531, 86.59797668457031, 90.78112030029297, 71.61246490478516, 341.41131591796875, 291.9564514160156, 124.43345642089844, 286.9115905761719, 144.82229614257812, -6.089197158813477, 322.1529541015625, -312.08734130859375, 192.7664794921875, 344.2444763183594, 437.61785888671875, 330.6478271484375, 79.30470275878906, -214.23619079589844, -107.60781860351562, 147.267822265625, 192.72314453125, 94.69780731201172, 158.09214782714844, -25.742855072021484, 127.85237121582031, 140.8088836669922, 369.1610107421875, -99.49903106689453, -22.36935806274414, 807.33984375, 445.3902282714844, 159.3251495361328, -16.149765014648438, 235.25506591796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000324.npy"}
|
|
{"epoch": 0.47577092511013214, "step": 325, "batch_size": 64, "mean": 137.27645874023438, "std": 165.57345581054688, "min": -311.2183837890625, "p10": -25.161759948730463, "median": 125.48783111572266, "p90": 352.65723876953143, "max": 552.50390625, "pos_frac": 0.859375, "sample": [-244.25076293945312, -311.2183837890625, 212.6336669921875, 6.107093811035156, 280.7339782714844, 232.72622680664062, 88.56232452392578, -67.14643859863281, 200.1522216796875, 302.35430908203125, 235.67982482910156, 374.21563720703125, 265.25653076171875, 130.66464233398438, 186.22120666503906, 82.2470474243164, 58.691017150878906, 147.4149932861328, 25.645950317382812, 206.40052795410156, -28.29857635498047, 274.2364807128906, 259.1097106933594, 396.3554992675781, 18.492084503173828, -17.84252166748047, 120.31101989746094, 58.161766052246094, 97.76161193847656, 5.437126159667969, 284.19696044921875, 31.218338012695312, 552.50390625, 0.027744293212890625, 444.3695068359375, -34.426856994628906, 426.52777099609375, 1.2168693542480469, 202.66409301757812, 214.47097778320312, -12.342157363891602, 88.6749496459961, 220.49435424804688, -42.552398681640625, 240.49444580078125, 233.02706909179688, 66.56685638427734, 190.82479858398438, 54.025238037109375, 173.7452392578125, 237.73281860351562, -210.94989013671875, 18.855091094970703, 536.2310791015625, 27.25714111328125, 108.10245513916016, 378.372802734375, 106.78260040283203, 25.685211181640625, 185.16415405273438, 145.89501953125, 33.58562088012695, 45.30724334716797, 215.12832641601562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000325.npy"}
|
|
{"epoch": 0.47723935389133626, "step": 326, "batch_size": 64, "mean": 141.99554443359375, "std": 143.06173706054688, "min": -180.05096435546875, "p10": -9.71599712371826, "median": 124.07331085205078, "p90": 351.5285064697268, "max": 523.932861328125, "pos_frac": 0.828125, "sample": [22.523193359375, 53.926517486572266, 99.15574645996094, 375.7966003417969, 184.2567596435547, -2.4034423828125, 89.87799072265625, 102.83323669433594, 257.6988830566406, -150.97903442382812, -2.9685707092285156, -180.05096435546875, 451.2969970703125, 38.293724060058594, -37.371421813964844, 146.90713500976562, 146.34097290039062, 61.39157485961914, 182.43670654296875, 243.37155151367188, 83.22366333007812, 136.43429565429688, 232.50242614746094, 230.87722778320312, -8.825775146484375, 115.93766784667969, 523.932861328125, 160.2107696533203, 58.122894287109375, 67.21879577636719, 249.109130859375, 173.55487060546875, 238.93296813964844, 143.42471313476562, 183.75711059570312, -7.938751220703125, 61.86492919921875, 204.27288818359375, -42.359378814697266, 165.6312255859375, 130.61207580566406, 63.99371337890625, 117.5345458984375, 108.12051391601562, 26.83270263671875, 87.34727478027344, 171.3070068359375, 267.8326721191406, 115.43330383300781, 208.2886199951172, -32.76527404785156, 385.1408386230469, 497.9736328125, 59.94054412841797, 281.32958984375, 199.45086669921875, 395.72613525390625, 246.18170166015625, -10.09752082824707, -61.09434509277344, 380.1474914550781, 294.9029541015625, 15.592578887939453, 85.76543426513672], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000326.npy"}
|
|
{"epoch": 0.4787077826725404, "step": 327, "batch_size": 64, "mean": 198.74432373046875, "std": 186.60940551757812, "min": -200.93783569335938, "p10": -54.38105163574217, "median": 206.63174438476562, "p90": 440.6451110839844, "max": 632.880615234375, "pos_frac": 0.859375, "sample": [185.74728393554688, 149.52870178222656, 438.03399658203125, 218.1657257080078, 100.04579162597656, 334.68719482421875, -37.331016540527344, 278.8616943359375, 266.3592529296875, 142.8537139892578, 302.1434326171875, 310.99658203125, 84.41334533691406, 341.1654052734375, -61.688209533691406, 282.252685546875, 16.22641372680664, 451.1748962402344, 309.1605224609375, 276.278564453125, -200.93783569335938, 518.8818969726562, 501.822021484375, 330.34722900390625, 39.07207489013672, 441.76416015625, 240.74221801757812, -82.04548645019531, 271.9579772949219, 318.68701171875, 362.9321594238281, -162.5413818359375, -4.67578125, 133.92788696289062, 270.95574951171875, 115.79670715332031, 249.84202575683594, 195.09776306152344, 175.935302734375, 179.0894775390625, 29.162429809570312, 91.58810424804688, 28.027711868286133, 395.0585632324219, 126.99774169921875, -72.50376892089844, 59.67311477661133, 291.4879150390625, 369.217041015625, 98.14382934570312, 475.79998779296875, 61.46797561645508, 632.880615234375, -125.01017761230469, 429.9246520996094, 7.524045944213867, 118.60359191894531, 468.460693359375, 72.52500915527344, 358.2164611816406, 247.45187377929688, 370.47900390625, 89.43488311767578, -190.70114135742188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000327.npy"}
|
|
{"epoch": 0.4801762114537445, "step": 328, "batch_size": 64, "mean": 171.35028076171875, "std": 191.02642822265625, "min": -261.0424499511719, "p10": -20.726442146301267, "median": 137.06407165527344, "p90": 435.7946228027344, "max": 669.7787475585938, "pos_frac": 0.8125, "sample": [159.2313232421875, -75.60995483398438, 101.5552749633789, 130.33334350585938, 72.34762573242188, 280.64154052734375, 248.13551330566406, 73.16539001464844, 34.277740478515625, -261.0424499511719, 60.940467834472656, 304.66619873046875, 440.16302490234375, 346.0475158691406, 154.34156799316406, 387.73919677734375, 32.266578674316406, -149.8996124267578, -16.176088333129883, 168.57135009765625, -33.298702239990234, -16.128387451171875, 291.7486572265625, 669.7787475585938, 199.92510986328125, 307.3914489746094, 310.05902099609375, 127.45498657226562, 519.8493041992188, 344.29144287109375, -12.447816848754883, 148.64584350585938, 248.6594696044922, -22.676593780517578, 58.67091369628906, 108.14425659179688, 267.6468505859375, 264.0536193847656, 180.42337036132812, 129.24176025390625, -4.676935195922852, 143.7947998046875, -76.36225891113281, 192.9072265625, -13.161186218261719, 363.1863708496094, 105.32499694824219, 202.32553100585938, 241.17333984375, 111.9575424194336, 481.2649841308594, 507.8161926269531, 71.63883972167969, 25.592864990234375, 3.8856239318847656, 612.7462768554688, 595.9390258789062, 52.232696533203125, 425.6016845703125, 10.120929718017578, 41.49152374267578, 296.0981750488281, 97.79227447509766, -105.40117645263672], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000328.npy"}
|
|
{"epoch": 0.48164464023494863, "step": 329, "batch_size": 64, "mean": 187.00729370117188, "std": 201.5757598876953, "min": -382.6728210449219, "p10": -24.44702930450439, "median": 186.5170669555664, "p90": 438.19190368652346, "max": 761.8298950195312, "pos_frac": 0.859375, "sample": [117.80429077148438, 347.0924072265625, 338.6156005859375, 285.64019775390625, 256.6304931640625, 1.247354507446289, 384.2339172363281, 15.548377990722656, 364.71246337890625, 214.10411071777344, 88.152587890625, 289.2867736816406, 78.4489517211914, -0.9771881103515625, 84.29652404785156, 163.21286010742188, 178.02745056152344, 218.05328369140625, 23.750078201293945, 588.245361328125, 88.38158416748047, 74.82855224609375, 224.20603942871094, 134.52008056640625, -382.6728210449219, 195.00668334960938, 38.55754852294922, 761.8298950195312, 459.5470275878906, -20.323165893554688, 266.3193054199219, 5.275089263916016, 312.3879699707031, 52.328277587890625, 213.373779296875, 325.0616455078125, 440.43182373046875, 308.05108642578125, 577.7732543945312, 197.2266845703125, 99.17878723144531, -60.88948440551758, -138.90638732910156, 380.515869140625, 432.9654235839844, -79.11492919921875, 329.27606201171875, -71.53596496582031, 232.4619140625, 143.94081115722656, 163.62710571289062, -26.214399337768555, 243.4842987060547, 242.64389038085938, 71.9540786743164, 50.60533905029297, 195.28831481933594, 347.13775634765625, 557.0088500976562, 40.3212890625, -230.93984985351562, 170.25344848632812, 493.14105224609375, 74.02672576904297], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000329.npy"}
|
|
{"epoch": 0.4831130690161527, "step": 330, "batch_size": 64, "mean": 183.4234619140625, "std": 185.03890991210938, "min": -194.76498413085938, "p10": -32.0986213684082, "median": 178.0016326904297, "p90": 408.3428009033204, "max": 648.4697265625, "pos_frac": 0.84375, "sample": [420.68206787109375, -27.6575927734375, -148.93701171875, 6.874874114990234, 250.32070922851562, 414.28948974609375, 113.60185241699219, 147.05465698242188, -32.305931091308594, 49.662071228027344, -91.98966979980469, 132.18504333496094, 111.26553344726562, 353.6361083984375, 373.8466796875, 69.92713165283203, 206.02120971679688, 239.26248168945312, 225.6219482421875, 370.7120666503906, 43.01145935058594, 117.68414306640625, 239.3221435546875, 75.79014587402344, 168.981689453125, 248.5915069580078, -144.50296020507812, 257.1746826171875, 211.22044372558594, 220.0615234375, 253.7200927734375, 11.944927215576172, -194.76498413085938, 139.75830078125, 394.4671936035156, 152.65451049804688, 87.35835266113281, 556.7066650390625, 30.88733673095703, 171.7930908203125, -12.451240539550781, 375.22747802734375, 119.56202697753906, 648.4697265625, -138.12071228027344, 634.2015380859375, 464.94921875, 288.1510009765625, 290.8494873046875, 101.74100494384766, 283.564453125, 282.51165771484375, 117.45111083984375, 522.2612915039062, 68.2934799194336, 253.5215301513672, 184.21017456054688, 296.31707763671875, -141.59194946289062, 336.4486083984375, 124.34994506835938, 201.9140167236328, 242.95187377929688, -31.614898681640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000330.npy"}
|
|
{"epoch": 0.4845814977973568, "step": 331, "batch_size": 64, "mean": 194.73370361328125, "std": 141.28611755371094, "min": -97.89571380615234, "p10": 55.26029930114746, "median": 182.07650756835938, "p90": 374.1411285400392, "max": 581.8545532226562, "pos_frac": 0.953125, "sample": [153.78407287597656, 543.10107421875, 267.1125793457031, -4.610595703125, 216.46717834472656, 187.00738525390625, 94.0465087890625, 9.194625854492188, 186.2466583251953, 81.3221435546875, 521.3372802734375, 340.7190246582031, 581.8545532226562, 187.86358642578125, 155.4237823486328, 228.84304809570312, 552.279541015625, 109.44810485839844, 201.68161010742188, 150.96084594726562, 344.6131896972656, 386.79595947265625, 405.100341796875, 58.13356399536133, 61.32804870605469, 92.10139465332031, 146.3853302001953, 151.7471923828125, 243.112060546875, 22.289306640625, 171.48880004882812, 8.155498504638672, 288.80303955078125, 102.68943786621094, 177.90635681152344, 105.01732635498047, -97.89571380615234, 129.93341064453125, 270.25238037109375, 467.2356262207031, 320.1798400878906, 145.7003936767578, 87.91316986083984, 272.70306396484375, 62.856483459472656, 120.05563354492188, 200.43499755859375, 204.464111328125, 247.0513916015625, 119.84873962402344, -58.46893310546875, 226.113037109375, 74.50767517089844, 125.72471618652344, 234.99945068359375, 225.37440490722656, 249.85386657714844, 117.95380401611328, 89.32005310058594, 304.2964782714844, 206.07550048828125, 266.927490234375, 265.7674560546875, 54.028900146484375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000331.npy"}
|
|
{"epoch": 0.48604992657856094, "step": 332, "batch_size": 64, "mean": 140.563232421875, "std": 166.491943359375, "min": -151.91357421875, "p10": -63.76346817016601, "median": 112.85050201416016, "p90": 386.51971435546875, "max": 544.5001831054688, "pos_frac": 0.796875, "sample": [479.97662353515625, 151.64230346679688, 117.43128967285156, 305.1898193359375, 50.82029724121094, 18.595748901367188, 89.83897399902344, 398.7110595703125, 369.69195556640625, 147.27027893066406, 100.13264465332031, 47.00042724609375, -65.84878540039062, 544.5001831054688, 153.3184814453125, 91.3722152709961, 77.22891235351562, 34.00840759277344, 91.99250030517578, 2.975759506225586, 194.3885498046875, 88.03044128417969, 388.8587951660156, -25.517990112304688, 337.240478515625, 93.74699401855469, -79.20498657226562, 45.169029235839844, 420.6728515625, 225.18118286132812, 146.98419189453125, 188.96633911132812, 190.22032165527344, -58.897727966308594, 4.794151306152344, 69.97447967529297, 431.19091796875, 381.0618591308594, -68.46444702148438, 211.80047607421875, 159.95632934570312, -145.22357177734375, 198.30511474609375, 98.26145935058594, 281.2268371582031, 519.8305053710938, 97.19212341308594, 363.1958312988281, -21.594818115234375, 128.49900817871094, -122.51345825195312, 128.346923828125, 269.68975830078125, -36.46775817871094, 323.8300476074219, 14.630023956298828, -33.74501037597656, -5.046195983886719, -75.93977355957031, 115.52022552490234, 250.2466278076172, 110.18077850341797, 137.53456115722656, -151.91357421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000332.npy"}
|
|
{"epoch": 0.48751835535976507, "step": 333, "batch_size": 64, "mean": 186.94793701171875, "std": 224.4123077392578, "min": -232.14486694335938, "p10": -87.76011352539061, "median": 189.7742156982422, "p90": 467.03565368652346, "max": 647.6998291015625, "pos_frac": 0.734375, "sample": [0.8054656982421875, 215.3861083984375, 647.6998291015625, 9.097030639648438, 187.12948608398438, 80.05290222167969, 257.91741943359375, 241.40162658691406, 112.759033203125, 58.12307357788086, 602.051513671875, -112.54065704345703, 192.4189453125, -155.2862548828125, 463.1444091796875, -69.43142700195312, 614.9243774414062, -95.61526489257812, 557.529541015625, 239.24642944335938, -64.14625549316406, 255.07574462890625, -26.565628051757812, 327.9555358886719, 382.2037353515625, 355.91119384765625, -56.088951110839844, 59.07196044921875, 120.58123779296875, 348.7626037597656, 88.6756362915039, -14.089494705200195, -42.624542236328125, 465.2138977050781, 444.63336181640625, 330.1860046386719, -232.14486694335938, 6.938423156738281, 373.04962158203125, 299.3392639160156, 367.774658203125, -3.9115142822265625, -4.9398956298828125, 460.4614562988281, 388.0599365234375, -44.493873596191406, 170.49026489257812, 179.8495330810547, 487.036865234375, 264.9466552734375, 467.81640625, 345.4351501464844, 553.070068359375, 132.80126953125, 208.59762573242188, -125.94886779785156, 376.59033203125, -24.29958724975586, 255.78057861328125, -169.2223663330078, 345.99774169921875, -205.96600341796875, 12.195388793945312, 57.79359436035156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000333.npy"}
|
|
{"epoch": 0.4889867841409692, "step": 334, "batch_size": 64, "mean": 165.01031494140625, "std": 211.95608520507812, "min": -339.5321044921875, "p10": -38.191849517822256, "median": 161.2049331665039, "p90": 492.59862976074226, "max": 564.9453735351562, "pos_frac": 0.828125, "sample": [478.52008056640625, 101.20571899414062, -158.41009521484375, 424.8133544921875, 208.46792602539062, 30.538742065429688, -24.61595916748047, 97.96408081054688, 207.82284545898438, 23.192550659179688, 198.25613403320312, 47.577049255371094, 306.49591064453125, 190.18698120117188, -41.80693054199219, 173.8418426513672, 13.15965461730957, 353.88641357421875, -339.5321044921875, 498.6322937011719, 58.029850006103516, 261.9256286621094, 30.085006713867188, -264.3405456542969, 310.5623779296875, 185.29312133789062, 447.6075134277344, -185.82569885253906, -138.6995391845703, 1.4648895263671875, 23.17877197265625, -10.899639129638672, 80.82000732421875, 74.00308227539062, 519.8807983398438, 562.1375732421875, 290.6482238769531, 5.804208755493164, 142.95310974121094, 229.11578369140625, 114.7426986694336, 251.58905029296875, 444.7596435546875, 172.7461395263672, 22.135482788085938, 54.89169692993164, 506.4468994140625, 226.49172973632812, 513.715576171875, 352.53289794921875, -29.75666046142578, -247.66688537597656, -15.24444580078125, 257.47100830078125, 165.26576232910156, 534.7398681640625, 382.63116455078125, 127.70428466796875, 48.32353210449219, 336.4451904296875, 34.949851989746094, 564.9453735351562, 157.14410400390625, 169.7154541015625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000334.npy"}
|
|
{"epoch": 0.49045521292217326, "step": 335, "batch_size": 64, "mean": 127.66055297851562, "std": 197.8296661376953, "min": -235.70266723632812, "p10": -125.34148101806639, "median": 122.12545013427734, "p90": 374.33812866210945, "max": 732.6494750976562, "pos_frac": 0.734375, "sample": [218.147705078125, 293.0875549316406, 344.1998291015625, 24.8201904296875, 145.74708557128906, -85.84168243408203, 79.93683624267578, 339.46319580078125, 64.32145690917969, 28.97818946838379, -183.5941162109375, 128.0215606689453, 223.79193115234375, 210.86329650878906, 50.106971740722656, 411.07537841796875, 9.546806335449219, 175.06600952148438, -214.0199737548828, 196.9739990234375, -60.735687255859375, 97.39460754394531, 146.56246948242188, 94.94436645507812, 417.8695373535156, 406.30084228515625, 351.404541015625, 196.6388397216797, 9.535371780395508, 496.164306640625, 732.6494750976562, -199.30450439453125, 178.57839965820312, -235.70266723632812, 339.529052734375, 37.45098114013672, 384.16680908203125, 41.37870788574219, 66.2498779296875, 278.57177734375, -102.8388671875, 288.5567932128906, 47.813514709472656, -13.757568359375, 291.0027160644531, -43.55834197998047, 172.50657653808594, -1.469207763671875, -182.14047241210938, -52.2470703125, 116.22933959960938, -129.78013610839844, 348.685302734375, -114.984619140625, -52.20177459716797, 318.5296630859375, 166.083740234375, 407.5738525390625, -27.365768432617188, -171.22772216796875, 195.01800537109375, 322.4739685058594, 145.24624633789062, 1.7880287170410156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000335.npy"}
|
|
{"epoch": 0.4919236417033774, "step": 336, "batch_size": 64, "mean": 200.35910034179688, "std": 236.41217041015625, "min": -373.47125244140625, "p10": -59.018446731567366, "median": 178.89165496826172, "p90": 542.4290405273438, "max": 815.3667602539062, "pos_frac": 0.859375, "sample": [173.0965576171875, 46.48749542236328, 655.7020263671875, 30.0521183013916, 371.7430114746094, 152.6318359375, 211.83944702148438, -94.15145111083984, -28.100154876708984, 728.375, 158.99717712402344, 272.73760986328125, 154.80227661132812, 346.5559387207031, 786.5819091796875, 39.2615966796875, 576.5360107421875, 304.0751953125, 282.8397216796875, 267.42694091796875, -85.09471130371094, 90.4808349609375, 163.50331115722656, 275.4603576660156, 51.302642822265625, 220.0687255859375, 93.82455444335938, 112.66226196289062, 525.1976928710938, 252.03005981445312, 184.68675231933594, 51.549957275390625, 105.64891052246094, 189.53286743164062, 129.9083709716797, 157.76223754882812, -189.5146942138672, 549.8139038085938, -45.170169830322266, -64.95342254638672, 330.78759765625, 101.34400177001953, 618.5831298828125, 68.64315795898438, 382.95751953125, 58.564056396484375, -373.47125244140625, 34.22504425048828, 210.63836669921875, 291.2579345703125, 108.33009338378906, 131.16995239257812, 190.15809631347656, 69.9026870727539, 201.1891632080078, 436.982666015625, 318.3372497558594, 815.3667602539062, -296.97381591796875, 241.5105438232422, -254.3023681640625, 298.4309387207031, 293.86785888671875, 339.2926025390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000336.npy"}
|
|
{"epoch": 0.4933920704845815, "step": 337, "batch_size": 64, "mean": 203.09140014648438, "std": 210.69859313964844, "min": -234.06739807128906, "p10": -39.840779495239246, "median": 190.35309600830078, "p90": 452.01453857421876, "max": 771.098388671875, "pos_frac": 0.859375, "sample": [705.6519775390625, 279.8490295410156, -234.06739807128906, 63.152000427246094, 418.84857177734375, 394.8671875, 29.597434997558594, 258.9305725097656, 68.74159240722656, 397.3145751953125, 11.862556457519531, 240.71234130859375, 223.08868408203125, 344.6055908203125, 443.35955810546875, -107.2009048461914, 254.3765106201172, 231.07772827148438, -25.132614135742188, 386.0460205078125, 119.86323547363281, 188.57077026367188, 70.4335708618164, 455.72381591796875, 139.37962341308594, 41.0340576171875, 771.098388671875, 40.7379150390625, -79.81873321533203, -144.41644287109375, 180.33596801757812, 192.1354217529297, 259.7477722167969, -211.94920349121094, 185.31455993652344, 111.77220153808594, 357.0614013671875, 15.433273315429688, 234.37380981445312, 305.3577880859375, 60.07908248901367, 584.1690063476562, 106.21710205078125, 177.927001953125, -212.0832061767578, 117.49578857421875, 584.3234252929688, 384.64044189453125, 338.7669372558594, 40.022621154785156, 187.2548828125, 92.79467010498047, 480.37750244140625, 333.054443359375, 99.32890319824219, 528.8160400390625, 320.0067138671875, 204.57821655273438, -43.44428253173828, 284.2008056640625, 294.5734558105469, 106.83953857421875, -31.432605743408203, 341.471923828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000337.npy"}
|
|
{"epoch": 0.4948604992657856, "step": 338, "batch_size": 64, "mean": 199.4775848388672, "std": 184.94252014160156, "min": -195.63482666015625, "p10": -6.723152923583982, "median": 189.01654052734375, "p90": 441.9676361083985, "max": 676.3937377929688, "pos_frac": 0.875, "sample": [143.2681427001953, 453.7519226074219, -7.6968231201171875, -195.63482666015625, 193.52871704101562, 89.79287719726562, -113.70120239257812, 215.28440856933594, 546.5109252929688, 248.42819213867188, 97.91165161132812, 79.57051086425781, -35.259918212890625, 408.5190734863281, 92.3758544921875, 289.686279296875, 67.48989868164062, 331.1292419433594, 343.7117919921875, 405.59521484375, 576.66650390625, 37.484764099121094, 230.3843994140625, 218.76527404785156, 236.3179473876953, 47.96662902832031, 354.0815124511719, -4.451255798339844, -9.172798156738281, 42.384361267089844, 236.83111572265625, 445.7995300292969, 407.2860412597656, 76.47064208984375, 184.50436401367188, 323.1412048339844, 109.18505096435547, 9.50204086303711, 310.66162109375, 148.82423400878906, 378.5919189453125, 69.42694091796875, 105.55026245117188, 86.17207336425781, 421.258056640625, 55.04762268066406, 210.79299926757812, 103.79769134521484, 297.03045654296875, 117.73237609863281, -37.09405517578125, 13.447345733642578, 464.196044921875, 676.3937377929688, 22.17450714111328, 242.571044921875, 433.02655029296875, 209.6868896484375, 327.306640625, 153.91961669921875, 493.09796142578125, -154.27584838867188, 58.56733703613281, 381.25213623046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000338.npy"}
|
|
{"epoch": 0.49632892804698975, "step": 339, "batch_size": 64, "mean": 189.7136993408203, "std": 188.45521545410156, "min": -132.3587188720703, "p10": -52.93168411254883, "median": 170.65679168701172, "p90": 466.96052551269537, "max": 646.9182739257812, "pos_frac": 0.84375, "sample": [104.60284423828125, -64.79639434814453, -55.09830093383789, -52.833892822265625, 496.0478515625, 6.5416717529296875, 120.08143615722656, 139.77757263183594, 281.2721252441406, 18.08490753173828, 646.9182739257812, 71.60914611816406, 80.02008056640625, 104.39779663085938, 286.6939697265625, 290.54315185546875, 9.316682815551758, 68.47539520263672, 55.609779357910156, 538.25634765625, 207.8627166748047, -4.264961242675781, 215.50335693359375, -47.39752960205078, 188.30325317382812, 146.7607879638672, 531.5, 29.49591064453125, 312.95965576171875, 595.40380859375, -132.3587188720703, 221.89584350585938, -115.82598114013672, 95.2891845703125, 472.15802001953125, 239.5521697998047, 159.2435302734375, 13.610570907592773, 193.73446655273438, 376.8472595214844, 116.31201171875, 181.0948486328125, 331.78961181640625, 370.4228820800781, 92.31290435791016, 442.7264709472656, -87.81385040283203, 534.0384521484375, 231.51144409179688, -127.77131652832031, 240.96685791015625, 238.16213989257812, 127.26258850097656, 213.967529296875, 80.21173095703125, 160.21873474121094, 153.90847778320312, -52.973594665527344, 307.718505859375, 260.40667724609375, 426.2630615234375, 454.8330383300781, 343.93585205078125, 256.3779602050781], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000339.npy"}
|
|
{"epoch": 0.4977973568281938, "step": 340, "batch_size": 64, "mean": 169.98484802246094, "std": 203.7632598876953, "min": -277.9409484863281, "p10": -31.898941040039062, "median": 145.39053344726562, "p90": 439.1630004882813, "max": 747.1547241210938, "pos_frac": 0.78125, "sample": [191.7923583984375, 263.8185729980469, 114.98113250732422, 181.3978271484375, 136.5886688232422, 388.44476318359375, 491.5372009277344, -264.06475830078125, 206.809814453125, -17.081323623657227, 246.5830535888672, 747.1547241210938, 167.33950805664062, -162.2198486328125, 463.77130126953125, -2.8062667846679688, 346.2702331542969, 123.32734680175781, -31.59778594970703, 77.71542358398438, 166.06820678710938, -28.7364501953125, -277.9409484863281, 429.5173645019531, -110.62727355957031, 144.0478515625, 19.185781478881836, 363.3014221191406, -0.991912841796875, 251.67654418945312, 397.20965576171875, 86.87544250488281, 668.5941772460938, 468.29486083984375, 10.643735885620117, 127.47824096679688, -6.477962493896484, 241.62579345703125, 67.28813171386719, 187.408447265625, 230.8927001953125, 116.833740234375, 178.74630737304688, 4.505031585693359, -78.19529724121094, 266.19903564453125, 146.73321533203125, -32.02800750732422, 457.75347900390625, 13.985641479492188, 128.47537231445312, 390.3174133300781, 391.9617614746094, 443.2968444824219, 259.91766357421875, 212.105224609375, 109.5355224609375, 108.9090576171875, 330.2754211425781, 136.07098388671875, -181.85914611816406, 323.2392578125, -24.793960571289062, 71.95038604736328], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000340.npy"}
|
|
{"epoch": 0.49926578560939794, "step": 341, "batch_size": 64, "mean": 193.81198120117188, "std": 203.6765899658203, "min": -231.6540985107422, "p10": -47.17550582885741, "median": 191.84841918945312, "p90": 421.92333984375, "max": 906.7955322265625, "pos_frac": 0.796875, "sample": [906.7955322265625, -51.585205078125, -27.599594116210938, -23.964235305786133, 274.097900390625, 420.17803955078125, 518.712646484375, 433.0440979003906, 204.38143920898438, 153.45501708984375, 25.093992233276367, 195.90908813476562, 165.97410583496094, 234.088623046875, 54.80863952636719, -64.00495910644531, -83.61787414550781, 310.614990234375, 60.30077362060547, 239.36605834960938, -141.24951171875, 313.9710693359375, 394.9521789550781, 247.44158935546875, 110.92955780029297, 477.8367919921875, 164.908935546875, 222.45352172851562, -2.7153778076171875, -110.59471130371094, 110.40406799316406, 301.82421875, 511.5251159667969, 251.79171752929688, 242.95396423339844, 344.0139465332031, 135.09312438964844, 11.535867691040039, -33.122581481933594, 279.39105224609375, 384.51416015625, 246.7831268310547, 99.08465576171875, 94.66678619384766, -231.6540985107422, 404.68792724609375, -10.130050659179688, 187.78775024414062, -185.36785888671875, 306.8540954589844, 153.18576049804688, 267.8760986328125, 292.61273193359375, -36.886207580566406, 596.9962158203125, 74.94834899902344, 349.00286865234375, 375.99652099609375, 381.19482421875, 78.83216857910156, 179.72933959960938, 114.70380401611328, 422.67132568359375, 76.48287200927734], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000341.npy"}
|
|
{"epoch": 0.5007342143906021, "step": 342, "batch_size": 64, "mean": 171.60679626464844, "std": 210.73953247070312, "min": -350.827392578125, "p10": -65.23833770751953, "median": 137.06170654296875, "p90": 407.9123657226564, "max": 840.160400390625, "pos_frac": 0.78125, "sample": [561.6155395507812, -350.827392578125, -156.93780517578125, 444.9706726074219, 29.42282485961914, 68.17030334472656, 185.32135009765625, 277.66796875, 228.7808380126953, 285.9249267578125, 66.17558288574219, 123.59156799316406, 196.04086303710938, 195.63133239746094, 375.7224426269531, 264.6593017578125, 86.71604919433594, 375.8199462890625, -89.73685455322266, -22.295425415039062, 286.3326721191406, -64.6937484741211, 78.99850463867188, -65.47173309326172, -111.76648712158203, 207.56805419921875, 840.160400390625, 322.5868835449219, 111.83793640136719, -119.18101501464844, 9.579669952392578, 787.3087158203125, 166.43228149414062, 115.58934020996094, 43.73658752441406, -55.541893005371094, 421.666259765625, 364.7261657714844, 459.814453125, 165.31256103515625, -29.25893211364746, 132.39894104003906, 339.3161315917969, -12.322418212890625, 333.9021911621094, 347.61639404296875, 116.86297607421875, 337.1365966796875, 152.87303161621094, 355.56854248046875, 422.05218505859375, 102.32221221923828, -103.46997833251953, 68.18771362304688, 296.77349853515625, -35.987815856933594, 127.30657958984375, -6.008428573608398, 124.44864654541016, 91.71360778808594, 184.99786376953125, 335.0648498535156, 141.72447204589844, 48.186546325683594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000342.npy"}
|
|
{"epoch": 0.5022026431718062, "step": 343, "batch_size": 64, "mean": 206.11268615722656, "std": 166.5856475830078, "min": -189.43206787109375, "p10": 5.753721427917482, "median": 187.26013946533203, "p90": 374.9249298095703, "max": 648.1650390625, "pos_frac": 0.90625, "sample": [353.9530944824219, 174.28607177734375, 178.89930725097656, 129.7392578125, 252.85247802734375, 5.0466156005859375, 188.03016662597656, 285.229248046875, 152.88229370117188, 574.5546264648438, 91.65913391113281, -38.910888671875, 197.84689331054688, 20.586212158203125, 173.34832763671875, 57.949459075927734, 79.97708892822266, 213.46694946289062, 229.80679321289062, 539.5703125, 189.1942901611328, 308.7065124511719, 77.90486145019531, 330.37017822265625, 311.3472900390625, 285.8817138671875, 75.37619018554688, 93.60678100585938, 582.4993286132812, 173.22157287597656, -57.31841278076172, 80.50563049316406, 350.3970642089844, 87.63421630859375, 175.66763305664062, 247.21441650390625, 180.781005859375, 75.30757141113281, 7.403635025024414, 372.1159362792969, -189.43206787109375, 269.8825378417969, 186.4901123046875, 304.2870788574219, 315.3189392089844, 648.1650390625, 366.148681640625, 278.8766174316406, 223.52447509765625, 302.4104919433594, 192.9296875, 289.4355163574219, 108.85946655273438, -26.842952728271484, 376.1287841796875, 336.7622985839844, 157.62059020996094, 157.4932098388672, 134.20620727539062, 134.3896484375, -25.15711212158203, 442.6108093261719, 516.012451171875, -117.49939727783203], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000343.npy"}
|
|
{"epoch": 0.5036710719530103, "step": 344, "batch_size": 64, "mean": 183.69131469726562, "std": 202.1725311279297, "min": -177.24636840820312, "p10": -2.668616485595683, "median": 146.28269958496094, "p90": 430.7065856933594, "max": 1110.4622802734375, "pos_frac": 0.890625, "sample": [50.55939483642578, 50.637611389160156, -147.6492156982422, -11.88534927368164, 80.80430603027344, 142.83563232421875, 179.1807098388672, 112.53153228759766, 43.88310241699219, 166.9176788330078, 31.670839309692383, 573.0779418945312, 69.85582733154297, 205.01866149902344, 106.03038024902344, 173.17526245117188, 179.02845764160156, -50.74888610839844, -93.67809295654297, 235.74920654296875, 204.4089813232422, 45.45210266113281, 119.62061309814453, 27.744922637939453, 149.72976684570312, 450.8917541503906, 431.49908447265625, 399.0355529785156, 90.79733276367188, 22.033477783203125, 398.2692565917969, 128.60409545898438, 228.16357421875, 39.402496337890625, 176.99960327148438, 62.966796875, 297.25164794921875, -11.5899658203125, 260.13946533203125, 43.09465026855469, 456.7684020996094, 279.522216796875, 226.23216247558594, 91.94109344482422, 69.21379089355469, 1110.4622802734375, 366.84942626953125, 275.9696350097656, 90.24150085449219, 460.00555419921875, 18.147865295410156, 328.0130920410156, 196.24295043945312, 428.857421875, 91.29829406738281, 28.939865112304688, -127.30858612060547, 403.30145263671875, 457.8371276855469, -177.24636840820312, 285.59344482421875, 326.9079895019531, 93.92488098144531, 313.0184020996094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000344.npy"}
|
|
{"epoch": 0.5051395007342144, "step": 345, "batch_size": 64, "mean": 193.00807189941406, "std": 209.54295349121094, "min": -280.698974609375, "p10": -39.08269348144531, "median": 165.22408294677734, "p90": 460.2702087402346, "max": 749.1371459960938, "pos_frac": 0.828125, "sample": [39.94158172607422, 480.8235778808594, 170.12246704101562, 286.82598876953125, 67.19942474365234, -141.427734375, 568.9949340820312, -108.56977844238281, 147.55081176757812, 374.78155517578125, -6.968435287475586, 208.4728546142578, 124.33978271484375, 344.7697448730469, 412.3123474121094, 99.88650512695312, -277.8335876464844, 194.28411865234375, 74.83216857910156, 123.84761047363281, 169.31472778320312, 349.0059814453125, 69.65916442871094, 264.3987731933594, 232.70751953125, 341.662353515625, 107.0003890991211, 333.3641662597656, 80.68243408203125, 161.13343811035156, 59.560028076171875, 643.4605712890625, 306.0732116699219, 71.26744079589844, 59.792236328125, 327.1084289550781, 373.64111328125, 371.5565185546875, 249.47744750976562, 294.970703125, 278.76031494140625, 749.1371459960938, 120.57879638671875, 94.46328735351562, 364.32196044921875, 263.4766540527344, -1.2292900085449219, 673.6173706054688, 108.77227783203125, 318.2357177734375, -37.683067321777344, -39.682533264160156, 158.69781494140625, 541.7914428710938, -10.269079208374023, -118.5182876586914, -55.99767303466797, 98.94561767578125, 519.490478515625, 155.19918823242188, 184.564208984375, 205.30294799804688, -280.698974609375, 11.217575073242188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000345.npy"}
|
|
{"epoch": 0.5066079295154186, "step": 346, "batch_size": 64, "mean": 185.19598388671875, "std": 203.9821319580078, "min": -441.831298828125, "p10": -63.013983917236324, "median": 212.43323516845703, "p90": 454.7567138671876, "max": 657.8998413085938, "pos_frac": 0.796875, "sample": [234.03778076171875, 127.94221496582031, 8.552665710449219, 288.92626953125, 429.1550598144531, 365.201416015625, 558.5394287109375, 196.7725067138672, 217.8917236328125, 657.8998413085938, -441.831298828125, 101.04209899902344, 293.7618408203125, 21.16318130493164, 261.0600891113281, 465.7288513183594, -4.349395751953125, 280.98162841796875, 577.123046875, 3.130525588989258, 278.8660888671875, 140.74237060546875, -78.19305419921875, 428.990234375, 239.0247802734375, 256.6788330078125, 255.80148315429688, 140.367919921875, 304.7962646484375, 81.69161987304688, -24.539674758911133, -64.27312469482422, 178.5342559814453, 88.24031066894531, 528.966064453125, 206.97474670410156, 244.51011657714844, 152.99212646484375, 280.9453430175781, -103.07984924316406, -20.62683868408203, 503.7823486328125, 258.04345703125, 467.11309814453125, 133.20701599121094, 276.273681640625, 292.3656005859375, 296.37835693359375, 28.496868133544922, -23.694576263427734, 234.08135986328125, -39.82631301879883, 34.333404541015625, -132.20994567871094, 125.03645324707031, 371.07659912109375, 108.24429321289062, 142.1185760498047, 419.2039489746094, -146.89871215820312, -60.07598876953125, -131.3538360595703, 269.6934814453125, 267.0130615234375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000346.npy"}
|
|
{"epoch": 0.5080763582966226, "step": 347, "batch_size": 64, "mean": 188.865478515625, "std": 220.07676696777344, "min": -300.8305358886719, "p10": -70.51843872070312, "median": 152.2061004638672, "p90": 460.43767700195315, "max": 687.4068603515625, "pos_frac": 0.8125, "sample": [-300.8305358886719, 283.50604248046875, 375.20013427734375, 321.0683898925781, 79.51876068115234, 16.55492401123047, 147.91116333007812, 247.0957794189453, 525.7728271484375, 97.80272674560547, 350.1737060546875, 29.743431091308594, 102.30116271972656, -14.1219482421875, 352.4630126953125, -211.28306579589844, -265.5421447753906, -71.73916625976562, 681.6702880859375, 91.38286590576172, 391.9576110839844, 67.3167724609375, 153.36798095703125, 312.86248779296875, -67.67007446289062, 462.23150634765625, 445.09991455078125, 201.04638671875, -39.130043029785156, 456.2520751953125, 325.6373291015625, 387.71539306640625, 322.7513427734375, -8.574562072753906, 230.64016723632812, 40.44391632080078, 465.86004638671875, 157.1981201171875, 84.221435546875, -78.85916137695312, 149.40585327148438, 244.53985595703125, 79.28759002685547, 264.2886047363281, 133.98880004882812, 151.04421997070312, 82.65670776367188, 46.13207244873047, 687.4068603515625, 68.91259765625, 294.785400390625, 471.75799560546875, 236.34446716308594, 671.4034423828125, -126.41028594970703, 146.24453735351562, 232.89956665039062, -195.1427459716797, 381.00970458984375, 67.67191314697266, 400.7625427246094, -66.60240173339844, 107.69926452636719, 408.28704833984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000347.npy"}
|
|
{"epoch": 0.5095447870778267, "step": 348, "batch_size": 64, "mean": 165.63992309570312, "std": 197.6311798095703, "min": -233.0101776123047, "p10": -53.00797042846679, "median": 159.75501251220703, "p90": 379.04403991699223, "max": 783.9830322265625, "pos_frac": 0.78125, "sample": [288.1201477050781, 31.002098083496094, 158.15872192382812, 334.95660400390625, 352.1225891113281, -4.029212951660156, 49.72113800048828, 331.34954833984375, 303.8575439453125, -225.96617126464844, 134.4008331298828, 60.607421875, 463.2819519042969, 185.91354370117188, 161.35130310058594, 239.16253662109375, 18.851083755493164, 366.9555358886719, 148.39462280273438, 71.29173278808594, 233.11672973632812, 221.11453247070312, 9.943733215332031, 783.9830322265625, 364.5905456542969, 307.7369384765625, -35.43358612060547, 245.98959350585938, 40.487213134765625, 118.0370101928711, -1.4228057861328125, -81.59178161621094, -46.54949951171875, 650.8690795898438, 382.9781799316406, 225.45379638671875, 279.3697814941406, 152.51882934570312, 224.39002990722656, 13.34919548034668, -233.0101776123047, 369.8643798828125, -55.77588653564453, 308.7323913574219, 231.40353393554688, 205.59735107421875, -32.39458465576172, 9.680784225463867, 478.12994384765625, 221.63038635253906, 128.50186157226562, -25.981674194335938, 420.35906982421875, -95.0411148071289, 302.246337890625, 166.5610809326172, 162.65484619140625, -1.4606590270996094, 22.234588623046875, 540.755859375, 70.66543579101562, 50.30670166015625, -96.42933654785156, -106.70980834960938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000348.npy"}
|
|
{"epoch": 0.5110132158590308, "step": 349, "batch_size": 64, "mean": 172.01544189453125, "std": 232.6883544921875, "min": -371.257080078125, "p10": -67.9125877380371, "median": 114.58097839355469, "p90": 458.5131927490236, "max": 856.6049194335938, "pos_frac": 0.8125, "sample": [20.341270446777344, -326.40948486328125, 352.9374694824219, 119.9992446899414, 91.09909057617188, 94.35791015625, 323.3099365234375, -59.45738220214844, 227.3494873046875, 307.28533935546875, 116.45458984375, 390.8138427734375, 149.07923889160156, 43.723854064941406, 34.850006103515625, 28.185667037963867, 9.031274795532227, 45.08312225341797, 228.43344116210938, -111.88114166259766, 348.80126953125, 288.4622802734375, 260.90234375, 7.977170944213867, 320.3614501953125, -11.183338165283203, 313.9530029296875, -71.53624725341797, 71.25029754638672, 108.67022705078125, 716.1025390625, -2.5028762817382812, -24.935649871826172, 414.9216613769531, 234.5325469970703, -371.257080078125, -145.77267456054688, 98.3588638305664, 505.4652099609375, 51.541202545166016, 550.8289794921875, 17.257476806640625, -171.60519409179688, 51.69401550292969, 470.2474060058594, 559.6447143554688, 665.005859375, 147.3478240966797, 431.13336181640625, 112.70736694335938, 322.861328125, 403.3134765625, -16.28209686279297, 247.68240356445312, 216.2333221435547, 161.28619384765625, -137.58294677734375, 55.40696716308594, 78.21409606933594, 356.3486022949219, 292.8643493652344, 80.60989379882812, 58.46746826171875, 856.6049194335938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000349.npy"}
|
|
{"epoch": 0.5124816446402349, "step": 350, "batch_size": 64, "mean": 183.8451690673828, "std": 223.78988647460938, "min": -501.65802001953125, "p10": -109.40437316894531, "median": 190.64501953125, "p90": 461.2583801269531, "max": 685.6036987304688, "pos_frac": 0.828125, "sample": [623.990234375, -309.8296203613281, 95.37470245361328, 301.7021789550781, 287.4000549316406, 169.91407775878906, 411.3861083984375, 52.71600341796875, -9.318252563476562, 85.40916442871094, 102.20166015625, 373.2599792480469, -266.3289794921875, 151.8546905517578, 218.28079223632812, 18.885284423828125, 605.1126098632812, 346.13262939453125, 685.6036987304688, 268.90631103515625, 195.31875610351562, 243.07948303222656, -58.04497528076172, 389.4943542480469, 135.46240234375, 553.8597412109375, 83.9281005859375, -115.7486572265625, 208.40350341796875, 128.493896484375, 460.30352783203125, 91.08529663085938, 191.02569580078125, 212.54049682617188, 262.4703063964844, 240.15769958496094, 132.94271850585938, 234.12635803222656, -183.903076171875, 461.6676025390625, 212.86671447753906, 465.43963623046875, 259.87310791015625, -111.56414794921875, 94.77509307861328, 173.03250122070312, 529.4010620117188, 304.8727111816406, 149.74899291992188, 241.40023803710938, -501.65802001953125, 416.095458984375, 130.30345153808594, 245.38095092773438, 86.85202026367188, 163.98849487304688, -102.19139099121094, -104.36489868164062, 48.17418670654297, 190.26434326171875, 370.40081787109375, 135.44308471679688, -118.70793151855469, 406.94775390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000350.npy"}
|
|
{"epoch": 0.5139500734214391, "step": 351, "batch_size": 64, "mean": 204.92010498046875, "std": 229.70767211914062, "min": -300.3514099121094, "p10": -46.423078918457016, "median": 194.4784927368164, "p90": 528.0199829101564, "max": 858.5684204101562, "pos_frac": 0.828125, "sample": [84.453857421875, 241.62457275390625, 384.6097412109375, 716.8243408203125, 858.5684204101562, 165.039794921875, 334.3447570800781, 97.82504272460938, 304.8692626953125, 257.5172119140625, 103.0670166015625, 201.92623901367188, 552.9119873046875, 660.1539306640625, 244.96839904785156, 315.32623291015625, 304.83740234375, 435.44696044921875, 539.7506713867188, 4.441339492797852, -300.3514099121094, 285.3092956542969, 114.63992309570312, 199.1656494140625, 154.54495239257812, 296.7790832519531, 379.9095764160156, 234.53965759277344, -208.4333038330078, -29.647720336914062, -106.08163452148438, 187.10333251953125, -145.6317901611328, 127.84270477294922, 1.05902099609375, 598.85205078125, 24.93310546875, 118.04733276367188, 171.42327880859375, 33.856285095214844, -53.612518310546875, 326.0526123046875, -22.821147918701172, 604.593505859375, 11.53619384765625, 25.93575096130371, 189.7913360595703, -0.940673828125, 70.31854248046875, -211.1973114013672, -15.346414566040039, 428.419921875, 154.06820678710938, 297.17767333984375, 500.64837646484375, -68.11994934082031, 241.46926879882812, 494.7337951660156, 262.54193115234375, 209.37118530273438, 170.2091522216797, 309.2664794921875, 209.8431854248047, 34.580543518066406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000351.npy"}
|
|
{"epoch": 0.5154185022026432, "step": 352, "batch_size": 64, "mean": 135.69808959960938, "std": 211.0836181640625, "min": -315.68206787109375, "p10": -111.67146835327148, "median": 109.75311279296875, "p90": 412.2062011718751, "max": 737.8418579101562, "pos_frac": 0.765625, "sample": [305.2453918457031, 345.0202331542969, 134.077392578125, 418.61199951171875, -229.2117919921875, 58.29509735107422, 114.07718658447266, 296.38665771484375, 33.88267517089844, 217.0196075439453, 495.93017578125, 397.25933837890625, -65.57356262207031, 87.30638885498047, 200.4209442138672, -55.084503173828125, -112.19403076171875, 737.8418579101562, 225.59654235839844, -110.45215606689453, 177.81304931640625, 625.1839599609375, 122.32865905761719, 357.5550537109375, 33.4661865234375, 8.721641540527344, -76.51165771484375, -231.39801025390625, 49.77051544189453, 79.09161376953125, 74.36538696289062, 450.76995849609375, 18.277013778686523, 10.125129699707031, 428.3637390136719, 143.6783447265625, -315.68206787109375, 74.15142059326172, 200.2772216796875, 322.6647033691406, 162.77352905273438, 52.546051025390625, 194.41087341308594, -160.05239868164062, -67.01982116699219, -62.086639404296875, -161.89370727539062, 393.32586669921875, 355.03729248046875, -83.71006774902344, 173.23854064941406, -1.7275619506835938, -131.7522430419922, 378.550537109375, 202.35794067382812, 189.46475219726562, 345.87335205078125, 13.711433410644531, 442.57220458984375, 105.42903900146484, 88.56002807617188, 3.7358551025390625, 153.97952270507812, 49.88632583618164], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000352.npy"}
|
|
{"epoch": 0.5168869309838473, "step": 353, "batch_size": 64, "mean": 166.03475952148438, "std": 216.22071838378906, "min": -320.27777099609375, "p10": -82.77623748779294, "median": 112.44722747802734, "p90": 478.30876770019535, "max": 647.8434448242188, "pos_frac": 0.765625, "sample": [268.81268310546875, 291.6244201660156, 34.506797790527344, 431.08404541015625, 179.01507568359375, 305.2544250488281, 76.08692932128906, -53.74693298339844, 463.4332275390625, 192.7376708984375, 99.7359619140625, 1.2811203002929688, 158.3075408935547, 454.4731140136719, 82.4407958984375, 338.7392578125, -98.37214660644531, -54.61334228515625, 521.0576171875, 534.4174194335938, 104.72840881347656, 257.4414978027344, 0.7547721862792969, 92.1796875, -320.27777099609375, 361.4268798828125, -201.99285888671875, 19.51778221130371, -39.25696563720703, 247.10897827148438, -16.625410079956055, 25.033899307250977, 273.01092529296875, 193.6361541748047, 17.389862060546875, -102.11746215820312, 94.08505249023438, -106.24331665039062, 569.116943359375, 1.4699153900146484, 47.852783203125, 533.9631958007812, -19.72040557861328, 116.37000274658203, 108.52445220947266, 207.59217834472656, 272.20672607421875, 132.7007293701172, -33.640960693359375, 482.2613830566406, 224.73699951171875, -58.24638366699219, 469.08599853515625, 330.2695007324219, 647.8434448242188, 591.826171875, -128.14341735839844, 303.97235107421875, 425.23663330078125, -93.28903198242188, 58.91954803466797, 102.79280090332031, -28.43438720703125, 234.88116455078125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000353.npy"}
|
|
{"epoch": 0.5183553597650514, "step": 354, "batch_size": 64, "mean": 188.36825561523438, "std": 178.61997985839844, "min": -149.78604125976562, "p10": -45.67686691284177, "median": 168.29510498046875, "p90": 403.74033203125003, "max": 629.116455078125, "pos_frac": 0.84375, "sample": [272.49578857421875, 376.26239013671875, 287.17095947265625, 23.66058349609375, 376.8597106933594, 165.01605224609375, 554.4461669921875, -75.95808410644531, 198.36566162109375, -75.84190368652344, -16.013282775878906, -149.78604125976562, 189.0840301513672, 629.116455078125, 404.3614807128906, 136.201171875, 5.027778625488281, 248.73333740234375, 364.73919677734375, 246.46238708496094, 16.261146545410156, 334.22467041015625, 171.57415771484375, 129.64581298828125, -131.60055541992188, 94.28187561035156, 279.6391296386719, 217.3178253173828, 105.9465103149414, 278.2641906738281, 433.5322265625, -58.38983154296875, 311.5687561035156, 257.79620361328125, 150.30520629882812, 136.02650451660156, -88.15763092041016, 387.2250671386719, 155.12191772460938, 53.36439514160156, 353.76898193359375, 44.756996154785156, 276.1777648925781, 132.80572509765625, 81.69696807861328, 219.7167510986328, 96.55176544189453, 609.1454467773438, 325.47161865234375, -2.4378662109375, 45.801048278808594, 402.2909851074219, 204.59120178222656, 100.63885498046875, -61.36048126220703, 500.5279541015625, 32.60930633544922, 288.09552001953125, 289.878662109375, 43.750457763671875, 417.6424865722656, 152.75140380859375, 121.62918090820312, -15.28335952758789], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000354.npy"}
|
|
{"epoch": 0.5198237885462555, "step": 355, "batch_size": 64, "mean": 200.06072998046875, "std": 182.92294311523438, "min": -182.80429077148438, "p10": -29.31412696838378, "median": 187.54563903808594, "p90": 470.1395568847657, "max": 764.71044921875, "pos_frac": 0.84375, "sample": [142.17654418945312, 481.94390869140625, 369.02777099609375, 487.4095764160156, 135.64468383789062, 239.0620574951172, 217.97320556640625, 764.71044921875, -102.8185806274414, 211.63734436035156, 248.0213623046875, 522.5184936523438, 86.56583404541016, -10.0020751953125, 349.47161865234375, 458.36871337890625, -19.787673950195312, 520.2390747070312, 116.06639862060547, 152.18212890625, 314.99658203125, 191.10491943359375, -46.80500793457031, -3.0558815002441406, 263.6302795410156, 419.613525390625, 381.41796875, 183.22268676757812, 102.98573303222656, 10.090087890625, 17.609378814697266, 131.4183349609375, 146.7610321044922, -182.80429077148438, 187.24197387695312, -122.71238708496094, 308.16766357421875, 475.1842041015625, 269.5365295410156, 550.8699951171875, 90.62134552001953, 241.29200744628906, 62.60080337524414, 111.44552612304688, 192.56948852539062, 255.8974609375, 72.2399673461914, 310.4836120605469, 71.71769714355469, 182.8424072265625, 374.5487060546875, 262.06805419921875, 211.5977020263672, 180.73104858398438, 145.7102813720703, -34.156341552734375, 128.2239532470703, -99.81399536132812, 187.84930419921875, -33.39689254760742, 87.11019897460938, 316.21368408203125, 275.5994567871094, 241.00677490234375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000355.npy"}
|
|
{"epoch": 0.5212922173274597, "step": 356, "batch_size": 64, "mean": 122.82986450195312, "std": 240.24685668945312, "min": -404.4898681640625, "p10": -160.29486846923828, "median": 101.25813293457031, "p90": 430.4805541992188, "max": 721.10546875, "pos_frac": 0.65625, "sample": [205.42202758789062, -14.706306457519531, 283.19012451171875, 158.19024658203125, -43.93547439575195, -109.53684997558594, -92.19313049316406, 288.1344299316406, -15.121612548828125, 175.29156494140625, 359.96771240234375, 5.753414154052734, 721.10546875, 435.3789367675781, -66.17927551269531, 458.8797607421875, -154.2454376220703, -167.99855041503906, -404.4898681640625, -145.21640014648438, 76.45140838623047, 542.1712036132812, -41.39888000488281, 93.03106689453125, 278.4779052734375, 419.0509948730469, -69.78255462646484, 375.3537902832031, -223.54605102539062, 82.4553451538086, -138.2218475341797, 568.8829956054688, 101.22807312011719, -191.3883056640625, 339.20635986328125, 101.28819274902344, 280.2698059082031, 298.58111572265625, -162.88748168945312, 61.71669006347656, -21.902660369873047, -129.50302124023438, 180.88037109375, 218.45172119140625, 7.572919845581055, 317.9547119140625, 284.2559814453125, -137.7316436767578, 506.9775390625, 179.47482299804688, 190.73385620117188, 22.351409912109375, -172.8247833251953, -42.730735778808594, 159.18804931640625, 156.06570434570312, 93.00505065917969, 240.9019775390625, -334.1764831542969, 716.1112060546875, 285.00592041015625, 337.35626220703125, 20.47925567626953, 114.58316802978516], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000356.npy"}
|
|
{"epoch": 0.5227606461086637, "step": 357, "batch_size": 64, "mean": 180.96995544433594, "std": 189.15882873535156, "min": -325.4189453125, "p10": -21.414439392089843, "median": 163.14852142333984, "p90": 440.4057373046876, "max": 659.162109375, "pos_frac": 0.84375, "sample": [305.38079833984375, 449.4688720703125, 72.74299621582031, 536.792724609375, 373.67822265625, 64.23243713378906, 80.08601379394531, 326.0804443359375, 97.60369873046875, 452.7628479003906, 279.47235107421875, 86.70882415771484, -66.1705093383789, 419.2584228515625, -40.4569091796875, 111.59947967529297, 193.37527465820312, 147.60110473632812, 286.816162109375, -134.5232391357422, 548.9981079101562, 52.60521697998047, 156.46853637695312, 15.262435913085938, -325.4189453125, 42.873931884765625, 211.67958068847656, 107.3825454711914, -45.6368408203125, 457.41192626953125, 377.7880554199219, -12.469917297363281, 394.5068664550781, -69.90145874023438, 298.59954833984375, 14.421710968017578, 367.1820373535156, 659.162109375, 213.5428466796875, 274.98828125, 32.727081298828125, -22.410303115844727, 275.8077087402344, 346.31976318359375, 81.17591094970703, 112.35520935058594, 108.39073181152344, 240.91397094726562, -15.91098403930664, 183.755859375, -19.090757369995117, 295.435302734375, 175.11404418945312, 29.948196411132812, 273.1198425292969, 252.0797119140625, 74.05655670166016, 333.9873962402344, 7.943634033203125, 169.82850646972656, 69.52409362792969, 581.1204833984375, 36.36322021484375, 177.5655517578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000357.npy"}
|
|
{"epoch": 0.5242290748898678, "step": 358, "batch_size": 64, "mean": 154.9437255859375, "std": 187.53564453125, "min": -277.2663879394531, "p10": -29.11322784423828, "median": 109.27006912231445, "p90": 384.8438995361329, "max": 744.0596313476562, "pos_frac": 0.8125, "sample": [78.82403564453125, 403.40435791015625, 88.36451721191406, 227.56373596191406, 219.2070770263672, 279.21368408203125, 65.51139831542969, 207.45602416992188, 104.79905700683594, 149.66476440429688, -27.312660217285156, 14.280181884765625, -40.848968505859375, -24.861068725585938, 16.177167892456055, 599.4205322265625, 5.795326232910156, 260.06475830078125, 83.85345458984375, 79.58824157714844, -62.34021759033203, -29.839492797851562, 11.690986633300781, 57.41199493408203, 85.01199340820312, -277.2663879394531, 246.28497314453125, -172.16482543945312, 662.5260009765625, 370.9003601074219, 350.12646484375, 204.56674194335938, 275.33355712890625, -1.4636001586914062, 440.31787109375, 182.12753295898438, 40.960792541503906, 77.27749633789062, -58.556053161621094, 230.152099609375, 267.02081298828125, 321.4754333496094, -78.55268859863281, 247.99908447265625, 320.55908203125, -25.856597900390625, 297.9753723144531, 472.8155212402344, 33.50258255004883, 108.22972106933594, 250.93182373046875, 162.8453369140625, 50.653656005859375, 11.910629272460938, 390.8197021484375, 744.0596313476562, 82.90795135498047, 61.542015075683594, 209.75401306152344, -27.418609619140625, 169.7555694580078, 110.31041717529297, 198.92111206054688, 111.0126953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000358.npy"}
|
|
{"epoch": 0.5256975036710719, "step": 359, "batch_size": 64, "mean": 189.73448181152344, "std": 217.5166473388672, "min": -207.9223175048828, "p10": -62.7234504699707, "median": 146.488525390625, "p90": 494.87040405273456, "max": 791.45458984375, "pos_frac": 0.8125, "sample": [-207.9223175048828, 240.67520141601562, 125.39541625976562, 375.09222412109375, -156.49142456054688, 590.5405883789062, -134.76019287109375, 791.45458984375, 99.91777801513672, -16.85600471496582, -66.45796203613281, -53.23413848876953, 345.05194091796875, 320.3042297363281, -130.18942260742188, 124.41712951660156, 252.98513793945312, -45.7906494140625, 374.80145263671875, -63.308021545410156, 210.16998291015625, 146.46493530273438, 443.819580078125, 94.06002044677734, 77.37413024902344, 126.62123107910156, 150.03765869140625, 270.3626708984375, 104.45050048828125, 87.80330657958984, 393.09857177734375, 277.00927734375, 777.904296875, 282.3380126953125, -61.35945129394531, 86.42044830322266, 227.12957763671875, 98.03881072998047, -83.22870635986328, 583.60546875, 102.69683837890625, 335.1365661621094, 172.1046600341797, 292.0108337402344, 6.24920654296875, 107.72833251953125, 390.11029052734375, 146.51211547851562, 82.4029312133789, 519.2083129882812, 32.812599182128906, 21.518783569335938, -20.505172729492188, 228.03216552734375, 126.35514831542969, 516.7493286132812, 2.6622314453125, 164.7626190185547, 602.856689453125, 408.57196044921875, 180.41473388671875, 410.595703125, 90.48661804199219, 165.78717041015625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000359.npy"}
|
|
{"epoch": 0.527165932452276, "step": 360, "batch_size": 64, "mean": 145.714599609375, "std": 158.34866333007812, "min": -151.21014404296875, "p10": -57.90260887145995, "median": 107.01808166503906, "p90": 366.37043457031257, "max": 491.4551086425781, "pos_frac": 0.8125, "sample": [40.249481201171875, 300.08831787109375, 23.849571228027344, 107.020263671875, -43.411033630371094, 313.22247314453125, 52.179664611816406, 185.88812255859375, 301.69720458984375, 60.53700256347656, 96.8250503540039, 263.8341369628906, -66.31195068359375, 214.53680419921875, 461.1441345214844, 138.45745849609375, 31.12131118774414, -5.7788238525390625, 327.2294921875, 51.061004638671875, 303.2338562011719, 195.34701538085938, -37.03826141357422, 211.1630401611328, 75.81552124023438, 259.99566650390625, 342.54779052734375, 65.97380065917969, 40.24773025512695, 376.58013916015625, 491.4551086425781, 287.7806701660156, 425.57928466796875, -69.28443908691406, 43.35614776611328, 91.15556335449219, 183.67178344726562, 162.765380859375, 418.63348388671875, -69.72088623046875, 56.76286697387695, 477.26995849609375, -151.21014404296875, -78.48440551757812, 8.187652587890625, 82.23173522949219, -84.14385223388672, 107.96989440917969, 130.67477416992188, -63.907649993896484, 30.758895874023438, 235.5076904296875, 43.119117736816406, 277.5224609375, 275.8997802734375, 242.5689697265625, 221.8888397216797, -0.7183303833007812, 107.01589965820312, 380.35015869140625, 66.93940734863281, -43.890846252441406, 51.933982849121094, 298.7891845703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000360.npy"}
|
|
{"epoch": 0.5286343612334802, "step": 361, "batch_size": 64, "mean": 152.30418395996094, "std": 201.8604736328125, "min": -328.0196838378906, "p10": -86.78822174072265, "median": 162.4267120361328, "p90": 415.74290466308605, "max": 624.5262451171875, "pos_frac": 0.75, "sample": [12.552999496459961, 160.3717041015625, 21.667522430419922, 198.88238525390625, -28.26184844970703, -146.24765014648438, 55.786277770996094, 165.5841522216797, 472.2335510253906, 179.26927185058594, 173.95050048828125, 217.09310913085938, -328.0196838378906, 392.60699462890625, -22.985977172851562, -46.450462341308594, 179.40325927734375, 130.951171875, -182.2982940673828, 81.67691040039062, 215.42898559570312, -88.8415756225586, 83.0638427734375, -12.074504852294922, 206.0349578857422, 589.2396240234375, 188.88522338867188, 377.8561096191406, 145.34811401367188, 337.8173828125, -21.04034423828125, -20.544673919677734, 185.00599670410156, 77.10775756835938, 452.49761962890625, 376.429443359375, -114.35234069824219, 624.5262451171875, 615.1649169921875, 314.60101318359375, 205.62042236328125, 246.67649841308594, 466.9443054199219, 345.129638671875, 137.94737243652344, 175.5458526611328, 67.82530975341797, 164.48171997070312, -29.674774169921875, -81.99706268310547, 14.662639617919922, -45.375675201416016, 425.6582946777344, 6.063232421875, 195.8599395751953, 323.598388671875, 145.0203399658203, -97.18232727050781, 238.98501586914062, 36.86997985839844, 128.129150390625, -230.23818969726562, 295.5277099609375, 391.4696350097656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000361.npy"}
|
|
{"epoch": 0.5301027900146843, "step": 362, "batch_size": 64, "mean": 226.57217407226562, "std": 210.6799774169922, "min": -90.48887634277344, "p10": -30.795259094238272, "median": 196.4655532836914, "p90": 486.3828308105469, "max": 999.1611328125, "pos_frac": 0.859375, "sample": [74.77151489257812, 174.59701538085938, 481.04742431640625, 405.6882019042969, 222.23330688476562, 7.192787170410156, 999.1611328125, 255.62002563476562, 130.94525146484375, 472.2675476074219, 416.7600402832031, 77.94200897216797, 83.4828872680664, 178.04510498046875, 217.4837188720703, 386.7830505371094, 113.19190216064453, -66.06998443603516, 112.77459716796875, 285.72637939453125, 287.6844482421875, 81.52261352539062, 39.80231475830078, 15.449676513671875, -90.48887634277344, -7.978532791137695, 114.39385986328125, 163.3833770751953, -82.76094818115234, 633.5194091796875, 518.8137817382812, 120.69206237792969, -43.148345947265625, 279.0280456542969, 249.97950744628906, 200.19320678710938, 514.50390625, -34.52787780761719, 223.0009002685547, 488.66943359375, 202.68255615234375, 37.951744079589844, 607.7244262695312, 405.7458190917969, 90.91226959228516, 138.64529418945312, 287.359619140625, 182.04653930664062, 56.18873596191406, 662.9564819335938, 192.73789978027344, 472.52069091796875, 401.9178161621094, 278.51641845703125, 179.39761352539062, 370.6184387207031, 421.731201171875, -50.79736328125, 155.0864715576172, 282.6545104980469, 230.77267456054688, 270.6375427246094, -22.0858154296875, -56.679290771484375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000362.npy"}
|
|
{"epoch": 0.5315712187958884, "step": 363, "batch_size": 64, "mean": 237.05946350097656, "std": 209.82504272460938, "min": -261.86785888671875, "p10": -5.594107818603515, "median": 219.4942169189453, "p90": 496.59907836914067, "max": 936.4427490234375, "pos_frac": 0.875, "sample": [-261.86785888671875, 284.6164855957031, 484.1031799316406, 197.875244140625, 113.88230895996094, 232.9718017578125, 247.20989990234375, 249.714599609375, 164.47926330566406, 145.6903839111328, 373.2059326171875, 138.7640838623047, 341.4405822753906, 127.74625396728516, 292.79888916015625, 702.2318115234375, -49.18041229248047, 444.8601989746094, 105.54689025878906, 615.3734130859375, 356.51123046875, 150.84640502929688, 242.65768432617188, 355.2357177734375, 234.8587646484375, 334.024658203125, 936.4427490234375, 390.68389892578125, 15.0714111328125, 499.059814453125, 122.50584411621094, 193.01792907714844, 51.848388671875, 186.58551025390625, 401.8557434082031, 121.9411392211914, 192.3040771484375, -16.097871780395508, 490.85736083984375, 412.4742431640625, 356.18719482421875, -57.116539001464844, 31.03058624267578, 393.60894775390625, -19.994665145874023, 318.8642578125, 512.0816040039062, -5.9315185546875, 206.01663208007812, 446.0259704589844, 3.8014450073242188, 99.46110534667969, -35.95946502685547, 313.21502685546875, 276.3587951660156, -4.806816101074219, 12.773307800292969, 196.84490966796875, 43.16707992553711, 83.20502471923828, 596.5239868164062, 503.0498046875, 30.512535095214844, 248.73858642578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000363.npy"}
|
|
{"epoch": 0.5330396475770925, "step": 364, "batch_size": 64, "mean": 175.40988159179688, "std": 207.92481994628906, "min": -314.14788818359375, "p10": -21.096369171142577, "median": 184.96182250976562, "p90": 440.3346252441406, "max": 765.6275634765625, "pos_frac": 0.859375, "sample": [305.652099609375, 280.3165588378906, -221.5150604248047, 55.48979949951172, -314.14788818359375, 438.65191650390625, -196.70806884765625, 293.9517822265625, 262.64520263671875, 765.6275634765625, 153.54978942871094, 509.9059143066406, 349.85284423828125, 212.04869079589844, -21.42554473876953, 49.98960876464844, 31.092391967773438, -161.91195678710938, 256.2455139160156, -20.328292846679688, 0.5619850158691406, 303.89813232421875, 194.3970947265625, 351.8048095703125, 245.97052001953125, 44.46177673339844, 11.508296966552734, 441.0557861328125, 210.04803466796875, 223.09475708007812, 1.3324127197265625, 513.1948852539062, 91.7490463256836, 86.33642578125, 83.30741119384766, 65.10514831542969, 289.79425048828125, 220.0931396484375, 329.9770812988281, -72.92733764648438, 58.468238830566406, 473.022705078125, 502.154296875, 83.80876159667969, 200.4550018310547, -14.986763000488281, 38.56224822998047, 294.99859619140625, 408.28125, 241.12216186523438, 55.3293571472168, 175.52655029296875, 22.644865036010742, 162.68850708007812, 699.832763671875, 70.69320678710938, 225.4111328125, 270.0235290527344, 29.421865463256836, 236.28518676757812, 72.28302001953125, 121.5755615234375, 337.789794921875, -202.90591430664062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000364.npy"}
|
|
{"epoch": 0.5345080763582967, "step": 365, "batch_size": 64, "mean": 177.15826416015625, "std": 203.67063903808594, "min": -176.1370849609375, "p10": -31.097274780273434, "median": 128.21404266357422, "p90": 436.13062438964874, "max": 933.8004150390625, "pos_frac": 0.828125, "sample": [165.1220703125, 100.00296783447266, 34.6512451171875, 114.28352355957031, 119.95353698730469, 113.37501525878906, 306.01861572265625, 31.484169006347656, 180.678955078125, -16.702423095703125, 209.5637969970703, 359.38507080078125, -32.27239990234375, 241.65130615234375, -13.680587768554688, -20.234039306640625, 79.84336853027344, 725.7061767578125, 483.42694091796875, 251.31317138671875, 61.89099884033203, 101.68006134033203, 24.021263122558594, 10.6610107421875, 165.18270874023438, -83.458740234375, 121.11528015136719, 110.07186889648438, 178.76510620117188, 135.31280517578125, -76.56532287597656, -46.190670013427734, 117.0693130493164, 36.195648193359375, 331.35845947265625, 469.0215759277344, 199.83047485351562, 62.507816314697266, -28.355316162109375, 273.86834716796875, 213.58058166503906, -113.3728256225586, 222.59669494628906, 19.3431339263916, 13.740835189819336, 294.40960693359375, 249.1070556640625, 231.3392333984375, 283.466064453125, -78.00958251953125, 54.459632873535156, 328.6819763183594, 57.46105194091797, 271.1085205078125, 228.4604949951172, 226.4403076171875, 933.8004150390625, 503.23309326171875, 347.379638671875, -176.1370849609375, 638.5066528320312, 540.151611328125, 104.98085021972656, 345.8484191894531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000365.npy"}
|
|
{"epoch": 0.5359765051395007, "step": 366, "batch_size": 64, "mean": 182.03890991210938, "std": 208.67599487304688, "min": -200.4481658935547, "p10": -89.04701461791993, "median": 176.84416961669922, "p90": 435.00901794433594, "max": 713.92529296875, "pos_frac": 0.78125, "sample": [-99.55159759521484, 165.6317596435547, 281.7289123535156, 79.18540954589844, -159.87222290039062, 30.282758712768555, -75.09764862060547, 74.57598876953125, -40.22796630859375, 344.85296630859375, 83.50736236572266, -189.17294311523438, 387.4900817871094, 499.68182373046875, 190.994873046875, -1.7132110595703125, 253.30935668945312, 29.374401092529297, 106.4992446899414, -88.01441955566406, 324.5094909667969, 363.55230712890625, 259.02996826171875, 26.697097778320312, 428.318359375, 713.92529296875, 308.9852294921875, 83.77523040771484, 377.3584899902344, 188.05657958984375, 134.51123046875, -36.546417236328125, -104.18425750732422, 448.82867431640625, 340.00689697265625, 328.86358642578125, 666.7859497070312, 312.2642517089844, 223.30474853515625, 120.41838073730469, 190.46664428710938, 136.92593383789062, 62.049774169921875, -118.35855102539062, 416.119873046875, -69.9303970336914, 299.00677490234375, 47.39400100708008, 432.31854248046875, 436.1620788574219, 337.5775146484375, 229.58377075195312, 486.1064147949219, 195.76962280273438, 561.7506713867188, 261.40203857421875, -89.48955535888672, 95.08039855957031, -61.44482421875, 133.12521362304688, 125.13127899169922, 86.7625503540039, 275.50189208984375, -200.4481658935547], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000366.npy"}
|
|
{"epoch": 0.5374449339207048, "step": 367, "batch_size": 64, "mean": 154.576416015625, "std": 178.8810577392578, "min": -323.8350524902344, "p10": -46.80113143920897, "median": 111.06131744384766, "p90": 412.8476287841797, "max": 507.0007019042969, "pos_frac": 0.84375, "sample": [64.59078216552734, 115.44400024414062, 437.8763427734375, 57.27280044555664, 491.3495178222656, 25.565549850463867, 313.5649719238281, 81.18850708007812, 427.05517578125, 18.249717712402344, 33.280517578125, -4.234418869018555, -50.7626953125, 428.7479553222656, 348.03125, 227.1446533203125, 297.7861328125, 86.53192138671875, 191.20101928710938, -55.819061279296875, 225.859619140625, 413.3957824707031, -38.79578399658203, 7.772941589355469, 181.37164306640625, 291.4207458496094, 507.0007019042969, 63.63169860839844, 503.0364990234375, 68.4031982421875, 381.57684326171875, 106.67863464355469, 23.08597183227539, 256.90228271484375, 68.06172180175781, 176.2752685546875, 281.94989013671875, 347.18359375, 297.40850830078125, 71.06192016601562, 166.10198974609375, -153.18505859375, -131.41114807128906, -50.23199462890625, -31.24005126953125, 1.1406383514404297, 21.685943603515625, 226.59957885742188, 155.9685516357422, 71.7650146484375, -97.20647430419922, -323.8350524902344, 136.2325439453125, 59.76887512207031, 257.112548828125, 299.89080810546875, 359.160400390625, 191.9089813232422, 362.479736328125, 24.176498413085938, 96.72164916992188, 9.671913146972656, 411.568603515625, 60.699005126953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000367.npy"}
|
|
{"epoch": 0.5389133627019089, "step": 368, "batch_size": 64, "mean": 216.4133758544922, "std": 165.5128631591797, "min": -136.6330108642578, "p10": 12.735290908813486, "median": 228.97915649414062, "p90": 424.18435974121104, "max": 764.6953125, "pos_frac": 0.9375, "sample": [309.79815673828125, 267.7571105957031, 201.40682983398438, 262.6351013183594, 282.42254638671875, 442.3084716796875, 61.981876373291016, 24.096054077148438, 454.8722229003906, 131.70037841796875, 7.669467926025391, 335.80145263671875, 333.520751953125, 70.90585327148438, 273.3719177246094, 293.3136901855469, 220.45901489257812, 156.12388610839844, -88.16333770751953, 286.5568542480469, 156.71267700195312, 161.5185546875, 474.4323425292969, 289.6129455566406, 281.44781494140625, 69.12442016601562, 286.640625, 4.9074249267578125, 28.53750991821289, 8.921001434326172, 80.81846618652344, 328.2386474609375, 281.99267578125, 214.08978271484375, 230.23736572265625, 298.9498596191406, 253.67779541015625, 59.9383544921875, 163.80174255371094, -22.876609802246094, 231.9910888671875, 433.60296630859375, 305.7821960449219, 438.2623596191406, 143.1779327392578, 288.8596496582031, 608.3596801757812, 21.635299682617188, -136.6330108642578, 66.63569641113281, 93.5950698852539, 390.1291809082031, 295.8343505859375, 402.2076110839844, 349.6897277832031, 73.19259643554688, 163.75433349609375, 764.6953125, 179.9100341796875, 227.720947265625, 181.70883178710938, 30.673995971679688, -28.816478729248047, 345.22491455078125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000368.npy"}
|
|
