Model: W-61/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851 Source: Original Platform
662 lines
1.1 MiB
662 lines
1.1 MiB
{"epoch": 0.0, "step": 1, "batch_size": 64, "mean": -0.0013527870178222656, "std": 0.2564818859100342, "min": -0.736083984375, "p10": -0.3432229995727539, "median": 0.038166046142578125, "p90": 0.29227676391601565, "max": 0.645111083984375, "pos_frac": 0.578125, "sample": [0.1120758056640625, 0.12518310546875, 0.31621551513671875, 0.13765716552734375, -0.12592506408691406, 0.23141098022460938, -0.21887779235839844, 0.21950721740722656, 0.04480743408203125, 0.020877838134765625, 0.0570220947265625, 0.058269500732421875, -0.4338226318359375, -0.030628204345703125, 0.645111083984375, -0.395477294921875, 0.09050941467285156, 0.0007190704345703125, -0.34615325927734375, 0.016077041625976562, -0.33638572692871094, 0.293853759765625, 0.17610931396484375, 0.22386932373046875, 0.21470260620117188, -0.08536529541015625, 0.0907745361328125, -0.03816986083984375, 0.39190101623535156, 0.16336441040039062, 0.08024787902832031, -0.031158447265625, 0.08477020263671875, 0.002460479736328125, -0.242034912109375, 0.07232666015625, -0.60186767578125, 0.20531463623046875, 0.155731201171875, -0.14299774169921875, -0.25698089599609375, 0.12331962585449219, -0.26497650146484375, 0.15140533447265625, -0.0920257568359375, -0.18599319458007812, 0.19028091430664062, 0.2496490478515625, 0.42162322998046875, 0.17873382568359375, -0.1525421142578125, -0.4972076416015625, 0.32010650634765625, -0.10365867614746094, -0.233795166015625, -0.19828224182128906, -0.4018898010253906, -0.13407135009765625, -0.09596633911132812, 0.031524658203125, 0.28859710693359375, -0.192962646484375, -0.736083984375, 0.3026123046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000001.npy"}
|
|
{"epoch": 0.0015117157974300832, "step": 2, "batch_size": 64, "mean": 0.03744968771934509, "std": 0.2875921130180359, "min": -0.7604827880859375, "p10": -0.2812448501586914, "median": 0.03963661193847656, "p90": 0.3654294967651367, "max": 0.8134727478027344, "pos_frac": 0.5625, "sample": [0.30594635009765625, -0.24289894104003906, -0.11509323120117188, -0.13417816162109375, 0.06942558288574219, 0.36568641662597656, -0.14640045166015625, 0.1497650146484375, 0.30261993408203125, 0.10124588012695312, 0.13028717041015625, -0.0031890869140625, 0.0361480712890625, 0.5662612915039062, 0.09694290161132812, -0.01091766357421875, 0.1128997802734375, 0.0411834716796875, -0.21860504150390625, -0.1236419677734375, -0.08812713623046875, 0.10360527038574219, 0.1790008544921875, -0.5114288330078125, 0.3056755065917969, -0.14553451538085938, 0.28168487548828125, 0.26990509033203125, 0.1686878204345703, 0.038089752197265625, 0.19541168212890625, -0.10783576965332031, -0.2644004821777344, -0.19707489013671875, -0.140472412109375, 0.1349811553955078, 0.19672012329101562, -0.0714111328125, 0.53369140625, 0.1271820068359375, 0.8134727478027344, 0.2990264892578125, -0.7604827880859375, -0.08274078369140625, 0.05890846252441406, 0.029361724853515625, 0.4510040283203125, -0.1599273681640625, -0.29346656799316406, 0.10005569458007812, -0.27509117126464844, -0.1937713623046875, 0.19167327880859375, 0.28173065185546875, -0.09406471252441406, -0.3380699157714844, -0.29186248779296875, 0.36483001708984375, 0.009979248046875, 0.44391632080078125, -0.126708984375, -0.6550216674804688, 0.6160736083984375, -0.28388214111328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000002.npy"}
|
|
{"epoch": 0.0030234315948601664, "step": 3, "batch_size": 64, "mean": -0.00468367338180542, "std": 0.31731459498405457, "min": -0.7940406799316406, "p10": -0.35069503784179684, "median": -0.0012054443359375, "p90": 0.3970069885253909, "max": 0.867828369140625, "pos_frac": 0.484375, "sample": [0.011333465576171875, 0.030282974243164062, -0.095489501953125, -0.5462646484375, 0.2355194091796875, -0.0477294921875, 0.022510528564453125, 0.0534515380859375, 0.463409423828125, -0.25984954833984375, -0.3600654602050781, -0.025848388671875, -0.1795501708984375, -0.03455162048339844, 0.015472412109375, 0.25701904296875, 0.04119873046875, -0.2780609130859375, -0.000392913818359375, -0.4187164306640625, -0.6969451904296875, -0.10743904113769531, -0.07501220703125, 0.03575897216796875, -0.118072509765625, 0.45575523376464844, 0.7013092041015625, 0.867828369140625, -0.09910202026367188, 0.1788787841796875, 0.055606842041015625, -0.3288307189941406, 0.05745697021484375, -0.2378692626953125, -0.079437255859375, -0.002017974853515625, -0.035247802734375, 0.0213775634765625, 0.19405174255371094, 0.42784881591796875, 0.6042556762695312, -0.26770782470703125, -0.3101654052734375, 0.623260498046875, 0.19634246826171875, 0.22673797607421875, -0.12972640991210938, 0.32180023193359375, 0.18674087524414062, -0.6961822509765625, -0.7940406799316406, 0.20313262939453125, 0.325042724609375, 0.31229591369628906, 0.07370376586914062, -0.01328277587890625, 0.0140533447265625, -0.44873046875, -0.18184661865234375, 0.03936004638671875, -0.25455474853515625, -0.15972137451171875, -0.1046905517578125, -0.16540908813476562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000003.npy"}
|
|
{"epoch": 0.0045351473922902496, "step": 4, "batch_size": 64, "mean": -0.006213396787643433, "std": 0.35557445883750916, "min": -0.8669891357421875, "p10": -0.46047286987304686, "median": -0.010596275329589844, "p90": 0.3808746337890625, "max": 0.9934368133544922, "pos_frac": 0.46875, "sample": [0.241485595703125, 0.30228424072265625, -0.1314544677734375, -0.09781646728515625, 0.5660247802734375, -0.5127105712890625, 0.30121803283691406, 0.9934368133544922, 0.38158416748046875, 0.1392364501953125, 0.12959671020507812, 0.21739768981933594, -0.2494354248046875, 0.1019439697265625, -0.44571685791015625, -0.4265899658203125, 0.37921905517578125, -0.0621185302734375, 0.7480621337890625, -0.8669891357421875, -0.28536224365234375, -0.10849380493164062, -0.12432861328125, -0.50634765625, 0.20138168334960938, 0.1246490478515625, 0.012935638427734375, 0.3427085876464844, -0.3942832946777344, -0.002655029296875, 0.80242919921875, 0.17810821533203125, 0.04363250732421875, -0.247894287109375, -0.6033935546875, 0.694305419921875, 0.322998046875, -0.2712516784667969, 0.018035888671875, -0.48909759521484375, -0.08678817749023438, -0.012477874755859375, 0.12348175048828125, -0.4059162139892578, -0.0338897705078125, 0.05853271484375, -0.6009979248046875, -0.008714675903320312, -0.466796875, 0.01045989990234375, 0.1928558349609375, -0.060535430908203125, -0.21191787719726562, -0.050930023193359375, -0.233856201171875, 0.1461944580078125, -0.3114776611328125, 0.3930397033691406, -0.06331634521484375, -0.3429832458496094, 0.11978912353515625, -0.068878173828125, -0.15631484985351562, 0.2570457458496094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000004.npy"}
|
|
{"epoch": 0.006046863189720333, "step": 5, "batch_size": 64, "mean": -0.02648913860321045, "std": 0.32702869176864624, "min": -0.7305183410644531, "p10": -0.4004495620727539, "median": -0.041484832763671875, "p90": 0.32037200927734377, "max": 0.9317855834960938, "pos_frac": 0.421875, "sample": [-0.07032203674316406, -0.38759613037109375, -0.0039920806884765625, 0.3170166015625, -0.15374755859375, -0.0605010986328125, -0.002727508544921875, 0.26721954345703125, 0.44210052490234375, 0.2739105224609375, -0.01987457275390625, -0.453277587890625, -0.25772857666015625, 0.3873176574707031, -0.30548858642578125, -0.078948974609375, 0.05206298828125, -0.19632911682128906, 0.02587890625, -0.17833328247070312, 0.11361885070800781, -0.21454429626464844, 0.2035961151123047, -0.048492431640625, -0.72552490234375, 0.1196746826171875, -0.5057220458984375, -0.4059581756591797, 0.3185882568359375, -0.061611175537109375, -0.0255126953125, -0.0482330322265625, -0.13184356689453125, 0.11541748046875, 0.0875701904296875, -0.6821060180664062, -0.12910842895507812, -0.30272674560546875, 0.05990791320800781, -0.08150672912597656, 0.13130950927734375, -0.280487060546875, 0.11067962646484375, -0.03551483154296875, 0.35715484619140625, -0.09682464599609375, 0.321136474609375, 0.0104522705078125, 0.894561767578125, -0.19712066650390625, 0.08791351318359375, -0.714019775390625, 0.14661407470703125, 0.9317855834960938, 0.090545654296875, -0.3436622619628906, -0.7305183410644531, -0.2026081085205078, -0.1290149688720703, 0.04925537109375, 0.1278705596923828, -0.047454833984375, -0.14904403686523438, 0.7195625305175781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000005.npy"}
|
|