{"epoch": 0.540381791483113, "step": 369, "batch_size": 64, "mean": 136.9704132080078, "std": 202.93460083007812, "min": -357.65264892578125, "p10": -104.77685546875, "median": 120.19867324829102, "p90": 411.094711303711, "max": 617.0467529296875, "pos_frac": 0.734375, "sample": [-104.17721557617188, -69.7066650390625, 607.9087524414062, -56.11481475830078, -206.1917266845703, 0.1961650848388672, -21.767112731933594, 272.76812744140625, 3.8090171813964844, 425.9104919433594, 228.51473999023438, -105.03384399414062, 183.00259399414062, 235.22119140625, 243.55921936035156, 617.0467529296875, 374.5132141113281, 196.97091674804688, -23.728313446044922, 293.4466857910156, 281.609130859375, 120.46293640136719, 123.65939331054688, 168.09182739257812, 215.6475830078125, 117.01910400390625, -341.53753662109375, 184.48501586914062, 142.26681518554688, 100.58605194091797, 242.5093536376953, 286.1619873046875, 253.31765747070312, 6.5910797119140625, 108.4267578125, 136.87551879882812, 277.2991027832031, 98.5483627319336, -2.8751373291015625, 316.1101379394531, 419.32965087890625, -133.24778747558594, 525.5965576171875, 88.66627502441406, 391.8798522949219, 221.25567626953125, 307.7398681640625, -137.47024536132812, -9.59786605834961, 441.41510009765625, 99.04803466796875, 95.24887084960938, -357.65264892578125, 72.68634796142578, 116.92174530029297, -65.77700805664062, -150.22471618652344, 54.322120666503906, 119.93441009521484, 498.0048828125, 186.11386108398438, -35.37842559814453, 89.85989379882812, -3.9712600708007812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000369.npy"}
|
|
{"epoch": 0.5418502202643172, "step": 370, "batch_size": 64, "mean": 121.46247100830078, "std": 187.2166748046875, "min": -277.1711730957031, "p10": -123.90952072143554, "median": 130.26538848876953, "p90": 392.5730651855471, "max": 551.4860229492188, "pos_frac": 0.71875, "sample": [77.3291244506836, 216.2713623046875, 132.7830047607422, -151.38792419433594, 551.4860229492188, 243.2939453125, 142.4368896484375, 177.2624053955078, 36.12275695800781, -14.615394592285156, 284.6063232421875, 232.56405639648438, 28.084808349609375, 143.1366729736328, -120.92680358886719, 417.62786865234375, 288.7930908203125, 87.63801574707031, 188.54180908203125, 486.9519958496094, 535.9102783203125, 77.64385986328125, 300.4625244140625, -67.12047576904297, 237.58078002929688, -37.89295196533203, 62.6212158203125, 106.52487182617188, -125.18782806396484, 160.14111328125, 417.68927001953125, 41.244659423828125, -19.112533569335938, 232.92945861816406, -19.24191665649414, 242.76629638671875, 214.80001831054688, -88.76666259765625, -248.2998504638672, 19.63411521911621, 156.38748168945312, 198.00592041015625, -125.19679260253906, 123.59677124023438, 336.4990234375, 127.74777221679688, 90.92041015625, 303.01385498046875, 161.8333282470703, -277.1711730957031, -72.31843566894531, -32.1046142578125, 92.0771484375, 161.03172302246094, -167.7032470703125, -99.77535247802734, 17.03797149658203, -175.11129760742188, -39.944881439208984, 281.3544921875, 451.0711364746094, 212.53811645507812, 140.87850952148438, 416.60479736328125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000370.npy"}
|
|
{"epoch": 0.5433186490455213, "step": 371, "batch_size": 64, "mean": 167.2401885986328, "std": 209.6616668701172, "min": -397.03863525390625, "p10": -78.27060546874998, "median": 161.70526123046875, "p90": 443.6722778320313, "max": 614.0314331054688, "pos_frac": 0.78125, "sample": [315.6446228027344, 235.31460571289062, 388.1095886230469, 56.83147430419922, -397.03863525390625, 213.57479858398438, 480.740966796875, 180.97271728515625, 545.2908935546875, 270.49884033203125, 400.3514404296875, 28.75960350036621, 287.889892578125, 169.07254028320312, -211.8216552734375, 60.39503479003906, 50.29402160644531, 151.78553771972656, 446.18902587890625, 155.41390991210938, -26.785139083862305, -215.9390411376953, 372.3565673828125, -102.96235656738281, -107.5130615234375, 169.8118133544922, 607.8402099609375, 346.36236572265625, -45.49604797363281, 107.2447509765625, 571.9129028320312, -53.87217712402344, 213.75277709960938, 175.71644592285156, -58.665916442871094, -39.515106201171875, 307.0033264160156, -86.67261505126953, 167.99661254882812, 353.23583984375, -52.998260498046875, 126.60929107666016, 478.2085876464844, 118.98126983642578, 26.611736297607422, 241.89529418945312, 79.98727416992188, 339.01953125, 388.188232421875, 3.1301116943359375, 306.45208740234375, 118.13426971435547, 154.52781677246094, 271.3365478515625, 174.42955017089844, 60.463653564453125, 144.86683654785156, 68.06773376464844, -99.39857482910156, 437.79986572265625, 209.49668884277344, 614.0314331054688, -2.9611968994140625, 12.411148071289062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000371.npy"}
|
|
{"epoch": 0.5447870778267254, "step": 372, "batch_size": 64, "mean": 215.7982635498047, "std": 239.59451293945312, "min": -179.02252197265625, "p10": -85.99593811035155, "median": 176.17650604248047, "p90": 517.8936828613281, "max": 840.3820190429688, "pos_frac": 0.78125, "sample": [317.595703125, 298.30560302734375, 511.68743896484375, -117.48170471191406, 37.94917297363281, 518.1171264648438, 517.372314453125, 227.70852661132812, 242.39248657226562, -7.7715911865234375, 840.3820190429688, 322.5032958984375, 85.094482421875, 179.86883544921875, -16.427276611328125, 113.73017120361328, 92.10679626464844, 100.95781707763672, 628.2177734375, -179.02252197265625, 214.36402893066406, 172.4841766357422, 504.6572570800781, -149.95407104492188, 471.5087890625, 193.9774627685547, 85.69356536865234, 118.94485473632812, 70.63751220703125, 134.1636962890625, 94.06771850585938, 228.20150756835938, 380.3889465332031, -30.20775604248047, 162.38079833984375, -66.69286346435547, 368.291015625, 541.5619506835938, 104.29764556884766, 753.7131958007812, 445.4671936035156, 326.9722595214844, 199.13497924804688, -159.33316040039062, 370.99462890625, 416.4582214355469, 447.57794189453125, 151.1663055419922, -21.503524780273438, 227.46832275390625, 141.81802368164062, 250.5708465576172, 123.353759765625, -45.376609802246094, 662.278076171875, 199.5295867919922, -10.810396194458008, 170.35833740234375, -94.26868438720703, 746.8544921875, -129.53831481933594, -169.16864013671875, 370.866455078125, 124.45272064208984], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000372.npy"}
|
|
{"epoch": 0.5462555066079295, "step": 373, "batch_size": 64, "mean": 204.6912841796875, "std": 204.4394073486328, "min": -323.6785583496094, "p10": -24.763478469848607, "median": 173.09322357177734, "p90": 489.4676483154297, "max": 692.9193115234375, "pos_frac": 0.875, "sample": [112.67098236083984, -135.1645050048828, -74.17153930664062, 144.43218994140625, 90.60125732421875, 640.7772216796875, 366.96173095703125, 133.38304138183594, 241.890380859375, 143.85366821289062, 523.7476196289062, 172.8551483154297, 11.744293212890625, 119.35382843017578, 329.92779541015625, 99.43000030517578, 14.8289155960083, 559.2835693359375, 219.18197631835938, 74.97017669677734, -215.12698364257812, -35.327484130859375, 207.9840087890625, -323.6785583496094, 142.62844848632812, 46.002227783203125, 173.331298828125, 170.7411346435547, 474.0306091308594, 277.19598388671875, 149.9459991455078, 349.17926025390625, 292.3697509765625, 303.57452392578125, 345.8747863769531, -0.11413192749023438, -101.81937408447266, 477.9189453125, 225.9375, 65.30415344238281, 692.9193115234375, 102.90404510498047, 322.0301208496094, 263.18255615234375, 360.34039306640625, 323.70050048828125, 267.1749572753906, 325.1939697265625, 484.184814453125, 491.7317199707031, 109.1588134765625, 91.3143310546875, 116.09136962890625, 125.89483642578125, 26.259384155273438, 225.42514038085938, 185.9595489501953, 129.26852416992188, 535.1126708984375, 293.83721923828125, 337.9887390136719, 544.2088623046875, -177.3914031982422, 107.24169158935547], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000373.npy"}
|
|
{"epoch": 0.5477239353891337, "step": 374, "batch_size": 64, "mean": 183.78297424316406, "std": 246.6298370361328, "min": -366.37701416015625, "p10": -70.35374755859371, "median": 145.8946533203125, "p90": 501.5327392578125, "max": 928.4658203125, "pos_frac": 0.78125, "sample": [-91.12346649169922, 130.74609375, -22.322158813476562, 928.4658203125, 173.60557556152344, 893.7086181640625, 176.95327758789062, 521.0230712890625, 282.40380859375, 19.493553161621094, 89.02883911132812, 245.51925659179688, 26.405303955078125, 137.37234497070312, 52.9739990234375, -21.918479919433594, 397.1333312988281, 264.4830627441406, 504.4879150390625, 256.56158447265625, -31.342071533203125, 89.22409057617188, 397.8365173339844, -24.192428588867188, -20.32103157043457, 541.248779296875, 7.167062759399414, 113.52098083496094, -171.7275848388672, 270.1490783691406, 224.12982177734375, 274.2821960449219, -102.5775146484375, 635.0734252929688, -228.403076171875, 185.26768493652344, 35.68473815917969, 440.99566650390625, 161.58291625976562, -27.53594970703125, 123.58918762207031, 107.13434600830078, 87.1955337524414, 70.76504516601562, 154.41696166992188, 175.43624877929688, -37.16072082519531, 447.4313659667969, 221.5035858154297, 299.3331298828125, 63.25559997558594, 494.6373291015625, -84.57933044433594, 342.6973876953125, 261.3428649902344, 14.287910461425781, -105.48119354248047, 287.1927185058594, 35.94273376464844, 73.40678405761719, 725.8513793945312, -366.37701416015625, 156.87213134765625, 478.3510437011719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000374.npy"}
|
|
{"epoch": 0.5491923641703378, "step": 375, "batch_size": 64, "mean": 178.8221893310547, "std": 236.7273712158203, "min": -482.8109130859375, "p10": -86.22624053955077, "median": 177.3263931274414, "p90": 488.8805419921876, "max": 778.763916015625, "pos_frac": 0.796875, "sample": [522.8854370117188, 352.7475280761719, 31.45367431640625, 227.91622924804688, -66.27549743652344, -123.89114379882812, 364.8148193359375, 197.332275390625, 120.85527038574219, -15.621448516845703, 33.173667907714844, 99.53562927246094, 126.40218353271484, -482.8109130859375, 60.241310119628906, 672.707275390625, 288.736572265625, 40.89186096191406, 306.68365478515625, 209.7803955078125, 27.6865234375, 249.4830322265625, -303.00604248046875, 103.80304718017578, 299.51123046875, -88.52168273925781, 513.6057739257812, 90.20204162597656, -17.67529296875, 129.68515014648438, -147.31405639648438, 15.486129760742188, 498.9137878417969, 531.8456420898438, 0.22835159301757812, 229.8614501953125, 332.4673156738281, 404.70794677734375, 339.52392578125, 526.7503662109375, 260.5952453613281, 162.31524658203125, 85.9541244506836, -273.9791259765625, 366.9942626953125, 337.1791076660156, 289.0300598144531, -189.80340576171875, 778.763916015625, 421.20880126953125, 334.5880126953125, -80.87020874023438, 2.8782958984375, 249.51181030273438, 94.5601806640625, 441.516357421875, 192.33753967285156, 465.4696350097656, -25.869041442871094, 94.30913543701172, 303.61578369140625, 346.2377624511719, 138.10647583007812, -54.83323287963867], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000375.npy"}
|
|
{"epoch": 0.5506607929515418, "step": 376, "batch_size": 64, "mean": 222.34678649902344, "std": 240.71397399902344, "min": -441.1722412109375, "p10": -37.28169403076172, "median": 164.29679107666016, "p90": 545.7631896972656, "max": 708.9410400390625, "pos_frac": 0.84375, "sample": [59.660545349121094, 536.2614135742188, 208.68807983398438, 60.12154006958008, 708.9410400390625, 94.95989990234375, 327.13299560546875, -185.4980926513672, 10.394393920898438, -24.417556762695312, 268.7818603515625, 1.53253173828125, 553.8235473632812, 97.6487045288086, -98.8659439086914, 419.960205078125, 439.00823974609375, -61.756927490234375, -35.881683349609375, 155.4164581298828, 73.16285705566406, 53.609458923339844, -106.40875244140625, 651.1985473632812, 630.8191528320312, 90.17098236083984, 325.3323974609375, 297.81939697265625, 474.3888244628906, -0.7999153137207031, 508.94732666015625, 546.75146484375, 93.99684143066406, 344.87677001953125, 202.82247924804688, 438.4311828613281, 439.7919006347656, 169.58975219726562, 351.99981689453125, 153.8905029296875, 123.55438232421875, 415.2985534667969, -37.88169860839844, 267.03802490234375, 494.4161071777344, 159.0038299560547, 644.3211669921875, 61.689186096191406, -105.67301940917969, 65.12097930908203, 6.2075958251953125, 58.482810974121094, 562.683837890625, 543.4572143554688, 247.6156005859375, 498.2413330078125, 118.44377136230469, 250.7501220703125, 156.40463256835938, 200.7301025390625, 52.533851623535156, 72.48031616210938, 540.144775390625, -441.1722412109375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000376.npy"}
|
|
{"epoch": 0.5521292217327459, "step": 377, "batch_size": 64, "mean": 203.26560974121094, "std": 218.23130798339844, "min": -301.8624267578125, "p10": -47.38658676147461, "median": 201.25446319580078, "p90": 507.8097778320313, "max": 615.0985107421875, "pos_frac": 0.8125, "sample": [368.91632080078125, 615.0985107421875, 197.28448486328125, 32.494842529296875, 202.0372772216797, -122.2861328125, 234.23452758789062, 1.416168212890625, 282.35150146484375, -46.484397888183594, 509.80474853515625, 545.7642822265625, -7.391868591308594, 600.033203125, 26.215627670288086, 204.5689697265625, 185.02073669433594, 417.3710021972656, 200.47164916992188, -35.76396179199219, 182.93914794921875, 492.42352294921875, 211.47830200195312, -13.158622741699219, 378.1341857910156, 613.7276000976562, 328.7239074707031, -167.44651794433594, 548.9716186523438, 169.38232421875, 342.0106506347656, 208.417236328125, 24.285057067871094, 118.86032104492188, 37.53532791137695, 326.086669921875, 132.12306213378906, 301.72821044921875, 247.2123260498047, 569.975341796875, 236.6114959716797, -47.77323913574219, -301.8624267578125, 267.22137451171875, 43.069549560546875, 95.05730438232422, 500.01922607421875, 191.5187530517578, -144.62367248535156, 268.68994140625, 31.319534301757812, 503.15484619140625, -46.03639221191406, -62.17254638671875, 55.43511199951172, 438.9732666015625, 164.377685546875, 501.8156433105469, 360.454345703125, 4.304130554199219, 264.4930419921875, 160.70147705078125, 226.43511962890625, -166.75137329101562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000377.npy"}
|
|
{"epoch": 0.55359765051395, "step": 378, "batch_size": 64, "mean": 225.80239868164062, "std": 214.27725219726562, "min": -352.0785827636719, "p10": -42.05877075195312, "median": 219.4757308959961, "p90": 491.85057678222665, "max": 850.9337158203125, "pos_frac": 0.828125, "sample": [558.5414428710938, 341.615966796875, 503.3017272949219, 570.277587890625, 465.1312255859375, 130.15406799316406, 211.32904052734375, 283.85784912109375, 216.35675048828125, 212.91183471679688, 344.59967041015625, 264.1448974609375, 398.6294860839844, 201.0169219970703, 238.3343505859375, -23.947952270507812, 233.28372192382812, 583.0508422851562, 172.70281982421875, 270.0484313964844, 180.05471801757812, -45.086204528808594, -31.845577239990234, 129.13970947265625, 211.17950439453125, 218.9024658203125, -67.79405212402344, 358.7496032714844, 120.09453582763672, 331.1772766113281, 405.07611083984375, 396.7471618652344, 178.25485229492188, 23.767026901245117, -34.99475860595703, 420.681884765625, 187.57464599609375, 850.9337158203125, 29.07856559753418, -142.69677734375, 278.2778625488281, -0.7192306518554688, 220.0489959716797, 505.2743225097656, 58.5777587890625, -65.05622863769531, 336.7082824707031, 288.63226318359375, 457.3580322265625, 367.0670166015625, 104.52284240722656, 282.2333984375, 451.50103759765625, 253.60491943359375, -352.0785827636719, 178.1432647705078, 201.419189453125, 241.05975341796875, -64.0477523803711, 519.22021484375, 81.44969177246094, -287.93951416015625, 84.65067291259766, 417.1098327636719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000378.npy"}
|
|
{"epoch": 0.5550660792951542, "step": 379, "batch_size": 64, "mean": 210.429443359375, "std": 239.05787658691406, "min": -297.3737487792969, "p10": -90.58422393798827, "median": 182.8217544555664, "p90": 500.89421997070315, "max": 903.1521606445312, "pos_frac": 0.8125, "sample": [213.31973266601562, 338.7689208984375, 109.44711303710938, 209.46923828125, -76.05095672607422, 75.1552963256836, 158.15985107421875, 523.58447265625, 128.54733276367188, 130.71383666992188, 302.83416748046875, 797.716796875, 115.76060485839844, 363.5378112792969, 368.59765625, 196.81100463867188, -297.3737487792969, 689.5221557617188, 57.89259338378906, 266.7670593261719, 398.26043701171875, 381.30780029296875, 281.37603759765625, 4.998676300048828, 129.57955932617188, -23.753202438354492, -206.31353759765625, 102.60031127929688, 40.271446228027344, 274.10504150390625, 502.3289794921875, 377.8001708984375, -214.1829833984375, 112.85089111328125, 407.5799255371094, 514.6002197265625, 290.8089599609375, 333.16790771484375, 466.7743835449219, 74.45960998535156, -96.8127670288086, 497.54644775390625, 146.43984985351562, -45.84403610229492, 370.2677001953125, -109.2210693359375, 127.00498962402344, 433.1665954589844, 366.19342041015625, 108.27137756347656, -120.72093200683594, 224.56622314453125, 528.9561767578125, 111.28858947753906, 903.1521606445312, 195.0596923828125, 107.27306365966797, 405.0773620605469, 170.5838165283203, -35.82082748413086, 17.890121459960938, 438.3175964355469, -135.30886840820312, -61.64393615722656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000379.npy"}
|
|
{"epoch": 0.5565345080763583, "step": 380, "batch_size": 64, "mean": 236.06048583984375, "std": 252.06959533691406, "min": -586.9954833984375, "p10": -63.274081420898405, "median": 243.71558380126953, "p90": 509.3045196533203, "max": 828.669189453125, "pos_frac": 0.859375, "sample": [125.91555786132812, 156.71388244628906, -29.790817260742188, 828.669189453125, 192.65597534179688, 92.11344146728516, 152.0604705810547, 384.395263671875, 353.1335144042969, 241.0800018310547, 100.15306854248047, 397.39239501953125, -29.927085876464844, 514.470947265625, 702.5834350585938, 417.8735656738281, 306.32769775390625, 337.2230224609375, 406.8670349121094, 108.60498046875, 263.580810546875, -260.68829345703125, 188.377685546875, 209.4742431640625, 81.00337219238281, -77.5656509399414, 416.91839599609375, 257.8346252441406, 187.8160858154297, 511.76129150390625, 317.4989318847656, 503.5720520019531, 480.9648742675781, 461.2938232421875, 427.4194641113281, 362.9396057128906, 95.78547668457031, 36.02105712890625, 233.6272735595703, 75.1493911743164, 709.2337646484375, 32.92042922973633, -193.54739379882812, 115.90347290039062, -102.78101348876953, 201.35362243652344, 387.06658935546875, 489.3721008300781, 45.67805480957031, -259.93695068359375, 368.4732666015625, 37.369293212890625, 643.9010620117188, 270.59429931640625, 246.35116577148438, -96.0163345336914, 400.6478576660156, 302.78045654296875, 149.34556579589844, 79.94635009765625, -586.9954833984375, 307.85565185546875, 687.9161376953125, 339.13861083984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000380.npy"}
|
|
{"epoch": 0.5580029368575624, "step": 381, "batch_size": 64, "mean": 177.61572265625, "std": 263.2547912597656, "min": -477.1943359375, "p10": -118.41689834594725, "median": 162.79911041259766, "p90": 524.419027709961, "max": 766.4384765625, "pos_frac": 0.734375, "sample": [213.13491821289062, 49.37835693359375, 5.800233840942383, 488.40789794921875, 427.5458984375, 599.1605834960938, 441.80584716796875, -36.12846374511719, 103.03588104248047, 505.4396667480469, 402.6697082519531, 78.41703796386719, 355.95294189453125, 267.5888671875, 766.4384765625, -126.32598114013672, 398.84228515625, 372.70977783203125, 584.7837524414062, 274.70263671875, 748.5807495117188, 609.41796875, -183.931884765625, 1.2859954833984375, -137.87693786621094, 149.24942016601562, 532.5530395507812, 18.221691131591797, -279.9515380859375, -5.246007919311523, 130.72386169433594, 225.16143798828125, -10.435379028320312, -37.072566986083984, 171.0159149169922, -78.51803588867188, 232.03515625, 109.22220611572266, 204.445556640625, -48.8428955078125, 43.615440368652344, 74.02792358398438, 741.6720581054688, 225.2538604736328, 315.22723388671875, 270.6009216308594, -24.90191650390625, -99.96237182617188, 383.8087158203125, -55.82954788208008, 30.64699363708496, 362.9397888183594, -10.445556640625, 181.94468688964844, 66.04496002197266, -201.61697387695312, 154.73846435546875, 334.18621826171875, 330.7760009765625, 125.50690460205078, 170.85975646972656, -477.1943359375, -379.9587707519531, 282.06781005859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000381.npy"}
|
|
{"epoch": 0.5594713656387665, "step": 382, "batch_size": 64, "mean": 204.6187744140625, "std": 197.6926727294922, "min": -246.59938049316406, "p10": -36.00947341918944, "median": 201.68577575683594, "p90": 507.229949951172, "max": 696.296142578125, "pos_frac": 0.84375, "sample": [696.296142578125, 472.67919921875, 107.59343719482422, 100.71929168701172, 95.92118072509766, 554.538330078125, -40.17597198486328, 64.27063751220703, 226.47698974609375, -83.89958953857422, 141.53431701660156, 222.27090454101562, 346.67779541015625, 330.18499755859375, 354.05487060546875, 378.7303161621094, 132.2238006591797, -19.896102905273438, 378.08880615234375, 94.58345031738281, 352.9715576171875, -26.287643432617188, 87.96937561035156, 258.2469482421875, 568.8829345703125, -46.716575622558594, 323.94952392578125, -80.69540405273438, 438.2113342285156, 219.195068359375, 530.453857421875, -81.39202117919922, 9.30856704711914, 270.28875732421875, 346.7427673339844, -246.59938049316406, -84.97250366210938, 289.68023681640625, -11.977798461914062, 251.11627197265625, 97.54535675048828, 197.39926147460938, 277.419189453125, 50.397544860839844, 180.4889678955078, 98.15733337402344, 229.75050354003906, 266.0903015136719, 210.8849639892578, 175.4064483642578, 522.0374145507812, 524.8733520507812, 291.26220703125, 25.576501846313477, 205.9722900390625, 565.447265625, 61.473392486572266, 28.74569320678711, 389.283203125, 455.1476745605469, 120.07378387451172, 87.73638153076172, 19.067672729492188, 94.11639404296875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000382.npy"}
|
|
{"epoch": 0.5609397944199707, "step": 383, "batch_size": 64, "mean": 207.67013549804688, "std": 264.35821533203125, "min": -609.5059204101562, "p10": -69.64219360351562, "median": 211.39759826660156, "p90": 548.7997680664063, "max": 822.691162109375, "pos_frac": 0.78125, "sample": [-20.887168884277344, -8.065616607666016, 76.74789428710938, -226.47451782226562, 439.0228271484375, -126.59141540527344, -70.00084686279297, 543.275146484375, 292.9830322265625, -20.210826873779297, 212.5828857421875, 221.80474853515625, 80.08514404296875, -28.599716186523438, -244.6116180419922, 210.21231079101562, 822.691162109375, 23.444923400878906, 230.97039794921875, 224.78932189941406, 604.7926025390625, -151.44869995117188, 58.212284088134766, 567.04541015625, 159.77841186523438, -68.80533599853516, 601.1910400390625, 114.18150329589844, -61.6593017578125, 231.83815002441406, 814.208251953125, 223.45863342285156, 525.897705078125, 197.34555053710938, 368.8087158203125, 270.6798095703125, 388.9942321777344, 59.28556823730469, 259.3916320800781, 270.26629638671875, 8.251495361328125, 6.98577880859375, 193.4471435546875, 93.12251281738281, 300.3012390136719, 479.902587890625, -56.59649658203125, 438.3916015625, 71.59828186035156, 255.95401000976562, 549.3753051757812, 136.73330688476562, -91.56507873535156, 241.70608520507812, 364.3212890625, 86.70264434814453, 681.5001220703125, 484.0638427734375, 142.54998779296875, 547.4568481445312, -609.5059204101562, 96.1724853515625, 371.30029296875, 432.0884094238281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000383.npy"}
|
|
{"epoch": 0.5624082232011748, "step": 384, "batch_size": 64, "mean": 168.70809936523438, "std": 173.3642578125, "min": -260.3528137207031, "p10": -65.07669067382811, "median": 153.8032684326172, "p90": 370.09532775878915, "max": 663.4417724609375, "pos_frac": 0.84375, "sample": [175.71554565429688, 64.98159790039062, 625.2213134765625, -69.62956237792969, 10.027252197265625, 96.10780334472656, -134.2141571044922, -1.2040863037109375, 663.4417724609375, 135.1385498046875, 433.3465270996094, 252.84683227539062, 208.6511688232422, -87.81783294677734, 247.1575164794922, 85.07579040527344, -84.41462707519531, 38.0725212097168, 286.0931396484375, 335.7427673339844, 120.68899536132812, 126.83267211914062, 104.63145446777344, 265.8973388671875, 214.67730712890625, 123.32286071777344, 76.7312240600586, -91.92725372314453, -260.3528137207031, 156.41786193847656, 272.8504638671875, 133.61900329589844, 37.73890686035156, 185.03866577148438, 183.46372985839844, -149.13711547851562, 393.2100830078125, 260.664306640625, 329.02130126953125, 40.63739013671875, 110.12034606933594, 196.26370239257812, 260.5970153808594, 319.09820556640625, 261.8928527832031, 244.08294677734375, 410.060546875, 149.35269165039062, 156.30606079101562, 60.51219940185547, 251.6768798828125, -11.604850769042969, 151.30047607421875, 497.15411376953125, 111.60025024414062, -54.45332336425781, 48.822513580322266, 344.3947448730469, 264.9834289550781, 149.10548400878906, 381.10986328125, 290.7548522949219, 104.11196899414062, 295.7090759277344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000384.npy"}
|
|
{"epoch": 0.5638766519823789, "step": 385, "batch_size": 64, "mean": 196.7218475341797, "std": 187.79299926757812, "min": -233.01248168945312, "p10": -23.504855346679676, "median": 169.57543182373047, "p90": 474.1608612060548, "max": 611.6703491210938, "pos_frac": 0.859375, "sample": [61.533935546875, -72.26586151123047, 72.88158416748047, 244.73587036132812, 394.61248779296875, 256.9559326171875, 383.91253662109375, 147.22015380859375, -64.0781021118164, 376.68121337890625, 164.7935333251953, 34.6822509765625, 97.71021270751953, 155.32827758789062, 439.5277099609375, -8.794746398925781, 541.5194702148438, 300.1987609863281, 100.74894714355469, 100.77176666259766, 288.9869689941406, 164.1066131591797, -120.05625915527344, 292.90087890625, -151.45401000976562, 194.01690673828125, 103.34443664550781, 493.303955078125, 46.30580139160156, 58.857025146484375, 265.40185546875, 190.8126678466797, 97.66363525390625, 581.6593017578125, 611.6703491210938, 127.31243896484375, 166.33871459960938, -13.204208374023438, 485.0098571777344, 245.31275939941406, -27.919418334960938, 102.69667053222656, 353.1819152832031, 190.2951202392578, 184.25047302246094, 482.0435791015625, 310.2826843261719, 51.1580810546875, 455.7678527832031, 246.2866973876953, -186.33456420898438, 46.03291320800781, 131.98345947265625, -233.01248168945312, 172.81214904785156, 236.30368041992188, 418.6120910644531, 132.56405639648438, 230.4660186767578, 551.1148681640625, 145.7909698486328, 307.77081298828125, 160.91433715820312, 270.1708068847656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000385.npy"}
|
|
{"epoch": 0.5653450807635829, "step": 386, "batch_size": 64, "mean": 211.4788055419922, "std": 235.0734405517578, "min": -339.0523681640625, "p10": -91.39491310119625, "median": 177.6403579711914, "p90": 512.8452362060547, "max": 830.0233764648438, "pos_frac": 0.84375, "sample": [-56.02237319946289, 308.6789245605469, -126.76488494873047, 346.3421325683594, 507.8154602050781, 227.39735412597656, 106.43732452392578, 348.2574462890625, 155.79962158203125, 354.4352111816406, 296.1111755371094, 742.1904296875, -191.38243103027344, -5.046113967895508, 515.0008544921875, 287.377685546875, 268.77593994140625, 388.2842712402344, 10.38591194152832, 107.01215362548828, 129.62417602539062, 258.347900390625, 143.91497802734375, 170.5421142578125, 205.10755920410156, 280.2630615234375, 126.36774444580078, 15.126468658447266, 830.0233764648438, 559.9236450195312, -48.11029052734375, 89.55014038085938, 172.37826538085938, -106.55457305908203, 152.8867645263672, 649.60302734375, 786.7505493164062, 307.75360107421875, 594.085693359375, 136.0464324951172, 213.00128173828125, 400.885009765625, -176.13063049316406, 139.41856384277344, 33.030914306640625, 171.33941650390625, 122.53327941894531, 185.63726806640625, -127.18794250488281, 296.5892333984375, -339.0523681640625, 393.19268798828125, 18.75696563720703, 87.43329620361328, 108.76177978515625, 107.01837921142578, 430.859619140625, 346.19140625, -194.9772491455078, 260.6918640136719, 305.68389892578125, 182.90245056152344, 409.23919677734375, 114.10822296142578], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000386.npy"}
|
|
{"epoch": 0.566813509544787, "step": 387, "batch_size": 64, "mean": 174.63983154296875, "std": 187.49375915527344, "min": -262.43731689453125, "p10": -14.359082412719722, "median": 166.18634033203125, "p90": 406.72687988281257, "max": 715.3258666992188, "pos_frac": 0.8125, "sample": [222.84527587890625, 10.676948547363281, 139.47776794433594, 143.05174255371094, 102.52511596679688, 715.3258666992188, 144.03506469726562, 80.7406005859375, 337.736083984375, 392.1739807128906, -262.43731689453125, 301.3709716796875, 100.640869140625, 135.87278747558594, -193.61688232421875, 230.73410034179688, 183.0203857421875, 419.97515869140625, -0.19849205017089844, 96.80746459960938, 590.0286254882812, 102.07080078125, 195.15377807617188, 612.599853515625, 329.3496398925781, 150.62765502929688, 124.6708755493164, 181.74502563476562, 46.040794372558594, 115.30686950683594, 45.147926330566406, 196.74757385253906, 283.0058898925781, 63.8714714050293, 435.8060607910156, 185.2978515625, 285.01287841796875, 118.93879699707031, 240.60836791992188, 316.71356201171875, 66.64730072021484, -16.283519744873047, 494.33673095703125, 344.489990234375, 278.8274841308594, 120.30645751953125, 243.17245483398438, -58.05764389038086, 412.9638366699219, -77.01509094238281, 266.2939453125, 91.27975463867188, -4.389167785644531, 262.9573669433594, -210.9481201171875, 195.725830078125, -6.350013732910156, 270.23956298828125, -4.05963134765625, -9.868728637695312, 247.26206970214844, -164.90585327148438, 222.18539428710938, 292.63690185546875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000387.npy"}
|
|
{"epoch": 0.5682819383259912, "step": 388, "batch_size": 64, "mean": 232.57754516601562, "std": 253.74038696289062, "min": -434.18963623046875, "p10": -2.82339859008788, "median": 222.12498474121094, "p90": 555.9727844238283, "max": 1152.4134521484375, "pos_frac": 0.890625, "sample": [1152.4134521484375, 329.6654968261719, -39.83955764770508, 440.8485412597656, 443.6901550292969, 687.5802001953125, 197.82850646972656, -234.24411010742188, 107.27709197998047, 270.2947082519531, 280.125, 727.580810546875, 337.49066162109375, 227.948974609375, -434.18963623046875, 22.26441192626953, 235.93572998046875, 264.04852294921875, 358.90850830078125, 7.986061096191406, 145.423095703125, 582.93017578125, 134.05206298828125, -91.30237579345703, 176.71022033691406, 241.03573608398438, 47.61376190185547, 41.67469787597656, 131.39730834960938, 216.30099487304688, 84.83892059326172, 126.61216735839844, 243.30943298339844, 22.00518798828125, 383.0579833984375, 184.77761840820312, 115.28312683105469, 94.44702911376953, 424.01568603515625, 115.4988784790039, 257.0018615722656, 124.70814514160156, 316.1437072753906, 77.65290069580078, 184.09881591796875, 103.29410552978516, 347.657470703125, 240.04248046875, -171.49221801757812, 338.21392822265625, 321.04132080078125, 888.6461791992188, -123.92698669433594, 347.09710693359375, 181.02603149414062, 21.372360229492188, -7.456024169921875, 322.93255615234375, 579.2433471679688, 594.7310791015625, 501.6748046875, 377.4790344238281, 237.38182067871094, 23.103864669799805], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000388.npy"}
|
|
{"epoch": 0.5697503671071953, "step": 389, "batch_size": 64, "mean": 199.80288696289062, "std": 243.26571655273438, "min": -256.6145935058594, "p10": -134.15240402221673, "median": 204.1396942138672, "p90": 485.1147735595704, "max": 798.1640625, "pos_frac": 0.8125, "sample": [53.240234375, 489.28863525390625, 367.93524169921875, -256.6145935058594, 66.48342895507812, 257.67498779296875, 184.33331298828125, 439.5965270996094, 270.66424560546875, 133.29888916015625, 245.88864135742188, -61.13611602783203, 211.63148498535156, -184.34072875976562, 148.60997009277344, 294.2998046875, 133.1229248046875, 390.0736083984375, 367.8873291015625, -54.0380859375, -254.67782592773438, 19.94241714477539, 494.65478515625, 28.821041107177734, 231.67242431640625, 475.3757629394531, -186.7451171875, 39.36820983886719, 146.08782958984375, 155.0958251953125, 315.7568664550781, 34.415924072265625, 124.43867492675781, 269.02862548828125, -235.67935180664062, 293.11053466796875, 196.6479034423828, 291.3248596191406, 308.1017150878906, 762.8887939453125, 388.81878662109375, -26.97816276550293, 363.3164978027344, 155.1101837158203, 258.2484130859375, 348.5089416503906, 48.491485595703125, 123.30030822753906, 408.05987548828125, 118.24243927001953, -30.068077087402344, 678.1015625, -0.0707550048828125, -165.44509887695312, 234.46864318847656, 698.4642333984375, 24.59307861328125, 798.1640625, 414.1184387207031, 8.823413848876953, 227.15045166015625, -196.69407653808594, 212.1278533935547, 691.0026245117188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000389.npy"}
|
|
{"epoch": 0.5712187958883994, "step": 390, "batch_size": 64, "mean": 196.91912841796875, "std": 288.4324951171875, "min": -394.5043029785156, "p10": -137.67621688842772, "median": 191.40157318115234, "p90": 440.2764068603516, "max": 1585.3828125, "pos_frac": 0.8125, "sample": [278.9395751953125, 260.94097900390625, 552.772705078125, 432.17864990234375, -383.119140625, -119.30216217041016, -182.84909057617188, 113.03889465332031, 323.53253173828125, 294.1398010253906, 417.3983459472656, 63.69306945800781, 248.98992919921875, 325.4630126953125, 149.36343383789062, 17.515716552734375, -31.62084197998047, 1585.3828125, -145.55081176757812, 400.8406982421875, 163.96170043945312, 332.79705810546875, 439.9151916503906, 100.88215637207031, 142.5876007080078, -64.23724365234375, 160.47093200683594, 4.07659912109375, 207.21702575683594, 260.11895751953125, 135.186279296875, 11.87176513671875, -195.381103515625, -39.956520080566406, 142.3325958251953, 494.2633056640625, 686.2738647460938, 497.00531005859375, -371.2799377441406, 317.9517822265625, 448.9306945800781, 440.43121337890625, 284.0418395996094, 128.72970581054688, 405.27435302734375, 161.43568420410156, 162.7882080078125, 72.47135925292969, 437.27618408203125, 416.09930419921875, 314.2395935058594, 155.6976776123047, 271.307373046875, 175.58612060546875, 320.8963928222656, -68.80721282958984, 18.499114990234375, -290.49627685546875, -394.5043029785156, 245.4403839111328, 270.7107238769531, 100.72494506835938, 253.35513305664062, 244.88973999023438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000390.npy"}
|
|
{"epoch": 0.5726872246696035, "step": 391, "batch_size": 64, "mean": 133.54165649414062, "std": 211.03897094726562, "min": -308.411865234375, "p10": -79.87749710083008, "median": 92.9854621887207, "p90": 400.56010742187505, "max": 829.953857421875, "pos_frac": 0.703125, "sample": [5.969646453857422, 12.08660888671875, 43.72443389892578, -11.960067749023438, 101.29100799560547, 240.03009033203125, 31.978403091430664, 32.56421661376953, 30.592578887939453, 259.0853576660156, -81.1617660522461, 185.17494201660156, -185.33901977539062, -1.3898849487304688, 84.67991638183594, 68.79354095458984, 269.27288818359375, 144.69403076171875, -2.196737289428711, -80.113037109375, 120.09379577636719, 402.8631896972656, 329.86199951171875, 395.1862487792969, -46.63483428955078, 114.77052307128906, 829.953857421875, -36.19880676269531, -28.19573974609375, 221.19393920898438, 263.5401611328125, -54.67656707763672, 82.86898803710938, 35.988101959228516, -79.3279037475586, 253.6116485595703, 119.03218841552734, -6.137199401855469, 474.7146301269531, 138.43817138671875, -41.64569854736328, -49.154258728027344, 159.00296020507812, 267.8804931640625, 644.48388671875, -148.95156860351562, 462.21136474609375, -0.6732692718505859, -205.5232391357422, 37.00011444091797, 48.59388732910156, 159.18460083007812, -135.40927124023438, 376.85003662109375, 415.60113525390625, 26.645889282226562, 640.86669921875, 329.95458984375, 276.57958984375, -308.411865234375, 231.41677856445312, 228.51202392578125, 244.04931640625, 208.8786163330078], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000391.npy"}
|
|
{"epoch": 0.5741556534508077, "step": 392, "batch_size": 64, "mean": 225.9717559814453, "std": 211.59774780273438, "min": -191.7513885498047, "p10": -11.626945495605455, "median": 199.06331634521484, "p90": 473.56088562011723, "max": 827.6825561523438, "pos_frac": 0.890625, "sample": [88.86518096923828, 127.76412200927734, -109.24610900878906, 116.7220458984375, 261.30535888671875, -45.483497619628906, 485.6778564453125, 164.27760314941406, 345.3390197753906, 326.71673583984375, 386.48876953125, -191.7513885498047, 262.87371826171875, 89.43002319335938, 68.52333068847656, 30.417905807495117, 478.1766357421875, 616.4075927734375, 397.52880859375, 206.40841674804688, -59.24944305419922, 267.24700927734375, 459.24993896484375, 67.68745422363281, 325.00830078125, 262.54168701171875, 224.16587829589844, 827.6825561523438, -17.284149169921875, 209.08795166015625, 191.7182159423828, 70.58026123046875, 357.93310546875, 149.2559814453125, 410.0582275390625, 177.68345642089844, 275.138916015625, 172.10943603515625, 187.5175018310547, 90.27955627441406, 342.09088134765625, 365.75555419921875, 31.783660888671875, 42.68671417236328, 462.7908020019531, 626.4736328125, 309.2230224609375, 422.7316589355469, 334.43548583984375, -148.67027282714844, 95.8718490600586, 154.49549865722656, 56.890377044677734, 291.43438720703125, 152.56576538085938, 135.08023071289062, 249.61082458496094, 21.886241912841797, 798.92236328125, 1.5731964111328125, -93.74584197998047, 68.29074096679688, 368.9639587402344, 616.197509765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000392.npy"}
|
|
{"epoch": 0.5756240822320118, "step": 393, "batch_size": 64, "mean": 179.87525939941406, "std": 274.1910095214844, "min": -252.09078979492188, "p10": -134.85715866088864, "median": 174.3599624633789, "p90": 451.4325866699219, "max": 1204.4541015625, "pos_frac": 0.703125, "sample": [257.09808349609375, 526.2568969726562, 36.92087936401367, -91.58026885986328, 295.533203125, 412.5795593261719, -112.88928985595703, 365.67144775390625, 249.22657775878906, 223.52166748046875, 79.16766357421875, 195.22210693359375, 109.06046295166016, 340.91790771484375, 146.23204040527344, 48.084983825683594, 837.7454833984375, 855.2996215820312, 439.1423645019531, -5.12847900390625, 68.58503723144531, 266.458740234375, -227.41864013671875, -208.2615966796875, -61.40576171875, -15.316600799560547, -187.58421325683594, 188.3025360107422, 270.2403564453125, 276.55841064453125, 160.41738891601562, 371.8382263183594, 451.564697265625, 248.51565551757812, 451.12432861328125, 229.18080139160156, 221.04330444335938, 123.8943862915039, 301.695068359375, -23.069168090820312, 14.379959106445312, 142.99855041503906, 534.5071411132812, -100.55814361572266, -142.52249145507812, 432.043701171875, 316.50555419921875, -174.51986694335938, -2.7930679321289062, 94.8415756225586, 90.34894561767578, 212.5391082763672, -113.63655853271484, -116.97138214111328, -169.61236572265625, 365.9390563964844, 286.2906494140625, -252.09078979492188, 224.14404296875, 69.8565444946289, -95.22859954833984, 1204.4541015625, 621.6331787109375, -44.97821044921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000393.npy"}
|
|
{"epoch": 0.5770925110132159, "step": 394, "batch_size": 64, "mean": 195.913330078125, "std": 245.18992614746094, "min": -238.8067626953125, "p10": -75.72322921752928, "median": 131.7893180847168, "p90": 532.7522277832031, "max": 910.23681640625, "pos_frac": 0.765625, "sample": [117.91374206542969, 302.0465087890625, -187.51072692871094, 343.6026916503906, 183.8067626953125, -32.76678466796875, 127.43003845214844, 190.3203887939453, 533.7696533203125, 68.31268310546875, 515.3343505859375, 38.07689666748047, -238.8067626953125, 50.44013977050781, -50.463348388671875, -175.21170043945312, -137.86383056640625, 640.031005859375, 488.0699768066406, -204.63291931152344, 465.880859375, -128.165283203125, 281.27154541015625, 309.0726318359375, 135.74729919433594, 266.043212890625, 234.2632598876953, 40.25691223144531, -57.92424774169922, 530.3782348632812, 502.34906005859375, 599.942626953125, -83.35136413574219, 409.4112548828125, 123.79037475585938, 45.35298156738281, 127.83133697509766, -18.79553985595703, 89.60089111328125, 579.650390625, 6.866828918457031, 407.2560729980469, -47.62217330932617, 154.84095764160156, 109.77323913574219, -42.90826416015625, 454.8561096191406, 559.6656494140625, 217.2659454345703, 124.15493774414062, 391.231201171875, 244.5155487060547, 910.23681640625, 629.6441650390625, 290.9810791015625, 56.66613006591797, 63.79759216308594, -33.373531341552734, 105.4295425415039, 380.263427734375, 206.12057495117188, -42.82243347167969, 114.03426361083984, 253.07388305664062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000394.npy"}
|
|
{"epoch": 0.57856093979442, "step": 395, "batch_size": 64, "mean": 205.683837890625, "std": 269.38922119140625, "min": -449.3671569824219, "p10": -68.11011123657225, "median": 172.16547393798828, "p90": 589.2660156250003, "max": 842.2890014648438, "pos_frac": 0.765625, "sample": [420.4974060058594, -153.627197265625, 163.25341796875, 64.35023498535156, 157.63760375976562, 303.30560302734375, 306.4285888671875, 247.94564819335938, -260.0373229980469, 202.4142303466797, -34.70976257324219, 340.3180236816406, 395.5809020996094, 17.185523986816406, 66.4770736694336, 9.434520721435547, 523.1445922851562, 269.2588195800781, 705.5535888671875, -48.67102813720703, 534.4451293945312, -13.330131530761719, 473.98101806640625, -23.892318725585938, 158.43014526367188, 46.274261474609375, 251.78814697265625, 451.0751037597656, -36.207054138183594, 19.518821716308594, 767.1883544921875, 842.2890014648438, 246.72511291503906, 93.76332092285156, 248.35006713867188, 13.218017578125, 134.74237060546875, 181.07752990722656, 684.624755859375, 203.5211181640625, 400.8934326171875, 415.3439025878906, -449.3671569824219, -89.40919494628906, -181.55545043945312, 270.8157958984375, -29.968521118164062, 668.3778076171875, -36.58214569091797, -76.44114685058594, 86.48736572265625, -200.53256225585938, 42.962120056152344, 805.2130126953125, 612.7606811523438, 254.1957244873047, 191.38528442382812, 398.3985290527344, 140.7653045654297, 402.7583923339844, 424.11444091796875, 51.43995666503906, 121.25273895263672, -32.86630630493164], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000395.npy"}
|
|
{"epoch": 0.580029368575624, "step": 396, "batch_size": 64, "mean": 194.87359619140625, "std": 211.51609802246094, "min": -347.803955078125, "p10": -11.895048904418932, "median": 167.4471435546875, "p90": 427.65292358398443, "max": 881.10009765625, "pos_frac": 0.890625, "sample": [183.29100036621094, 415.9134826660156, 62.825889587402344, 183.19387817382812, 111.77699279785156, 490.60009765625, 119.9103775024414, 101.53167724609375, 119.24102020263672, 211.65475463867188, 313.2381896972656, 249.55606079101562, 186.33847045898438, 351.9453430175781, -112.51995849609375, 178.82798767089844, 175.53884887695312, 81.03605651855469, 56.24876403808594, 188.5859375, 132.18521118164062, 356.72344970703125, 822.792724609375, 273.1612243652344, 161.14523315429688, 369.5210266113281, 310.01727294921875, 163.06634521484375, 73.94770050048828, 559.7803344726562, -94.48954772949219, 347.7259216308594, 1.4558334350585938, 258.6771545410156, 62.52035140991211, 152.1515350341797, 257.19970703125, -84.16715240478516, 353.15655517578125, 569.4974365234375, 293.3126220703125, 27.662017822265625, -95.5614013671875, 432.6841125488281, 32.87261962890625, -17.61685562133789, 360.1500549316406, 13.766975402832031, 104.1619873046875, 196.36380004882812, 652.4765014648438, 61.248756408691406, 69.830078125, -94.14382934570312, 74.97892761230469, 210.2714080810547, 293.9405822753906, 159.43728637695312, 31.147499084472656, -347.803955078125, 127.30143737792969, 881.10009765625, 171.82794189453125, 117.69821166992188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000396.npy"}
|
|
{"epoch": 0.5814977973568282, "step": 397, "batch_size": 64, "mean": 249.26467895507812, "std": 221.98751831054688, "min": -289.2301025390625, "p10": -4.850967407226539, "median": 238.56871795654297, "p90": 512.5455505371094, "max": 777.7557373046875, "pos_frac": 0.890625, "sample": [23.597171783447266, 451.1710205078125, 17.7374267578125, 312.5832824707031, 126.0287857055664, 121.56155395507812, 198.5831298828125, 240.0655517578125, 445.05712890625, 143.1058807373047, 400.1218566894531, 110.73088073730469, 454.904541015625, -14.531707763671875, 110.5087890625, 447.8410949707031, 283.6368408203125, -80.96612548828125, 367.52142333984375, 511.311279296875, 301.41876220703125, 513.0745239257812, 282.5975341796875, 341.341552734375, 777.7557373046875, 39.81947326660156, 548.5255126953125, 437.70611572265625, -98.23654174804688, 165.3502960205078, 503.4580078125, 177.3349151611328, 106.90432739257812, 670.47119140625, 396.811767578125, 251.95703125, 38.7828254699707, 274.647216796875, 172.947509765625, 177.11721801757812, -289.2301025390625, -126.22765350341797, 71.43822479248047, -132.67767333984375, 446.8208923339844, 153.63844299316406, 24.57605743408203, 178.90576171875, -27.12955093383789, 463.04241943359375, 40.38444137573242, 573.2791137695312, 282.2226257324219, 765.0499877929688, 152.67916870117188, 666.569580078125, 127.34877014160156, 130.85494995117188, 429.34130859375, 162.07200622558594, 238.32177734375, 297.56121826171875, 334.927978515625, 238.81565856933594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000397.npy"}
|
|
{"epoch": 0.5829662261380323, "step": 398, "batch_size": 64, "mean": 211.71212768554688, "std": 241.58135986328125, "min": -282.0902099609375, "p10": -29.89179000854492, "median": 152.55103302001953, "p90": 483.7974975585938, "max": 853.07275390625, "pos_frac": 0.84375, "sample": [152.35736083984375, 71.19493103027344, 338.9681701660156, 488.738525390625, -38.26023864746094, 96.41544342041016, 245.42066955566406, -174.51971435546875, 68.60899353027344, 67.76966094970703, 433.01300048828125, -282.0902099609375, 464.41522216796875, 346.56915283203125, 362.98724365234375, 810.6458740234375, 152.7447052001953, 172.9124755859375, 490.7385559082031, 268.33612060546875, 348.52874755859375, 109.65618896484375, 472.2684326171875, 340.5210876464844, 828.4815063476562, 12.17718505859375, 148.2025146484375, 104.77635192871094, 441.810302734375, 272.67724609375, 47.318450927734375, 173.73007202148438, -95.188232421875, -9.941543579101562, 256.2960510253906, 65.51823425292969, 21.42144012451172, 267.16119384765625, 82.95319366455078, 312.33367919921875, 135.8896942138672, 447.54107666015625, 112.84807586669922, 69.7387466430664, 126.59310913085938, 0.30651092529296875, 301.0092468261719, 53.694793701171875, -6.9934234619140625, 124.56515502929688, 167.10739135742188, 853.07275390625, 493.9049377441406, 703.7452392578125, -245.24664306640625, 372.7529296875, 357.100830078125, 437.9483337402344, 433.77337646484375, 83.67105102539062, -26.74981689453125, 24.68719482421875, -31.23834991455078, -177.81378173828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000398.npy"}
|
|
{"epoch": 0.5844346549192364, "step": 399, "batch_size": 64, "mean": 201.39321899414062, "std": 243.2018280029297, "min": -275.90216064453125, "p10": -129.47829666137693, "median": 200.0589370727539, "p90": 491.0795288085938, "max": 821.55615234375, "pos_frac": 0.75, "sample": [216.18728637695312, 215.0109100341797, 82.1393051147461, -105.85235595703125, 414.8080139160156, -81.40511322021484, 548.5814208984375, 69.85833740234375, 133.7017822265625, 59.726165771484375, 242.87850952148438, -134.62985229492188, 482.4254455566406, -39.770233154296875, 494.7884216308594, 136.7281951904297, 310.91717529296875, -117.45800018310547, -55.145606994628906, 66.96804809570312, 450.3963928222656, 80.97892761230469, 336.90911865234375, 212.56443786621094, 469.07958984375, 615.65966796875, 113.0160140991211, 360.4488525390625, -199.0889434814453, 187.55343627929688, 45.58277893066406, 388.6395568847656, 243.83840942382812, 135.50196838378906, 348.6586608886719, -166.76495361328125, 821.55615234375, 404.5701599121094, 555.4910278320312, -135.24916076660156, 418.3480224609375, 362.88800048828125, 158.95965576171875, -275.90216064453125, -43.00143051147461, 358.06121826171875, -5.433286666870117, 560.7301025390625, 123.59581756591797, -169.92611694335938, 369.5043029785156, 731.2887573242188, 382.4449462890625, 177.7687225341797, 402.47479248046875, 215.85853576660156, 139.39883422851562, 241.40408325195312, -176.66232299804688, 252.81954956054688, 414.80389404296875, -71.25151062011719, 178.5582275390625, -67.3642807006836], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000399.npy"}
|
|
{"epoch": 0.5859030837004405, "step": 400, "batch_size": 64, "mean": 284.9051818847656, "std": 237.66815185546875, "min": -376.32025146484375, "p10": -12.671077919006345, "median": 287.54673767089844, "p90": 548.2846191406251, "max": 942.4893798828125, "pos_frac": 0.828125, "sample": [240.47286987304688, 410.71453857421875, 326.38995361328125, 198.9622802734375, 554.967529296875, 39.256378173828125, 410.7335205078125, 221.41799926757812, -50.78759002685547, 491.9295654296875, 202.04638671875, 152.32162475585938, 392.5268249511719, 231.58164978027344, -376.32025146484375, 504.73175048828125, 529.299560546875, 194.8186798095703, 211.2039337158203, 316.27801513671875, 663.282470703125, 476.52056884765625, 200.12155151367188, 380.207763671875, 18.389394760131836, 197.5348663330078, 120.13865661621094, 476.373291015625, 302.05078125, 359.5498046875, 307.705078125, 127.71176147460938, -71.89786529541016, 286.04278564453125, 223.35617065429688, -158.80946350097656, -7.746925354003906, 585.333984375, 305.634033203125, 422.72198486328125, -13.707382202148438, 233.8524169921875, 269.6810302734375, 856.1324462890625, -10.253034591674805, 515.971923828125, 227.44400024414062, 184.73553466796875, -8.013774871826172, 942.4893798828125, 524.161376953125, -9.7666015625, 294.67291259765625, 390.3328857421875, 289.0506896972656, 532.691162109375, -42.63621520996094, 597.990478515625, 513.2593994140625, 326.506103515625, 183.8392333984375, 489.0513000488281, 555.4093017578125, -25.729110717773438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000400.npy"}
|
|
{"epoch": 0.5873715124816447, "step": 401, "batch_size": 64, "mean": 157.9160614013672, "std": 288.9046936035156, "min": -566.7479248046875, "p10": -180.43320465087888, "median": 125.3333740234375, "p90": 626.3911865234376, "max": 777.3399658203125, "pos_frac": 0.703125, "sample": [89.46421813964844, 391.49365234375, -23.860149383544922, 220.4375, 403.9228515625, -10.452066421508789, -343.3404541015625, -321.62554931640625, -566.7479248046875, -28.781780242919922, 93.59661102294922, 88.52987670898438, -187.467041015625, 67.47725677490234, -10.242919921875, -2.6500606536865234, 357.37664794921875, 183.1268310546875, 151.97836303710938, 84.96626281738281, -65.18377685546875, -164.0209197998047, 59.902915954589844, 728.98193359375, 33.346046447753906, 499.25225830078125, 619.505859375, 331.3542175292969, 629.342041015625, 171.1710662841797, 114.41061401367188, 245.81719970703125, -38.580257415771484, 73.8962173461914, -66.99018096923828, -351.22314453125, 680.2992553710938, 26.617847442626953, 243.0229034423828, 353.006591796875, 174.31475830078125, 467.7301940917969, 179.5562286376953, -35.70395278930664, 206.39036560058594, 148.39404296875, 167.8358917236328, -52.14768981933594, 777.3399658203125, -347.56805419921875, 568.903564453125, 382.738525390625, 252.93690490722656, 79.53160858154297, 646.7342529296875, 702.9921875, 57.893226623535156, 279.6740417480469, 136.25613403320312, 31.900577545166016, 208.9779052734375, 650.1100463867188, -151.73855590820312, -187.5554656982422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000401.npy"}
|
|
{"epoch": 0.5888399412628488, "step": 402, "batch_size": 64, "mean": 291.61102294921875, "std": 233.02163696289062, "min": -209.42123413085938, "p10": 37.30738067626954, "median": 250.67058563232422, "p90": 646.2081176757814, "max": 1006.888671875, "pos_frac": 0.921875, "sample": [250.1961212158203, 33.05644989013672, 47.226219177246094, 429.2281494140625, 229.89437866210938, 303.415283203125, 8.285354614257812, 225.8466796875, -1.7276058197021484, 391.2643737792969, 367.671630859375, 479.2466735839844, 552.5299682617188, 203.2544403076172, -67.3714599609375, 478.4983215332031, 48.14781188964844, 202.16395568847656, 220.7300567626953, 340.2115783691406, 478.0856018066406, 214.13546752929688, 520.2136840820312, -91.1677017211914, 167.36865234375, 441.7279357910156, 56.228302001953125, 735.4646606445312, 68.06303405761719, 740.5692138671875, 182.81375122070312, 313.0633544921875, 119.11683654785156, 308.34423828125, 148.4168701171875, 356.4674377441406, 662.1158447265625, 126.1202392578125, 179.4396514892578, 490.401611328125, 66.99272155761719, -18.784526824951172, 728.6865844726562, 722.9957885742188, 181.037109375, 344.6624450683594, 609.090087890625, 154.30844116210938, 251.14505004882812, 1006.888671875, 430.1754150390625, 457.905029296875, 91.6185302734375, 671.468017578125, 121.43426513671875, 344.59771728515625, 296.8013916015625, 193.71481323242188, 173.82803344726562, -209.42123413085938, 334.644775390625, 269.7398681640625, 186.07318115234375, 294.7452697753906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000402.npy"}
|
|
{"epoch": 0.5903083700440529, "step": 403, "batch_size": 64, "mean": 221.64427185058594, "std": 233.1367645263672, "min": -270.992431640625, "p10": -82.69343757629393, "median": 190.78439331054688, "p90": 507.38984069824227, "max": 804.8327026367188, "pos_frac": 0.84375, "sample": [186.1795654296875, 135.17404174804688, 51.28681182861328, 156.14495849609375, 341.8973693847656, -90.89118957519531, 195.38922119140625, 369.8179931640625, -106.27434539794922, -116.54792785644531, 123.49044799804688, 31.782318115234375, -270.992431640625, 471.14984130859375, 160.9661102294922, 649.0296630859375, 211.6100311279297, 103.73230743408203, 417.19580078125, 257.219970703125, 317.723388671875, 96.01420593261719, 603.29541015625, 4.225700378417969, -153.3782501220703, 513.043212890625, 408.65771484375, 489.54632568359375, 238.0004425048828, 804.8327026367188, -63.56534957885742, 310.9239807128906, 490.7364501953125, 53.638423919677734, 32.08672332763672, 420.9952697753906, -92.5912094116211, 265.1058654785156, 494.1986389160156, 462.59197998046875, -193.71925354003906, 171.90536499023438, 654.879150390625, 1.8071956634521484, 406.29901123046875, 96.95301818847656, 420.4730529785156, 529.7760620117188, 233.29122924804688, 677.7352905273438, 289.86749267578125, 357.7254638671875, 279.8400573730469, 131.07887268066406, -40.077301025390625, 36.83497619628906, 4.075216293334961, 316.0138854980469, 101.77272033691406, 158.42324829101562, 290.79205322265625, 150.95416259765625, -49.82786560058594, 184.91770935058594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000403.npy"}
|
|
{"epoch": 0.591776798825257, "step": 404, "batch_size": 64, "mean": 229.15975952148438, "std": 265.5217590332031, "min": -364.8903503417969, "p10": -135.88765563964841, "median": 247.04950714111328, "p90": 552.6205810546876, "max": 859.6967163085938, "pos_frac": 0.78125, "sample": [-142.6547393798828, 236.0362548828125, 34.90917205810547, 468.3573913574219, 446.21368408203125, 151.41981506347656, 90.26628112792969, 739.2440185546875, 210.705810546875, 637.7427368164062, 859.6967163085938, 155.6190643310547, 237.9334259033203, 524.1829833984375, -63.39942169189453, -242.37298583984375, 618.7377319335938, 168.5228271484375, -183.9810028076172, 155.29257202148438, -65.33221435546875, 87.83191680908203, 13.66465950012207, 406.6241455078125, 325.6325988769531, 300.2313537597656, -238.7615966796875, 435.93426513671875, 250.2938995361328, 347.6260681152344, 455.21514892578125, -147.84481811523438, 352.29571533203125, 165.04315185546875, 466.6587829589844, 134.23614501953125, -108.19032287597656, 256.83349609375, 275.43878173828125, 403.3438720703125, -364.8903503417969, 342.2595520019531, 126.136474609375, 533.51123046875, 164.28671264648438, -166.04530334472656, -11.866636276245117, 360.7769470214844, 452.2388000488281, -120.09779357910156, 285.2904357910156, 138.82223510742188, 275.2547302246094, 296.5643005371094, 459.05963134765625, 653.9758911132812, 260.3242492675781, 15.913528442382812, -80.9474105834961, -37.434303283691406, 771.4374389648438, 243.80511474609375, 560.810302734375, 287.7912902832031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000404.npy"}
|
|
{"epoch": 0.593245227606461, "step": 405, "batch_size": 64, "mean": 184.99761962890625, "std": 276.7222900390625, "min": -369.8765563964844, "p10": -152.88018646240232, "median": 160.35669708251953, "p90": 586.8504333496094, "max": 852.853271484375, "pos_frac": 0.71875, "sample": [608.8914184570312, 536.8817138671875, 218.78482055664062, -125.10646057128906, -221.78150939941406, -13.1531982421875, -285.58331298828125, 184.91238403320312, 391.7970275878906, -271.62225341796875, 62.016746520996094, 726.3994750976562, 454.6880187988281, 119.09539794921875, 28.391151428222656, 762.2877197265625, -41.94342041015625, -21.93030548095703, 161.77801513671875, -112.34555053710938, 153.9254150390625, 229.3571319580078, 129.93614196777344, 385.046630859375, 156.47442626953125, 158.9353790283203, 85.16148376464844, 332.522216796875, -186.42144775390625, -105.44527435302734, 331.0431823730469, -18.562013626098633, 358.94415283203125, 39.11933135986328, 374.8753662109375, 43.100914001464844, 234.29217529296875, -162.5096893310547, -27.16733741760254, 23.979984283447266, 490.3583679199219, 233.35140991210938, 591.0841064453125, 328.4754333496094, 29.887584686279297, 175.5753936767578, 473.1297912597656, -61.894683837890625, 148.59469604492188, 213.08526611328125, 248.9072723388672, -227.2241668701172, 200.65463256835938, 695.234130859375, 89.7255630493164, -369.8765563964844, 264.07135009765625, 306.0719299316406, -20.391721725463867, 852.853271484375, 666.9672241210938, -130.41134643554688, 365.58038330078125, 576.9718627929688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000405.npy"}
|
|
{"epoch": 0.5947136563876652, "step": 406, "batch_size": 64, "mean": 229.93914794921875, "std": 264.2333984375, "min": -259.5608825683594, "p10": -51.38692855834959, "median": 194.50747680664062, "p90": 560.616015625, "max": 946.5745239257812, "pos_frac": 0.828125, "sample": [386.0277099609375, 142.44602966308594, 231.694580078125, 67.94728088378906, 281.43707275390625, 182.875732421875, 148.47604370117188, -259.5608825683594, 903.978759765625, 163.73208618164062, 387.5679016113281, 183.46461486816406, 946.5745239257812, 501.85504150390625, 288.37066650390625, 0.5542411804199219, 390.8125305175781, 296.6502380371094, 278.1947021484375, -103.0950698852539, 676.0604248046875, 340.3166809082031, 294.819091796875, -232.4038848876953, 79.17350006103516, 153.55931091308594, 57.767333984375, -0.6612892150878906, 377.6802062988281, -61.2021484375, 26.608566284179688, 221.99253845214844, 348.0300598144531, -229.10423278808594, -99.53326416015625, 421.03240966796875, 654.1595458984375, 186.415771484375, 119.69750213623047, -14.500879287719727, 158.24615478515625, 202.59918212890625, 372.5669860839844, -28.48474884033203, 8.646026611328125, 36.243568420410156, 563.705810546875, 256.2826843261719, 27.285491943359375, 139.13232421875, 553.406494140625, 522.3736572265625, 267.615234375, -3.8047637939453125, -213.76898193359375, 16.227115631103516, 681.54736328125, 391.1690673828125, 885.9573364257812, 330.1943359375, 325.0531005859375, 340.4658203125, 109.55485534667969, 33.98059844970703], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000406.npy"}
|
|
{"epoch": 0.5961820851688693, "step": 407, "batch_size": 64, "mean": 159.05093383789062, "std": 257.32049560546875, "min": -716.1044311523438, "p10": -169.9386993408203, "median": 141.77674865722656, "p90": 497.0245269775391, "max": 678.9402465820312, "pos_frac": 0.75, "sample": [-207.4911346435547, 475.28839111328125, 309.6662292480469, -241.72450256347656, -229.8963165283203, -5.977134704589844, 344.4486083984375, -192.33209228515625, -129.89810180664062, 132.625244140625, 389.6429443359375, 151.83055114746094, 264.8558349609375, 581.6093139648438, 550.3761596679688, 6.590822219848633, -12.60506820678711, 437.84259033203125, 85.55689239501953, -52.44438171386719, 310.2274475097656, 39.8486328125, 170.5167999267578, 113.42312622070312, 206.01759338378906, 401.68731689453125, 234.152587890625, 143.7408905029297, 99.69508361816406, 153.66404724121094, -169.92227172851562, 215.0078125, 23.12317657470703, -46.90253448486328, 6.265556335449219, 96.97364807128906, 385.35467529296875, 630.5066528320312, 121.95033264160156, -716.1044311523438, 489.43084716796875, 306.595703125, 262.5216064453125, 29.08606719970703, 407.7030029296875, 133.7013702392578, 51.17144775390625, 678.9402465820312, 353.61785888671875, 206.54071044921875, 544.497802734375, -169.94573974609375, -66.27743530273438, 500.2789611816406, -41.74567413330078, 103.95542907714844, 139.81260681152344, 441.7233581542969, 34.50666809082031, -232.61656188964844, 502.1150207519531, 297.78399658203125, -106.46998596191406, 235.14080810546875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000407.npy"}
|
|
{"epoch": 0.5976505139500734, "step": 408, "batch_size": 64, "mean": 157.62081909179688, "std": 199.70608520507812, "min": -247.26649475097656, "p10": -90.10501708984374, "median": 146.36318969726562, "p90": 421.99759826660164, "max": 644.9171752929688, "pos_frac": 0.75, "sample": [159.21607971191406, 581.5859375, 644.9171752929688, 59.33184814453125, 318.9095458984375, 297.37445068359375, -80.94387817382812, 100.61620330810547, -70.14459228515625, -100.04554748535156, -29.12843132019043, -190.84266662597656, -45.43608856201172, 69.22918701171875, -8.611839294433594, -163.39520263671875, 120.38908386230469, 288.0144958496094, 237.61990356445312, 81.19126892089844, 133.5102996826172, 261.9114685058594, 10.8909912109375, 291.60595703125, -154.7938690185547, 251.93630981445312, 31.151113510131836, 172.96652221679688, 72.47293853759766, 486.44158935546875, 4.387655258178711, 127.69256591796875, 431.5734558105469, -16.671661376953125, -247.26649475097656, -94.03121948242188, 263.7375793457031, 261.5362548828125, 81.63566589355469, 215.75540161132812, 285.2016296386719, 274.4012145996094, 246.53025817871094, 257.010498046875, 190.9167938232422, -175.95913696289062, 536.5651245117188, -10.384805679321289, -20.32540512084961, 42.507232666015625, 131.51715087890625, 298.5697937011719, 127.68311309814453, -35.021575927734375, 160.45346069335938, 260.1328125, 102.26297760009766, 376.44500732421875, 209.5587921142578, 208.22161865234375, 552.4594116210938, 499.1869812011719, 399.6539306640625, 313.8549499511719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000408.npy"}
|
|
{"epoch": 0.5991189427312775, "step": 409, "batch_size": 64, "mean": 172.93191528320312, "std": 210.62550354003906, "min": -285.46722412109375, "p10": -60.560817718505845, "median": 155.71721649169922, "p90": 464.7268280029297, "max": 680.3016967773438, "pos_frac": 0.8125, "sample": [-19.391876220703125, -216.5015869140625, 203.20114135742188, 227.6820068359375, 219.9456024169922, 225.52737426757812, -0.0071506500244140625, 41.212158203125, 77.67254638671875, 157.63108825683594, 470.27313232421875, 389.9876403808594, -194.84408569335938, 97.37728881835938, 336.6370849609375, 128.45721435546875, 192.14175415039062, 253.94920349121094, 150.73623657226562, 561.205322265625, 227.89840698242188, -64.90496063232422, 263.9306640625, 48.95530319213867, 458.0250244140625, -89.2618637084961, 38.20842742919922, 29.9559326171875, 80.30535888671875, 134.9062957763672, 212.00282287597656, 602.5541381835938, 407.0902099609375, -285.46722412109375, 64.3372573852539, 622.903564453125, -42.119224548339844, 86.93724060058594, -204.7610626220703, 71.23487091064453, 467.5990295410156, 64.82328033447266, 388.33111572265625, 25.429088592529297, 244.62429809570312, 257.9656066894531, 74.08965301513672, 57.36621856689453, 680.3016967773438, 336.274658203125, -19.231576919555664, 164.84390258789062, -83.91963195800781, 213.8107452392578, 252.79029846191406, 414.4642028808594, 329.72430419921875, 234.09967041015625, 168.46585083007812, 76.62281799316406, 623.4454345703125, -50.42448425292969, 26.72021484375, 153.8033447265625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000409.npy"}
|
|
{"epoch": 0.6005873715124816, "step": 410, "batch_size": 64, "mean": 188.38836669921875, "std": 212.0001983642578, "min": -370.393310546875, "p10": -50.100121307373044, "median": 198.96470642089844, "p90": 472.1693450927735, "max": 650.2265625, "pos_frac": 0.8125, "sample": [236.12640380859375, 205.45645141601562, -44.563743591308594, 69.37979888916016, 396.23162841796875, 220.30941772460938, 346.0525207519531, 459.0378112792969, 316.4618225097656, 15.790800094604492, 109.00981140136719, 163.45639038085938, 15.857765197753906, 371.557861328125, -79.01959991455078, -87.6817626953125, 462.9356384277344, -321.9490966796875, -33.716373443603516, 261.48797607421875, 292.5846252441406, 553.891845703125, -52.47285461425781, 147.5615692138672, 498.1983642578125, 547.5625610351562, 208.403076171875, -10.06290054321289, 238.43443298339844, 650.2265625, 291.8043518066406, 27.851001739501953, 291.23248291015625, 90.36306762695312, 312.03021240234375, 38.692230224609375, -19.825353622436523, 40.972015380859375, 112.65716552734375, -146.682861328125, 260.93365478515625, 273.2168884277344, 140.59628295898438, 328.07525634765625, 172.7941131591797, -370.393310546875, 140.53057861328125, 175.68125915527344, 476.12664794921875, 247.7044219970703, 98.3358154296875, 67.99689483642578, 340.19403076171875, 320.53985595703125, 218.3005828857422, 625.9966430664062, 191.74293518066406, 378.3970947265625, -232.3740997314453, -7.3447418212890625, 192.47296142578125, 86.5671157836914, 488.469970703125, 246.65162658691406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000410.npy"}
|
|
{"epoch": 0.6020558002936858, "step": 411, "batch_size": 64, "mean": 189.33250427246094, "std": 238.69956970214844, "min": -344.26953125, "p10": -51.57650451660156, "median": 131.37474060058594, "p90": 525.3145782470705, "max": 929.675048828125, "pos_frac": 0.796875, "sample": [245.00387573242188, 31.13311004638672, -25.88640594482422, 91.2109603881836, 319.5263366699219, 128.76376342773438, 65.53488159179688, 188.60369873046875, 290.00244140625, 295.21661376953125, 125.8423080444336, 36.04753875732422, 187.65496826171875, -127.83306884765625, 39.97008514404297, -18.971864700317383, 265.88775634765625, 264.1263732910156, 61.11602783203125, 319.0908508300781, 652.7909545898438, 316.939453125, -52.824432373046875, -17.73731803894043, 81.4627456665039, 130.74932861328125, 56.00343322753906, 253.50401306152344, 124.48654174804688, 809.4049682617188, 65.32931518554688, 331.3580627441406, 635.3563232421875, -83.83910369873047, 396.40032958984375, 15.65753173828125, 237.89024353027344, 319.4393310546875, 544.7392578125, 111.4050064086914, 54.036231994628906, 300.73529052734375, 212.35113525390625, 274.4461669921875, 136.68524169921875, 360.3503723144531, 21.28458023071289, -344.26953125, 929.675048828125, 347.1827087402344, 61.85591506958008, 338.2067565917969, 459.783935546875, 479.9903259277344, 571.143310546875, -134.39288330078125, 26.152976989746094, -151.95823669433594, 132.00015258789062, -48.6646728515625, 582.14453125, -171.81460571289062, -21.972457885742188, -8.228008270263672], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000411.npy"}
|
|
{"epoch": 0.6035242290748899, "step": 412, "batch_size": 64, "mean": 184.47250366210938, "std": 231.1099853515625, "min": -291.82110595703125, "p10": -103.26314849853514, "median": 196.890869140625, "p90": 466.2486083984377, "max": 848.7590942382812, "pos_frac": 0.75, "sample": [-130.22616577148438, 195.89453125, 258.46392822265625, 281.12701416015625, 323.06463623046875, -109.89196014404297, 574.4589233398438, -65.39725494384766, 73.57656860351562, 361.6867370605469, 81.1031494140625, 204.44943237304688, 56.60111999511719, 209.3128662109375, -61.77973937988281, 367.57318115234375, 314.99176025390625, 133.39224243164062, -291.82110595703125, 266.72021484375, 42.49542999267578, -2.0406417846679688, 385.8685607910156, 212.93601989746094, 86.0294189453125, 247.5982208251953, 64.28353881835938, 236.6807861328125, -87.7959213256836, 158.783935546875, -37.726646423339844, 84.7753677368164, 197.88720703125, 152.990478515625, 161.85740661621094, 65.91352844238281, -274.0555419921875, -36.6242561340332, 91.5998764038086, 655.109619140625, 41.424346923828125, -209.04710388183594, 376.45391845703125, 419.6610107421875, 592.96044921875, -61.44087600708008, 290.3144836425781, 655.9578857421875, -3.262664794921875, 278.4883728027344, 318.0710754394531, -1.0671405792236328, 82.41136169433594, 380.3298034667969, 486.2147216796875, 848.7590942382812, -122.38465881347656, 269.1661376953125, 339.3409423828125, -122.75949096679688, 508.04962158203125, 285.28887939453125, 385.1833801269531, 318.2599792480469], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000412.npy"}
|
|
{"epoch": 0.604992657856094, "step": 413, "batch_size": 64, "mean": 178.3287353515625, "std": 229.4633331298828, "min": -619.7184448242188, "p10": -74.78552551269532, "median": 171.2074737548828, "p90": 408.0618560791016, "max": 816.4364624023438, "pos_frac": 0.828125, "sample": [48.029754638671875, 12.254138946533203, 8.502012252807617, 220.33700561523438, 154.7972412109375, -75.82142639160156, 90.92562866210938, 28.572959899902344, 65.47510528564453, 202.98587036132812, 390.3245544433594, 747.0469970703125, -0.8909072875976562, 324.1902160644531, -29.66326141357422, -72.36842346191406, 258.35174560546875, 52.428524017333984, 24.665054321289062, 595.9376831054688, 99.98661804199219, 411.64447021484375, 227.25418090820312, -619.7184448242188, 44.93902587890625, 253.1050262451172, 221.51829528808594, 112.20024871826172, 369.43853759765625, 816.4364624023438, 186.6450653076172, 237.38682556152344, 426.86285400390625, 319.0003662109375, -98.07101440429688, 393.66180419921875, 319.4482727050781, 73.37796020507812, 301.5750732421875, 349.0982666015625, 82.50022888183594, 112.31440734863281, 421.10150146484375, -79.13034057617188, 399.7024230957031, 391.7938232421875, 242.27197265625, -115.61996459960938, -177.69749450683594, -320.89013671875, 162.3256378173828, 398.18902587890625, 330.7737731933594, 393.1556396484375, 7.004222869873047, 173.42030334472656, 137.53213500976562, 431.8422546386719, 168.99464416503906, 253.7618408203125, 153.97085571289062, 126.67860412597656, 273.4124755859375, -46.238800048828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000413.npy"}
|
|
{"epoch": 0.6064610866372981, "step": 414, "batch_size": 64, "mean": 229.0056610107422, "std": 228.24557495117188, "min": -556.7830200195312, "p10": -9.586811256408689, "median": 213.16127014160156, "p90": 522.4053771972657, "max": 779.6090087890625, "pos_frac": 0.875, "sample": [351.943359375, 216.89476013183594, 256.4530944824219, 199.0746612548828, 42.554386138916016, 680.3758544921875, 195.65020751953125, -556.7830200195312, 128.2919464111328, 232.2510986328125, 234.2772216796875, 209.83184814453125, 425.19512939453125, 726.3087158203125, 182.41290283203125, -140.37025451660156, 321.85174560546875, 96.51644897460938, -39.41631317138672, 83.10330963134766, 251.3572998046875, 351.39801025390625, 21.051250457763672, 216.49069213867188, 422.3026123046875, 146.4677276611328, 147.10009765625, -7.83049201965332, 331.720947265625, 89.61885070800781, 525.0582275390625, 363.8759765625, -10.339519500732422, 270.6767578125, 630.1488037109375, 134.75039672851562, 321.9271240234375, -21.562515258789062, -158.4909210205078, 338.4535217285156, 206.5810089111328, 198.17410278320312, 516.2153930664062, -127.4288330078125, 74.24932861328125, 423.7392883300781, 28.943008422851562, 500.4256286621094, 156.22900390625, 59.3297233581543, 290.2569580078125, 208.20228576660156, 156.24545288085938, 176.3185577392578, 227.47213745117188, 255.0824737548828, 652.4468994140625, 12.728179931640625, 335.5527038574219, 606.1497802734375, 779.6090087890625, 264.29071044921875, 113.15229797363281, 331.80462646484375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000414.npy"}
|
|
{"epoch": 0.6079295154185022, "step": 415, "batch_size": 64, "mean": 165.65188598632812, "std": 220.51922607421875, "min": -427.8155212402344, "p10": -54.12539863586425, "median": 158.91706085205078, "p90": 456.6364776611329, "max": 736.77392578125, "pos_frac": 0.78125, "sample": [204.38543701171875, 6.64825439453125, 308.1751403808594, 630.537109375, -39.52061462402344, 516.031005859375, 437.7096252441406, 145.5964813232422, -44.70610809326172, 216.84751892089844, 426.3354187011719, 275.8307189941406, 375.755859375, 333.8442077636719, 13.967567443847656, -55.93305587768555, 223.84417724609375, -49.90753173828125, 264.35394287109375, 126.04179382324219, 7.0628662109375, 83.40591430664062, 14.151947021484375, 432.8252258300781, -78.7898178100586, 45.78534698486328, 619.0611572265625, -27.743751525878906, 159.48211669921875, 464.74798583984375, -32.27299499511719, 284.1606140136719, 188.81619262695312, 63.273223876953125, 736.77392578125, -223.71238708496094, -139.86131286621094, 168.13829040527344, 344.3082275390625, 496.0694274902344, 94.8876953125, 288.734130859375, 80.99163818359375, 76.03448486328125, -18.193912506103516, 158.3520050048828, 0.7328128814697266, 174.9580841064453, 413.2541809082031, -77.50760650634766, 96.48541259765625, 346.2962646484375, 183.50526428222656, 29.0091552734375, 162.29486083984375, -24.627546310424805, 177.04757690429688, 7.794952392578125, 305.80609130859375, -183.99478149414062, 511.9245910644531, 55.16315841674805, -427.8155212402344, 249.0675048828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000415.npy"}
|
|
{"epoch": 0.6093979441997063, "step": 416, "batch_size": 64, "mean": 258.1584167480469, "std": 216.19906616210938, "min": -145.72512817382812, "p10": -35.355855178833, "median": 248.82337951660156, "p90": 535.2752868652344, "max": 804.464111328125, "pos_frac": 0.859375, "sample": [567.4931640625, 193.70706176757812, 97.926513671875, 352.49102783203125, -50.19744110107422, 531.6257934570312, 156.1683349609375, 111.17498016357422, 394.98419189453125, 660.4819946289062, -45.064552307128906, 95.7867431640625, 42.98700714111328, 299.34063720703125, 239.76663208007812, 387.43487548828125, 347.60272216796875, 36.892662048339844, 453.85595703125, 162.13282775878906, 324.9491271972656, 318.2102966308594, 59.00654220581055, 193.49966430664062, 429.9230651855469, 240.5821533203125, 215.89181518554688, 591.4561157226562, 50.202545166015625, 507.5305480957031, 255.3798370361328, 425.93768310546875, 804.464111328125, -67.1534194946289, -145.72512817382812, 88.3145980834961, 505.2328796386719, 373.22467041015625, 56.73174285888672, 573.4525146484375, 58.244895935058594, 778.1515502929688, 220.09637451171875, 182.97618103027344, 357.4036865234375, 364.1043701171875, 283.25653076171875, 447.57757568359375, 168.15975952148438, 69.01062774658203, -51.73079299926758, 242.2669219970703, -82.28823852539062, 468.1070556640625, 338.8328552246094, -35.873504638671875, 389.09307861328125, -34.148006439208984, 289.7395935058594, -19.115890502929688, 120.92333221435547, 301.2897033691406, 291.520263671875, 536.83935546875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000416.npy"}
|
|
{"epoch": 0.6108663729809104, "step": 417, "batch_size": 64, "mean": 207.35751342773438, "std": 191.4608154296875, "min": -185.775634765625, "p10": -32.52237167358396, "median": 208.1901092529297, "p90": 487.16687316894564, "max": 686.2222290039062, "pos_frac": 0.875, "sample": [86.5352554321289, 61.03547668457031, 0.5300540924072266, 196.76683044433594, 28.259822845458984, 380.0688171386719, 4.1098785400390625, 165.39071655273438, -43.119529724121094, 589.2743530273438, 7.606784820556641, 408.4781188964844, 242.33230590820312, 87.8778076171875, 171.5231170654297, -7.7956695556640625, -108.03428649902344, 558.1532592773438, 583.9754028320312, 97.56796264648438, 575.3231201171875, 238.39724731445312, 526.0616455078125, 79.8004150390625, 387.22515869140625, 686.2222290039062, 314.85455322265625, 206.37893676757812, 230.20718383789062, 299.51995849609375, 141.09390258789062, 128.6260528564453, -56.00531768798828, 133.43991088867188, 242.340087890625, 351.00311279296875, 87.71923828125, 192.86390686035156, 329.09771728515625, 268.48907470703125, 51.61800765991211, 238.92759704589844, 218.24244689941406, 286.4834289550781, -181.26242065429688, 194.9017333984375, 210.00128173828125, 133.82052612304688, 295.11663818359375, 212.20460510253906, 375.4823303222656, -185.775634765625, 215.37051391601562, 231.65936279296875, 101.68192291259766, -133.36587524414062, 156.545654296875, 145.31707763671875, 520.890625, 306.62188720703125, 390.2479248046875, 279.04473876953125, 383.54718017578125, -49.63506317138672], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000417.npy"}
|
|
{"epoch": 0.6123348017621145, "step": 418, "batch_size": 64, "mean": 216.85317993164062, "std": 262.18505859375, "min": -315.1510925292969, "p10": -90.2038475036621, "median": 174.4568328857422, "p90": 542.961755371094, "max": 1015.75732421875, "pos_frac": 0.796875, "sample": [161.7833251953125, -99.73185729980469, 234.933837890625, -27.86490249633789, 150.25978088378906, 725.4087524414062, 48.56102752685547, 70.39166259765625, 478.3592529296875, 389.0158386230469, 47.4857177734375, -92.36099243164062, 152.82049560546875, 108.1905288696289, 756.7567749023438, 338.727294921875, 405.45526123046875, -182.0949249267578, 389.3844909667969, 171.22189331054688, -110.45016479492188, 421.0019836425781, -26.651092529296875, 385.8592224121094, 177.6917724609375, 40.08357238769531, 268.43963623046875, 566.2742309570312, 319.7095947265625, 205.96658325195312, 459.0106201171875, 58.37449645996094, 123.30469512939453, 488.56597900390625, -315.1510925292969, 43.63074493408203, -85.1705093383789, 194.31658935546875, 238.28721618652344, 387.1595458984375, 1015.75732421875, -40.370323181152344, 96.34323120117188, 112.27944946289062, 284.7247009277344, 194.9721221923828, 103.87065887451172, 146.94371032714844, -217.96339416503906, -193.27626037597656, 328.736328125, 150.6920928955078, 474.0478210449219, -34.27607727050781, 583.2825317382812, 227.43069458007812, 439.83038330078125, 834.5460205078125, 69.13433074951172, -23.095996856689453, 370.8559265136719, 635.79296875, 247.38784790039062, 4.0000457763671875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000418.npy"}
|
|
{"epoch": 0.6138032305433186, "step": 419, "batch_size": 64, "mean": 233.06138610839844, "std": 233.52256774902344, "min": -321.26837158203125, "p10": -45.60826263427733, "median": 221.98600006103516, "p90": 557.1385559082033, "max": 756.9490966796875, "pos_frac": 0.859375, "sample": [607.858154296875, -321.26837158203125, 243.34573364257812, 399.0084228515625, 156.59596252441406, 453.216064453125, -88.1953125, -68.61662292480469, 10.0150146484375, 516.9435424804688, 580.6478271484375, 162.47677612304688, 204.97622680664062, 409.77813720703125, 305.7403869628906, 322.5901794433594, 95.72134399414062, 118.47409057617188, -273.36419677734375, 92.01061248779297, 10.389015197753906, 574.364990234375, -121.56703186035156, 221.32467651367188, 270.9874572753906, -193.27801513671875, 121.55772399902344, 107.0468521118164, 453.6547546386719, 689.3164672851562, 274.4696044921875, 147.88644409179688, 354.98712158203125, 304.41668701171875, 254.6877899169922, 128.09286499023438, 126.33857727050781, 79.604736328125, 493.1318664550781, 288.741455078125, 490.4521789550781, 173.42892456054688, 508.4589538574219, 108.78569030761719, 56.79521942138672, 222.64732360839844, 48.16558837890625, -33.875946044921875, 585.6058349609375, 452.8262634277344, 501.7298278808594, 260.83807373046875, -5.197296142578125, 756.9490966796875, 324.4169921875, 364.3610534667969, 350.9508056640625, 117.04204559326172, 83.20846557617188, 613.9310302734375, -50.63639831542969, 7.1751708984375, 123.47124481201172, 340.2911071777344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000419.npy"}
|
|
{"epoch": 0.6152716593245228, "step": 420, "batch_size": 64, "mean": 161.28619384765625, "std": 265.73040771484375, "min": -401.95697021484375, "p10": -140.8005157470703, "median": 117.0654067993164, "p90": 470.84471435546874, "max": 895.1903686523438, "pos_frac": 0.71875, "sample": [-202.52719116210938, 169.63284301757812, 165.40255737304688, 84.6998291015625, 470.1761169433594, -88.67012786865234, -39.501731872558594, -9.73193359375, 233.57737731933594, 505.7431335449219, -42.08052062988281, 51.216651916503906, -29.010059356689453, 82.78759765625, -41.76025390625, 95.76602935791016, 350.2474060058594, 276.1530456542969, 78.74284362792969, 179.95864868164062, 21.945432662963867, 279.57806396484375, -148.385498046875, 330.5237731933594, -19.089996337890625, 123.09011840820312, 82.46751403808594, 372.1612548828125, 74.31551361083984, 4.788726806640625, 338.5852966308594, 373.69189453125, 348.10028076171875, -145.15626525878906, 135.3579559326172, 221.48211669921875, -130.63710021972656, 7.990577697753906, 471.1312561035156, 615.7178955078125, -38.796775817871094, -228.2362060546875, 540.4109497070312, 334.0857849121094, 412.7279052734375, 135.16720581054688, 419.6351623535156, 387.0194091796875, 335.3377685546875, 264.0943908691406, 67.0649642944336, -401.3806457519531, 134.8079071044922, 144.47483825683594, -28.300251007080078, 895.1903686523438, -401.95697021484375, 52.67500686645508, 111.04069519042969, 70.06220245361328, 879.9382934570312, -25.39592742919922, -199.87869262695312, 814.0482177734375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000420.npy"}
|
|
{"epoch": 0.6167400881057269, "step": 421, "batch_size": 64, "mean": 163.34597778320312, "std": 212.4307861328125, "min": -420.3016662597656, "p10": -144.75990295410156, "median": 201.9530792236328, "p90": 410.48419189453125, "max": 690.8939208984375, "pos_frac": 0.78125, "sample": [276.40771484375, 78.49542999267578, 122.53836822509766, -144.76150512695312, 210.49395751953125, 328.1884460449219, 483.2024230957031, -39.34808349609375, 109.40406799316406, 39.18877029418945, 209.664794921875, -11.81442642211914, 409.5380554199219, 501.33953857421875, 373.57147216796875, 380.9896545410156, 270.9091796875, -157.8898468017578, 160.5921630859375, 195.75784301757812, 547.363037109375, 386.05328369140625, 79.75238037109375, -179.18722534179688, 225.85850524902344, 119.07243347167969, 298.4464111328125, 239.92855834960938, 63.64112091064453, 208.1483154296875, 410.8896789550781, 280.345458984375, 449.58416748046875, 236.495361328125, 223.2678680419922, -144.75616455078125, 264.157470703125, 268.7830810546875, -299.1265563964844, 498.0738220214844, 97.4416275024414, -19.23470115661621, 244.8084259033203, 690.8939208984375, 277.0970153808594, 276.940673828125, 144.49319458007812, 24.484983444213867, 24.58945655822754, 311.770263671875, 85.78500366210938, -40.54888153076172, 258.92828369140625, -420.3016662597656, -194.2555389404297, 100.03584289550781, 90.631591796875, -56.33699417114258, -47.78814697265625, 225.72869873046875, 104.01139068603516, -190.67515563964844, 172.576904296875, 319.8065185546875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000421.npy"}
|
|
{"epoch": 0.618208516886931, "step": 422, "batch_size": 64, "mean": 182.31634521484375, "std": 225.27139282226562, "min": -400.1595764160156, "p10": -87.62969284057617, "median": 187.14364624023438, "p90": 468.78415527343765, "max": 761.2682495117188, "pos_frac": 0.84375, "sample": [162.2711639404297, 529.82421875, 617.1503295898438, 336.59527587890625, 246.6316680908203, -112.05513000488281, -380.6671142578125, 484.4881591796875, 143.82305908203125, 761.2682495117188, -136.17494201660156, 192.46954345703125, 92.86436462402344, 45.236480712890625, 285.19775390625, 126.98892211914062, 94.8427963256836, -87.04232025146484, 14.281986236572266, 189.9174346923828, 99.02526092529297, 577.2078247070312, 235.4099884033203, 37.46600341796875, 383.70892333984375, 226.15243530273438, 116.52398681640625, 508.1343688964844, 121.52010345458984, 432.1414794921875, 596.0000610351562, 313.44976806640625, 205.9674072265625, 137.4839630126953, 210.2654571533203, 239.84629821777344, 283.19329833984375, -303.11474609375, 138.53390502929688, 13.15435791015625, -28.442447662353516, 260.0517578125, 144.31585693359375, -400.1595764160156, -34.17977523803711, 419.62420654296875, 323.8260498046875, 150.77554321289062, 404.4901123046875, 40.67179870605469, 383.86993408203125, 184.36985778808594, 43.151611328125, 43.95882034301758, 307.1962890625, 281.1900939941406, 255.5843505859375, -162.2459259033203, 279.2085266113281, 340.0163269042969, -87.88142395019531, 124.59262084960938, 194.23956298828125, 20.039749145507812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000422.npy"}
|
|
{"epoch": 0.6196769456681351, "step": 423, "batch_size": 64, "mean": 206.6036376953125, "std": 212.82229614257812, "min": -366.338623046875, "p10": -34.38019409179687, "median": 179.33145904541016, "p90": 493.6349304199219, "max": 772.323974609375, "pos_frac": 0.8125, "sample": [121.12903594970703, 102.60079193115234, 34.38017272949219, 111.2269287109375, -59.294769287109375, -189.53237915039062, 603.4102783203125, 66.21804809570312, 420.6177673339844, 324.8867492675781, 445.1166076660156, -58.53472137451172, 434.6197814941406, 228.52532958984375, 362.0934143066406, -6.841266632080078, 575.8760375976562, 496.22735595703125, 534.194580078125, 516.158447265625, -13.32305908203125, 28.22320556640625, 335.1087646484375, 124.58477020263672, 447.9320068359375, 171.95071411132812, 143.6795196533203, -20.15423583984375, -366.338623046875, -37.989349365234375, 249.887939453125, 31.375995635986328, 305.9186706542969, 298.83685302734375, 525.3795166015625, 201.60606384277344, 42.34393310546875, 381.12799072265625, -58.66644287109375, 106.45639038085938, 203.56417846679688, 487.5859375, 201.68399047851562, 172.6684112548828, 772.323974609375, 140.73890686035156, 281.0733947753906, 5.009300231933594, 279.9647216796875, 102.13938903808594, -7.333179473876953, 205.36155700683594, 333.23199462890625, 299.6603698730469, 185.9945068359375, 229.9988250732422, -25.958831787109375, 171.6629638671875, 122.14138793945312, -57.493865966796875, 476.3123779296875, 93.05606842041016, 159.58358764648438, 428.6439208984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000423.npy"}
|
|
{"epoch": 0.6211453744493393, "step": 424, "batch_size": 64, "mean": 187.33949279785156, "std": 236.42970275878906, "min": -275.382080078125, "p10": -81.33141059875486, "median": 186.64505004882812, "p90": 394.0647857666016, "max": 900.341796875, "pos_frac": 0.796875, "sample": [-225.68295288085938, 315.56109619140625, 25.317031860351562, 0.07841873168945312, -40.731719970703125, 358.039794921875, 336.27032470703125, 242.2496337890625, 55.620216369628906, -238.50726318359375, 280.81146240234375, 200.64205932617188, 69.32502746582031, -90.74764251708984, 271.8819885253906, 606.045654296875, 304.63739013671875, 568.6815185546875, 100.95347595214844, 359.87744140625, 392.3533020019531, -1.8661632537841797, 376.09228515625, 362.72576904296875, 257.0683288574219, -106.0720443725586, 101.0577392578125, 164.46237182617188, -148.08596801757812, -22.87509536743164, 10.076894760131836, 900.341796875, -59.36020278930664, 186.77493286132812, 394.79827880859375, -28.042236328125, 318.4796142578125, -179.5789337158203, 115.70648956298828, 208.32907104492188, -7.886713027954102, 119.3745346069336, -275.382080078125, 40.715030670166016, 189.6011962890625, 186.51516723632812, 31.295188903808594, 73.9922866821289, 199.17306518554688, 819.6334228515625, 218.59426879882812, 375.42431640625, 338.1776428222656, 207.8763427734375, 310.90802001953125, 47.00527572631836, 592.533447265625, 81.26168060302734, 120.89985656738281, 715.1683349609375, 286.2101745605469, 305.1669616699219, 147.4990997314453, 123.26141357421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000424.npy"}
|
|
{"epoch": 0.6226138032305433, "step": 425, "batch_size": 64, "mean": 164.8294677734375, "std": 205.5462188720703, "min": -377.1182861328125, "p10": -101.82470855712889, "median": 165.99496459960938, "p90": 400.8649047851563, "max": 593.654052734375, "pos_frac": 0.8125, "sample": [-42.954185485839844, -59.80000305175781, 349.4775695800781, -216.98028564453125, -310.29022216796875, 203.84091186523438, 100.58306884765625, 258.2539367675781, 58.261199951171875, 254.5055694580078, 58.78009796142578, 394.1845703125, -377.1182861328125, 9.347576141357422, 332.94915771484375, 123.65252685546875, 120.83792877197266, 26.5360107421875, 371.45977783203125, -49.06111145019531, 296.9585876464844, 142.96466064453125, 61.84427261352539, 315.89776611328125, 468.1810302734375, -39.42858123779297, 340.91534423828125, 60.32331848144531, 46.9827995300293, 566.1710205078125, 106.86002349853516, 421.1988525390625, 371.9315490722656, -86.01390075683594, 202.12567138671875, 149.14578247070312, 403.7279052734375, 139.8774871826172, 115.19821166992188, 593.654052734375, -165.7838134765625, 307.14080810546875, 89.00096130371094, -176.83120727539062, 306.921630859375, 252.8561248779297, 226.91429138183594, 182.84414672851562, 21.193252563476562, 364.35491943359375, 360.3548583984375, 215.087890625, -108.60076904296875, 3.43878173828125, 146.03659057617188, 202.6988983154297, 384.2222595214844, 521.5699462890625, 407.98394775390625, 303.4122009277344, 219.12228393554688, -113.57067108154297, 206.31182861328125, 107.42499542236328], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000425.npy"}
|
|
{"epoch": 0.6240822320117474, "step": 426, "batch_size": 64, "mean": 201.99559020996094, "std": 214.2150421142578, "min": -378.427490234375, "p10": -19.00898475646971, "median": 144.0979766845703, "p90": 517.5415924072266, "max": 652.2062377929688, "pos_frac": 0.859375, "sample": [241.76870727539062, 43.148597717285156, 78.8017349243164, 134.799072265625, 153.47665405273438, 104.41435241699219, 124.380126953125, 210.5649871826172, 586.5091552734375, 189.83091735839844, 121.77253723144531, 132.6566162109375, 143.15023803710938, -0.36656761169433594, 466.7692565917969, -56.37637710571289, 114.66259765625, 161.5386962890625, 357.0992126464844, 79.12493896484375, -81.8752212524414, 623.8302612304688, -41.12397003173828, 138.05532836914062, 100.30430603027344, 80.83145141601562, 54.91572189331055, 112.89444732666016, 640.050537109375, 328.4928894042969, 247.9886016845703, 535.8734741210938, 71.60967254638672, 145.04571533203125, 159.54844665527344, 529.6880493164062, 11.02142333984375, -271.9879150390625, 127.9483871459961, 387.1741943359375, 3.7354736328125, 292.49078369140625, 652.2062377929688, 211.72491455078125, 352.30303955078125, 84.71532440185547, 266.8752136230469, 460.63763427734375, -25.24481201171875, 107.90003967285156, -85.6236801147461, 341.0885314941406, -4.458721160888672, 306.7419738769531, 374.57464599609375, 293.3394775390625, 22.01137351989746, 418.6292724609375, 489.1998596191406, -378.427490234375, 479.83148193359375, 532.821533203125, 52.225860595703125, 390.407958984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000426.npy"}
|
|
{"epoch": 0.6255506607929515, "step": 427, "batch_size": 64, "mean": 143.31716918945312, "std": 184.3173370361328, "min": -240.8046875, "p10": -57.30793228149413, "median": 114.63858032226562, "p90": 374.3894836425781, "max": 695.3132934570312, "pos_frac": 0.8125, "sample": [149.48385620117188, 41.12235641479492, 217.60739135742188, 362.2636413574219, 548.43017578125, 142.19192504882812, 69.67022705078125, -41.10944366455078, 247.202880859375, 211.09683227539062, -31.9134521484375, 126.66836547851562, 80.4890365600586, 65.16877746582031, 120.96170043945312, 236.9717559814453, 423.6212158203125, 610.6178588867188, 78.4023666381836, -240.8046875, 77.4242172241211, 57.467437744140625, 198.0823974609375, 181.569580078125, 53.62914276123047, 351.7225036621094, 127.7679214477539, 695.3132934570312, 44.056427001953125, 101.34146881103516, -47.687889099121094, 401.1122741699219, 301.66510009765625, -76.69254302978516, 18.035137176513672, 249.26593017578125, 82.64960479736328, 19.71799087524414, 373.71575927734375, 120.87479400634766, 257.0982971191406, 135.85411071777344, 83.19934844970703, 143.98948669433594, -189.56857299804688, 97.3108901977539, 339.73028564453125, 293.847412109375, -183.94793701171875, -124.62114715576172, 397.68798828125, 70.88125610351562, 218.37191772460938, 13.618583679199219, 374.67822265625, -32.516510009765625, 38.94944763183594, 225.23556518554688, 293.75360107421875, 108.4023666381836, -51.21588897705078, -59.91880798339844, 74.8907241821289, -102.58586120605469], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000427.npy"}
|
|
{"epoch": 0.6270190895741556, "step": 428, "batch_size": 64, "mean": 185.44857788085938, "std": 239.77325439453125, "min": -429.95263671875, "p10": -77.16749038696288, "median": 154.33734130859375, "p90": 502.40422973632815, "max": 718.3468627929688, "pos_frac": 0.796875, "sample": [-57.79111862182617, -29.431819915771484, 38.738365173339844, 293.3037109375, 635.6064453125, 248.1439971923828, 83.97184753417969, 422.82928466796875, 353.649169921875, 664.771484375, -112.0006103515625, 47.395660400390625, 391.7969970703125, 75.77871704101562, 418.4099426269531, -112.5245361328125, -425.0500183105469, 558.6829833984375, 51.85139465332031, 410.5124816894531, 344.36981201171875, 151.8739013671875, -18.545957565307617, 72.01589965820312, 341.9217529296875, 551.7542724609375, 252.42095947265625, 215.37887573242188, -63.25766372680664, 230.27906799316406, 125.92302703857422, 199.71807861328125, 495.0491027832031, -81.4135971069336, 264.1617126464844, 40.68766784667969, 310.7411804199219, 315.08148193359375, 35.54567337036133, 13.785770416259766, 100.67926025390625, 121.04280090332031, 30.69353485107422, -69.08234405517578, -203.60592651367188, 94.59971618652344, 12.825515747070312, 156.80078125, 718.3468627929688, 122.8425064086914, 309.1676330566406, 299.65948486328125, 311.0733642578125, -429.95263671875, 387.6859130859375, 238.17848205566406, 505.5564270019531, 576.7423706054688, 256.6978759765625, 134.0425567626953, 488.18597412109375, -80.63255310058594, 63.17106628417969, -32.144630432128906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000428.npy"}
|
|
{"epoch": 0.6284875183553598, "step": 429, "batch_size": 64, "mean": 157.81967163085938, "std": 239.58853149414062, "min": -511.3569030761719, "p10": -123.5286018371582, "median": 166.89752960205078, "p90": 451.4903869628906, "max": 727.109375, "pos_frac": 0.734375, "sample": [126.98238372802734, 117.6329345703125, -206.31568908691406, 258.7217102050781, -136.5032958984375, 405.0792236328125, -38.10436248779297, 37.501792907714844, 168.20147705078125, 237.3223114013672, -83.30686950683594, 233.94869995117188, 21.23784637451172, 175.13002014160156, -120.74794006347656, 249.9105224609375, 101.13143920898438, -116.24040222167969, 532.974365234375, -117.84571838378906, 620.1510009765625, 727.109375, -4.198699951171875, 352.2876892089844, 156.37063598632812, 37.211124420166016, 259.40228271484375, 19.006938934326172, 483.8961181640625, 258.91668701171875, 65.22563171386719, 202.81834411621094, 367.700927734375, -305.813720703125, 2.3014144897460938, -161.6201629638672, -124.7203140258789, 165.5935821533203, 211.00685119628906, 11.569574356079102, 420.6126708984375, 305.9620361328125, 175.65396118164062, -1.251007080078125, 632.5374145507812, 147.05270385742188, 444.8411560058594, -70.88983154296875, 69.28758239746094, -251.43759155273438, 487.6125793457031, 132.23728942871094, 368.38690185546875, 266.22552490234375, -0.38862037658691406, -511.3569030761719, -16.169158935546875, 262.44451904296875, 171.14364624023438, 454.3400573730469, 257.5753479003906, 438.9594421386719, 392.98431396484375, 333.16943359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000429.npy"}
|
|
{"epoch": 0.6299559471365639, "step": 430, "batch_size": 64, "mean": 176.12640380859375, "std": 192.52781677246094, "min": -149.30035400390625, "p10": -31.63998565673826, "median": 144.02594757080078, "p90": 388.70733032226565, "max": 772.5195922851562, "pos_frac": 0.828125, "sample": [255.87649536132812, 135.19866943359375, 128.77536010742188, -87.12344360351562, 514.3214111328125, 291.8175048828125, -12.264984130859375, 174.37118530273438, 73.61627197265625, 335.55120849609375, 51.94253158569336, -5.008949279785156, 51.299781799316406, 153.87008666992188, 215.98680114746094, 392.3558349609375, -121.62353515625, 343.5362548828125, 73.86747741699219, 64.28215026855469, 168.07798767089844, 42.137794494628906, 142.64366149902344, 147.6126708984375, 29.566518783569336, 210.5794677734375, 334.4460754394531, 287.5536804199219, -39.94355773925781, 142.71580505371094, 117.73356628417969, -136.29652404785156, 772.5195922851562, 94.61183166503906, 150.74603271484375, 327.9671936035156, 42.21221160888672, -79.69876861572266, 20.45833396911621, 199.06748962402344, 380.19415283203125, 456.0446472167969, 344.7676696777344, 106.12055206298828, 189.41409301757812, 579.0534057617188, 5.9018402099609375, -40.29461669921875, 330.7052307128906, 70.38987731933594, 103.23643493652344, -2.372905731201172, 365.0600280761719, 171.30393981933594, -0.31908416748046875, 224.82891845703125, 289.25299072265625, -149.30035400390625, 293.9986572265625, 613.6361083984375, 660.5380249023438, 145.33609008789062, 48.690773010253906, 80.54444885253906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000430.npy"}
|
|
{"epoch": 0.631424375917768, "step": 431, "batch_size": 64, "mean": 244.80091857910156, "std": 223.73458862304688, "min": -214.6218719482422, "p10": -12.450842285156249, "median": 238.5251693725586, "p90": 517.0964813232423, "max": 852.2569580078125, "pos_frac": 0.84375, "sample": [-64.35224914550781, 235.0375518798828, -66.49665832519531, 364.7687683105469, 195.82122802734375, 117.64122772216797, 669.47216796875, 131.36911010742188, 242.01278686523438, 427.4416809082031, 182.74468994140625, 329.1615295410156, 28.795854568481445, -13.291946411132812, -136.11184692382812, 96.43260192871094, 10.69921875, -149.9285888671875, 373.5107116699219, -214.6218719482422, 366.03692626953125, 567.1083374023438, 727.8806762695312, 530.3721313476562, 148.28953552246094, 486.1199645996094, 359.60369873046875, 345.458740234375, 218.41001892089844, 316.85992431640625, 558.3104858398438, 296.33770751953125, 228.5616455078125, 115.08406066894531, 425.81402587890625, -208.95278930664062, 852.2569580078125, -10.488265991210938, 251.33279418945312, -3.4863204956054688, 405.3390808105469, 282.30584716796875, 636.0401611328125, -3.5777931213378906, 123.69633483886719, 107.86074829101562, 319.15667724609375, 47.25550079345703, 215.7580108642578, 284.6710205078125, 466.9257507324219, 414.90667724609375, 392.1168518066406, 332.4088134765625, 374.5772399902344, 340.26300048828125, 146.4384307861328, 98.78104400634766, 11.51727294921875, 161.65728759765625, 479.00738525390625, 186.07296752929688, 129.484130859375, 383.579345703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000431.npy"}
|
|
{"epoch": 0.6328928046989721, "step": 432, "batch_size": 64, "mean": 215.7953338623047, "std": 216.55331420898438, "min": -207.41371154785156, "p10": -6.698566436767575, "median": 158.2133560180664, "p90": 496.0015563964844, "max": 1006.3038330078125, "pos_frac": 0.875, "sample": [77.08079528808594, 106.32254028320312, 96.70903015136719, 220.291259765625, 109.33297729492188, 46.23747253417969, 92.84558868408203, 107.54689025878906, 175.85281372070312, -8.129135131835938, 55.79460525512695, 9.77713394165039, 420.3908386230469, 374.7884826660156, -46.33220291137695, 1006.3038330078125, 424.01171875, 255.17788696289062, 57.98817443847656, 194.28265380859375, 82.75289916992188, 21.672637939453125, 396.85205078125, 22.631134033203125, 301.97802734375, 289.7773132324219, 372.28656005859375, 93.82646179199219, 303.86077880859375, 155.44735717773438, 241.57666015625, -3.3605728149414062, -72.43333435058594, 105.22750854492188, -76.14312744140625, 498.45819091796875, 160.97935485839844, -207.41371154785156, 297.3874206542969, 78.83354187011719, 151.31692504882812, -121.39627075195312, 359.3310241699219, 298.37457275390625, 271.090087890625, 373.8174743652344, 464.5370788574219, 403.0889892578125, 322.0455322265625, 130.5315704345703, 117.08708190917969, 455.7114562988281, 59.90534973144531, 490.2694091796875, 127.99201202392578, 101.34030151367188, 640.9228515625, 253.0590362548828, 520.166748046875, 547.397216796875, -128.67849731445312, 510.15631103515625, 39.555503845214844, 582.8094482421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000432.npy"}
|
|
{"epoch": 0.6343612334801763, "step": 433, "batch_size": 64, "mean": 158.52906799316406, "std": 225.87457275390625, "min": -507.59197998046875, "p10": -145.81087493896484, "median": 189.11246490478516, "p90": 413.76783447265626, "max": 565.1204223632812, "pos_frac": 0.75, "sample": [18.976959228515625, 377.3594970703125, 388.5816345214844, 326.21710205078125, 410.6894836425781, -227.81105041503906, 244.76222229003906, -142.9370574951172, 382.2060241699219, 371.54132080078125, -507.59197998046875, 40.74977111816406, 447.8565368652344, 265.3200378417969, -2.2919921875, 125.00410461425781, 260.40557861328125, 252.0341796875, 510.10064697265625, -354.912109375, -17.34528350830078, 415.0871276855469, 370.98553466796875, 311.5660400390625, 196.75161743164062, 189.23907470703125, 48.08128356933594, 166.55441284179688, 71.20066833496094, 366.8295593261719, -20.227386474609375, 478.0411682128906, -158.41464233398438, 202.7042236328125, 113.36961364746094, 387.0235595703125, 391.3375549316406, -153.21414184570312, 204.94354248046875, -4.197227478027344, 565.1204223632812, 454.6247863769531, 202.56613159179688, -126.27806854248047, 1.11053466796875, 85.0667953491211, 258.9241638183594, 52.08466339111328, -86.27662658691406, -147.04251098632812, 557.4326171875, 260.4697570800781, 137.78378295898438, -163.16014099121094, 149.2479248046875, -126.87705993652344, 68.35001373291016, 299.9460144042969, -63.294708251953125, 83.76705169677734, 263.0962219238281, 336.5533447265625, 147.08274841308594, 188.98585510253906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000433.npy"}
|
|
{"epoch": 0.6358296622613803, "step": 434, "batch_size": 64, "mean": 179.4976348876953, "std": 242.32501220703125, "min": -194.51290893554688, "p10": -47.39075012207031, "median": 131.5177459716797, "p90": 436.6409912109375, "max": 1411.492919921875, "pos_frac": 0.796875, "sample": [361.2344970703125, 223.29421997070312, 1411.492919921875, 53.48252487182617, -87.57966613769531, 9.066070556640625, 135.81326293945312, 289.80853271484375, -11.387319564819336, -36.30235290527344, 222.4143829345703, 1.3719329833984375, 73.89501190185547, 438.3163146972656, -25.521221160888672, 65.60546875, -49.56719207763672, -194.51290893554688, 328.9630432128906, 503.38458251953125, 549.361328125, -42.31238555908203, 204.15176391601562, 131.67242431640625, 311.9073181152344, 210.87408447265625, 8.918649673461914, 47.45208740234375, 139.4933624267578, 118.40282440185547, 125.74862670898438, 293.6180725097656, 262.30035400390625, -128.0874481201172, 85.81391906738281, 409.0847473144531, 82.50149536132812, 68.8697509765625, 166.16372680664062, 131.36306762695312, 16.792327880859375, 340.4002380371094, 432.7319030761719, 61.97981262207031, 168.184326171875, 283.03021240234375, 295.6139831542969, 96.38175964355469, 140.64389038085938, 495.16180419921875, 232.14967346191406, -13.447237014770508, 518.9202270507812, 88.00718688964844, 61.74162292480469, -51.939666748046875, 425.7483215332031, 181.88491821289062, 53.126953125, -75.9816665649414, 640.8214111328125, 383.27398681640625, -20.436813354492188, -157.5107879638672], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000434.npy"}
|
|
{"epoch": 0.6372980910425844, "step": 435, "batch_size": 64, "mean": 205.54232788085938, "std": 233.9084930419922, "min": -402.11083984375, "p10": -17.399559211730953, "median": 183.55968475341797, "p90": 528.8649841308595, "max": 809.7113037109375, "pos_frac": 0.828125, "sample": [809.7113037109375, 249.15838623046875, -337.14556884765625, 299.6578063964844, 456.90350341796875, 234.78903198242188, 390.9290771484375, 429.76629638671875, 300.92376708984375, 172.0424346923828, 443.82843017578125, 109.68763732910156, 65.9447250366211, 264.1214904785156, -0.4881439208984375, 498.67547607421875, 693.8714599609375, 590.4623413085938, 217.24456787109375, -13.649215698242188, -8.178558349609375, 19.64373779296875, -13.181961059570312, 84.74116516113281, -145.76821899414062, 118.02717590332031, 156.62513732910156, 218.46969604492188, -149.26388549804688, 170.4927978515625, 237.83889770507812, -19.00684928894043, 164.1085205078125, 114.05459594726562, 321.02227783203125, 322.89508056640625, 330.16162109375, -402.11083984375, 348.74462890625, -226.71192932128906, 127.82977294921875, 159.19961547851562, 541.8033447265625, 26.367935180664062, 141.1366729736328, 210.04415893554688, 550.1712036132812, 392.82952880859375, -49.526607513427734, 7.472776412963867, 666.4154052734375, 41.91157531738281, 211.94479370117188, 103.07003784179688, 245.105224609375, 605.484619140625, 439.9795227050781, 131.172119140625, 143.5579071044922, 313.7596740722656, 120.53837585449219, 53.15989685058594, 257.167724609375, 195.07693481445312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000435.npy"}
|
|
{"epoch": 0.6387665198237885, "step": 436, "batch_size": 64, "mean": 191.78115844726562, "std": 206.94383239746094, "min": -272.9756164550781, "p10": -67.05700912475584, "median": 167.9971923828125, "p90": 459.17563171386723, "max": 719.690185546875, "pos_frac": 0.84375, "sample": [-22.61970329284668, 71.00566101074219, -123.1346435546875, 149.78005981445312, 381.7091369628906, 194.71202087402344, 231.8800048828125, 621.605224609375, 329.9085998535156, -46.687095642089844, 153.65562438964844, 713.9071044921875, 45.505157470703125, -97.58007049560547, 248.21205139160156, 157.9427490234375, -75.78697204589844, 53.285400390625, 565.982177734375, -146.36947631835938, 301.8851318359375, 216.53244018554688, 289.93084716796875, 300.84991455078125, 462.5805969238281, 138.24050903320312, 408.936767578125, 487.8011779785156, 74.93940734863281, 149.68031311035156, 719.690185546875, 54.483131408691406, 451.230712890625, 231.97438049316406, 233.12588500976562, 129.6783447265625, 83.92887878417969, -35.591331481933594, 202.9569091796875, 96.85366821289062, 379.0225830078125, 71.3065185546875, 151.08395385742188, 93.4920883178711, 227.21885681152344, 274.0173645019531, 142.00148010253906, -116.90019989013672, 170.71563720703125, 165.27874755859375, 239.15289306640625, 179.38967895507812, 222.81625366210938, 325.80560302734375, 667.1450805664062, 51.38987731933594, 114.92230987548828, 86.016845703125, -272.9756164550781, 188.9868621826172, -103.05814361572266, 182.02772521972656, 427.76641845703125, 0.750335693359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000436.npy"}
|
|
{"epoch": 0.6402349486049926, "step": 437, "batch_size": 64, "mean": 181.99075317382812, "std": 206.25338745117188, "min": -290.98687744140625, "p10": -86.18462448120115, "median": 179.95606994628906, "p90": 462.4987396240235, "max": 690.6074829101562, "pos_frac": 0.8125, "sample": [146.40838623046875, -188.64990234375, 203.57215881347656, 323.3234558105469, 242.42105102539062, -140.7656707763672, 465.4111328125, 176.8875732421875, 258.89990234375, 5.775970458984375, -290.98687744140625, 161.88929748535156, -90.20597076416016, 455.7031555175781, 52.905609130859375, 118.51539611816406, 254.30630493164062, -212.82870483398438, 78.37176513671875, 479.4265441894531, 48.328025817871094, 380.75653076171875, 158.69056701660156, 96.90824127197266, -103.73667907714844, 513.332763671875, 146.39288330078125, 79.04508209228516, 294.9781494140625, 289.85577392578125, 659.3654174804688, 352.28851318359375, 601.1397705078125, -143.9586944580078, 690.6074829101562, -76.80148315429688, -38.69322967529297, -32.76336669921875, 254.26858520507812, 183.02456665039062, 82.6897201538086, 102.5673828125, 144.99623107910156, 223.2771453857422, 317.41534423828125, 153.036865234375, 373.81195068359375, 353.20867919921875, 390.6224365234375, 292.83978271484375, 76.49183654785156, 299.4815368652344, 246.33082580566406, -2.295257568359375, 20.07135772705078, 478.79962158203125, 228.460693359375, 202.70213317871094, -12.623275756835938, 205.7687225341797, 121.08999633789062, 294.8499450683594, 195.8857421875, 4.519721984863281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000437.npy"}
|
|
{"epoch": 0.6417033773861968, "step": 438, "batch_size": 64, "mean": 217.51998901367188, "std": 196.9735870361328, "min": -206.74404907226562, "p10": -21.449842262268064, "median": 202.7464370727539, "p90": 523.2020385742188, "max": 634.6488647460938, "pos_frac": 0.859375, "sample": [37.7882080078125, 115.50824737548828, 336.49261474609375, -57.67692565917969, 150.03756713867188, 72.11500549316406, 570.2413330078125, -45.04881286621094, 133.38101196289062, 28.4915771484375, 269.3265380859375, 243.1806640625, 23.102413177490234, 96.18035125732422, 345.2843017578125, 361.3699951171875, 246.9268341064453, 634.6488647460938, 149.6622772216797, 260.36578369140625, 146.26663208007812, 518.8272705078125, 342.17425537109375, 520.18798828125, 116.98077392578125, 133.8953857421875, 67.23136138916016, 307.9278564453125, -206.74404907226562, 37.989898681640625, 12.418455123901367, -17.193647384643555, 287.3665771484375, 406.934814453125, 524.4937744140625, 231.56594848632812, 211.03683471679688, -143.754150390625, 351.30010986328125, 351.9421081542969, 524.6533203125, 173.8880615234375, -62.74485397338867, 607.4108276367188, -23.27392578125, 192.27798461914062, 430.1119689941406, 341.4320068359375, 282.8023681640625, 231.549072265625, 76.00436401367188, 296.8722229003906, 568.7570190429688, 92.04029846191406, -15.130647659301758, 127.8206787109375, 206.54248046875, -51.10939025878906, 78.08879852294922, 422.80206298828125, 132.19288635253906, 550.1356201171875, 198.9503936767578, 366.9804382324219], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000438.npy"}
|
|
{"epoch": 0.6431718061674009, "step": 439, "batch_size": 64, "mean": 174.4813232421875, "std": 224.18980407714844, "min": -203.12762451171875, "p10": -120.98984680175778, "median": 144.69598388671875, "p90": 486.6257019042969, "max": 743.5665283203125, "pos_frac": 0.765625, "sample": [177.03884887695312, 623.044189453125, 75.60385131835938, 292.1507263183594, 32.956817626953125, -94.52688598632812, 483.7067565917969, 110.54496765136719, 175.8763427734375, 123.19689178466797, -203.12762451171875, 487.8766784667969, 48.35462951660156, -3.0194015502929688, 17.162391662597656, 251.43826293945312, 237.11927795410156, 402.8464050292969, 321.3420104980469, 130.6781463623047, 138.7382354736328, 675.3192138671875, 228.59793090820312, -59.061737060546875, 127.60457611083984, -132.33111572265625, 116.78179931640625, 93.12516784667969, -3.089265823364258, 39.04558563232422, 288.7559509277344, -150.16876220703125, 350.7412414550781, 318.36663818359375, 743.5665283203125, 188.55389404296875, 425.6418151855469, 272.72576904296875, 72.45698547363281, 143.23651123046875, 148.60122680664062, -152.04629516601562, -194.98313903808594, -33.00421142578125, 179.72921752929688, 150.50515747070312, 146.15545654296875, 222.4142608642578, -157.23095703125, 415.75341796875, 623.9331665039062, 37.664791107177734, 146.5884246826172, 508.3446044921875, 106.52376556396484, -18.009565353393555, 406.7742919921875, 280.77874755859375, -9.901248931884766, 55.73722839355469, 678.3533325195312, 228.03805541992188, -172.67108154296875, -0.11405563354492188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000439.npy"}
|
|
{"epoch": 0.644640234948605, "step": 440, "batch_size": 64, "mean": 177.81304931640625, "std": 226.8293914794922, "min": -262.7265625, "p10": -65.75025405883787, "median": 146.18838500976562, "p90": 493.98973083496105, "max": 727.773193359375, "pos_frac": 0.78125, "sample": [380.29461669921875, 202.87554931640625, 508.7806701660156, 250.20936584472656, -96.1334228515625, 83.70891571044922, 438.33697509765625, -5.939355850219727, 61.76579666137695, 583.547119140625, 48.56676483154297, -29.8814697265625, 389.7065734863281, 80.6563720703125, 459.4775390625, -131.53616333007812, 132.55673217773438, -225.2296905517578, 55.05058288574219, 9.627159118652344, 139.66714477539062, 418.5630187988281, 73.04888916015625, 39.179054260253906, 258.15484619140625, 210.54917907714844, -99.15575408935547, 292.05426025390625, 289.2237548828125, 81.43977355957031, 185.84103393554688, 411.7049865722656, -49.630157470703125, 209.4441680908203, 6.124269485473633, -29.232208251953125, 659.41455078125, -262.7265625, 201.00198364257812, -72.65886688232422, 568.1998901367188, 184.25912475585938, 727.773193359375, 342.79779052734375, 152.70962524414062, 193.57763671875, 28.933502197265625, -37.40077209472656, 310.5987854003906, 372.1652526855469, 200.36883544921875, 531.6945190429688, 8.618064880371094, 158.51785278320312, 50.087398529052734, 413.2015686035156, 36.14931106567383, -38.69647216796875, -26.89630889892578, 27.848541259765625, 711.7942504882812, 387.2862854003906, -106.61272430419922, 24.6123046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000440.npy"}
|
|
{"epoch": 0.6461086637298091, "step": 441, "batch_size": 64, "mean": 206.31964111328125, "std": 283.3464660644531, "min": -457.52880859375, "p10": -67.2898193359375, "median": 173.59606170654297, "p90": 527.7760131835938, "max": 1301.76220703125, "pos_frac": 0.796875, "sample": [225.94395446777344, 85.75177001953125, 125.2540283203125, 450.42156982421875, 336.0975341796875, 681.0596923828125, -16.761383056640625, 33.65739440917969, 284.45831298828125, -364.38262939453125, -70.67964172363281, -26.02228355407715, 513.7064208984375, 655.1590576171875, 301.06756591796875, 20.09300994873047, 43.89729309082031, 524.2266235351562, 94.88555908203125, 17.564727783203125, 30.11875343322754, 87.99891662597656, 185.15292358398438, 581.136962890625, 596.6952514648438, 108.92413330078125, 162.03919982910156, 373.0043640136719, 282.6403503417969, 1301.76220703125, -17.6726016998291, -292.1717529296875, 271.9293212890625, 499.9898376464844, -258.2667236328125, 19.607484817504883, 104.56436920166016, 264.2835388183594, 529.2971801757812, -72.86093139648438, 12.20269775390625, 304.4577331542969, -59.38023376464844, -457.52880859375, 386.55914306640625, 272.5924072265625, 275.50958251953125, 474.686279296875, 356.9422607421875, 19.51529312133789, 50.31776428222656, -86.18595123291016, -28.92987060546875, 375.9298095703125, 322.15673828125, 105.24139404296875, 485.13653564453125, 234.48927307128906, 280.49517822265625, 625.1096801757812, 38.413143157958984, -27.62895965576172, 151.05618286132812, 419.7287902832031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000441.npy"}
|
|
{"epoch": 0.6475770925110133, "step": 442, "batch_size": 64, "mean": 223.28561401367188, "std": 236.46299743652344, "min": -232.6896209716797, "p10": -31.022761917114256, "median": 169.0298309326172, "p90": 580.1296325683595, "max": 825.78662109375, "pos_frac": 0.84375, "sample": [17.419471740722656, 143.05462646484375, 478.41650390625, 129.34429931640625, 280.74072265625, 111.85879516601562, -13.076591491699219, 225.08924865722656, 57.857421875, 90.38848876953125, 132.01133728027344, 285.7431335449219, 42.10710906982422, 199.86473083496094, 263.9207763671875, 534.3612060546875, 589.0733642578125, 235.6295623779297, 13.202301025390625, 119.78645324707031, 117.21985626220703, 431.7900390625, 49.54559326171875, 279.954833984375, -35.757102966308594, 168.90618896484375, 536.4603881835938, 219.9174346923828, 629.2757568359375, 196.06973266601562, 131.45767211914062, -232.6896209716797, 169.15347290039062, -28.871318817138672, 9.819023132324219, 411.24908447265625, -32.795387268066406, 378.1639404296875, 161.84652709960938, 748.8692016601562, -143.7208251953125, 338.0574645996094, 559.2609252929688, -12.251474380493164, 36.5711669921875, 658.8961181640625, 604.0782470703125, 247.70144653320312, 425.2220764160156, 78.94377899169922, 10.00775146484375, -108.39199829101562, 22.877777099609375, 179.7687225341797, -31.944808959960938, -89.60050964355469, 409.4268493652344, 354.54193115234375, 825.78662109375, 295.61920166015625, 758.1942138671875, 358.7832336425781, 104.30043029785156, 161.7729034423828], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000442.npy"}
|
|
{"epoch": 0.6490455212922174, "step": 443, "batch_size": 64, "mean": 148.06362915039062, "std": 262.1896667480469, "min": -509.5323181152344, "p10": -145.7830841064453, "median": 141.46268463134766, "p90": 420.1353515625001, "max": 939.8646850585938, "pos_frac": 0.796875, "sample": [915.1964721679688, 52.133827209472656, 502.80670166015625, 14.171432495117188, 235.25448608398438, 92.14045715332031, 15.563955307006836, -126.71345520019531, -176.0981903076172, 135.78575134277344, 26.788436889648438, 24.177444458007812, 141.3656005859375, 276.2373046875, 72.74002075195312, -509.5323181152344, 145.46620178222656, -140.00547790527344, 164.43167114257812, 103.203857421875, 939.8646850585938, 375.45318603515625, 328.40374755859375, 228.39950561523438, -245.89358520507812, -504.84307861328125, 202.21932983398438, 173.23597717285156, 430.9820251464844, -375.5711975097656, 317.06982421875, 314.02960205078125, 106.52082824707031, 103.47477722167969, 70.93644714355469, -66.42194366455078, 168.028076171875, 373.8900146484375, 319.4106750488281, 108.26551818847656, -178.43431091308594, -137.7737579345703, 258.77508544921875, -57.43105697631836, 149.02020263671875, 76.38896179199219, 280.3135986328125, 611.8321533203125, 141.5597686767578, 147.13418579101562, 437.1947937011719, 96.05508422851562, 129.25172424316406, 192.48280334472656, -148.2592010498047, 296.7210388183594, -47.26109313964844, 298.42095947265625, 163.49636840820312, 51.694427490234375, 394.8264465332031, 590.1919555664062, 105.6418228149414, 291.661865234375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000443.npy"}
|
|
{"epoch": 0.6505139500734214, "step": 444, "batch_size": 64, "mean": 230.2260284423828, "std": 261.7230529785156, "min": -351.4384460449219, "p10": -44.09638881683348, "median": 200.88668823242188, "p90": 499.2125000000001, "max": 1012.1304931640625, "pos_frac": 0.84375, "sample": [275.9033203125, -101.47052001953125, 87.60306549072266, 411.252685546875, 68.06584930419922, 265.9222412109375, 79.00770568847656, 1012.1304931640625, 422.9971008300781, -194.2342987060547, 10.51174545288086, 205.37448120117188, 354.2633056640625, 448.58026123046875, 469.19732666015625, 224.302734375, 394.71636962890625, 343.0823974609375, -186.6109619140625, 15.294998168945312, -51.680809020996094, -26.3994083404541, 339.7085266113281, 713.4453125, 49.4771728515625, 353.95147705078125, 253.81549072265625, -21.11767578125, 71.40892028808594, -258.73828125, 19.89472007751465, 980.4910888671875, 269.06280517578125, 196.39889526367188, 125.17959594726562, 476.02166748046875, 509.15142822265625, 352.17803955078125, 397.63665771484375, 195.1807861328125, -17.32403564453125, 613.6760864257812, 531.6813354492188, 116.4940414428711, 430.74957275390625, 415.1055603027344, 443.8214111328125, 23.54929542541504, 226.40443420410156, 126.12979125976562, 434.7093505859375, 99.48973083496094, 120.8529281616211, 123.0093765258789, 393.8798828125, 194.09744262695312, 19.84556770324707, 73.00187683105469, -158.0975341796875, 112.45431518554688, 597.3516235351562, -351.4384460449219, 192.96853637695312, 421.09735107421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000444.npy"}
|
|
{"epoch": 0.6519823788546255, "step": 445, "batch_size": 64, "mean": 218.43179321289062, "std": 253.35916137695312, "min": -314.345458984375, "p10": -53.938861083984335, "median": 171.59364318847656, "p90": 576.0533752441406, "max": 905.3009033203125, "pos_frac": 0.84375, "sample": [189.76422119140625, 32.938690185546875, 66.22368621826172, 669.2459716796875, 95.70399475097656, 410.6733703613281, 301.7488098144531, 106.04717254638672, 108.36770629882812, 179.7493438720703, 0.2665367126464844, 563.3831787109375, -19.019760131835938, 267.3031005859375, 36.333412170410156, -314.345458984375, 150.29409790039062, 414.7421569824219, 106.65494537353516, 136.7367401123047, 294.0254211425781, 328.9366760253906, 465.3018798828125, 581.4834594726562, 45.90141677856445, 114.74724578857422, 818.6055297851562, 476.6691589355469, 273.9268493652344, 594.1395874023438, 428.05047607421875, 623.0078125, 79.21632385253906, 496.6595153808594, 93.05058288574219, 5.73089599609375, 238.72238159179688, 291.7134704589844, -135.37161254882812, 404.63629150390625, -10.804786682128906, 776.7156982421875, 380.0109558105469, -94.45292663574219, 381.22955322265625, 269.05987548828125, 163.4379425048828, -193.00885009765625, 34.70142364501953, 31.262357711791992, 276.7669982910156, 15.635002136230469, 27.913570404052734, 393.4405212402344, 301.6981201171875, 118.10810852050781, -15.855804443359375, -83.44266510009766, 905.3009033203125, -68.90419006347656, 261.4788818359375, 212.25433349609375, 105.55661010742188, -230.43295288085938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000445.npy"}
|
|
{"epoch": 0.6534508076358296, "step": 446, "batch_size": 64, "mean": 230.83352661132812, "std": 216.98951721191406, "min": -293.8396911621094, "p10": -11.344997406005852, "median": 207.28736114501953, "p90": 536.7461791992188, "max": 704.2887573242188, "pos_frac": 0.859375, "sample": [-3.6513900756835938, 106.90570068359375, 111.34809875488281, 636.775634765625, 688.4513549804688, 290.31170654296875, 29.065601348876953, -293.8396911621094, 134.70814514160156, 456.06988525390625, 704.2887573242188, 267.1588134765625, 9.329933166503906, 489.84136962890625, 83.16783905029297, 477.1357727050781, 453.85845947265625, 280.06695556640625, 137.3687744140625, 333.24359130859375, -2.6129150390625, 86.60125732421875, 142.98289489746094, 157.78346252441406, 158.8697052001953, -132.90525817871094, 115.04518127441406, 283.4349365234375, 380.10076904296875, 233.78826904296875, 319.6357727050781, -46.94367980957031, 115.86550903320312, 177.74163818359375, 7.117528915405273, 215.30209350585938, -14.642257690429688, 411.3792724609375, 146.60415649414062, 542.44921875, 298.19647216796875, 49.76087951660156, 369.12823486328125, 685.9134521484375, 250.32601928710938, 46.08246612548828, 659.697021484375, 353.293212890625, 253.7153778076172, 322.65045166015625, 306.1543884277344, -41.06221008300781, -94.18865966796875, 179.0230712890625, 35.157554626464844, 226.658935546875, 114.91806030273438, 281.898193359375, 630.1361694335938, 199.2726287841797, -31.498451232910156, 523.4390869140625, 189.28640747070312, 276.1841735839844], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000446.npy"}
|
|
{"epoch": 0.6549192364170338, "step": 447, "batch_size": 64, "mean": 232.25527954101562, "std": 257.3019714355469, "min": -195.94154357910156, "p10": -42.90697364807128, "median": 179.8773193359375, "p90": 518.8571502685548, "max": 1393.240234375, "pos_frac": 0.84375, "sample": [82.15414428710938, 146.5966339111328, 251.0506591796875, 215.4342803955078, 57.417388916015625, -118.85661315917969, 351.0954895019531, 173.00119018554688, 275.25384521484375, -31.78204345703125, -47.674800872802734, 439.12152099609375, 241.11227416992188, 172.19200134277344, 462.0511779785156, 452.84393310546875, 642.4259033203125, 115.10649108886719, 1393.240234375, 623.7002563476562, 142.25216674804688, 682.0518798828125, 335.4877014160156, 392.81634521484375, -98.64833068847656, 146.38998413085938, 248.23046875, -173.24258422851562, 199.30886840820312, 103.87663269042969, -116.21258544921875, 135.1333770751953, 86.83252716064453, -80.51736450195312, 335.9002990722656, -17.673418045043945, 174.49639892578125, 479.6427307128906, 46.2864990234375, 25.01580047607422, 457.8936767578125, 131.79095458984375, 133.00294494628906, -195.94154357910156, 589.1502685546875, 225.90643310546875, 335.5519104003906, -30.290739059448242, 185.25823974609375, 32.88201141357422, 419.95684814453125, 307.47003173828125, 257.20697021484375, 191.8056640625, 41.422149658203125, 368.4871826171875, 171.19496154785156, 55.16718292236328, 632.996826171875, 535.663330078125, 158.046875, 476.54205322265625, 430.52288818359375, 9.739715576171875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000447.npy"}
|
|
{"epoch": 0.6563876651982379, "step": 448, "batch_size": 64, "mean": 244.29922485351562, "std": 238.98329162597656, "min": -128.55760192871094, "p10": -59.71023406982421, "median": 188.2368392944336, "p90": 521.1963623046875, "max": 921.4364013671875, "pos_frac": 0.859375, "sample": [7.071434020996094, 72.64102172851562, -86.95462036132812, 143.12249755859375, 606.9560546875, 56.85791015625, 114.96058654785156, -63.30876922607422, 313.8927307128906, 82.48004150390625, 401.14971923828125, 114.87667083740234, 499.2516784667969, 59.515838623046875, 248.7288818359375, 364.42041015625, 194.1354522705078, 106.56892395019531, 521.6727294921875, -67.60855102539062, -128.55760192871094, 90.43972778320312, 486.8908386230469, 320.2164306640625, 168.83953857421875, 182.33822631835938, 176.06100463867188, 157.8621826171875, 450.8979797363281, 664.9991455078125, 304.28363037109375, 520.0848388671875, 56.635337829589844, 368.31707763671875, 921.4364013671875, 23.73493766784668, 70.32881164550781, 148.01417541503906, 81.82547760009766, 263.675048828125, 424.9378662109375, 484.4488525390625, 170.1666717529297, -92.99246978759766, 799.7286987304688, 277.67010498046875, 90.86383056640625, -77.14971923828125, 271.7317199707031, 239.18502807617188, 276.5792541503906, -124.20332336425781, 855.625732421875, -51.31365203857422, 501.34576416015625, 588.1690673828125, -24.056915283203125, 288.2774658203125, 104.57459259033203, 171.32199096679688, 454.82659912109375, 423.9205627441406, 352.9280090332031, 209.81024169921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000448.npy"}
|
|
{"epoch": 0.657856093979442, "step": 449, "batch_size": 64, "mean": 219.3227081298828, "std": 306.81890869140625, "min": -625.185791015625, "p10": -118.89217071533203, "median": 175.86514282226562, "p90": 565.2874877929688, "max": 1339.1268310546875, "pos_frac": 0.765625, "sample": [181.2052459716797, -31.42974090576172, 305.0107727050781, 30.138290405273438, 546.7950439453125, 541.966064453125, 144.6087646484375, -8.028085708618164, 160.44244384765625, 216.28829956054688, 606.431396484375, -118.28388977050781, -49.611000061035156, 124.8592300415039, 568.2220458984375, 41.0194206237793, 390.91497802734375, 132.78839111328125, 89.33444213867188, -147.2379608154297, 529.1678466796875, 206.56858825683594, 463.8301086425781, 483.9258117675781, 183.52012634277344, 558.440185546875, -119.15286254882812, 30.275985717773438, 257.82952880859375, 263.7398986816406, 170.52503967285156, 133.15060424804688, -64.47003173828125, 445.20806884765625, 399.39892578125, 654.9263916015625, -151.14834594726562, 85.43302917480469, -153.3927001953125, -199.1787567138672, 1339.1268310546875, 495.0228576660156, -136.1913604736328, 308.5684509277344, 244.73562622070312, 159.49496459960938, -51.648529052734375, -625.185791015625, 4.972927093505859, 806.4249877929688, 240.73558044433594, 740.2728271484375, 321.95166015625, -65.34001159667969, 251.8720245361328, 142.2769775390625, 378.199951171875, 866.7130126953125, 254.43063354492188, 28.505050659179688, -45.09071350097656, 3.735015869140625, 149.0317840576172, 320.0072326660156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000449.npy"}
|
|
{"epoch": 0.6593245227606461, "step": 450, "batch_size": 64, "mean": 254.86483764648438, "std": 230.20799255371094, "min": -541.601318359375, "p10": 7.8627571105957035, "median": 253.96186065673828, "p90": 515.2291259765626, "max": 966.609375, "pos_frac": 0.921875, "sample": [158.26251220703125, 464.37109375, 580.148193359375, 316.71484375, 80.029052734375, 549.4000854492188, 966.609375, -76.11534118652344, 299.4779052734375, -13.757835388183594, 491.309326171875, 437.5391845703125, 478.32843017578125, 208.17367553710938, 330.01690673828125, 438.2073974609375, 101.83963012695312, 177.50433349609375, 260.235595703125, 85.00869750976562, 284.87213134765625, 437.37066650390625, 408.1876220703125, 525.48046875, 247.89149475097656, -4.463893890380859, 134.0387420654297, 663.297119140625, 209.03123474121094, 427.435302734375, 164.70143127441406, 379.504638671875, 260.0322265625, 142.19911193847656, 293.6088562011719, 33.60872268676758, 112.68034362792969, 383.9659729003906, 220.70816040039062, 116.87429809570312, 641.4570922851562, -123.65632629394531, 187.79486083984375, 363.9321594238281, 170.53549194335938, 270.54144287109375, 25.407447814941406, 362.9902648925781, 226.6199951171875, 69.17045593261719, 362.09979248046875, 7.892669677734375, 323.4950866699219, 269.57061767578125, 31.195775985717773, -541.601318359375, 308.8558349609375, 264.0221862792969, 1.3508987426757812, 187.78375244140625, 166.18597412109375, 800.6605224609375, 7.849937438964844, 152.8668212890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000450.npy"}
|
|
{"epoch": 0.6607929515418502, "step": 451, "batch_size": 64, "mean": 196.84994506835938, "std": 259.9522399902344, "min": -675.92919921875, "p10": -110.84138946533201, "median": 191.1505126953125, "p90": 492.23395385742197, "max": 789.558837890625, "pos_frac": 0.84375, "sample": [106.92352294921875, 46.86456298828125, 83.41357421875, 203.70028686523438, 390.59893798828125, 164.80502319335938, 201.8803253173828, 471.4433898925781, 61.888145446777344, 325.7488708496094, 462.8672180175781, 639.4452514648438, 455.392333984375, 638.4796142578125, 156.05157470703125, 177.56039428710938, 67.79297637939453, -675.92919921875, -127.642578125, -160.62689208984375, 130.87432861328125, 85.95194244384766, 438.1894226074219, 382.2703857421875, 299.2044372558594, 202.44496154785156, 349.8104553222656, 169.1769561767578, 218.34201049804688, 529.6162109375, 113.13601684570312, 789.558837890625, 618.3502197265625, 175.45162963867188, 347.7031555175781, 292.85870361328125, 343.70794677734375, -92.00251770019531, 255.10098266601562, 187.60560607910156, -47.85313415527344, -0.9512062072753906, 416.16058349609375, -580.6826171875, 180.39659118652344, 208.262451171875, 194.69541931152344, 171.759521484375, 260.839111328125, 29.13271141052246, 196.63336181640625, -145.9381561279297, -118.91519165039062, 565.0711059570312, 429.2425842285156, 142.1459503173828, 384.55859375, -314.0049133300781, 176.06475830078125, 8.79090404510498, 501.1441955566406, 4.979351043701172, 237.27981567382812, 171.57733154296875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000451.npy"}
|
|
{"epoch": 0.6622613803230544, "step": 452, "batch_size": 64, "mean": 221.6055450439453, "std": 206.95066833496094, "min": -207.90335083007812, "p10": -23.325763702392575, "median": 197.98135375976562, "p90": 480.9881958007813, "max": 774.5558471679688, "pos_frac": 0.875, "sample": [440.4073486328125, -124.68155670166016, 408.2107849121094, 351.2796325683594, -24.588768005371094, 377.3509826660156, 337.6849365234375, 327.43505859375, 326.0884094238281, -49.375022888183594, 560.8881225585938, 251.50030517578125, 87.95904541015625, 91.0404281616211, 365.5111083984375, 192.12002563476562, 414.9319152832031, 277.3631286621094, 353.83203125, 24.34956169128418, 308.5203552246094, 9.348747253417969, 348.8393249511719, 139.70709228515625, 202.8133087158203, 612.6538696289062, 47.494144439697266, -136.91259765625, 358.93994140625, 90.32763671875, -20.378753662109375, 404.1558837890625, 76.43418884277344, 197.00991821289062, 611.5960693359375, 96.14421081542969, 256.6925354003906, 445.4609375, 129.65283203125, 149.77859497070312, 55.47537612915039, 6.74066162109375, 107.81643676757812, 658.23193359375, 148.10382080078125, 321.9289855957031, 96.366455078125, 471.3711853027344, 774.5558471679688, 131.3020782470703, 124.05464172363281, 220.32925415039062, 326.89129638671875, 25.038814544677734, -207.90335083007812, 181.1119384765625, -41.0804443359375, 485.1097717285156, 206.14956665039062, 522.2044067382812, 28.656471252441406, 198.95278930664062, -29.134414672851562, 52.89519500732422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000452.npy"}
|
|
{"epoch": 0.6637298091042585, "step": 453, "batch_size": 64, "mean": 198.021484375, "std": 232.22840881347656, "min": -201.91897583007812, "p10": -74.87649688720703, "median": 139.96949768066406, "p90": 477.78968811035173, "max": 1174.074462890625, "pos_frac": 0.8125, "sample": [386.1348571777344, 434.6182861328125, 9.181159973144531, 101.98241424560547, 290.82861328125, 208.30140686035156, 164.2333984375, 127.59007263183594, 115.63145446777344, 120.0062255859375, 212.49853515625, 251.29718017578125, 393.0088195800781, 428.7341613769531, 132.4066162109375, 81.44117736816406, 303.440185546875, 250.54635620117188, -87.06370544433594, 40.18000793457031, 324.61004638671875, -112.78916931152344, 148.97445678710938, 152.6947021484375, 525.4022216796875, -144.33578491210938, 82.23579406738281, 401.6907043457031, -201.91897583007812, 285.2059631347656, -19.701522827148438, 125.09262084960938, 292.4212341308594, 496.2917175292969, 588.09228515625, 57.62117385864258, -16.097976684570312, 123.255126953125, 1174.074462890625, 586.6792602539062, 93.26626586914062, -119.48091888427734, 641.0765991210938, 28.04888153076172, 196.7084197998047, 66.05909729003906, 272.19842529296875, 57.49243927001953, 382.44757080078125, 294.483154296875, -72.64955139160156, 333.66705322265625, -75.83090209960938, 393.5876159667969, 104.10235595703125, -82.32132720947266, -12.697860717773438, 147.53237915039062, 542.052734375, 405.71044921875, -33.39334487915039, 51.914451599121094, 96.42306518554688, 128.48291015625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000453.npy"}
|
|
{"epoch": 0.6651982378854625, "step": 454, "batch_size": 64, "mean": 234.9840087890625, "std": 267.9895935058594, "min": -391.00494384765625, "p10": 6.44961433410645, "median": 187.4669647216797, "p90": 557.916961669922, "max": 1118.0220947265625, "pos_frac": 0.90625, "sample": [143.15750122070312, 4.833984375, 708.5319213867188, 98.24635314941406, 433.1022033691406, 27.902786254882812, 538.93505859375, 248.3040008544922, 75.00344848632812, 180.5087432861328, -391.00494384765625, 426.2791748046875, 256.9678649902344, 261.252197265625, 216.07107543945312, 133.12744140625, 223.30076599121094, 278.61181640625, -141.767333984375, -316.1764221191406, 65.91435241699219, 432.94085693359375, 12.98150634765625, 715.8035888671875, -27.059425354003906, 25.403167724609375, 356.1328125, 49.762123107910156, 130.57203674316406, 175.02447509765625, 375.7145690917969, 67.45280456542969, 265.50457763671875, 108.48636627197266, 400.40435791015625, 145.98605346679688, 246.06427001953125, 89.05874633789062, 314.10150146484375, 194.42518615722656, 355.4501953125, 387.3739929199219, 54.82893371582031, 291.7328796386719, 24.642791748046875, 566.0520629882812, 1118.0220947265625, 225.66018676757812, 58.676551818847656, 134.81674194335938, 976.6354370117188, 920.159423828125, 83.8641128540039, 379.28155517578125, 106.76790618896484, 362.94061279296875, 416.76922607421875, -50.055877685546875, 321.7983093261719, -7.5940399169921875, 151.2291717529297, 10.219417572021484, 579.27783203125, 20.566089630126953], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000454.npy"}
|
|
{"epoch": 0.6666666666666666, "step": 455, "batch_size": 64, "mean": 262.0626220703125, "std": 278.8086242675781, "min": -366.41705322265625, "p10": -44.04685287475583, "median": 211.2845230102539, "p90": 666.1114868164062, "max": 942.16357421875, "pos_frac": 0.875, "sample": [51.48965072631836, 195.92697143554688, 399.1062316894531, 593.895751953125, -197.0411834716797, 292.7742614746094, 542.8111572265625, -58.47035217285156, 800.7919921875, -125.5324935913086, 726.7573852539062, 185.2588348388672, 621.7371826171875, -103.36984252929688, 805.9639892578125, -97.60729217529297, 284.66326904296875, 658.1396484375, 132.90072631835938, 127.65046691894531, 4.98297119140625, 378.7523498535156, 228.54991149902344, 54.224159240722656, 283.01025390625, 838.373291015625, 283.9024658203125, -10.392021179199219, 88.74131774902344, 183.71531677246094, 299.656982421875, 420.734619140625, 213.74874877929688, 121.43850708007812, 73.94621276855469, 152.95260620117188, 251.58544921875, 307.447021484375, 73.95098876953125, -366.41705322265625, 339.40618896484375, 29.95339584350586, 265.8232421875, 441.04998779296875, 28.01885986328125, 71.6937026977539, -98.24212646484375, 375.6470947265625, 668.325927734375, 207.10440063476562, 600.5543823242188, 56.953857421875, 208.82029724121094, 362.3672180175781, 117.55402374267578, 98.09609985351562, 6.868946075439453, 660.9444580078125, 275.64190673828125, 270.10540771484375, 841.4469604492188, 106.90892791748047, 174.0502471923828, 942.16357421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000455.npy"}
|
|
{"epoch": 0.6681350954478708, "step": 456, "batch_size": 64, "mean": 254.17987060546875, "std": 227.615478515625, "min": -137.98291015625, "p10": 0.741769790649415, "median": 211.57376861572266, "p90": 606.7820739746094, "max": 776.1741333007812, "pos_frac": 0.90625, "sample": [141.696044921875, 105.00401306152344, 604.9699096679688, 176.16574096679688, 1.61993408203125, 239.296142578125, 0.3654136657714844, 86.79193115234375, 181.68475341796875, 533.10498046875, 321.39984130859375, 32.38905334472656, 134.91395568847656, 600.208740234375, 217.05953979492188, 75.20439910888672, 146.4465789794922, 331.6634216308594, 256.49200439453125, 272.738525390625, 118.20271301269531, 52.043678283691406, 21.33343505859375, 697.902587890625, -86.44827270507812, 330.0025939941406, 776.1741333007812, 271.4236755371094, 450.9844970703125, 52.283966064453125, 261.17999267578125, 444.8128967285156, 524.4940185546875, 213.2097930908203, 103.67757415771484, 77.11722564697266, -14.73084831237793, 342.097900390625, 709.8381958007812, 60.66224670410156, 621.686767578125, 293.59576416015625, 245.2118682861328, 75.6670150756836, -137.98291015625, 209.937744140625, -1.0789432525634766, -29.40545654296875, 607.5587158203125, 165.36843872070312, 748.0494384765625, 593.477783203125, 177.43643188476562, -5.61285400390625, 100.21688842773438, 124.47550964355469, 275.3575134277344, 81.225341796875, 370.279052734375, 83.93140411376953, 375.8232116699219, 573.1270751953125, 618.9993896484375, 234.68943786621094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000456.npy"}
|
|
{"epoch": 0.6696035242290749, "step": 457, "batch_size": 64, "mean": 245.4560546875, "std": 226.9004364013672, "min": -289.177490234375, "p10": -12.044242095947261, "median": 263.79437255859375, "p90": 506.5214477539063, "max": 822.1519775390625, "pos_frac": 0.875, "sample": [54.469886779785156, 373.6402893066406, 362.99267578125, 250.19808959960938, 309.65142822265625, 293.702392578125, 421.4283447265625, 540.2784423828125, 133.29421997070312, 53.08238220214844, 822.1519775390625, 467.41162109375, 38.35080337524414, 63.71653366088867, 119.46363830566406, 191.0529327392578, 425.26617431640625, 235.66131591796875, 488.2370300292969, 385.85107421875, 563.219482421875, -8.474964141845703, 226.4604034423828, 402.628173828125, 277.3906555175781, 481.12786865234375, -289.177490234375, 0.8676567077636719, 490.2593688964844, 493.8125, 71.09159851074219, 311.7458190917969, 123.49247741699219, 308.2953796386719, 281.3853759765625, -152.45692443847656, 428.3303527832031, 378.0830383300781, 101.99337768554688, -26.9945068359375, 192.1999053955078, 606.1714477539062, 225.6608123779297, -13.573932647705078, 40.1922721862793, 61.49299621582031, 436.6785583496094, 156.4979248046875, 385.39501953125, 348.6275939941406, -156.4144287109375, 347.3215026855469, 173.35877990722656, 624.238037109375, -279.025390625, 511.9681396484375, 153.51205444335938, -120.76580047607422, 292.87762451171875, 40.03157424926758, 127.32450103759766, 351.4215087890625, 91.90648651123047, 619.1094360351562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000457.npy"}
|
|