{"epoch": 0.007558578987150416, "step": 6, "batch_size": 64, "mean": -0.07354229688644409, "std": 0.31148895621299744, "min": -0.7928009033203125, "p10": -0.5221153259277344, "median": -0.0594482421875, "p90": 0.2326023101806641, "max": 0.9345703125, "pos_frac": 0.40625, "sample": [-0.45786285400390625, -0.2069835662841797, 0.3841552734375, 0.15856170654296875, -0.12569427490234375, -0.6103096008300781, -0.1622295379638672, 0.32080078125, -0.2120513916015625, -0.13692283630371094, -0.1454296112060547, -0.549652099609375, -0.3060264587402344, 0.202972412109375, -0.063873291015625, -0.022430419921875, 0.42256927490234375, 0.12262725830078125, 0.152191162109375, -0.5605716705322266, -0.189208984375, -0.7928009033203125, -0.08258247375488281, 0.23739051818847656, 0.08416366577148438, 0.18061065673828125, 0.18927764892578125, -0.1493682861328125, -0.029836654663085938, 0.04573631286621094, 0.1600818634033203, -0.09673309326171875, 0.1609954833984375, 0.1561126708984375, -0.39080810546875, -0.4414520263671875, -0.041553497314453125, -0.4160919189453125, -0.16351318359375, 2.288818359375e-05, -0.376129150390625, 0.05112266540527344, -0.29988861083984375, -0.000858306884765625, -0.42612457275390625, -0.04520416259765625, 0.9345703125, 0.08606719970703125, -0.1262683868408203, 0.0523834228515625, -0.084381103515625, 0.3872833251953125, -0.6162586212158203, -0.30864906311035156, 0.1693878173828125, 0.22142982482910156, -0.13423728942871094, -0.278533935546875, -0.6173095703125, -0.560150146484375, -0.055023193359375, 0.014982223510742188, 0.4824562072753906, 0.19834327697753906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000006.npy"}
|
|
{"epoch": 0.009070294784580499, "step": 7, "batch_size": 64, "mean": -0.03844526410102844, "std": 0.25930771231651306, "min": -0.5935745239257812, "p10": -0.3589138031005859, "median": 0.008675575256347656, "p90": 0.30114593505859394, "max": 0.5105133056640625, "pos_frac": 0.515625, "sample": [-0.2519989013671875, 0.22034454345703125, 0.1512298583984375, 0.03883934020996094, 0.19370079040527344, -0.02310943603515625, 0.08989906311035156, -0.13698577880859375, -0.18621826171875, 0.5105133056640625, 0.051891326904296875, -0.29338645935058594, 0.08709907531738281, 0.042034149169921875, -0.00446319580078125, -0.4337177276611328, -0.06892776489257812, -0.3162345886230469, 0.09468841552734375, -0.32074737548828125, 0.021753311157226562, 0.1857452392578125, -0.24735260009765625, 0.046539306640625, 0.0180511474609375, 0.17964935302734375, -0.1125030517578125, 0.31884765625, 0.3471832275390625, 0.008069992065429688, -0.28205108642578125, 0.194580078125, -0.3354206085205078, 0.4213371276855469, -0.4839286804199219, 0.35297393798828125, 0.15360641479492188, 0.4216270446777344, 0.10123443603515625, -0.3770599365234375, -0.23355865478515625, 0.034709930419921875, 0.2598419189453125, -0.4749908447265625, -0.1978759765625, -0.286041259765625, -0.5935745239257812, 0.079681396484375, -0.142120361328125, 0.1530303955078125, -0.282745361328125, -0.1489715576171875, 0.2543010711669922, -0.20046615600585938, 0.06293487548828125, 0.13567352294921875, -0.28993988037109375, -0.487701416015625, -0.2808837890625, 0.4710845947265625, -0.19965362548828125, -0.11086273193359375, 0.009281158447265625, -0.36898231506347656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000007.npy"}
|
|
{"epoch": 0.010582010582010581, "step": 8, "batch_size": 64, "mean": 0.06668540835380554, "std": 0.3351237177848816, "min": -0.5222015380859375, "p10": -0.31957015991210935, "median": 0.000240325927734375, "p90": 0.5464920043945314, "max": 1.0146713256835938, "pos_frac": 0.5, "sample": [0.20256614685058594, 0.4048919677734375, 0.509246826171875, -7.62939453125e-06, -0.27590179443359375, -0.19362640380859375, 0.6476821899414062, 0.381195068359375, -0.14661407470703125, 0.263397216796875, 0.01116943359375, 0.104248046875, -0.4301433563232422, -0.06893539428710938, 0.5624542236328125, 0.87420654296875, 1.0146713256835938, 0.14599227905273438, -0.3622093200683594, 0.5730743408203125, -0.011257171630859375, -0.07152557373046875, -0.46829986572265625, -0.118743896484375, 0.1429920196533203, -0.10405540466308594, -0.2324657440185547, -0.07450103759765625, 0.08556365966796875, -0.11370086669921875, 0.26548004150390625, -0.15313720703125, -0.16637802124023438, 0.039066314697265625, -0.184356689453125, -0.1909637451171875, 0.07366561889648438, 0.4477996826171875, 0.765411376953125, 0.1695098876953125, 0.6576347351074219, -0.07622909545898438, 0.17974853515625, -0.315399169921875, -0.07636260986328125, -0.18442153930664062, 0.3635578155517578, 0.00048828125, -0.3381996154785156, 0.11507415771484375, -0.0010204315185546875, 0.16038894653320312, -0.13591384887695312, -0.19343948364257812, -0.5222015380859375, 0.38179779052734375, -0.2820568084716797, -0.32135772705078125, 0.23077392578125, 0.15734481811523438, -0.05356788635253906, 0.48950958251953125, 0.1478424072265625, -0.43358612060546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000008.npy"}
|
|
{"epoch": 0.012093726379440665, "step": 9, "batch_size": 64, "mean": -0.004131674766540527, "std": 0.3330223560333252, "min": -0.9198150634765625, "p10": -0.4280975341796875, "median": 0.003814697265625, "p90": 0.418209457397461, "max": 0.9535980224609375, "pos_frac": 0.515625, "sample": [-0.35575103759765625, 0.14667510986328125, -0.12858200073242188, 0.0643768310546875, -0.23634910583496094, -0.395263671875, -0.193115234375, -0.21985626220703125, 0.1835174560546875, -0.6475830078125, -0.442169189453125, -0.4427509307861328, 0.6022300720214844, -0.001079559326171875, 0.330657958984375, -0.4557647705078125, -0.1879119873046875, 0.16864013671875, 0.167236328125, 0.039226531982421875, 0.3125, 0.04129791259765625, 0.9535980224609375, -0.05611419677734375, -0.2380237579345703, -0.60430908203125, 0.002437591552734375, -0.0098419189453125, -0.08281707763671875, 0.021572113037109375, 0.05364227294921875, -0.07262420654296875, 0.080810546875, 0.682281494140625, -0.0040760040283203125, -0.17667579650878906, 0.16152572631835938, -0.1498565673828125, -0.6642608642578125, 0.233001708984375, 0.4246673583984375, -0.0232696533203125, -0.2038726806640625, -0.2725982666015625, -0.0843963623046875, 0.06406402587890625, 0.08257102966308594, 0.3963146209716797, -0.2718849182128906, -0.35511016845703125, 0.3025321960449219, -0.003021240234375, 0.005191802978515625, 0.5213165283203125, 0.4031410217285156, 0.4897308349609375, 0.07580184936523438, -0.12332916259765625, 0.0838165283203125, 0.09641075134277344, 0.428985595703125, -0.9198150634765625, 0.025341033935546875, 0.1125335693359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000009.npy"}
|
|
{"epoch": 0.013605442176870748, "step": 10, "batch_size": 64, "mean": -0.015134930610656738, "std": 0.40784332156181335, "min": -1.0358123779296875, "p10": -0.3925758361816406, "median": -0.0268096923828125, "p90": 0.4113906860351563, "max": 1.3630523681640625, "pos_frac": 0.421875, "sample": [-0.2853240966796875, -0.15361404418945312, -0.103302001953125, 0.09737777709960938, 0.215667724609375, -0.3195953369140625, -0.37099456787109375, 0.11525726318359375, -0.0674591064453125, 0.03911590576171875, -0.05179595947265625, -0.07426071166992188, 0.38802337646484375, -0.02294921875, -0.030670166015625, -0.23767852783203125, 0.053600311279296875, -0.401824951171875, 0.83709716796875, -0.12808609008789062, -0.015336990356445312, -0.05182647705078125, 0.155517578125, 0.2101593017578125, -0.021465301513671875, -0.010593414306640625, -0.1154632568359375, 0.016754150390625, 0.23600006103515625, -0.334228515625, -0.14240264892578125, 0.6298446655273438, -0.17425537109375, -0.20183563232421875, 0.12382125854492188, 0.13738250732421875, -0.8232040405273438, 0.2862434387207031, 0.62139892578125, 0.08923912048339844, -0.24962806701660156, -0.14648056030273438, -1.0358123779296875, -0.4269866943359375, -0.3203582763671875, 0.029338836669921875, -0.9081535339355469, -0.2684326171875, 1.3630523681640625, -0.619232177734375, -0.1999664306640625, 0.43320465087890625, -0.21651077270507812, 0.38794708251953125, -0.2841796875, -0.00881195068359375, 1.0786285400390625, 0.15249252319335938, 0.1762542724609375, -0.326812744140625, 0.15581321716308594, -0.6607589721679688, 0.3932647705078125, 0.419158935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000010.npy"}
|
|
{"epoch": 0.015117157974300832, "step": 11, "batch_size": 64, "mean": -0.005576759576797485, "std": 0.26791393756866455, "min": -0.7640838623046875, "p10": -0.36079559326171873, "median": 0.0015010833740234375, "p90": 0.33624229431152347, "max": 0.6376953125, "pos_frac": 0.5, "sample": [0.03420257568359375, -0.020586013793945312, -0.15644454956054688, -0.178070068359375, -0.18603515625, 0.08254241943359375, -0.049591064453125, -0.04944801330566406, -0.27081298828125, -0.0839080810546875, -0.182098388671875, 0.3202247619628906, 0.04662322998046875, 0.2142047882080078, 0.09922218322753906, -0.26317596435546875, 0.3515357971191406, 0.4010009765625, -0.08758544921875, 0.19519805908203125, 0.3323783874511719, -0.089813232421875, 0.6376953125, 0.22054290771484375, -0.03545951843261719, -0.4628143310546875, 0.04691314697265625, 0.02555084228515625, 0.26852989196777344, -0.37950897216796875, -0.09507369995117188, 0.11500167846679688, -0.12656021118164062, -0.24626922607421875, 0.023588180541992188, 0.16846466064453125, -0.21074295043945312, 0.23089599609375, 0.1839141845703125, 0.039928436279296875, -0.5554656982421875, 0.33029937744140625, -0.4204864501953125, -0.11130523681640625, -0.11965179443359375, -0.7640838623046875, -0.4223747253417969, 0.162567138671875, -0.0812530517578125, 0.33789825439453125, 0.4230079650878906, -0.400146484375, -0.19832611083984375, 0.3847999572753906, 0.11502838134765625, -0.293182373046875, 0.07537269592285156, 0.25940895080566406, 0.056812286376953125, -0.0686187744140625, -0.31713104248046875, -0.1808624267578125, 0.1883697509765625, 0.3782501220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000011.npy"}
|
|
{"epoch": 0.016628873771730914, "step": 12, "batch_size": 64, "mean": -0.00015756487846374512, "std": 0.2934369444847107, "min": -0.958251953125, "p10": -0.38398895263671873, "median": 0.03884124755859375, "p90": 0.36255912780761723, "max": 0.5716590881347656, "pos_frac": 0.515625, "sample": [-0.10160064697265625, -0.40128326416015625, 0.14442062377929688, 0.0711212158203125, -0.2520275115966797, -0.958251953125, 0.21175193786621094, -0.115142822265625, -0.042232513427734375, -0.025806427001953125, -0.34363555908203125, -0.07899093627929688, 0.0775909423828125, -0.55743408203125, 0.2360820770263672, -0.42237091064453125, -0.1390533447265625, -0.18092727661132812, 0.17742156982421875, 0.5482349395751953, -0.07355499267578125, 0.0394744873046875, -0.295654296875, 0.336578369140625, 0.16231346130371094, -0.466461181640625, -0.09753990173339844, -0.05001068115234375, 0.05739593505859375, 0.19071388244628906, -0.15275955200195312, 0.1127471923828125, -0.24333572387695312, 0.1175994873046875, -0.1822052001953125, 0.06019020080566406, -0.45757293701171875, 0.5716590881347656, 0.0382080078125, 0.05467987060546875, 0.5362319946289062, 0.1490306854248047, -0.25042724609375, -0.16815948486328125, 0.3549461364746094, 0.4515666961669922, 0.36582183837890625, 0.06941986083984375, 0.48903656005859375, -0.21004867553710938, 0.2882843017578125, -0.024129867553710938, 0.16222381591796875, -0.002349853515625, -0.1397552490234375, 0.06641387939453125, -0.5369911193847656, 0.428741455078125, -0.3374652862548828, 0.17611122131347656, -0.00678253173828125, 0.11578750610351562, 0.1607666015625, 0.28131103515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000012.npy"}
|
|
{"epoch": 0.018140589569160998, "step": 13, "batch_size": 64, "mean": -0.027852505445480347, "std": 0.32535678148269653, "min": -1.052947998046875, "p10": -0.39260940551757806, "median": 0.010778427124023438, "p90": 0.32614021301269536, "max": 0.5130729675292969, "pos_frac": 0.515625, "sample": [-1.052947998046875, 0.04265594482421875, 0.3706207275390625, 0.26663970947265625, -0.035247802734375, -0.8120079040527344, 0.002079010009765625, 0.2206439971923828, 0.1655731201171875, 0.12006378173828125, 0.4244060516357422, 0.1243438720703125, 0.01947784423828125, -0.34009552001953125, 0.4816017150878906, 0.38010597229003906, -0.172271728515625, -0.15473556518554688, -0.2101593017578125, -0.0060272216796875, -0.2563018798828125, -0.42902374267578125, -0.0629730224609375, 0.06732177734375, 0.2965812683105469, -0.128753662109375, -0.007968902587890625, -0.28650474548339844, -0.1024322509765625, -0.03723907470703125, -0.05061912536621094, 0.10501861572265625, 0.5130729675292969, -0.8355026245117188, 0.12128448486328125, -0.022434234619140625, 0.2557525634765625, 0.3163642883300781, 0.1425495147705078, 0.02869415283203125, -0.0287933349609375, 0.17322540283203125, -0.6929931640625, -0.3029670715332031, -0.4151153564453125, 0.09209632873535156, 0.19069290161132812, -0.06082916259765625, -0.148773193359375, 0.33032989501953125, 0.38677215576171875, -0.17899322509765625, 0.05615425109863281, 0.09639167785644531, 0.05529022216796875, -0.85150146484375, -0.245147705078125, 0.2709999084472656, 0.18424224853515625, 0.2642993927001953, 0.22411346435546875, -0.285919189453125, -0.06820297241210938, -0.2895374298095703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000013.npy"}
|
|
{"epoch": 0.019652305366591082, "step": 14, "batch_size": 64, "mean": 0.07302093505859375, "std": 0.3871995210647583, "min": -0.81695556640625, "p10": -0.2837486267089843, "median": 0.03735542297363281, "p90": 0.5161396026611336, "max": 1.7301864624023438, "pos_frac": 0.515625, "sample": [0.029903411865234375, -0.683074951171875, 0.04480743408203125, 0.6260833740234375, -0.06399154663085938, 0.1654682159423828, -0.0739288330078125, -0.307861328125, 0.10157012939453125, 0.3231830596923828, -0.1244964599609375, 0.28278350830078125, -0.019388198852539062, -0.22748565673828125, 0.11287307739257812, 0.046962738037109375, -0.08075332641601562, 0.244384765625, 0.205718994140625, -0.074462890625, -0.013498306274414062, -0.6332855224609375, -0.14321327209472656, 0.11583709716796875, 0.1428680419921875, -0.22681427001953125, -0.81695556640625, 0.19788360595703125, 0.22991371154785156, 1.7301864624023438, 0.18878936767578125, 1.1290740966796875, 0.21857452392578125, 0.58935546875, 0.18030929565429688, -0.1789703369140625, 0.3453025817871094, -0.0344696044921875, -0.18554115295410156, -0.02886962890625, 0.31170654296875, 0.7167472839355469, -0.21268463134765625, 0.69287109375, 0.060916900634765625, -0.012804031372070312, 0.05078887939453125, 0.3162841796875, 0.6072158813476562, -0.13611412048339844, -0.41033935546875, -0.051624298095703125, 0.2122211456298828, -0.4990386962890625, -0.029508590698242188, 0.20486068725585938, -0.1137847900390625, -0.058559417724609375, -0.09650993347167969, -0.33924293518066406, 0.09830665588378906, -0.0647735595703125, -0.14026260375976562, 0.23189544677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000014.npy"}
|
|
{"epoch": 0.021164021164021163, "step": 15, "batch_size": 64, "mean": 0.0031641721725463867, "std": 0.427504301071167, "min": -1.21038818359375, "p10": -0.6025375366210937, "median": 0.05574607849121094, "p90": 0.46403350830078127, "max": 1.1126632690429688, "pos_frac": 0.5625, "sample": [0.24779891967773438, -1.0634307861328125, 0.03230857849121094, 0.10051155090332031, -0.20520591735839844, 0.6331863403320312, -0.4096527099609375, -0.07310867309570312, 0.489898681640625, -0.75274658203125, 0.21201324462890625, -0.06875038146972656, 0.421722412109375, -0.04363822937011719, 0.7346649169921875, 0.10947036743164062, 1.1126632690429688, -0.25659942626953125, 0.1398468017578125, -0.7610931396484375, 0.0691375732421875, -0.2706642150878906, -0.15784454345703125, -0.0828399658203125, 0.03248786926269531, -0.6406631469726562, 0.17813873291015625, 0.4820556640625, 0.0971221923828125, 0.368988037109375, 0.092620849609375, -0.12863731384277344, -0.61932373046875, -0.1654205322265625, 0.13521957397460938, 0.34847259521484375, 0.14948272705078125, 0.1838226318359375, -0.001312255859375, -0.043731689453125, 0.259124755859375, 0.15826416015625, -0.9384994506835938, 0.8010330200195312, 0.3324909210205078, -1.21038818359375, 0.4213409423828125, 0.10752105712890625, 0.07135391235351562, -0.5633697509765625, -0.488861083984375, 0.03884315490722656, -0.2172393798828125, -0.01300048828125, -0.14228057861328125, -0.1701202392578125, 0.46075439453125, -0.2910041809082031, 0.10786628723144531, 0.14002609252929688, 0.4306488037109375, 0.4654388427734375, -0.2267608642578125, 0.042354583740234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000015.npy"}
|
|
{"epoch": 0.022675736961451247, "step": 16, "batch_size": 64, "mean": 0.003903329372406006, "std": 0.2404957413673401, "min": -0.5147743225097656, "p10": -0.26294670104980467, "median": -0.01056671142578125, "p90": 0.33955497741699225, "max": 0.5726432800292969, "pos_frac": 0.46875, "sample": [0.06932640075683594, -0.0106964111328125, -0.0567779541015625, 0.04941749572753906, 0.038112640380859375, -0.06711578369140625, -0.16828346252441406, -0.35430908203125, 0.20729827880859375, -0.1503467559814453, -0.03533172607421875, 0.14865875244140625, -0.10956573486328125, -0.258575439453125, 0.1352977752685547, 0.16581344604492188, 0.4362468719482422, 0.029193878173828125, -0.5147743225097656, 0.076324462890625, -0.17813873291015625, -0.4711761474609375, 0.40155029296875, 0.5726432800292969, -0.01697540283203125, -0.2928619384765625, 0.3469123840332031, 0.0749664306640625, 0.04504203796386719, 0.3802642822265625, 0.23919677734375, 0.2547760009765625, -0.2381744384765625, -0.20261764526367188, -0.01043701171875, 0.30144500732421875, 0.457122802734375, 0.3223876953125, -0.2648200988769531, -0.17180633544921875, -0.0076923370361328125, 0.3074226379394531, -0.148895263671875, 0.3154335021972656, 0.362823486328125, 0.02989959716796875, 0.17852401733398438, -0.201385498046875, 0.0279693603515625, -0.2548484802246094, -0.038173675537109375, -0.07989311218261719, -0.054195404052734375, -0.23654556274414062, 0.098541259765625, 0.12332344055175781, -0.2969970703125, -0.4165802001953125, -0.2122344970703125, 0.14655494689941406, -0.016077041625976562, -0.06630325317382812, -0.238677978515625, -0.2513923645019531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000016.npy"}
|
|