{"epoch": 0.671071953010279, "step": 458, "batch_size": 64, "mean": 255.62677001953125, "std": 262.8251037597656, "min": -196.67678833007812, "p10": -78.49713973999023, "median": 275.1033020019531, "p90": 451.99920654296886, "max": 1273.1038818359375, "pos_frac": 0.796875, "sample": [-196.67678833007812, 256.1904296875, 355.89642333984375, 225.35452270507812, 192.78892517089844, -1.0712127685546875, 1273.1038818359375, -153.7606658935547, 335.0057373046875, 379.2656555175781, 137.03018188476562, 274.38275146484375, 398.4608154296875, 322.122314453125, 633.2858276367188, -126.0592041015625, -74.01856994628906, -2.267171859741211, 202.77081298828125, 344.949951171875, 428.677978515625, 378.3653259277344, 160.99758911132812, 461.9940185546875, 308.32427978515625, 284.3204345703125, 252.70248413085938, 40.10333251953125, 277.92431640625, 285.50048828125, 316.7998352050781, 742.0220336914062, 390.6772766113281, 422.225341796875, 235.11807250976562, -29.42291259765625, 275.8238525390625, 59.41136169433594, 506.2098388671875, 212.08209228515625, 558.7451171875, 292.0184326171875, -80.4165267944336, -106.3323745727539, 243.78121948242188, -156.70263671875, 408.9501953125, -13.288688659667969, 121.45167541503906, 1144.6746826171875, 268.642578125, 314.8078308105469, 228.0677490234375, 79.73153686523438, 140.6661376953125, 401.2370910644531, 343.9215087890625, 349.7627868652344, 199.28370666503906, 334.9764404296875, -13.807138442993164, 398.00030517578125, -165.17202758789062, 280.5023193359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000458.npy"}
|
|
{"epoch": 0.6725403817914831, "step": 459, "batch_size": 64, "mean": 237.4794158935547, "std": 257.1190490722656, "min": -514.702392578125, "p10": -11.409993743896477, "median": 190.12843322753906, "p90": 562.635906982422, "max": 903.7509155273438, "pos_frac": 0.875, "sample": [53.04039001464844, 86.73977661132812, 167.1658935546875, 308.98809814453125, 49.42485809326172, 306.0713806152344, -104.75689697265625, 136.6627655029297, 384.89385986328125, 153.9816131591797, 449.9693298339844, 285.0656433105469, 428.7236633300781, 340.49810791015625, -4.104387283325195, 304.0167236328125, 63.3133544921875, 352.4216003417969, 69.64446258544922, 656.2894287109375, 37.19744873046875, 617.6327514648438, 481.0936279296875, -85.55413818359375, 332.4879455566406, -14.54096794128418, 454.5800476074219, 0.8858184814453125, 191.69960021972656, 118.07889556884766, 121.75309753417969, 135.35147094726562, 593.55078125, -514.702392578125, 903.7509155273438, 232.5225372314453, 854.029541015625, 298.6848449707031, 176.1376190185547, -151.829833984375, 110.01187896728516, 148.90219116210938, 172.05133056640625, 83.18890380859375, 13.53094482421875, 254.00228881835938, -260.77606201171875, 177.73178100585938, 169.1685028076172, 467.6033630371094, 554.4067993164062, 90.13479614257812, 203.47039794921875, 806.23876953125, 260.9592590332031, 188.55726623535156, 396.2887878417969, -119.8569564819336, 564.6676635742188, 557.8951416015625, 362.795166015625, 253.0614013671875, 32.63521194458008, 441.1552734375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000459.npy"}
|
|
{"epoch": 0.6740088105726872, "step": 460, "batch_size": 64, "mean": 248.22824096679688, "std": 251.23037719726562, "min": -245.91897583007812, "p10": -58.23565330505371, "median": 228.20926666259766, "p90": 577.215447998047, "max": 935.0209350585938, "pos_frac": 0.796875, "sample": [591.7991943359375, 359.70538330078125, 230.90603637695312, -41.40496826171875, 78.92631530761719, 543.1867065429688, 620.5147705078125, 267.3605041503906, -245.91897583007812, 183.7193145751953, -52.048545837402344, 517.719482421875, 141.7246856689453, 462.2974548339844, 142.3384246826172, 188.32806396484375, 208.9832763671875, -59.09772872924805, -23.558639526367188, 935.0209350585938, 629.2189331054688, 191.0238037109375, 407.3359069824219, 110.54544067382812, 665.585693359375, 317.6008605957031, 800.0972900390625, 255.1177520751953, 92.96631622314453, 19.96795654296875, 224.65716552734375, -149.59585571289062, 531.5584716796875, -197.3417510986328, 324.5531921386719, 601.2599487304688, -56.224143981933594, 405.9087829589844, 295.8250427246094, -0.03972625732421875, 407.98394775390625, 372.0821228027344, 167.3909149169922, 320.13983154296875, 363.92529296875, 65.5318603515625, 225.5124969482422, 177.7699737548828, 320.74267578125, 222.8334197998047, -209.68064880371094, -29.833900451660156, 63.00536346435547, 181.86465454101562, -163.87257385253906, 325.40625, 446.80548095703125, 300.6584167480469, -115.4762954711914, 521.7178344726562, 389.28900146484375, 299.3932800292969, 222.1397705078125, 490.7562255859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000460.npy"}
|
|
{"epoch": 0.6754772393538914, "step": 461, "batch_size": 64, "mean": 205.47274780273438, "std": 256.77386474609375, "min": -349.170166015625, "p10": -62.49787368774413, "median": 158.18441009521484, "p90": 466.34665832519534, "max": 945.2571411132812, "pos_frac": 0.796875, "sample": [93.6258544921875, -47.39659118652344, 350.62823486328125, 235.25564575195312, -9.685310363769531, 466.6692810058594, -349.170166015625, 738.4639892578125, 59.004241943359375, 319.08782958984375, 115.95369720458984, -136.0814208984375, 120.00350952148438, 359.0271301269531, 736.5842895507812, 354.81317138671875, 632.6776123046875, 19.854524612426758, 137.60841369628906, 175.3927764892578, 133.41993713378906, 414.5705871582031, 465.5938720703125, 771.1644287109375, -54.38874816894531, 4.8536834716796875, 763.7138061523438, 945.2571411132812, 72.08390808105469, 39.53717041015625, 364.2856140136719, 153.01719665527344, 286.78399658203125, 25.650253295898438, 316.9542541503906, 128.5200958251953, -172.18862915039062, 51.9312744140625, -83.80453491210938, -8.864471435546875, 397.5869140625, -26.32874298095703, 385.64947509765625, 403.6350402832031, -185.0972900390625, 260.016845703125, 226.7598114013672, 437.357177734375, 22.238250732421875, 81.959228515625, 394.1153259277344, 94.73189544677734, 88.07461547851562, 163.35162353515625, 370.26275634765625, -65.97321319580078, -33.856048583984375, 211.68399047851562, 218.1807861328125, 203.77313232421875, 61.828453063964844, -210.4987335205078, 306.22271728515625, 354.1741943359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000461.npy"}
|
|
{"epoch": 0.6769456681350955, "step": 462, "batch_size": 64, "mean": 169.3271026611328, "std": 242.47805786132812, "min": -321.05389404296875, "p10": -119.90933227539061, "median": 164.58853149414062, "p90": 430.93475341796875, "max": 812.1295166015625, "pos_frac": 0.75, "sample": [212.46124267578125, -22.131675720214844, -62.07598114013672, -74.53703308105469, 361.6529235839844, 316.3314208984375, 30.092073440551758, 348.52496337890625, 343.9710693359375, 411.14971923828125, 283.700927734375, -271.1803283691406, -216.01759338378906, -126.70396423339844, 139.032470703125, 319.0569152832031, 517.5249633789062, 193.74417114257812, -63.52471923828125, 382.3409423828125, -139.71658325195312, 146.39483642578125, 258.210205078125, -321.05389404296875, -51.52189254760742, 117.40597534179688, 431.68731689453125, -104.05519104003906, 32.5423698425293, 41.45448684692383, 766.2840576171875, 153.46063232421875, -53.1447868347168, 235.01597595214844, 392.86065673828125, 812.1295166015625, 9.599611282348633, -77.76590728759766, 548.43896484375, 299.52557373046875, 73.14649963378906, -72.59172058105469, 253.11668395996094, 127.63080596923828, 30.983671188354492, 45.16252136230469, 290.4665832519531, 509.423095703125, -234.98306274414062, 99.25402069091797, 33.40032958984375, 302.7052001953125, 709.3990478515625, 321.43939208984375, 429.17877197265625, 5.14508056640625, 368.3549499511719, 215.3853302001953, 175.7164306640625, 267.7550048828125, 107.12342834472656, 176.26565551757812, -165.42356872558594, 247.71559143066406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000462.npy"}
|
|
{"epoch": 0.6784140969162996, "step": 463, "batch_size": 64, "mean": 224.18450927734375, "std": 291.0583190917969, "min": -299.1660461425781, "p10": -125.55717697143554, "median": 237.64759063720703, "p90": 602.5517700195314, "max": 965.7678833007812, "pos_frac": 0.75, "sample": [37.2308349609375, 460.6828308105469, -44.92798614501953, 82.50994873046875, 75.71503448486328, 295.2572326660156, -160.16867065429688, 3.7121200561523438, 301.33721923828125, -218.91847229003906, 441.3948974609375, -129.95834350585938, 350.61859130859375, 514.8253173828125, 113.06704711914062, 224.780517578125, 251.041259765625, 387.2082214355469, 282.6700744628906, -4.412342071533203, 86.55548095703125, 965.7678833007812, 315.52197265625, 207.49349975585938, 619.7425537109375, 487.37725830078125, 112.13137817382812, 93.27518463134766, -102.24288940429688, 256.4466857910156, -246.27084350585938, -9.490211486816406, -299.1660461425781, -45.00524139404297, 674.9812622070312, -218.96615600585938, 1.3413543701171875, 131.14752197265625, 316.4948425292969, -66.39540100097656, 474.94537353515625, 274.0623779296875, 549.6652221679688, 253.34988403320312, 831.21142578125, 729.8088989257812, 36.00469970703125, 295.9659423828125, 548.47607421875, -88.56971740722656, 255.8401336669922, 852.2482299804688, 155.72128295898438, 562.43994140625, -115.28778839111328, -7.61676025390625, 250.51466369628906, 141.22268676757812, -212.86778259277344, 251.21151733398438, 364.1347961425781, 451.2102966308594, 799.7239990234375, 149.98793029785156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000463.npy"}
|
|
{"epoch": 0.6798825256975036, "step": 464, "batch_size": 64, "mean": 317.8370361328125, "std": 295.8577575683594, "min": -347.9139709472656, "p10": -38.87569789886473, "median": 294.50160217285156, "p90": 652.2101501464844, "max": 1348.5615234375, "pos_frac": 0.875, "sample": [-121.62629699707031, 574.0216064453125, 298.8545837402344, 87.98072814941406, 160.917236328125, 520.4476318359375, 331.08721923828125, -25.977609634399414, 133.72286987304688, 320.30242919921875, 551.3453979492188, 553.9033203125, -92.64932250976562, 650.4561767578125, 242.66653442382812, 405.71588134765625, 422.046142578125, 87.64494323730469, 143.064453125, 652.9618530273438, 534.2529296875, 277.7776794433594, 676.611328125, -159.07334899902344, 790.7008666992188, 110.81021881103516, 353.4276428222656, 495.01251220703125, -124.0110092163086, 89.77307891845703, 264.70416259765625, 211.03326416015625, 381.6356201171875, 54.27359390258789, 181.05935668945312, 291.7660827636719, 596.8726806640625, 297.23712158203125, 163.0941162109375, 202.38418579101562, 52.029052734375, 242.0895233154297, 613.9259643554688, 83.5904769897461, 66.68382263183594, 429.401611328125, -125.21549224853516, 214.71517944335938, 15.059806823730469, 543.66748046875, 522.1526489257812, 913.5670776367188, 1348.5615234375, 687.4119262695312, 336.9548034667969, 550.4804077148438, 611.4862060546875, -44.40345001220703, 747.0214233398438, 566.4437255859375, 413.8415832519531, -347.9139709472656, 273.33172607421875, 40.458717346191406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000464.npy"}
|
|
{"epoch": 0.6813509544787077, "step": 465, "batch_size": 64, "mean": 278.3427734375, "std": 259.730224609375, "min": -274.5818786621094, "p10": -90.90707550048825, "median": 289.21844482421875, "p90": 572.1370483398438, "max": 1152.5567626953125, "pos_frac": 0.859375, "sample": [513.2918701171875, 557.2196044921875, 293.9562683105469, 136.88624572753906, -167.06597900390625, 66.03884887695312, 442.11883544921875, 160.706298828125, 219.37371826171875, 222.89498901367188, 440.7889709472656, 191.49078369140625, 142.32284545898438, 179.88043212890625, 520.384033203125, 629.1874389648438, 309.51055908203125, 316.5061340332031, 313.9898681640625, 102.27132415771484, -134.54794311523438, 481.6198425292969, 562.3624267578125, 192.2620849609375, 1152.5567626953125, -64.04682922363281, 286.69091796875, 88.6944580078125, 172.88925170898438, 361.4173583984375, 465.3722229003906, 320.12579345703125, 63.942665100097656, 155.8629150390625, 94.75468444824219, 454.99151611328125, -25.86193084716797, 299.7163391113281, -102.41860961914062, 337.1173400878906, 576.326171875, 232.63998413085938, 291.7459716796875, -221.75668334960938, 455.5222473144531, 313.05413818359375, 103.94195556640625, 441.4556884765625, 96.95249938964844, 271.4336242675781, 660.218505859375, -115.75454711914062, 783.3233642578125, -274.5818786621094, 662.8905029296875, 270.7471923828125, -139.7884979248047, 218.3546142578125, 380.59320068359375, 445.7718811035156, 37.98628234863281, 399.68084716796875, 637.8545532226562, 530.07275390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000465.npy"}
|
|
{"epoch": 0.6828193832599119, "step": 466, "batch_size": 64, "mean": 222.0611114501953, "std": 265.2637939453125, "min": -475.1767883300781, "p10": -72.10706291198728, "median": 214.48110961914062, "p90": 518.5747222900392, "max": 860.3522338867188, "pos_frac": 0.796875, "sample": [200.98269653320312, -11.153396606445312, 168.40869140625, -134.66464233398438, 396.725341796875, 314.69940185546875, 302.73651123046875, 225.26145935058594, 430.3422546386719, -50.10121154785156, -19.348289489746094, 269.9868469238281, -254.25131225585938, 758.4735107421875, -80.09197998046875, 215.5341796875, 184.8928680419922, -8.024761199951172, 310.6651611328125, 137.44049072265625, 329.1458740234375, 192.6339111328125, 211.2248077392578, 371.96051025390625, 502.4584655761719, 116.13601684570312, 277.12493896484375, 525.481689453125, 308.91632080078125, 358.07696533203125, 213.42803955078125, 183.26556396484375, 383.740478515625, 673.1449584960938, -475.1767883300781, 860.3522338867188, 296.5248718261719, 703.4891357421875, 485.4892578125, 81.22473907470703, 353.8973388671875, 7.482913970947266, 158.4544677734375, 171.20266723632812, 406.844970703125, 790.8438110351562, -53.475589752197266, -153.79800415039062, 28.57799530029297, 156.42422485351562, 453.3319091796875, 224.09107971191406, 229.45639038085938, 35.92998504638672, 388.2642517089844, 358.9540710449219, -447.9672546386719, 39.77537536621094, 663.0718994140625, -112.18177795410156, 365.79144287109375, -38.85381317138672, 95.17781829833984, 133.45846557617188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000466.npy"}
|
|
{"epoch": 0.684287812041116, "step": 467, "batch_size": 64, "mean": 237.80014038085938, "std": 334.0853576660156, "min": -483.5992431640625, "p10": -162.2244964599609, "median": 210.16908264160156, "p90": 734.2256591796877, "max": 945.9102783203125, "pos_frac": 0.734375, "sample": [393.31103515625, 681.6246337890625, 520.83203125, 691.9267578125, -285.0010986328125, -483.5992431640625, 224.5862274169922, 359.3289489746094, 200.12193298339844, 196.16995239257812, 191.64328002929688, 379.38641357421875, 752.353759765625, 4.310041427612305, 387.16876220703125, 417.44512939453125, 91.15186309814453, 220.2162322998047, 805.4328002929688, 561.1474609375, 330.455078125, 69.13182067871094, -17.243099212646484, 456.0705261230469, -102.07673645019531, -172.68008422851562, -109.80622863769531, 244.47042846679688, 895.2680053710938, 272.74200439453125, 641.591064453125, 135.65223693847656, 412.260009765625, -107.00324249267578, 755.6827392578125, -13.22845458984375, 945.9102783203125, -29.330081939697266, 113.06109619140625, 543.2056884765625, 140.958740234375, -372.84747314453125, 130.23171997070312, -137.828125, 153.78480529785156, 222.87841796875, -112.772216796875, -127.5672607421875, -194.6607666015625, 539.1107177734375, 498.4332275390625, 418.7669677734375, 387.7976989746094, -24.83304214477539, 886.75830078125, 229.97470092773438, 418.31005859375, 58.975555419921875, -184.5437469482422, 816.8257446289062, -309.8869323730469, 51.198299407958984, 47.81237030029297, 108.6409912109375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000467.npy"}
|
|
{"epoch": 0.6857562408223201, "step": 468, "batch_size": 64, "mean": 260.9195556640625, "std": 272.6064453125, "min": -364.09185791015625, "p10": -49.60289993286132, "median": 231.35992431640625, "p90": 639.5100524902347, "max": 912.8483276367188, "pos_frac": 0.84375, "sample": [389.1835021972656, -41.89618682861328, 912.8483276367188, 212.65802001953125, 578.7855224609375, 522.9158325195312, 277.45977783203125, 560.9703369140625, 392.5322570800781, 298.44146728515625, 299.0581970214844, 298.484130859375, 114.8201904296875, 185.6956787109375, 101.43508911132812, 81.03045654296875, 107.04072570800781, 24.078405380249023, 407.3133239746094, 225.1798095703125, 168.59878540039062, 31.845237731933594, -16.19696807861328, 257.2809753417969, 326.01910400390625, 463.2054138183594, 205.4488525390625, 813.9755249023438, 686.2701416015625, 57.074371337890625, -204.0946807861328, -52.90577697753906, 466.4105529785156, 283.4638366699219, 427.4366149902344, 303.47357177734375, 157.16232299804688, 104.35543823242188, 577.1295166015625, -364.09185791015625, 833.8222045898438, 75.466064453125, 258.28289794921875, 219.95431518554688, 313.4490966796875, 225.10470581054688, -20.06060791015625, 145.46743774414062, 4.5031890869140625, 893.655517578125, 237.5400390625, 665.5348510742188, 116.92650604248047, 384.8707580566406, 84.04865264892578, 474.97088623046875, 516.646484375, 722.7975463867188, 306.416748046875, -80.24798583984375, 109.54145812988281, -204.9718017578125, -159.4056854248047, -65.35901641845703], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000468.npy"}
|
|
{"epoch": 0.6872246696035242, "step": 469, "batch_size": 64, "mean": 272.3406982421875, "std": 270.6866149902344, "min": -289.7607421875, "p10": -1.3710098266601505, "median": 218.20164489746094, "p90": 689.96572265625, "max": 887.4579467773438, "pos_frac": 0.890625, "sample": [543.4367065429688, 268.0107421875, 110.8995590209961, 227.2063446044922, 676.9290771484375, 130.4570770263672, 539.3824462890625, 135.18673706054688, 54.66532897949219, 178.7421875, 424.6266174316406, 86.25840759277344, 178.46519470214844, 47.942047119140625, 850.2333984375, 98.73082733154297, 442.55230712890625, 530.2496337890625, 353.942138671875, 364.0110778808594, 648.246826171875, 277.67620849609375, 119.52323913574219, 173.1658172607422, 887.4579467773438, -3.85284423828125, 447.7176513671875, 225.31021118164062, 760.3344116210938, -171.50616455078125, 180.08566284179688, 185.61436462402344, 401.1285095214844, 47.49663543701172, 357.8939208984375, 102.28777313232422, 88.85848236083984, -63.51300048828125, 4.4199371337890625, -153.9584197998047, 319.1247253417969, 247.97671508789062, 695.5528564453125, 8.458812713623047, -222.1139678955078, 161.87026977539062, 333.5408020019531, 393.7559814453125, 300.645263671875, -29.34429359436035, 553.3570556640625, 737.3733520507812, 284.2655944824219, 830.2443237304688, 820.1434326171875, 442.0417785644531, 87.47936248779297, 396.1432800292969, 68.94206237792969, 133.09457397460938, 211.09307861328125, 60.75779724121094, 128.8469696044922, -289.7607421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000469.npy"}
|
|
{"epoch": 0.6886930983847284, "step": 470, "batch_size": 64, "mean": 258.62164306640625, "std": 256.94970703125, "min": -610.1917724609375, "p10": -41.62600669860838, "median": 240.31944274902344, "p90": 577.2301330566407, "max": 791.8897705078125, "pos_frac": 0.875, "sample": [276.48052978515625, 345.068115234375, 130.6890869140625, 66.24195098876953, 157.13668823242188, 61.777557373046875, 251.51312255859375, 100.19844818115234, 99.53739929199219, -213.6217498779297, 316.27593994140625, 103.23893737792969, 445.00701904296875, 186.36717224121094, 446.65985107421875, 632.599609375, 11.375532150268555, 583.9760131835938, -610.1917724609375, 385.73291015625, 119.41876220703125, 204.15383911132812, 63.95458984375, -217.68710327148438, 158.92266845703125, 176.19802856445312, -18.020320892333984, 279.42584228515625, 227.63037109375, 182.0380859375, 479.70501708984375, -122.59593200683594, 696.57861328125, 145.78221130371094, 57.479637145996094, 436.58599853515625, 256.20098876953125, 513.14111328125, 791.8897705078125, 452.86334228515625, 531.0440063476562, 158.08087158203125, 429.0685729980469, 561.48974609375, -63.646697998046875, 689.8597412109375, 356.0599670410156, 262.0581970214844, 509.7734680175781, 401.3937683105469, 293.6380615234375, -51.74272918701172, 209.46038818359375, 229.12576293945312, 522.566650390625, 674.5528564453125, -116.03704833984375, 702.0621948242188, 449.4195251464844, 118.86326599121094, 491.981689453125, 208.13531494140625, 73.21178436279297, 251.63815307617188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000470.npy"}
|
|
{"epoch": 0.6901615271659325, "step": 471, "batch_size": 64, "mean": 219.40579223632812, "std": 251.19482421875, "min": -359.22564697265625, "p10": -69.28296890258788, "median": 226.5001449584961, "p90": 481.17858581542976, "max": 1138.895751953125, "pos_frac": 0.8125, "sample": [581.73046875, -15.159080505371094, 157.91842651367188, 559.2626953125, 109.330322265625, 236.706298828125, 162.8030548095703, 841.831298828125, 458.22998046875, 101.82656860351562, 129.5807647705078, 148.19085693359375, 216.2939910888672, 329.7304382324219, 315.767822265625, 97.70208740234375, -73.7918701171875, -92.61788940429688, 244.32424926757812, 261.6720886230469, 1138.895751953125, -13.682601928710938, 171.74951171875, 363.3058166503906, 366.3177185058594, -56.35114288330078, -117.528564453125, 36.169464111328125, -59.98711395263672, 241.59439086914062, -83.86425018310547, 653.1311645507812, 208.15225219726562, 456.349609375, 187.01585388183594, 491.0137023925781, 332.8813781738281, 245.76011657714844, 329.495849609375, 453.3402099609375, 164.53616333007812, 79.55696868896484, 201.88414001464844, 108.29670715332031, 364.1331787109375, 643.3442993164062, 266.06317138671875, -348.1686706542969, 419.8883056640625, 32.206634521484375, -359.22564697265625, 245.23587036132812, 26.283029556274414, 278.43475341796875, 55.516998291015625, 320.85113525390625, -32.26342010498047, 12.778121948242188, 255.09075927734375, 345.4607849121094, 275.601806640625, 367.0477600097656, -73.26690673828125, 277.59381103515625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000471.npy"}
|
|
{"epoch": 0.6916299559471366, "step": 472, "batch_size": 64, "mean": 270.7828063964844, "std": 322.0876770019531, "min": -316.18719482421875, "p10": -45.80129318237303, "median": 183.87969207763672, "p90": 788.0534790039063, "max": 1154.8863525390625, "pos_frac": 0.875, "sample": [-316.18719482421875, -73.07276916503906, 530.81494140625, -172.74118041992188, 59.59336853027344, 133.93466186523438, 223.60321044921875, -111.03179931640625, 95.04013061523438, -286.0344543457031, 156.35836791992188, 195.23583984375, 114.13381958007812, 327.9329528808594, 65.29776000976562, 797.7933349609375, 50.0458984375, 221.831787109375, 170.30792236328125, 187.37298583984375, 817.7373657226562, 81.31787109375, 267.12921142578125, 33.13542175292969, 184.43173217773438, 140.7548065185547, 117.12969970703125, 50.683631896972656, 209.78466796875, 614.50927734375, 29.855819702148438, 484.6697692871094, 223.93905639648438, 587.15869140625, 1154.8863525390625, 587.4054565429688, 216.8757781982422, 985.36376953125, 223.3902130126953, 164.9217529296875, 449.48577880859375, 686.9844970703125, 92.16140747070312, 261.32440185546875, -159.38050842285156, 843.667724609375, 1152.4521484375, 183.32765197753906, 765.3271484375, 3.4170989990234375, 891.0888671875, -31.852706909179688, 66.00874328613281, -51.779258728027344, 355.2345275878906, 326.7431945800781, 564.2258911132812, 439.8785705566406, 76.18475341796875, 164.56629943847656, 174.23646545410156, 98.84446716308594, 278.66888427734375, 153.9727783203125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000472.npy"}
|
|
{"epoch": 0.6930983847283406, "step": 473, "batch_size": 64, "mean": 183.41294860839844, "std": 322.2508850097656, "min": -496.2951354980469, "p10": -225.2379425048828, "median": 196.09993743896484, "p90": 453.50875244140633, "max": 1577.56689453125, "pos_frac": 0.765625, "sample": [17.393577575683594, 259.99658203125, 262.1869812011719, 438.6746826171875, -64.66452026367188, 369.8240966796875, 106.4512939453125, 238.26370239257812, 396.7188720703125, 78.04548645019531, 459.8662109375, 466.8622741699219, 74.10227966308594, 271.24884033203125, -60.73284912109375, -114.07816314697266, 232.89199829101562, 91.88746643066406, 250.7841796875, 319.78717041015625, -238.0230712890625, 170.68519592285156, 300.96484375, 175.1389923095703, 143.20762634277344, 28.900915145874023, 296.4610900878906, 529.139404296875, 174.6160888671875, 174.1392822265625, -463.0791320800781, 169.8807373046875, -318.59564208984375, 169.71119689941406, -496.2951354980469, -307.49371337890625, 1577.56689453125, 400.54498291015625, 379.7394104003906, 936.6261596679688, 311.905517578125, 233.24676513671875, 426.81927490234375, 345.55126953125, 189.0835418701172, 254.71151733398438, -120.72676849365234, -368.5528259277344, 67.57408142089844, 327.3605041503906, 430.9890441894531, 369.39996337890625, -195.40597534179688, -107.19107818603516, 203.1163330078125, -101.6303482055664, 568.7570190429688, -357.8341064453125, 369.6774597167969, 187.21969604492188, -92.06636047363281, 315.3217468261719, 5.916345596313477, 575.8401489257812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000473.npy"}
|
|
{"epoch": 0.6945668135095447, "step": 474, "batch_size": 64, "mean": 196.49615478515625, "std": 275.48638916015625, "min": -563.4605712890625, "p10": -125.49281997680663, "median": 168.21288299560547, "p90": 533.7492736816406, "max": 909.1593627929688, "pos_frac": 0.8125, "sample": [-292.6951904296875, 238.11273193359375, 34.10939407348633, 502.3642883300781, 425.0998840332031, 130.61544799804688, 633.748779296875, 1.2873516082763672, 150.71920776367188, 364.36712646484375, 445.9633483886719, 661.418701171875, 23.191804885864258, 17.899391174316406, 117.70931243896484, 375.4266662597656, 130.30825805664062, 1.1881961822509766, -294.63543701171875, 253.1227569580078, 23.883255004882812, 372.3885803222656, -138.6295166015625, -19.17428970336914, -27.242650985717773, 448.44970703125, 140.73388671875, 165.67605590820312, 302.26641845703125, 170.7497100830078, 131.34747314453125, -64.12033081054688, 172.76199340820312, 118.09868621826172, 391.5843505859375, 233.8544464111328, 530.4388427734375, 98.50567626953125, 18.486310958862305, -174.43289184570312, 394.41455078125, 75.35357666015625, 392.45947265625, 53.23565673828125, 39.27001190185547, 202.43923950195312, 522.42626953125, 823.0620727539062, -132.20957946777344, -305.8009338378906, 202.09783935546875, 535.1680297851562, 258.0676574707031, 344.2873229980469, 590.1195068359375, 207.44593811035156, 282.0768737792969, -109.82038116455078, 378.6698913574219, -82.92977905273438, 143.7562255859375, 601.5184936523438, -563.4605712890625, 909.1593627929688], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000474.npy"}
|
|
{"epoch": 0.6960352422907489, "step": 475, "batch_size": 64, "mean": 272.72491455078125, "std": 265.3132629394531, "min": -297.7701416015625, "p10": -40.518988800048824, "median": 279.59161376953125, "p90": 579.3848205566408, "max": 1051.0535888671875, "pos_frac": 0.78125, "sample": [-5.306854248046875, -3.306377410888672, 327.6946105957031, 129.0724334716797, -21.77490234375, 413.5604553222656, 914.5362548828125, 236.75857543945312, 763.1162719726562, 465.0020751953125, 210.85147094726562, 214.46725463867188, 431.61981201171875, 44.0675048828125, 301.34356689453125, -42.22479248046875, 519.66748046875, 85.13557434082031, 375.4119873046875, -36.538780212402344, 254.09422302246094, 329.06988525390625, 382.1705627441406, 79.94679260253906, -78.84976196289062, 244.18055725097656, 299.31640625, 386.1606140136719, 53.89716720581055, 402.5452880859375, 76.51227569580078, 453.4073791503906, -297.7701416015625, 411.3698425292969, 390.5024108886719, 593.7140502929688, 291.1799011230469, 349.15496826171875, 130.9733123779297, 884.3015747070312, 284.20147705078125, -10.829399108886719, 200.33309936523438, -102.46006774902344, 424.89544677734375, 171.1121063232422, 132.794677734375, -21.583511352539062, 274.98175048828125, 334.22784423828125, -43.841156005859375, 408.0042419433594, 424.97247314453125, -21.109771728515625, 719.5525512695312, -84.62397766113281, 185.84315490722656, 1051.0535888671875, 422.93731689453125, 706.1359252929688, 258.4505920410156, 545.949951171875, -86.15992736816406, 320.52392578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000475.npy"}
|
|
{"epoch": 0.697503671071953, "step": 476, "batch_size": 64, "mean": 231.2842254638672, "std": 232.137451171875, "min": -229.99200439453125, "p10": -23.18261661529541, "median": 183.99356079101562, "p90": 561.2708312988282, "max": 958.5745849609375, "pos_frac": 0.8125, "sample": [465.0195617675781, 1.569192886352539, -24.327478408813477, 176.97853088378906, 211.08079528808594, 372.3510437011719, 676.2460327148438, 343.8193664550781, 115.22366333007812, -142.36984252929688, 568.1061401367188, -59.69567108154297, 655.3275146484375, 405.880859375, 423.2724914550781, 483.090576171875, 369.3515625, 26.27220916748047, 180.79205322265625, 190.90536499023438, 290.3253479003906, 303.3118896484375, -4.4080963134765625, 445.8834228515625, 253.64266967773438, 368.4635009765625, -229.99200439453125, -34.05237579345703, 203.06072998046875, 68.09839630126953, 14.685798645019531, 131.851806640625, -40.351314544677734, -0.7643966674804688, 368.4700927734375, 78.41419982910156, 71.50846099853516, -7.298093795776367, 222.6009521484375, 99.41402435302734, -1.5260276794433594, 394.55853271484375, 145.48779296875, -94.92625427246094, 98.11394500732422, 182.3186492919922, 545.32177734375, 610.8869018554688, 958.5745849609375, 638.7446899414062, 113.32548522949219, 125.04255676269531, 159.10467529296875, 281.23260498046875, 115.91145324707031, 98.87550354003906, 378.6228332519531, 344.5804443359375, 99.90623474121094, 352.7632751464844, 643.6834716796875, 404.6716003417969, 185.66847229003906, -20.511272430419922], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000476.npy"}
|
|
{"epoch": 0.6989720998531571, "step": 477, "batch_size": 64, "mean": 212.0760498046875, "std": 267.9363708496094, "min": -714.6406860351562, "p10": -60.45518188476562, "median": 183.7178497314453, "p90": 578.3512084960938, "max": 853.5048828125, "pos_frac": 0.859375, "sample": [-54.87744140625, 328.49945068359375, 376.92962646484375, -53.61540985107422, 620.0164184570312, 273.85882568359375, 230.36807250976562, 276.34844970703125, 311.7876892089844, 317.7135925292969, 154.2257080078125, 82.15616607666016, 201.7128143310547, 471.7021789550781, 678.679443359375, 31.227869033813477, 423.2794189453125, 264.8959655761719, 823.5277099609375, 144.43081665039062, 181.6654052734375, 33.433990478515625, 10.087631225585938, 108.1581039428711, -144.64663696289062, 324.3580627441406, 231.78878784179688, 117.20498657226562, 25.871797561645508, -96.51988220214844, -209.4208221435547, 289.6417541503906, -714.6406860351562, -254.25747680664062, 641.92138671875, 126.472412109375, 187.33193969726562, 206.10848999023438, 457.5079345703125, 95.90484619140625, 441.88970947265625, 42.707035064697266, 386.71356201171875, 606.5217895507812, 90.26364135742188, 185.77029418945312, 139.79959106445312, 586.3630981445312, 372.1747741699219, 157.03463745117188, 559.6567993164062, -146.24496459960938, 47.617713928222656, 179.28302001953125, 0.94610595703125, 519.5540771484375, 147.56936645507812, -62.84564208984375, 42.481353759765625, 17.21792221069336, 46.053428649902344, 325.4393005371094, 853.5048828125, 512.556640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000477.npy"}
|
|
{"epoch": 0.7004405286343612, "step": 478, "batch_size": 64, "mean": 256.3935241699219, "std": 237.7212677001953, "min": -154.91693115234375, "p10": 14.496442413330081, "median": 220.839111328125, "p90": 582.4578552246095, "max": 1026.8953857421875, "pos_frac": 0.921875, "sample": [-66.50944519042969, 213.66551208496094, 284.810302734375, 90.12467193603516, 157.35299682617188, 296.26983642578125, 284.1697692871094, 2.8819732666015625, 55.77263641357422, 33.375946044921875, 134.59153747558594, 337.39068603515625, 59.047386169433594, 152.43186950683594, 75.12390899658203, 590.2906494140625, 382.2039489746094, 736.6069946289062, 200.6852569580078, 159.224853515625, -85.50227355957031, 86.57882690429688, -154.91693115234375, 270.5219421386719, 311.94097900390625, 709.9025268554688, 444.66473388671875, 172.05599975585938, -17.189178466796875, 174.89393615722656, 1026.8953857421875, 478.4002990722656, 284.56890869140625, 239.99127197265625, 263.61370849609375, 396.356689453125, 615.36962890625, 292.387451171875, 407.4066467285156, 32.77833557128906, 417.3800354003906, 31.740501403808594, 78.94183349609375, -63.024696350097656, 564.1813354492188, 347.859619140625, 117.51866912841797, 827.2596435546875, 188.2698211669922, 259.83209228515625, 437.0536804199219, 95.39473724365234, 238.49993896484375, 440.5317687988281, 48.76243591308594, 221.95535278320312, 388.6286926269531, 39.06709289550781, 836.8668823242188, 17.04346466064453, 219.72286987304688, 194.84226989746094, 317.22247314453125, 13.404861450195312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000478.npy"}
|
|
{"epoch": 0.7019089574155654, "step": 479, "batch_size": 64, "mean": 209.93270874023438, "std": 264.6244201660156, "min": -597.675537109375, "p10": -65.83391609191894, "median": 181.6835479736328, "p90": 548.6159545898438, "max": 890.5396728515625, "pos_frac": 0.796875, "sample": [50.024208068847656, 17.846397399902344, 541.2130126953125, 331.4411926269531, 271.64398193359375, 495.6910400390625, -22.893329620361328, -146.03298950195312, 332.88360595703125, 9.034513473510742, 378.2332763671875, 54.578426361083984, -44.29425811767578, 332.6688232421875, 583.772705078125, 156.9876708984375, 250.54209899902344, 59.207672119140625, 23.42980194091797, -0.6673851013183594, -58.61716079711914, 421.86541748046875, -33.10456848144531, 292.369140625, 890.5396728515625, 686.3576049804688, 541.5559692382812, 331.0246276855469, 478.51373291015625, 116.231201171875, 384.3360290527344, 194.30325317382812, 298.8287353515625, 507.4425048828125, -219.56698608398438, 585.8118896484375, -165.16757202148438, 169.0638427734375, 135.66476440429688, 44.3543815612793, 303.064453125, 551.6416625976562, 397.8570556640625, 301.9770812988281, 81.46723937988281, -597.675537109375, 247.88916015625, 107.77850341796875, -68.92681121826172, 68.5672607421875, 121.05659484863281, -56.983192443847656, -70.57968139648438, -147.353759765625, 242.9980010986328, 21.035438537597656, 398.95465087890625, 80.03616333007812, 115.2843017578125, 853.166748046875, 207.08226013183594, 568.2720336914062, 300.18017578125, 131.78628540039062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000479.npy"}
|
|
{"epoch": 0.7033773861967695, "step": 480, "batch_size": 64, "mean": 179.13446044921875, "std": 269.76220703125, "min": -482.31353759765625, "p10": -147.1012008666992, "median": 136.8493194580078, "p90": 512.8842407226564, "max": 921.8410034179688, "pos_frac": 0.78125, "sample": [-77.65721130371094, 31.56307029724121, -127.28414916992188, 105.86541748046875, 75.43011474609375, 237.12484741210938, 215.9188995361328, 528.138427734375, 129.77706909179688, 523.25, 210.98123168945312, 684.29052734375, 104.19090270996094, -150.8453826904297, 92.4464111328125, -266.9643859863281, 24.799388885498047, 432.6500549316406, 123.21267700195312, 578.4202880859375, 77.5158920288086, -161.35279846191406, 17.582351684570312, 318.7635803222656, 78.62422943115234, 218.8240203857422, 492.88262939453125, 59.67805480957031, 441.3375549316406, -157.93678283691406, 475.64801025390625, 662.3949584960938, 442.4975280761719, -30.565174102783203, -425.9595031738281, 281.5355529785156, 321.7663879394531, 20.241470336914062, 336.9688720703125, -62.96868896484375, -257.2039794921875, -482.31353759765625, 469.2153625488281, 462.1856384277344, 921.8410034179688, 158.22671508789062, 83.93570709228516, 365.55450439453125, 4.224170684814453, 46.586151123046875, -4.388275146484375, -138.36477661132812, 72.3570327758789, 147.92202758789062, 143.92156982421875, 156.20664978027344, 281.9898681640625, 113.703125, 391.126220703125, 521.4563598632812, 448.41046142578125, -41.90985870361328, 264.49981689453125, 452.6379089355469], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000480.npy"}
|
|
{"epoch": 0.7048458149779736, "step": 481, "batch_size": 64, "mean": 248.92288208007812, "std": 261.1285400390625, "min": -329.424072265625, "p10": -46.27878952026364, "median": 210.65425872802734, "p90": 582.4370239257813, "max": 1011.4415283203125, "pos_frac": 0.84375, "sample": [463.0674133300781, -329.424072265625, 474.1800537109375, 129.08633422851562, 147.0093231201172, 130.49684143066406, -58.508941650390625, 209.68341064453125, 292.4348449707031, 617.3681030273438, -17.74176788330078, -189.8162841796875, 124.25587463378906, -174.00732421875, 243.15455627441406, 311.75384521484375, 35.33099365234375, 441.97235107421875, 749.9154052734375, 30.316909790039062, 176.74481201171875, 365.31317138671875, 235.84112548828125, 451.3847961425781, 243.09149169921875, 165.36337280273438, 77.3721923828125, 440.9086608886719, 108.49392700195312, -189.99563598632812, -90.04449462890625, 399.9443359375, 179.81982421875, -240.87933349609375, 512.1160278320312, 348.64959716796875, -12.375144958496094, 185.06825256347656, 559.8245849609375, 273.18475341796875, -13.264345169067383, 211.62510681152344, 334.037109375, 585.5418701171875, 107.7756118774414, 173.69009399414062, 756.741943359375, 157.19094848632812, 1011.4415283203125, 648.5114135742188, 356.8383483886719, 9.717788696289062, 152.93038940429688, 454.9961242675781, 169.07989501953125, 271.40679931640625, 43.051143646240234, 186.830322265625, 575.1923828125, 372.9076232910156, 721.5875854492188, 56.40129089355469, 273.01629638671875, 493.46209716796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000481.npy"}
|
|
{"epoch": 0.7063142437591777, "step": 482, "batch_size": 64, "mean": 168.63145446777344, "std": 255.32916259765625, "min": -679.9314575195312, "p10": -75.73720397949218, "median": 134.23970794677734, "p90": 470.1533447265625, "max": 943.0126342773438, "pos_frac": 0.78125, "sample": [401.3948669433594, -87.54426574707031, 324.6061096191406, -15.149078369140625, 484.67333984375, 471.2098083496094, -174.35577392578125, 457.49005126953125, 457.0633544921875, 176.58160400390625, 176.7188720703125, 218.11068725585938, -1.045440673828125, 3.963836669921875, 553.7329711914062, 446.41473388671875, 201.74554443359375, 149.3651580810547, 134.7252655029297, 101.35487365722656, -63.08842849731445, 103.79757690429688, 268.95916748046875, 472.7831726074219, -212.49964904785156, 12.68450927734375, 121.34996795654297, 366.10687255859375, 263.0584716796875, 81.77885437011719, 285.43951416015625, 285.35882568359375, 467.6882629394531, 132.54373168945312, -77.35952758789062, 78.08282470703125, 290.5448913574219, 39.954010009765625, 715.5184936523438, -262.3072509765625, 434.8058776855469, 56.52086639404297, -29.59209632873535, 55.53412628173828, 232.44972229003906, 57.739959716796875, -679.9314575195312, 284.2255859375, 943.0126342773438, -3.8812942504882812, -21.271453857421875, 16.127120971679688, 15.22760009765625, 211.52391052246094, -243.22662353515625, 231.05661010742188, 133.754150390625, 579.8033447265625, -71.9517822265625, 177.4252166748047, 23.545372009277344, 12.675498962402344, 453.38507080078125, 72.00375366210938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000482.npy"}
|
|
{"epoch": 0.7077826725403817, "step": 483, "batch_size": 64, "mean": 281.00067138671875, "std": 285.5683898925781, "min": -244.35256958007812, "p10": -44.78669357299804, "median": 238.82766723632812, "p90": 645.5563110351562, "max": 1416.13134765625, "pos_frac": 0.859375, "sample": [-48.94670104980469, 108.38400268554688, 529.5317993164062, 427.39776611328125, -83.29020690917969, 583.2581176757812, 189.57550048828125, 235.24114990234375, 527.1055297851562, 508.28265380859375, 426.96942138671875, 501.42572021484375, 385.857421875, 40.14064025878906, 12.408409118652344, 78.94096374511719, 719.1140747070312, 276.3192138671875, -174.55130004882812, 311.33154296875, 114.8249740600586, -35.08000946044922, 750.8271484375, 644.2490234375, 162.2789764404297, 75.31185150146484, 855.1005859375, 394.2586975097656, 47.386444091796875, 586.4623413085938, -75.70191955566406, 650.1819458007812, 43.902000427246094, 699.8082885742188, 90.88993835449219, 646.1165771484375, 367.76861572265625, 425.7362060546875, 128.72586059570312, 350.314208984375, 242.4141845703125, 362.4986267089844, 180.58958435058594, -11.605697631835938, 91.40738677978516, 156.8582763671875, 221.28573608398438, 135.50511169433594, 400.8687438964844, -142.2490997314453, 329.03863525390625, -244.35256958007812, 142.03099060058594, 1416.13134765625, 65.60380554199219, 232.1235809326172, 341.745361328125, 246.15069580078125, 419.08428955078125, -53.984458923339844, 308.1536560058594, 460.533935546875, 172.213134765625, 34.141963958740234], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000483.npy"}
|
|
{"epoch": 0.7092511013215859, "step": 484, "batch_size": 64, "mean": 242.32603454589844, "std": 243.64588928222656, "min": -255.42369079589844, "p10": -12.88018360137939, "median": 202.0984115600586, "p90": 578.0735900878907, "max": 1008.0640869140625, "pos_frac": 0.875, "sample": [318.76373291015625, 145.0492706298828, 589.0718383789062, 220.97889709472656, -73.93629455566406, 3.1625518798828125, 284.55450439453125, 85.15752410888672, 601.3442993164062, 166.08331298828125, 64.97560119628906, 118.93778228759766, 297.14593505859375, 280.610107421875, 458.62420654296875, 314.6351318359375, 221.715576171875, 124.69845581054688, -143.74346923828125, -255.42369079589844, 128.51641845703125, 115.13038635253906, 329.1886901855469, 211.93466186523438, 1008.0640869140625, 34.51374053955078, 213.51339721679688, 425.684326171875, 2.552196502685547, -53.12841796875, -36.08883285522461, 180.7174072265625, 487.267822265625, 667.5899047851562, -14.875358581542969, 250.21035766601562, 99.492431640625, 152.74330139160156, 208.04721069335938, 35.495338439941406, 77.15372467041016, 36.491783142089844, 548.274169921875, 283.22833251953125, 552.4110107421875, 81.80126953125, 424.6737365722656, 487.4283447265625, 60.177146911621094, -25.59600067138672, -8.224775314331055, 133.89459228515625, 600.37890625, 474.22723388671875, 196.1496124267578, 476.24639892578125, 794.385009765625, 278.4866943359375, 528.6050415039062, 5.967687606811523, 626.6121826171875, 91.17494201660156, 23.3638973236084, 492.5806579589844], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000484.npy"}
|
|
{"epoch": 0.71071953010279, "step": 485, "batch_size": 64, "mean": 238.25384521484375, "std": 269.02093505859375, "min": -436.0274658203125, "p10": -36.88506469726561, "median": 198.80915069580078, "p90": 626.0136596679689, "max": 874.176513671875, "pos_frac": 0.859375, "sample": [267.9656677246094, 394.749755859375, 105.04487609863281, 257.68743896484375, 310.24896240234375, 103.81592559814453, 391.1334533691406, 12.632110595703125, 258.33868408203125, 359.24761962890625, 155.94259643554688, 13.619560241699219, 464.8957214355469, 461.85357666015625, 147.56463623046875, 805.75390625, -294.37933349609375, 12.806020736694336, 668.4801025390625, 593.5303955078125, 485.4416198730469, 510.5600891113281, 102.0640869140625, 291.9499816894531, 869.4329223632812, 874.176513671875, -25.597511291503906, 11.236825942993164, 239.41473388671875, -148.79681396484375, -73.5528335571289, 98.90858459472656, -41.72258758544922, 15.626014709472656, 239.345703125, 300.9951171875, 499.7962646484375, 342.08306884765625, 107.86000061035156, 391.3623046875, 70.32183837890625, 113.7366943359375, 698.9149169921875, 188.880615234375, -130.73031616210938, 268.9532775878906, 47.48442840576172, 162.5367431640625, 208.73768615722656, -436.0274658203125, 742.20166015625, 462.0587463378906, 59.513980865478516, -14.0673828125, 133.8771209716797, -57.55769348144531, 639.93505859375, 42.43009948730469, 57.36693572998047, 254.6451416015625, 184.5411376953125, 74.23326110839844, 473.3332214355469, 421.4108581542969], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000485.npy"}
|
|
{"epoch": 0.7121879588839941, "step": 486, "batch_size": 64, "mean": 245.2418975830078, "std": 307.4638671875, "min": -407.5664367675781, "p10": -103.44321975708007, "median": 291.80615234375, "p90": 605.253466796875, "max": 968.9013671875, "pos_frac": 0.78125, "sample": [733.116943359375, 297.1549987792969, 82.23750305175781, 88.50772094726562, 286.4573059082031, 42.69783020019531, 95.37820434570312, -105.3593978881836, 527.928466796875, -98.97213745117188, 152.52615356445312, 442.3809814453125, 494.3004150390625, -56.53986358642578, -395.2911071777344, 929.9737548828125, 569.2279052734375, 304.2774963378906, -176.9111785888672, 594.0662841796875, 620.0946655273438, -222.08616638183594, 448.292236328125, -26.803192138671875, -97.30796813964844, -2.3628158569335938, 408.58612060546875, 401.0978088378906, 61.311065673828125, 457.1556701660156, 525.5413818359375, 85.81195831298828, -327.3524169921875, -300.6130676269531, 116.78895568847656, 320.8260192871094, 41.620452880859375, 610.0479736328125, 451.09747314453125, 383.5438537597656, 307.32049560546875, 368.2015075683594, 373.5770263671875, -83.85800170898438, 463.1144714355469, 829.35693359375, 572.9569091796875, 398.9671936035156, 299.0772399902344, 717.17333984375, -47.794776916503906, 238.46951293945312, 196.034912109375, 299.63262939453125, 420.42437744140625, 126.57267761230469, 66.60298919677734, 325.1390075683594, -407.5664367675781, 155.6598663330078, 276.3829345703125, 2.77008056640625, 65.91844940185547, 968.9013671875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000486.npy"}
|
|
{"epoch": 0.7136563876651982, "step": 487, "batch_size": 64, "mean": 224.89999389648438, "std": 233.16722106933594, "min": -464.763916015625, "p10": -83.42331619262694, "median": 243.94121551513672, "p90": 481.86084899902346, "max": 788.5643920898438, "pos_frac": 0.8125, "sample": [335.9801025390625, 443.158447265625, 158.9432373046875, 139.28097534179688, 788.5643920898438, 485.35272216796875, 471.2767333984375, 340.0984802246094, 573.5020751953125, 12.520088195800781, 239.77728271484375, -164.36410522460938, 520.6787719726562, -281.0204162597656, 89.04332733154297, 462.40716552734375, 438.71136474609375, -143.56961059570312, 409.1958312988281, 644.6229858398438, 368.96051025390625, 315.0303955078125, -146.15399169921875, 163.07168579101562, 64.73590850830078, 210.60601806640625, 51.47405242919922, 297.8481750488281, 466.15521240234375, -9.810373306274414, 248.1051483154297, 59.71382141113281, 222.7359619140625, 393.17681884765625, 142.54940795898438, -91.58967590332031, 194.45724487304688, -464.763916015625, 396.7508544921875, 332.8288269042969, 299.56719970703125, 248.86642456054688, 229.4713897705078, -64.36847686767578, 282.53546142578125, 334.4051513671875, -39.92652893066406, 180.06918334960938, 309.025146484375, 128.1754150390625, 147.5852508544922, 163.73138427734375, 360.6510009765625, 353.5311279296875, 143.8610382080078, -50.1988525390625, 161.57955932617188, 522.2758178710938, 482.55133056640625, 480.2497253417969, -61.074432373046875, 426.7998352050781, -123.85977172851562, 298.05340576171875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000487.npy"}
|
|
{"epoch": 0.7151248164464024, "step": 488, "batch_size": 64, "mean": 224.94769287109375, "std": 196.17205810546875, "min": -148.69183349609375, "p10": -17.675538635253893, "median": 224.9545135498047, "p90": 470.52380676269536, "max": 778.5567016601562, "pos_frac": 0.859375, "sample": [-0.41739654541015625, 103.53384399414062, 482.076416015625, 117.14643859863281, 664.7618408203125, 5.9785614013671875, 407.1941223144531, 35.304603576660156, 12.533782958984375, 235.45712280273438, 200.73855590820312, 171.9803924560547, 351.9928894042969, 374.0523681640625, 334.9743957519531, 384.2298583984375, 36.701934814453125, -23.895410537719727, -141.3024139404297, 96.63140869140625, 178.9883575439453, -81.37374114990234, 316.63909912109375, 276.5260314941406, 46.761077880859375, 182.3201904296875, 162.57244873046875, 286.88922119140625, 778.5567016601562, 197.4730987548828, 214.451904296875, 328.3052062988281, -76.2900161743164, 314.3447265625, 520.1080932617188, 196.70326232910156, 475.1573181152344, 327.6111145019531, 339.839111328125, 244.16786193847656, 289.82940673828125, 459.7122802734375, 290.9178161621094, 56.54139709472656, 256.1608581542969, 159.15713500976562, 334.981201171875, 530.691650390625, -148.69183349609375, -57.683204650878906, 195.32785034179688, 391.2597351074219, 282.8878479003906, 129.59368896484375, 291.710205078125, 390.0230712890625, -3.162504196166992, 262.9761657714844, 31.416156768798828, 58.782188415527344, -85.04962158203125, 390.2390441894531, 186.62466430664062, 622.9827880859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000488.npy"}
|
|
{"epoch": 0.7165932452276065, "step": 489, "batch_size": 64, "mean": 211.18069458007812, "std": 228.2081298828125, "min": -103.02001190185547, "p10": -0.5226108551025286, "median": 132.77862548828125, "p90": 487.5967803955079, "max": 888.5758056640625, "pos_frac": 0.890625, "sample": [347.59710693359375, -69.77592468261719, 28.224533081054688, 82.58363342285156, 52.574851989746094, 133.75054931640625, 59.16744613647461, 425.76953125, 75.7493896484375, 169.02117919921875, 21.459232330322266, 125.15499877929688, 369.865966796875, 132.06475830078125, 275.6472473144531, -68.53034973144531, 18.400188446044922, 260.76165771484375, 101.2519760131836, 617.5374755859375, 134.3962860107422, 97.47128295898438, 65.38069152832031, 226.66812133789062, 73.0101318359375, 94.94482421875, 52.92424011230469, 67.63247680664062, 100.73820495605469, 147.6940460205078, 137.56964111328125, 58.90571594238281, 25.46168327331543, -24.202512741088867, 734.193115234375, 471.45587158203125, 188.44140625, 90.77497863769531, 257.43963623046875, 133.49249267578125, 352.7420654296875, -7.603385925292969, 698.2268676757812, -4.885280609130859, 494.5143127441406, 362.6026916503906, 361.44915771484375, 742.9564819335938, 131.9103240966797, 79.11672973632812, 888.5758056640625, 245.80300903320312, 229.8248748779297, 816.7798461914062, -103.02001190185547, 104.76348114013672, 458.0771179199219, 291.2877502441406, 461.5788269042969, 42.658477783203125, 9.656951904296875, 449.3516845703125, 215.32919311523438, -98.7998046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000489.npy"}
|
|
{"epoch": 0.7180616740088106, "step": 490, "batch_size": 64, "mean": 233.45465087890625, "std": 295.6266174316406, "min": -538.7109375, "p10": -83.18940353393549, "median": 163.6039047241211, "p90": 673.4225463867189, "max": 964.016845703125, "pos_frac": 0.796875, "sample": [57.7445068359375, -538.7109375, 749.9533081054688, -104.85376739501953, 112.5413818359375, 57.782005310058594, 158.0751953125, 411.2052001953125, 450.171630859375, -154.18331909179688, 45.85133361816406, -32.63922119140625, 596.851806640625, 123.53662109375, 462.1448974609375, 28.31916046142578, -175.715576171875, 193.76373291015625, 491.98779296875, -171.85464477539062, -5.425327301025391, 103.93673706054688, -25.72657012939453, 67.44477844238281, -158.09593200683594, 701.59619140625, 159.0141143798828, 220.205810546875, 136.74664306640625, 444.423828125, 92.29755401611328, 600.9098510742188, 129.6304931640625, 488.0306701660156, 168.19369506835938, 59.725669860839844, 689.9930419921875, 281.810791015625, 964.016845703125, 28.346698760986328, 337.6805419921875, 236.98617553710938, 634.758056640625, 720.2861328125, 199.37594604492188, 781.3901977539062, -276.469970703125, 316.7582702636719, 30.230133056640625, 409.19219970703125, 198.30128479003906, -2.603677749633789, 817.439453125, 145.24005126953125, 328.6404113769531, 490.31463623046875, -28.562469482421875, -21.327747344970703, 114.44497680664062, 598.4822387695312, 52.08530044555664, 466.2852478027344, 311.14447021484375, 171.97860717773438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000490.npy"}
|
|
{"epoch": 0.7195301027900147, "step": 491, "batch_size": 64, "mean": 174.71432495117188, "std": 249.136474609375, "min": -384.7040100097656, "p10": -93.5887664794922, "median": 167.4868392944336, "p90": 505.07870788574235, "max": 899.6659545898438, "pos_frac": 0.71875, "sample": [288.21563720703125, -94.03186798095703, 547.7449951171875, 221.69512939453125, 410.26116943359375, 157.41831970214844, 217.95770263671875, -13.992704391479492, 179.7923583984375, 15.13006591796875, 151.17005920410156, 220.57679748535156, -49.315940856933594, -25.250228881835938, 122.45852661132812, 14.323322296142578, 282.9302978515625, 190.03997802734375, -73.55674743652344, 300.122314453125, 153.34933471679688, -92.55486297607422, 522.9542236328125, -156.59317016601562, 202.5295867919922, 136.88088989257812, 450.9366455078125, -296.39862060546875, 463.3691711425781, 69.27510833740234, 646.23486328125, 300.5758972167969, -384.7040100097656, 179.398681640625, 108.07121276855469, 177.55535888671875, -3.7474746704101562, 557.2098388671875, 254.71852111816406, 544.8886108398438, 216.60646057128906, -240.7015380859375, 850.4364013671875, -20.045652389526367, -31.981155395507812, 245.99574279785156, -41.11041259765625, 187.19309997558594, -32.482818603515625, 366.18511962890625, 105.68002319335938, -172.93690490722656, 436.734130859375, 106.51329040527344, 104.98462677001953, 269.5987548828125, 331.0993957519531, 416.3897399902344, 88.40275573730469, -28.944061279296875, 899.6659545898438, 93.80980682373047, 288.4190673828125, -155.43472290039062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000491.npy"}
|
|
{"epoch": 0.7209985315712188, "step": 492, "batch_size": 64, "mean": 138.69735717773438, "std": 266.39569091796875, "min": -468.8843688964844, "p10": -182.02791442871091, "median": 92.8935775756836, "p90": 513.7326721191407, "max": 741.3782958984375, "pos_frac": 0.75, "sample": [69.6989974975586, 58.53016662597656, 267.58837890625, -136.68899536132812, 110.8291015625, 60.55839538574219, 57.355072021484375, 373.2095947265625, 34.15565872192383, -92.84022521972656, 30.281517028808594, 343.1397705078125, -152.99822998046875, -80.19750213623047, -466.16217041015625, 527.54736328125, 36.890342712402344, 322.7269287109375, 38.75054931640625, 655.7342529296875, -38.381500244140625, 148.43646240234375, 237.87265014648438, -43.112335205078125, 343.9214782714844, 199.13385009765625, 363.7117919921875, -50.38258361816406, 105.29649353027344, 73.61184692382812, -48.960968017578125, 39.533103942871094, 71.76519012451172, -280.1138916015625, -256.93145751953125, 407.6584777832031, 681.9603881835938, 121.52536010742188, -468.8843688964844, 69.5091781616211, 741.3782958984375, -310.2651672363281, 486.29736328125, 6.9112091064453125, -143.6450958251953, 525.4906616210938, 183.23829650878906, 65.26931762695312, 670.454345703125, 99.80548095703125, 281.4737854003906, 69.32698059082031, 277.58782958984375, -194.46920776367188, 355.4044189453125, 250.1702423095703, 313.28692626953125, 161.6075897216797, 85.98167419433594, 321.5777893066406, -262.12335205078125, 617.4821166992188, 183.49908447265625, 355.612548828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000492.npy"}
|
|
{"epoch": 0.7224669603524229, "step": 493, "batch_size": 64, "mean": 172.2346954345703, "std": 217.24761962890625, "min": -265.2001647949219, "p10": -45.8697280883789, "median": 163.29092407226562, "p90": 464.98071289062506, "max": 766.77099609375, "pos_frac": 0.78125, "sample": [203.99598693847656, 72.75902557373047, 368.4914245605469, 220.5286865234375, -47.55010986328125, 766.77099609375, 430.101318359375, -9.713836669921875, 304.8428039550781, -194.4251708984375, 312.2967224121094, 79.01136779785156, 470.5537109375, 184.37945556640625, 144.4696502685547, 157.47549438476562, 31.037811279296875, 188.27560424804688, 193.1334228515625, 40.337249755859375, 182.89292907714844, 297.3170471191406, 390.06292724609375, 226.4608154296875, 150.3421630859375, 187.61175537109375, 451.97705078125, -156.79293823242188, -25.472253799438477, -131.96810913085938, 15.650543212890625, -265.2001647949219, 318.44818115234375, 4.4888153076171875, 489.1786193847656, 64.87850189208984, 475.1240234375, 503.013916015625, 48.39177703857422, -41.94883728027344, -29.079010009765625, 254.8924560546875, 394.13824462890625, 451.17510986328125, 175.32916259765625, 616.4024658203125, -4.348394393920898, 210.64230346679688, 695.2017822265625, 352.5018310546875, 182.38296508789062, 134.02633666992188, -178.65122985839844, 169.10635375976562, 5.74053955078125, -145.39736938476562, -36.052345275878906, 66.05360412597656, 337.50286865234375, 149.58497619628906, 44.465003967285156, 25.011077880859375, -33.84307861328125, 85.00630187988281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000493.npy"}
|
|
{"epoch": 0.723935389133627, "step": 494, "batch_size": 64, "mean": 256.29547119140625, "std": 224.6760711669922, "min": -346.2223205566406, "p10": 8.292983245849612, "median": 245.71573638916016, "p90": 563.8530029296875, "max": 776.62255859375, "pos_frac": 0.90625, "sample": [-1.9645748138427734, 193.7890167236328, 206.23204040527344, 322.7591247558594, 776.62255859375, 682.96630859375, -282.3046875, 509.11322021484375, 132.11465454101562, 56.474639892578125, 381.4042053222656, 249.9244842529297, 330.8849792480469, 406.299072265625, 54.50585174560547, 582.4149780273438, 370.3096923828125, -97.6112060546875, 407.32843017578125, 419.6776428222656, 93.26459503173828, -346.2223205566406, 417.8156433105469, 571.804443359375, 471.0552978515625, 127.07329559326172, 338.74603271484375, 567.2735595703125, 236.6859893798828, 92.32086181640625, -165.48757934570312, 30.558273315429688, 225.9413604736328, 11.090873718261719, 555.8717041015625, 248.212158203125, 404.582763671875, 91.53148651123047, 256.70172119140625, 719.2965087890625, 612.9783325195312, 220.77345275878906, -30.160415649414062, 51.85942840576172, 176.3092041015625, 268.15045166015625, 500.88433837890625, 289.9707336425781, 139.84652709960938, 357.7711486816406, 270.1204528808594, 7.0938873291015625, 445.08984375, 170.43020629882812, 225.89239501953125, 113.18499755859375, 383.4194641113281, 309.5892639160156, 213.47732543945312, 348.643798828125, 229.18215942382812, 243.2193145751953, 182.83331298828125, 23.294326782226562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000494.npy"}
|
|
{"epoch": 0.7254038179148311, "step": 495, "batch_size": 64, "mean": 245.58688354492188, "std": 263.2423400878906, "min": -209.31739807128906, "p10": -69.83875274658203, "median": 226.66244506835938, "p90": 598.7249755859375, "max": 1030.8726806640625, "pos_frac": 0.828125, "sample": [288.73516845703125, -72.56159973144531, 338.4417419433594, -20.95489501953125, 30.81427001953125, 140.726318359375, -159.2602996826172, 47.420013427734375, 303.37750244140625, 28.048385620117188, 202.8124542236328, 565.2633666992188, 566.53857421875, -209.31739807128906, 261.03753662109375, 21.4075870513916, -63.485443115234375, 514.932861328125, 78.80972290039062, 478.1110534667969, 134.10101318359375, -153.707763671875, 292.8019714355469, 132.40728759765625, 232.34677124023438, 603.847900390625, 520.4611206054688, 357.6501159667969, 100.02892303466797, 402.8121337890625, 301.4393310546875, 318.59857177734375, 257.2986755371094, 207.91683959960938, 237.4961700439453, 626.9435424804688, 586.771484375, -122.43952941894531, 735.3150634765625, 660.890380859375, 417.5125427246094, 95.15281677246094, 104.52347564697266, 171.593017578125, 260.760498046875, -19.064727783203125, 14.099960327148438, 306.21270751953125, 50.91655731201172, -146.0301513671875, 122.27136993408203, -3.9896774291992188, 266.18414306640625, 220.97811889648438, 519.5830078125, 107.86747741699219, 122.88186645507812, 514.3945922851562, -130.06996154785156, 846.4339599609375, 666.05859375, 234.99769592285156, 1030.8726806640625, 169.54312133789062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000495.npy"}
|
|
{"epoch": 0.7268722466960352, "step": 496, "batch_size": 64, "mean": 170.2359619140625, "std": 217.91583251953125, "min": -291.96295166015625, "p10": -63.34344329833984, "median": 155.62442016601562, "p90": 470.79795227050784, "max": 640.2535400390625, "pos_frac": 0.78125, "sample": [70.34939575195312, -124.12056732177734, 560.8741455078125, 598.022705078125, -202.84274291992188, 582.3456420898438, 164.65072631835938, 241.466552734375, 470.9213562011719, 157.65182495117188, 188.2418670654297, 61.56122589111328, 280.8306884765625, 168.81243896484375, 94.43645477294922, 153.59701538085938, 32.60736083984375, 242.25967407226562, -248.80987548828125, 378.04052734375, 456.3668212890625, 378.9309387207031, 640.2535400390625, 218.3544921875, 547.33447265625, 54.36252212524414, -64.45982360839844, 275.16265869140625, 73.28228759765625, 40.47650146484375, 166.88174438476562, -13.786054611206055, 231.33956909179688, 294.5959167480469, 40.029014587402344, -60.738555908203125, 212.14361572265625, -29.93614959716797, -291.96295166015625, 139.74989318847656, -183.31149291992188, -60.465057373046875, 275.3607482910156, 51.956275939941406, 124.34516906738281, 516.05908203125, -60.04890441894531, 380.0953369140625, 470.510009765625, 406.9100646972656, 48.660430908203125, -44.25492858886719, 407.49090576171875, 132.1753387451172, 75.80343627929688, -9.149795532226562, 198.71421813964844, 362.484130859375, 68.0176010131836, 80.08057403564453, 225.12814331054688, 364.6003723144531, 51.27935028076172, -166.61639404296875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000496.npy"}
|
|
{"epoch": 0.7283406754772394, "step": 497, "batch_size": 64, "mean": 164.8284912109375, "std": 211.4091339111328, "min": -335.25274658203125, "p10": -94.82327499389648, "median": 160.4282455444336, "p90": 468.66783447265624, "max": 612.5396118164062, "pos_frac": 0.78125, "sample": [-67.21382141113281, 500.12701416015625, 372.96441650390625, 525.0902709960938, 9.210977554321289, 111.86090087890625, 404.07305908203125, -101.76438903808594, 0.342071533203125, -52.72846984863281, 322.112060546875, 329.77606201171875, 612.5396118164062, 215.24317932128906, 37.550567626953125, 7.009765625, 363.7664489746094, 117.32991027832031, 348.0816345214844, 188.53549194335938, 56.36592102050781, 58.38416290283203, 466.4244079589844, 266.6614990234375, -50.04679870605469, 60.35809326171875, 317.2259216308594, -87.88453674316406, 34.66209411621094, -79.81700134277344, 160.0262908935547, 100.3875732421875, -27.915935516357422, 379.3179931640625, 89.02783203125, 543.5933837890625, 136.77386474609375, 575.8189697265625, 160.8302001953125, 42.60182189941406, -11.161293029785156, 193.02598571777344, 24.754072189331055, 421.89752197265625, -108.61491394042969, 210.38143920898438, 229.529541015625, 166.92532348632812, 294.194091796875, 333.66864013671875, 298.2945251464844, 171.95115661621094, 469.6293029785156, 125.14781951904297, -97.7970199584961, 230.36721801757812, -225.7433319091797, 279.9726867675781, -153.3131561279297, 39.226593017578125, 481.4916076660156, -335.25274658203125, 197.4556121826172, -133.71018981933594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000497.npy"}
|
|
{"epoch": 0.7298091042584435, "step": 498, "batch_size": 64, "mean": 205.400390625, "std": 263.642578125, "min": -484.3830871582031, "p10": -102.98683853149413, "median": 201.63510131835938, "p90": 510.12695922851566, "max": 801.818359375, "pos_frac": 0.796875, "sample": [-184.12559509277344, 240.1455078125, 219.91213989257812, 64.22909545898438, 77.39120483398438, 261.9790954589844, 513.2833862304688, 518.8431396484375, 267.3382263183594, 131.85498046875, 291.55224609375, 627.0262451171875, 102.24374389648438, 365.5162353515625, -350.04449462890625, 464.634765625, 323.8359375, 612.8150024414062, 40.71100616455078, 423.296630859375, 43.63933563232422, 420.7461242675781, -59.235809326171875, 217.60073852539062, -104.97549438476562, 145.98504638671875, 139.17112731933594, 49.18531036376953, -70.81855010986328, 71.87162017822266, 129.12059020996094, 349.03802490234375, -484.3830871582031, 741.7245483398438, 181.4368896484375, 28.98370361328125, -98.34664154052734, 502.761962890625, 496.65478515625, 457.900634765625, 801.818359375, -79.56694793701172, 3.7305908203125, 59.39921188354492, 477.9643859863281, 278.98577880859375, -91.78146362304688, 293.36767578125, 404.9277648925781, 367.7082214355469, 246.72642517089844, -127.51702880859375, 115.14236450195312, 185.66946411132812, -133.83335876464844, 8.388214111328125, 419.59307861328125, 146.13058471679688, 368.3272399902344, 297.03948974609375, -51.42003631591797, 358.7930908203125, 798.9165649414062, -173.385009765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000498.npy"}
|
|
{"epoch": 0.7312775330396476, "step": 499, "batch_size": 64, "mean": 255.0259552001953, "std": 268.60174560546875, "min": -493.9228515625, "p10": -110.0764236450195, "median": 275.7036437988281, "p90": 572.2710083007813, "max": 928.04443359375, "pos_frac": 0.84375, "sample": [574.8250732421875, 88.94013977050781, 209.2857666015625, -318.32635498046875, 23.18360137939453, 468.8377380371094, 267.30633544921875, 213.97674560546875, 117.31456756591797, -84.70938110351562, 325.60870361328125, 236.11473083496094, 142.96710205078125, 566.3115234375, 320.0884704589844, 385.3337707519531, 374.926025390625, -120.94801330566406, 928.04443359375, 235.7772674560547, 187.85342407226562, 112.19956970214844, 537.970947265625, 121.45539093017578, -493.9228515625, 476.0782165527344, 265.2646789550781, 284.1009521484375, -236.41896057128906, 306.6396179199219, 187.9427947998047, 26.69769287109375, 364.2812194824219, 180.55630493164062, 181.9539031982422, 110.4511489868164, 447.331298828125, 588.181640625, 395.5840148925781, -139.08164978027344, 23.365644454956055, 414.7193298339844, 532.7535400390625, 382.2090148925781, 566.0537719726562, 214.16281127929688, 608.0407104492188, 739.979248046875, 383.190185546875, 124.91287994384766, 303.5111389160156, 554.7257080078125, 292.5403137207031, 386.1639404296875, 2.462535858154297, 406.29705810546875, 526.574951171875, 305.2023620605469, -246.431884765625, -36.92091369628906, -172.03004455566406, 580.0101318359375, -39.398231506347656, 609.5887451171875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000499.npy"}
|
|
{"epoch": 0.7327459618208517, "step": 500, "batch_size": 64, "mean": 221.58847045898438, "std": 275.05316162109375, "min": -342.06597900390625, "p10": -69.97883758544921, "median": 216.10433959960938, "p90": 603.2657836914063, "max": 996.477294921875, "pos_frac": 0.796875, "sample": [-8.832643508911133, -70.94642639160156, -62.93395233154297, 996.477294921875, -222.77072143554688, 349.5943908691406, 341.7574157714844, 364.14794921875, 330.5745544433594, 214.33416748046875, 652.49267578125, 442.0938720703125, 374.4495849609375, -25.050018310546875, -112.44725036621094, 238.20217895507812, 43.76611328125, -67.72113037109375, 114.06672668457031, 515.1704711914062, 593.6494140625, 4.6627044677734375, 357.44671630859375, 761.7835083007812, 106.3203353881836, 680.4196166992188, -342.06597900390625, 433.08306884765625, 521.0021362304688, -292.12371826171875, 152.0195770263672, 607.3870849609375, 275.40814208984375, 49.438987731933594, 150.78802490234375, 320.1616516113281, -337.3963623046875, 151.66867065429688, 373.0148620605469, -233.25376892089844, -36.31409454345703, 247.06333923339844, 137.8575439453125, 335.95721435546875, 6.150245666503906, 86.66527557373047, 268.0322265625, 58.45330047607422, 217.87451171875, 206.44717407226562, 649.1311645507812, 277.0058898925781, 507.0917053222656, 186.93817138671875, 313.39337158203125, 304.0404357910156, 473.955078125, 244.47271728515625, 125.56163787841797, 187.43289184570312, 0.04975700378417969, 693.4625854492188, -56.6546630859375, 7.7552032470703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000500.npy"}
|
|
{"epoch": 0.7342143906020558, "step": 501, "batch_size": 64, "mean": 237.33590698242188, "std": 274.39794921875, "min": -443.07757568359375, "p10": -96.51120376586914, "median": 242.76119995117188, "p90": 563.5489807128907, "max": 913.6885375976562, "pos_frac": 0.796875, "sample": [-97.1902847290039, 318.52813720703125, 293.74554443359375, -124.84464263916016, -8.555683135986328, 700.87841796875, 96.28999328613281, 468.9232482910156, 197.19810485839844, 81.48544311523438, 202.5340576171875, -11.834287643432617, 368.69287109375, 200.53541564941406, 820.1436157226562, 570.0234985351562, 666.1341552734375, 232.96170043945312, 154.59156799316406, 476.0008544921875, 458.6307678222656, 330.9705810546875, 0.6957244873046875, 88.13443756103516, 537.7549438476562, -137.27255249023438, -94.92668151855469, 165.81500244140625, 627.1295776367188, 252.56069946289062, -443.07757568359375, 190.081298828125, 333.7630615234375, 260.7450866699219, 377.50872802734375, 913.6885375976562, -113.44627380371094, 324.57159423828125, 276.56903076171875, 375.67425537109375, 383.3218994140625, 29.50841522216797, 124.19764709472656, -61.174224853515625, 427.9539794921875, 280.7174377441406, -92.8262939453125, 101.92979431152344, 17.993892669677734, -362.8731994628906, 153.37112426757812, 391.892578125, -101.04597473144531, 10.112640380859375, 118.10223388671875, -25.95854949951172, 323.1842041015625, 451.17547607421875, 548.4417724609375, 325.2066345214844, 901.4990234375, 448.3654479980469, 300.3927001953125, 164.197021484375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000501.npy"}
|
|
{"epoch": 0.73568281938326, "step": 502, "batch_size": 64, "mean": 239.15948486328125, "std": 186.28744506835938, "min": -209.78628540039062, "p10": 27.99856262207032, "median": 230.2044219970703, "p90": 489.63659973144536, "max": 786.0161743164062, "pos_frac": 0.90625, "sample": [345.113525390625, -110.69633483886719, 162.65066528320312, 351.5787353515625, 40.8729248046875, 463.87896728515625, -64.6458511352539, 35.101806640625, 510.67449951171875, 317.3843078613281, 264.4219665527344, 258.5005798339844, 182.20191955566406, 385.4176330566406, 409.5888671875, 56.132476806640625, 101.39617919921875, 65.35359191894531, -209.78628540039062, 212.78701782226562, 447.1953430175781, 452.456298828125, 113.4045639038086, 224.48043823242188, 24.954315185546875, 214.9185333251953, 118.96842956542969, 243.3939971923828, 388.10784912109375, 275.4026184082031, 83.93440246582031, 506.56927490234375, -38.381935119628906, 272.5475769042969, 430.2336730957031, 497.2890930175781, 117.36970520019531, 502.2291564941406, -33.67411804199219, 65.72169494628906, 492.967041015625, 183.03587341308594, 168.44949340820312, 241.23681640625, 298.685791015625, 382.8870849609375, 786.0161743164062, -20.634389877319336, 324.96258544921875, 161.89044189453125, 434.5249328613281, 96.35142517089844, 235.92840576171875, 74.8336181640625, 542.0416870117188, 74.74136352539062, 216.0193328857422, 377.6929016113281, 481.8655700683594, 250.7587432861328, 177.60565185546875, 192.75131225585938, 272.053955078125, 174.4924774169922], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000502.npy"}
|
|
{"epoch": 0.737151248164464, "step": 503, "batch_size": 64, "mean": 217.50338745117188, "std": 219.0126495361328, "min": -166.3697509765625, "p10": -31.128821563720695, "median": 205.57588958740234, "p90": 457.10881958007815, "max": 963.8088989257812, "pos_frac": 0.859375, "sample": [20.778911590576172, -35.425331115722656, 599.1244506835938, 211.42127990722656, -140.6837615966797, -116.76882934570312, 425.05279541015625, 199.73049926757812, 25.90612030029297, -21.971031188964844, -166.3697509765625, 223.0087127685547, 96.72258758544922, 89.91780090332031, 714.2557373046875, 349.895751953125, 294.92169189453125, 397.3780822753906, 121.90048217773438, 299.08978271484375, 39.81044006347656, 97.06538391113281, 83.23536682128906, 135.74343872070312, 348.11761474609375, 167.07460021972656, -125.45476531982422, 457.6053466796875, 33.98027801513672, 295.54156494140625, 224.92633056640625, -35.0535888671875, 81.99180603027344, 353.5856628417969, -14.254631042480469, 113.03229522705078, 615.0709228515625, 963.8088989257812, 300.82415771484375, 368.2710266113281, 143.37884521484375, 455.95025634765625, 293.92376708984375, 147.5024871826172, -54.642425537109375, 277.46246337890625, 79.32221221923828, 408.1329650878906, 257.8888244628906, 286.1211853027344, 54.285011291503906, 256.2050476074219, 357.120849609375, 174.15650939941406, 696.1047973632812, 413.83990478515625, 27.28429412841797, 368.3901062011719, 29.28075408935547, 60.6533203125, 139.44552612304688, 250.90440368652344, 493.1552734375, 211.54190063476562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000503.npy"}
|
|
{"epoch": 0.7386196769456681, "step": 504, "batch_size": 64, "mean": 269.34735107421875, "std": 256.5953674316406, "min": -278.8019104003906, "p10": -94.84624557495117, "median": 256.33223724365234, "p90": 569.2632507324219, "max": 944.1251220703125, "pos_frac": 0.84375, "sample": [145.81723022460938, 223.39141845703125, 569.5802612304688, 162.8229522705078, 21.600204467773438, 187.676025390625, 353.25164794921875, -138.90771484375, 944.1251220703125, 741.21826171875, 779.6573486328125, -135.83697509765625, 437.3747253417969, 385.5849609375, -28.69561767578125, 33.73966979980469, 420.67279052734375, 161.18853759765625, 62.42205047607422, 96.87216186523438, 281.99908447265625, 251.5693359375, 258.05511474609375, 64.28410339355469, 264.01910400390625, 128.6956329345703, -93.8793716430664, 440.09637451171875, 67.56698608398438, 218.85487365722656, -278.8019104003906, 557.712158203125, 255.37135314941406, 645.91015625, 555.4669799804688, 481.7349853515625, 568.5235595703125, 471.7935791015625, 98.22634887695312, 175.21163940429688, 343.2164306640625, 720.8199462890625, 460.95947265625, 275.7961730957031, 344.8272399902344, -111.76516723632812, -161.448486328125, 315.56011962890625, 140.3465118408203, 555.0914306640625, 193.45809936523438, 416.4393310546875, 430.0774230957031, 441.804931640625, 240.69395446777344, 136.3572998046875, -97.01799011230469, 232.620849609375, 657.128173828125, 257.2931213378906, -95.2606201171875, 352.89422607421875, 442.1761779785156, -89.80438232421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000504.npy"}
|
|
{"epoch": 0.7400881057268722, "step": 505, "batch_size": 64, "mean": 152.27745056152344, "std": 269.1082763671875, "min": -442.06671142578125, "p10": -151.82197265624998, "median": 115.43742370605469, "p90": 478.57625427246103, "max": 935.2919311523438, "pos_frac": 0.6875, "sample": [1.3218002319335938, 444.46856689453125, -117.89236450195312, 399.1596374511719, -33.61158752441406, 486.7019348144531, -43.37150573730469, 4.6816558837890625, 108.47979736328125, 43.77244567871094, 565.9031372070312, 453.6350402832031, 935.2919311523438, -4.863250732421875, 315.80377197265625, 357.42218017578125, 24.749282836914062, 5.20452880859375, 248.3856201171875, -141.2856903076172, 117.54164123535156, 424.635009765625, 168.50872802734375, 68.97561645507812, -207.3739013671875, -236.62570190429688, 96.96046447753906, 230.41302490234375, 452.23431396484375, 284.70587158203125, -100.73194885253906, 290.4161376953125, 342.6883544921875, -147.42642211914062, 36.86602020263672, 550.8017578125, 240.91122436523438, 152.83213806152344, -18.153770446777344, 179.93540954589844, -192.1148681640625, -442.06671142578125, 191.02378845214844, 113.33320617675781, 18.8670654296875, 302.2745666503906, 715.4719848632812, 296.0511169433594, -35.14664840698242, 645.32861328125, -181.1695098876953, -130.45480346679688, -153.70578002929688, 637.4067993164062, 164.4498748779297, -87.67765808105469, 214.884521484375, -379.1468505859375, 76.85723876953125, 233.6874542236328, 459.6163330078125, -1.4760894775390625, 300.4228820800781, -3.030773162841797], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000505.npy"}
|
|
{"epoch": 0.7415565345080763, "step": 506, "batch_size": 64, "mean": 233.715087890625, "std": 244.8497314453125, "min": -295.2010498046875, "p10": -65.17940063476563, "median": 242.9813995361328, "p90": 564.1444458007812, "max": 926.1643676757812, "pos_frac": 0.796875, "sample": [-295.2010498046875, 37.203094482421875, 355.6211242675781, -30.876846313476562, 127.298583984375, 261.4697265625, -118.65617370605469, 302.73779296875, 456.096435546875, 68.25430297851562, 402.2026672363281, 137.0942840576172, 110.84980773925781, 308.13922119140625, -178.12344360351562, 699.8699951171875, 340.0626525878906, 48.86738586425781, 186.06686401367188, 334.52838134765625, 177.67999267578125, 609.3283081054688, 364.7242126464844, 124.58220672607422, 926.1643676757812, -127.11309814453125, -38.61071014404297, 309.1407470703125, 368.9124755859375, 317.51458740234375, 202.04664611816406, 497.41839599609375, 403.2972412109375, 99.22119140625, 445.7303466796875, -207.40792846679688, 380.3828430175781, 207.4793701171875, 273.6145324707031, 605.58349609375, 457.2783203125, 348.9683837890625, -53.67784118652344, 297.102294921875, 201.68324279785156, -66.8821029663086, 224.49307250976562, 85.63066864013672, 619.0201416015625, 446.9718322753906, 386.544189453125, -2.1430740356445312, 120.096923828125, 559.0180053710938, 566.3414916992188, 636.3795166015625, -54.62483215332031, 261.4765625, 121.05805969238281, 24.690650939941406, -144.30484008789062, 184.82040405273438, 305.83770751953125, -61.20642852783203], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000506.npy"}
|
|
{"epoch": 0.7430249632892805, "step": 507, "batch_size": 64, "mean": 200.48583984375, "std": 257.65283203125, "min": -294.4805908203125, "p10": -67.53376693725585, "median": 189.42889404296875, "p90": 491.83185424804697, "max": 1157.3289794921875, "pos_frac": 0.765625, "sample": [51.111839294433594, 204.3682098388672, -33.57341003417969, -294.4805908203125, 684.9769287109375, 155.8206787109375, 110.25682067871094, 279.8119201660156, -34.25334930419922, 141.153076171875, 29.873184204101562, 293.7325134277344, -90.70817565917969, 300.51220703125, 407.8194885253906, 159.98406982421875, -222.4818572998047, 215.91290283203125, 174.4895782470703, 1157.3289794921875, 205.12815856933594, 271.8282470703125, -27.775012969970703, 281.6288146972656, 391.5391845703125, 270.96551513671875, 252.94793701171875, 324.85699462890625, 23.702510833740234, 476.3453369140625, 294.2336120605469, 19.662506103515625, -106.7204818725586, 50.04591369628906, -9.318719863891602, 169.58346557617188, 224.01394653320312, 385.5017395019531, 226.88775634765625, 53.00133514404297, 223.85740661621094, 18.916915893554688, 46.926177978515625, -72.92337799072266, 979.9967041015625, 560.2816772460938, 87.88842010498047, 227.6487274169922, -61.558746337890625, 433.12921142578125, 385.5205078125, -70.09449005126953, -41.53837585449219, 498.46893310546875, 558.7698364257812, 419.1023254394531, 559.3275756835938, 407.8986511230469, 261.63336181640625, -2.1374149322509766, 23.088882446289062, -83.77279663085938, -61.29063415527344, 62.24132537841797], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000507.npy"}
|
|
{"epoch": 0.7444933920704846, "step": 508, "batch_size": 64, "mean": 187.93719482421875, "std": 270.62628173828125, "min": -484.7939758300781, "p10": -60.18796119689941, "median": 130.00976181030273, "p90": 488.7281799316406, "max": 1163.9990234375, "pos_frac": 0.765625, "sample": [133.89483642578125, 92.69732666015625, -291.98980712890625, 332.75103759765625, 296.5679931640625, 11.470367431640625, -49.209903717041016, 1163.9990234375, 376.2018127441406, 489.24957275390625, 273.8836364746094, 546.3831787109375, 189.60800170898438, 833.2470703125, 423.67510986328125, 256.5, 268.75823974609375, -215.40829467773438, 609.267578125, 338.91937255859375, -62.16079330444336, 145.8848419189453, 118.73210144042969, 11.903167724609375, 645.7972412109375, 112.09085083007812, 169.59912109375, -33.928916931152344, -0.3582592010498047, 126.12468719482422, 30.425460815429688, 235.68704223632812, 39.15126037597656, -24.196517944335938, 791.61474609375, -484.7939758300781, 21.925254821777344, 125.9588851928711, 94.92037963867188, 261.6319885253906, 53.420196533203125, 275.852783203125, 70.1326904296875, 121.15606689453125, 329.01947021484375, 213.63858032226562, 455.54058837890625, 152.7070770263672, 458.3884582519531, 410.0310974121094, 274.4609680175781, -10.148651123046875, -95.85310363769531, 487.5115966796875, 399.7798767089844, -17.30261993408203, -55.584686279296875, 7.67291259765625, -109.89198303222656, 162.94256591796875, -45.99220275878906, 97.73259735107422, -103.82866668701172, 90.11983489990234], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000508.npy"}
|
|
{"epoch": 0.7459618208516887, "step": 509, "batch_size": 64, "mean": 190.60777282714844, "std": 233.42396545410156, "min": -218.85911560058594, "p10": -71.75232620239254, "median": 154.43143463134766, "p90": 507.15971984863296, "max": 1044.103759765625, "pos_frac": 0.828125, "sample": [87.6372299194336, -37.46531677246094, -207.38851928710938, 533.1575927734375, 119.8447265625, 265.756591796875, 59.89289093017578, 12.705085754394531, 317.82806396484375, 258.26739501953125, 407.63189697265625, 229.345947265625, 209.6358642578125, 83.8212890625, 84.83714294433594, 53.15376281738281, 722.5601806640625, -218.85911560058594, -195.05120849609375, 269.5145568847656, 471.1336669921875, 269.74249267578125, 129.2812042236328, 109.25261688232422, 291.8773193359375, 158.91534423828125, 170.65487670898438, 168.97821044921875, 301.45068359375, 331.8457336425781, 125.54335021972656, -168.3726043701172, -19.499645233154297, -100.66731262207031, -38.871429443359375, 666.1639404296875, 215.24598693847656, 149.94752502441406, 235.77157592773438, -26.478897094726562, 62.574371337890625, 243.08351135253906, 75.44155883789062, 130.09878540039062, 37.68199157714844, -125.76789855957031, 585.8723754882812, 474.1698913574219, 95.59471130371094, 1044.103759765625, -85.8441390991211, 161.24295043945312, 66.01921081542969, 58.247222900390625, 282.9793701171875, 111.36039733886719, 103.36531829833984, 115.70846557617188, 242.39080810546875, 341.839111328125, 521.2982177734375, 656.3810424804688, 182.798828125, 319.5167236328125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000509.npy"}
|
|
{"epoch": 0.7474302496328928, "step": 510, "batch_size": 64, "mean": 228.685302734375, "std": 210.82870483398438, "min": -276.4483947753906, "p10": -47.959822463989255, "median": 241.58431243896484, "p90": 520.9692932128906, "max": 626.5381469726562, "pos_frac": 0.796875, "sample": [-53.623043060302734, 293.95245361328125, 626.1323852539062, 330.1395263671875, 113.84506225585938, 126.22349548339844, 483.691162109375, 465.18939208984375, -48.14533996582031, 353.976318359375, 344.50341796875, 276.1944885253906, -15.854997634887695, 215.537109375, 514.2821655273438, 393.6054992675781, 366.5679016113281, -276.4483947753906, -60.81859588623047, 452.1257629394531, 207.15951538085938, 243.8606719970703, 260.751953125, 69.39547729492188, 202.12008666992188, 307.55682373046875, 371.0924072265625, 626.5381469726562, 422.0281066894531, -184.330322265625, 523.835205078125, 375.8095397949219, -0.4322242736816406, 111.78905487060547, 377.82574462890625, 269.1684265136719, 239.30795288085938, 95.10307312011719, -48.02898025512695, 186.35797119140625, 108.68305969238281, 233.46487426757812, 527.6177978515625, 260.235107421875, -54.72053527832031, -38.807655334472656, 68.98527526855469, 224.3817138671875, 189.8767547607422, 329.39093017578125, 587.502685546875, -18.77378273010254, 303.85791015625, 533.6600952148438, -17.471267700195312, 393.6383361816406, 264.3828125, 440.1773376464844, -47.79845428466797, 32.86644744873047, 57.28400421142578, 584.5708618164062, 87.31314086914062, 27.55648422241211], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000510.npy"}
|
|
{"epoch": 0.748898678414097, "step": 511, "batch_size": 64, "mean": 228.15643310546875, "std": 279.3397216796875, "min": -548.4164428710938, "p10": -56.56517715454099, "median": 202.11420440673828, "p90": 565.7629455566407, "max": 1023.7711181640625, "pos_frac": 0.8125, "sample": [78.21817779541016, 729.2109985351562, 98.04287719726562, 185.34454345703125, 210.41363525390625, 148.57627868652344, 529.74169921875, 638.2638549804688, 141.6302032470703, 174.31683349609375, 266.0529479980469, 430.861572265625, 543.5891723632812, 627.34814453125, -12.79005241394043, 780.4837036132812, -548.4164428710938, 459.7685546875, 122.057861328125, 452.11517333984375, 412.3330383300781, 402.8330383300781, 62.56874084472656, 381.02685546875, 142.11181640625, 56.513099670410156, -376.8782958984375, 291.5784912109375, -32.400421142578125, 478.84423828125, 470.9951171875, 85.86598205566406, -30.837862014770508, 193.8147735595703, 283.0463562011719, 280.5666198730469, 431.3210144042969, 162.82815551757812, -26.81756591796875, 144.07708740234375, 452.8467712402344, 235.62655639648438, 216.52891540527344, -197.96585083007812, -66.92150115966797, 575.2659912109375, 267.6857604980469, -8.618757247924805, -85.2408218383789, 69.40480041503906, -161.79592895507812, 176.39334106445312, 287.15521240234375, 0.16600608825683594, 106.43051147460938, 269.6744384765625, 221.92874145507812, 2.8763198852539062, 1023.7711181640625, 86.8475341796875, 248.03109741210938, -123.0733642578125, 242.45596313476562, 894.319091796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000511.npy"}
|
|
{"epoch": 0.750367107195301, "step": 512, "batch_size": 64, "mean": 245.87234497070312, "std": 219.16929626464844, "min": -136.19015502929688, "p10": -11.083580589294424, "median": 209.0545425415039, "p90": 591.5253601074219, "max": 748.429443359375, "pos_frac": 0.875, "sample": [637.8095703125, 609.0892333984375, 457.548095703125, -136.19015502929688, -2.1942596435546875, -69.74087524414062, 620.5704956054688, 155.5946044921875, 3.870847702026367, 82.33467864990234, 86.37165832519531, 374.5525207519531, 3.0372238159179688, 241.38145446777344, -83.9080810546875, 96.83882141113281, 405.0186767578125, 286.5914611816406, 180.7288818359375, -14.893289566040039, 17.04602813720703, 422.7848205566406, 38.36639404296875, 338.64361572265625, 748.429443359375, 309.48028564453125, 159.8824462890625, 241.68353271484375, 63.860801696777344, 419.7845153808594, 196.99288940429688, 295.6787109375, 261.520263671875, 321.9141845703125, 208.42982482910156, 155.95913696289062, 166.2134552001953, 209.67926025390625, -35.261573791503906, 515.267578125, 63.87188720703125, 484.2073974609375, 348.10980224609375, 594.2542724609375, 16.93756103515625, -108.30677795410156, 380.68182373046875, 69.55850219726562, 65.71556091308594, 299.5065002441406, 695.4899291992188, 42.82579803466797, 205.3258056640625, 435.06072998046875, 114.76839447021484, 186.16400146484375, 347.778076171875, 162.85433959960938, 303.66314697265625, 338.63287353515625, -25.578815460205078, 585.1578979492188, 660.0501098632812, 478.33392333984375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000512.npy"}
|
|
{"epoch": 0.7518355359765051, "step": 513, "batch_size": 64, "mean": 172.156982421875, "std": 240.09603881835938, "min": -347.0347900390625, "p10": -85.78386840820312, "median": 179.85171508789062, "p90": 496.88312683105477, "max": 756.8375244140625, "pos_frac": 0.75, "sample": [483.2757873535156, 177.086181640625, 184.8495635986328, -24.896102905273438, 192.26519775390625, 264.3527526855469, 9.468130111694336, 182.61724853515625, -245.27342224121094, 553.273681640625, 342.7054443359375, 26.853851318359375, 419.344482421875, 102.23934173583984, -46.238868713378906, 431.09356689453125, 61.922943115234375, -87.5848388671875, 160.3920135498047, 231.01907348632812, 243.5919189453125, 427.32781982421875, 604.5557861328125, 276.4959716796875, 194.40530395507812, 113.76113891601562, 246.35630798339844, -81.58160400390625, 294.36181640625, -16.73080825805664, 95.95382690429688, 170.3340606689453, 355.37835693359375, 382.5926818847656, 163.27301025390625, -13.422393798828125, 61.99430847167969, -304.78192138671875, 270.07769775390625, 756.8375244140625, -72.80744934082031, 3.5705184936523438, 242.23297119140625, -314.33001708984375, 568.3872680664062, 12.3482666015625, -111.96404266357422, -19.06351661682129, 502.71484375, 509.6410827636719, -308.406005859375, -9.279548645019531, 596.30126953125, 408.3349914550781, 32.88117218017578, 229.2528076171875, -28.527780532836914, 123.4028549194336, 233.78128051757812, 241.8268585205078, -347.0347900390625, 440.0494384765625, 136.42166137695312, 288.76513671875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000513.npy"}
|
|
{"epoch": 0.7533039647577092, "step": 514, "batch_size": 64, "mean": 186.24154663085938, "std": 250.3760223388672, "min": -276.0170593261719, "p10": -111.79317169189453, "median": 157.8502197265625, "p90": 516.0884826660157, "max": 963.291015625, "pos_frac": 0.78125, "sample": [-276.0170593261719, -111.03156280517578, 171.98866271972656, -176.28408813476562, 512.8140258789062, 151.84463500976562, 385.3889465332031, 348.1112060546875, 963.291015625, 571.1615600585938, 68.89799499511719, 284.69110107421875, -149.21560668945312, 225.124267578125, 418.22760009765625, -46.05830383300781, -92.88224792480469, 84.80549621582031, 567.7770385742188, 662.9774780273438, -143.3486785888672, 220.5494384765625, 275.6668395996094, -88.37712860107422, 24.32042694091797, 163.85580444335938, 84.38258361816406, 92.5640869140625, 236.44989013671875, 44.65129852294922, 360.8128662109375, 270.59283447265625, -46.770545959472656, 106.538330078125, 65.24799346923828, 88.11143493652344, 599.2808837890625, -131.43792724609375, 517.4918212890625, -107.81590270996094, 151.2962646484375, -224.4957733154297, -112.11957550048828, 293.542724609375, 181.51312255859375, 60.13621139526367, -76.05142211914062, 3.0941619873046875, 94.6452865600586, 393.05499267578125, 225.42764282226562, 381.03497314453125, 37.57225036621094, 489.3030090332031, 447.67034912109375, 78.54026794433594, 268.76788330078125, 172.45907592773438, 512.2022094726562, 208.8091278076172, 394.95703125, 39.91835403442383, 42.557037353515625, 657.2447509765625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000514.npy"}
|
|
{"epoch": 0.7547723935389133, "step": 515, "batch_size": 64, "mean": 176.40408325195312, "std": 221.39749145507812, "min": -524.193115234375, "p10": -70.41390228271483, "median": 164.91222381591797, "p90": 481.38923645019537, "max": 589.5636596679688, "pos_frac": 0.84375, "sample": [56.70784378051758, 123.20957946777344, 302.0945129394531, 564.8616943359375, 174.11383056640625, -524.193115234375, 382.0685729980469, 249.39671325683594, 91.56402587890625, -215.8898468017578, -79.43647766113281, 106.93580627441406, 16.11328125, 535.2216796875, 317.24920654296875, 270.021240234375, 49.36613464355469, 472.4161071777344, 9.79307746887207, -12.21014404296875, 224.83392333984375, -56.86192321777344, 155.48143005371094, 340.0008544921875, 156.92359924316406, 76.93165588378906, 485.23486328125, -80.39630126953125, 191.68435668945312, 149.5694122314453, 557.7042236328125, 22.589365005493164, -3.0066165924072266, 223.53701782226562, 363.6311950683594, 435.657958984375, 322.18756103515625, 10.616432189941406, 540.7138671875, -76.22189331054688, 139.47994995117188, 589.5636596679688, -327.9447021484375, 33.926551818847656, -183.67181396484375, 205.64593505859375, 69.71463775634766, 569.9627685546875, 60.4166145324707, 68.46725463867188, 190.61895751953125, 468.3385314941406, 34.444129943847656, 129.68084716796875, 356.7405700683594, 407.75006103515625, 177.66104125976562, 42.837982177734375, 253.67335510253906, 347.072021484375, 172.90084838867188, 112.96310424804688, 183.78924560546875, 255.6151123046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000515.npy"}
|
|
{"epoch": 0.7562408223201175, "step": 516, "batch_size": 64, "mean": 235.27685546875, "std": 210.7569580078125, "min": -199.10597229003906, "p10": -65.41987533569335, "median": 237.50296020507812, "p90": 507.70304260253914, "max": 758.3977661132812, "pos_frac": 0.84375, "sample": [106.63895416259766, 162.48825073242188, 307.2921447753906, -50.95093536376953, 365.0552062988281, 289.98358154296875, 378.972412109375, -88.30619812011719, 223.2059783935547, 430.85589599609375, 533.1293334960938, 205.10987854003906, 238.28900146484375, 361.201416015625, 203.12046813964844, -79.87551879882812, 365.0603332519531, 208.6171112060547, 69.7265625, 657.2149658203125, -199.10597229003906, 100.21774291992188, -71.620849609375, 106.40602111816406, 253.77334594726562, 238.1753692626953, 58.074729919433594, 580.1293334960938, 756.355224609375, 190.5027313232422, 96.34640502929688, 578.3539428710938, 758.3977661132812, 244.66949462890625, 489.1774597167969, 270.8413391113281, 291.50677490234375, 35.97806930541992, 153.8249053955078, 169.26568603515625, 181.47702026367188, -90.491455078125, 363.63140869140625, 163.02755737304688, -36.22292709350586, 473.47113037109375, 229.00782775878906, 256.16632080078125, 54.39293670654297, 236.83055114746094, 335.82305908203125, 375.6385192871094, 58.844512939453125, 261.60821533203125, 397.31109619140625, -88.26302337646484, -89.49647521972656, 36.12628173828125, 319.82806396484375, 374.8760681152344, 371.4444580078125, 515.642578125, 394.368408203125, -25.422107696533203], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000516.npy"}
|
|
{"epoch": 0.7577092511013216, "step": 517, "batch_size": 64, "mean": 235.31272888183594, "std": 251.65916442871094, "min": -231.67755126953125, "p10": -65.49091110229492, "median": 209.00360107421875, "p90": 545.9819396972657, "max": 1061.1676025390625, "pos_frac": 0.8125, "sample": [207.26486206054688, 387.62371826171875, 240.52117919921875, 82.92257690429688, 611.2794189453125, 104.91485595703125, 163.4607391357422, 325.66009521484375, 373.39013671875, -149.42559814453125, 340.4328308105469, 799.9751586914062, 224.88507080078125, 217.89739990234375, 268.7218017578125, -36.694419860839844, 191.80198669433594, 256.94891357421875, 310.95489501953125, 556.61474609375, 521.1720581054688, 439.1898193359375, 77.56432342529297, -6.186859130859375, -5.432228088378906, -95.31451416015625, 100.53447723388672, 210.74234008789062, -67.10263061523438, 1061.1676025390625, 168.97018432617188, 69.76200866699219, -32.78614807128906, 508.9205322265625, 18.211669921875, 382.15350341796875, -182.38262939453125, 372.32049560546875, -76.76216888427734, 206.644287109375, 418.98974609375, 89.90747833251953, 181.13719177246094, 331.0654602050781, 357.0492248535156, 307.065185546875, 360.7496032714844, 292.2276916503906, 5.274055480957031, 844.345458984375, 121.97042083740234, 396.55145263671875, 292.1343688964844, 643.691650390625, -231.67755126953125, 646.4361572265625, 106.87461853027344, 194.34713745117188, -61.73023223876953, 185.1474151611328, -126.33438110351562, 390.55010986328125, 149.66268920898438, 14.043113708496094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000517.npy"}
|
|
{"epoch": 0.7591776798825257, "step": 518, "batch_size": 64, "mean": 206.73744201660156, "std": 258.8006286621094, "min": -468.00604248046875, "p10": -106.01330108642573, "median": 193.2380599975586, "p90": 593.970233154297, "max": 758.3052978515625, "pos_frac": 0.796875, "sample": [24.67959213256836, 676.8961181640625, 298.057373046875, 176.45787048339844, 210.4309539794922, 210.6613006591797, 200.33953857421875, 382.603515625, 39.12471389770508, 88.52230834960938, 274.0655517578125, 492.333251953125, 122.354248046875, 37.49311828613281, 324.2846984863281, 127.02205657958984, 178.77313232421875, 145.36676025390625, -5.01031494140625, 453.75347900390625, 517.439453125, -17.72571563720703, -154.77890014648438, 192.01182556152344, -147.28146362304688, 268.2526550292969, 614.4566040039062, 159.2305908203125, 252.07179260253906, 676.5133666992188, 702.7409057617188, -1.2258453369140625, 636.312255859375, 546.168701171875, 117.51184844970703, 349.5044860839844, 194.46429443359375, -45.309356689453125, 304.1265563964844, 423.5123291015625, 111.73701477050781, -55.79371643066406, 318.51629638671875, 83.16631317138672, 246.5895538330078, -47.81830596923828, -210.15513610839844, -468.00604248046875, 335.5594787597656, 316.3735656738281, 0.6893157958984375, 84.99530029296875, -244.05653381347656, 180.54241943359375, -295.72723388671875, 758.3052978515625, 378.9151306152344, 67.38906860351562, -127.53598022460938, 748.9768676757812, 253.22593688964844, 87.74345397949219, 409.3859558105469, 251.9726104736328], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000518.npy"}
|
|
{"epoch": 0.7606461086637298, "step": 519, "batch_size": 64, "mean": 179.88906860351562, "std": 236.11048889160156, "min": -369.8384704589844, "p10": -83.66320037841795, "median": 149.43231201171875, "p90": 498.46814880371096, "max": 901.9027099609375, "pos_frac": 0.8125, "sample": [26.31170654296875, 219.82911682128906, 177.0692596435547, -152.95989990234375, 627.9298706054688, -123.15518951416016, 179.49656677246094, 488.886962890625, 588.2298583984375, 313.0023193359375, 126.65660095214844, 502.5743713378906, -98.50518798828125, 565.5336303710938, 225.74493408203125, -259.4275817871094, 688.2257690429688, 378.2539367675781, 85.97069549560547, 125.25679016113281, 235.61451721191406, 18.373550415039062, 53.4005012512207, 150.85443115234375, 151.62664794921875, 54.811981201171875, 172.33914184570312, 118.08677673339844, 185.08351135253906, 67.94417572021484, -3.80206298828125, 461.06298828125, 125.50906372070312, -87.27987670898438, 282.5592956542969, 165.40670776367188, 275.60888671875, -8.096351623535156, -71.44872283935547, 74.76173400878906, 148.01019287109375, 48.698211669921875, 393.1365051269531, 123.3216552734375, -369.8384704589844, 448.4869384765625, 901.9027099609375, 36.614051818847656, 113.89105224609375, 100.72499084472656, 190.93643188476562, 632.5445556640625, -75.22428894042969, 418.9542236328125, -47.177001953125, 128.38229370117188, 420.9944152832031, 12.92813491821289, 175.28067016601562, 262.8700256347656, 245.7576446533203, 69.73954772949219, 218.82809448242188, -194.20425415039062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000519.npy"}
|
|
{"epoch": 0.762114537444934, "step": 520, "batch_size": 64, "mean": 198.79312133789062, "std": 227.2389373779297, "min": -313.5616149902344, "p10": -24.7774314880371, "median": 146.7658920288086, "p90": 500.8487365722657, "max": 826.8521118164062, "pos_frac": 0.859375, "sample": [89.4994125366211, -10.26849365234375, 756.5743408203125, 89.28984069824219, 243.40870666503906, 72.54803466796875, 383.2441711425781, -14.079322814941406, 3.5605087280273438, 151.638427734375, 220.57814025878906, -29.362335205078125, 441.9176940917969, 37.48735046386719, 37.71538543701172, 133.86349487304688, 536.506591796875, 826.8521118164062, -129.52017211914062, 132.16726684570312, 505.03472900390625, 824.1461791992188, -313.5616149902344, 180.080322265625, -59.577850341796875, 427.533203125, 108.9739761352539, 525.0179443359375, 164.14486694335938, 287.21148681640625, 26.55135726928711, 192.9740447998047, 559.9923706054688, 93.06761169433594, 97.6629638671875, 294.15771484375, 491.0814208984375, 358.3373718261719, 335.7564697265625, 107.58935546875, 321.61944580078125, 108.78214263916016, 141.8933563232422, 184.94879150390625, 38.58135986328125, 481.5592956542969, -87.6166763305664, 189.92376708984375, 446.55322265625, 57.232635498046875, 155.787353515625, 12.142669677734375, -62.551483154296875, 188.07337951660156, 60.930755615234375, 453.0437927246094, 50.633056640625, 270.748046875, -75.24723815917969, 320.17510986328125, 154.17724609375, 1.2803497314453125, 28.1343994140625, 102.15997314453125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000520.npy"}
|
|
{"epoch": 0.7635829662261381, "step": 521, "batch_size": 64, "mean": 195.25540161132812, "std": 226.31771850585938, "min": -192.18917846679688, "p10": -56.385371017456045, "median": 171.3476333618164, "p90": 483.70603027343753, "max": 937.8847045898438, "pos_frac": 0.71875, "sample": [406.66522216796875, 542.1688232421875, 214.00765991210938, 23.662147521972656, -63.24781036376953, -164.85794067382812, -26.233245849609375, -4.42132568359375, -35.820037841796875, 341.4861755371094, 53.28875732421875, -60.97808074951172, -19.866573333740234, 35.875282287597656, 407.6217956542969, 509.892822265625, 58.56182861328125, -13.16115951538086, -133.9781036376953, 580.53125, -1.3333473205566406, -44.11064147949219, -44.135292053222656, 204.5076141357422, 406.66632080078125, 474.6769104003906, 355.2521667480469, 205.25265502929688, 81.4711685180664, 67.0238037109375, 349.37030029296875, 274.7695007324219, -8.770713806152344, 439.418212890625, -9.911354064941406, 136.63113403320312, 241.22476196289062, 144.21029663085938, 468.7512512207031, -81.0875473022461, 274.05950927734375, 399.68304443359375, 161.13796997070312, -45.66904830932617, 141.67791748046875, 332.60064697265625, -62.9629020690918, 350.8251953125, 41.50579833984375, -192.18917846679688, 337.4001159667969, 222.50181579589844, 536.6828002929688, 87.58262634277344, 121.43597412109375, 381.3656921386719, 296.7830810546875, 487.5756530761719, 17.331283569335938, 937.8847045898438, 303.54669189453125, 181.5572967529297, 280.6531677246094, 592.3017578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000521.npy"}
|
|
{"epoch": 0.7650513950073421, "step": 522, "batch_size": 64, "mean": 172.7684326171875, "std": 213.8473663330078, "min": -475.7686767578125, "p10": -82.7793918609619, "median": 175.1966323852539, "p90": 474.29831848144534, "max": 612.3250732421875, "pos_frac": 0.8125, "sample": [277.123779296875, 237.16966247558594, 396.0538330078125, -28.12322235107422, 527.0946044921875, 34.341060638427734, 475.7294616699219, 143.95828247070312, 71.68212890625, -31.015064239501953, 2.6004371643066406, 113.14000701904297, 427.81658935546875, 26.678482055664062, 353.2584228515625, 184.90997314453125, 298.4359130859375, -58.06230545043945, 264.95611572265625, 189.04495239257812, 281.51123046875, 462.7867431640625, 187.74325561523438, 198.35865783691406, -198.59832763671875, 612.3250732421875, 122.7743911743164, 139.3073272705078, 416.88629150390625, 169.98635864257812, 88.1903076171875, 179.3162078857422, 127.84257507324219, 27.981781005859375, 285.83624267578125, -475.7686767578125, 187.1460418701172, -4.793004989624023, 213.15725708007812, -150.55523681640625, -129.9374237060547, 237.82131958007812, 479.72430419921875, 22.72917938232422, 44.50618362426758, 325.0799560546875, 392.0207214355469, 171.07705688476562, -283.5255432128906, 523.568603515625, 295.2687072753906, 39.29693603515625, 358.5708923339844, 124.4660415649414, 91.86644744873047, -120.58422088623047, -93.37242889404297, -13.686820983886719, 490.21612548828125, 236.36317443847656, 470.958984375, 33.399723052978516, 477.98583984375, 105.13859558105469], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000522.npy"}
|
|
{"epoch": 0.7665198237885462, "step": 523, "batch_size": 64, "mean": 189.81651306152344, "std": 230.30813598632812, "min": -556.6541748046875, "p10": -64.97952194213866, "median": 202.4580841064453, "p90": 498.4781890869141, "max": 677.8579711914062, "pos_frac": 0.75, "sample": [-44.00849914550781, 503.59844970703125, 140.79432678222656, 280.12225341796875, 340.08941650390625, -115.98420715332031, 299.17181396484375, 204.44192504882812, 415.3487548828125, 218.751220703125, -244.6532745361328, 265.8617858886719, 17.25339698791504, 252.52743530273438, -32.8079719543457, 74.64350891113281, 200.25906372070312, 552.347412109375, 170.94911193847656, 200.4742431640625, 171.27430725097656, 582.6654052734375, 148.5569610595703, -83.92033386230469, 238.14683532714844, 120.69346618652344, 146.9065399169922, 102.92001342773438, -23.66698455810547, 491.91656494140625, 233.82028198242188, 11.507280349731445, -26.365873336791992, -556.6541748046875, 276.7692565917969, 423.9181823730469, -163.07235717773438, -26.711042404174805, 298.4477844238281, -69.82613372802734, 367.0928649902344, 23.71103858947754, 582.713623046875, 264.33343505859375, 332.76483154296875, 61.648223876953125, 431.7571105957031, 247.3858642578125, 548.2564086914062, 122.39205169677734, 436.9917297363281, 144.33583068847656, 343.9714660644531, -12.400306701660156, 385.7218017578125, 371.09967041015625, 228.8675994873047, -53.67076110839844, 362.5060119628906, 677.8579711914062, 501.2903137207031, -23.257404327392578, -157.27880859375, -36.339412689208984], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000523.npy"}
|
|
{"epoch": 0.7679882525697503, "step": 524, "batch_size": 64, "mean": 213.60824584960938, "std": 217.58135986328125, "min": -422.7045593261719, "p10": -35.51767272949219, "median": 233.02706909179688, "p90": 451.14876098632817, "max": 674.0765380859375, "pos_frac": 0.828125, "sample": [82.98957824707031, 611.9444580078125, 252.00839233398438, 210.05548095703125, 171.3857421875, 188.61354064941406, 147.69869995117188, 26.48918914794922, 368.96917724609375, 240.5377197265625, 375.4111328125, 57.464874267578125, 277.2914733886719, 453.017333984375, 397.2786560058594, 221.78073120117188, 7.312774658203125, 417.251708984375, 175.17405700683594, 333.4034423828125, -422.7045593261719, 225.51641845703125, 212.09976196289062, 658.2872314453125, 20.081375122070312, -72.75613403320312, 297.52685546875, 208.61798095703125, 446.78875732421875, -30.59280776977539, 109.62823486328125, 89.42536926269531, 273.6712646484375, 57.53749084472656, 519.4947509765625, -31.724607467651367, 473.0596923828125, 674.0765380859375, 361.5842590332031, 273.3070068359375, 101.5297622680664, -187.767578125, 14.949172973632812, -37.56658172607422, 433.31988525390625, -35.7418212890625, 358.71087646484375, 12.275642395019531, 340.3533935546875, 288.2835388183594, 542.328125, 434.52581787109375, 286.67572021484375, -307.3126525878906, -6.983482360839844, 191.26551818847656, 441.0888977050781, 327.9609375, -40.170310974121094, 260.73260498046875, 391.0138244628906, 247.02989196777344, 290.4173583984375, -34.994659423828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000524.npy"}
|
|
{"epoch": 0.7694566813509545, "step": 525, "batch_size": 64, "mean": 251.78009033203125, "std": 225.4392547607422, "min": -155.45721435546875, "p10": -8.960177803039548, "median": 211.0677490234375, "p90": 609.9954040527344, "max": 808.2232666015625, "pos_frac": 0.859375, "sample": [191.0748748779297, 637.7606811523438, 202.33807373046875, 168.96841430664062, 65.36446380615234, 411.15869140625, 522.6650390625, 208.90536499023438, 518.7340087890625, 225.98904418945312, 194.54647827148438, -155.45721435546875, 382.46533203125, 105.83723449707031, 166.37887573242188, -84.81480407714844, 129.25462341308594, 157.19122314453125, 605.9510498046875, 362.78424072265625, 383.85894775390625, 177.23965454101562, -3.20123291015625, 273.9222106933594, 59.618896484375, 323.55743408203125, 280.260986328125, 338.07928466796875, 278.3962707519531, 636.4357299804688, 361.3774108886719, 42.379234313964844, 620.437255859375, 80.5736312866211, 552.73876953125, 188.68896484375, 21.78852081298828, 239.99346923828125, 336.0618591308594, -10.074066162109375, 808.2232666015625, -123.98583984375, -64.43365478515625, -6.361104965209961, 33.871978759765625, 160.54275512695312, 320.8435363769531, 49.93276596069336, 431.74774169921875, 193.43325805664062, 493.543701171875, 676.8853149414062, -70.83148193359375, 119.68838500976562, 472.36956787109375, 228.8441925048828, 213.23013305664062, 58.45360565185547, -146.07098388671875, 375.9806213378906, 266.3401794433594, 611.7286987304688, 173.22923278808594, 637.4912719726562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000525.npy"}
|
|
{"epoch": 0.7709251101321586, "step": 526, "batch_size": 64, "mean": 242.66949462890625, "std": 239.4748077392578, "min": -242.36996459960938, "p10": -53.91685714721679, "median": 240.10370635986328, "p90": 527.9705017089844, "max": 884.135986328125, "pos_frac": 0.84375, "sample": [763.5477294921875, 515.4324951171875, 74.577880859375, 146.35543823242188, 178.99957275390625, 178.995849609375, -0.16089820861816406, -97.85492706298828, 300.59857177734375, -242.36996459960938, 363.51202392578125, 333.7679443359375, 174.31289672851562, 181.94534301757812, 602.6239624023438, 779.22900390625, 11.971250534057617, 415.90228271484375, 211.78076171875, 436.82879638671875, 171.2960662841797, 256.0085754394531, -45.10752868652344, 533.3439331054688, 321.5733642578125, 321.7772521972656, 884.135986328125, 191.59071350097656, 274.8941955566406, 22.865671157836914, -158.85545349121094, 419.0306091308594, -57.692283630371094, 96.65599822998047, -104.69168090820312, 26.343292236328125, 469.6401672363281, 313.99945068359375, 81.13461303710938, -215.48770141601562, 396.13812255859375, 314.7517395019531, 323.63494873046875, 397.52734375, -82.3961181640625, 291.8337097167969, 509.0560302734375, 420.2381591796875, 57.92340850830078, 326.6910705566406, 20.22674560546875, 133.72779846191406, 118.74673461914062, 611.2667236328125, 268.3377685546875, 337.9013366699219, 465.5304870605469, 224.19883728027344, 343.6834716796875, 150.39991760253906, 27.74933624267578, 162.530029296875, -44.79377746582031, 623.493408203125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000526.npy"}
|
|
{"epoch": 0.7723935389133627, "step": 527, "batch_size": 64, "mean": 185.0948486328125, "std": 206.33306884765625, "min": -210.9775848388672, "p10": -71.8046920776367, "median": 166.86638641357422, "p90": 509.65101318359376, "max": 759.5501708984375, "pos_frac": 0.796875, "sample": [11.576103210449219, 445.9684753417969, 350.76904296875, 250.19720458984375, 326.95220947265625, 167.34201049804688, 759.5501708984375, -34.614166259765625, -49.665802001953125, 144.4609375, 27.577377319335938, -26.816131591796875, 546.6995849609375, 169.3655548095703, -75.89826202392578, -129.31561279296875, 240.71734619140625, -16.439064025878906, 100.89458465576172, -82.23858642578125, -210.9775848388672, 530.5254516601562, 161.5589141845703, 140.4971466064453, 102.15093994140625, 399.4748229980469, 2.8629302978515625, -52.37718963623047, 308.7681579589844, 416.6988525390625, 284.67926025390625, 323.3017272949219, 241.57156372070312, 270.80364990234375, 553.9589233398438, 569.5422973632812, 519.3558349609375, -62.253028869628906, 34.70481872558594, 282.3434143066406, 507.15478515625, 164.96380615234375, 510.7208251953125, 166.39076232910156, -110.51089477539062, 211.58724975585938, 277.14776611328125, 164.36184692382812, 146.04669189453125, -146.92958068847656, 231.87335205078125, 63.98879623413086, 186.1471710205078, 191.0767822265625, 115.72482299804688, 85.30711364746094, 386.37451171875, 245.5329132080078, 230.1666259765625, 180.8256378173828, 122.54281616210938, -106.50611114501953, 71.0501480102539, 6.7584381103515625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000527.npy"}
|
|
{"epoch": 0.7738619676945668, "step": 528, "batch_size": 64, "mean": 212.27090454101562, "std": 286.2279357910156, "min": -313.346923828125, "p10": -101.1538604736328, "median": 207.52323150634766, "p90": 599.8544006347656, "max": 1175.343994140625, "pos_frac": 0.75, "sample": [-1.4620742797851562, 279.829833984375, 425.6782531738281, 192.402099609375, 130.23638916015625, -313.346923828125, -207.79656982421875, 543.927978515625, 455.63397216796875, 248.9188232421875, 368.00067138671875, -53.83927917480469, 255.2091064453125, -86.67262268066406, 219.35714721679688, 65.55066680908203, 287.14691162109375, -63.907073974609375, 683.8038940429688, 1175.343994140625, -106.79255676269531, 480.077392578125, 129.993896484375, 308.92779541015625, 216.228759765625, 635.4364624023438, -130.46075439453125, 775.11181640625, -36.20195770263672, 313.5727844238281, -257.28973388671875, 69.430419921875, 226.19512939453125, 36.88156509399414, 366.4471435546875, 167.5471954345703, 123.23392486572266, 255.61537170410156, 343.8718566894531, 298.8423156738281, 7.013723373413086, 560.91259765625, 5.038909912109375, -161.22036743164062, 650.973388671875, 58.232810974121094, 294.3308410644531, 788.2398071289062, 198.8177032470703, 305.8184814453125, -87.99690246582031, -38.69013977050781, 238.10801696777344, 587.8434448242188, -65.36251831054688, 605.001953125, 308.4673156738281, 43.98530197143555, -204.73951721191406, -3.6446075439453125, 54.330596923828125, 22.634502410888672, 522.5479125976562, 74.01065826416016], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000528.npy"}
|
|
{"epoch": 0.775330396475771, "step": 529, "batch_size": 64, "mean": 161.04995727539062, "std": 241.33319091796875, "min": -380.197265625, "p10": -144.55311584472656, "median": 165.78802490234375, "p90": 426.5684783935547, "max": 845.5852661132812, "pos_frac": 0.75, "sample": [145.44532775878906, -117.42027282714844, 191.847900390625, -145.20465087890625, 172.07968139648438, 218.72874450683594, 341.2595520019531, 394.79296875, 64.68002319335938, 845.5852661132812, 110.96385955810547, 417.9026184082031, 189.381591796875, -74.96979522705078, 276.9875183105469, -32.47174835205078, 126.76331329345703, 19.408836364746094, 264.21771240234375, 97.13253021240234, -90.96068572998047, 154.1121826171875, 515.3508911132812, 499.37261962890625, 499.23785400390625, -317.56463623046875, -28.32453155517578, 411.2773132324219, -200.1844024658203, 498.83544921875, 227.98187255859375, 159.49636840820312, 38.418479919433594, 374.66168212890625, 44.37060546875, -11.84544563293457, 119.11126708984375, 254.16343688964844, 424.11334228515625, 380.143310546875, -380.197265625, 199.11065673828125, 393.7987365722656, -295.6053466796875, 374.43304443359375, -238.42469787597656, 262.1611022949219, 152.270263671875, 507.63189697265625, 12.512550354003906, 18.542768478393555, 346.76727294921875, -30.228612899780273, 224.55332946777344, -230.2868194580078, 173.89642333984375, 55.621604919433594, 425.03173828125, 427.2270812988281, 345.5368347167969, -103.26973724365234, 40.10841369628906, -143.03286743164062, 310.1612548828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000529.npy"}
|
|
{"epoch": 0.7767988252569751, "step": 530, "batch_size": 64, "mean": 253.9014892578125, "std": 288.2153015136719, "min": -731.5819702148438, "p10": -56.088294601440424, "median": 215.51998901367188, "p90": 625.290966796875, "max": 876.6268920898438, "pos_frac": 0.84375, "sample": [456.9662780761719, 78.38943481445312, 8.60198974609375, 170.80618286132812, 203.87551879882812, -73.39025115966797, 585.3994750976562, 481.5776062011719, 171.47012329101562, -138.25241088867188, 199.8800811767578, -138.666259765625, 77.898681640625, 19.76947021484375, 474.822998046875, 423.33660888671875, 52.58885192871094, 196.27601623535156, -58.79210662841797, 497.077880859375, 625.71826171875, 371.12548828125, 770.3416748046875, 194.38800048828125, 36.20674133300781, 365.62982177734375, -42.70733642578125, 440.0715637207031, -49.77939987182617, 519.0531005859375, 7.05693244934082, 711.68359375, 691.234619140625, 288.42401123046875, 442.212890625, 127.53883361816406, 624.2939453125, 176.7150115966797, -120.77828979492188, -127.87843322753906, 62.85740661621094, 494.2481994628906, 314.7908935546875, 876.6268920898438, -24.257322311401367, 227.16445922851562, 513.0645751953125, 43.427696228027344, 793.7787475585938, 91.9511489868164, 292.8116455078125, 41.27507781982422, 598.3566284179688, -731.5819702148438, 700.73095703125, 254.7848358154297, 43.551239013671875, 461.118896484375, 376.3932189941406, 414.1557922363281, 75.68635559082031, 243.83819580078125, 246.60946655273438, 98.1254653930664], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000530.npy"}
|
|
{"epoch": 0.7782672540381792, "step": 531, "batch_size": 64, "mean": 218.2183837890625, "std": 248.6769561767578, "min": -402.9283752441406, "p10": -39.28515396118163, "median": 195.38721466064453, "p90": 605.3448120117188, "max": 821.0076904296875, "pos_frac": 0.78125, "sample": [305.1995544433594, -9.313730239868164, 440.9230041503906, 721.8766479492188, 365.57012939453125, -14.826309204101562, 88.41950225830078, -4.043540954589844, 821.0076904296875, -23.62567138671875, 34.74742126464844, 289.00946044921875, 11.022176742553711, 299.0246887207031, 227.49490356445312, 152.6002655029297, 252.92620849609375, -5.7063446044921875, -155.40390014648438, 220.08441162109375, 721.5611572265625, 123.39634704589844, 450.874755859375, 73.714599609375, 428.46185302734375, 188.3552703857422, 73.10124206542969, 347.19482421875, 328.13330078125, -45.996360778808594, 607.9824829101562, 307.9725341796875, 292.154052734375, 105.54299926757812, 62.17073059082031, 202.41915893554688, -7.708137512207031, -143.58163452148438, -10.367864608764648, 430.466064453125, 513.3028564453125, 391.14141845703125, 119.11693572998047, 73.00605773925781, 701.8988037109375, -402.9283752441406, 176.47525024414062, 405.665771484375, 414.2994079589844, 609.841796875, 612.816650390625, 82.39852142333984, 599.1902465820312, 71.60965728759766, 255.42977905273438, 30.536529541015625, 225.81204223632812, 85.64942932128906, 81.76773834228516, -67.03556823730469, 252.76577758789062, 421.4106140136719, -131.58102416992188, -109.44802856445312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000531.npy"}
|
|
{"epoch": 0.7797356828193832, "step": 532, "batch_size": 64, "mean": 227.61959838867188, "std": 298.1923522949219, "min": -472.5634460449219, "p10": -91.60445175170898, "median": 211.65110778808594, "p90": 642.8784423828125, "max": 1010.0601806640625, "pos_frac": 0.78125, "sample": [650.023681640625, 332.96490478515625, 572.5650024414062, 4.567558288574219, 714.9532470703125, 189.59307861328125, 607.0752563476562, 474.0739440917969, -472.5634460449219, 160.5695343017578, -168.1655731201172, 341.43914794921875, 188.38865661621094, 540.299560546875, -31.774734497070312, 163.13290405273438, 429.4411315917969, -333.1434326171875, 325.1968994140625, 235.90493774414062, 392.8445739746094, 753.5929565429688, 735.4685668945312, 648.2037353515625, 284.72125244140625, 323.0262145996094, -9.683059692382812, 687.7184448242188, 517.9501953125, 57.8955078125, 2.1962203979492188, -63.353553771972656, -82.40570068359375, 625.1082763671875, 242.96046447753906, 81.968994140625, 221.51513671875, 630.4527587890625, 130.72415161132812, 282.1352844238281, -55.72156524658203, 455.4779052734375, 136.98721313476562, 341.4551086425781, 287.15997314453125, 75.18022918701172, 40.20026397705078, 447.198486328125, 103.3424301147461, -41.450050354003906, -82.59996795654297, -152.87472534179688, 64.44996643066406, 1010.0601806640625, 250.06141662597656, 201.78707885742188, 27.02146339416504, -416.88043212890625, 122.74453735351562, 227.25213623046875, -95.46351623535156, 235.4190673828125, -178.79852294921875, 178.06423950195312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000532.npy"}
|
|
{"epoch": 0.7812041116005873, "step": 533, "batch_size": 64, "mean": 186.47169494628906, "std": 274.09326171875, "min": -362.71514892578125, "p10": -138.28216705322262, "median": 172.05514526367188, "p90": 549.2430419921875, "max": 948.9483642578125, "pos_frac": 0.71875, "sample": [-48.703704833984375, -177.9103240966797, 537.2493896484375, -70.99483489990234, 72.89823913574219, 201.135986328125, 759.4234008789062, 471.4876708984375, 238.02740478515625, 232.16754150390625, 580.037353515625, 486.5138854980469, 480.5379638671875, 528.0586547851562, -173.84793090820312, 437.2374267578125, 510.25799560546875, 216.02645874023438, 119.16619873046875, -159.56460571289062, 429.24957275390625, 27.35162353515625, 126.69158935546875, 454.9400329589844, 126.36297607421875, 321.890625, 55.45879364013672, -362.71514892578125, 61.072879791259766, 948.9483642578125, 174.7578125, 290.7239074707031, -47.0081787109375, 9.183570861816406, 53.93136215209961, -316.56134033203125, -75.21578216552734, 235.2540740966797, -102.01908874511719, 174.26556396484375, 31.229389190673828, 152.50550842285156, 604.4567260742188, 554.3831787109375, -52.028167724609375, 194.8334197998047, -153.823486328125, -5.267608642578125, 227.42593383789062, 143.36593627929688, 237.65689086914062, -35.18351745605469, 443.0195007324219, 55.594688415527344, -98.35346221923828, -62.249244689941406, -222.0943145751953, 169.8447265625, 231.27685546875, 621.9698486328125, -92.22896575927734, 181.40908813476562, 671.6653442382812, 309.0134582519531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000533.npy"}
|
|
{"epoch": 0.7826725403817915, "step": 534, "batch_size": 64, "mean": 197.724365234375, "std": 245.4235382080078, "min": -484.81280517578125, "p10": -69.56170768737792, "median": 174.89361572265625, "p90": 519.1689544677735, "max": 783.1356201171875, "pos_frac": 0.796875, "sample": [178.4031982421875, 437.51190185546875, -18.45068359375, 746.6603393554688, 68.52767944335938, 199.47470092773438, -243.60186767578125, 301.0776672363281, 262.17767333984375, -149.92156982421875, 77.63387298583984, 692.778076171875, -31.686683654785156, 102.66384887695312, 510.4074401855469, 660.190185546875, 274.11651611328125, -75.44945526123047, 255.30825805664062, 426.3159484863281, 157.50404357910156, -56.067623138427734, 150.72244262695312, 358.35107421875, 252.66453552246094, 226.4017791748047, 783.1356201171875, 29.531681060791016, 97.80314636230469, 174.33352661132812, 178.44677734375, 571.191162109375, 144.62344360351562, -75.34488677978516, 384.2158508300781, -484.81280517578125, -43.949798583984375, 430.0183410644531, 522.9238891601562, 368.5054016113281, 119.94143676757812, 91.02159118652344, -10.691652297973633, 139.24880981445312, 410.10211181640625, 163.2247772216797, 234.10496520996094, 238.34242248535156, 328.0704345703125, 647.8518676757812, 363.70928955078125, 44.892738342285156, 170.60366821289062, 78.88609313964844, 217.44680786132812, 50.90478515625, 175.45370483398438, 121.8700180053711, 325.8985595703125, -53.557861328125, 305.39300537109375, -309.19329833984375, -113.40821075439453, 69.9080810546875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000534.npy"}
|
|
{"epoch": 0.7841409691629956, "step": 535, "batch_size": 64, "mean": 248.56649780273438, "std": 239.7193603515625, "min": -190.57472229003906, "p10": -44.71237754821773, "median": 199.56495666503906, "p90": 561.5038024902345, "max": 882.184814453125, "pos_frac": 0.875, "sample": [-106.05482482910156, 882.184814453125, 65.36579132080078, 42.2991943359375, -132.95594787597656, 154.32516479492188, 96.14103698730469, 529.6860961914062, 77.133056640625, 573.9291381835938, 192.27455139160156, 304.3988952636719, 478.7986145019531, 727.376953125, 343.1247863769531, 192.3440704345703, 10.875617980957031, 672.9108276367188, 163.39730834960938, 305.38092041015625, 575.203125, 352.67236328125, 218.4263916015625, 392.8482360839844, 183.71287536621094, 527.7069702148438, 295.16796875, 322.68316650390625, 1.8371295928955078, 183.39901733398438, 404.7559814453125, 426.80206298828125, 446.611083984375, -0.9279975891113281, 58.807106018066406, -190.57472229003906, 481.9919738769531, 422.76934814453125, 142.27395629882812, 250.21270751953125, 314.22332763671875, 744.091552734375, 87.32979583740234, 194.04632568359375, 267.9678955078125, 38.94648742675781, 142.13888549804688, 532.5113525390625, -78.54624938964844, -63.47711181640625, 430.05364990234375, 52.85124969482422, 246.09178161621094, 120.38668060302734, -101.99699401855469, -125.0526123046875, 642.310791015625, 64.96371459960938, 531.78369140625, 110.35540008544922, 21.82112693786621, 382.77532958984375, 78.2823486328125, 205.08358764648438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000535.npy"}
|
|
{"epoch": 0.7856093979441997, "step": 536, "batch_size": 64, "mean": 293.05352783203125, "std": 304.9113464355469, "min": -258.0986328125, "p10": -0.5267147064208975, "median": 238.24129486083984, "p90": 645.3651428222657, "max": 1689.8536376953125, "pos_frac": 0.890625, "sample": [123.34590911865234, 202.9470977783203, 252.1962432861328, 364.2900085449219, 970.9573364257812, 132.9099578857422, 191.98556518554688, 0.49395751953125, 5.890071868896484, -1.2201080322265625, 264.7508239746094, 255.24810791015625, 555.3134765625, 662.44287109375, 87.64274597167969, 460.57568359375, 843.8231811523438, 83.53724670410156, 549.27734375, 440.493896484375, 630.6326293945312, 329.4111022949219, -136.02474975585938, 355.4382629394531, 300.9538269042969, 745.66796875, 27.36008071899414, 475.34173583984375, 393.1312561035156, 99.90927124023438, 1689.8536376953125, 548.6522827148438, 266.80047607421875, 291.3258361816406, 357.5408020019531, 203.46600341796875, 160.5779266357422, 651.6790771484375, -0.9641456604003906, -62.220645904541016, 177.11807250976562, -133.6273956298828, 606.6298217773438, 273.5579833984375, 311.025634765625, -157.7539520263672, 97.369873046875, 146.33395385742188, 243.20164489746094, 123.203857421875, 226.71792602539062, 36.387718200683594, 151.79656982421875, 211.24880981445312, 207.6341552734375, 614.9508056640625, 176.50437927246094, 5.894329071044922, -258.0986328125, 181.19699096679688, 381.19671630859375, 436.7512512207031, 233.28094482421875, 687.4697265625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000536.npy"}
|
|
{"epoch": 0.7870778267254038, "step": 537, "batch_size": 64, "mean": 226.4695281982422, "std": 275.76800537109375, "min": -379.6746520996094, "p10": -43.954516601562474, "median": 189.97844696044922, "p90": 601.513720703125, "max": 806.9483642578125, "pos_frac": 0.84375, "sample": [-367.5252380371094, -302.161376953125, 14.037696838378906, 60.568336486816406, -52.96553039550781, 140.69215393066406, 39.85737609863281, 649.685302734375, 22.8138427734375, 15.819149017333984, 367.09039306640625, -20.924789428710938, 392.69891357421875, 113.91859436035156, -213.87940979003906, 530.1007690429688, 657.1239013671875, -97.52365112304688, 288.8479309082031, 202.11328125, 29.000215530395508, 528.7141723632812, 66.29239654541016, -215.69882202148438, 368.94573974609375, -21.449844360351562, 806.9483642578125, 595.3394165039062, 261.8236999511719, 345.4373779296875, 700.7136840820312, 173.7501220703125, 545.0275268554688, 346.2109375, 273.58331298828125, 186.4920654296875, 122.60475158691406, 368.6051940917969, 194.92245483398438, -22.928817749023438, 27.961883544921875, 397.6643981933594, 24.832839965820312, -379.6746520996094, 106.06092834472656, 496.8814697265625, 76.47317504882812, 193.46482849121094, 233.92861938476562, 631.344970703125, 67.69059753417969, 436.9513244628906, 29.078529357910156, 365.2696533203125, 531.0237426757812, 28.301536560058594, 788.7965698242188, 124.39434814453125, 572.0540161132812, 304.55242919921875, 150.76315307617188, 604.1598510742188, 81.42684936523438, 505.927490234375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000537.npy"}
|
|
{"epoch": 0.788546255506608, "step": 538, "batch_size": 64, "mean": 239.66844177246094, "std": 252.4333953857422, "min": -269.494140625, "p10": -47.605553817749026, "median": 249.5909194946289, "p90": 559.0314575195313, "max": 886.4642333984375, "pos_frac": 0.828125, "sample": [12.235916137695312, 344.39501953125, 444.2093505859375, 602.985107421875, 505.32147216796875, -145.39703369140625, 577.14794921875, 132.56533813476562, 287.7735595703125, 136.24609375, -32.68889617919922, 491.855224609375, 474.41497802734375, 313.06597900390625, 121.09648132324219, 544.4780883789062, 295.6917724609375, -47.247962951660156, 259.7706298828125, 583.880126953125, 256.60028076171875, 120.44699096679688, 50.572601318359375, 451.1715087890625, 131.27468872070312, 178.20668029785156, 85.201904296875, 826.4749755859375, 0.8507232666015625, -41.723724365234375, -269.494140625, 385.8135681152344, 129.7926483154297, -110.28385925292969, 886.4642333984375, 305.325439453125, 9.845394134521484, 323.345703125, 227.6563262939453, 497.5874938964844, 78.992919921875, -127.70563507080078, 14.327938079833984, 484.686279296875, 455.7243957519531, 463.59259033203125, -47.465206146240234, 282.41973876953125, -86.08163452148438, 110.70510864257812, 283.20062255859375, 315.67572021484375, 242.58155822753906, 283.5563659667969, 13.834554672241211, 87.2822036743164, 283.11895751953125, 151.35108947753906, 818.529541015625, 565.2686157226562, -98.01301574707031, -47.66570281982422, 442.8441467285156, 21.08998680114746], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000538.npy"}
|
|
{"epoch": 0.7900146842878121, "step": 539, "batch_size": 64, "mean": 196.82864379882812, "std": 235.99432373046875, "min": -215.4361572265625, "p10": -87.42426376342772, "median": 195.7155532836914, "p90": 482.08575134277345, "max": 794.8954467773438, "pos_frac": 0.734375, "sample": [220.74929809570312, 273.835693359375, 66.42430877685547, 19.284215927124023, 197.07125854492188, 344.87481689453125, -175.79209899902344, -47.399009704589844, -19.40875244140625, 464.544677734375, -92.28931427001953, -17.739267349243164, 465.1446533203125, 194.35984802246094, 358.89288330078125, 451.5705261230469, 528.6262817382812, 168.4365997314453, 24.16314697265625, 440.1761474609375, 794.8954467773438, 230.90365600585938, 459.71453857421875, 468.0206604003906, 108.60753631591797, 5.229892730712891, -62.91160583496094, 505.697021484375, 133.84359741210938, 11.959678649902344, -210.04891967773438, 274.90374755859375, -215.4361572265625, 544.72705078125, 219.08157348632812, -8.3154296875, 38.80200958251953, 419.4090576171875, 79.53056335449219, -29.209747314453125, -109.41682434082031, 710.5147094726562, -138.63211059570312, 282.7916564941406, 142.6884307861328, -76.07247924804688, 485.0235595703125, -36.45500183105469, 227.29794311523438, -135.50442504882812, 206.91012573242188, -14.383659362792969, 607.5390625, 363.3531799316406, 264.68621826171875, 231.12998962402344, 475.2308654785156, 334.4522705078125, 184.32901000976562, 359.5557861328125, -32.31129455566406, 406.73370361328125, 135.61424255371094, 87.02854919433594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000539.npy"}
|
|
{"epoch": 0.7914831130690162, "step": 540, "batch_size": 64, "mean": 291.5929260253906, "std": 301.5835876464844, "min": -284.1638488769531, "p10": -88.55144424438475, "median": 280.3258361816406, "p90": 726.8307434082033, "max": 1049.198486328125, "pos_frac": 0.875, "sample": [376.41461181640625, 369.56793212890625, 460.8470153808594, 26.511215209960938, 1003.0068969726562, 56.747039794921875, 278.8914489746094, 16.19073486328125, -284.1638488769531, -261.25982666015625, -143.40899658203125, 387.0436706542969, 116.48524475097656, 418.91961669921875, 917.9993896484375, 66.44921112060547, 206.92108154296875, 179.71621704101562, 365.1006774902344, 470.06549072265625, -91.04006958007812, 43.08831024169922, 163.712158203125, 296.51495361328125, 448.6911926269531, 583.860595703125, 376.3822326660156, 828.5177001953125, 535.42431640625, 292.2234802246094, 281.7602233886719, 269.0460205078125, 251.2958221435547, 101.22721099853516, 174.12600708007812, 384.5994567871094, 55.084625244140625, 1.0597381591796875, 513.336181640625, 215.72256469726562, 523.9093017578125, 1049.198486328125, 843.0721435546875, 824.36328125, -209.96145629882812, 602.7618408203125, 140.536865234375, 340.9481201171875, 315.24334716796875, 292.9713134765625, -127.35453796386719, 130.40261840820312, 461.46942138671875, 463.3600769042969, 199.57708740234375, 214.9976348876953, 697.778564453125, 178.65040588378906, 739.2816772460938, 23.20781707763672, -82.7446517944336, 398.6969299316406, 79.10258483886719, -190.19976806640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000540.npy"}
|
|
{"epoch": 0.7929515418502202, "step": 541, "batch_size": 64, "mean": 182.57830810546875, "std": 279.73248291015625, "min": -341.9859619140625, "p10": -181.41219635009762, "median": 173.08106994628906, "p90": 559.9781250000001, "max": 1099.7921142578125, "pos_frac": 0.734375, "sample": [-247.23171997070312, 167.69393920898438, 305.68914794921875, 171.220458984375, 116.8899154663086, 65.491943359375, 254.68008422851562, 120.61518859863281, 32.9119873046875, -88.2900390625, -148.15524291992188, 797.9613647460938, -197.98158264160156, 85.36402893066406, 285.9195861816406, 12.573432922363281, 177.266845703125, 885.8599853515625, 80.72835540771484, 283.9751892089844, 1099.7921142578125, 455.7244567871094, 132.54702758789062, -251.3665313720703, 174.94168090820312, -204.6802978515625, 190.2667236328125, 158.13348388671875, -341.9859619140625, 528.9212646484375, 303.1705627441406, 52.023826599121094, 433.8522644042969, 132.51466369628906, -21.454803466796875, 154.77438354492188, 240.25534057617188, 188.508544921875, 428.78875732421875, 177.44801330566406, -62.39329528808594, 224.0946502685547, 418.64471435546875, 616.0465698242188, 573.2882080078125, 256.1205749511719, -106.09367370605469, -198.571044921875, 306.4929504394531, 410.60662841796875, -71.48145294189453, 218.16317749023438, 330.1047058105469, 658.5457153320312, 591.8809204101562, -0.6423606872558594, -195.66517639160156, 126.7494125366211, -145.14883422851562, -69.15120697021484, -9.671501159667969, 222.54013061523438, 219.98336791992188, 175.2103729248047], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000541.npy"}
|
|
{"epoch": 0.7944199706314243, "step": 542, "batch_size": 64, "mean": 246.30075073242188, "std": 283.7776794433594, "min": -383.94940185546875, "p10": -78.11454162597656, "median": 190.77896881103516, "p90": 605.5169982910156, "max": 1114.7796630859375, "pos_frac": 0.8125, "sample": [195.7036590576172, 167.76947021484375, 180.69119262695312, 82.57263946533203, -81.4643783569336, 693.103515625, -104.84832763671875, 1114.7796630859375, 585.2448120117188, 215.95550537109375, 91.22453308105469, 419.58551025390625, 480.6408386230469, -111.80538940429688, 80.15912628173828, 369.85284423828125, 691.093017578125, 354.96490478515625, -23.706932067871094, 751.4717407226562, 183.62759399414062, 264.61761474609375, 277.4171447753906, 109.31854248046875, -193.89309692382812, 181.9073944091797, 30.84088134765625, -37.36949920654297, -41.53471374511719, 606.6492309570312, -50.07842254638672, 548.1754150390625, 345.72247314453125, 92.86038208007812, 773.8430786132812, 18.70455551147461, 459.9718017578125, 216.56857299804688, -199.31826782226562, 452.8508605957031, 286.6796569824219, 280.6935729980469, 174.9151611328125, -70.29825592041016, 658.89404296875, 177.56951904296875, 155.97784423828125, -383.94940185546875, 475.4293212890625, 82.67342376708984, 185.85427856445312, 304.83447265625, 59.81874084472656, 364.7398986816406, 565.931396484375, 375.102783203125, 99.57843780517578, -363.0968322753906, 518.3189697265625, 177.36495971679688, 86.00162506103516, 337.2708740234375, 416.19866943359375, 602.8751220703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000542.npy"}
|
|
{"epoch": 0.7958883994126285, "step": 543, "batch_size": 64, "mean": 174.83230590820312, "std": 287.4061279296875, "min": -401.25665283203125, "p10": -105.12508392333984, "median": 114.38514709472656, "p90": 545.2955505371094, "max": 973.6298828125, "pos_frac": 0.75, "sample": [18.15045166015625, -88.4935302734375, 367.0841369628906, -3.350982666015625, 110.00391387939453, 934.4085693359375, 5.081520080566406, 7.7672576904296875, -401.25665283203125, 472.1539611816406, -46.078224182128906, -294.5390625, 34.144466400146484, 219.97189331054688, 782.2928466796875, 106.75660705566406, 44.55726623535156, 812.3330078125, -292.6275939941406, -6.588279724121094, 306.03076171875, -26.18163299560547, 25.685420989990234, 30.829084396362305, -201.21542358398438, 118.7663803100586, -108.1483154296875, 421.71905517578125, 148.990234375, 14.47430419921875, 480.15087890625, 143.75746154785156, -282.31475830078125, 560.8939208984375, -94.74594116210938, 71.70280456542969, 973.6298828125, 437.9714050292969, 546.3450927734375, 28.76325225830078, 544.8538208007812, -27.491439819335938, -98.07087707519531, 143.18589782714844, 500.08599853515625, 31.01903533935547, 302.50762939453125, 545.48486328125, -137.3992462158203, 320.9685974121094, 184.41888427734375, -32.196250915527344, 223.4464874267578, 46.499855041503906, 318.2270202636719, 405.1055908203125, 214.65658569335938, 118.9779281616211, 54.23316955566406, 206.99020385742188, 20.37769317626953, 177.41470336914062, 353.6148986816406, 393.48046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000543.npy"}
|
|
{"epoch": 0.7973568281938326, "step": 544, "batch_size": 64, "mean": 255.83843994140625, "std": 249.7470245361328, "min": -254.34368896484375, "p10": 12.50353450775147, "median": 234.8062515258789, "p90": 530.9213195800781, "max": 1251.6666259765625, "pos_frac": 0.90625, "sample": [447.53271484375, 203.10955810546875, 336.86102294921875, 535.281005859375, 1251.6666259765625, -167.12637329101562, 341.9200134277344, 30.75916290283203, 467.2118225097656, -3.095399856567383, 800.2247314453125, 42.00880432128906, 315.70562744140625, 201.5797882080078, 10.296669006347656, 373.499755859375, 275.9037780761719, 260.1513671875, 212.9959716796875, 348.1884765625, 240.66246032714844, 303.11676025390625, 54.219261169433594, 254.56561279296875, 119.94876098632812, 157.8932342529297, 683.1109008789062, 126.0322265625, -178.54405212402344, 17.65288734436035, 476.2633056640625, 228.95004272460938, 31.883398056030273, 400.91741943359375, -62.51225280761719, 166.69554138183594, 385.9041748046875, 225.53904724121094, 328.6112976074219, -39.611473083496094, 248.90487670898438, 135.97174072265625, 210.68902587890625, 520.7487182617188, 387.4701843261719, 159.14707946777344, 279.69512939453125, 685.1040649414062, 348.778564453125, 47.174747467041016, 27.513408660888672, 128.68173217773438, 382.1517028808594, 696.9195556640625, 409.53607177734375, 23.301555633544922, 319.20819091796875, 30.792617797851562, 445.22479248046875, 99.28190612792969, 123.13307189941406, 598.833251953125, -254.34368896484375, 113.76726531982422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000544.npy"}
|
|
{"epoch": 0.7988252569750367, "step": 545, "batch_size": 64, "mean": 252.58734130859375, "std": 258.9729919433594, "min": -249.03640747070312, "p10": -57.003440093994115, "median": 271.70338439941406, "p90": 601.3989990234376, "max": 756.30419921875, "pos_frac": 0.78125, "sample": [-6.032503128051758, -28.71148681640625, -248.43179321289062, -20.198211669921875, 122.06600952148438, 316.08892822265625, 341.30401611328125, -1.1522960662841797, 286.5147705078125, 130.02255249023438, 453.7720642089844, 668.0472412109375, -66.24652862548828, 389.34735107421875, 606.7408447265625, 40.29944610595703, 582.5839233398438, 480.3262634277344, 721.14306640625, 362.0736999511719, 402.70172119140625, 186.27418518066406, 402.4681396484375, 237.86915588378906, -22.220436096191406, 93.16653442382812, -149.57705688476562, 268.38055419921875, 461.7713317871094, -0.6624393463134766, 707.46142578125, 167.8680419921875, 184.2696075439453, -35.43623352050781, 722.998779296875, 427.1858825683594, 98.67250061035156, 588.070556640625, -180.99183654785156, 309.1448974609375, 6.272346496582031, 461.1059265136719, 157.91363525390625, 275.3039245605469, 255.23239135742188, 451.60247802734375, 399.54107666015625, 756.30419921875, 250.69021606445312, 289.9168701171875, -132.78843688964844, 61.51344299316406, 93.2385025024414, 275.0262145996094, 306.5685729980469, 588.9346923828125, 613.3678588867188, -191.85549926757812, 316.427978515625, 91.46257781982422, 419.9689025878906, -249.03640747070312, 104.8304672241211, 565.0751953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000545.npy"}
|
|
{"epoch": 0.8002936857562408, "step": 546, "batch_size": 64, "mean": 208.7684326171875, "std": 222.02651977539062, "min": -474.12744140625, "p10": -44.36222152709961, "median": 215.06877899169922, "p90": 460.8604736328125, "max": 718.2546997070312, "pos_frac": 0.859375, "sample": [-107.06076049804688, 87.53388214111328, 384.58056640625, 158.88253784179688, 276.1946105957031, 376.87286376953125, 90.4916000366211, 215.79771423339844, 181.45108032226562, 148.78948974609375, -45.29924774169922, -176.7984619140625, 15.687850952148438, 43.3912353515625, 234.576904296875, 91.62318420410156, 28.99597930908203, 357.3451232910156, 522.609619140625, 107.81753540039062, -2.1099472045898438, 528.3411865234375, -130.21243286132812, 388.1195373535156, 314.89886474609375, 355.7774658203125, 146.0330047607422, 281.665283203125, 456.922119140625, 214.33984375, 462.54833984375, -169.2367706298828, 150.61825561523438, 143.80227661132812, 354.05889892578125, -349.52056884765625, 373.29644775390625, 318.8336181640625, 420.1738586425781, 718.2546997070312, -474.12744140625, 483.2608642578125, 361.4386901855469, 132.68698120117188, 349.5752258300781, 646.0432739257812, -42.17582702636719, 383.6051025390625, 114.5641098022461, 233.75186157226562, 139.1995849609375, 260.98406982421875, 198.804443359375, 310.52398681640625, 47.132667541503906, 444.0571594238281, 105.3179931640625, 364.86981201171875, 3.7842025756835938, 105.02376556396484, 54.7413330078125, 406.73956298828125, 286.27227783203125, 475.0183410644531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000546.npy"}
|
|
{"epoch": 0.801762114537445, "step": 547, "batch_size": 64, "mean": 206.67840576171875, "std": 276.90045166015625, "min": -693.2713012695312, "p10": -65.91431503295898, "median": 151.2628936767578, "p90": 520.5647277832031, "max": 971.9298706054688, "pos_frac": 0.78125, "sample": [-5.555051803588867, 245.0703125, -693.2713012695312, 667.474609375, 47.8570442199707, -16.5408935546875, 140.2616729736328, 141.82498168945312, -21.1983642578125, 517.2294311523438, 130.88912963867188, 762.4274291992188, 96.70077514648438, 971.9298706054688, -31.718154907226562, 27.89275360107422, 468.11248779296875, 46.79798126220703, -180.3142547607422, 503.589599609375, 29.83936309814453, 83.53113555908203, 354.3550109863281, 160.7008056640625, 401.8552551269531, 91.19927978515625, 433.391357421875, 321.87646484375, 196.2237091064453, 209.11483764648438, 521.994140625, 96.78663635253906, 210.18569946289062, 484.9806823730469, 273.17828369140625, 136.38064575195312, -29.590167999267578, 320.58404541015625, 287.5524597167969, -65.32293701171875, 607.2469482421875, 744.58544921875, 288.365478515625, -188.2066192626953, -80.55722045898438, -66.16776275634766, -251.41050720214844, 405.07025146484375, 90.9383544921875, 287.3144836425781, 475.26055908203125, 18.196273803710938, 210.48635864257812, 337.6369934082031, 123.39241027832031, 100.49879455566406, 476.935546875, 8.83404541015625, 371.3039855957031, -41.376434326171875, 493.223388671875, 541.1664428710938, 124.9345703125, -188.53082275390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000547.npy"}