{"epoch": 0.02418745275888133, "step": 17, "batch_size": 64, "mean": 0.021797776222229004, "std": 0.34068769216537476, "min": -1.3570404052734375, "p10": -0.3081157684326172, "median": -0.014655113220214844, "p90": 0.3985074996948244, "max": 0.7150650024414062, "pos_frac": 0.484375, "sample": [0.055164337158203125, -0.2806243896484375, -0.034271240234375, -0.0047397613525390625, -0.455963134765625, 0.41608619689941406, -0.102630615234375, 9.72747802734375e-05, 0.00054931640625, -0.0873260498046875, 0.3120269775390625, -0.23966217041015625, -0.21393775939941406, 0.06660842895507812, -0.3005561828613281, -0.397613525390625, -0.2421112060546875, -0.11806869506835938, 0.04920768737792969, -1.3570404052734375, 0.33809661865234375, 0.2892265319824219, -0.199737548828125, 0.13701629638671875, -0.141815185546875, 0.16379928588867188, 0.601654052734375, -0.032726287841796875, -0.24801063537597656, -0.12908172607421875, -0.49875640869140625, -0.34265899658203125, 0.35749053955078125, 0.011524200439453125, 0.6868972778320312, 0.632232666015625, 0.256622314453125, 0.24747467041015625, -0.29915618896484375, 0.354248046875, -0.22857666015625, -0.111663818359375, 0.3327178955078125, 0.11769866943359375, 0.11591148376464844, -0.05556488037109375, -0.1215667724609375, 0.22747802734375, 0.3354301452636719, 0.677337646484375, -0.024570465087890625, -0.3113555908203125, 0.10874176025390625, -0.03798103332519531, 0.688140869140625, 0.07065200805664062, -0.06867790222167969, -0.030057907104492188, 0.7150650024414062, -0.0986175537109375, 0.10076141357421875, 0.2337646484375, -0.3148689270019531, -0.17467498779296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000017.npy"}
|
|
{"epoch": 0.025699168556311415, "step": 18, "batch_size": 64, "mean": 0.05831030011177063, "std": 0.2819819450378418, "min": -0.9905166625976562, "p10": -0.3010814666748047, "median": 0.07980537414550781, "p90": 0.3515539169311524, "max": 0.6141281127929688, "pos_frac": 0.609375, "sample": [0.07690811157226562, 0.19686508178710938, 0.0247955322265625, 0.18976593017578125, 0.21698379516601562, 0.329833984375, 0.02838134765625, 0.00553131103515625, 0.5944900512695312, 0.2447662353515625, 0.21544456481933594, -0.272735595703125, 0.11037635803222656, -0.1537151336669922, -0.10927581787109375, 0.44205474853515625, 0.040966033935546875, -0.08679962158203125, 0.06793594360351562, 0.08270263671875, 0.20600509643554688, 0.3584747314453125, -0.4502716064453125, 0.11809539794921875, 0.23053741455078125, -0.03652191162109375, -0.28778076171875, 0.2129669189453125, 0.16245651245117188, 0.2527580261230469, 0.09940528869628906, 0.09279632568359375, -0.12688446044921875, 0.1775360107421875, -0.20987892150878906, 0.19985198974609375, -0.14941787719726562, 0.06453132629394531, 0.29611968994140625, -0.000335693359375, 0.5031967163085938, 0.2771148681640625, -0.3970947265625, -0.32891845703125, -0.10985565185546875, 0.6141281127929688, -0.13964271545410156, 0.3354053497314453, -0.003021240234375, -0.9905166625976562, 0.1902313232421875, -0.09373283386230469, -0.2169628143310547, 0.25798797607421875, 0.3038749694824219, -0.00232696533203125, -0.06830406188964844, -0.3067817687988281, 0.2877349853515625, -0.041019439697265625, -0.3397026062011719, -0.4453277587890625, 0.42479705810546875, 0.5648765563964844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000018.npy"}
|
|
{"epoch": 0.027210884353741496, "step": 19, "batch_size": 64, "mean": 0.08134375512599945, "std": 0.2967988848686218, "min": -0.796112060546875, "p10": -0.3221328735351562, "median": 0.06918716430664062, "p90": 0.44999465942382816, "max": 0.7204437255859375, "pos_frac": 0.65625, "sample": [0.593505859375, 0.15483856201171875, -0.4719505310058594, 0.0439453125, 0.2350006103515625, 0.3219108581542969, -0.172607421875, 0.2211761474609375, -0.796112060546875, 0.2955322265625, -0.06689071655273438, 0.076416015625, 0.4464111328125, -0.19182395935058594, 0.1506500244140625, 0.15369606018066406, 0.016721725463867188, -0.023324966430664062, 0.008968353271484375, 0.3874378204345703, 0.1358642578125, -0.35150146484375, -0.060489654541015625, -0.18048095703125, 0.2130279541015625, 0.30925750732421875, -0.5130615234375, 0.090118408203125, -0.4886627197265625, 0.12191390991210938, 0.7204437255859375, 0.0045928955078125, -0.3837127685546875, 0.23468017578125, -0.283050537109375, -0.12353515625, 0.45153045654296875, 0.1493053436279297, 0.3557586669921875, -0.027187347412109375, 0.011163711547851562, -0.041110992431640625, -0.016815185546875, 0.5157546997070312, 0.06195831298828125, 0.2604637145996094, 0.02739715576171875, -0.1475677490234375, -0.0070037841796875, 0.20711517333984375, 0.5802993774414062, -0.08572006225585938, 0.4762725830078125, 0.005597114562988281, 0.16143798828125, 0.3518524169921875, -0.16681671142578125, -0.3388824462890625, 0.12180900573730469, 0.43682861328125, 0.024370193481445312, 0.012065887451171875, 0.6870651245117188, 0.3101539611816406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000019.npy"}
|
|
{"epoch": 0.02872260015117158, "step": 20, "batch_size": 64, "mean": -0.005038082599639893, "std": 0.27014678716659546, "min": -0.5095710754394531, "p10": -0.34262619018554685, "median": -0.00762176513671875, "p90": 0.375326156616211, "max": 0.5809860229492188, "pos_frac": 0.484375, "sample": [0.15015411376953125, 0.1956787109375, 0.17173004150390625, -0.35986328125, 0.09720230102539062, -0.43607330322265625, -0.5095710754394531, 0.496917724609375, 0.23474884033203125, -0.06548309326171875, 0.23938751220703125, 0.026554107666015625, -0.2519073486328125, -0.06083106994628906, -0.2050018310546875, -0.2960796356201172, -0.19832611083984375, 0.06411170959472656, -0.0222320556640625, -0.4496917724609375, 0.23925209045410156, 0.06820106506347656, -0.3883075714111328, -0.06473541259765625, 0.13339805603027344, -0.2272491455078125, 0.4697990417480469, -0.3224945068359375, 0.05328369140625, 0.057494163513183594, -0.223358154296875, -0.2988777160644531, -0.07171630859375, -0.15934181213378906, -0.507843017578125, 0.11190414428710938, 0.08234024047851562, 0.3921966552734375, -0.049304962158203125, 0.37987518310546875, 0.06995582580566406, -0.3346099853515625, -0.14646148681640625, -0.2275238037109375, -0.01059722900390625, 0.14691162109375, 0.102752685546875, -0.00464630126953125, 0.28403472900390625, -0.235321044921875, -0.34606170654296875, 0.5809860229492188, -0.02899932861328125, 0.48911285400390625, 0.01349639892578125, -0.21792984008789062, 0.5548362731933594, 0.2841644287109375, -0.2220306396484375, 0.30919456481933594, -0.02053070068359375, 0.3647117614746094, -0.2253894805908203, 0.0015668869018554688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000020.npy"}
|
|
{"epoch": 0.030234315948601664, "step": 21, "batch_size": 64, "mean": -0.019357353448867798, "std": 0.3326614499092102, "min": -0.9317855834960938, "p10": -0.3563129425048828, "median": -0.06764984130859375, "p90": 0.4134042739868167, "max": 1.044342041015625, "pos_frac": 0.421875, "sample": [0.021387100219726562, -0.24616050720214844, 0.06481742858886719, -0.5063934326171875, 0.440582275390625, -0.2122650146484375, 0.08281517028808594, 0.2035198211669922, 0.018707275390625, -0.3003692626953125, -0.22818374633789062, -0.06941604614257812, 0.066864013671875, 0.17748260498046875, 0.05168914794921875, 0.6319847106933594, 0.0857086181640625, 0.5490798950195312, -0.3610038757324219, -0.06783866882324219, -0.20855712890625, 0.07846260070800781, 0.06270408630371094, -0.23381805419921875, -0.158843994140625, -0.41412353515625, -0.19611549377441406, 0.7961883544921875, 0.7692337036132812, -0.10039138793945312, 1.044342041015625, -0.058315277099609375, -0.22589111328125, -0.06170082092285156, -0.2511749267578125, 0.3499889373779297, 0.11016845703125, 0.2244415283203125, -0.345367431640625, -0.08218574523925781, 0.44258880615234375, -0.9317855834960938, -0.32147216796875, 0.2749595642089844, 0.28504180908203125, -0.18441009521484375, -0.2686576843261719, 0.030628204345703125, -0.4034843444824219, -0.06558418273925781, -0.1376800537109375, 0.292510986328125, -0.019437789916992188, -0.12922286987304688, -0.16484451293945312, -0.483001708984375, -0.1390399932861328, -0.06746101379394531, 0.09545516967773438, -0.32415771484375, -0.16495895385742188, -0.1999969482421875, 0.21111297607421875, -0.3680248260498047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000021.npy"}
|
|
{"epoch": 0.031746031746031744, "step": 22, "batch_size": 64, "mean": -0.016939282417297363, "std": 0.22836941480636597, "min": -0.5416889190673828, "p10": -0.28940353393554685, "median": 0.028264999389648438, "p90": 0.26851463317871094, "max": 0.5019454956054688, "pos_frac": 0.5625, "sample": [-0.18316650390625, -0.23733139038085938, -0.28913116455078125, -0.06669998168945312, 0.2018413543701172, 0.4569587707519531, 0.07350349426269531, 0.032318115234375, 0.08617973327636719, 0.11555862426757812, 0.2501182556152344, -0.2943878173828125, -0.08322334289550781, 0.041576385498046875, -0.14306259155273438, 0.08774566650390625, -0.286651611328125, -0.11426925659179688, 0.052459716796875, -0.2703857421875, 0.0148468017578125, 0.08551025390625, 0.18662261962890625, 0.40303993225097656, 0.08217239379882812, 0.0372772216796875, 0.12246894836425781, -0.5416889190673828, -0.12250137329101562, 0.04057502746582031, -0.1467437744140625, -0.40924072265625, 0.08933639526367188, 0.13425445556640625, -0.08539581298828125, 0.020771026611328125, 0.10476875305175781, -0.348052978515625, -0.48804473876953125, -0.27069854736328125, 0.3239555358886719, -0.09417724609375, 0.5019454956054688, -0.1166534423828125, 0.06525802612304688, -0.27862548828125, -0.026012420654296875, 0.1607818603515625, -0.15341949462890625, 0.2675933837890625, 0.015239715576171875, -0.289520263671875, 0.10137939453125, 0.13552093505859375, -0.1506195068359375, 0.2689094543457031, 0.29591941833496094, 0.1424579620361328, 0.0345611572265625, -0.239654541015625, -0.2587432861328125, 0.024211883544921875, 0.31271934509277344, -0.46636962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000022.npy"}