|
|
{"epoch": 0.8032305433186491, "step": 548, "batch_size": 64, "mean": 215.37667846679688, "std": 280.3435363769531, "min": -504.5769958496094, "p10": -69.11094589233399, "median": 188.74560546875, "p90": 495.22872009277353, "max": 1350.515380859375, "pos_frac": 0.8125, "sample": [77.64900207519531, 280.521240234375, 130.78012084960938, 286.5506286621094, -129.7806396484375, 8.975133895874023, -504.5769958496094, 786.3550415039062, 316.8880615234375, 282.1042785644531, 355.59857177734375, 293.57843017578125, 434.93072509765625, 572.3225708007812, 335.27484130859375, -95.41083526611328, 29.45991325378418, 388.1745910644531, 465.8053283691406, 24.0472469329834, 533.0868530273438, 313.72064208984375, 176.57766723632812, 59.56250762939453, 476.0269775390625, -68.86075592041016, 281.00067138671875, 160.11553955078125, 262.5255126953125, 201.02040100097656, -57.7933349609375, 1350.515380859375, 421.75830078125, 135.28277587890625, 148.4644317626953, -9.334051132202148, -49.63751220703125, 433.13372802734375, -83.41867065429688, 647.7060546875, 42.464717864990234, -69.21817016601562, -87.15702819824219, 251.83132934570312, 11.158597946166992, 7.144866943359375, 233.4166717529297, 101.947509765625, 330.6373596191406, -60.323150634765625, 236.58447265625, 328.3958435058594, -90.22769165039062, 1036.5390625, 229.75106811523438, 124.24468994140625, 105.47257995605469, 122.73281860351562, 167.71482849121094, 200.91354370117188, 120.7730941772461, 259.2912292480469, 5.860099792480469, 503.4580383300781], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000548.npy"}
|
|
{"epoch": 0.8046989720998532, "step": 549, "batch_size": 64, "mean": 188.5994873046875, "std": 251.0821075439453, "min": -413.82855224609375, "p10": -82.40077209472655, "median": 165.56330108642578, "p90": 504.66812744140634, "max": 939.2162475585938, "pos_frac": 0.796875, "sample": [175.1341552734375, 622.0585327148438, 67.87779235839844, 91.31465911865234, 134.3684539794922, 371.8257141113281, -58.81312561035156, 314.2888488769531, 148.66207885742188, 299.99041748046875, 22.367816925048828, 341.873779296875, -7.894187927246094, 181.91073608398438, 38.12105941772461, -161.44346618652344, 511.60760498046875, -157.22750854492188, -99.23168182373047, 20.586326599121094, 320.24969482421875, 252.17544555664062, -87.79224395751953, 250.25115966796875, 260.3362121582031, 138.17694091796875, 170.73129272460938, 151.37332153320312, 252.47373962402344, 488.47601318359375, 14.66363525390625, 88.93540954589844, 237.5224609375, -15.207878112792969, 288.3425598144531, -234.37051391601562, 202.95181274414062, 559.6302490234375, 160.44656372070312, -20.576763153076172, -191.9298095703125, -6.906608581542969, 310.9658203125, 196.07583618164062, 939.2162475585938, 170.68003845214844, 51.59898376464844, 325.8960876464844, 66.219970703125, 849.2635498046875, 584.9164428710938, -413.82855224609375, 6.311140060424805, 457.555419921875, 345.96514892578125, -69.82067108154297, 143.2474365234375, 57.47787857055664, 344.76446533203125, 817.0455932617188, 49.89875793457031, 324.99139404296875, 86.40695190429688, 288.1884460449219], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000549.npy"}
|
|
{"epoch": 0.8061674008810573, "step": 550, "batch_size": 64, "mean": 189.18296813964844, "std": 206.87850952148438, "min": -270.49609375, "p10": -71.50550155639648, "median": 172.00103759765625, "p90": 432.57814025878906, "max": 697.035888671875, "pos_frac": 0.828125, "sample": [169.60220336914062, 123.07806396484375, 481.63177490234375, 160.41323852539062, 275.6874084472656, 428.83856201171875, 322.7853698730469, 96.93678283691406, 31.06643295288086, 41.95835876464844, 232.64471435546875, 247.43572998046875, 381.4393615722656, 242.15380859375, 152.58302307128906, 481.1293029785156, 119.22908782958984, -65.99513244628906, 69.93525695800781, 333.3917541503906, -270.49609375, -22.148574829101562, 376.5629577636719, 245.3078155517578, -155.20127868652344, 502.127685546875, 302.06207275390625, 174.39987182617188, 252.83328247070312, 43.35234069824219, 25.539962768554688, -190.8431396484375, 206.489501953125, -73.8670883178711, 163.71014404296875, 697.035888671875, 323.97308349609375, 399.53411865234375, 623.8694458007812, 434.1808166503906, 280.6508483886719, 110.38136291503906, -14.630046844482422, 20.089019775390625, 338.31085205078125, 164.00994873046875, 377.795166015625, 279.1727294921875, 145.95870971679688, 323.80340576171875, 206.17376708984375, 504.061767578125, -24.328418731689453, 421.0224609375, -233.462890625, 60.08097839355469, 158.48634338378906, 416.3594055175781, 195.26852416992188, -157.293212890625, 165.1886749267578, -201.45526123046875, 140.8511962890625, 46.84700012207031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000550.npy"}
|
|
{"epoch": 0.8076358296622613, "step": 551, "batch_size": 64, "mean": 284.74884033203125, "std": 248.57144165039062, "min": -260.6473693847656, "p10": -20.187883758544917, "median": 288.4998474121094, "p90": 615.413720703125, "max": 801.1253662109375, "pos_frac": 0.859375, "sample": [19.701480865478516, -14.263961791992188, 297.2040710449219, 263.90802001953125, 302.76025390625, 457.6236267089844, -164.6298828125, 28.642078399658203, 675.2998046875, 181.2047576904297, -8.924049377441406, 189.0001678466797, -54.44728088378906, 470.60986328125, 496.65130615234375, 525.1221923828125, 424.8446044921875, 239.4310302734375, 277.0689697265625, 801.1253662109375, 665.47314453125, 238.5206298828125, 538.9840698242188, 603.0572509765625, 340.5882568359375, 160.88536071777344, 622.9651489257812, 233.48780822753906, 91.18871307373047, 318.319580078125, 131.2747039794922, 62.87331771850586, 596.9130249023438, 87.88580322265625, 299.8651123046875, 334.0445251464844, 390.8394775390625, 431.16351318359375, 389.7539978027344, 68.40394592285156, -171.78857421875, 462.94512939453125, 452.132080078125, 324.7347106933594, -137.78904724121094, 751.30078125, -116.99252319335938, -260.6473693847656, 620.7093505859375, 79.3498306274414, -22.726707458496094, 110.66116333007812, 378.2358093261719, 580.70849609375, 279.7956237792969, 89.71940612792969, 562.3134765625, 350.154541015625, 238.92770385742188, 345.83721923828125, 190.83714294433594, 775.2969360351562, 197.21136474609375, 128.57945251464844], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000551.npy"}
|
|
{"epoch": 0.8091042584434655, "step": 552, "batch_size": 64, "mean": 313.2353820800781, "std": 310.28558349609375, "min": -287.5774230957031, "p10": 1.4756729125976626, "median": 250.48731994628906, "p90": 761.0413818359376, "max": 1257.85302734375, "pos_frac": 0.890625, "sample": [199.21612548828125, 143.59844970703125, 217.333740234375, 963.743896484375, 292.1107482910156, 776.6410522460938, 499.38592529296875, 354.3149108886719, 61.191009521484375, 181.1121826171875, 480.0758056640625, 412.8759460449219, 50.723388671875, 412.2391357421875, 107.87236022949219, 533.6078491210938, 204.66677856445312, 656.3148193359375, 934.257568359375, 7.763763427734375, 130.75634765625, 341.8260498046875, 417.5732727050781, 233.63735961914062, -287.5774230957031, 22.641746520996094, 34.189292907714844, 474.7900695800781, -68.93638610839844, -164.86724853515625, 711.8807373046875, -121.72795104980469, 212.7670135498047, 78.2217025756836, 308.46295166015625, 147.2516326904297, 399.35784912109375, 986.0834350585938, 929.1800537109375, 612.51318359375, 368.13177490234375, 44.724098205566406, 174.27857971191406, 430.7144470214844, 588.4031982421875, 1257.85302734375, 176.8916015625, 10.456802368164062, 806.5285034179688, 724.6421508789062, 267.3372802734375, -1.2192230224609375, 184.70623779296875, 29.19217300415039, -4.708404541015625, 162.83847045898438, -48.47410583496094, 156.8987579345703, 271.3172302246094, 343.1124572753906, 447.59686279296875, 422.810302734375, 323.9908447265625, 21.97338104248047], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000552.npy"}
|
|
{"epoch": 0.8105726872246696, "step": 553, "batch_size": 64, "mean": 202.9493408203125, "std": 259.1405029296875, "min": -238.7708740234375, "p10": -136.40412750244138, "median": 197.11134338378906, "p90": 560.3975646972657, "max": 1182.318359375, "pos_frac": 0.8125, "sample": [1182.318359375, 655.6991577148438, -42.643707275390625, 254.75839233398438, 307.9093017578125, -69.2433853149414, 42.742584228515625, -27.31067657470703, 98.03466796875, 216.39892578125, 57.25027847290039, 110.45726013183594, 147.18463134765625, 178.2783203125, 59.61687469482422, 248.1293487548828, -100.21139526367188, 17.880828857421875, 286.77288818359375, -174.61785888671875, 196.24447631835938, 200.11843872070312, 271.7478942871094, -162.5757293701172, 349.0079345703125, 63.49687194824219, 551.9696655273438, 96.336669921875, 44.707069396972656, 411.81207275390625, -151.91529846191406, 352.778076171875, 103.20632934570312, -177.3999481201172, 257.6682434082031, 198.8228759765625, 172.28341674804688, 187.99609375, 537.3097534179688, 89.73502349853516, 202.42620849609375, -93.50254821777344, 286.5152282714844, 293.88104248046875, 38.887107849121094, 135.6346435546875, -174.18150329589844, 241.2569580078125, 524.0955810546875, 104.52045440673828, 564.009521484375, 206.38766479492188, 572.0006103515625, 232.65406799316406, 633.3448486328125, 566.7108154296875, 43.67938995361328, 411.58544921875, -205.3416748046875, 766.750732421875, -238.7708740234375, 316.8423767089844, 197.97821044921875, 318.6396179199219], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000553.npy"}
|
|
{"epoch": 0.8120411160058737, "step": 554, "batch_size": 64, "mean": 264.222900390625, "std": 272.8453674316406, "min": -400.44677734375, "p10": -75.11426239013672, "median": 249.19900512695312, "p90": 649.614892578125, "max": 857.0433349609375, "pos_frac": 0.828125, "sample": [174.2868194580078, -79.39570617675781, 258.25299072265625, 111.03143310546875, 397.3499450683594, 140.75277709960938, 417.51947021484375, 383.7297058105469, 208.44546508789062, 335.29071044921875, 627.447998046875, 231.080078125, 442.3466491699219, 251.97061157226562, -75.82264709472656, 372.78656005859375, 482.5843505859375, 194.01904296875, 311.018798828125, 190.31777954101562, -85.7232894897461, -243.57595825195312, -62.372039794921875, 28.956398010253906, 857.0433349609375, -49.59680938720703, 226.54788208007812, 385.69073486328125, 215.132568359375, 388.9486083984375, 324.8271484375, 780.4086303710938, 786.1304931640625, 129.8428192138672, 511.02587890625, 208.0157470703125, 701.050048828125, 121.8482666015625, 651.5188598632812, -143.3701629638672, 141.93655395507812, 294.03765869140625, 288.90423583984375, 493.1687927246094, 476.6107482910156, 74.68220520019531, 231.8555908203125, 10.68017578125, 332.18402099609375, 246.42739868164062, -73.46136474609375, 266.5234680175781, 254.97555541992188, -400.44677734375, 645.1723022460938, 10.37139892578125, -194.22610473632812, 82.2050552368164, 506.33184814453125, 482.55352783203125, 827.9879760742188, 742.792236328125, 114.46080780029297, -52.824161529541016], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000554.npy"}
|
|
{"epoch": 0.8135095447870778, "step": 555, "batch_size": 64, "mean": 266.2066650390625, "std": 306.276123046875, "min": -422.2078552246094, "p10": -44.363905334472605, "median": 181.8758773803711, "p90": 768.9130126953127, "max": 1022.2822265625, "pos_frac": 0.890625, "sample": [249.3540802001953, 258.60491943359375, 414.8717956542969, 398.1337890625, 403.5706787109375, 638.7294921875, 171.80714416503906, 287.6919250488281, 17.782989501953125, 105.5306167602539, -67.10655212402344, 647.0940551757812, 584.9515380859375, 267.53594970703125, 60.975341796875, 10.978914260864258, 121.03095245361328, 34.8065185546875, 994.872802734375, 781.0279541015625, 153.82135009765625, 743.233642578125, 176.96633911132812, 513.3139038085938, 256.873291015625, 129.32644653320312, 464.78155517578125, 42.05537414550781, 143.13873291015625, 9.145309448242188, 286.9248046875, 348.74542236328125, 101.85892486572266, 303.5940856933594, 20.292572021484375, 243.27207946777344, 186.78541564941406, -131.83975219726562, 140.7135772705078, 46.46300506591797, -422.2078552246094, 426.8315124511719, 811.0679931640625, 779.91845703125, -122.08831787109375, 105.43196105957031, 873.8692626953125, 22.371137619018555, 335.2769470214844, 169.68690490722656, 888.9158325195312, -250.069091796875, 601.5910034179688, -116.02445983886719, 16.04149627685547, 216.98236083984375, 1022.2822265625, 147.92303466796875, 8.7022705078125, 165.50482177734375, 536.3352661132812, 245.90621948242188, -94.45823669433594, 105.725830078125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000555.npy"}
|
|
{"epoch": 0.8149779735682819, "step": 556, "batch_size": 64, "mean": 191.5848388671875, "std": 262.7500305175781, "min": -287.22265625, "p10": -80.24335174560545, "median": 118.78096008300781, "p90": 611.6067138671875, "max": 1056.40576171875, "pos_frac": 0.78125, "sample": [610.697021484375, 76.94245147705078, 63.044830322265625, 108.9779052734375, 295.0965576171875, 103.3821029663086, 429.1757507324219, 314.00994873046875, -104.11199188232422, 310.77825927734375, 5.282402038574219, 383.65478515625, 257.490234375, 78.10931396484375, 293.0709228515625, 125.3773193359375, 801.8231201171875, 112.18460083007812, 350.8714904785156, -37.228614807128906, -287.22265625, 49.162818908691406, -230.95114135742188, 84.72660064697266, -36.5157470703125, 337.7287292480469, 244.08786010742188, -58.29713439941406, 515.1936645507812, -66.0687255859375, 267.461669921875, 21.84074592590332, 83.89616394042969, -52.38353729248047, 619.5518798828125, -29.9932804107666, 163.26922607421875, 104.15879821777344, 79.3005142211914, 3.6356754302978516, -88.218505859375, 306.7432556152344, 204.099365234375, 201.735595703125, -164.35720825195312, 250.53749084472656, 714.7852783203125, 66.82060241699219, 671.7149658203125, 69.2635269165039, 228.49880981445312, -112.76612091064453, 416.88092041015625, 143.20941162109375, 1056.40576171875, 12.791145324707031, 611.99658203125, 744.3607177734375, 94.98639678955078, 149.76670837402344, 168.07325744628906, -38.92927932739258, -86.31819152832031, 248.1378173828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000556.npy"}
|
|
{"epoch": 0.8164464023494861, "step": 557, "batch_size": 64, "mean": 232.67929077148438, "std": 274.2548828125, "min": -349.78173828125, "p10": -29.192052459716788, "median": 183.75379943847656, "p90": 563.2677673339845, "max": 917.1387939453125, "pos_frac": 0.875, "sample": [-307.098876953125, 208.67601013183594, -349.78173828125, 136.99684143066406, 91.90324401855469, 258.3977966308594, 22.896774291992188, 378.8808288574219, 376.12066650390625, 69.10281372070312, 377.25323486328125, 465.16802978515625, 474.7628479003906, 22.99749755859375, 66.19418334960938, 199.8169708251953, 364.24200439453125, 131.75645446777344, 49.065799713134766, 22.45978546142578, 54.59794616699219, 850.0983276367188, 530.1824951171875, 551.3863525390625, 110.91482543945312, 443.2610168457031, 99.05722045898438, -20.205459594726562, -141.96881103515625, 723.3203125, 159.84368896484375, 262.82220458984375, 568.3598022460938, 271.71771240234375, 138.4669647216797, 88.62054443359375, 283.41748046875, 70.82213592529297, 458.52459716796875, 172.01150512695312, 302.8316345214844, -114.07099151611328, 259.5120849609375, 113.17543029785156, 50.38935852050781, -205.7677001953125, 420.6810607910156, 816.2933349609375, 22.174102783203125, 804.1318359375, 115.90107727050781, 776.35302734375, 917.1387939453125, 230.44668579101562, 269.6306457519531, 32.90470886230469, 29.132686614990234, 442.6590881347656, -183.2118377685547, 528.6567993164062, -33.04344940185547, 270.2970886230469, 195.49609375, 94.70060729980469], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000557.npy"}
|
|
{"epoch": 0.8179148311306902, "step": 558, "batch_size": 64, "mean": 314.78643798828125, "std": 298.69915771484375, "min": -546.8016357421875, "p10": -51.705421447753906, "median": 266.81004333496094, "p90": 714.7025390625, "max": 909.8635864257812, "pos_frac": 0.875, "sample": [245.28636169433594, 182.1538543701172, 240.71766662597656, 456.16729736328125, 138.502685546875, 237.9380645751953, 416.2602233886719, 119.51670837402344, 152.64382934570312, -54.815216064453125, -108.74520874023438, 551.2445678710938, 893.9337768554688, 714.0517578125, 172.81581115722656, 385.8260192871094, 262.6304931640625, 277.63934326171875, 714.9814453125, 567.0091552734375, 143.11654663085938, 188.43060302734375, 182.7447509765625, 360.74395751953125, -63.984046936035156, 205.35409545898438, 701.20361328125, 101.13996887207031, 560.5016479492188, 289.45050048828125, 127.98270416259766, 398.8367614746094, 509.03741455078125, 881.1858520507812, 754.131103515625, 492.60595703125, -387.6960754394531, 66.40084838867188, 402.1341857910156, 228.57676696777344, 225.1868896484375, 98.42221069335938, 148.95018005371094, 320.06671142578125, 169.21717834472656, -51.84181213378906, -546.8016357421875, 909.8635864257812, 586.837646484375, 3.730804443359375, 523.9693603515625, 607.1906127929688, 438.7762756347656, 430.9778747558594, 809.341064453125, -139.8683319091797, 305.1853942871094, 175.36428833007812, 165.49734497070312, -51.387176513671875, 270.9895935058594, 889.590576171875, 597.1963500976562, 552.2216796875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000558.npy"}
|
|
{"epoch": 0.8193832599118943, "step": 559, "batch_size": 64, "mean": 241.2025146484375, "std": 260.89739990234375, "min": -327.40289306640625, "p10": -56.265205383300774, "median": 232.50496673583984, "p90": 570.8150939941407, "max": 904.4193725585938, "pos_frac": 0.875, "sample": [480.9198303222656, 401.7635498046875, 305.2044677734375, 488.56610107421875, 51.03876495361328, 40.76158142089844, 103.83399963378906, 87.56581115722656, 253.4123077392578, 904.4193725585938, 440.9444580078125, 355.97027587890625, 303.091552734375, 270.9176330566406, 84.97390747070312, 498.0828552246094, 655.9105224609375, 542.1808471679688, -247.66598510742188, -201.8134765625, 489.2026062011719, 134.8343048095703, 39.908172607421875, 159.02159118652344, 152.7698516845703, 409.78143310546875, -58.51085662841797, 16.271728515625, 438.52703857421875, 177.93438720703125, 57.56330108642578, 40.544986724853516, 62.354591369628906, 97.64118194580078, 20.227367401123047, 328.36614990234375, -178.79354858398438, -51.025352478027344, 265.55206298828125, 192.0372314453125, 629.0420532226562, 228.2491912841797, 274.6627197265625, 53.613624572753906, -327.40289306640625, 322.07464599609375, 115.07437133789062, 185.03506469726562, 583.0869140625, 104.6093521118164, 301.08343505859375, 432.61077880859375, 291.83758544921875, 433.1005859375, 717.2755126953125, 825.51025390625, -240.21092224121094, 638.7738037109375, 350.4696350097656, 236.7607421875, 206.91238403320312, 135.3059539794922, -204.7583465576172, 529.958251953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000559.npy"}
|
|
{"epoch": 0.8208516886930984, "step": 560, "batch_size": 64, "mean": 178.22430419921875, "std": 192.11102294921875, "min": -264.50726318359375, "p10": -81.44376220703124, "median": 160.00439453125, "p90": 457.83251953125006, "max": 623.7275390625, "pos_frac": 0.859375, "sample": [323.1044921875, 9.274711608886719, 48.49712371826172, 405.6034851074219, 46.432899475097656, 348.3877868652344, 91.12288665771484, -146.22003173828125, 137.02960205078125, 217.73623657226562, 125.4775390625, 98.73867797851562, 265.37384033203125, 324.073974609375, 534.2930908203125, 22.103614807128906, 516.859130859375, 450.03271484375, 207.40365600585938, 461.17529296875, 218.27035522460938, 158.834716796875, 178.65228271484375, 504.7525634765625, 154.9069366455078, -113.04329681396484, 187.13034057617188, 267.8563232421875, 228.4597930908203, -100.13543701171875, 429.90472412109375, 14.977401733398438, 27.273714065551758, 218.1820068359375, 159.9845733642578, -264.50726318359375, 62.21845245361328, 623.7275390625, 175.91256713867188, 471.29534912109375, 160.0242156982422, -85.76959228515625, 21.17487335205078, 29.322540283203125, 333.89068603515625, 325.98297119140625, 109.00825500488281, 40.504310607910156, -14.439226150512695, 533.867431640625, 301.63983154296875, 119.85287475585938, 275.2442932128906, 68.9172134399414, 200.64927673339844, 75.25333404541016, -71.35015869140625, 350.85345458984375, 307.69390869140625, 308.70782470703125, -137.2857208251953, 60.220008850097656, -118.6760025024414, 119.91412353515625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000560.npy"}
|
|
{"epoch": 0.8223201174743024, "step": 561, "batch_size": 64, "mean": 180.79673767089844, "std": 228.63671875, "min": -535.4813232421875, "p10": -49.053321838378906, "median": 156.4789276123047, "p90": 458.1491394042971, "max": 868.0372314453125, "pos_frac": 0.796875, "sample": [387.22003173828125, 276.41485595703125, -27.987525939941406, -48.52759552001953, -316.5172119140625, 304.6021423339844, 251.81304931640625, -50.45808410644531, 367.0660400390625, 484.42999267578125, -0.5799598693847656, 356.0433349609375, 594.0069580078125, 169.38978576660156, 116.86109161376953, 118.24057006835938, 71.69461059570312, -9.883995056152344, 326.81494140625, 205.37754821777344, 651.5553588867188, 63.58719253540039, 273.5429992675781, 240.23455810546875, 70.94601440429688, 228.8182830810547, 529.3572998046875, 142.29718017578125, 107.27336883544922, 507.0461730957031, 321.85614013671875, 868.0372314453125, 3.9857330322265625, 219.40676879882812, 364.3004150390625, -9.103157043457031, 1.4827690124511719, 125.45777893066406, 20.990280151367188, 349.58831787109375, 111.43512725830078, -71.79737854003906, -49.27863311767578, 396.8271484375, 145.28347778320312, -179.65261840820312, 71.9407958984375, 542.6450805664062, 360.3946838378906, 339.0680236816406, -535.4813232421875, 381.38262939453125, 198.16688537597656, 97.74518585205078, 122.86906433105469, 69.44287109375, 258.4459533691406, 107.11949157714844, 195.0332489013672, 105.8187255859375, 167.67437744140625, -155.17044067382812, 262.3774108886719, -27.980201721191406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000561.npy"}
|
|
{"epoch": 0.8237885462555066, "step": 562, "batch_size": 64, "mean": 262.9320068359375, "std": 277.76165771484375, "min": -239.7957763671875, "p10": -15.610922813415518, "median": 227.29379272460938, "p90": 569.483935546875, "max": 1194.24169921875, "pos_frac": 0.859375, "sample": [393.5260925292969, 1194.24169921875, 388.8619689941406, 216.51882934570312, 121.86337280273438, 388.0813293457031, 197.82757568359375, 121.76036071777344, 301.1736755371094, 382.2628173828125, 415.9706726074219, 664.5556030273438, 104.49628448486328, 337.4049987792969, 530.4367065429688, -75.04020690917969, 161.45562744140625, 288.7197265625, 91.09811401367188, 341.4605407714844, 235.3640594482422, 260.2789306640625, 408.45941162109375, -59.642173767089844, -0.23859405517578125, 496.859375, 573.151611328125, -6.904388427734375, 760.3392333984375, 92.60687255859375, -109.18609619140625, -46.19123077392578, 18.88160514831543, 384.3463439941406, -196.8731689453125, 1188.9923095703125, 90.1962890625, 407.69720458984375, 315.6626892089844, 525.3819580078125, 212.91656494140625, 123.78242492675781, 428.5473937988281, 251.18093872070312, -239.7957763671875, 178.08322143554688, 37.51402282714844, 560.926025390625, 304.4189453125, 800.6643676757812, 56.149784088134766, 248.64625549316406, 336.1574401855469, 160.22715759277344, 252.23300170898438, 219.22352600097656, 34.02531051635742, 50.59314727783203, -19.342294692993164, 55.62246322631836, 115.46760559082031, 75.15577697753906, 676.6801147460938, 2.7139663696289062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000562.npy"}
|
|
{"epoch": 0.8252569750367107, "step": 563, "batch_size": 64, "mean": 185.759521484375, "std": 264.2390441894531, "min": -479.7585754394531, "p10": -132.88982696533202, "median": 199.61490631103516, "p90": 491.24861450195317, "max": 820.8643188476562, "pos_frac": 0.78125, "sample": [125.1167221069336, -433.9439392089844, -408.33465576171875, 233.9559326171875, 308.8922424316406, -142.87420654296875, 511.9029235839844, 159.64968872070312, 271.42547607421875, 325.8998718261719, 273.49725341796875, 205.180908203125, -76.38469696044922, 412.2912902832031, 445.6644287109375, 352.04388427734375, 77.33647155761719, 820.8643188476562, 350.37725830078125, 15.012161254882812, 404.0406799316406, -65.83136749267578, -99.35882568359375, -109.59294128417969, 188.7266387939453, 8.110279083251953, 342.6837158203125, 472.6376953125, 562.1458740234375, 106.81858825683594, 54.36442565917969, 78.88005828857422, -311.7669677734375, 321.9393615722656, -60.86476135253906, 297.3700866699219, -35.28341293334961, 48.534446716308594, 475.3886413574219, 295.30291748046875, 457.7288513183594, 448.5528869628906, 557.878173828125, 135.4771728515625, 401.1619567871094, 187.53057861328125, -479.7585754394531, 654.7671508789062, -172.5224609375, 498.0457458496094, -107.11383056640625, 541.3507690429688, 58.41394805908203, 79.05784606933594, 278.90478515625, -156.18161010742188, 380.6536560058594, 14.400623321533203, 138.9198455810547, 268.2421569824219, 292.0699462890625, 168.8409423828125, 246.32107543945312, 194.0489044189453], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000563.npy"}
|
|
{"epoch": 0.8267254038179148, "step": 564, "batch_size": 64, "mean": 232.22918701171875, "std": 196.8150634765625, "min": -262.03564453125, "p10": 11.59458274841309, "median": 236.46741485595703, "p90": 506.4056549072267, "max": 670.447509765625, "pos_frac": 0.9375, "sample": [236.9744110107422, 235.96041870117188, 517.5679931640625, 15.804428100585938, 480.3601989746094, 534.7698364257812, 429.588623046875, 51.66136932373047, 211.24319458007812, 54.855743408203125, 227.3817138671875, 303.56781005859375, 177.35641479492188, 251.17428588867188, 360.77740478515625, 27.361099243164062, 57.19596481323242, 219.99996948242188, 420.0377197265625, 251.0497283935547, 237.62974548339844, 125.35057067871094, 309.1448974609375, -254.87347412109375, 76.86894226074219, 246.77737426757812, -114.423828125, -0.44931793212890625, 16.737457275390625, -262.03564453125, 223.60595703125, 631.61669921875, 354.46343994140625, 0.38404083251953125, 153.82833862304688, 262.56396484375, 193.44471740722656, 181.27587890625, 143.23004150390625, 9.790363311767578, 434.67645263671875, 670.447509765625, 70.468017578125, 158.8018798828125, 403.83111572265625, 265.2752380371094, 612.5426635742188, 327.18060302734375, 570.117431640625, 313.820556640625, 266.504150390625, 307.0860900878906, 95.87305450439453, 74.64187622070312, 237.05897521972656, 147.5635223388672, 253.9408416748047, 454.8917541503906, 314.5917053222656, 153.07884216308594, 648.0286865234375, 1.45806884765625, 283.1474609375, 198.0230712890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000564.npy"}
|
|
{"epoch": 0.8281938325991189, "step": 565, "batch_size": 64, "mean": 239.86215209960938, "std": 317.02044677734375, "min": -291.52349853515625, "p10": -150.07431030273438, "median": 187.50580596923828, "p90": 690.3908203125002, "max": 1130.71044921875, "pos_frac": 0.75, "sample": [-154.71827697753906, 920.9400024414062, 719.7818603515625, -291.52349853515625, 161.4122314453125, 400.1491394042969, 659.6326904296875, -8.603713989257812, 22.145797729492188, -44.32609558105469, 408.6128845214844, 449.50189208984375, -43.797027587890625, 51.841339111328125, -30.675846099853516, -185.18565368652344, -139.23838806152344, 36.21038818359375, -60.04638671875, 347.2607421875, 397.93701171875, 221.98800659179688, 354.2379455566406, 778.8344116210938, -56.81822204589844, 155.8594970703125, 188.19747924804688, 139.16162109375, -275.0843200683594, 547.3112182617188, 284.34771728515625, 270.65533447265625, 347.65777587890625, 426.715087890625, 38.16221618652344, -1.1309967041015625, 601.64892578125, 337.6709289550781, -166.48361206054688, 73.57343292236328, -30.526473999023438, 302.08056640625, 396.02392578125, 882.9180908203125, 214.014404296875, 1130.71044921875, 136.12173461914062, 914.79833984375, 469.3923645019531, -218.39239501953125, 119.58099365234375, 17.878265380859375, 517.080078125, 356.06622314453125, 703.5728759765625, 216.04510498046875, 427.8101806640625, 26.686145782470703, 186.8141326904297, 75.59668731689453, 86.76809692382812, 53.90546798706055, -158.05067443847656, 640.4680786132812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000565.npy"}
|
|
{"epoch": 0.8296622613803231, "step": 566, "batch_size": 64, "mean": 242.91168212890625, "std": 312.4317321777344, "min": -420.94598388671875, "p10": -138.33148498535155, "median": 272.7509002685547, "p90": 668.6410034179688, "max": 930.21240234375, "pos_frac": 0.71875, "sample": [-70.37380981445312, -194.38299560546875, 341.93829345703125, 641.7396240234375, -63.58197021484375, -176.19967651367188, 497.0861511230469, 214.21463012695312, -275.7671813964844, 384.49609375, 329.8115539550781, -100.83586120605469, -10.915618896484375, 242.3914794921875, 324.32470703125, 271.65911865234375, 438.960693359375, 63.58864212036133, 13.121707916259766, -420.94598388671875, 480.1398620605469, 464.3206787109375, 255.20130920410156, -228.182373046875, 376.9269714355469, -88.05461883544922, 272.22906494140625, 583.5296630859375, 25.29136848449707, 680.170166015625, 435.6739807128906, 930.21240234375, 536.6817016601562, 66.33528137207031, 727.8548583984375, 21.417884826660156, -75.207763671875, 794.1724853515625, 319.4775695800781, -110.83399963378906, -33.771297454833984, 385.69122314453125, 110.65990447998047, 593.4424438476562, 501.758056640625, 698.8737182617188, 740.876708984375, 457.9750671386719, -100.11576843261719, -30.12708282470703, -140.07550048828125, 689.2595825195312, 273.2727355957031, 569.5626831054688, 8.729684829711914, -183.96145629882812, 470.0439147949219, 445.89178466796875, 491.117431640625, -134.26211547851562, 2.9626541137695312, 20.73974609375, 338.2313232421875, 451.8855895996094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000566.npy"}
|
|
{"epoch": 0.8311306901615272, "step": 567, "batch_size": 64, "mean": 223.38568115234375, "std": 245.63865661621094, "min": -413.3139343261719, "p10": -29.890642547607417, "median": 210.5444107055664, "p90": 575.9289794921875, "max": 784.2719116210938, "pos_frac": 0.859375, "sample": [-25.13721466064453, 149.5088348388672, 621.7446899414062, 454.5907287597656, 289.286376953125, 375.0502014160156, 70.74915313720703, 126.41460418701172, 628.1952514648438, 227.9747314453125, 515.3172607421875, -34.87010955810547, 784.2719116210938, 320.358154296875, 128.23123168945312, 194.321044921875, 565.2742309570312, 439.9851989746094, 426.2436218261719, 57.4073371887207, 383.7667541503906, 207.35580444335938, 11.154279708862305, 227.8324737548828, 211.34774780273438, 271.9769287109375, 166.25267028808594, 624.317626953125, 12.859052658081055, 400.9010925292969, 11.329946517944336, 463.90106201171875, 219.6690216064453, -413.3139343261719, 222.46405029296875, 22.506175994873047, 60.34502029418945, 378.6860656738281, 723.843017578125, 611.1848754882812, -31.927825927734375, -209.7744903564453, 72.345458984375, -24.115699768066406, 50.92109680175781, 580.4953002929688, 269.11016845703125, 128.6961669921875, 41.20604705810547, 125.63398742675781, -163.77955627441406, 534.3421630859375, 10.976768493652344, 391.3624267578125, -123.73530578613281, 218.34048461914062, 62.83063507080078, 411.668701171875, 1.3947372436523438, 150.17633056640625, 209.74107360839844, 329.41192626953125, -218.8473358154297, 346.9134521484375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000567.npy"}
|
|
{"epoch": 0.8325991189427313, "step": 568, "batch_size": 64, "mean": 264.66583251953125, "std": 253.11590576171875, "min": -452.82989501953125, "p10": -30.943200302123998, "median": 236.85745239257812, "p90": 589.131805419922, "max": 952.540283203125, "pos_frac": 0.875, "sample": [292.3620910644531, -67.05744934082031, 603.8154907226562, 321.4984130859375, 325.57696533203125, 312.0944519042969, 341.18353271484375, 401.6951599121094, 476.564208984375, -452.82989501953125, 70.83470153808594, -141.2115020751953, 426.4550476074219, 736.9630126953125, 83.36634063720703, 177.39527893066406, 183.42324829101562, 22.39236831665039, 150.68470764160156, 149.30039978027344, 235.75100708007812, 235.36953735351562, 670.2515869140625, 552.9441528320312, 251.93881225585938, 237.53567504882812, 207.11212158203125, 952.540283203125, 53.677406311035156, 12.194049835205078, 315.4449768066406, 356.7344970703125, 376.212158203125, 464.7706298828125, 382.422119140625, 369.62518310546875, 99.72947692871094, 260.42645263671875, 127.19523620605469, 520.6552124023438, 797.2363891601562, -41.442867279052734, 213.42225646972656, 435.76873779296875, 309.222412109375, 145.61566162109375, -106.44564819335938, -59.47743606567383, 157.39523315429688, 652.5908813476562, 71.87694549560547, 721.27685546875, 231.256591796875, -252.92483520507812, 185.8319091796875, 236.17922973632812, 283.0591125488281, 359.75579833984375, 232.3406982421875, 545.6031494140625, 104.9792709350586, 554.869873046875, -6.443977355957031, 70.03141784667969], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000568.npy"}
|
|
{"epoch": 0.8340675477239354, "step": 569, "batch_size": 64, "mean": 195.7487030029297, "std": 265.715576171875, "min": -506.18865966796875, "p10": -72.1861171722412, "median": 163.7097625732422, "p90": 527.1037414550782, "max": 1075.52001953125, "pos_frac": 0.8125, "sample": [-258.087890625, 234.97561645507812, 1075.52001953125, 161.51211547851562, 496.21710205078125, -39.67292785644531, 264.7344970703125, 53.339630126953125, -232.953369140625, 36.01469039916992, 469.25128173828125, 623.09375, 165.90740966796875, 374.998779296875, 338.0931701660156, -52.230987548828125, 132.7990264892578, 511.70233154296875, 92.56169128417969, -3.8590030670166016, -506.18865966796875, 547.4718017578125, 98.22830200195312, -54.247562408447266, 31.563325881958008, 443.4124755859375, 562.6483154296875, 17.051742553710938, 46.338165283203125, 103.81549072265625, 301.08099365234375, 510.1321716308594, 401.2119445800781, 20.830673217773438, 533.704345703125, 288.9160461425781, 59.941925048828125, 225.59474182128906, 412.2664794921875, -185.68359375, 214.89431762695312, 30.274028778076172, -79.87406921386719, 481.41986083984375, 149.0543670654297, 131.0279541015625, 347.4449768066406, 180.55567932128906, -140.94888305664062, 59.10606384277344, -24.162063598632812, 597.0386352539062, 34.99901580810547, 170.20687866210938, 136.03990173339844, -288.8355712890625, 126.53262329101562, 201.7417755126953, 116.88218688964844, 305.005615234375, 183.43380737304688, 231.77706909179688, 347.5186767578125, 714.7789306640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000569.npy"}
|
|
{"epoch": 0.8355359765051396, "step": 570, "batch_size": 64, "mean": 238.44491577148438, "std": 240.86587524414062, "min": -204.64529418945312, "p10": -68.52125473022461, "median": 194.59124755859375, "p90": 641.1330627441406, "max": 705.11376953125, "pos_frac": 0.8125, "sample": [111.84886169433594, 636.0457153320312, 288.9537658691406, 682.755615234375, 705.11376953125, 111.03179931640625, 381.0232238769531, 620.2738647460938, 657.58642578125, 274.58905029296875, 182.5885009765625, 452.2163391113281, 158.34603881835938, 344.7376708984375, -25.172555923461914, 249.82325744628906, 698.7201538085938, 267.595947265625, 392.1672058105469, -105.70938110351562, 199.84963989257812, -132.07489013671875, -18.969240188598633, -76.71260833740234, 92.8724594116211, 643.3133544921875, -13.270709991455078, 172.49224853515625, 198.37841796875, 294.64569091796875, 160.30636596679688, 277.37701416015625, -87.66405487060547, 59.9066162109375, 167.76373291015625, 406.32452392578125, 159.11666870117188, 349.68218994140625, 29.183128356933594, 652.9842529296875, 161.141357421875, -66.1277847290039, 37.071014404296875, 258.19708251953125, 428.751220703125, 266.53631591796875, 118.33491516113281, 441.3470458984375, 466.5161437988281, 551.3668212890625, -56.08095169067383, 116.43917846679688, 554.0391235351562, 313.24371337890625, 190.8040771484375, -204.64529418945312, 22.117393493652344, 204.99964904785156, 167.2212371826172, -69.54702758789062, -146.85174560546875, 660.43212890625, 143.69937133789062, 81.42893981933594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000570.npy"}
|
|
{"epoch": 0.8370044052863436, "step": 571, "batch_size": 64, "mean": 258.15692138671875, "std": 294.2375183105469, "min": -493.62799072265625, "p10": -89.53960876464843, "median": 257.68990325927734, "p90": 606.6851989746094, "max": 998.9759521484375, "pos_frac": 0.84375, "sample": [543.386962890625, 372.61541748046875, 123.17796325683594, 297.7232360839844, 156.86026000976562, 172.50686645507812, 58.85111618041992, 519.6831665039062, 317.7318420410156, 661.4961547851562, 998.9759521484375, 52.8397102355957, 74.05536651611328, 470.6927185058594, 73.65859985351562, 246.9110565185547, 410.1124267578125, 563.395751953125, 197.06948852539062, 532.1450805664062, 763.3780517578125, 759.685546875, 607.705810546875, 290.63702392578125, -239.85400390625, 33.68975830078125, 391.0394287109375, 186.6464080810547, 317.3897399902344, 515.1626586914062, 175.84092712402344, 30.32818603515625, 313.3280029296875, 870.4794311523438, 521.321044921875, 181.0723876953125, -163.210205078125, 107.49209594726562, 604.3037719726562, 430.0627136230469, 336.4542236328125, -55.3562126159668, 616.7822265625, 189.98744201660156, -18.706546783447266, 201.22084045410156, 571.0550537109375, 272.6829833984375, 306.9386901855469, 122.59104919433594, -206.9060821533203, -417.5084228515625, 268.46875, 185.750244140625, -493.62799072265625, 309.46795654296875, 150.97254943847656, 356.6580505371094, -328.2488098144531, 225.77085876464844, 102.0887451171875, 460.5645751953125, -83.17990112304688, -92.26519775390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000571.npy"}
|
|
{"epoch": 0.8384728340675477, "step": 572, "batch_size": 64, "mean": 235.5412139892578, "std": 326.050048828125, "min": -443.9429626464844, "p10": -96.38732376098629, "median": 163.8994140625, "p90": 721.0860778808594, "max": 1134.2864990234375, "pos_frac": 0.78125, "sample": [-176.28179931640625, -320.8265380859375, -16.10748291015625, 397.9117431640625, 901.1672973632812, 188.52227783203125, 16.943023681640625, 62.730628967285156, 52.53158950805664, -61.92389678955078, 95.8675765991211, 309.4789123535156, 522.0823974609375, 131.3325653076172, 385.8995361328125, 404.4661560058594, 545.4517822265625, 704.8584594726562, 406.61187744140625, 357.2405700683594, 85.35035705566406, 67.94367980957031, 105.4593505859375, 383.9695739746094, 29.94292449951172, 24.579198837280273, 92.08856201171875, -111.15736389160156, -201.62680053710938, 461.46636962890625, 254.474365234375, -42.636878967285156, 250.55642700195312, 235.72537231445312, 91.36233520507812, 1134.2864990234375, 107.1422348022461, 216.71218872070312, -56.254478454589844, 17.416641235351562, 581.8175048828125, 102.80130004882812, -151.951171875, 386.5121154785156, -16.070140838623047, -443.9429626464844, 1062.040283203125, 220.6448516845703, 200.91722106933594, -8.527076721191406, 603.9513549804688, 139.27655029296875, 330.8711242675781, 883.7435913085938, 46.43873977661133, 266.0140686035156, 412.13336181640625, 745.9608764648438, -28.0133056640625, -222.29315185546875, 246.85716247558594, 728.040771484375, 875.4058227539062, 57.25233459472656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000572.npy"}
|
|
{"epoch": 0.8399412628487518, "step": 573, "batch_size": 64, "mean": 230.921142578125, "std": 258.00311279296875, "min": -392.78167724609375, "p10": -27.80372467041014, "median": 226.39759063720703, "p90": 603.2265869140626, "max": 996.4599609375, "pos_frac": 0.84375, "sample": [312.7055969238281, -12.669355392456055, 76.27574920654297, 107.5635986328125, 25.946701049804688, 774.0609130859375, 199.52523803710938, 9.450813293457031, 289.41156005859375, 337.22296142578125, 266.2843017578125, 166.88600158691406, 12.500120162963867, 996.4599609375, 241.56369018554688, 248.7146453857422, 805.7066040039062, -392.78167724609375, 118.22718048095703, -13.037651062011719, -213.19293212890625, 747.0498657226562, 148.27093505859375, 243.892578125, 22.02948760986328, 240.76321411132812, 555.8970336914062, 623.5106811523438, 225.41598510742188, 308.28436279296875, -208.71275329589844, 163.4459686279297, 241.4768829345703, 145.1006622314453, 663.9219970703125, -200.98928833007812, 304.3565979003906, -83.55001831054688, 183.5841064453125, -34.132041931152344, 728.244384765625, -10.322654724121094, 308.26287841796875, 23.80607032775879, 176.05792236328125, 50.13951110839844, 269.90545654296875, 445.2023620605469, 376.4744873046875, -57.97735595703125, 346.63623046875, 146.96603393554688, 171.87020874023438, 323.70526123046875, 246.8355255126953, 111.94596099853516, 240.58114624023438, 362.8935852050781, 407.9464111328125, 227.3791961669922, 162.2705078125, 460.3788146972656, 505.4992980957031, 107.81150817871094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000573.npy"}
|
|
{"epoch": 0.8414096916299559, "step": 574, "batch_size": 64, "mean": 263.2662658691406, "std": 268.7518310546875, "min": -359.83721923828125, "p10": -47.268352890014626, "median": 235.1339874267578, "p90": 673.5486572265626, "max": 907.0743408203125, "pos_frac": 0.828125, "sample": [43.76300811767578, 232.94651794433594, -101.15841674804688, 253.85137939453125, 377.76361083984375, 107.04441833496094, -13.051332473754883, 907.0743408203125, 581.1105346679688, 435.2220458984375, 104.00359344482422, 674.798095703125, 693.8482055664062, 495.61090087890625, 147.69976806640625, 276.2215576171875, 77.96483612060547, 293.4600830078125, -277.40997314453125, 745.9012451171875, 371.093505859375, 413.07177734375, 115.06483459472656, 514.9124145507812, 199.77688598632812, 530.82666015625, 207.68853759765625, -176.88900756835938, -55.70698928833008, -3.8082618713378906, 129.67239379882812, -56.28395462036133, 240.37222290039062, 386.37664794921875, 87.80144500732422, 483.7581481933594, -165.95431518554688, 703.725830078125, 177.9210968017578, 419.0898742675781, 606.6278076171875, -27.578201293945312, 544.598876953125, 382.7274475097656, -0.241180419921875, 122.24093627929688, 283.45068359375, 165.2686004638672, -359.83721923828125, 6.864349365234375, 209.3413848876953, 94.8919906616211, 70.01268005371094, 670.63330078125, 323.4098205566406, 344.3434143066406, 228.47665405273438, 701.3975830078125, 134.5341339111328, 343.2353210449219, 237.3214569091797, 770.5742797851562, 278.2828674316406, 139.2906036376953], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000574.npy"}
|
|
{"epoch": 0.8428781204111601, "step": 575, "batch_size": 64, "mean": 219.3174285888672, "std": 318.01397705078125, "min": -555.8516845703125, "p10": -202.82124633789056, "median": 239.98040008544922, "p90": 655.4403686523439, "max": 791.9276733398438, "pos_frac": 0.78125, "sample": [506.8682556152344, 240.4240264892578, 405.05029296875, 388.86187744140625, 356.0992431640625, -229.82571411132812, 271.66937255859375, 295.0228271484375, 228.543701171875, 84.05171203613281, 145.14918518066406, 37.658836364746094, -139.81082153320312, 416.64471435546875, 286.14288330078125, 37.386924743652344, -56.750274658203125, 679.7752685546875, 59.54142379760742, 239.53677368164062, 323.4819641113281, 280.748046875, 588.3055419921875, 99.62739562988281, 163.0703125, 27.399436950683594, -555.8516845703125, 243.58914184570312, -247.9556884765625, 331.87322998046875, -134.60897827148438, 786.4054565429688, 382.1593933105469, 736.3474731445312, 33.50617218017578, -9.515853881835938, 99.57318878173828, -67.9299545288086, 626.542724609375, 586.2598876953125, -12.899658203125, -29.39958953857422, -329.2432556152344, 667.8250732421875, 784.7333374023438, -479.95361328125, 395.1016845703125, 396.82086181640625, 467.0322265625, 763.72265625, 92.91292572021484, 516.6851806640625, 395.2209167480469, 26.12109375, 505.855712890625, 224.72256469726562, 257.6890869140625, -418.65216064453125, 791.9276733398438, 2.746379852294922, -316.3493957519531, 151.78628540039062, 141.9315643310547, 494.9099426269531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000575.npy"}
|
|
{"epoch": 0.8443465491923642, "step": 576, "batch_size": 64, "mean": 208.48135375976562, "std": 292.9903869628906, "min": -527.3912353515625, "p10": -159.48949279785157, "median": 207.28125, "p90": 633.5790405273438, "max": 811.4396362304688, "pos_frac": 0.75, "sample": [331.66015625, 711.8751220703125, -283.362548828125, 9.941471099853516, -102.74815368652344, 137.84742736816406, -101.54930114746094, 673.9440307617188, 393.27239990234375, 49.61054229736328, 252.87242126464844, 627.50146484375, 167.4932861328125, 130.4983673095703, 379.5498962402344, 257.095703125, 636.1837158203125, -379.843505859375, 310.56463623046875, 287.91351318359375, 365.8164367675781, 357.88873291015625, 538.1707763671875, 651.7318115234375, 118.10905456542969, 784.45263671875, -162.64170837402344, -179.33111572265625, 248.83792114257812, 515.4406127929688, -152.1343231201172, 197.64247131347656, 128.21743774414062, 29.056236267089844, 358.4044494628906, 572.1787109375, 811.4396362304688, 191.26193237304688, -27.62250328063965, 386.532958984375, 53.36082458496094, 477.69610595703125, 347.4947814941406, 318.47601318359375, -119.05704498291016, -527.3912353515625, -29.98792266845703, 300.91876220703125, -202.925537109375, 216.92002868652344, 252.29214477539062, 691.6074829101562, 527.4352416992188, 455.34454345703125, -21.456958770751953, 11.88197135925293, 92.47049713134766, 107.45850372314453, 87.95536041259766, 89.69244384765625, -196.6468505859375, -62.97018051147461, -35.47186279296875, 283.93719482421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000576.npy"}
|
|
{"epoch": 0.8458149779735683, "step": 577, "batch_size": 64, "mean": 232.3063507080078, "std": 243.85682678222656, "min": -526.5806884765625, "p10": -19.580031585693348, "median": 219.21044158935547, "p90": 525.9484771728517, "max": 949.3372802734375, "pos_frac": 0.859375, "sample": [-526.5806884765625, 949.3372802734375, 423.2200012207031, 348.00604248046875, 66.82735443115234, 67.62803649902344, 258.0393981933594, 157.0758514404297, 84.32634735107422, -10.25091552734375, 540.852783203125, 233.52130126953125, 254.51576232910156, 404.11798095703125, 154.4267120361328, 454.183837890625, 552.69873046875, 373.41033935546875, -46.43183898925781, 396.0890197753906, 89.49613952636719, 192.73947143554688, -226.99136352539062, 154.0350341796875, 309.922119140625, 297.2390441894531, 134.5943603515625, 403.8750915527344, 301.15753173828125, 382.544189453125, 352.0399475097656, 144.10836791992188, 199.6536865234375, 293.19085693359375, 536.2417602539062, 50.035400390625, 319.23779296875, -66.56295776367188, 144.26283264160156, -359.36077880859375, 65.2236328125, 176.12130737304688, 367.79815673828125, 345.1207275390625, -23.578224182128906, 122.08558654785156, 102.83372497558594, 501.9308166503906, 401.642578125, 608.59716796875, 124.21658325195312, 82.60051727294922, 604.4075927734375, 77.74371337890625, 709.6527099609375, 344.0377197265625, 342.0403747558594, -2.3497314453125, -221.29747009277344, 417.58624267578125, 146.0974578857422, 424.6675109863281, 159.05770874023438, 204.8995819091797], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000577.npy"}
|
|
{"epoch": 0.8472834067547724, "step": 578, "batch_size": 64, "mean": 239.20761108398438, "std": 221.272216796875, "min": -205.80352783203125, "p10": -16.68137702941894, "median": 224.058349609375, "p90": 558.4211303710938, "max": 720.1430053710938, "pos_frac": 0.859375, "sample": [720.1430053710938, 156.08290100097656, 353.3645324707031, 199.24136352539062, 298.4601745605469, 224.91012573242188, 643.89892578125, 344.8649597167969, 117.27969360351562, 10.493114471435547, 54.75791931152344, 658.2317504882812, 260.4971923828125, 223.20657348632812, 75.883056640625, -201.69589233398438, 278.5291748046875, 61.04924774169922, 588.2890625, 416.6207275390625, 425.47845458984375, 360.14208984375, 413.6871337890625, -19.789199829101562, -132.79428100585938, 241.7709503173828, 427.0977783203125, 31.73578643798828, -9.429790496826172, 124.48737335205078, -5.9553680419921875, 94.62017822265625, 283.8387451171875, 195.49807739257812, 88.59770202636719, 470.75872802734375, 84.58312225341797, 62.39501190185547, -205.80352783203125, 101.97510528564453, 244.5543670654297, 239.28988647460938, 231.52294921875, 212.0937957763672, 423.11639404296875, 299.2537841796875, 94.75511932373047, 21.664827346801758, 459.9881591796875, 183.5504150390625, 153.13833618164062, 568.931640625, 201.60499572753906, 407.3682556152344, 405.1151123046875, 533.8966064453125, -50.420562744140625, 458.2784423828125, 677.6123657226562, 592.5460205078125, 214.1074981689453, -132.0989990234375, -79.92988586425781, 432.345947265625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000578.npy"}
|
|
{"epoch": 0.8487518355359766, "step": 579, "batch_size": 64, "mean": 195.37242126464844, "std": 245.53936767578125, "min": -467.89251708984375, "p10": -92.37208023071287, "median": 189.1178741455078, "p90": 450.8032989501953, "max": 800.3854370117188, "pos_frac": 0.8125, "sample": [63.597801208496094, -69.17395782470703, -296.92010498046875, 344.9872741699219, -23.920867919921875, 169.24081420898438, 431.1223449707031, 208.16473388671875, 299.9713439941406, 400.30267333984375, -29.47278594970703, 80.42497253417969, -467.89251708984375, 280.8538818359375, 226.82391357421875, 362.4051818847656, 233.9080047607422, 356.5587158203125, 766.1973876953125, 57.91789245605469, 64.75540924072266, 89.86693572998047, 183.78598022460938, 313.5781555175781, 180.3558807373047, 290.0230407714844, 333.2276611328125, 23.33954620361328, 296.7260437011719, 252.7400665283203, 189.2847900390625, 138.93431091308594, 429.56036376953125, 445.31756591796875, 800.3854370117188, 591.1962280273438, 221.8052215576172, -220.14036560058594, -23.04043960571289, 358.8080139160156, 689.67529296875, 65.17819213867188, 453.1543273925781, -230.8084716796875, 718.9466552734375, 82.04612731933594, 158.55906677246094, 17.246200561523438, -249.64317321777344, 140.5557403564453, 463.30743408203125, 188.95095825195312, 161.62022399902344, 225.751708984375, 302.2386474609375, -166.63485717773438, 301.6864929199219, 167.72637939453125, 91.04803466796875, -102.31413269042969, -43.58750915527344, 321.986328125, 214.623779296875, 176.91412353515625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000579.npy"}
|
|
{"epoch": 0.8502202643171806, "step": 580, "batch_size": 64, "mean": 270.815673828125, "std": 287.83526611328125, "min": -475.7402038574219, "p10": -18.644701385498042, "median": 238.75189971923828, "p90": 633.5441406250001, "max": 1191.49560546875, "pos_frac": 0.875, "sample": [112.44825744628906, 639.0013427734375, 49.94651794433594, 212.7946014404297, -34.610103607177734, 158.31597900390625, 196.7384490966797, 765.0927124023438, 256.0658874511719, 264.7406311035156, 363.95965576171875, 25.64013671875, 1191.49560546875, 205.21121215820312, 52.58565139770508, 538.356201171875, 179.21707153320312, 99.89329528808594, 392.9434814453125, 297.30511474609375, 233.99095153808594, 329.5910339355469, 507.6134948730469, 289.9456481933594, 299.1306457519531, 114.77181243896484, 329.7878723144531, 384.1599426269531, 389.964111328125, 105.94613647460938, 178.92465209960938, 186.07960510253906, 210.01730346679688, 62.57258605957031, 608.4638671875, 547.513427734375, 85.32206726074219, -311.9674377441406, 620.8106689453125, 445.1492614746094, 644.98046875, 37.136383056640625, -475.7402038574219, 400.921630859375, 63.676475524902344, 308.3447265625, -111.99276733398438, -94.32320404052734, 367.4180908203125, 1064.87646484375, 158.5136260986328, -20.13165283203125, 850.1464233398438, 143.44351196289062, 281.90435791015625, 185.6171875, -15.175148010253906, 243.51284790039062, 3.346780776977539, -36.77360916137695, 255.54385375976562, 741.3379516601562, 506.7261047363281, 243.96307373046875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000580.npy"}
|
|
{"epoch": 0.8516886930983847, "step": 581, "batch_size": 64, "mean": 193.22781372070312, "std": 243.687744140625, "min": -320.4969482421875, "p10": -88.19098281860346, "median": 184.91483306884766, "p90": 462.57442626953133, "max": 792.9628295898438, "pos_frac": 0.828125, "sample": [347.0155944824219, 79.64741516113281, 86.83596801757812, -27.115707397460938, 249.04974365234375, 160.58499145507812, -320.4969482421875, 239.13815307617188, 76.2335433959961, 1.6722373962402344, 97.49139404296875, 336.6767883300781, 348.1671142578125, 119.30413055419922, 411.6259765625, -276.9371032714844, 747.6018676757812, 338.50811767578125, 469.6947021484375, 792.9628295898438, -110.29905700683594, 414.1707458496094, 36.25628662109375, 414.86053466796875, 200.22264099121094, 445.6171569824219, 171.9185333251953, 87.24735260009766, 197.9111328125, 398.2938537597656, -155.24720764160156, 762.0003051757812, 1.3224754333496094, 402.09808349609375, 203.81182861328125, 234.01727294921875, 471.6402282714844, 56.03645324707031, 106.6932144165039, 350.6524353027344, 148.42031860351562, 49.24298095703125, 54.31256103515625, 200.47071838378906, 251.03631591796875, 113.40318298339844, 445.96044921875, 498.90771484375, 648.98486328125, 236.4461212158203, -285.54559326171875, -276.7708435058594, 28.28543472290039, -167.91049194335938, 370.30657958984375, 22.042953491210938, -28.074886322021484, -33.62721252441406, 166.975341796875, -36.60547637939453, 272.5396423339844, 301.48309326171875, 62.34241485595703, 357.0696105957031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000581.npy"}
|
|
{"epoch": 0.8531571218795888, "step": 582, "batch_size": 64, "mean": 268.527587890625, "std": 292.97393798828125, "min": -621.6290283203125, "p10": -28.355145645141594, "median": 256.09385681152344, "p90": 697.6870910644532, "max": 794.3663330078125, "pos_frac": 0.875, "sample": [-621.6290283203125, 188.3516387939453, 764.0327758789062, 410.24078369140625, 610.8436279296875, -20.622100830078125, 276.4753723144531, 415.932861328125, 307.29046630859375, 66.30673217773438, 157.2932586669922, 191.69647216796875, 10.569564819335938, 519.7374267578125, 319.68353271484375, 451.1191101074219, 43.62046813964844, 35.98603820800781, 225.44129943847656, 663.212890625, 332.0611572265625, 109.75770568847656, 449.2084655761719, -219.94216918945312, 105.68999481201172, 794.3663330078125, 67.78519439697266, 371.9681091308594, 745.0911254882812, 729.5950927734375, 235.71234130859375, 346.5388488769531, 560.0361938476562, 12.450859069824219, -251.6844482421875, 108.12618255615234, 11.662727355957031, -71.22648620605469, 295.9089660644531, 443.87359619140625, 90.78628540039062, -62.352813720703125, 700.889404296875, 690.2150268554688, 282.1634521484375, 713.0578002929688, 20.588401794433594, 227.6985626220703, 72.84251403808594, 129.37477111816406, 30.926563262939453, 665.4774169921875, 564.5654296875, 350.07415771484375, -31.669307708740234, -312.46356201171875, 328.175048828125, 99.2103500366211, 501.78570556640625, 443.2261657714844, 733.9895629882812, 129.1778564453125, 452.5113525390625, 172.95143127441406], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000582.npy"}
|
|
{"epoch": 0.8546255506607929, "step": 583, "batch_size": 64, "mean": 187.08334350585938, "std": 282.05462646484375, "min": -314.45782470703125, "p10": -233.87679138183591, "median": 158.47075653076172, "p90": 522.877313232422, "max": 979.6691284179688, "pos_frac": 0.796875, "sample": [187.33648681640625, 134.09173583984375, 251.89614868164062, 221.54774475097656, 50.34515380859375, 172.72889709472656, 590.38427734375, 215.0782012939453, 979.6691284179688, 100.80977630615234, 17.52569580078125, 57.56890869140625, -272.42291259765625, 191.5552520751953, 153.47412109375, -215.2945556640625, 533.5870971679688, 415.9190368652344, 253.51950073242188, 345.0552978515625, 305.7092590332031, 62.508209228515625, 76.90621948242188, -267.01519775390625, 56.65624237060547, -248.55780029296875, 287.64642333984375, 46.76483154296875, -254.93585205078125, -41.770931243896484, 91.40419006347656, 11.84521484375, 259.10400390625, 432.00750732421875, 336.9302978515625, 862.4114990234375, 244.94158935546875, 9.610481262207031, 492.3055725097656, 64.2804946899414, -6.72125244140625, 163.46739196777344, 290.96258544921875, -77.01193237304688, -98.123779296875, -255.85731506347656, 857.8139038085938, 354.63092041015625, -314.45782470703125, 489.13482666015625, 497.8878173828125, 634.5581665039062, -12.836967468261719, 316.9513244628906, 369.941162109375, -241.84060668945312, 92.75346374511719, 374.2611999511719, 131.6732177734375, 211.60055541992188, 743.7540893554688, 26.375885009765625, 93.50955963134766, 117.77957153320312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000583.npy"}
|
|
{"epoch": 0.856093979441997, "step": 584, "batch_size": 64, "mean": 288.3196105957031, "std": 242.595947265625, "min": -180.8719940185547, "p10": 36.570678710937514, "median": 233.66082763671875, "p90": 664.4008056640627, "max": 916.1005249023438, "pos_frac": 0.953125, "sample": [299.65374755859375, 213.6533966064453, 55.83695602416992, 287.2209167480469, 29.61871337890625, -180.8719940185547, 467.04852294921875, 369.5355224609375, 330.18780517578125, 54.58222579956055, 12.053955078125, 84.61400604248047, 77.95147705078125, 229.75491333007812, 345.8338317871094, 559.7049560546875, 286.876220703125, 704.3075561523438, 480.4527893066406, 82.23482513427734, 251.6280517578125, 181.08477783203125, 122.51469421386719, 173.75941467285156, 166.64404296875, -49.451416015625, 626.6906127929688, 118.2161636352539, 133.433837890625, 173.4368896484375, 213.7362823486328, 320.6758728027344, 121.88168334960938, 52.79193115234375, 422.5308532714844, 300.03509521484375, 113.83464050292969, 220.1332550048828, 82.92633056640625, 375.3111877441406, 387.19036865234375, 29.185672760009766, 216.49441528320312, 680.5623168945312, 304.5444030761719, 188.2657470703125, 472.1213684082031, 845.1339111328125, -140.01785278320312, 844.9088134765625, 624.71337890625, 57.39881896972656, 916.1005249023438, 782.2965698242188, 394.3130798339844, 459.9959411621094, 172.45465087890625, 496.62530517578125, 19.558509826660156, 237.56674194335938, 350.486083984375, 318.44354248046875, 726.0953979492188, 155.95346069335938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000584.npy"}
|
|
{"epoch": 0.8575624082232012, "step": 585, "batch_size": 64, "mean": 225.02957153320312, "std": 269.5071105957031, "min": -423.5154724121094, "p10": -96.58874816894529, "median": 205.37845611572266, "p90": 547.2297607421875, "max": 1051.3448486328125, "pos_frac": 0.75, "sample": [205.64039611816406, -2.4153594970703125, 428.6141052246094, 223.67897033691406, -126.35890197753906, 1051.3448486328125, 403.8512878417969, -117.1361083984375, 205.11651611328125, 206.9661865234375, 610.3869018554688, 50.83116149902344, 156.15274047851562, 342.59521484375, 161.3709716796875, 415.247314453125, 457.9127502441406, 440.34637451171875, 885.7208251953125, 292.7557067871094, 259.32293701171875, 550.936767578125, -15.008094787597656, 264.4547424316406, 366.11737060546875, 33.17369079589844, 202.28428649902344, 283.7730712890625, 456.8276062011719, 108.32623291015625, 2.9433746337890625, 382.3232421875, 280.4781494140625, -125.93585205078125, 593.3232421875, 125.62261199951172, 611.405517578125, -3.0732154846191406, -73.24919128417969, 392.9217834472656, 143.40219116210938, -129.20977783203125, 461.8236999511719, -46.578224182128906, 372.4862060546875, 180.5074920654297, 273.44940185546875, -66.13235473632812, 451.23468017578125, 10.984033584594727, -26.469083786010742, -423.5154724121094, -57.66011428833008, -242.65621948242188, -106.59141540527344, 717.8544921875, 406.77459716796875, 126.01310729980469, 409.1219787597656, -21.331207275390625, 115.87068176269531, 161.42385864257812, 162.91981506347656, 538.580078125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000585.npy"}
|
|
{"epoch": 0.8590308370044053, "step": 586, "batch_size": 64, "mean": 195.86416625976562, "std": 270.7097473144531, "min": -418.60491943359375, "p10": -96.39696731567382, "median": 150.94602966308594, "p90": 528.4643615722656, "max": 942.31982421875, "pos_frac": 0.796875, "sample": [699.904052734375, 156.0540771484375, 2.2467689514160156, 64.21392059326172, 320.03192138671875, 242.28598022460938, -31.381568908691406, 231.0302276611328, 52.57286834716797, 359.6712951660156, 15.823627471923828, 145.83798217773438, 339.2183837890625, 673.6233520507812, -52.980079650878906, 69.30223083496094, 442.6663513183594, -418.60491943359375, 50.084388732910156, 79.55464172363281, 130.14088439941406, -20.459325790405273, 93.18392181396484, 496.396484375, 529.7717895507812, -100.41996765136719, 447.73876953125, -247.11428833007812, 167.07379150390625, 4.973907470703125, 394.96917724609375, 1.0825576782226562, 539.3819580078125, -97.53730010986328, 127.89741516113281, 195.20523071289062, 456.3287353515625, -66.16886138916016, 137.3828125, 9.875249862670898, 556.40283203125, -93.73619079589844, 525.4136962890625, 348.04852294921875, -344.2876892089844, -322.9610290527344, 196.64041137695312, 400.68524169921875, 351.2165832519531, 86.29351043701172, -121.23680114746094, 27.200035095214844, 450.0601806640625, 942.31982421875, 517.325927734375, 218.74537658691406, 460.83563232421875, 685.2498779296875, 69.6578369140625, -75.9212417602539, 232.97900390625, 462.023681640625, 207.86187744140625, 113.63153076171875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000586.npy"}
|
|
{"epoch": 0.8604992657856094, "step": 587, "batch_size": 64, "mean": 228.1727294921875, "std": 262.16961669921875, "min": -548.162109375, "p10": -111.02851715087888, "median": 213.6801986694336, "p90": 587.7978942871094, "max": 892.0816650390625, "pos_frac": 0.8125, "sample": [189.19644165039062, 567.4297485351562, 339.57659912109375, 50.45088195800781, 367.41094970703125, 35.44994354248047, 330.13958740234375, -41.0356559753418, -4.6904144287109375, 421.4605712890625, -92.26742553710938, 229.66879272460938, 694.6522216796875, 135.2991943359375, 211.32351684570312, 306.9842224121094, 7.472145080566406, 216.03688049316406, 178.2742919921875, 327.14080810546875, -146.3716583251953, 318.6695251464844, 611.056396484375, 18.409582138061523, 194.34776306152344, 784.7173461914062, -131.93258666992188, 472.19293212890625, 157.19741821289062, -119.06898498535156, 226.60430908203125, 88.45081329345703, 225.82301330566406, -14.059600830078125, 446.36468505859375, 183.71543884277344, 107.69820404052734, -548.162109375, 182.45362854003906, 106.72357177734375, -192.837646484375, 649.35888671875, 96.646728515625, 539.8214111328125, -4.972944259643555, 717.4646606445312, 225.0409393310547, -153.05873107910156, 222.16079711914062, 196.87222290039062, 372.5724792480469, 892.0816650390625, 299.0965576171875, 445.36480712890625, 305.8462219238281, 431.9907531738281, 493.9598693847656, -124.6451644897461, 306.1380920410156, 33.27400588989258, 201.53924560546875, 72.6109619140625, 345.3992919921875, 596.527099609375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000587.npy"}
|
|
{"epoch": 0.8619676945668135, "step": 588, "batch_size": 64, "mean": 217.41366577148438, "std": 259.8033142089844, "min": -276.2165832519531, "p10": -67.16600418090819, "median": 189.5819549560547, "p90": 539.2708007812503, "max": 912.4559326171875, "pos_frac": 0.796875, "sample": [225.9441680908203, 294.68499755859375, -56.292030334472656, 60.6961784362793, 268.518310546875, 459.6930847167969, 423.27777099609375, 70.29324340820312, 455.978271484375, 562.885986328125, -27.07452392578125, -13.476005554199219, 61.18634033203125, 117.28820037841797, -120.18746948242188, -133.6259765625, 75.86056518554688, 15.682342529296875, 97.79434967041016, 130.35140991210938, 484.168701171875, -34.973785400390625, -207.7050018310547, 463.8416442871094, 218.48829650878906, 183.0998992919922, 782.35205078125, 37.133480072021484, -274.012939453125, 25.995079040527344, 188.358642578125, 61.16471862792969, 184.0336456298828, 569.0411987304688, 483.3460693359375, 201.6510467529297, 245.68421936035156, -42.45806121826172, 369.0979919433594, 294.9173583984375, 595.276123046875, 325.07177734375, 2.71661376953125, 431.41693115234375, 153.0928497314453, 64.88416290283203, 101.16275024414062, 721.4886474609375, -276.2165832519531, 360.739990234375, -8.2294921875, 912.4559326171875, 893.7155151367188, 349.9337158203125, 401.45806884765625, 343.725830078125, 259.977783203125, 238.59072875976562, -95.05137634277344, 117.14372253417969, 190.80526733398438, 235.26657104492188, 464.17242431640625, -71.82627868652344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000588.npy"}
|
|
{"epoch": 0.8634361233480177, "step": 589, "batch_size": 64, "mean": 287.77593994140625, "std": 289.5597229003906, "min": -284.6943359375, "p10": -37.27410125732421, "median": 258.83599853515625, "p90": 658.8710266113283, "max": 918.22265625, "pos_frac": 0.8125, "sample": [608.0330200195312, 286.35009765625, 260.80950927734375, 96.24302673339844, 350.75225830078125, -22.742816925048828, -33.63401794433594, 558.1142578125, 259.527587890625, -86.16259002685547, 29.190826416015625, -5.314483642578125, 189.635986328125, 258.1444091796875, 224.03082275390625, 918.22265625, 913.1636962890625, -14.927482604980469, 259.65985107421875, 80.41104125976562, 506.241455078125, -97.10052490234375, 110.70216369628906, 769.4268798828125, 134.91824340820312, 202.029052734375, 251.11318969726562, 265.4073791503906, 615.7243041992188, -16.958555221557617, 628.9989624023438, 548.6730346679688, 173.79818725585938, 454.8453674316406, 872.7451171875, -38.834136962890625, -120.1907958984375, 109.1731948852539, 553.364501953125, 90.89439392089844, 420.6187744140625, 18.999481201171875, 402.1550598144531, 388.5166015625, 335.1517639160156, 141.65011596679688, 671.67333984375, 192.2346649169922, 277.6014709472656, 886.0960083007812, 611.19775390625, 153.9612274169922, 440.56219482421875, -168.42755126953125, 216.4093475341797, -284.6943359375, 578.2922973632812, 811.3231201171875, 28.832231521606445, 225.256103515625, 306.9838562011719, 328.92041015625, -196.27220153808594, 486.13995361328125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000589.npy"}
|
|
{"epoch": 0.8649045521292217, "step": 590, "batch_size": 64, "mean": 261.2105712890625, "std": 266.9997863769531, "min": -435.638671875, "p10": -61.449471282958974, "median": 249.3218994140625, "p90": 591.5802368164065, "max": 963.741943359375, "pos_frac": 0.859375, "sample": [222.78062438964844, 873.397216796875, 609.8284912109375, 108.34148406982422, 244.8089599609375, -151.700927734375, 275.68927001953125, 410.4329833984375, 472.70159912109375, -234.44021606445312, 142.90716552734375, -125.832763671875, 375.18231201171875, 666.910400390625, 190.62738037109375, 87.69474792480469, -52.47888946533203, 52.942970275878906, 483.00958251953125, 152.123779296875, 124.39840698242188, -70.5995101928711, 11.0167236328125, 253.1300506591797, -65.29400634765625, 154.5164794921875, 528.2133178710938, 92.82134246826172, 911.9014892578125, -7.435598373413086, 225.85800170898438, 365.4237060546875, 127.05309295654297, 712.95654296875, 365.3961486816406, 325.4483947753906, 245.5137481689453, 377.1746520996094, 449.4490966796875, 271.5108642578125, 310.33197021484375, 36.03155517578125, -164.65699768066406, -435.638671875, 276.9149169921875, 54.16395568847656, 358.22900390625, 360.546142578125, 191.9602813720703, 220.852783203125, 417.46148681640625, 462.16229248046875, 205.5203857421875, 103.64212799072266, 112.45803833007812, 651.3010864257812, 549.0009765625, 963.741943359375, 338.3621520996094, 366.5771789550781, 524.6427001953125, 288.77191162109375, 27.391094207763672, 294.3286437988281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000590.npy"}
|
|
{"epoch": 0.8663729809104258, "step": 591, "batch_size": 64, "mean": 245.3155059814453, "std": 240.54800415039062, "min": -341.860107421875, "p10": -32.398414802551265, "median": 246.72650146484375, "p90": 501.7909210205078, "max": 904.106689453125, "pos_frac": 0.859375, "sample": [15.869392395019531, 904.106689453125, 486.4418640136719, 438.5687255859375, 657.75, 35.54414367675781, 195.21104431152344, -283.5142517089844, 336.943603515625, 689.5098876953125, 448.29779052734375, 180.84274291992188, 469.03369140625, 273.39459228515625, 137.0640106201172, 703.28662109375, 497.5749206542969, 381.0882873535156, 119.87126159667969, 200.68218994140625, 149.1685333251953, 503.5977783203125, -26.060869216918945, 93.79593658447266, 181.74520874023438, 157.60519409179688, -132.34901428222656, 280.5994567871094, 179.61358642578125, 360.12158203125, -190.39901733398438, 104.86102294921875, 309.91546630859375, -59.83586120605469, 428.867919921875, 47.78208923339844, 111.73591613769531, 78.281005859375, 331.7421875, 40.808204650878906, -341.860107421875, 485.51861572265625, 278.437255859375, 351.3326110839844, -2.8719711303710938, 540.796875, 411.6672668457031, -35.114505767822266, 360.1944274902344, 266.759033203125, 230.14678955078125, 328.87396240234375, 416.2060241699219, 99.23452758789062, 486.4259033203125, 263.30621337890625, 575.564453125, 413.3963317871094, -163.40713500976562, 373.8497009277344, 122.53605651855469, 63.571876525878906, 185.90687561035156, 150.55767822265625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000591.npy"}
|
|
{"epoch": 0.8678414096916299, "step": 592, "batch_size": 64, "mean": 224.35015869140625, "std": 325.40155029296875, "min": -661.6262817382812, "p10": -177.0004455566406, "median": 249.7310333251953, "p90": 569.8513427734375, "max": 949.7931518554688, "pos_frac": 0.78125, "sample": [101.95204162597656, 585.9234619140625, -105.14834594726562, 235.23748779296875, -293.2950439453125, 8.210891723632812, 358.199951171875, 611.6890258789062, 650.7771606445312, 390.5605163574219, -126.24766540527344, 949.7931518554688, 47.717994689941406, 557.4363403320312, -508.6743469238281, -20.94705581665039, 200.7486572265625, 269.60699462890625, 8.13739013671875, 860.185302734375, 132.63424682617188, 472.8684997558594, -281.5370178222656, 591.731201171875, -99.81809997558594, 82.50830078125, 163.25045776367188, 525.4605712890625, 496.51544189453125, 345.2950134277344, 32.869197845458984, -6.8149871826171875, 45.2825927734375, 375.4505310058594, 516.6551513671875, 103.33071899414062, -6.85137939453125, 331.9908752441406, -159.56863403320312, 383.99395751953125, 264.2245788574219, 186.34402465820312, 424.46795654296875, 538.076904296875, 215.08282470703125, -184.47122192382812, 432.3980407714844, -661.6262817382812, 492.7005310058594, 330.2178649902344, 469.33843994140625, 118.56123352050781, 537.5303955078125, 471.4670104980469, -259.3068542480469, 528.8515014648438, 469.96795654296875, 134.56297302246094, 575.1720581054688, 132.49322509765625, -624.9326782226562, 361.99169921875, 358.6544189453125, 219.53323364257812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000592.npy"}