|
|
{"epoch": 0.03325774754346183, "step": 23, "batch_size": 64, "mean": -0.007070481777191162, "std": 0.33784642815589905, "min": -0.996978759765625, "p10": -0.4088676452636718, "median": 0.01938915252685547, "p90": 0.48117103576660164, "max": 0.5992317199707031, "pos_frac": 0.5625, "sample": [-0.5915374755859375, -0.996978759765625, 0.09996795654296875, -0.52545166015625, -0.01294708251953125, 0.15023231506347656, -0.53704833984375, -0.09429931640625, -0.01148223876953125, -0.0252685546875, 0.17241668701171875, 0.0028705596923828125, 0.11814117431640625, 0.000629425048828125, 0.4630851745605469, -0.7471923828125, 0.1914520263671875, 0.22058868408203125, -0.1555938720703125, 0.2094879150390625, 0.3094215393066406, 0.10249519348144531, 0.009840011596679688, -0.29842376708984375, 0.09132766723632812, 0.37213134765625, -0.2661590576171875, -0.10836410522460938, 0.5418777465820312, 0.02893829345703125, -0.7979583740234375, 0.1111297607421875, -0.0080718994140625, 0.13028335571289062, 0.5735359191894531, 0.05633544921875, -0.3108673095703125, 0.0463714599609375, 0.13625717163085938, -0.21857070922851562, -0.42555999755859375, -0.13008880615234375, -0.350128173828125, 0.004486083984375, -0.017486572265625, -0.3699188232421875, 0.136444091796875, -0.2517242431640625, -0.26842498779296875, -0.271270751953125, 0.0887451171875, -0.3093891143798828, 0.13654327392578125, 0.07740592956542969, -0.15621566772460938, -0.10691261291503906, 0.488922119140625, 0.1158599853515625, 0.494171142578125, 0.586761474609375, 0.5936050415039062, 0.5992317199707031, 0.25518226623535156, 0.19464874267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000023.npy"}
|
|
{"epoch": 0.03476946334089191, "step": 24, "batch_size": 64, "mean": -0.04283055663108826, "std": 0.2619958519935608, "min": -0.7613067626953125, "p10": -0.3680049896240234, "median": 0.01651763916015625, "p90": 0.21822280883789066, "max": 0.5383453369140625, "pos_frac": 0.515625, "sample": [-0.3825492858886719, -0.20439910888671875, 0.05637359619140625, 0.0616607666015625, -0.22937393188476562, 0.09152603149414062, 0.13638687133789062, 0.513275146484375, -0.12172698974609375, -0.09622955322265625, 0.11042976379394531, 0.027990341186523438, -0.16404342651367188, 0.1393585205078125, 0.138519287109375, 0.152008056640625, -0.20074844360351562, -0.24793624877929688, 0.0562286376953125, 0.11128425598144531, -0.31002044677734375, 0.22258758544921875, 0.30010223388671875, -0.046718597412109375, 0.5383453369140625, -0.23208236694335938, 0.19276046752929688, -0.3043708801269531, -0.7613067626953125, -0.195831298828125, -0.2638378143310547, -0.2750091552734375, -0.4190826416015625, -0.4288463592529297, 0.107635498046875, -0.3906402587890625, -0.3222503662109375, 0.08795547485351562, -0.28907012939453125, -0.2531013488769531, 0.1118011474609375, 0.0308990478515625, 0.3659038543701172, 0.14661407470703125, 0.157135009765625, 0.319732666015625, -0.0384063720703125, -0.33406829833984375, -0.029605865478515625, 0.208038330078125, -0.4498310089111328, 0.15299224853515625, -0.14635467529296875, 0.012218475341796875, -0.697021484375, 0.4134368896484375, -0.15363693237304688, 0.05938720703125, 0.0890350341796875, 0.020816802978515625, 0.12290191650390625, -0.0982818603515625, -0.113922119140625, 0.20380783081054688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000024.npy"}
|
|
{"epoch": 0.036281179138321996, "step": 25, "batch_size": 64, "mean": -0.006348103284835815, "std": 0.28701674938201904, "min": -1.0926971435546875, "p10": -0.3079065322875976, "median": -0.007877349853515625, "p90": 0.2594017028808594, "max": 0.7977142333984375, "pos_frac": 0.5, "sample": [-0.09095001220703125, -0.177459716796875, -0.18084144592285156, 0.17339324951171875, -0.1204986572265625, -0.1024017333984375, 0.5242767333984375, 0.013418197631835938, 0.224456787109375, 0.08073043823242188, -0.40604400634765625, -0.048442840576171875, -0.409454345703125, 0.0394744873046875, 0.2376861572265625, 0.22261428833007812, -0.26496124267578125, 0.17853164672851562, -0.1434478759765625, 0.1419830322265625, -0.056591033935546875, 0.06867790222167969, -0.43714141845703125, 0.7977142333984375, 0.19311141967773438, 0.1886463165283203, 0.017620086669921875, 0.005340576171875, -0.2569694519042969, 0.031829833984375, 0.2497406005859375, 0.24384498596191406, 0.1492919921875, 0.0728302001953125, 0.101837158203125, -0.2691326141357422, -0.2415771484375, -0.19990158081054688, 0.15905189514160156, -0.4017791748046875, -1.0926971435546875, -0.14499282836914062, -0.17586135864257812, -0.17531204223632812, 0.36705780029296875, -0.02109527587890625, 0.2791633605957031, -0.14679718017578125, -0.40236663818359375, -0.17200469970703125, -0.21947860717773438, 0.158477783203125, 0.17800140380859375, -0.07474899291992188, -0.04340362548828125, 0.11124992370605469, -0.09392547607421875, 0.311920166015625, 0.26354217529296875, -0.32452392578125, 0.2355194091796875, -0.0345916748046875, -0.22113990783691406, 0.7232208251953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000025.npy"}
|
|
{"epoch": 0.03779289493575208, "step": 26, "batch_size": 64, "mean": -0.00837475061416626, "std": 0.30521219968795776, "min": -0.931976318359375, "p10": -0.36809349060058594, "median": -0.0018701553344726562, "p90": 0.37049484252929693, "max": 0.6955318450927734, "pos_frac": 0.5, "sample": [0.010578155517578125, -0.549224853515625, -0.2064056396484375, 0.0765380859375, 0.26880645751953125, 0.43469810485839844, 0.36138916015625, 0.4052543640136719, -0.12505722045898438, 0.20891189575195312, 0.005123138427734375, 0.4237060546875, -0.021453857421875, -0.050838470458984375, -0.931976318359375, -0.164764404296875, 0.37439727783203125, 0.5440826416015625, -0.0472259521484375, 0.20786285400390625, -0.15618896484375, -0.37060546875, 0.1251068115234375, -0.44512367248535156, 0.163360595703125, -0.32575225830078125, 0.3299102783203125, -0.008863449096679688, 0.019632339477539062, -0.09313201904296875, -0.1342182159423828, 0.6955318450927734, 0.29045867919921875, 0.3426666259765625, -0.02751922607421875, -0.476593017578125, 0.11806488037109375, 0.12799072265625, 0.1268768310546875, -0.3130645751953125, -0.05636405944824219, -0.5222206115722656, -0.17801856994628906, -0.2939949035644531, 0.5383224487304688, -0.28472900390625, 0.0682373046875, -0.16996383666992188, -0.297332763671875, -0.12137413024902344, 0.33330535888671875, -0.5336456298828125, 0.21507644653320312, -0.2215118408203125, -0.09611892700195312, 0.14847564697265625, -0.3604469299316406, 0.10339927673339844, 0.1297149658203125, 0.101837158203125, -0.3622322082519531, -0.029201507568359375, 0.0201263427734375, 0.1197357177734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000026.npy"}
|
|
{"epoch": 0.039304610733182165, "step": 27, "batch_size": 64, "mean": 0.0015683174133300781, "std": 0.31072792410850525, "min": -0.6072311401367188, "p10": -0.392578125, "median": -0.008910179138183594, "p90": 0.39340744018554685, "max": 0.85394287109375, "pos_frac": 0.484375, "sample": [0.17502975463867188, 0.26643943786621094, -0.3995819091796875, 0.09502410888671875, -0.012464523315429688, 0.43157196044921875, -0.01770782470703125, -0.12896728515625, -0.06465911865234375, 0.234466552734375, 0.578826904296875, -0.59857177734375, 0.22110748291015625, 0.34601783752441406, -0.14331436157226562, -0.132598876953125, 0.057281494140625, -0.3372802734375, -0.31574249267578125, -0.6072311401367188, 0.19902801513671875, 0.46270751953125, -0.4046478271484375, 0.239654541015625, 0.378448486328125, -0.3762359619140625, 0.12982940673828125, 0.24026870727539062, 0.07940673828125, 0.0015850067138671875, -0.35410308837890625, 0.08513832092285156, -0.08410072326660156, -0.277008056640625, 0.39398193359375, -0.43862152099609375, 0.4179534912109375, 0.47092437744140625, 0.114410400390625, -0.13470458984375, 0.39206695556640625, -0.5076828002929688, -0.17827415466308594, -0.14031219482421875, -0.21688461303710938, 0.28314971923828125, 0.0461273193359375, 0.125732421875, -0.072235107421875, 0.08612632751464844, -0.3063507080078125, -0.0158843994140625, -0.0053558349609375, -0.25787353515625, 0.3884429931640625, -0.2772102355957031, 0.85394287109375, -0.21355438232421875, 0.165191650390625, -0.3506050109863281, -0.2131500244140625, -0.5026168823242188, 0.3015632629394531, -0.07554244995117188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000027.npy"}
|
|
{"epoch": 0.04081632653061224, "step": 28, "batch_size": 64, "mean": 0.006865113973617554, "std": 0.3029409945011139, "min": -1.031768798828125, "p10": -0.33432235717773434, "median": 0.06401920318603516, "p90": 0.3422508239746094, "max": 0.8313674926757812, "pos_frac": 0.546875, "sample": [0.33757781982421875, -1.031768798828125, 0.0892333984375, 0.07274246215820312, -0.2004547119140625, 0.09065628051757812, -0.3140716552734375, -0.32503509521484375, 0.1934967041015625, -0.1512908935546875, -0.28005218505859375, 0.42778778076171875, -0.37026214599609375, -0.3383026123046875, 0.043212890625, -0.26995086669921875, -0.13396835327148438, -0.1597423553466797, -0.08803176879882812, 0.18547821044921875, 0.244659423828125, 0.3521461486816406, -0.13799667358398438, 0.034698486328125, 0.12748146057128906, 0.3032054901123047, 0.19356918334960938, 0.25539398193359375, -0.6027679443359375, 0.16632843017578125, 0.1576080322265625, -0.1641387939453125, 0.177215576171875, 0.110321044921875, 0.29767608642578125, -0.008758544921875, 0.10167694091796875, -0.36075592041015625, 0.20721435546875, 0.1508026123046875, 0.0940704345703125, -0.11847686767578125, -0.23341751098632812, 0.3923454284667969, -0.03942108154296875, -0.06625556945800781, 0.16958236694335938, -0.6884765625, -0.23189544677734375, -0.14985275268554688, -0.0880126953125, 0.8313674926757812, 0.4891242980957031, 0.19901275634765625, 0.453216552734375, 0.3442535400390625, -0.18020248413085938, -0.34075164794921875, -0.23548126220703125, 0.05529594421386719, 0.1165924072265625, 0.33535194396972656, 0.13208961486816406, -0.18352508544921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000028.npy"}
|
|
{"epoch": 0.042328042328042326, "step": 29, "batch_size": 64, "mean": 0.021755367517471313, "std": 0.2664119303226471, "min": -0.7596969604492188, "p10": -0.27954559326171874, "median": 0.031067848205566406, "p90": 0.3364208221435547, "max": 0.773284912109375, "pos_frac": 0.5625, "sample": [0.353668212890625, -0.2850799560546875, 0.038143157958984375, -0.7596969604492188, 0.773284912109375, 0.34702491760253906, 0.09197235107421875, 0.05724143981933594, -0.5827789306640625, 0.373687744140625, 0.026094436645507812, 0.21492767333984375, -0.12511444091796875, -0.08048629760742188, 0.114593505859375, 0.33977508544921875, -0.53765869140625, -0.4019908905029297, 0.0881500244140625, 0.2868499755859375, -0.01830291748046875, 0.116790771484375, 0.13263702392578125, -0.23276329040527344, 0.12064361572265625, -0.1486358642578125, -0.08615875244140625, -0.266632080078125, 0.036041259765625, 0.17945098876953125, -0.04582977294921875, -0.1393890380859375, -0.0615997314453125, 0.204254150390625, -0.5072021484375, 0.06279182434082031, 0.12862396240234375, 0.206146240234375, 0.08895683288574219, -0.004058837890625, -0.047607421875, -0.22397232055664062, -0.05586433410644531, 0.18840408325195312, 0.1631641387939453, -0.03873443603515625, 0.000499725341796875, 0.333282470703125, 0.0644073486328125, 0.0024166107177734375, 0.2783203125, -0.16527557373046875, 0.09777069091796875, -0.04739570617675781, -0.055145263671875, 0.008209228515625, -0.34084320068359375, 0.14116668701171875, 0.3339080810546875, 0.659912109375, -0.0551300048828125, -0.2310791015625, 0.3374977111816406, -0.0539398193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000029.npy"}
|
|
{"epoch": 0.04383975812547241, "step": 30, "batch_size": 64, "mean": 0.05786845088005066, "std": 0.4193798303604126, "min": -1.0318222045898438, "p10": -0.37311439514160144, "median": 0.03969001770019531, "p90": 0.4311111450195314, "max": 1.4803314208984375, "pos_frac": 0.546875, "sample": [-0.07469367980957031, -0.0281524658203125, -0.25711822509765625, -1.0318222045898438, 0.3372936248779297, 0.11698150634765625, 0.07994842529296875, 0.38094329833984375, -0.5968017578125, -0.5850830078125, -0.2730560302734375, 0.35710906982421875, 0.00884246826171875, -0.098846435546875, -0.46063995361328125, 0.213165283203125, -0.08818435668945312, 1.4803314208984375, -0.5048866271972656, 0.65423583984375, -0.16668701171875, -0.060699462890625, -0.2061634063720703, 0.14292526245117188, 0.6685791015625, -0.1303558349609375, 0.059906005859375, -0.11878585815429688, 0.8188400268554688, 0.39357757568359375, 0.027740478515625, -0.0650177001953125, 0.20333099365234375, 0.37306976318359375, 0.484527587890625, 0.1830291748046875, -0.25052642822265625, -0.06373214721679688, 0.0071258544921875, -0.24883079528808594, 0.10268402099609375, 0.09040069580078125, 1.29400634765625, 0.22039031982421875, -0.09918594360351562, 0.3871917724609375, 0.1877288818359375, 0.07106399536132812, 0.17531394958496094, -0.01043701171875, -0.09017181396484375, 0.32056427001953125, 0.24264907836914062, 0.2478790283203125, 0.44719696044921875, -0.23295211791992188, -0.2585296630859375, 0.3227958679199219, -1.031707763671875, 0.051639556884765625, -0.13650131225585938, -0.4159965515136719, -0.18003082275390625, 0.31616973876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000030.npy"}
|
|
{"epoch": 0.045351473922902494, "step": 31, "batch_size": 64, "mean": -0.011627346277236938, "std": 0.32182377576828003, "min": -0.99969482421875, "p10": -0.34514808654785156, "median": -0.011913299560546875, "p90": 0.3317556381225586, "max": 1.0313720703125, "pos_frac": 0.5, "sample": [0.19832611083984375, -0.103912353515625, -0.2399444580078125, 0.24189376831054688, -0.22110557556152344, 0.0111083984375, -0.3779296875, 0.010608673095703125, 0.3787078857421875, -0.06197357177734375, -0.21600914001464844, -0.48935699462890625, 0.4186134338378906, -0.5130767822265625, 0.32635498046875, 0.25067138671875, 0.33407020568847656, 0.00550079345703125, 0.43981170654296875, -0.2888832092285156, 1.0313720703125, 0.16867828369140625, 0.07734870910644531, -0.2901439666748047, 0.22339248657226562, 0.3955230712890625, 0.2476654052734375, -0.21368408203125, 0.02855682373046875, 0.173797607421875, 0.047119140625, -0.029327392578125, 0.22931671142578125, -0.10860252380371094, -0.08501434326171875, 0.22052383422851562, -0.02984619140625, -0.12081146240234375, -0.3180046081542969, -0.14324951171875, -0.2380523681640625, -0.8356475830078125, 0.052822113037109375, 0.7054290771484375, -0.259490966796875, -0.2488555908203125, -0.36611366271972656, -0.20291900634765625, 0.033931732177734375, -0.034999847412109375, -0.99969482421875, -0.29052734375, -0.356781005859375, 0.12358474731445312, -0.125213623046875, 0.1391754150390625, -0.1013946533203125, 0.2999420166015625, 0.10510635375976562, -0.04734039306640625, 0.27630615234375, 0.10988616943359375, 0.08138656616210938, -0.1727752685546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000031.npy"}
|
|
{"epoch": 0.04686318972033258, "step": 32, "batch_size": 64, "mean": -0.014784663915634155, "std": 0.2988406717777252, "min": -0.879974365234375, "p10": -0.35516891479492185, "median": -0.039188385009765625, "p90": 0.2759056091308594, "max": 0.9469985961914062, "pos_frac": 0.453125, "sample": [-0.23529815673828125, -0.2370758056640625, 0.20073699951171875, -0.3388214111328125, -0.15875244140625, 0.34284210205078125, -0.3627777099609375, -0.1383514404296875, 0.2633514404296875, 0.05849456787109375, 0.3535423278808594, 0.7110214233398438, 0.028827667236328125, -0.30389404296875, 0.23935890197753906, -0.879974365234375, -0.2805633544921875, -0.0480194091796875, -0.3991127014160156, -0.21248245239257812, 0.058780670166015625, -0.0771942138671875, 0.2400341033935547, 0.27368927001953125, -0.00211334228515625, 0.08882713317871094, -0.03035736083984375, 0.1134490966796875, -0.1638031005859375, 0.2549171447753906, 0.21595001220703125, 0.27685546875, -0.11180305480957031, -0.09162139892578125, -0.05194854736328125, 0.25023651123046875, 0.0767364501953125, -0.023624420166015625, -0.0491943359375, 0.9469985961914062, 0.030391693115234375, -0.08237838745117188, -0.10869598388671875, -0.21782302856445312, 0.011688232421875, 0.2718963623046875, 0.2021617889404297, -0.5810585021972656, -0.5033340454101562, -0.07252693176269531, 0.169525146484375, 0.06570053100585938, -0.06766319274902344, -0.06151390075683594, -0.29721832275390625, -0.17669296264648438, -0.5424575805664062, 0.4215869903564453, -0.35828399658203125, 0.2170257568359375, 0.087615966796875, 0.37244415283203125, -0.347900390625, -0.17657470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000032.npy"}
|
|
{"epoch": 0.04837490551776266, "step": 33, "batch_size": 64, "mean": 0.08834587037563324, "std": 0.3150263726711273, "min": -0.6475601196289062, "p10": -0.23259544372558594, "median": 0.044750213623046875, "p90": 0.507900619506836, "max": 1.0697975158691406, "pos_frac": 0.59375, "sample": [-0.1518707275390625, -0.3813934326171875, 0.1841278076171875, -0.4485015869140625, -0.053157806396484375, -0.010833740234375, -0.15869140625, -0.17359161376953125, -0.14864349365234375, 0.7204742431640625, 0.5247955322265625, 0.3815879821777344, 0.16037654876708984, -0.1269817352294922, 1.0697975158691406, -0.223876953125, -0.100860595703125, 0.19178390502929688, 0.0116729736328125, 0.8032989501953125, 0.17047119140625, 0.7430038452148438, 0.2105255126953125, 0.06085014343261719, -0.6475601196289062, 0.2881927490234375, 0.10243988037109375, 0.07523345947265625, 0.491363525390625, -0.022735595703125, 0.2559814453125, 0.03415679931640625, 0.5149879455566406, -0.046749114990234375, -0.2916431427001953, -0.04309654235839844, -0.0030231475830078125, 0.311279296875, 0.09050369262695312, 0.12666702270507812, -0.197906494140625, 0.15224456787109375, -0.03661155700683594, 0.3926544189453125, 0.00537872314453125, -0.2552947998046875, 0.35260009765625, 0.18010902404785156, -0.10345840454101562, -0.23633193969726562, -0.04279327392578125, 0.318756103515625, 0.015546798706054688, 0.1312713623046875, -0.130279541015625, 0.0239105224609375, 0.20419692993164062, 0.18666458129882812, -0.643585205078125, 0.0553436279296875, 0.011150360107421875, 0.6380157470703125, 0.14321517944335938, -0.0010223388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000033.npy"}