|
|
{"epoch": 0.869309838472834, "step": 593, "batch_size": 64, "mean": 167.77935791015625, "std": 251.9904327392578, "min": -461.2398681640625, "p10": -116.02500381469726, "median": 142.1826171875, "p90": 464.7361297607422, "max": 876.8524169921875, "pos_frac": 0.75, "sample": [-12.753498077392578, 289.7443542480469, -116.84188079833984, -109.76898193359375, 2.6289443969726562, -114.11895751953125, 350.388671875, -173.6453399658203, 460.40582275390625, 502.8826599121094, 210.8599853515625, 465.0455322265625, 436.2933349609375, 483.0764465332031, 43.028465270996094, -59.77665710449219, 494.08612060546875, 397.94512939453125, 389.40869140625, 143.13099670410156, 15.045116424560547, 56.02044677734375, 119.14791107177734, 62.52631378173828, 286.6170349121094, 876.8524169921875, 288.3150329589844, 141.97946166992188, 240.2054901123047, -289.1908874511719, -30.6802921295166, 464.0141906738281, -102.85746002197266, 120.21595764160156, 315.6153564453125, 438.8890380859375, -59.366180419921875, 551.0, 4.678672790527344, 8.306995391845703, -461.2398681640625, 513.5778198242188, -245.63485717773438, -304.289794921875, 96.11264038085938, -0.66943359375, 32.095603942871094, 140.4996337890625, 407.8070373535156, 423.66156005859375, 223.2764129638672, 196.55136108398438, 327.1304626464844, 324.8898620605469, 382.2149963378906, 296.04034423828125, 142.38577270507812, 39.54698944091797, -209.82290649414062, 386.2455749511719, -3.971832275390625, 35.23229217529297, 25.26955795288086, 381.6145324707031], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000593.npy"}
|
|
{"epoch": 0.8707782672540382, "step": 594, "batch_size": 64, "mean": 231.80612182617188, "std": 289.9068908691406, "min": -448.410400390625, "p10": -145.13975906372067, "median": 224.44214630126953, "p90": 658.7178100585938, "max": 1039.40283203125, "pos_frac": 0.796875, "sample": [387.5720520019531, -40.741790771484375, 20.730573654174805, 4.515871047973633, -40.93125915527344, -245.00390625, 534.3970336914062, 228.21853637695312, 246.44766235351562, 189.909912109375, 165.86904907226562, 270.5980529785156, 754.3710327148438, 495.50628662109375, 133.25274658203125, 644.993896484375, 463.1873779296875, 220.66575622558594, 413.82537841796875, -93.77835083007812, -256.7755126953125, 250.53431701660156, -186.361572265625, 664.5994873046875, -1.4298820495605469, 364.9207458496094, -194.54324340820312, 322.9808044433594, 434.8076171875, 158.40322875976562, 164.29518127441406, 335.05731201171875, 329.5364685058594, 538.1675415039062, 353.31689453125, 283.2281494140625, 317.11968994140625, 13.616107940673828, 171.44607543945312, -448.410400390625, 143.68875122070312, 724.883544921875, 216.6331024169922, 147.54019165039062, 335.6482849121094, 668.3262939453125, 578.4209594726562, 325.44793701171875, 23.401824951171875, 300.43524169921875, -122.6687240600586, 26.617996215820312, 63.033939361572266, -264.78564453125, 368.81793212890625, -32.63179397583008, 369.04742431640625, 682.5894775390625, 730.0267944335938, 169.1840057373047, 1039.40283203125, -154.77020263671875, 64.17901611328125, 65.00753021240234], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000594.npy"}
|
|
{"epoch": 0.8722466960352423, "step": 595, "batch_size": 64, "mean": 221.48776245117188, "std": 258.86444091796875, "min": -169.41720581054688, "p10": -58.190211486816395, "median": 162.93838500976562, "p90": 545.2584960937501, "max": 1135.0946044921875, "pos_frac": 0.8125, "sample": [39.69804382324219, 324.6441345214844, 253.88623046875, 127.03089141845703, -34.046485900878906, 632.9697875976562, 137.00161743164062, 461.2479553222656, -42.34963607788086, 199.6749267578125, 474.8231201171875, 136.90794372558594, 149.61959838867188, 597.6300048828125, -139.3380584716797, 314.424560546875, 160.23016357421875, 1077.788818359375, 191.4522247314453, 14.72829818725586, -77.443115234375, -136.86151123046875, 260.7714538574219, 308.6222229003906, -100.20978546142578, -68.14889526367188, 238.04754638671875, 80.4327163696289, 290.84686279296875, 192.25912475585938, 89.10790252685547, 159.70248413085938, 107.55587768554688, 162.6221923828125, 556.2843017578125, 114.49842071533203, 163.25457763671875, 129.17445373535156, 502.1549072265625, 429.39105224609375, -39.26716613769531, -64.072021484375, 294.67132568359375, -169.41720581054688, 171.52725219726562, 165.9076690673828, 237.38548278808594, -44.46598815917969, 312.4971618652344, 438.7115173339844, 21.240440368652344, 633.1399536132812, 5.854724884033203, 61.46549987792969, 230.68722534179688, 155.9811553955078, 519.5316162109375, 71.54672241210938, -25.310302734375, 89.51852416992188, 588.4078369140625, 423.43341064453125, 481.06024169921875, 1135.0946044921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000595.npy"}
|
|
{"epoch": 0.8737151248164464, "step": 596, "batch_size": 64, "mean": 193.6279296875, "std": 269.7957763671875, "min": -571.0958251953125, "p10": -96.83414840698241, "median": 158.69281005859375, "p90": 567.2750549316407, "max": 896.7317504882812, "pos_frac": 0.796875, "sample": [-221.97235107421875, -28.0035400390625, 209.16824340820312, 31.819168090820312, 328.60662841796875, 485.1383361816406, 54.425804138183594, 391.69720458984375, 400.3537292480469, 174.53311157226562, 620.8311157226562, 107.07003784179688, -78.7220458984375, 336.76312255859375, 545.3742065429688, 576.6611328125, 205.4479217529297, 97.05595397949219, 420.73162841796875, 348.43017578125, -61.02512741088867, 456.416015625, 258.1029968261719, 173.8108673095703, -134.9822998046875, 578.8858642578125, 55.89997100830078, 230.61666870117188, 201.3497772216797, 81.38487243652344, -103.34028625488281, 119.86775970458984, -81.65316009521484, 719.55517578125, 94.2330322265625, 427.7352294921875, 660.521484375, 353.0057678222656, 131.46951293945312, -9.511283874511719, 228.15774536132812, 34.12316131591797, 133.2739715576172, 121.0653305053711, 76.0801773071289, 431.552490234375, 587.0762329101562, 143.5747528076172, 0.40981101989746094, 101.33641052246094, 472.7073059082031, 39.52924346923828, -571.0958251953125, 141.43182373046875, 199.47586059570312, -243.67263793945312, 36.66157531738281, -77.85493469238281, 353.70452880859375, 286.37109375, 896.7317504882812, -178.24545288085938, 408.9778747558594, -386.9376220703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000596.npy"}
|
|
{"epoch": 0.8751835535976505, "step": 597, "batch_size": 64, "mean": 258.4862060546875, "std": 280.88848876953125, "min": -333.20660400390625, "p10": -39.70517959594726, "median": 224.6937484741211, "p90": 547.6481384277344, "max": 1123.0653076171875, "pos_frac": 0.84375, "sample": [358.7165832519531, 293.6220703125, 289.8665771484375, 231.5620574951172, 217.825439453125, 870.7626953125, -8.43032455444336, 446.8156433105469, 359.81671142578125, 202.86839294433594, -30.19762420654297, -22.89581298828125, 20.47216033935547, 45.24892807006836, 376.10382080078125, 185.822021484375, 393.439453125, 163.64259338378906, 350.8108215332031, 701.5301513671875, 202.89222717285156, 320.37652587890625, 66.82583618164062, 546.3564453125, 83.35203552246094, 169.31918334960938, 22.993938446044922, 66.20884704589844, 74.07319641113281, -315.4306640625, -333.20660400390625, -71.4039306640625, 363.7896728515625, 132.0912322998047, 384.49072265625, 1029.103515625, 535.9773559570312, -43.77984619140625, -68.60527038574219, 147.16519165039062, 474.1319885253906, 469.7491455078125, 104.60977935791016, 355.9619140625, 74.1799545288086, 293.0180969238281, 299.1166076660156, 445.1405029296875, 37.85862731933594, 594.1007690429688, 151.45758056640625, -179.70526123046875, -133.648193359375, 1123.0653076171875, 758.7044677734375, 170.95436096191406, 361.0821838378906, 267.2083740234375, 536.0064697265625, 548.2017211914062, 60.17362976074219, 459.05712890625, 100.3337173461914, 412.3644714355469], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000597.npy"}
|
|
{"epoch": 0.8766519823788547, "step": 598, "batch_size": 64, "mean": 227.01986694335938, "std": 265.7610778808594, "min": -508.8982238769531, "p10": -73.76621360778805, "median": 229.87300872802734, "p90": 552.3398498535156, "max": 757.1102905273438, "pos_frac": 0.859375, "sample": [142.5913543701172, 568.237548828125, 151.2638702392578, 542.10009765625, -367.8826599121094, 380.1926574707031, 480.403076171875, 405.69378662109375, 131.39373779296875, 554.8717041015625, 72.97732543945312, -304.22344970703125, -508.8982238769531, 41.30445861816406, 507.69964599609375, 5.212362289428711, 546.4321899414062, 319.99127197265625, 757.1102905273438, 342.6522521972656, 149.167724609375, 509.27349853515625, 544.08984375, 153.3399658203125, 181.00746154785156, 344.5685729980469, 397.301513671875, 160.6340789794922, -417.64080810546875, 34.91614532470703, 277.1813659667969, 222.67288208007812, 237.07313537597656, -90.29788970947266, -139.4694366455078, 355.3042297363281, 49.48994827270508, 543.40771484375, 131.40086364746094, 257.1956787109375, 156.37283325195312, 153.52801513671875, -228.33334350585938, 404.4555358886719, -34.477684020996094, 586.796142578125, 113.05720520019531, 237.7760772705078, 120.37858581542969, 99.55243682861328, 364.2398376464844, 587.0626220703125, 555.3014526367188, 353.2364501953125, 394.37335205078125, 107.53284454345703, 295.78863525390625, 500.9222412109375, 339.83685302734375, 85.37216186523438, 28.005691528320312, 63.16643524169922, 610.77978515625, -35.19230270385742], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000598.npy"}
|
|
{"epoch": 0.8781204111600588, "step": 599, "batch_size": 64, "mean": 209.8470458984375, "std": 211.64028930664062, "min": -347.9669494628906, "p10": -77.44976882934569, "median": 223.93124389648438, "p90": 475.6362487792969, "max": 668.6626586914062, "pos_frac": 0.859375, "sample": [271.5143737792969, 284.3775634765625, 166.1893768310547, 173.0618896484375, 182.2049560546875, 422.48931884765625, 352.9661865234375, 117.00894165039062, 271.9701232910156, 319.19879150390625, 470.1204833984375, 62.45844268798828, 8.699302673339844, 174.08021545410156, 0.19765281677246094, 230.2754364013672, 380.6178283691406, 269.1380615234375, 267.0011291503906, 470.97589111328125, 119.6195068359375, 376.0575866699219, 343.4111022949219, 263.78741455078125, 253.3756866455078, -102.26660919189453, 128.86627197265625, 248.5021514892578, 190.8164825439453, -347.9669494628906, 501.24700927734375, 56.33174133300781, 530.6909790039062, 394.7447509765625, 555.7660522460938, 477.633544921875, 425.066162109375, -44.732940673828125, 609.5274047851562, 7.940834045410156, 242.263427734375, -109.387939453125, 542.1297607421875, 55.01264572143555, 271.1767578125, 212.44961547851562, 399.9528503417969, -136.09121704101562, 217.58705139160156, 196.57174682617188, 383.5468444824219, 49.49160385131836, -85.81682586669922, 12.419578552246094, -252.95535278320312, 152.77285766601562, 295.67156982421875, 83.55831909179688, 259.07000732421875, 668.6626586914062, -57.9266357421875, -176.60665893554688, 144.14146423339844, 179.55191040039062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000599.npy"}
|
|
{"epoch": 0.8795888399412628, "step": 600, "batch_size": 64, "mean": 224.59609985351562, "std": 280.3115234375, "min": -408.26629638671875, "p10": -79.26737670898437, "median": 181.49508666992188, "p90": 614.9307617187502, "max": 897.92626953125, "pos_frac": 0.828125, "sample": [86.33280181884766, 181.35682678222656, 406.76727294921875, -360.3917541503906, 493.2366638183594, 351.9237060546875, 58.936771392822266, 350.65924072265625, -71.3915023803711, 432.5506286621094, 124.23986053466797, 239.63674926757812, 573.4593505859375, 67.4263687133789, 195.07913208007812, 558.3719482421875, -109.83386993408203, 687.7085571289062, 37.79094696044922, 190.40838623046875, 422.2778015136719, 657.8818359375, 79.71903991699219, 159.15713500976562, 123.85633850097656, -42.49409484863281, 384.7426452636719, 791.93359375, 688.6959228515625, 179.07150268554688, 119.475341796875, 309.2489929199219, 542.673583984375, 181.6333465576172, 632.7042236328125, 257.70928955078125, -77.31684875488281, 213.00477600097656, 11.430242538452148, 428.16424560546875, 897.92626953125, 118.20140075683594, -80.10331726074219, 78.05098724365234, 753.5575561523438, 136.822021484375, 510.183837890625, 153.1613006591797, -245.7550506591797, 248.63638305664062, 203.5771026611328, 305.1005859375, 423.7747802734375, 67.99675750732422, 264.9476318359375, 559.0901489257812, -273.0518798828125, 117.07708740234375, 85.90514373779297, -248.373779296875, -408.26629638671875, 58.14436340332031, -50.66742706298828, 140.37791442871094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000600.npy"}
|
|
{"epoch": 0.8810572687224669, "step": 601, "batch_size": 64, "mean": 241.43040466308594, "std": 276.5715026855469, "min": -369.5078125, "p10": -56.930786514282225, "median": 222.59337615966797, "p90": 615.4133422851563, "max": 999.7489013671875, "pos_frac": 0.828125, "sample": [221.68426513671875, 408.03814697265625, 999.7489013671875, 484.5508117675781, -27.233905792236328, 609.9093017578125, 255.15536499023438, 223.16680908203125, -1.6563339233398438, 356.270263671875, -96.64620208740234, -318.2930908203125, 627.6170043945312, 297.56854248046875, 161.45912170410156, 370.5287780761719, 60.94977569580078, 339.8465881347656, 145.37765502929688, 860.9024047851562, 348.97540283203125, 417.4849853515625, 17.11944580078125, 89.06272888183594, 278.4057312011719, 222.0199432373047, 20.310758590698242, 259.57989501953125, 638.9103393554688, 171.2822265625, 212.84115600585938, 179.3501434326172, -364.0385437011719, 225.74688720703125, -176.86398315429688, 402.95263671875, 69.75794219970703, 216.85890197753906, 69.20516967773438, 216.55087280273438, 176.34048461914062, 374.80816650390625, 104.39176177978516, -267.5323791503906, 84.70622253417969, -369.5078125, 34.341522216796875, 839.31201171875, 455.92266845703125, -43.477073669433594, 430.5688171386719, 314.4299621582031, 87.43174743652344, 364.7081604003906, -53.28714370727539, 679.6227416992188, -58.492347717285156, 617.772216796875, 478.5619201660156, 409.43341064453125, 379.70458984375, 192.657470703125, 378.13067626953125, 346.5411376953125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000601.npy"}
|
|
{"epoch": 0.882525697503671, "step": 602, "batch_size": 64, "mean": 274.4678955078125, "std": 261.5825500488281, "min": -243.04190063476562, "p10": -18.222309112548817, "median": 201.21173095703125, "p90": 566.6651611328126, "max": 1005.3060913085938, "pos_frac": 0.875, "sample": [166.64129638671875, 188.84494018554688, -25.091022491455078, 144.8396453857422, 232.20053100585938, 151.98590087890625, 581.9557495117188, 390.7004699707031, 122.32417297363281, 148.95191955566406, 290.06695556640625, 141.01927185058594, 204.00228881835938, 324.0777282714844, 782.5856323242188, 101.73829650878906, 523.568603515625, 179.89248657226562, 322.53662109375, 521.1380615234375, 12.325576782226562, 195.7344970703125, -23.2325439453125, -243.04190063476562, 155.03395080566406, 343.20452880859375, 429.0591125488281, 308.0791931152344, -161.99761962890625, -74.05575561523438, 47.612449645996094, 429.9974060058594, 364.1201477050781, 105.49650573730469, 284.14105224609375, 530.9871215820312, 161.02310180664062, 391.880615234375, 87.44223022460938, -187.581298828125, 824.541259765625, 462.3500061035156, 391.05615234375, 286.322998046875, 486.61956787109375, 483.361328125, 1005.3060913085938, 58.99583435058594, 515.4271240234375, -83.45783996582031, 616.046875, 408.69622802734375, 140.88650512695312, 147.30935668945312, 805.8590087890625, 81.03103637695312, -6.531761169433594, 304.7116394042969, 188.11956787109375, 121.46987915039062, 509.32708740234375, 957.1438598632812, 198.42117309570312, 12.724014282226562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000602.npy"}
|
|
{"epoch": 0.8839941262848752, "step": 603, "batch_size": 64, "mean": 235.269287109375, "std": 282.40631103515625, "min": -312.207763671875, "p10": -104.95238876342772, "median": 190.01280975341797, "p90": 649.7604980468752, "max": 909.811767578125, "pos_frac": 0.796875, "sample": [-113.2457275390625, -55.70188903808594, 343.4354248046875, 35.397682189941406, 700.43701171875, -3.57684326171875, 469.5831298828125, 46.34233093261719, 304.4049072265625, 469.876220703125, 133.62295532226562, 253.07388305664062, -266.0491943359375, 423.185546875, 48.53947830200195, 169.68032836914062, 37.312896728515625, 558.0802001953125, 679.8796997070312, 191.70755004882812, 777.0326538085938, 416.36767578125, 290.10333251953125, 10.211292266845703, 14.257431030273438, 221.36233520507812, 452.44659423828125, 141.09437561035156, 63.31509780883789, 162.18206787109375, 579.0635986328125, -312.207763671875, 573.73193359375, 115.914306640625, 15.307487487792969, 372.9103088378906, 579.4823608398438, -164.50726318359375, 219.17910766601562, 251.3682861328125, 422.5346984863281, 103.47260284423828, -8.442611694335938, 45.019832611083984, -85.60126495361328, 261.8175354003906, 188.3180694580078, 142.67417907714844, -152.33322143554688, 237.1043701171875, -29.620534896850586, 529.3635864257812, 441.2498779296875, -138.44973754882812, 219.93048095703125, -169.79864501953125, 909.811767578125, 186.6402587890625, 824.1487426757812, 11.01544189453125, 693.0367431640625, 535.06787109375, 687.4793701171875, -1.8061485290527344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000603.npy"}
|
|
{"epoch": 0.8854625550660793, "step": 604, "batch_size": 64, "mean": 279.2680358886719, "std": 281.4589538574219, "min": -517.2169799804688, "p10": -24.186060523986807, "median": 277.1787414550781, "p90": 661.4248962402345, "max": 921.6649780273438, "pos_frac": 0.84375, "sample": [-28.354970932006836, 165.43719482421875, 16.27100372314453, -48.72447204589844, 453.43280029296875, 360.919677734375, 506.5333251953125, 510.4151916503906, 256.80645751953125, 354.01617431640625, -327.11016845703125, 638.7669677734375, 183.1506805419922, -98.32535552978516, 555.171630859375, 380.7520751953125, 395.50799560546875, 19.841541290283203, 287.9233093261719, 457.3682861328125, 186.48695373535156, -290.8402404785156, 116.21870422363281, 540.3606567382812, 77.06700897216797, -517.2169799804688, 719.1624755859375, -14.458602905273438, 190.69192504882812, 77.171630859375, -70.99654388427734, 266.4341735839844, 213.9145050048828, 447.6335144042969, 223.94668579101562, 746.573486328125, 457.94183349609375, 41.361427307128906, 190.3956756591797, 252.65579223632812, 474.2851867675781, 330.5971984863281, 313.8421325683594, 90.88768768310547, 499.0339660644531, 760.5814819335938, 125.82231140136719, 921.6649780273438, 305.3667907714844, 147.80079650878906, 504.05279541015625, 671.1354370117188, 465.4781494140625, 151.42071533203125, 716.8711547851562, 316.9092712402344, 34.25383758544922, -3.14703369140625, 198.45681762695312, 358.06805419921875, -3.400146484375, 363.0789794921875, 896.524658203125, 339.2651062011719], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000604.npy"}
|
|
{"epoch": 0.8869309838472834, "step": 605, "batch_size": 64, "mean": 232.27914428710938, "std": 293.14508056640625, "min": -466.6174621582031, "p10": -49.55627708435058, "median": 200.3551788330078, "p90": 564.7755798339845, "max": 1476.4161376953125, "pos_frac": 0.828125, "sample": [-20.575393676757812, -49.66155242919922, 8.142181396484375, 16.815500259399414, 253.30020141601562, 101.07893371582031, 157.6743927001953, 414.1075439453125, 510.983642578125, 125.96218872070312, 369.4042663574219, -15.551788330078125, 48.496429443359375, -94.74578094482422, 237.0384979248047, 219.55545043945312, 129.79747009277344, 706.1557006835938, 853.9754638671875, 13.851913452148438, 1476.4161376953125, 109.01370239257812, 211.72564697265625, 193.64443969726562, 242.6807098388672, 276.1886291503906, 78.607421875, 299.6572570800781, -2.422748565673828, 262.3146057128906, 85.96550750732422, -158.80348205566406, 487.0790710449219, -171.84140014648438, -93.69998168945312, 326.2169494628906, 704.1917114257812, 87.22123718261719, 161.60076904296875, 583.0418701171875, 74.74885559082031, 974.36376953125, -154.06504821777344, -466.6174621582031, 485.8876953125, 266.0450134277344, -49.31063461303711, 402.69024658203125, 288.92791748046875, 153.6794891357422, 229.84341430664062, 265.7456970214844, 346.4775085449219, 215.92062377929688, 148.54034423828125, 197.02679443359375, 128.15231323242188, 191.63046264648438, 293.1356201171875, 73.16804504394531, 533.6048583984375, 339.8485412597656, 203.68356323242188, 578.1344604492188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000605.npy"}
|
|
{"epoch": 0.8883994126284875, "step": 606, "batch_size": 64, "mean": 249.30398559570312, "std": 257.77728271484375, "min": -346.8890075683594, "p10": -39.36994819641112, "median": 250.98041534423828, "p90": 581.7471313476562, "max": 979.3059692382812, "pos_frac": 0.859375, "sample": [-30.728572845458984, -43.073394775390625, 302.1540222167969, -136.8203125, 266.9024353027344, 581.940673828125, 23.815006256103516, 130.9808349609375, 465.407470703125, 581.2955322265625, 121.55207824707031, 911.1104736328125, 428.4107666015625, 3.46575927734375, 13.379241943359375, 59.1768798828125, 272.76873779296875, 159.087646484375, 280.5855712890625, 521.808349609375, 326.26361083984375, -99.90129089355469, 685.9716186523438, 14.273246765136719, 713.8656005859375, 387.86773681640625, 255.45660400390625, 102.05305480957031, 29.097984313964844, -139.2327423095703, 264.0247802734375, 389.1551818847656, -78.9197998046875, 133.38377380371094, -346.8890075683594, 95.02909851074219, 226.9599609375, 332.003662109375, 38.78154754638672, -48.18603515625, 86.46592712402344, 211.61447143554688, 979.3059692382812, 358.6804504394531, 197.09136962890625, 281.48358154296875, 286.34783935546875, 382.3551025390625, 403.67578125, 160.56167602539062, 107.50770568847656, 597.5436401367188, -11.531570434570312, 246.5042266845703, 268.85272216796875, 851.7628173828125, 131.51441955566406, 426.87835693359375, 101.5957260131836, 318.7247314453125, 528.56982421875, 273.9338073730469, 423.58953857421875, 148.14927673339844], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000606.npy"}
|
|
{"epoch": 0.8898678414096917, "step": 607, "batch_size": 64, "mean": 207.49200439453125, "std": 242.16004943847656, "min": -332.6207580566406, "p10": -98.4650909423828, "median": 185.7003173828125, "p90": 520.7430053710938, "max": 752.20947265625, "pos_frac": 0.796875, "sample": [237.33619689941406, 160.03892517089844, 256.218017578125, 107.93937683105469, 46.024681091308594, 133.1295623779297, 695.6502685546875, 462.88433837890625, 7.9476318359375, 687.8261108398438, -155.55328369140625, 257.7778015136719, 231.65704345703125, -28.948402404785156, 107.3583984375, 728.688720703125, -25.666515350341797, 31.38555335998535, 78.88703155517578, 43.43067169189453, -80.59492492675781, -106.12373352050781, 95.3122787475586, 319.44659423828125, -163.49020385742188, 428.76544189453125, -62.23670196533203, 458.489990234375, 397.8498840332031, 175.545654296875, 308.2530517578125, 80.03665924072266, 537.3114013671875, 104.3082275390625, 525.10498046875, 420.84490966796875, 435.9450378417969, -19.05841064453125, 510.5650634765625, 161.8240966796875, -332.6207580566406, 365.8744812011719, -129.57855224609375, 461.20989990234375, 392.34613037109375, 84.85090637207031, -11.354148864746094, 254.18923950195312, -148.20504760742188, 220.27955627441406, 116.38809204101562, 228.97183227539062, 22.504425048828125, 199.80502319335938, 101.91747283935547, 17.786575317382812, 352.72174072265625, 752.20947265625, 571.1168212890625, 499.57135009765625, 195.85498046875, -108.56038665771484, 376.5274963378906, 203.5706024169922], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000607.npy"}
|
|
{"epoch": 0.8913362701908958, "step": 608, "batch_size": 64, "mean": 231.10963439941406, "std": 227.98318481445312, "min": -232.4120635986328, "p10": -67.3290802001953, "median": 222.7414779663086, "p90": 516.554248046875, "max": 863.5923461914062, "pos_frac": 0.8125, "sample": [429.9652404785156, 484.3162841796875, 124.2376480102539, 147.78228759765625, -98.0062026977539, 485.98919677734375, 96.6006088256836, 64.66792297363281, 237.5742950439453, 406.06085205078125, 440.490478515625, -72.16090393066406, 96.71581268310547, 395.28857421875, -56.05482482910156, 188.5791473388672, -188.60018920898438, 529.8240356445312, 1.409158706665039, 358.5685729980469, 378.606201171875, 244.4541473388672, 71.32806396484375, 259.384033203125, -97.2593002319336, 227.4528045654297, 332.8159484863281, 344.7275390625, 517.8003540039062, -94.90980529785156, 48.888851165771484, -26.084091186523438, 192.8572540283203, 137.27450561523438, 167.70936584472656, 297.72454833984375, 88.12126159667969, 76.09623718261719, 572.8751220703125, -4.242881774902344, 863.5923461914062, 300.19073486328125, 551.701904296875, 218.0301513671875, 377.76171875, 191.77972412109375, 15.326416015625, 388.6515808105469, 356.9363708496094, 86.65068054199219, 474.3980407714844, -93.11882781982422, 345.1534423828125, 513.6466674804688, -232.4120635986328, 508.0821533203125, -49.08567428588867, 109.49765014648438, 611.0110473632812, 409.03790283203125, 335.4071044921875, 572.9379272460938, 140.4540252685547, -13.48199462890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000608.npy"}
|
|
{"epoch": 0.8928046989720999, "step": 609, "batch_size": 64, "mean": 270.2357177734375, "std": 275.0284729003906, "min": -556.1978149414062, "p10": -33.8574161529541, "median": 244.85995483398438, "p90": 564.650732421875, "max": 1210.37060546875, "pos_frac": 0.875, "sample": [168.7458038330078, 182.4228515625, 88.970703125, -176.6757354736328, 266.41168212890625, 292.65142822265625, 565.912841796875, 276.20733642578125, 299.46295166015625, -59.6781005859375, 413.396728515625, 561.705810546875, 71.67787170410156, 309.78350830078125, 449.0955810546875, -556.1978149414062, -40.779510498046875, 309.396484375, 413.2181701660156, 152.5235595703125, 191.18338012695312, 229.9132080078125, 258.67791748046875, 429.0326232910156, 105.69449615478516, 418.6915283203125, 805.6098022460938, 93.12686157226562, 124.93099975585938, 365.1280822753906, 244.083251953125, 331.22076416015625, 794.2057495117188, 472.9495849609375, 1210.37060546875, 648.1676025390625, -34.95648956298828, 377.740234375, 67.53534698486328, 320.6806945800781, 208.13461303710938, 233.8790740966797, 379.90228271484375, 1084.76904296875, 365.9404602050781, 245.63665771484375, 255.69210815429688, -31.292911529541016, -69.52473449707031, 110.80374145507812, 239.78672790527344, 141.52053833007812, 143.56825256347656, 155.40357971191406, 20.46869659423828, 191.3624267578125, 232.0347900390625, 115.39703369140625, 491.4731140136719, 697.5096435546875, 6.690114974975586, 386.3545837402344, 285.900146484375, -38.5611686706543], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000609.npy"}
|
|
{"epoch": 0.8942731277533039, "step": 610, "batch_size": 64, "mean": 206.88259887695312, "std": 246.189697265625, "min": -348.3037414550781, "p10": -114.30789337158203, "median": 229.71805572509766, "p90": 506.585531616211, "max": 792.2100830078125, "pos_frac": 0.796875, "sample": [253.51654052734375, 431.6557922363281, 237.82655334472656, 238.92811584472656, -104.89328002929688, 367.14141845703125, 57.71290588378906, 325.304443359375, 269.5782470703125, -164.59564208984375, 13.773628234863281, 266.21063232421875, 16.35924530029297, 624.6080932617188, 569.5828857421875, 279.24859619140625, 161.97024536132812, 160.2135772705078, -232.60818481445312, -216.81019592285156, 152.6259307861328, 493.8350830078125, 206.02178955078125, 591.0735473632812, -163.38616943359375, 260.45379638671875, 40.107845306396484, -32.90571594238281, 700.3815307617188, 504.7464294433594, 792.2100830078125, 68.89320373535156, 36.63001251220703, 446.75115966796875, 310.59832763671875, 392.12371826171875, -17.492395401000977, 270.0908508300781, 110.17070770263672, 330.0726013183594, 341.4999084472656, 144.72169494628906, -118.34272766113281, 123.79948425292969, 335.1064758300781, 153.5257568359375, 437.08819580078125, 149.03634643554688, -91.97830200195312, 290.4607238769531, 28.38226318359375, 423.7451171875, 126.66734313964844, -14.806652069091797, 352.67095947265625, 240.32809448242188, 259.062255859375, 221.60955810546875, -39.916542053222656, 202.03916931152344, -249.27325439453125, -348.3037414550781, 507.37371826171875, 718.2639770507812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000610.npy"}
|
|
{"epoch": 0.895741556534508, "step": 611, "batch_size": 64, "mean": 256.1167907714844, "std": 263.9600830078125, "min": -226.79656982421875, "p10": -75.32529525756833, "median": 237.34295654296875, "p90": 629.4976013183594, "max": 821.8001098632812, "pos_frac": 0.828125, "sample": [511.4783020019531, 210.96182250976562, 821.8001098632812, 586.091064453125, 469.5376281738281, 280.0411682128906, 22.93523406982422, 481.39117431640625, 179.25360107421875, 235.51678466796875, 247.5286865234375, 662.4404907226562, -110.32914733886719, -20.339702606201172, -131.41043090820312, -41.714149475097656, -4.022886276245117, -216.58348083496094, 124.4411849975586, 625.8945922851562, -226.79656982421875, 487.0770568847656, 502.0867919921875, -150.3468780517578, 488.9013366699219, 333.67529296875, 238.09378051757812, 762.4996948242188, 258.3713073730469, 144.28477478027344, 41.064910888671875, 631.041748046875, 279.60986328125, 12.248956680297852, 125.47216796875, 386.59271240234375, 126.50616455078125, 154.13461303710938, -89.73007202148438, 236.59213256835938, 434.35968017578125, 274.541748046875, -6.8246612548828125, 681.357421875, 256.10498046875, 258.52484130859375, 9.904485702514648, 199.82290649414062, 516.6006469726562, 50.099632263183594, 405.31463623046875, -186.91551208496094, 187.9488525390625, 241.8359375, 121.33346557617188, 42.941650390625, 407.7250061035156, 554.6315307617188, 234.65512084960938, 557.2660522460938, 716.2310180664062, 19.400466918945312, 712.6541137695312, 25.66909408569336], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000611.npy"}
|
|
{"epoch": 0.8972099853157122, "step": 612, "batch_size": 64, "mean": 269.17584228515625, "std": 242.01010131835938, "min": -190.49700927734375, "p10": -7.724334335327147, "median": 274.2870635986328, "p90": 604.2285278320312, "max": 868.02685546875, "pos_frac": 0.875, "sample": [641.0316772460938, 277.1440124511719, 356.0645446777344, -190.49700927734375, -8.27225112915039, 199.48648071289062, 546.0835571289062, 330.3731384277344, 166.94895935058594, -117.15614318847656, -128.52175903320312, 632.6002807617188, 93.67922973632812, 809.9630126953125, 362.2159423828125, 560.8859252929688, 430.7213134765625, 193.35316467285156, 292.5472106933594, 604.6029052734375, 423.9891052246094, 48.02873992919922, -180.33302307128906, 439.35076904296875, 642.1469116210938, 65.41253662109375, 297.2793273925781, -49.4305419921875, 303.6495056152344, 462.4788818359375, 363.09661865234375, 104.04244995117188, 125.67854309082031, 100.94147491455078, 398.0037536621094, 33.2643928527832, 311.08172607421875, 262.6299133300781, 88.8328857421875, 87.425048828125, 130.89991760253906, 271.43011474609375, 36.51068115234375, 217.43875122070312, 664.7359008789062, 38.49069595336914, 603.35498046875, 419.08087158203125, 868.02685546875, 264.00335693359375, 72.51002502441406, 427.45159912109375, 600.6840209960938, 96.98316192626953, -27.322372436523438, 78.48088073730469, 376.4767150878906, 47.35810852050781, 313.88262939453125, 135.97425842285156, 585.9506225585938, 306.03436279296875, 324.4401550292969, -6.44586181640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000612.npy"}
|
|
{"epoch": 0.8986784140969163, "step": 613, "batch_size": 64, "mean": 251.95529174804688, "std": 218.29322814941406, "min": -254.81085205078125, "p10": 26.279509925842284, "median": 226.24600982666016, "p90": 502.5793090820313, "max": 839.274169921875, "pos_frac": 0.9375, "sample": [430.0169677734375, 331.09326171875, 412.449462890625, -29.118663787841797, 359.63311767578125, 295.5543212890625, 58.50264358520508, 244.10260009765625, 400.4409484863281, 70.02262115478516, 623.602783203125, 26.163352966308594, 346.5459289550781, 148.14724731445312, 807.0891723632812, 217.4099884033203, -76.98762512207031, 495.64996337890625, 189.2167205810547, 65.88426208496094, 310.19195556640625, 76.02220916748047, 308.3968200683594, 500.995849609375, 456.737060546875, 223.1339874267578, 503.2579345703125, 446.5909423828125, 114.9417724609375, 327.98553466796875, 303.351806640625, 105.58065795898438, 26.5505428314209, 429.51348876953125, 194.9233856201172, 630.2920532226562, 127.73107147216797, 316.0198974609375, 839.274169921875, 1.0125923156738281, -36.340538024902344, 554.5513305664062, 337.68707275390625, 36.85320281982422, 139.3201904296875, 75.56277465820312, 137.7013702392578, 784.4779663085938, 6.338478088378906, 136.63238525390625, 126.66728210449219, 140.3778076171875, -254.81085205078125, 258.0748596191406, 374.024169921875, 277.3145751953125, 165.7288818359375, 229.3580322265625, 31.83592987060547, 296.14605712890625, 60.59519958496094, 110.7297134399414, 426.5871276855469, 51.80281066894531], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000613.npy"}
|
|
{"epoch": 0.9001468428781204, "step": 614, "batch_size": 64, "mean": 263.7240905761719, "std": 259.4837951660156, "min": -380.30157470703125, "p10": -72.95443267822264, "median": 256.5888977050781, "p90": 607.4689208984375, "max": 930.4140625, "pos_frac": 0.8125, "sample": [104.3849105834961, 448.995849609375, -1.241729736328125, 460.8059387207031, 193.9984893798828, 578.021240234375, 49.24472427368164, 338.01104736328125, 177.2678985595703, -54.08102035522461, 409.50628662109375, 82.84449768066406, 302.3153381347656, 205.20623779296875, 241.1961669921875, -135.203125, 504.03302001953125, 600.6221313476562, 210.78468322753906, -35.80275344848633, 564.5054931640625, 3.68310546875, 272.90338134765625, 390.85345458984375, -115.05389404296875, -46.474998474121094, 398.4921875, 190.53875732421875, 273.0028381347656, -60.06227111816406, 271.98162841796875, 590.3093872070312, -78.47964477539062, 302.12420654296875, -145.03555297851562, 10.186515808105469, 397.1108093261719, -157.7403106689453, 154.6615753173828, 486.36114501953125, 472.7424621582031, 128.0901641845703, 610.4032592773438, 234.0714111328125, 635.3905639648438, 639.7806396484375, 289.3871765136719, 202.86102294921875, 312.4541320800781, 486.5795593261719, 155.79150390625, 194.12828063964844, 561.1339111328125, 101.93434143066406, 620.4754638671875, 930.4140625, 638.5227661132812, 94.72283935546875, -107.8362045288086, -380.30157470703125, 650.248291015625, 237.54571533203125, 316.9925231933594, 468.0321960449219], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000614.npy"}
|
|
{"epoch": 0.9016152716593245, "step": 615, "batch_size": 64, "mean": 204.82322692871094, "std": 241.33505249023438, "min": -417.26739501953125, "p10": -18.596644783020007, "median": 155.91200256347656, "p90": 474.9652404785156, "max": 955.5394287109375, "pos_frac": 0.875, "sample": [82.92367553710938, 465.5, 91.24092102050781, 517.350341796875, 100.3243408203125, 411.4256286621094, 2.9460601806640625, 86.85584259033203, 61.010223388671875, 75.93609619140625, 531.6207275390625, 195.5619659423828, 180.6183624267578, -66.99774169921875, 343.83917236328125, -224.57925415039062, 56.030731201171875, -175.106201171875, 228.9834442138672, 246.33343505859375, 147.06341552734375, 74.39016723632812, 806.1141357421875, 89.82003021240234, 65.17992401123047, 420.3612976074219, 1.0246047973632812, 476.81268310546875, 312.7556457519531, 114.90252685546875, 210.9273681640625, 303.81597900390625, 353.0462646484375, 336.36517333984375, 193.62869262695312, 15.176856994628906, 265.35797119140625, 106.81893920898438, -24.128158569335938, 52.81481170654297, -417.26739501953125, 470.654541015625, 244.615966796875, 8.304916381835938, 441.5047607421875, 318.9993591308594, 164.76058959960938, 778.9823608398438, 110.73822021484375, 334.2949523925781, 423.2822570800781, -129.71229553222656, 195.8094024658203, 95.71981811523438, 39.318138122558594, -5.689779281616211, 137.02804565429688, 274.85870361328125, 298.41766357421875, 675.490478515625, -71.43852233886719, 955.5394287109375, 142.6824188232422, 87.72700500488281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000615.npy"}
|
|
{"epoch": 0.9030837004405287, "step": 616, "batch_size": 64, "mean": 157.602294921875, "std": 258.00836181640625, "min": -309.9073486328125, "p10": -202.5070861816406, "median": 130.61426544189453, "p90": 480.0932830810547, "max": 1042.2681884765625, "pos_frac": 0.765625, "sample": [468.12738037109375, -56.111083984375, 146.51010131835938, 366.1862487792969, 161.1805419921875, 67.02527618408203, 118.64642333984375, -125.01882934570312, 136.53973388671875, 668.9066772460938, 485.2215270996094, 350.7463684082031, 38.64842987060547, 145.10142517089844, -68.85963439941406, 241.44886779785156, 207.54736328125, 168.4735107421875, 175.739013671875, -48.71288299560547, 593.0989990234375, 410.65618896484375, 65.99392700195312, 52.495582580566406, -206.83529663085938, -281.0555725097656, 306.0323181152344, 108.2891845703125, 128.70523071289062, 532.6163330078125, 1042.2681884765625, 110.85865020751953, -309.9073486328125, 87.07909393310547, 0.21049118041992188, -271.7412109375, -45.05211639404297, 14.518192291259766, 159.32424926757812, 378.68890380859375, 215.8730926513672, 313.3056945800781, 105.13478088378906, 54.23841094970703, -238.03695678710938, 347.1421813964844, 132.52330017089844, -82.45973205566406, 50.46302032470703, 342.54693603515625, 39.95674133300781, -192.40792846679688, 464.1651306152344, 125.23484802246094, 332.410400390625, 117.18867492675781, 441.0667724609375, 161.23561096191406, -65.66375732421875, 299.55694580078125, 558.5941162109375, -251.75010681152344, 525.9068603515625, -233.26834106445312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000616.npy"}
|
|
{"epoch": 0.9045521292217328, "step": 617, "batch_size": 64, "mean": 164.12026977539062, "std": 250.6628875732422, "min": -321.33197021484375, "p10": -99.76354370117187, "median": 129.61108779907227, "p90": 423.90352478027353, "max": 962.669921875, "pos_frac": 0.734375, "sample": [105.95341491699219, 89.73043823242188, -92.89804077148438, 338.2294616699219, 58.51118469238281, -85.85881805419922, 43.08123016357422, -284.78350830078125, 374.25274658203125, 72.86013793945312, 44.081947326660156, 153.783935546875, 312.3019104003906, 362.2419128417969, 61.673030853271484, 176.36691284179688, -37.793548583984375, 353.639892578125, 15.793228149414062, 652.3678588867188, 402.7103271484375, 190.1178741455078, 571.0934448242188, 20.677724838256836, 183.55299377441406, 402.9092712402344, 96.4688949584961, -31.130756378173828, 356.997314453125, -20.858291625976562, 0.2163410186767578, 333.1259765625, -110.67402648925781, -27.187889099121094, 289.97027587890625, 289.4107360839844, 342.2859802246094, 322.07379150390625, 324.6947937011719, 294.6183776855469, 210.4062957763672, -195.5989990234375, 458.1018981933594, 341.2127380371094, 917.0450439453125, 432.90106201171875, 137.17811584472656, -265.10308837890625, 75.27969360351562, 191.5893096923828, -321.33197021484375, 31.616531372070312, 69.05028533935547, 163.369384765625, 122.04405975341797, -115.67691040039062, 498.1991882324219, -27.6907958984375, -87.02230072021484, -66.92010498046875, 962.669921875, 161.7760009765625, -31.300399780273438, -102.70590209960938], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000617.npy"}
|
|
{"epoch": 0.9060205580029369, "step": 618, "batch_size": 64, "mean": 245.20562744140625, "std": 271.394775390625, "min": -359.482666015625, "p10": -104.35360641479491, "median": 240.62266540527344, "p90": 594.5861267089845, "max": 886.3411254882812, "pos_frac": 0.828125, "sample": [-359.482666015625, 433.090087890625, 133.89337158203125, 209.53811645507812, 193.79757690429688, 185.77781677246094, 238.6563720703125, 63.09291458129883, 779.9830322265625, 188.69558715820312, 266.9783935546875, -226.53321838378906, 259.60260009765625, 198.56813049316406, 329.4816589355469, 273.8536071777344, -234.726318359375, 607.5156860351562, 149.3758544921875, -217.52938842773438, -80.64098358154297, 24.593753814697266, -112.33136749267578, 52.59648132324219, -85.73883056640625, -178.9656524658203, 220.20816040039062, 607.462158203125, 710.5969848632812, 170.53591918945312, 140.41188049316406, 74.98003387451172, 485.5335693359375, 492.0279235839844, 549.2553100585938, -284.78387451171875, -65.09988403320312, 283.8716735839844, 17.900156021118164, 296.166015625, 316.29083251953125, 500.1282958984375, 653.4246826171875, 541.2413940429688, 467.4642333984375, -34.94361114501953, 263.43084716796875, 317.9568176269531, 270.4203796386719, 652.3211669921875, 347.6120300292969, 3.1811580657958984, 886.3411254882812, 504.69305419921875, 543.3203125, 241.68663024902344, 179.3164825439453, 339.1615905761719, 500.41241455078125, 75.15145874023438, 132.24502563476562, 564.5420532226562, 395.9947509765625, 239.55870056152344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000618.npy"}
|
|
{"epoch": 0.9074889867841409, "step": 619, "batch_size": 64, "mean": 197.93942260742188, "std": 299.8244934082031, "min": -409.4754943847656, "p10": -112.2905288696289, "median": 175.16942596435547, "p90": 609.0356567382812, "max": 1025.761962890625, "pos_frac": 0.6875, "sample": [-366.56109619140625, -71.69679260253906, 276.0478515625, 249.87811279296875, 358.72760009765625, -68.8262710571289, 182.08250427246094, -74.62906646728516, 291.95330810546875, 36.13550567626953, 290.758056640625, 424.269287109375, 175.8948211669922, 897.5089111328125, -110.57286071777344, 188.3358612060547, 609.5396728515625, 105.67718505859375, 640.7523803710938, 574.9688720703125, -240.39404296875, -78.06364440917969, 587.8528442382812, 537.8023681640625, -66.16142272949219, -214.51986694335938, -29.422935485839844, 501.998291015625, 316.6362609863281, -15.74859619140625, -26.2716064453125, 271.4958801269531, -35.44422149658203, -161.0631103515625, 536.634765625, 41.5789794921875, 93.40399169921875, -103.02033996582031, 275.3796691894531, 178.72256469726562, 107.54341125488281, 755.6787109375, 385.31195068359375, 607.859619140625, 331.56494140625, 71.64791870117188, -113.02667236328125, 637.8463745117188, 543.52978515625, 174.44403076171875, 45.91230010986328, 41.75716781616211, 238.41622924804688, 1025.761962890625, -409.4754943847656, -127.880126953125, 277.15570068359375, -44.273677825927734, 623.5630493164062, 33.425228118896484, 89.29557800292969, 114.47283935546875, -11.369440078735352, 287.32305908203125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000619.npy"}
|
|
{"epoch": 0.908957415565345, "step": 620, "batch_size": 64, "mean": 199.90643310546875, "std": 266.7724609375, "min": -310.9432373046875, "p10": -68.55966796874998, "median": 198.13742065429688, "p90": 534.2791381835938, "max": 1007.9276123046875, "pos_frac": 0.796875, "sample": [228.174560546875, -266.0859680175781, 338.2057800292969, 287.9342346191406, 190.7255859375, 357.6688232421875, 222.35874938964844, 234.9425048828125, -42.576148986816406, 613.1134033203125, -85.75899505615234, 28.02545928955078, -43.995262145996094, 534.176513671875, 435.248779296875, 141.09555053710938, 510.1667175292969, 88.90574645996094, 244.8826141357422, 627.5979614257812, 444.96881103515625, 90.90567016601562, 89.83966064453125, 332.06439208984375, -210.96014404296875, 205.54925537109375, 317.94610595703125, 47.25946044921875, 225.5264129638672, 43.43153381347656, -264.5521545410156, 565.176025390625, 345.91156005859375, 999.0747680664062, 13.473876953125, 261.61614990234375, -43.274879455566406, 210.57823181152344, 222.58001708984375, 326.71429443359375, 90.73613739013672, 210.574462890625, -44.284976959228516, -52.26298522949219, 821.6898803710938, 66.46612548828125, 534.3231201171875, -310.9432373046875, 121.6048355102539, 278.19537353515625, -41.549007415771484, -75.54396057128906, 1007.9276123046875, 77.75887298583984, 67.29777526855469, 16.822784423828125, 257.2413635253906, 144.92233276367188, -102.03120422363281, 353.1961669921875, 66.89867401123047, 272.474609375, 1.192840576171875, 162.66868591308594], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000620.npy"}
|
|
{"epoch": 0.9104258443465492, "step": 621, "batch_size": 64, "mean": 285.97552490234375, "std": 273.928955078125, "min": -399.77862548828125, "p10": -11.493375968933089, "median": 249.53997802734375, "p90": 680.6499572753908, "max": 976.818359375, "pos_frac": 0.890625, "sample": [500.718994140625, -207.49224853515625, -399.77862548828125, -64.35417175292969, 556.8079223632812, 78.5547866821289, 181.26229858398438, 180.7807159423828, 252.84603881835938, 698.6964111328125, 509.12030029296875, 145.16908264160156, 180.67308044433594, 496.7763671875, 193.73806762695312, 52.85906219482422, 5.255395889282227, 302.56158447265625, 30.748085021972656, 180.0115966796875, 860.2891845703125, 337.58111572265625, 655.3302001953125, 244.2924041748047, 457.7898254394531, 928.2003784179688, 72.8774185180664, 436.4377136230469, 693.309814453125, -18.67142105102539, 154.14019775390625, 422.1194152832031, 479.92999267578125, 126.93172454833984, 290.1811218261719, 741.9033203125, 135.36807250976562, 423.3841552734375, 309.58013916015625, 136.58346557617188, -41.17405700683594, 533.7697143554688, -121.34896850585938, 369.6353759765625, 221.68667602539062, 152.337890625, 523.3470458984375, 252.47776794433594, 214.269775390625, 610.4664916992188, -258.900634765625, 115.83721160888672, 62.459678649902344, 268.3255920410156, 152.6531524658203, 246.60218811035156, 418.65960693359375, 691.5012817382812, 55.016475677490234, 302.32037353515625, 976.818359375, 278.4538879394531, 342.8222961425781, 171.88290405273438], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000621.npy"}
|
|
{"epoch": 0.9118942731277533, "step": 622, "batch_size": 64, "mean": 229.01025390625, "std": 255.44955444335938, "min": -277.88873291015625, "p10": -35.994467926025386, "median": 204.58377075195312, "p90": 535.7964050292969, "max": 1165.9765625, "pos_frac": 0.828125, "sample": [279.5188293457031, 325.64678955078125, 136.628173828125, 305.14569091796875, 235.16871643066406, 358.5446472167969, -186.59519958496094, 169.08563232421875, 911.7716064453125, 163.11399841308594, 152.06198120117188, 78.58930206298828, 135.5496368408203, -107.03584289550781, 61.881103515625, 390.4268798828125, 85.62242126464844, 247.48846435546875, 166.18618774414062, 166.01893615722656, 434.1025085449219, 42.418975830078125, 88.13545227050781, 1165.9765625, 223.60040283203125, 56.002685546875, 257.2869873046875, 370.3086242675781, 326.1780090332031, -242.31735229492188, 238.03619384765625, -35.07318115234375, 635.6718139648438, 105.69807434082031, 122.97305297851562, 660.62939453125, -11.943954467773438, -93.29082489013672, 489.27862548828125, 293.6107482910156, 545.4493408203125, 262.48028564453125, -36.389305114746094, 132.4224853515625, 246.24050903320312, 185.567138671875, 314.2132873535156, -4.378265380859375, 20.083572387695312, 150.48028564453125, 568.0171508789062, 428.20062255859375, 513.2728881835938, 338.6644592285156, 337.0176696777344, 293.4523620605469, -277.88873291015625, -25.851966857910156, 113.25215148925781, 798.2429809570312, 266.257080078125, 248.7378692626953, 77.43003845214844, -40.41851043701172], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000622.npy"}
|
|
{"epoch": 0.9133627019089574, "step": 623, "batch_size": 64, "mean": 181.05374145507812, "std": 266.7269287109375, "min": -553.5941772460938, "p10": -83.20799713134765, "median": 120.99246215820312, "p90": 578.6808898925783, "max": 1046.3770751953125, "pos_frac": 0.8125, "sample": [85.646484375, 216.27264404296875, -296.82598876953125, 643.1315307617188, 165.20233154296875, 42.55458450317383, 346.080322265625, 390.8904113769531, 239.59811401367188, 106.2442626953125, 840.4845581054688, 68.62367248535156, 274.4154357910156, -18.456619262695312, 255.84744262695312, -12.859333038330078, 595.5614013671875, 123.68756103515625, 54.46246337890625, 87.19232177734375, 88.95079040527344, 194.731201171875, 203.06048583984375, 87.6597900390625, -553.5941772460938, 190.7642822265625, 291.67120361328125, -49.769065856933594, 217.52613830566406, 272.9871826171875, -83.88021850585938, 127.29959106445312, 48.634315490722656, 37.11170196533203, -85.77175903320312, 48.52769470214844, 605.6861572265625, 304.2264404296875, 328.2413024902344, 167.09365844726562, -81.63948059082031, 97.19570922851562, 217.6658172607422, 539.2930297851562, -146.85897827148438, 0.8342475891113281, 1046.3770751953125, 399.1907958984375, 29.736286163330078, -132.183349609375, 796.0210571289062, 336.7775573730469, 61.90602111816406, 106.95963287353516, 118.29736328125, 645.6959838867188, -210.28460693359375, 112.53062438964844, 189.980712890625, 92.39601135253906, 369.6611328125, -21.25823974609375, 364.39794921875, 5.8360595703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000623.npy"}
|
|
{"epoch": 0.9148311306901615, "step": 624, "batch_size": 64, "mean": 256.51275634765625, "std": 265.1474914550781, "min": -476.1198425292969, "p10": -56.85379257202148, "median": 263.57696533203125, "p90": 613.5181396484376, "max": 772.700927734375, "pos_frac": 0.828125, "sample": [257.8168029785156, -27.180500030517578, 496.40716552734375, 3.1883621215820312, 406.3304443359375, 464.93145751953125, 446.0667724609375, 772.700927734375, 649.207275390625, 59.30481719970703, 13.49554443359375, 28.8677978515625, 217.3768310546875, 444.03131103515625, 244.04702758789062, -98.52728271484375, 397.2893981933594, 585.865234375, 645.077392578125, -476.1198425292969, 176.75404357910156, 484.8006591796875, -54.19280242919922, 136.8517303466797, 269.3371276855469, 192.46014404296875, 460.7462463378906, 577.1898193359375, 414.9381408691406, 43.5081787109375, 224.56239318847656, 477.0231018066406, -150.84539794921875, 282.9327087402344, -17.8619384765625, 363.219482421875, 389.7643737792969, 63.09226989746094, 270.8040466308594, 358.78057861328125, -109.7414779663086, 73.76229095458984, 299.4610595703125, -57.99421691894531, 578.2735595703125, 423.1447448730469, 144.84864807128906, 392.44793701171875, 637.8714599609375, 379.3426208496094, 162.4043731689453, -251.79685974121094, 40.10601806640625, 754.045654296875, 70.98247528076172, 75.43058776855469, -46.139862060546875, 137.7034912109375, -103.98875427246094, 280.3564453125, 710.4425048828125, 163.89434814453125, 625.369384765625, 542.5474853515625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000624.npy"}
|
|
{"epoch": 0.9162995594713657, "step": 625, "batch_size": 64, "mean": 236.09140014648438, "std": 257.5205993652344, "min": -258.5920715332031, "p10": -36.30623302459716, "median": 201.11408233642578, "p90": 568.2857604980469, "max": 1234.2470703125, "pos_frac": 0.84375, "sample": [-18.225685119628906, -39.769683837890625, 386.0308532714844, 391.7301330566406, 69.79354858398438, 177.71702575683594, 561.06201171875, 914.7244873046875, 124.02699279785156, 379.3965148925781, 84.83514404296875, -258.5920715332031, 55.204132080078125, 152.19969177246094, 571.3816528320312, 380.33331298828125, 24.604299545288086, -51.19334030151367, 1234.2470703125, 358.84320068359375, 726.395751953125, 175.18174743652344, 93.97968292236328, 221.4624786376953, -95.3134536743164, 319.10467529296875, 720.8593139648438, 46.25495910644531, 0.1649322509765625, 208.53042602539062, 138.3953857421875, 225.920166015625, 443.6408996582031, -47.48874282836914, 381.3372802734375, 85.6239013671875, 62.269691467285156, 296.1360778808594, 195.15887451171875, 95.05897521972656, 252.2240447998047, 225.90753173828125, 530.5455322265625, 238.24876403808594, 213.8112335205078, 204.09347534179688, 29.64503288269043, 370.43548583984375, 232.35301208496094, 198.1346893310547, -28.2248477935791, -3.4311065673828125, 425.55926513671875, 204.58363342285156, 111.00630187988281, 48.577537536621094, 194.81809997558594, 186.10336303710938, 614.3424682617188, -145.17141723632812, 328.0286865234375, -63.810516357421875, 607.7726440429688, 343.2744140625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000625.npy"}
|
|
{"epoch": 0.9177679882525698, "step": 626, "batch_size": 64, "mean": 190.15899658203125, "std": 248.3304901123047, "min": -252.22805786132812, "p10": -84.03362808227538, "median": 133.34564208984375, "p90": 529.7066101074221, "max": 940.7040405273438, "pos_frac": 0.734375, "sample": [405.2207946777344, 94.06585693359375, -86.24078369140625, 269.583740234375, 135.0283660888672, 213.62547302246094, 468.2701721191406, 64.65548706054688, 455.01605224609375, -120.86719512939453, 580.6529541015625, 362.56378173828125, 594.8282470703125, 940.7040405273438, -130.12908935546875, -6.543479919433594, 255.22369384765625, -78.88359832763672, 35.484161376953125, 40.39019775390625, 87.34991455078125, -20.091781616210938, 71.67601013183594, 15.450845718383789, -27.52137565612793, 131.6629180908203, 122.94732666015625, -179.35865783691406, 53.51413345336914, 492.2861328125, 8.710039138793945, 363.5365295410156, -73.71659851074219, -120.68830871582031, 217.15487670898438, 438.98553466796875, 547.2926635742188, 164.92330932617188, 148.44932556152344, 450.7137145996094, 458.76715087890625, -40.69837951660156, 722.426513671875, 353.902587890625, 732.049560546875, 101.49530792236328, -67.30962371826172, 238.18716430664062, 60.67897033691406, 340.49114990234375, -3.10931396484375, 545.7439575195312, 331.8715515136719, 398.4874572753906, -5.151031494140625, 106.41783142089844, 204.60247802734375, 242.9463653564453, -252.22805786132812, -119.57328796386719, 140.76365661621094, 41.231483459472656, 259.70281982421875, -7.445003509521484], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000626.npy"}
|
|
{"epoch": 0.9192364170337739, "step": 627, "batch_size": 64, "mean": 273.4734191894531, "std": 346.5963134765625, "min": -406.1734619140625, "p10": -83.69394607543944, "median": 196.96349334716797, "p90": 756.624938964844, "max": 1533.6268310546875, "pos_frac": 0.78125, "sample": [283.2281188964844, 169.72573852539062, 818.1624755859375, 334.09539794921875, 80.74030303955078, -21.684402465820312, 179.60400390625, 441.4068908691406, 702.109619140625, 202.77655029296875, 642.2684936523438, -227.71234130859375, 163.99908447265625, -89.62747192382812, -69.84905242919922, 595.0775756835938, 21.699317932128906, 233.46746826171875, 493.3714599609375, 542.76611328125, -245.8887481689453, 368.90093994140625, 368.9860534667969, -0.9217662811279297, 957.774658203125, 159.99331665039062, 341.6966857910156, 779.9886474609375, 120.261962890625, 65.24198150634766, 195.85287475585938, 52.23935317993164, 194.10787963867188, 905.952392578125, 546.6647338867188, 418.8053283691406, 560.3261108398438, 182.88296508789062, 361.92645263671875, -140.04518127441406, 63.233253479003906, -8.190202713012695, 1533.6268310546875, 289.78155517578125, 164.91571044921875, -26.635902404785156, 999.084228515625, -132.04266357421875, 84.40579223632812, -22.059722900390625, 682.7600708007812, 3.1914901733398438, 390.77972412109375, -406.1734619140625, 121.11756896972656, 332.626708984375, 198.07411193847656, 65.18840026855469, -184.46755981445312, 881.2100219726562, 350.1098327636719, -7.3048248291015625, 238.041015625, 200.655029296875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000627.npy"}
|
|
{"epoch": 0.920704845814978, "step": 628, "batch_size": 64, "mean": 267.0002746582031, "std": 269.0215148925781, "min": -376.24853515625, "p10": -33.93421401977539, "median": 234.68248748779297, "p90": 616.3008728027344, "max": 964.187744140625, "pos_frac": 0.796875, "sample": [271.90338134765625, -8.831121444702148, 362.9329833984375, 779.860107421875, 622.6407470703125, 140.65267944335938, 382.0549621582031, 101.91924285888672, 22.996383666992188, -33.26483154296875, -29.425086975097656, 73.65052032470703, -376.24853515625, 607.2955932617188, 404.4050598144531, 620.1602783203125, 56.405609130859375, -17.72521209716797, 964.187744140625, 455.888427734375, 511.9072265625, 455.94964599609375, 193.59010314941406, 216.72103881835938, 57.607032775878906, -41.703712463378906, 219.83729553222656, 304.1059875488281, -37.95037078857422, -16.360515594482422, 637.2665405273438, 310.1761779785156, 502.00299072265625, 548.5816650390625, 546.1527709960938, 435.77508544921875, 280.5451354980469, 458.962158203125, 29.76271629333496, 126.37445068359375, -64.76424407958984, 285.78741455078125, 139.87222290039062, 248.4949951171875, 460.89190673828125, 374.470947265625, 497.1880798339844, 151.24061584472656, -34.221092224121094, -194.09234619140625, 187.62063598632812, 94.28226470947266, 575.3260498046875, 657.305419921875, -140.42462158203125, 56.94237518310547, 605.73486328125, 177.72216796875, 475.5921936035156, 120.10369873046875, 796.675048828125, 275.172119140625, 220.86997985839844, -20.535655975341797], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000628.npy"}
|
|
{"epoch": 0.922173274596182, "step": 629, "batch_size": 64, "mean": 228.7070770263672, "std": 218.15367126464844, "min": -239.12559509277344, "p10": -19.648378944396967, "median": 210.54822540283203, "p90": 485.0705169677736, "max": 1009.6665649414062, "pos_frac": 0.859375, "sample": [206.04489135742188, 325.9405517578125, -32.93196105957031, -54.104957580566406, 425.16876220703125, 95.99580383300781, 123.62379455566406, 272.4256896972656, 406.832275390625, 430.2806396484375, 334.5391845703125, 168.181396484375, 174.93621826171875, 110.1357421875, 200.71070861816406, -22.85511016845703, 186.66046142578125, 128.38088989257812, -12.166006088256836, 422.0279541015625, 28.350433349609375, 549.553466796875, 219.76840209960938, -210.3777618408203, 54.9245491027832, 224.31436157226562, -68.36256408691406, -45.247474670410156, 452.9170837402344, 313.75677490234375, 116.1154556274414, 309.84857177734375, 1009.6665649414062, 698.6546020507812, 400.52862548828125, 157.16514587402344, -239.12559509277344, 556.3469848632812, 369.5067138671875, 31.500808715820312, 0.1154937744140625, -10.7369384765625, 202.46278381347656, 47.75596618652344, 331.6070556640625, 581.625732421875, 207.35311889648438, 213.7433319091797, 303.9173583984375, 26.326078414916992, 290.1694641113281, 197.6525421142578, 234.10665893554688, 422.2939453125, 107.54074096679688, 498.7667236328125, 34.67658233642578, 507.2923278808594, 453.1127014160156, 308.0837707519531, 352.8607177734375, 241.4354248046875, 244.5162353515625, 20.9429931640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000629.npy"}
|
|
{"epoch": 0.9236417033773862, "step": 630, "batch_size": 64, "mean": 167.92913818359375, "std": 250.61485290527344, "min": -330.5079040527344, "p10": -102.28007354736326, "median": 123.8551254272461, "p90": 451.2589782714844, "max": 1071.91552734375, "pos_frac": 0.84375, "sample": [543.4852294921875, -281.2027282714844, -8.200231552124023, 174.35940551757812, -135.77743530273438, 440.5279235839844, 142.64236450195312, 87.0518798828125, 1071.91552734375, 44.125938415527344, 295.0828857421875, 52.69609069824219, 157.25454711914062, 414.3291015625, 455.8580017089844, -330.5079040527344, 46.678314208984375, 42.94818115234375, 127.40384674072266, 152.89962768554688, 124.45450592041016, 11.998340606689453, 96.82771301269531, 647.9349365234375, -173.81703186035156, 779.5478515625, 89.21434020996094, 431.74072265625, 82.80012512207031, 33.15824890136719, 30.115522384643555, 118.66082000732422, 263.93212890625, -292.1282653808594, 284.2414855957031, 17.190046310424805, -110.07884216308594, 329.52728271484375, 110.4736328125, 236.64991760253906, 82.05056762695312, 47.20156478881836, 105.0487289428711, -84.08294677734375, 143.49349975585938, 225.95086669921875, 128.78167724609375, 161.3087158203125, 386.5263366699219, 4.590202331542969, 124.46510314941406, 524.2717895507812, 428.5224304199219, 151.61134338378906, -69.08943176269531, 137.3937530517578, 123.25574493408203, 119.55158996582031, 421.5187683105469, -154.600830078125, 708.50732421875, 85.63182830810547, 322.85955810546875, 16.683164596557617], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000630.npy"}
|
|
{"epoch": 0.9251101321585903, "step": 631, "batch_size": 64, "mean": 201.62759399414062, "std": 284.2636413574219, "min": -600.216064453125, "p10": -162.83534088134763, "median": 214.02606964111328, "p90": 593.5474914550781, "max": 838.0641479492188, "pos_frac": 0.765625, "sample": [725.494140625, 318.9559326171875, -191.64694213867188, 246.94491577148438, 133.9053955078125, 344.5570068359375, 22.321441650390625, 67.5764389038086, -129.47076416015625, -129.8835906982422, 253.33596801757812, 838.0641479492188, -305.8123474121094, 407.5743103027344, -75.75331115722656, -24.40858268737793, 107.42827606201172, 586.778076171875, 26.961402893066406, 68.4347152709961, 56.89094543457031, -13.509771347045898, 279.72625732421875, -262.91827392578125, 271.1694641113281, 217.85989379882812, 385.42236328125, 24.833084106445312, 446.719482421875, 257.0686340332031, -600.216064453125, 355.6623229980469, 596.4486694335938, 216.5749053955078, 349.90411376953125, 141.91217041015625, 191.71249389648438, -101.87315368652344, 151.45213317871094, 369.1592712402344, 291.6005859375, -33.24890899658203, 818.3974609375, 427.6993408203125, 379.1207580566406, -201.75942993164062, 617.4300537109375, 219.89013671875, 703.2625122070312, 460.19146728515625, 103.12187957763672, 104.62178039550781, 16.36581039428711, 84.0815658569336, -178.16845703125, -21.7386474609375, 387.83355712890625, 341.1663513183594, 730.9251708984375, -176.95751953125, 211.47723388671875, 388.7762756347656, 197.098876953125, 407.6221923828125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000631.npy"}
|
|
{"epoch": 0.9265785609397944, "step": 632, "batch_size": 64, "mean": 201.32220458984375, "std": 278.7908630371094, "min": -532.7213134765625, "p10": -122.25747299194335, "median": 210.01346588134766, "p90": 599.2401062011719, "max": 761.8645629882812, "pos_frac": 0.796875, "sample": [192.0220947265625, -369.04541015625, 194.58770751953125, 168.8773193359375, 421.0623779296875, -532.7213134765625, 603.42041015625, 632.4273071289062, 624.5823364257812, 323.1346435546875, 369.13043212890625, 227.46067810058594, 219.495361328125, 602.5696411132812, 273.9439392089844, 146.64898681640625, 464.6695251464844, 273.2751159667969, -0.7748832702636719, -44.50701141357422, 540.4314575195312, 262.2623291015625, 431.7795715332031, 315.759033203125, -250.42921447753906, 671.2216796875, 343.74005126953125, -331.1863708496094, 278.4037780761719, 23.236303329467773, -387.27752685546875, -35.59442901611328, 591.47119140625, 200.5315704345703, 154.59793090820312, 253.38595581054688, 586.5526123046875, 307.0345153808594, 250.43844604492188, -104.0623550415039, 415.3486633300781, 69.0218734741211, -114.4257583618164, 8.10723876953125, 425.718505859375, 279.9444580078125, 417.9917297363281, 65.99755096435547, 613.2100830078125, 761.8645629882812, 182.20977783203125, -125.61392211914062, -262.41058349609375, 42.00415802001953, 88.32267761230469, 55.480995178222656, 45.978981018066406, 298.10455322265625, -9.279884338378906, 171.80645751953125, 53.68183898925781, 36.77912902832031, 61.449058532714844, 410.7744445800781], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000632.npy"}
|
|
{"epoch": 0.9280469897209985, "step": 633, "batch_size": 64, "mean": 173.24588012695312, "std": 260.1631164550781, "min": -357.3591613769531, "p10": -139.87426452636714, "median": 173.66622161865234, "p90": 472.11967163085944, "max": 904.4025268554688, "pos_frac": 0.765625, "sample": [-103.62759399414062, 366.5219421386719, 375.74151611328125, -87.96367645263672, 313.79815673828125, 399.67919921875, -179.8565216064453, -85.90695190429688, 771.933349609375, 209.88571166992188, 259.0882873535156, 172.16903686523438, 228.7038116455078, -283.7765808105469, -155.068603515625, 96.00556182861328, 175.1634063720703, 489.8494567871094, 153.45779418945312, 25.348358154296875, 63.85759735107422, 591.8790283203125, 272.07196044921875, -22.45196533203125, 429.57177734375, 52.030364990234375, 66.89405822753906, 131.05606079101562, -330.2613525390625, 109.00630950927734, 391.507568359375, 235.98739624023438, 245.75042724609375, 245.38067626953125, 83.17524719238281, -48.23332595825195, -104.42080688476562, -298.99749755859375, -301.453857421875, 140.51828002929688, 904.4025268554688, 353.7071838378906, -60.78472900390625, -357.3591613769531, 406.04730224609375, 351.4884338378906, 338.6670837402344, 605.8489379882812, 450.2796630859375, 219.25634765625, 4.932647705078125, 151.65823364257812, -40.48081970214844, 381.759765625, 47.74159240722656, 2.0883636474609375, 222.25904846191406, 630.8082275390625, 262.7589111328125, 80.23849487304688, 481.47967529296875, 47.995635986328125, 247.47412109375, 261.45452880859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000633.npy"}
|
|
{"epoch": 0.9295154185022027, "step": 634, "batch_size": 64, "mean": 183.81390380859375, "std": 247.04026794433594, "min": -344.3436279296875, "p10": -117.7514678955078, "median": 141.94908905029297, "p90": 525.2412841796876, "max": 720.452392578125, "pos_frac": 0.75, "sample": [-19.129226684570312, 375.98675537109375, 49.530555725097656, 693.7498168945312, 57.27836608886719, 13.111997604370117, 19.882078170776367, 363.33428955078125, 235.6119384765625, 438.2301025390625, 14.333433151245117, 156.3631591796875, 14.405328750610352, 175.5233154296875, 514.1351928710938, 205.30709838867188, 124.89892578125, -45.599700927734375, 103.71487426757812, 175.89590454101562, 575.9615478515625, -165.82823181152344, -164.51283264160156, 77.24374389648438, -134.5633544921875, 176.005126953125, -16.18865966796875, 530.0010375976562, 321.5381164550781, 95.51048278808594, 705.0997314453125, 167.17178344726562, 120.77338409423828, -344.3436279296875, 720.452392578125, 402.46270751953125, 717.64794921875, -111.83778381347656, 111.388671875, 165.26998901367188, 305.6721496582031, -120.28590393066406, -63.964149475097656, -31.96750259399414, 81.33857727050781, -129.38905334472656, 219.3678741455078, -49.74415588378906, 560.3048706054688, 448.4361267089844, 14.091035842895508, 426.905029296875, 379.065185546875, 484.970947265625, 432.6066589355469, 127.53501892089844, 223.449951171875, 426.5422058105469, -61.48204803466797, 176.62274169921875, -45.50727844238281, -133.4827117919922, 118.34293365478516, 358.8442687988281], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000634.npy"}
|
|
{"epoch": 0.9309838472834068, "step": 635, "batch_size": 64, "mean": 192.16888427734375, "std": 238.74911499023438, "min": -228.3656005859375, "p10": -62.77860870361327, "median": 152.10922241210938, "p90": 593.1296813964844, "max": 831.4476318359375, "pos_frac": 0.78125, "sample": [268.53253173828125, 266.702880859375, 764.1510620117188, -70.82196044921875, 203.83950805664062, -27.56029510498047, 54.384098052978516, 21.358226776123047, -1.8647994995117188, 164.40399169921875, 149.32147216796875, 620.3218994140625, 18.885086059570312, 240.56236267089844, -108.43765258789062, 230.21444702148438, 154.89697265625, -7.02813720703125, 610.4024047851562, 38.52888488769531, 233.25376892089844, 73.58708190917969, 546.9071044921875, 580.939453125, 242.47366333007812, 55.125099182128906, 8.20672607421875, 321.6145324707031, 218.63755798339844, 820.2308349609375, -156.00013732910156, 333.4710693359375, 83.13774871826172, 598.3540649414062, 347.2774658203125, 131.8973388671875, 72.08865356445312, -10.658512115478516, 74.2123794555664, 335.5693359375, 15.215412139892578, 370.6919250488281, 620.4585571289062, 161.40774536132812, -111.12757110595703, 81.97516632080078, 343.9591064453125, -26.611427307128906, -228.3656005859375, 288.7974548339844, -30.26422119140625, -79.58403015136719, 140.72604370117188, 297.49798583984375, 385.11920166015625, 219.54440307617188, 182.84634399414062, 137.47140502929688, 3.3756103515625, 831.4476318359375, -50.83502960205078, -67.89728546142578, 143.65631103515625, 168.18568420410156], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000635.npy"}
|
|
{"epoch": 0.9324522760646109, "step": 636, "batch_size": 64, "mean": 225.10348510742188, "std": 243.043701171875, "min": -217.13778686523438, "p10": -90.58636932373047, "median": 211.3942413330078, "p90": 471.66826477050785, "max": 1178.2886962890625, "pos_frac": 0.828125, "sample": [45.306976318359375, -80.64579772949219, 72.05545043945312, 18.54364013671875, 113.51739501953125, -81.21697998046875, 379.59381103515625, 124.35700225830078, 387.5426025390625, 58.128623962402344, 275.5506591796875, 110.87450408935547, 310.94384765625, 328.805419921875, 95.60289764404297, 203.4759521484375, -180.7886962890625, 776.3572998046875, 565.7906494140625, 400.3065490722656, 429.248779296875, 389.0680236816406, 156.15707397460938, -108.28236389160156, 370.8706359863281, 177.57528686523438, 405.8527526855469, 305.25982666015625, 206.47491455078125, 1178.2886962890625, -85.64164733886719, -95.05995178222656, 372.5458679199219, 394.209716796875, -92.70553588867188, 713.33935546875, 366.0402526855469, 252.57736206054688, 286.58917236328125, -129.639404296875, 508.849609375, 185.22251892089844, 216.31356811523438, 310.6299743652344, 226.72540283203125, 275.3670654296875, 456.4208984375, 307.1332092285156, 118.60045623779297, -217.13778686523438, 163.6555633544922, -42.7705078125, -149.062744140625, 134.63694763183594, 229.16558837890625, 185.7634735107422, 326.2393798828125, 551.1466064453125, 114.04508209228516, 270.86126708984375, 11.037918090820312, 163.27828979492188, 165.427001953125, 478.2028503417969], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000636.npy"}
|
|