|
|
{"epoch": 0.049886621315192746, "step": 34, "batch_size": 64, "mean": 0.09302344918251038, "std": 0.32941189408302307, "min": -0.620513916015625, "p10": -0.3735153198242187, "median": 0.1134185791015625, "p90": 0.5287403106689456, "max": 0.98553466796875, "pos_frac": 0.59375, "sample": [-0.1655864715576172, 0.6664962768554688, 0.6028213500976562, 0.17829513549804688, 0.046733856201171875, 0.3743438720703125, -0.08555030822753906, 0.14505386352539062, 0.30162620544433594, 0.256317138671875, 0.3144073486328125, -0.1697845458984375, 0.1524810791015625, 0.3428802490234375, -0.0633697509765625, -0.3807525634765625, 0.6364364624023438, -0.14104461669921875, 0.26242828369140625, -0.09707832336425781, -0.4055595397949219, 0.06954383850097656, -0.07215118408203125, -0.441864013671875, 0.2729225158691406, 0.17691802978515625, -0.35662841796875, 0.4148292541503906, 0.29631805419921875, -0.1926116943359375, 0.14170074462890625, 0.19714736938476562, 0.04958152770996094, 0.20928573608398438, -0.0612640380859375, 0.98553466796875, -0.00528717041015625, 0.459716796875, 0.16580963134765625, 0.5526390075683594, 0.344482421875, -0.020145416259765625, 0.14061737060546875, 0.022977828979492188, 0.160400390625, 0.3083038330078125, -0.4879646301269531, -0.22034454345703125, 0.2778358459472656, -0.10427665710449219, -0.007659912109375, 0.08621978759765625, 0.800537109375, 0.5762176513671875, 0.15572738647460938, -0.5334739685058594, -0.23014068603515625, -0.04737663269042969, -0.04839134216308594, 0.012804031372070312, 0.4729766845703125, -0.287445068359375, -0.43160247802734375, -0.620513916015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000034.npy"}
|
|
{"epoch": 0.05139833711262283, "step": 35, "batch_size": 64, "mean": -0.0008640885353088379, "std": 0.36521878838539124, "min": -0.7928390502929688, "p10": -0.44831962585449214, "median": 0.03327655792236328, "p90": 0.38117313385009766, "max": 1.0068511962890625, "pos_frac": 0.53125, "sample": [0.0808868408203125, -0.727630615234375, 0.3204345703125, 0.04914093017578125, -0.3825225830078125, 0.8408660888671875, -0.6421661376953125, 0.06459808349609375, -0.320648193359375, -0.1540699005126953, -0.463104248046875, -0.09359931945800781, -0.00513458251953125, -0.1998577117919922, -0.154266357421875, 0.38100624084472656, 1.0068511962890625, -0.07338714599609375, 0.1392822265625, -0.31253814697265625, 0.2041339874267578, 0.19617080688476562, 0.3561973571777344, -0.21897125244140625, 0.08349609375, 0.07321929931640625, -0.1275482177734375, -0.4138221740722656, 0.4863739013671875, 0.29082489013671875, -0.6985321044921875, 0.2489471435546875, -0.18536376953125, -0.23606109619140625, -0.14965438842773438, -0.7928390502929688, 0.1739501953125, -0.182037353515625, 0.14005470275878906, -0.253021240234375, 0.08753585815429688, 0.32317161560058594, 0.35996246337890625, -0.23336029052734375, 0.000263214111328125, -0.01953887939453125, 0.28386688232421875, -0.31948089599609375, 0.047405242919921875, 0.019147872924804688, -0.24285507202148438, -0.35056304931640625, 0.0889129638671875, -0.119110107421875, -0.5926666259765625, 0.3812446594238281, -0.5977783203125, 0.26763916015625, 0.5942344665527344, 0.573394775390625, 0.22185134887695312, 0.5889663696289062, 0.10349082946777344, 0.1293048858642578], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000035.npy"}
|
|
{"epoch": 0.05291005291005291, "step": 36, "batch_size": 64, "mean": 0.006588369607925415, "std": 0.3408361077308655, "min": -0.7123908996582031, "p10": -0.42115936279296873, "median": 0.01742267608642578, "p90": 0.44376564025878906, "max": 0.7706756591796875, "pos_frac": 0.53125, "sample": [-0.4100189208984375, 0.6541595458984375, 0.5987091064453125, -0.5453681945800781, -0.3232574462890625, -0.7123908996582031, 0.25733184814453125, 0.24236679077148438, 0.18680572509765625, 0.60052490234375, -0.27065277099609375, 0.1301288604736328, 0.20142364501953125, -0.2545166015625, 0.1558990478515625, 0.29620361328125, 0.182525634765625, 0.11461257934570312, 0.09317970275878906, 0.558624267578125, 0.11284637451171875, 0.26123809814453125, 0.230804443359375, -0.20638656616210938, -0.65655517578125, -0.4256134033203125, -0.02979278564453125, 0.2432708740234375, 0.2828254699707031, -0.69586181640625, -0.30374908447265625, -0.0005340576171875, 0.1425628662109375, 0.1676025390625, -0.203857421875, -0.14324188232421875, -0.24317550659179688, 0.1217193603515625, -0.25605201721191406, -0.22248458862304688, 0.061489105224609375, 0.4374237060546875, 0.6227798461914062, -0.33454132080078125, 0.0046062469482421875, -0.06192588806152344, 0.37424468994140625, 0.2791862487792969, 0.18114471435546875, 0.4464836120605469, 0.053630828857421875, -0.178924560546875, -0.21848106384277344, 0.00406646728515625, -0.5102348327636719, -0.46417999267578125, -0.10462188720703125, -0.01563262939453125, -0.4107666015625, -0.2270050048828125, -0.18992996215820312, 0.7706756591796875, 0.030239105224609375, -0.05992698669433594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000036.npy"}
|
|
{"epoch": 0.05442176870748299, "step": 37, "batch_size": 64, "mean": 0.10840204358100891, "std": 0.4210970103740692, "min": -0.854827880859375, "p10": -0.2925811767578125, "median": 0.03805732727050781, "p90": 0.5725044250488283, "max": 1.747802734375, "pos_frac": 0.546875, "sample": [0.03105926513671875, 0.5909042358398438, 0.18852996826171875, -0.1353912353515625, 0.529571533203125, -0.30994415283203125, 0.03284454345703125, 0.1002960205078125, 0.043270111083984375, -0.06573486328125, -0.331207275390625, 0.3548469543457031, 0.7856674194335938, 1.472198486328125, 0.17823410034179688, -0.854827880859375, -0.000598907470703125, 0.464996337890625, 0.3943901062011719, 0.0535125732421875, -0.35930824279785156, -0.09572792053222656, -0.319915771484375, -0.1264190673828125, -0.1542682647705078, -0.25206756591796875, -0.04790496826171875, -0.07890510559082031, 0.11653900146484375, -0.05108642578125, -0.1122283935546875, -0.15782928466796875, 0.43267059326171875, 0.12902069091796875, 0.708465576171875, 0.6126022338867188, 0.222686767578125, 0.08161354064941406, -0.16502952575683594, -0.17273330688476562, 1.747802734375, -0.32997894287109375, -0.09928512573242188, 0.05423736572265625, -0.8245697021484375, -0.0022430419921875, 0.07828521728515625, 0.4300537109375, 0.06258392333984375, 0.069091796875, 0.35185813903808594, -0.12914276123046875, -0.057331085205078125, -0.19091796875, 0.7910537719726562, 0.36450958251953125, 0.080596923828125, 0.20538902282714844, -0.052734375, -0.2066650390625, 0.4793853759765625, 0.46381378173828125, 0.020998001098632812, -0.1018524169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000037.npy"}
|
|
{"epoch": 0.055933484504913075, "step": 38, "batch_size": 64, "mean": 0.0775165855884552, "std": 0.398190438747406, "min": -0.8059310913085938, "p10": -0.4256084442138672, "median": 0.06411075592041016, "p90": 0.5780605316162111, "max": 0.9444618225097656, "pos_frac": 0.578125, "sample": [0.05863761901855469, 0.3938751220703125, 0.182952880859375, 0.9444618225097656, -0.17547607421875, 0.496002197265625, -0.05064201354980469, 0.5105857849121094, -0.1473846435546875, -0.4963226318359375, 0.1285114288330078, 0.06305313110351562, -0.0051288604736328125, 0.06516838073730469, -0.42852783203125, -0.06949234008789062, -0.4335479736328125, 0.015233993530273438, -0.11383819580078125, 0.232574462890625, -0.34946441650390625, 0.93524169921875, 0.123138427734375, -0.30626678466796875, 0.4123687744140625, 0.1335906982421875, 0.48421478271484375, -0.699981689453125, -0.196685791015625, -0.13922119140625, -0.4187965393066406, -0.8059310913085938, 0.2652130126953125, 0.24662399291992188, 0.7275352478027344, -0.076507568359375, 0.00975799560546875, -0.37813377380371094, 0.15171051025390625, 0.1388988494873047, 0.2600212097167969, 0.5518569946289062, -0.05314826965332031, 0.27060699462890625, 0.5892906188964844, -0.4089012145996094, -0.3808250427246094, -0.5444602966308594, 0.371124267578125, 0.820098876953125, -0.4153900146484375, 0.493408203125, 0.8711090087890625, -0.023040771484375, -0.1322154998779297, 0.6451587677001953, -0.03046417236328125, 0.17858123779296875, -0.5554122924804688, 0.22602081298828125, 0.2508583068847656, 0.33282470703125, 0.1616191864013672, 0.05433845520019531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000038.npy"}
|
|
{"epoch": 0.05744520030234316, "step": 39, "batch_size": 64, "mean": 0.03494417667388916, "std": 0.3678329885005951, "min": -0.947418212890625, "p10": -0.5388992309570313, "median": 0.08483219146728516, "p90": 0.39241504669189464, "max": 1.02337646484375, "pos_frac": 0.5625, "sample": [0.6535568237304688, -0.5325546264648438, 0.1826934814453125, 0.2716693878173828, -0.22705078125, -0.6170196533203125, 0.04813385009765625, 0.3373546600341797, -0.06537246704101562, 0.08460807800292969, -0.09378814697265625, -0.025209426879882812, 0.6483917236328125, 0.248626708984375, 0.27725982666015625, 0.19243240356445312, -0.06595420837402344, 0.4202098846435547, 0.5142421722412109, -0.1469573974609375, -0.5697307586669922, -0.5507965087890625, -0.1295642852783203, 0.347991943359375, 0.67010498046875, -0.1478862762451172, -0.858245849609375, 0.4061107635498047, -0.5416183471679688, -0.24782943725585938, -0.012409210205078125, 0.1624603271484375, -0.2890472412109375, -0.18309783935546875, -0.03282928466796875, 0.32639312744140625, 0.1256866455078125, 0.2580108642578125, 0.022815704345703125, -0.41455841064453125, -0.001434326171875, -0.6465072631835938, 0.09229278564453125, 0.3090667724609375, 0.28082275390625, -0.19616317749023438, 1.02337646484375, 0.163238525390625, 0.16043853759765625, -0.17095375061035156, 0.3604583740234375, -0.947418212890625, 0.2193145751953125, 0.11203575134277344, -0.04123687744140625, 0.2103729248046875, 0.33275604248046875, -0.237091064453125, 0.33751678466796875, 0.026424407958984375, 0.13128662109375, 0.08505630493164062, -0.08395957946777344, 0.269500732421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000039.npy"}