{"epoch": 0.933920704845815, "step": 637, "batch_size": 64, "mean": 236.39791870117188, "std": 248.39356994628906, "min": -207.3295440673828, "p10": -34.5251220703125, "median": 200.67762756347656, "p90": 569.2327514648438, "max": 934.3697509765625, "pos_frac": 0.828125, "sample": [209.054443359375, 97.27230072021484, -20.197845458984375, 207.89898681640625, 27.213581085205078, 390.21649169921875, 386.02392578125, -207.3295440673828, 508.28216552734375, 521.84423828125, 447.56072998046875, -60.013763427734375, 125.50271606445312, 177.14315795898438, -152.7932586669922, 568.1188354492188, 284.2962951660156, 293.4128112792969, 536.4059448242188, 274.50018310546875, 606.5795288085938, 46.73571014404297, 75.1822738647461, 934.3697509765625, 102.0386962890625, 382.1134033203125, 22.46335220336914, 38.18490219116211, 432.8569030761719, 144.20022583007812, 257.72589111328125, -181.36318969726562, 37.73908233642578, 268.4373779296875, 322.22808837890625, -28.1635799407959, 86.79060363769531, 277.72998046875, 461.80670166015625, 130.5205078125, 156.6456298828125, 528.3845825195312, 153.26348876953125, 193.45626831054688, 560.906982421875, 344.91766357421875, 577.0550537109375, -11.519645690917969, 620.172607421875, -35.025489807128906, 708.19775390625, -144.7505645751953, 214.43707275390625, 547.7354125976562, 272.6518249511719, 34.23948669433594, 92.68977355957031, 637.6187133789062, -59.96861267089844, 569.7101440429688, 13.57223892211914, 77.30979919433594, 78.5352783203125, -33.35759735107422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000637.npy"}
|
|
{"epoch": 0.9353891336270191, "step": 638, "batch_size": 64, "mean": 226.19699096679688, "std": 211.06434631347656, "min": -177.6313018798828, "p10": 14.522167015075693, "median": 183.1212615966797, "p90": 532.4740234375, "max": 764.5911254882812, "pos_frac": 0.90625, "sample": [655.1015625, -138.92083740234375, -166.08795166015625, 143.54013061523438, 288.40325927734375, 57.834014892578125, 105.32315063476562, 116.86700439453125, 374.7809143066406, 120.59269714355469, 281.67254638671875, 764.5911254882812, 75.60166931152344, 561.1069946289062, 224.3230438232422, 111.23390197753906, 297.34478759765625, 164.77383422851562, 319.828369140625, 172.35089111328125, 116.81804656982422, 44.749595642089844, 85.58324432373047, 221.09368896484375, 49.62837600708008, 469.8828125, 232.794677734375, 147.57179260253906, 622.10546875, 358.1866760253906, 459.5419921875, 366.7605895996094, 285.93023681640625, 207.23825073242188, 314.99530029296875, 185.30288696289062, 31.135055541992188, -29.55624008178711, 53.878265380859375, 10.487476348876953, -177.6313018798828, 216.27764892578125, 313.953369140625, 538.6419677734375, 65.93634796142578, -39.47087860107422, 628.7860107421875, 172.09230041503906, 296.3776550292969, 274.2281494140625, 64.2222900390625, 23.936445236206055, 518.0821533203125, 166.2322540283203, 282.1986389160156, 44.669525146484375, 76.48975372314453, 78.48383331298828, 457.78167724609375, 656.0514526367188, -62.21703338623047, 180.93963623046875, 504.513916015625, 431.6417236328125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000638.npy"}
|
|
{"epoch": 0.9368575624082232, "step": 639, "batch_size": 64, "mean": 167.3184356689453, "std": 240.71636962890625, "min": -358.1801452636719, "p10": -133.0756683349609, "median": 155.52477264404297, "p90": 487.1779998779297, "max": 738.2841796875, "pos_frac": 0.71875, "sample": [531.6989135742188, -101.46519470214844, -90.42228698730469, 514.7733154296875, 422.9316711425781, 143.86314392089844, 343.4958190917969, 222.17376708984375, -49.549461364746094, 280.88653564453125, 362.7584533691406, -107.80238342285156, 365.7395324707031, 103.98501586914062, -140.02964782714844, 47.580841064453125, -99.11665344238281, -184.33770751953125, 82.81155395507812, 6.793022155761719, 523.533447265625, -147.29779052734375, 401.98895263671875, 298.8447265625, 187.30882263183594, -104.78695678710938, -229.49867248535156, 219.39027404785156, 45.221473693847656, 294.3956298828125, 475.70941162109375, 45.464332580566406, -0.6250991821289062, -116.84971618652344, 492.0931091308594, 335.78924560546875, 140.9541778564453, 563.4805908203125, 196.21412658691406, 167.1864013671875, 44.001426696777344, 364.9500427246094, 460.21307373046875, 316.3796691894531, 738.2841796875, -268.3994140625, 114.8684310913086, 332.0234680175781, -199.61264038085938, 220.7440185546875, 139.84727478027344, -50.80513000488281, 299.4378662109375, 107.6873550415039, -358.1801452636719, -98.19412994384766, -7.4263916015625, 251.90199279785156, 185.70362854003906, 377.71478271484375, 124.49334716796875, 551.2064208984375, 473.91912841796875, 142.33685302734375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000639.npy"}
|
|
{"epoch": 0.9383259911894273, "step": 640, "batch_size": 64, "mean": 275.9643249511719, "std": 290.27294921875, "min": -196.8724365234375, "p10": -82.78165588378904, "median": 269.3448944091797, "p90": 621.0504333496095, "max": 1258.950927734375, "pos_frac": 0.859375, "sample": [641.5759887695312, 313.7934875488281, -13.182313919067383, 87.35902404785156, 163.22134399414062, 100.62112426757812, 466.36212158203125, 649.8883666992188, 320.4957275390625, 47.44055938720703, 532.9003295898438, 762.5239868164062, 472.54461669921875, 92.21737670898438, 42.78810501098633, 157.22787475585938, 346.44622802734375, -100.9725341796875, -94.31440734863281, 280.3638916015625, -92.23995971679688, 376.79144287109375, 253.15481567382812, 139.85874938964844, 40.63778305053711, -150.74757385253906, 62.970428466796875, 314.31201171875, 1258.950927734375, 510.20068359375, 363.88134765625, 334.06842041015625, -196.8724365234375, 381.44671630859375, 952.29052734375, 44.57307434082031, -171.55657958984375, 202.57530212402344, -60.7122802734375, 288.2042236328125, 160.3179473876953, 321.9017333984375, 113.53611755371094, 524.1127319335938, 322.9871826171875, 124.04104614257812, 268.48516845703125, 460.3323974609375, -137.76712036132812, 573.157470703125, 873.7755126953125, 296.30078125, 36.179229736328125, 388.3432922363281, 9.549942016601562, 270.2046203613281, 281.5245361328125, 249.63003540039062, 80.51222229003906, 487.1694641113281, 549.2568359375, 964.2701416015625, 236.0017547607422, 86.804931640625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000640.npy"}
|
|
{"epoch": 0.9397944199706314, "step": 641, "batch_size": 64, "mean": 195.2603759765625, "std": 279.43115234375, "min": -374.3211975097656, "p10": -140.60601730346679, "median": 195.51599884033203, "p90": 506.17774658203126, "max": 1451.2587890625, "pos_frac": 0.765625, "sample": [115.39556121826172, 193.7451934814453, 336.8454895019531, 622.9951782226562, -201.54513549804688, 84.21194458007812, 384.4239501953125, 486.0654296875, 192.1500244140625, -8.676559448242188, -100.25636291503906, 204.30416870117188, -37.47550964355469, -126.29334259033203, 1451.2587890625, -193.28369140625, -72.75794219970703, -217.4637451171875, 21.146865844726562, 273.4977111816406, 233.99612426757812, 118.55030059814453, 85.4742431640625, 238.5262908935547, 214.75369262695312, 202.97171020507812, -374.3211975097656, 143.96546936035156, 226.98907470703125, 134.7432098388672, 186.3512420654297, 221.7997589111328, 100.55596160888672, -34.87890625, 101.48493194580078, 209.02017211914062, 620.7857666015625, 117.43775939941406, 234.11231994628906, 197.28680419921875, 255.07809448242188, -57.7333984375, -33.319313049316406, -279.07183837890625, 153.5155029296875, 157.8840789794922, -281.35601806640625, 241.0196075439453, 407.44757080078125, 519.0892333984375, 659.9845581054688, 507.1577453613281, -146.74002075195312, 308.0683288574219, 190.86666870117188, 396.9759216308594, 376.40814208984375, 304.9644775390625, 218.1044464111328, 151.82272338867188, 333.3791198730469, 413.43896484375, 503.8910827636719, 607.8948364257812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000641.npy"}
|
|
{"epoch": 0.9412628487518355, "step": 642, "batch_size": 64, "mean": 235.2342987060547, "std": 330.4285583496094, "min": -363.4966125488281, "p10": -96.99999237060547, "median": 188.65003204345703, "p90": 627.1670349121094, "max": 1620.74072265625, "pos_frac": 0.765625, "sample": [162.41221618652344, -50.78755187988281, 360.0488586425781, 328.3733215332031, 181.90625, -363.4966125488281, 580.03076171875, 127.36978149414062, -57.052024841308594, 123.90821838378906, 15.044609069824219, 44.761383056640625, 584.5860595703125, -134.13580322265625, 353.482666015625, 760.4427490234375, 189.6969451904297, 465.323486328125, -325.29058837890625, -42.66830825805664, -211.6181640625, -35.554901123046875, 15.365898132324219, 82.91856384277344, 123.96490478515625, 294.31536865234375, 109.05486297607422, 214.87588500976562, 81.93862915039062, 294.9745788574219, 451.0680236816406, 232.07232666015625, 434.0864562988281, 838.3892822265625, -97.8004150390625, 187.60311889648438, 152.7183837890625, 436.12646484375, -271.7440185546875, 263.8271789550781, 452.014892578125, 218.5276336669922, 209.1365966796875, 628.5923461914062, -2.1113853454589844, 826.2330932617188, 459.0312194824219, 623.84130859375, -95.13233947753906, 222.82794189453125, 1620.74072265625, 285.81976318359375, -39.25636291503906, 89.00132751464844, -68.11836242675781, -110.05726623535156, 229.18141174316406, 888.124755859375, 131.30715942382812, 36.197532653808594, 296.5535888671875, 686.791748046875, 61.80682373046875, 503.40155029296875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000642.npy"}
|
|
{"epoch": 0.9427312775330396, "step": 643, "batch_size": 64, "mean": 218.98019409179688, "std": 246.63710021972656, "min": -342.1142578125, "p10": -80.88766479492188, "median": 219.64913940429688, "p90": 549.3936828613282, "max": 826.5286865234375, "pos_frac": 0.84375, "sample": [226.61578369140625, 52.50202560424805, 150.127197265625, 558.25537109375, 436.71282958984375, 422.1287536621094, 364.4275207519531, -98.49895477294922, -83.27059936523438, 195.0633087158203, 13.060291290283203, 5.8066253662109375, 254.87936401367188, 66.8250732421875, -101.22850036621094, 436.1170959472656, 556.8032836914062, 532.1046142578125, 256.2566833496094, 332.7714538574219, 160.7604522705078, 39.826637268066406, 34.990150451660156, 134.54400634765625, 147.69505310058594, 165.07571411132812, -284.50494384765625, 305.5874328613281, 276.4883728027344, 390.6015625, -342.1142578125, 15.97547721862793, -75.32748413085938, 447.9505615234375, 613.276611328125, 642.2214965820312, 288.37982177734375, -74.45254516601562, 253.37838745117188, 444.5528564453125, 42.3060188293457, 786.772705078125, 666.6134643554688, 348.18475341796875, 363.640869140625, 374.38458251953125, 3.397350311279297, 251.0184783935547, 826.5286865234375, 23.73757553100586, 490.1735534667969, 322.0940246582031, 288.66912841796875, -135.53179931640625, 401.87158203125, 38.98047637939453, 16.947097778320312, -35.336326599121094, 156.7890167236328, 190.85647583007812, 39.67313003540039, 284.3687744140625, -106.45406341552734, 212.6824951171875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000643.npy"}
|
|
{"epoch": 0.9441997063142438, "step": 644, "batch_size": 64, "mean": 191.19378662109375, "std": 290.8260192871094, "min": -716.1312255859375, "p10": -98.2561279296875, "median": 134.26103973388672, "p90": 618.8516723632813, "max": 940.1785888671875, "pos_frac": 0.765625, "sample": [802.0537109375, -82.2356185913086, 599.8885498046875, -716.1312255859375, 150.73829650878906, 263.9727783203125, -251.48907470703125, -37.30540466308594, 133.99533081054688, 119.494873046875, 37.708091735839844, 102.41572570800781, -95.66869354248047, 390.75115966796875, 22.618148803710938, 622.253662109375, 105.57764434814453, 148.93826293945312, 177.66339111328125, -76.8056411743164, 139.09921264648438, 692.1661376953125, 126.93983459472656, -164.11724853515625, 414.75933837890625, 448.1549377441406, -3.93597412109375, 35.97069549560547, 207.0233154296875, 584.9025268554688, 35.654930114746094, -155.57748413085938, 493.4334411621094, 82.69184875488281, 148.5936737060547, -0.42066001892089844, 940.1785888671875, 111.02597045898438, 134.52674865722656, 218.96878051757812, -51.963356018066406, -99.36502838134766, 344.3127746582031, 17.82122039794922, 429.1716613769531, 220.50747680664062, 63.90126037597656, -113.8208236694336, 89.59739685058594, 692.48583984375, 693.4066772460938, 393.2815246582031, -45.053531646728516, -260.43804931640625, 377.8240966796875, 610.9136962890625, 262.1190185546875, 326.4183349609375, 33.16180419921875, 251.58242797851562, 93.0870132446289, 669.0611572265625, 105.8916244506836, 224.02517700195312], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000644.npy"}
|
|
{"epoch": 0.9456681350954479, "step": 645, "batch_size": 64, "mean": 156.83892822265625, "std": 285.7796325683594, "min": -425.08062744140625, "p10": -159.06338424682613, "median": 86.75946044921875, "p90": 604.1718566894532, "max": 948.9190063476562, "pos_frac": 0.703125, "sample": [11.444000244140625, 436.3579406738281, -183.22720336914062, 614.9672241210938, -328.43927001953125, -73.12339782714844, -297.0244140625, 66.25018310546875, 245.08029174804688, 28.40115737915039, 57.3484992980957, 75.0010986328125, 57.206748962402344, -7.066648483276367, -28.12808609008789, 73.31868743896484, 254.54490661621094, -74.95278930664062, 629.6488037109375, 199.37326049804688, 335.0521240234375, -109.51892852783203, 139.52239990234375, 92.75132751464844, 106.94374084472656, 412.3552551269531, 578.982666015625, 160.1507568359375, -260.7979736328125, -425.08062744140625, 91.74564361572266, 271.12042236328125, -64.69747924804688, 449.8591003417969, 664.6701049804688, -34.001129150390625, 390.72216796875, 551.478271484375, -180.29672241210938, -9.511619567871094, 948.9190063476562, 57.306304931640625, 57.6180419921875, 0.7755928039550781, 104.85346984863281, 374.077880859375, 221.90309143066406, 232.9668426513672, 176.98983764648438, 330.01483154296875, 250.66827392578125, -38.85826110839844, 628.7020874023438, 41.34166717529297, 619.515625, 903.344482421875, 81.77327728271484, 310.74603271484375, -287.18780517578125, 76.3943099975586, 255.03857421875, -11.558156967163086, -107.17839050292969, -108.90521240234375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000645.npy"}
|
|
{"epoch": 0.947136563876652, "step": 646, "batch_size": 64, "mean": 198.30703735351562, "std": 213.97776794433594, "min": -356.131103515625, "p10": -23.925502395629877, "median": 182.91069793701172, "p90": 525.0828521728516, "max": 674.69384765625, "pos_frac": 0.84375, "sample": [584.1679077148438, 469.2567138671875, 185.48538208007812, 363.0885925292969, 379.0240173339844, 180.77967834472656, 643.7868041992188, 413.87884521484375, 674.69384765625, 14.514694213867188, 267.5103454589844, 246.9827423095703, -356.131103515625, 174.79592895507812, 260.2052307128906, 150.96726989746094, 111.93666076660156, 17.23126983642578, 454.70050048828125, 98.84619903564453, 295.4925537109375, 68.00263214111328, -10.830726623535156, 202.18338012695312, 80.61312866210938, 231.85250854492188, 297.4776611328125, 185.04171752929688, -65.83798217773438, 317.628662109375, 142.44610595703125, 203.75396728515625, 9.761444091796875, 267.18438720703125, -40.56562805175781, 216.30873107910156, 51.3988037109375, 471.2543640136719, 114.07691955566406, -161.1700897216797, 244.71583557128906, 671.66943359375, 101.15589904785156, -19.971721649169922, -116.56578063964844, 3.0874481201171875, 175.3889617919922, 194.67236328125, -5.5101165771484375, 550.1927490234375, 536.723388671875, 129.8846435546875, 220.82984924316406, 4.0045623779296875, 497.9216003417969, -25.619979858398438, -87.68772888183594, 62.49980163574219, 188.66990661621094, 160.95443725585938, 49.96643829345703, 611.0457763671875, 86.53142547607422, 245.29702758789062], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000646.npy"}
|
|
{"epoch": 0.9486049926578561, "step": 647, "batch_size": 64, "mean": 203.58551025390625, "std": 248.80775451660156, "min": -446.34527587890625, "p10": -92.45351791381835, "median": 206.0557098388672, "p90": 518.1011077880861, "max": 814.53564453125, "pos_frac": 0.78125, "sample": [296.8074951171875, 489.8002624511719, 212.5341796875, 188.0032501220703, 139.91836547851562, 437.9889221191406, 398.6033630371094, 371.7479553222656, 349.79510498046875, 156.76535034179688, 638.0165405273438, 100.07907104492188, 32.11669921875, 616.7388916015625, 461.0335693359375, 251.09686279296875, 35.29258728027344, 615.279296875, 376.1793212890625, 13.032869338989258, 135.55783081054688, -97.37286376953125, -44.89335632324219, -36.74901580810547, 338.4270935058594, 300.02154541015625, -48.00645446777344, 814.53564453125, -170.3647003173828, 552.4629516601562, 9.544486999511719, -5.959419250488281, 98.12899780273438, 357.32086181640625, 530.2300415039062, 117.47880554199219, 239.7154083251953, 32.81983947753906, -257.28289794921875, -211.47071838378906, 358.3955078125, -39.233497619628906, 589.5531005859375, -165.35504150390625, -185.9281768798828, 92.88996124267578, 361.2530822753906, 199.57723999023438, 92.59427642822266, 345.56024169921875, 121.01020812988281, 104.2181167602539, 478.56829833984375, 246.72592163085938, -40.90730285644531, 309.68963623046875, -446.34527587890625, 260.3761291503906, -80.97504425048828, 398.37335205078125, 321.4337463378906, 488.9871826171875, 122.86659240722656, 261.169921875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000647.npy"}
|
|
{"epoch": 0.9500734214390602, "step": 648, "batch_size": 64, "mean": 298.71856689453125, "std": 349.8816833496094, "min": -366.613037109375, "p10": -85.48562469482418, "median": 257.1703567504883, "p90": 821.728552246094, "max": 1418.6304931640625, "pos_frac": 0.828125, "sample": [231.39498901367188, 56.90757751464844, 73.93114471435547, 307.2173767089844, 183.90826416015625, 409.6447448730469, 770.89013671875, 1087.41943359375, 254.0753936767578, 919.1537475585938, 29.9676513671875, -162.56982421875, 124.00357055664062, 654.85205078125, 169.383544921875, -224.3881378173828, 441.7168884277344, 260.26531982421875, 189.10044860839844, -249.2880859375, -25.010679244995117, 363.4151306152344, 41.032432556152344, 496.4749755859375, -100.98741149902344, 773.3451538085938, 344.09149169921875, 160.40525817871094, 28.233413696289062, 12.708015441894531, 607.4150390625, -0.7068557739257812, 13.112648010253906, 1418.6304931640625, 592.6632080078125, 253.9350128173828, -140.4777374267578, 216.30003356933594, 87.63299560546875, -213.80935668945312, 83.98486328125, 295.5128173828125, 926.2940063476562, 338.60040283203125, 99.81637573242188, 421.75592041015625, 842.4642944335938, -49.22076416015625, 316.09423828125, 281.4550476074219, 534.7037963867188, 306.97918701171875, -49.314788818359375, 287.6495361328125, 629.7230224609375, 93.67424774169922, 386.5722961425781, 567.9730224609375, 159.05462646484375, 880.4208984375, 450.7476501464844, 326.5458068847656, -366.613037109375, 897.1256713867188], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000648.npy"}
|
|
{"epoch": 0.9515418502202643, "step": 649, "batch_size": 64, "mean": 167.37564086914062, "std": 267.22735595703125, "min": -267.0087890625, "p10": -100.14200668334959, "median": 70.16358757019043, "p90": 482.47088928222666, "max": 1126.1282958984375, "pos_frac": 0.671875, "sample": [-59.23652267456055, -26.37682342529297, 193.76254272460938, -54.979705810546875, 337.5451965332031, 1.4047660827636719, 56.79196548461914, 196.87045288085938, 382.9275817871094, 247.627197265625, 398.30401611328125, -83.7969970703125, 53.55797576904297, -24.910903930664062, -107.14701080322266, -68.90402221679688, -16.488998413085938, 400.34698486328125, -193.40530395507812, 279.23248291015625, 246.937255859375, 50.765724182128906, -230.2380828857422, -0.16209983825683594, 266.92144775390625, -200.4279022216797, 495.50848388671875, 332.88568115234375, -31.708293914794922, 299.25103759765625, 342.5464172363281, -134.15274047851562, 16.044189453125, -77.17684173583984, 619.1532592773438, 216.139404296875, 429.2292175292969, 195.66629028320312, 367.1883850097656, 29.28497314453125, 732.8631591796875, 42.89604949951172, 17.177913665771484, 560.525634765625, 382.79107666015625, 217.78738403320312, 437.025146484375, 1126.1282958984375, 241.73355102539062, 83.53520965576172, 11.381917953491211, 89.13996887207031, 8.098831176757812, 553.8172607421875, -61.7470703125, 391.8705749511719, 732.6405029296875, -174.87889099121094, -18.328723907470703, 28.14275550842285, -267.0087890625, -9.77471923828125, 452.0498352050781, -12.606300354003906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000649.npy"}
|
|
{"epoch": 0.9530102790014684, "step": 650, "batch_size": 64, "mean": 249.080322265625, "std": 300.02301025390625, "min": -325.3900146484375, "p10": -197.26286773681636, "median": 243.23587799072266, "p90": 569.2084533691407, "max": 970.5586547851562, "pos_frac": 0.796875, "sample": [-28.89684295654297, -295.41973876953125, 448.7447509765625, 491.29180908203125, 279.0744323730469, 462.1342468261719, 304.3507385253906, -271.9373474121094, 826.1319580078125, 243.1078338623047, 646.9859008789062, -210.16236877441406, 772.8631591796875, 464.58514404296875, 556.8145141601562, 482.2004699707031, 219.7230987548828, 746.1813354492188, 250.2157440185547, 970.5586547851562, 461.02313232421875, 91.53571319580078, 356.5330810546875, 299.9846496582031, 145.07679748535156, 88.14820861816406, 79.27716064453125, 15.781776428222656, 70.35714721679688, 316.1312561035156, 455.77471923828125, 98.05357360839844, -167.16403198242188, -306.89202880859375, 35.94446563720703, 574.5201416015625, 7.016544342041016, 880.5897216796875, -41.03019332885742, 185.4281005859375, 378.8112487792969, 537.052734375, -223.1514892578125, 455.54144287109375, -5.749977111816406, 366.4225158691406, 162.18870544433594, 433.9299621582031, 209.34573364257812, 547.8849487304688, 316.8988342285156, 198.03179931640625, 504.9111328125, 243.36392211914062, 341.8317565917969, 152.85992431640625, 148.14051818847656, -325.3900146484375, -236.23605346679688, 183.59115600585938, 556.72021484375, 62.03473663330078, -2.1286964416503906, -70.40294647216797], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000650.npy"}
|
|
{"epoch": 0.9544787077826725, "step": 651, "batch_size": 64, "mean": 263.5880126953125, "std": 242.4410858154297, "min": -169.773193359375, "p10": -10.510694313049306, "median": 238.51280975341797, "p90": 648.4383850097656, "max": 721.524169921875, "pos_frac": 0.859375, "sample": [480.79095458984375, 355.35174560546875, 671.4487915039062, 640.8648071289062, -85.50543212890625, 83.88607788085938, 218.24493408203125, 493.0641174316406, 99.93415832519531, 102.54908752441406, 466.89471435546875, 565.2102661132812, 191.1196746826172, -0.30236053466796875, 49.57366943359375, 235.15139770507812, 35.97580337524414, 110.34443664550781, -69.20721435546875, 300.078125, 682.9766235351562, -70.62159729003906, 651.6842041015625, 250.24862670898438, 436.74468994140625, 47.641510009765625, -14.88569450378418, -169.773193359375, 117.73681640625, 299.65338134765625, 145.83912658691406, -0.23964691162109375, 272.90155029296875, 130.30572509765625, 721.524169921875, 372.5125732421875, 455.5350341796875, 143.35301208496094, 348.1777648925781, 4.452207565307617, 266.5848388671875, 535.88916015625, -129.9050750732422, 263.60211181640625, 44.68743896484375, 380.92620849609375, 18.706024169921875, 657.7174682617188, 10.269207000732422, 357.56939697265625, 150.666015625, 228.41647338867188, 709.7685546875, 588.00146484375, 241.8742218017578, -83.4837646484375, 214.62083435058594, 14.616687774658203, 608.0899658203125, 289.8977966308594, 473.13800048828125, 450.0210876464844, 651.9241943359375, 154.80035400390625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000651.npy"}
|
|
{"epoch": 0.9559471365638766, "step": 652, "batch_size": 64, "mean": 228.38198852539062, "std": 280.81219482421875, "min": -360.4317626953125, "p10": -61.193504714965805, "median": 161.6780776977539, "p90": 604.1009887695313, "max": 1053.64501953125, "pos_frac": 0.796875, "sample": [229.4524688720703, 218.82064819335938, 558.6707153320312, -360.4317626953125, 57.64015579223633, 303.0672912597656, 288.7823486328125, -76.9798355102539, 513.0864868164062, 159.422607421875, 463.1396484375, 734.177490234375, 668.89453125, -259.42156982421875, 47.03936004638672, -66.55844116210938, 58.72100830078125, 510.320556640625, 781.4332275390625, -44.957862854003906, 238.64891052246094, 123.31600189208984, 88.88721466064453, 122.50189208984375, 144.66038513183594, 25.23931884765625, 315.3238525390625, 101.70893859863281, 163.9335479736328, 148.68569946289062, 72.02183532714844, 1053.64501953125, -136.37875366210938, 612.405517578125, 225.34945678710938, 464.3071594238281, 235.87294006347656, 13.17803955078125, 52.235137939453125, 23.41680908203125, 252.88670349121094, 45.888206481933594, -14.26351547241211, -13.562110900878906, 429.89703369140625, 309.2311706542969, 642.2281494140625, -74.30606842041016, 335.26763916015625, 318.1572570800781, 949.0904541015625, -26.209461212158203, 557.8002319335938, 108.2480697631836, -48.67531967163086, 87.43525695800781, 584.7237548828125, -176.86268615722656, 228.05230712890625, 527.1314697265625, 151.3507080078125, 276.3544921875, 304.3702392578125, -11.065105438232422], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000652.npy"}
|
|
{"epoch": 0.9574155653450808, "step": 653, "batch_size": 64, "mean": 238.23910522460938, "std": 278.5325927734375, "min": -267.7219543457031, "p10": -56.6212837219238, "median": 194.97510528564453, "p90": 615.230810546875, "max": 977.6492309570312, "pos_frac": 0.78125, "sample": [380.80023193359375, 3.8681182861328125, 54.01255798339844, 974.9927368164062, 417.65069580078125, -96.65238189697266, 75.94273376464844, -8.711669921875, -23.903076171875, 489.17852783203125, 655.236083984375, 228.3550262451172, -246.60101318359375, 588.8031005859375, 445.4372253417969, 363.08636474609375, 386.4482727050781, 469.2115783691406, 613.017333984375, 118.162109375, -26.76827621459961, 171.66612243652344, 514.977294921875, 279.5096435546875, 253.24574279785156, 249.52610778808594, 625.2679443359375, 125.24552154541016, 178.59405517578125, -267.7219543457031, 616.179443359375, 218.80714416503906, 455.37188720703125, -241.82418823242188, 221.62002563476562, 5.0049896240234375, 117.65927124023438, 352.17034912109375, -8.042015075683594, 742.1715698242188, 977.6492309570312, 578.4168090820312, -30.644426345825195, -34.55724334716797, 637.7928466796875, 338.55078125, 18.320953369140625, 51.46577453613281, 148.06240844726562, 159.09628295898438, -66.07730102539062, 113.29705810546875, 559.189697265625, 159.40402221679688, 211.3561553955078, 139.6256561279297, -21.92804718017578, 21.89935302734375, 453.653076171875, 212.04115295410156, 236.43814086914062, -129.3287811279297, 164.25112915039062, -121.66716003417969], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000653.npy"}
|
|
{"epoch": 0.9588839941262849, "step": 654, "batch_size": 64, "mean": 210.70352172851562, "std": 281.0909423828125, "min": -260.2887878417969, "p10": -132.40191650390622, "median": 174.73818969726562, "p90": 534.8576477050782, "max": 1169.8082275390625, "pos_frac": 0.75, "sample": [1.2184066772460938, 464.080078125, -112.20684814453125, 283.6332092285156, 149.43434143066406, 19.290054321289062, -166.54470825195312, 215.5994873046875, -29.842021942138672, 491.0305480957031, 201.69924926757812, 112.9271469116211, 177.6065673828125, 150.17266845703125, 141.7490234375, -93.61975860595703, 82.30921173095703, 750.134521484375, 282.4800109863281, 364.4703369140625, 183.14390563964844, 1169.8082275390625, 257.9716796875, 171.389892578125, 407.0470275878906, 301.0523681640625, -141.3656768798828, 334.4219665527344, 36.824615478515625, -66.95608520507812, -66.47880554199219, 462.5812683105469, 353.73138427734375, 240.31112670898438, 35.73326110839844, -10.481071472167969, 668.3281860351562, -144.45816040039062, 122.54280853271484, 266.8283386230469, 605.28759765625, 261.8148193359375, 474.93194580078125, -57.335899353027344, -242.92466735839844, 403.3927917480469, 203.52459716796875, 512.9215087890625, 589.6123046875, 105.86082458496094, 171.86981201171875, 544.2588500976562, 371.1558837890625, 118.07245635986328, 135.56646728515625, -185.34671020507812, 452.4275817871094, -106.24041748046875, 976.4056396484375, 364.6084289550781, 122.40451049804688, -141.05694580078125, -260.2887878417969, -3.4939346313476562], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000654.npy"}
|
|
{"epoch": 0.960352422907489, "step": 655, "batch_size": 64, "mean": 216.68463134765625, "std": 296.0732727050781, "min": -319.1158142089844, "p10": -138.25258483886716, "median": 186.0564727783203, "p90": 711.0082519531252, "max": 934.4561157226562, "pos_frac": 0.765625, "sample": [-245.16036987304688, -243.1352996826172, 741.3499755859375, 153.32986450195312, 34.217689514160156, 851.4088745117188, 190.851318359375, 538.3460083007812, 95.05010986328125, 22.6080265045166, 303.8371276855469, 29.38124656677246, 67.48919677734375, -319.1158142089844, 58.02788543701172, 360.56298828125, -75.71797943115234, 908.563720703125, -24.083511352539062, 576.54736328125, 80.970703125, 231.58132934570312, 555.7322998046875, 184.29562377929688, -8.226144790649414, 164.02642822265625, 274.24090576171875, 188.75234985351562, -144.2758026123047, 206.58261108398438, -92.0500717163086, 753.017333984375, 191.09439086914062, 27.706918716430664, 448.12139892578125, 287.7627258300781, -152.90109252929688, 409.375244140625, 367.853271484375, 418.7048645019531, 189.89581298828125, 54.94470977783203, 355.2149353027344, 934.4561157226562, 732.3538818359375, 140.8450164794922, 180.363037109375, -191.6022186279297, 319.857666015625, 335.75665283203125, -124.19841003417969, -27.77294921875, 34.04339599609375, -43.73768997192383, 661.2017822265625, 3.251697540283203, 785.7567138671875, -152.16529846191406, -28.656539916992188, 397.94793701171875, 187.81732177734375, 211.79469299316406, 136.52313232421875, 357.2017822265625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000655.npy"}
|
|
{"epoch": 0.9618208516886931, "step": 656, "batch_size": 64, "mean": 319.91790771484375, "std": 265.09027099609375, "min": -107.62384796142578, "p10": 10.250581359863306, "median": 282.6455841064453, "p90": 711.6368347167969, "max": 971.478759765625, "pos_frac": 0.890625, "sample": [849.2823486328125, 347.8646545410156, -107.62384796142578, -21.648914337158203, 128.2227783203125, 266.93621826171875, 173.0370635986328, 450.7607421875, 177.6810760498047, 412.6793518066406, 444.529541015625, 324.84112548828125, 284.682373046875, 267.7801818847656, 171.6795654296875, 698.190673828125, 44.12249755859375, 139.6746826171875, 552.9994506835938, 187.3359832763672, 471.20965576171875, -0.063232421875, 679.0977783203125, -95.5599365234375, -75.83078002929688, 128.3377685546875, 601.908447265625, 699.557861328125, 334.69366455078125, 806.4924926757812, 162.6277313232422, 971.478759765625, 280.6087951660156, -37.54690933227539, 873.9425659179688, 478.61419677734375, 353.5877990722656, 349.3086242675781, 742.7244873046875, 146.41708374023438, 879.6787719726562, 86.05192565917969, 531.8564453125, 137.64752197265625, 173.71791076660156, 555.6609497070312, 67.77093505859375, 716.8135375976562, 138.2920684814453, 402.9320373535156, 504.7353515625, 134.14935302734375, -51.676918029785156, 183.54396057128906, 449.05413818359375, 286.1486511230469, 34.31614685058594, 70.35711669921875, 96.83020782470703, 186.61024475097656, 327.6497802734375, 312.2507629394531, 259.65472412109375, 326.06494140625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000656.npy"}
|
|
{"epoch": 0.9632892804698973, "step": 657, "batch_size": 64, "mean": 298.534423828125, "std": 273.57275390625, "min": -124.66024780273438, "p10": 5.521052551269537, "median": 273.1789093017578, "p90": 641.0653564453128, "max": 1127.30029296875, "pos_frac": 0.90625, "sample": [264.8332214355469, 17.21912384033203, 74.1937255859375, 254.0178985595703, 521.1791381835938, 1035.6136474609375, 341.2048034667969, 95.38954162597656, 522.6911010742188, -100.9036865234375, -85.80160522460938, 213.3034210205078, 407.79852294921875, 31.607681274414062, 324.017578125, 431.8260498046875, 37.887359619140625, 339.5625305175781, 94.86263275146484, 682.5757446289062, 674.38916015625, 445.352783203125, 119.1583480834961, 1041.3056640625, 563.309814453125, 724.8580322265625, 126.36451721191406, 159.16864013671875, 446.27447509765625, 353.16455078125, 260.3357238769531, 410.6114501953125, 25.15465545654297, -113.83241271972656, 408.96295166015625, 553.8812255859375, 263.64190673828125, 3.2323074340820312, 141.1724090576172, 177.7178955078125, 244.3634033203125, -60.077728271484375, -10.047647476196289, 260.2528076171875, 798.0621337890625, 83.10417175292969, 166.9672393798828, 340.1044921875, 347.9173583984375, 10.861457824707031, 551.922607421875, 509.822998046875, 305.51739501953125, 423.2745361328125, 281.52459716796875, 158.60525512695312, 394.7528381347656, 1127.30029296875, 290.3233337402344, -124.66024780273438, 49.253929138183594, 75.32221984863281, 295.45623779296875, 298.97808837890625], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000657.npy"}
|
|
{"epoch": 0.9647577092511013, "step": 658, "batch_size": 64, "mean": 228.47642517089844, "std": 229.60252380371094, "min": -282.1343994140625, "p10": -84.6881622314453, "median": 215.4228515625, "p90": 475.98522644042976, "max": 907.1848754882812, "pos_frac": 0.859375, "sample": [125.05755615234375, 167.56072998046875, 93.38815307617188, -89.98005676269531, 16.248828887939453, 105.46087646484375, 401.09014892578125, 175.12876892089844, -72.34040832519531, 382.9336242675781, 350.7880859375, 724.8095703125, 167.9915313720703, 346.9320373535156, 202.16351318359375, -40.84367370605469, 75.37083435058594, 220.46917724609375, 100.46741485595703, -126.63955688476562, 454.7324523925781, 112.8016586303711, 301.484375, 145.54446411132812, -282.1343994140625, 457.724365234375, 191.03317260742188, 347.6800842285156, 504.2327880859375, -116.50556182861328, 36.62825012207031, 68.52850341796875, 45.11349105834961, 518.1365966796875, 256.26025390625, 907.1848754882812, 18.861120223999023, 459.771728515625, 268.3549499511719, 121.95162963867188, 207.54969787597656, 231.37387084960938, 440.2716064453125, 365.0327453613281, 391.9473876953125, 370.02691650390625, 482.9338684082031, 311.7900085449219, 291.1014404296875, 412.3131103515625, 218.8249969482422, 321.96661376953125, -110.1280517578125, 729.1636962890625, 232.93356323242188, 347.6373291015625, 380.13653564453125, 623.1553955078125, -243.641845703125, 212.0207061767578, 147.5856475830078, 133.90164184570312, 97.56829071044922, -116.41577911376953], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000658.npy"}
|
|
{"epoch": 0.9662261380323054, "step": 659, "batch_size": 64, "mean": 228.52227783203125, "std": 257.3695068359375, "min": -230.31381225585938, "p10": -67.09307327270507, "median": 179.06333923339844, "p90": 550.1593139648438, "max": 1009.1029663085938, "pos_frac": 0.78125, "sample": [395.5815124511719, -39.675262451171875, 116.13648986816406, 106.46504211425781, 201.08258056640625, 113.63169860839844, 553.2015380859375, 33.75018310546875, -230.31381225585938, -73.5928955078125, 747.8790283203125, 154.61138916015625, 412.26837158203125, -184.4687957763672, 299.14605712890625, 23.51793670654297, 324.76568603515625, 232.1109619140625, 309.366455078125, 132.14749145507812, 146.2236785888672, 271.727783203125, -127.7945556640625, 1009.1029663085938, -49.240638732910156, 827.8513793945312, 283.1064758300781, -57.923683166503906, 682.00341796875, 328.8807373046875, 227.568359375, 320.493408203125, -1.0736103057861328, 128.4711151123047, 522.0811157226562, 189.13246154785156, 323.6158447265625, 76.1668701171875, 37.450653076171875, 472.0147705078125, -39.28584289550781, -39.90605926513672, 324.3912353515625, -71.02281188964844, 509.50872802734375, -24.17404556274414, 4.404945373535156, 88.97076416015625, 573.3343505859375, 701.7236328125, 543.060791015625, 150.13453674316406, 161.40960693359375, 329.42578125, 266.63671875, 168.9942169189453, 445.03411865234375, 434.3321838378906, 475.6650085449219, -109.73109436035156, 326.23846435546875, 119.72691345214844, -117.076416015625, 166.1593017578125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000659.npy"}
|
|
{"epoch": 0.9676945668135095, "step": 660, "batch_size": 64, "mean": 264.0384521484375, "std": 248.5737762451172, "min": -230.3341064453125, "p10": -47.234814453125, "median": 258.5406799316406, "p90": 541.2628967285157, "max": 1017.748291015625, "pos_frac": 0.859375, "sample": [30.307479858398438, 159.90664672851562, 283.619873046875, 376.2548522949219, 101.64237976074219, 270.5744323730469, 448.849365234375, 307.7862548828125, 352.4376220703125, 412.462890625, 180.87890625, -62.785640716552734, 365.3290100097656, 317.61773681640625, 311.9837646484375, 821.6817626953125, 114.18092346191406, -49.09523010253906, 298.3404541015625, 335.5328063964844, 264.73388671875, 376.49578857421875, 156.77206420898438, 297.3117980957031, 548.9778442382812, 753.80615234375, -58.872108459472656, -96.39717864990234, 385.9485778808594, 384.1899108886719, -36.30921173095703, 663.9693603515625, -149.1751708984375, 926.66162109375, 117.39285278320312, 339.9595642089844, 482.4989929199219, 109.72340393066406, 252.34747314453125, 199.8660125732422, 142.05718994140625, 78.83576965332031, 382.5157775878906, 317.9234924316406, 44.65020751953125, 42.97967529296875, 61.033409118652344, 1017.748291015625, 476.48468017578125, 401.6024169921875, -42.89384460449219, 235.23129272460938, -223.6188201904297, 564.607421875, 523.2613525390625, 242.0199432373047, 189.5545654296875, 201.06661987304688, 248.7347412109375, 404.72882080078125, 248.51617431640625, -230.3341064453125, 65.61054229736328, 208.73797607421875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000660.npy"}
|
|
{"epoch": 0.9691629955947136, "step": 661, "batch_size": 64, "mean": 252.34420776367188, "std": 294.7525634765625, "min": -510.00115966796875, "p10": 0.7663808822631903, "median": 211.79244232177734, "p90": 554.8550415039062, "max": 1423.19287109375, "pos_frac": 0.890625, "sample": [100.28077697753906, 118.52220916748047, 365.0757141113281, 68.25398254394531, 177.96754455566406, 47.4007453918457, 306.5722351074219, -209.45883178710938, 345.8433837890625, 388.66619873046875, 556.8717651367188, 383.13665771484375, 141.81666564941406, 654.834716796875, 1423.19287109375, 198.03546142578125, 155.99954223632812, 9.1593017578125, -86.04263305664062, 65.12835693359375, 173.5765838623047, 457.70599365234375, 104.51324462890625, 186.55621337890625, 76.54940032958984, 254.9146270751953, 140.10581970214844, 405.4032897949219, 744.2737426757812, 39.43113708496094, 90.21241760253906, 550.1493530273438, 18.994781494140625, 631.7468872070312, 266.2630920410156, 137.5809326171875, 268.3071594238281, 239.71646118164062, 195.56427001953125, 132.4857177734375, -211.9563751220703, 258.06268310546875, -2.1298675537109375, 431.04571533203125, 398.5240173339844, -225.7747802734375, 384.0543518066406, 346.9188537597656, 8.679643630981445, 405.91156005859375, 992.0545654296875, -510.00115966796875, 225.54942321777344, -35.86515808105469, 457.0040588378906, 431.1808776855469, 68.73534393310547, 449.4449157714844, 7.524293899536133, 844.139892578125, 411.07708740234375, 374.61785888671875, 288.2449951171875, 27.67957305908203], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000661.npy"}
|
|
{"epoch": 0.9706314243759178, "step": 662, "batch_size": 64, "mean": 206.3653106689453, "std": 272.3310241699219, "min": -392.681396484375, "p10": -208.3612548828125, "median": 200.82430267333984, "p90": 569.4723999023438, "max": 797.8856201171875, "pos_frac": 0.78125, "sample": [61.508338928222656, 193.82122802734375, 688.2897338867188, -392.681396484375, 446.87933349609375, 57.92997741699219, 350.97235107421875, 169.88800048828125, 753.5650634765625, 210.47683715820312, 504.4820251464844, -58.417205810546875, -215.61524963378906, 76.86558532714844, 378.6574401855469, 75.63018035888672, 114.17748260498047, 175.8805389404297, 797.8856201171875, 282.7524108886719, 571.0182495117188, -32.845184326171875, 281.9947204589844, 1.5513267517089844, 176.52197265625, 462.2529296875, 141.8908233642578, 686.4229736328125, 153.42916870117188, -48.8162841796875, 230.28228759765625, -206.91827392578125, 207.82737731933594, -208.97967529296875, 157.16651916503906, 450.836669921875, 239.00225830078125, 398.2630310058594, 424.6922302246094, 354.2359619140625, 36.58119201660156, 173.2808380126953, 541.6595458984375, 597.08154296875, 413.069580078125, 439.6090087890625, -275.72088623046875, 267.9827880859375, 305.1546630859375, 161.4121551513672, 30.514562606811523, 11.17617416381836, -44.86869812011719, 322.39434814453125, -221.59173583984375, -71.4775161743164, 637.8629150390625, 565.8654174804688, 256.5128479003906, 271.4831237792969, 229.45285034179688, -269.2290954589844, -209.8514404296875, -73.75233459472656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000662.npy"}
|
|
{"epoch": 0.9720998531571219, "step": 663, "batch_size": 64, "mean": 245.96690368652344, "std": 285.4202880859375, "min": -484.52996826171875, "p10": -33.500791931152314, "median": 176.00513458251953, "p90": 577.23564453125, "max": 1225.541015625, "pos_frac": 0.875, "sample": [539.1674194335938, 281.85418701171875, 475.61572265625, 47.52064514160156, 303.7412109375, 508.921630859375, 96.21174621582031, 70.64004516601562, -47.140350341796875, 234.03372192382812, 136.6451416015625, 114.91888427734375, 556.1622924804688, 441.1041259765625, 56.07148742675781, 335.9372863769531, 397.544921875, 261.48394775390625, -383.4565734863281, 71.9630126953125, 177.21615600585938, 175.077880859375, -77.373046875, 171.25875854492188, 396.00823974609375, 308.6605224609375, 272.1189880371094, 135.87200927734375, 384.6055908203125, 1225.541015625, -128.6903839111328, 574.2071533203125, 910.4600830078125, 68.30989074707031, 362.3740234375, 876.1658325195312, 578.5335693359375, 95.179931640625, 54.150390625, 570.7666625976562, 626.8846435546875, 173.35671997070312, 124.49029541015625, 173.52508544921875, 245.33935546875, 286.578125, 172.6312713623047, 146.4038848876953, 176.93238830566406, 787.1043701171875, 243.2831268310547, 116.04735565185547, 609.5935668945312, -484.52996826171875, 335.5496826171875, 23.462627410888672, 21.842987060546875, -83.34369659423828, 220.26739501953125, 116.68726348876953, -1.6751556396484375, 44.73979949951172, -63.704864501953125, 101.03226470947266], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000663.npy"}
|
|
{"epoch": 0.973568281938326, "step": 664, "batch_size": 64, "mean": 279.1371765136719, "std": 317.1445007324219, "min": -738.1943969726562, "p10": -66.62538223266601, "median": 221.70770263671875, "p90": 686.7397155761719, "max": 1153.9920654296875, "pos_frac": 0.84375, "sample": [671.0361328125, 146.3683624267578, 258.0192565917969, 687.0205078125, 470.5799865722656, 211.53433227539062, 188.9959716796875, 625.6311645507812, -69.3434066772461, 802.5308227539062, -194.196533203125, 655.6477661132812, 298.52191162109375, 73.30977630615234, 675.2222290039062, 95.9017333984375, -217.30068969726562, 368.45562744140625, 61.47376251220703, 374.81756591796875, 385.6029968261719, 618.8001708984375, -8.48101806640625, 711.3441772460938, -80.93972778320312, 81.6497802734375, 686.0845336914062, 42.652774810791016, 139.5105438232422, -238.45413208007812, 743.4898681640625, 248.98104858398438, -60.2833251953125, 282.55938720703125, 218.88992309570312, 282.5721740722656, -738.1943969726562, 166.19430541992188, 605.09765625, 462.7616271972656, 192.45738220214844, 343.4427795410156, 204.12222290039062, 216.734130859375, 445.71160888671875, 1153.9920654296875, 430.55810546875, 61.09148406982422, 787.2366943359375, 40.15753173828125, -30.935760498046875, 825.8511962890625, 396.541015625, 107.82130432128906, 189.25860595703125, 313.2730407714844, 224.52548217773438, 141.6519317626953, 41.91343688964844, 52.17901611328125, 332.4650573730469, 132.19326782226562, 600.0075073242188, -71.53453826904297], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000664.npy"}
|
|
{"epoch": 0.9750367107195301, "step": 665, "batch_size": 64, "mean": 241.38986206054688, "std": 248.9109649658203, "min": -328.4434814453125, "p10": -30.375462341308584, "median": 186.07070922851562, "p90": 592.9858276367188, "max": 1024.853515625, "pos_frac": 0.859375, "sample": [-67.32024383544922, 46.69442367553711, 286.1510925292969, 150.18215942382812, 129.29476928710938, 398.8788757324219, 474.3302001953125, 95.85397338867188, 597.26708984375, 217.9040985107422, 250.35134887695312, 484.1317138671875, -33.8939208984375, 394.74249267578125, 262.57330322265625, 30.426637649536133, -60.545318603515625, 162.67691040039062, 133.38771057128906, 359.13330078125, 402.9408874511719, -190.20970153808594, 340.7850646972656, 348.3231201171875, -81.32798767089844, 105.71773529052734, -328.4434814453125, 661.9100341796875, 582.9962158203125, -8.99100112915039, 110.76964569091797, 120.79558563232422, 472.1705322265625, 432.1435546875, 601.32177734375, 125.24109649658203, 458.847412109375, 452.0272216796875, 135.1177978515625, 161.01336669921875, 156.92529296875, 666.8153076171875, 446.33807373046875, -232.11160278320312, 203.05465698242188, 702.961181640625, 26.1822509765625, 169.08676147460938, 255.86083984375, 314.91778564453125, 137.0748748779297, 304.2449645996094, 391.1784362792969, 61.701751708984375, 693.750732421875, -22.165725708007812, 103.63459777832031, 22.414573669433594, 139.54205322265625, 354.33154296875, 1024.853515625, 62.92753601074219, 203.16455078125, 46.89706039428711], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000665.npy"}
|
|
{"epoch": 0.9765051395007343, "step": 666, "batch_size": 64, "mean": 242.0909423828125, "std": 235.96170043945312, "min": -333.81829833984375, "p10": -47.208460998535145, "median": 241.78792572021484, "p90": 521.5551879882813, "max": 740.1636962890625, "pos_frac": 0.84375, "sample": [441.6349792480469, 111.61024475097656, 511.53985595703125, 337.7910461425781, -224.7220458984375, 360.14697265625, 525.8474731445312, 404.2784423828125, 101.8230972290039, 707.2005004882812, -22.81536102294922, 129.39462280273438, 212.63809204101562, -72.40790557861328, 200.6407470703125, 43.250030517578125, 392.01275634765625, 7.525627136230469, -52.1221923828125, 152.23843383789062, 578.2554931640625, 136.06039428710938, 394.1115417480469, 261.97625732421875, 740.1636962890625, -35.74308776855469, 431.36602783203125, 224.8250274658203, -21.26020050048828, -333.81829833984375, 318.1481018066406, 302.81170654296875, 157.61981201171875, 717.4205932617188, 469.6977233886719, 279.546630859375, 460.2260437011719, -123.36764526367188, 231.38978576660156, 404.4505310058594, 83.81367492675781, 297.0630798339844, -131.07095336914062, 618.2488403320312, 202.3736114501953, 306.1869812011719, -310.4241638183594, 457.8078918457031, 174.67160034179688, 108.54788208007812, 276.17230224609375, 329.02423095703125, 320.1884460449219, 68.71810913085938, 179.73977661132812, 105.52439880371094, 138.61395263671875, 322.55169677734375, 638.5291748046875, 227.3496856689453, 76.72846221923828, 452.38818359375, 252.18606567382812, 437.5018615722656], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000666.npy"}
|
|
{"epoch": 0.9779735682819384, "step": 667, "batch_size": 64, "mean": 232.80560302734375, "std": 229.7256622314453, "min": -286.883056640625, "p10": -79.34800567626954, "median": 240.34390258789062, "p90": 569.5171325683594, "max": 781.2150268554688, "pos_frac": 0.84375, "sample": [148.42288208007812, -79.63178253173828, 301.0726623535156, -221.62728881835938, 40.880950927734375, 304.205810546875, 250.98220825195312, 294.03265380859375, 323.95916748046875, 458.446533203125, 257.58746337890625, 50.28179931640625, 99.62551879882812, 180.2207489013672, 151.47842407226562, -76.04119110107422, 256.9005126953125, 229.70559692382812, 781.2150268554688, 265.6239013671875, 374.3566589355469, 150.1913604736328, 646.3966064453125, 190.02169799804688, 374.3634338378906, 297.76812744140625, 382.0941467285156, 102.62293243408203, 420.6060485839844, 373.5874938964844, 188.62966918945312, 265.0085144042969, 630.3126831054688, 426.17376708984375, 223.09756469726562, -85.76895141601562, 50.08148193359375, 177.6895294189453, 356.6151123046875, -78.68585968017578, -128.3437957763672, 724.341064453125, 611.19140625, 6.258689880371094, 576.4652709960938, 553.3048095703125, 275.088623046875, 105.11907196044922, 202.68603515625, 411.6952209472656, -22.87921905517578, 288.4588928222656, -286.883056640625, 353.388671875, 705.4906005859375, 171.6944580078125, 80.82776641845703, 304.78753662109375, -91.8987045288086, 129.04281616210938, -127.85185241699219, 74.67477416992188, 173.80734252929688, 326.58880615234375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000667.npy"}
|
|
{"epoch": 0.9794419970631424, "step": 668, "batch_size": 64, "mean": 212.354736328125, "std": 276.0991516113281, "min": -334.6983337402344, "p10": -124.4728630065918, "median": 171.6298599243164, "p90": 564.8216186523438, "max": 939.644775390625, "pos_frac": 0.796875, "sample": [398.0672912597656, 216.6494903564453, 939.644775390625, -334.26275634765625, 5.134838104248047, 351.2242431640625, 451.2064514160156, 96.10624694824219, 572.748291015625, 158.6671142578125, 66.23713684082031, -124.93452453613281, 128.49746704101562, 105.6607666015625, 306.3447265625, 414.2418212890625, 192.4550323486328, 32.60594177246094, -123.3956527709961, 42.4637565612793, -157.11822509765625, 464.1271057128906, -111.6475601196289, 428.50592041015625, 212.62423706054688, 123.1150131225586, 83.37570190429688, 126.01083374023438, 22.351974487304688, 456.9512939453125, 192.64520263671875, 307.6453857421875, 241.6358642578125, -129.16464233398438, 546.3260498046875, 164.41468811035156, -334.6983337402344, 448.73016357421875, 137.427001953125, -8.459035873413086, 340.81976318359375, 346.11236572265625, 642.3932495117188, 68.65650177001953, -322.2290344238281, 630.6200561523438, 178.84503173828125, -161.44203186035156, 160.16009521484375, -4.411771774291992, 314.18243408203125, 101.95537567138672, 197.4736328125, 262.0373840332031, -25.770906448364258, 545.6891479492188, 312.5295104980469, 317.30279541015625, 938.9030151367188, 138.1287384033203, -5.272574424743652, 712.9674072265625, 49.123382568359375, 741.768310546875], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000668.npy"}
|
|
{"epoch": 0.9809104258443465, "step": 669, "batch_size": 64, "mean": 265.5224609375, "std": 272.4607849121094, "min": -461.7530517578125, "p10": -64.85935058593749, "median": 271.2186737060547, "p90": 591.155731201172, "max": 1133.668212890625, "pos_frac": 0.859375, "sample": [206.6100616455078, 43.51123809814453, -50.944297790527344, 467.33087158203125, -270.19964599609375, 461.12646484375, 323.8836669921875, 418.9882507324219, 91.78795623779297, -70.82294464111328, 1133.668212890625, -115.70123291015625, 20.299697875976562, 296.8303527832031, -86.7003402709961, 272.4897155761719, 670.58349609375, -79.5202865600586, 117.96637725830078, 235.0238037109375, 426.8753662109375, 331.2054443359375, 828.0081787109375, 282.67645263671875, 188.1133270263672, 630.6362915039062, 87.56552124023438, 515.3222045898438, 608.6207885742188, 89.20939636230469, 303.6943359375, 709.4791870117188, 147.0472412109375, 543.490234375, 650.9202270507812, -409.6404113769531, 72.43115234375, 352.85760498046875, 500.2279968261719, 334.71881103515625, 325.5941162109375, 209.3055419921875, 89.3890380859375, 232.10009765625, -461.7530517578125, 409.0458068847656, 260.02569580078125, 550.4039306640625, 299.056640625, 234.87619018554688, -17.481422424316406, 345.3750915527344, 165.4959716796875, 229.05535888671875, 320.22772216796875, 382.4839172363281, 50.533546447753906, 244.32150268554688, 410.2123107910156, 484.4236755371094, 144.99832153320312, 384.9645690917969, 151.1641845703125, 269.9476318359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000669.npy"}
|
|
{"epoch": 0.9823788546255506, "step": 670, "batch_size": 64, "mean": 282.030517578125, "std": 278.4869079589844, "min": -300.3878479003906, "p10": -36.69942626953125, "median": 246.7250747680664, "p90": 697.1295166015625, "max": 905.79638671875, "pos_frac": 0.84375, "sample": [148.56488037109375, 327.96075439453125, 229.72622680664062, 683.0615844726562, 295.9770202636719, 15.775579452514648, -212.8212890625, 40.31282043457031, 437.0357666015625, 610.6295776367188, 189.6051483154297, 36.962188720703125, -300.3878479003906, 779.8634643554688, 497.21673583984375, 816.8607788085938, 458.52740478515625, 243.2823486328125, 555.6094970703125, 180.35133361816406, 178.8764190673828, -240.19918823242188, 793.5457153320312, 27.48082733154297, 327.01641845703125, 316.693603515625, 241.55133056640625, 115.16854858398438, 277.26580810546875, -35.515655517578125, 601.295166015625, 591.2570190429688, -75.23817443847656, 288.8497314453125, 250.1678009033203, 764.2672119140625, 483.25469970703125, -3.748729705810547, 340.3656921386719, 530.6577758789062, 3.757251739501953, 703.1586303710938, 240.6295623779297, -53.77667236328125, 905.79638671875, 355.12823486328125, 398.86639404296875, 241.74281311035156, -166.40040588378906, -2.3573532104492188, 139.91030883789062, 803.864013671875, 161.98089599609375, 101.58697509765625, 372.0513916015625, 109.1756820678711, 221.35504150390625, 486.36639404296875, 63.262603759765625, 350.46380615234375, 144.45957946777344, 426.7286376953125, -37.206756591796875, 272.2851867675781], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000670.npy"}
|
|
{"epoch": 0.9838472834067548, "step": 671, "batch_size": 64, "mean": 237.013427734375, "std": 268.9710693359375, "min": -352.05096435546875, "p10": -121.46142501831048, "median": 249.69773864746094, "p90": 599.4309692382815, "max": 893.4774780273438, "pos_frac": 0.8125, "sample": [3.5111541748046875, -262.69439697265625, 197.374267578125, 182.57342529296875, -11.503875732421875, 446.490966796875, 188.2496337890625, 307.58465576171875, 243.82699584960938, 502.66119384765625, 279.39697265625, 17.476341247558594, 262.8634033203125, 359.2227478027344, 398.98187255859375, 3.8815460205078125, 75.84065246582031, 216.3157958984375, 474.09259033203125, 298.97979736328125, -32.46044158935547, 893.4774780273438, -352.05096435546875, 684.5289306640625, 532.7763061523438, 149.1837921142578, 699.9739379882812, 356.1932067871094, 323.37078857421875, 332.7122802734375, 74.51820373535156, -60.271095275878906, -192.70550537109375, 284.3621520996094, 443.66375732421875, -147.68585205078125, 309.2564697265625, 201.00259399414062, 442.9925537109375, 439.47698974609375, 526.81298828125, 191.07733154296875, 29.204055786132812, 147.77288818359375, 414.4837646484375, 60.06358337402344, 266.4078063964844, 758.32861328125, -169.13003540039062, -28.84918975830078, 410.70391845703125, 342.8551940917969, -3.8018436431884766, 33.205078125, 435.1233825683594, 108.30702209472656, -218.92129516601562, -182.51348876953125, 109.68232727050781, 632.1112060546875, 77.73253631591797, 777.1971435546875, 627.9972534179688, 255.5684814453125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000671.npy"}
|
|
{"epoch": 0.9853157121879589, "step": 672, "batch_size": 64, "mean": 215.81500244140625, "std": 283.2297668457031, "min": -413.027587890625, "p10": -143.48916015624997, "median": 216.3318862915039, "p90": 592.1544006347656, "max": 868.0442504882812, "pos_frac": 0.796875, "sample": [24.021467208862305, -67.46517944335938, -161.88882446289062, 160.43453979492188, -30.380023956298828, 23.447708129882812, 439.3092041015625, 304.81768798828125, 600.91796875, 86.60037994384766, 29.260643005371094, 118.64629364013672, -239.06997680664062, 47.71192932128906, 266.0321960449219, -29.965843200683594, 89.55624389648438, -222.26478576660156, 374.6539306640625, 429.0358581542969, 80.78086853027344, 591.3319091796875, 26.415443420410156, -100.55661010742188, 493.56500244140625, -46.89374542236328, 800.8974609375, -188.05203247070312, 227.02728271484375, 868.0442504882812, -69.51455688476562, 697.16357421875, 645.3271484375, 38.20794677734375, 592.5068969726562, 369.2347412109375, 229.4750518798828, 327.62933349609375, 293.4340515136719, 543.29443359375, -203.96453857421875, 42.84534454345703, 238.9473419189453, 466.6678161621094, 325.42816162109375, 518.8118286132812, 104.8805923461914, 158.17071533203125, 315.95184326171875, 119.7467041015625, 435.3931579589844, 369.7231140136719, 7.6729736328125, 392.7266540527344, 83.35784912109375, 58.85665512084961, 379.3525390625, 271.43878173828125, 205.63648986816406, -413.027587890625, 381.8023986816406, 462.99560546875, 772.4530639648438, -346.4385681152344], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000672.npy"}
|
|
{"epoch": 0.986784140969163, "step": 673, "batch_size": 64, "mean": 265.8677673339844, "std": 224.85189819335938, "min": -192.05209350585938, "p10": -19.77282199859619, "median": 263.7817077636719, "p90": 543.3391540527344, "max": 806.3888549804688, "pos_frac": 0.859375, "sample": [-110.65513610839844, 35.229488372802734, 178.1804962158203, 403.30413818359375, 117.22602081298828, 408.63934326171875, 317.02630615234375, 194.69923400878906, 552.9183349609375, 299.88690185546875, -130.4984893798828, 549.4437255859375, 296.7585144042969, 684.615966796875, 262.708740234375, 424.03582763671875, -155.0318603515625, 323.0085754394531, 806.3888549804688, 276.2427062988281, 173.3117218017578, 215.60702514648438, 264.352783203125, -49.908721923828125, 374.665771484375, -54.853553771972656, 210.6092529296875, 376.79656982421875, 20.107728958129883, 80.88584899902344, 324.2274169921875, 483.0633239746094, 203.22071838378906, 210.70838928222656, 299.9827880859375, 654.5770263671875, 356.7620849609375, 45.32194519042969, 529.0951538085938, 337.9769592285156, 386.2662353515625, -10.785482406616211, 484.6481628417969, 427.93951416015625, 170.4776611328125, 69.77091217041016, 78.12825012207031, -16.25465202331543, 381.5356750488281, 802.2858276367188, 160.9244384765625, 116.84370422363281, 417.9390563964844, 448.16162109375, -192.05209350585938, 402.5201416015625, 263.21063232421875, 478.68414306640625, 220.516845703125, -21.280609130859375, 50.708717346191406, 195.457275390625, 670.6251220703125, 238.6273193359375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000673.npy"}
|
|
{"epoch": 0.9882525697503671, "step": 674, "batch_size": 64, "mean": 269.932861328125, "std": 266.0668029785156, "min": -209.64620971679688, "p10": -16.098453140258776, "median": 255.0480499267578, "p90": 587.173260498047, "max": 1295.146484375, "pos_frac": 0.875, "sample": [73.91995239257812, 40.29349899291992, 405.14263916015625, 336.86456298828125, 279.0663146972656, 450.62506103515625, 123.09492492675781, 124.83338928222656, -126.7547378540039, 640.4900512695312, 379.2837829589844, 668.8525390625, -189.21510314941406, 598.69921875, -82.35111999511719, 128.45265197753906, 18.7014217376709, 828.986328125, 347.5986633300781, 1295.146484375, 240.8539581298828, 221.28152465820312, -5.427295684814453, 556.0137329101562, 247.38290405273438, 107.56390380859375, 430.63751220703125, 359.0823974609375, 304.3702392578125, 69.32919311523438, 275.88494873046875, 790.5963745117188, 72.32217407226562, 307.7918701171875, 291.44281005859375, 527.8219604492188, -100.60614013671875, -20.67180633544922, 102.16413879394531, 513.4453735351562, 220.1328125, 216.59793090820312, 399.0635986328125, 356.1776123046875, 28.287702560424805, 612.4881591796875, 392.9297790527344, 349.9378662109375, 155.8024139404297, 262.71319580078125, -184.63983154296875, 57.59145736694336, 545.0531616210938, -209.64620971679688, 98.64432525634766, 370.423095703125, 214.68658447265625, 560.2793579101562, 319.8052673339844, 331.88275146484375, 246.17318725585938, 14.246049880981445, 150.84381103515625, 133.21922302246094], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000674.npy"}
|
|
{"epoch": 0.9897209985315712, "step": 675, "batch_size": 64, "mean": 285.0060119628906, "std": 282.5855712890625, "min": -157.06028747558594, "p10": -59.74741973876951, "median": 261.28326416015625, "p90": 663.8172729492189, "max": 1254.214599609375, "pos_frac": 0.84375, "sample": [117.76722717285156, 301.9661865234375, 381.7137756347656, 153.00726318359375, 161.59228515625, 151.8976287841797, 620.2445678710938, -4.405265808105469, 271.4002990722656, 295.13140869140625, 487.94110107421875, -145.67999267578125, 473.9291687011719, 732.4926147460938, 6.421272277832031, -18.700927734375, 9.6494140625, 595.345703125, 207.11297607421875, 236.96939086914062, 310.4345397949219, 62.93498992919922, 1254.214599609375, 428.1067810058594, 68.0732650756836, 188.81179809570312, -124.93218994140625, 394.962158203125, 300.13714599609375, 175.32603454589844, 354.14202880859375, 458.806396484375, 326.04425048828125, 519.6654052734375, 27.510406494140625, 363.4578857421875, 851.1870727539062, -71.02546691894531, 629.5377807617188, 745.3017578125, 127.14338684082031, -76.83840942382812, 20.48720359802246, 269.5066833496094, 324.121826171875, -33.431976318359375, -157.06028747558594, 271.73712158203125, 438.71875, 372.7413024902344, 199.58816528320312, 601.516845703125, 233.06576538085938, -85.86431884765625, -143.58200073242188, 253.05984497070312, 193.5081024169922, 371.3363952636719, 233.62232971191406, 678.5084838867188, 21.965301513671875, 868.2181396484375, 767.796875, 192.02590942382812], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000675.npy"}
|
|
{"epoch": 0.9911894273127754, "step": 676, "batch_size": 64, "mean": 247.96432495117188, "std": 244.14044189453125, "min": -307.80364990234375, "p10": -25.393661499023402, "median": 236.3093032836914, "p90": 515.5293762207032, "max": 810.2727661132812, "pos_frac": 0.890625, "sample": [251.21429443359375, 513.2889404296875, 399.77081298828125, 385.14501953125, 348.70721435546875, 13.476249694824219, 461.2806091308594, 266.6604309082031, -40.135711669921875, -307.80364990234375, 45.3861083984375, 47.739898681640625, 38.29899597167969, 439.3036193847656, 516.4895629882812, 809.8932495117188, 218.81887817382812, 208.4778594970703, 126.39207458496094, -46.96639633178711, 50.63001251220703, 51.22142028808594, 417.7322082519531, 9.00445556640625, 25.085006713867188, 426.8599853515625, 90.25906372070312, 260.59674072265625, 162.51693725585938, 762.1464233398438, 84.28640747070312, 376.07470703125, 198.99359130859375, 470.3646545410156, -54.40691375732422, 143.62033081054688, 589.4461669921875, 243.8194122314453, 293.4734802246094, 810.2727661132812, 376.684326171875, 72.7699966430664, 62.52922058105469, 150.32972717285156, 435.2112731933594, -151.85635375976562, 28.702178955078125, 487.78082275390625, 159.1048583984375, 163.96307373046875, 675.1681518554688, 387.3248291015625, 379.9228820800781, 321.2841796875, 42.14849090576172, 368.44000244140625, 634.40869140625, -306.04461669921875, 458.9604187011719, 63.487335205078125, 228.7991943359375, 356.495849609375, -69.68452453613281, 436.3516845703125], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000676.npy"}
|
|
{"epoch": 0.9926578560939795, "step": 677, "batch_size": 64, "mean": 236.56594848632812, "std": 218.70281982421875, "min": -267.72259521484375, "p10": -64.60383110046386, "median": 224.1123504638672, "p90": 484.30384521484376, "max": 722.966796875, "pos_frac": 0.796875, "sample": [-37.50279998779297, 722.966796875, 223.37265014648438, 113.20601654052734, 205.074951171875, 160.4437255859375, -98.01786041259766, -54.606056213378906, 481.8486328125, -68.8798828125, 172.79058837890625, 383.71551513671875, 42.92100524902344, 465.08038330078125, 208.92298889160156, 465.467041015625, 464.8545837402344, -81.28511047363281, 72.47361755371094, 217.6376190185547, 465.4997863769531, 377.9934387207031, 476.51934814453125, -76.06861877441406, 231.6060791015625, 453.57147216796875, 316.765625, 312.0777282714844, 308.0213623046875, 524.9681396484375, 199.78651428222656, 166.22879028320312, -42.76671600341797, 196.7449951171875, 310.0002136230469, -267.72259521484375, 547.880126953125, 181.43336486816406, 289.1688232421875, 247.81658935546875, 351.4200744628906, 224.85205078125, 430.57391357421875, 485.3560791015625, -18.23914337158203, 438.734619140625, 165.197509765625, 88.9423828125, 649.1761474609375, 18.348892211914062, 655.904541015625, -18.24759292602539, 318.6486511230469, 125.60384368896484, 454.48992919921875, 194.13583374023438, -120.1046142578125, 520.7255249023438, 330.4085388183594, 308.351806640625, 306.3679504394531, -68.19058227539062, -56.234745025634766, 73.98968505859375], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000677.npy"}
|
|
{"epoch": 0.9941262848751835, "step": 678, "batch_size": 64, "mean": 260.56817626953125, "std": 285.618896484375, "min": -481.6930847167969, "p10": -71.63183670043944, "median": 284.1085510253906, "p90": 638.1245971679688, "max": 922.625244140625, "pos_frac": 0.84375, "sample": [54.829322814941406, 644.1114501953125, 370.6681823730469, 379.74407958984375, 212.27842712402344, -77.04135131835938, 413.8230895996094, -481.6930847167969, 57.2010498046875, 103.79562377929688, -58.75563049316406, 512.6263427734375, 80.73286437988281, 624.1552734375, 51.22357940673828, 283.9869384765625, 401.97528076171875, -270.264404296875, 346.8260498046875, 656.9063110351562, 572.3861083984375, 87.78797912597656, 8.467472076416016, 818.8646240234375, 527.6458129882812, 826.8414916992188, 284.23016357421875, 2.9952850341796875, -12.921398162841797, 582.6002197265625, -134.31143188476562, -95.68106842041016, 52.38043212890625, 456.7525634765625, 99.9012451171875, 541.84326171875, 101.89508819580078, 727.6570434570312, 207.84414672851562, 374.16424560546875, -127.35487365722656, 320.47088623046875, 474.96258544921875, 463.306396484375, 291.8704528808594, 68.45634460449219, 707.0478515625, 922.625244140625, 150.98602294921875, 337.60137939453125, 388.92132568359375, -100.96842193603516, 339.6335144042969, 14.425735473632812, 47.94202423095703, 22.24107551574707, 155.30406188964844, 314.228759765625, 310.54693603515625, 222.95758056640625, 400.50616455078125, 72.94894409179688, 598.2392578125, -59.00963592529297], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000678.npy"}
|
|
{"epoch": 0.9955947136563876, "step": 679, "batch_size": 64, "mean": 216.64544677734375, "std": 261.2093811035156, "min": -324.7715759277344, "p10": -41.5997158050537, "median": 212.07521057128906, "p90": 530.5122619628906, "max": 1079.395263671875, "pos_frac": 0.78125, "sample": [389.99444580078125, 48.230079650878906, 275.75115966796875, 669.2465209960938, 71.31478881835938, 441.1845703125, 440.27880859375, 527.100830078125, 494.48712158203125, 224.99969482421875, 237.40283203125, 377.2016906738281, 194.5159149169922, 90.48905944824219, 743.3951416015625, -54.881004333496094, 55.143760681152344, -24.453948974609375, 230.13697814941406, 595.1840209960938, 330.86773681640625, 791.613037109375, -18.671463012695312, 3.133678436279297, 127.98713684082031, -223.84133911132812, 169.42352294921875, 419.63958740234375, -29.106433868408203, 214.092041015625, 240.00051879882812, 51.54681396484375, -46.9539794921875, 332.9891357421875, 210.05838012695312, 57.37030029296875, 592.1383056640625, 125.66342163085938, -18.114717483520508, 26.39569854736328, -9.155426025390625, -7.3498687744140625, 232.77389526367188, 111.86068725585938, 316.4366760253906, 451.24261474609375, 42.31452178955078, -28.608917236328125, -324.7715759277344, 22.28030014038086, 531.9743041992188, -259.24664306640625, 152.14950561523438, 9.394268035888672, 374.12493896484375, 291.2935791015625, 286.9952392578125, 299.4150390625, 392.4686584472656, 1079.395263671875, 252.39193725585938, 466.59710693359375, -128.7605438232422, -72.86089324951172], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000679.npy"}
|
|
{"epoch": 0.9970631424375918, "step": 680, "batch_size": 64, "mean": 252.61300659179688, "std": 296.1508483886719, "min": -740.4317626953125, "p10": -20.253433990478513, "median": 190.3773956298828, "p90": 609.4927368164062, "max": 1125.0469970703125, "pos_frac": 0.875, "sample": [614.0689086914062, 1.5096549987792969, 246.3198699951172, 465.94281005859375, 116.2767333984375, 440.0617980957031, 889.8616943359375, 283.69757080078125, 451.6877746582031, 199.37673950195312, 258.014892578125, 174.25624084472656, 167.29356384277344, -21.340057373046875, 487.2887268066406, 839.9564208984375, 240.6794891357422, -232.7093505859375, 11.487106323242188, 204.10342407226562, 818.8157348632812, 109.36832427978516, 12.440235137939453, 88.94341278076172, -17.717979431152344, 55.97096252441406, 114.40126037597656, 113.40462493896484, 370.4658203125, 417.89044189453125, 167.4065704345703, 181.3780517578125, 132.625, 573.9075927734375, 457.54986572265625, 105.58253479003906, 26.339500427246094, 487.227294921875, 454.08343505859375, -190.22796630859375, 164.22286987304688, 700.2611083984375, 308.40362548828125, 247.01800537109375, 139.58108520507812, -740.4317626953125, 538.9873046875, -40.12641143798828, 22.36977767944336, 598.8150024414062, 263.5599365234375, 1125.0469970703125, 12.521377563476562, 56.99143981933594, 146.90667724609375, 320.36016845703125, 433.96063232421875, 714.8766479492188, -24.749820709228516, 159.30584716796875, -62.257659912109375, 86.96271514892578, 206.02767944335938, 470.9308166503906], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000680.npy"}
|
|
{"epoch": 0.9985315712187959, "step": 681, "batch_size": 64, "mean": 226.318603515625, "std": 274.0923156738281, "min": -251.24725341796875, "p10": -104.45186386108398, "median": 220.95140075683594, "p90": 584.4341064453126, "max": 985.624755859375, "pos_frac": 0.765625, "sample": [115.53529357910156, -239.84228515625, 300.50750732421875, 182.99114990234375, 528.640380859375, -23.746437072753906, 358.656982421875, 304.9083557128906, -106.1684799194336, 214.59893798828125, -71.62274169921875, 355.72064208984375, -123.74613952636719, -100.44642639160156, 344.8153076171875, 985.624755859375, 237.63966369628906, -213.32730102539062, -74.90324401855469, 459.687744140625, 62.79241180419922, 0.8087234497070312, 1.1489944458007812, 157.8443603515625, 460.485107421875, 310.7039794921875, 227.30386352539062, 654.3934326171875, 395.8841552734375, 382.79345703125, 182.96261596679688, 354.10479736328125, 279.60760498046875, 80.07864379882812, -53.72435760498047, 197.27818298339844, 33.474952697753906, -251.24725341796875, 410.4859619140625, -88.90916442871094, -165.24937438964844, 826.2860107421875, 72.17681121826172, 427.27972412109375, -193.89163208007812, 93.63406372070312, -40.18684387207031, 155.6608428955078, 109.72846984863281, -31.791648864746094, 542.20654296875, 161.23049926757812, 481.28460693359375, 347.9662170410156, 285.82086181640625, 289.04815673828125, 614.3590698242188, 593.9277954101562, 562.2821655273438, 312.9087829589844, 88.09040069580078, 673.1627197265625, 794.2462768554688, 250.41636657714844], "npy": "/scratch/qu.yang1/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-helpful-4xh200-batch-64-q_t-0.5-s_star-0.6/margin_logs/step_0000681.npy"}
|