|
|
{"epoch": 0.05895691609977324, "step": 40, "batch_size": 64, "mean": 0.04155641794204712, "std": 0.3246128559112549, "min": -0.6251068115234375, "p10": -0.4206069946289062, "median": 0.041794776916503906, "p90": 0.43949909210205096, "max": 0.834686279296875, "pos_frac": 0.578125, "sample": [-0.4392242431640625, 0.5979080200195312, 0.1801300048828125, 0.1603240966796875, 0.20593643188476562, 0.26985740661621094, 0.34600067138671875, 0.39984130859375, -0.45941162109375, 0.7353057861328125, -0.06278228759765625, 0.09456443786621094, 0.10802459716796875, -0.12980079650878906, -0.3302497863769531, -0.04432106018066406, 0.2015533447265625, 0.6820831298828125, 0.4564952850341797, -0.023241043090820312, 0.17674636840820312, 0.6185760498046875, 0.15716171264648438, 0.3221015930175781, 0.0292816162109375, 0.017303466796875, 0.1112518310546875, 0.006229400634765625, 0.29354095458984375, -0.08549690246582031, 0.18447494506835938, -0.6251068115234375, 0.834686279296875, -0.15251922607421875, -0.08079910278320312, 0.19754981994628906, 0.2792243957519531, 0.64801025390625, -0.5153732299804688, -0.377166748046875, -0.20726776123046875, -0.0040283203125, 0.05430793762207031, -0.1002349853515625, -0.2889251708984375, -0.24299240112304688, 0.02227020263671875, 0.08202743530273438, -0.13720321655273438, 0.00286865234375, 0.16300201416015625, -0.2544403076171875, -0.45095062255859375, 0.349029541015625, -0.2165374755859375, -0.4802589416503906, -0.2769775390625, 0.07317352294921875, -0.44968414306640625, -0.33104705810546875, 0.2599906921386719, -0.22454261779785156, 0.1959075927734375, 0.133453369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000040.npy"}
|
|
{"epoch": 0.06046863189720333, "step": 41, "batch_size": 64, "mean": 0.08492866158485413, "std": 0.37898486852645874, "min": -0.9263877868652344, "p10": -0.4221778869628905, "median": 0.0531158447265625, "p90": 0.5932029724121094, "max": 0.9118537902832031, "pos_frac": 0.59375, "sample": [0.20827484130859375, 0.08279800415039062, -0.019962310791015625, -0.66046142578125, 0.184539794921875, -0.000579833984375, -0.4835205078125, 0.9118537902832031, 0.005870819091796875, -0.1372528076171875, 0.21541595458984375, 0.2273540496826172, 0.04463958740234375, 0.06159210205078125, 0.06792449951171875, 0.404388427734375, -0.092559814453125, 0.43646240234375, 0.14995956420898438, 0.07144546508789062, 0.9011917114257812, -0.53350830078125, -0.2281513214111328, 0.2924346923828125, 0.37188720703125, -0.05260467529296875, 0.3308868408203125, 0.40888404846191406, 0.3869171142578125, -0.4851531982421875, -0.1071929931640625, 0.4677276611328125, 0.774139404296875, 0.6744308471679688, -0.08815574645996094, -0.31365203857421875, 0.02394866943359375, -0.02713775634765625, 0.6029434204101562, -0.19618797302246094, -0.16297531127929688, -0.9263877868652344, -0.199951171875, -0.25264549255371094, 0.5899505615234375, -0.6769847869873047, -0.033496856689453125, 0.2445068359375, 0.47713470458984375, 0.506011962890625, 0.1346588134765625, -0.46868896484375, -0.10727691650390625, 0.00864410400390625, -0.057857513427734375, 0.1011505126953125, 0.06971359252929688, 0.0146484375, 0.6494045257568359, -0.219970703125, 0.48563385009765625, 0.5945968627929688, 0.011659622192382812, -0.227874755859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000041.npy"}
|
|
{"epoch": 0.06198034769463341, "step": 42, "batch_size": 64, "mean": 0.031011685729026794, "std": 0.34122297167778015, "min": -0.7912826538085938, "p10": -0.43906173706054685, "median": 0.041312217712402344, "p90": 0.43426589965820334, "max": 0.818695068359375, "pos_frac": 0.515625, "sample": [0.058349609375, 0.4547119140625, -0.08605194091796875, 0.10404205322265625, -0.12944793701171875, -0.1832427978515625, 0.05114555358886719, -0.12352657318115234, -0.4337158203125, 0.18925857543945312, -0.42327880859375, 0.0314788818359375, -0.18423843383789062, -0.17956924438476562, -0.385833740234375, 0.3070716857910156, 0.818695068359375, 0.330169677734375, 0.2431182861328125, 0.33013916015625, 0.11066818237304688, 0.30171966552734375, 0.11275863647460938, -0.47513580322265625, -0.5269508361816406, -0.01612091064453125, 0.3017158508300781, -0.006378173828125, -0.0031070709228515625, -0.4719963073730469, 0.2605018615722656, 0.29708099365234375, -0.05177497863769531, -0.39618682861328125, 0.5412979125976562, 0.14934539794921875, 0.5574493408203125, -0.0091400146484375, -0.27410125732421875, 0.6402969360351562, -0.15989112854003906, -0.32762908935546875, -0.5026702880859375, -0.44135284423828125, -0.7912826538085938, -0.147552490234375, -0.2576141357421875, 0.220367431640625, -0.19961166381835938, 0.3693695068359375, 0.10510826110839844, 0.5928077697753906, -0.01638031005859375, 0.2716827392578125, 0.3170318603515625, -0.023040771484375, -0.5705490112304688, -0.028850555419921875, 0.7495040893554688, 0.0971527099609375, 0.21830177307128906, 0.12444496154785156, 0.38655853271484375, 0.16762542724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000042.npy"}
|
|
{"epoch": 0.06349206349206349, "step": 43, "batch_size": 64, "mean": 0.07082346081733704, "std": 0.32684293389320374, "min": -0.8306884765625, "p10": -0.32928695678710934, "median": 0.15124130249023438, "p90": 0.40556793212890624, "max": 0.6911163330078125, "pos_frac": 0.625, "sample": [-0.1744537353515625, 0.37554931640625, -0.409454345703125, 0.1483325958251953, 0.25939178466796875, -0.28815650939941406, -0.031528472900390625, -0.12688827514648438, -0.1977081298828125, -0.21662139892578125, 0.15415000915527344, 0.1006011962890625, 0.0245208740234375, 0.2021026611328125, 0.315338134765625, 0.6911163330078125, 0.47714805603027344, -0.025981903076171875, 0.17211341857910156, 0.03582763671875, 0.3240795135498047, 0.27347564697265625, -0.40555572509765625, -0.015472412109375, 0.4059295654296875, 0.1960277557373047, 0.13153648376464844, -0.21419525146484375, 0.067230224609375, 0.218353271484375, 0.27020263671875, -0.07958984375, 0.42612457275390625, -0.012908935546875, 0.12757110595703125, 0.40472412109375, 0.3929290771484375, -0.7436027526855469, 0.6480178833007812, 0.27855682373046875, -0.23482513427734375, -0.21302032470703125, 0.13787841796875, -0.5911102294921875, 0.16733360290527344, 0.6451568603515625, 0.17707061767578125, 0.46170806884765625, -0.233642578125, 0.3022918701171875, 0.22126007080078125, 0.35437774658203125, 0.16330718994140625, -0.31967926025390625, 0.3276710510253906, 0.28271484375, -0.6353530883789062, 0.16769027709960938, 0.388031005859375, 0.29205322265625, -0.8306884765625, -0.18747329711914062, -0.15547943115234375, -0.333404541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000043.npy"}
|
|
{"epoch": 0.06500377928949358, "step": 44, "batch_size": 64, "mean": 0.042850494384765625, "std": 0.4277274012565613, "min": -1.0227928161621094, "p10": -0.467218017578125, "median": 0.08111095428466797, "p90": 0.49206314086914066, "max": 1.31805419921875, "pos_frac": 0.5625, "sample": [-0.5001220703125, -0.3533897399902344, -0.4775390625, 0.29491424560546875, 0.247802734375, -0.471466064453125, 0.18602752685546875, -0.1668853759765625, -0.4567527770996094, -0.32358551025390625, -0.3076820373535156, 0.5128021240234375, -0.8668975830078125, 0.07219696044921875, 0.2502288818359375, 0.010467529296875, 0.1821765899658203, 0.2417583465576172, 0.1774005889892578, 0.1287403106689453, -0.034412384033203125, 0.6944541931152344, 0.3933429718017578, -0.2593536376953125, -0.13327598571777344, 0.3983001708984375, 0.4100799560546875, -0.13629150390625, -0.457305908203125, 0.14223480224609375, 0.58465576171875, 0.2648773193359375, 0.09001922607421875, 0.49547576904296875, 0.08794212341308594, -0.128509521484375, 1.31805419921875, -0.13172531127929688, -0.1319732666015625, -0.44635009765625, -0.1522083282470703, -0.09244537353515625, 0.8275146484375, 0.02046966552734375, 0.206634521484375, 0.4451026916503906, -0.0892486572265625, 0.33026123046875, -0.6439437866210938, 0.29096412658691406, -0.6878585815429688, 0.484100341796875, 0.2499237060546875, 0.07427978515625, 0.24225616455078125, 0.23998260498046875, 0.4231414794921875, 1.0386505126953125, 0.23954200744628906, -0.19347381591796875, -0.07928466796875, -0.43990325927734375, -0.36966705322265625, -1.0227928161621094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000044.npy"}
|
|
{"epoch": 0.06651549508692366, "step": 45, "batch_size": 64, "mean": 0.07733985781669617, "std": 0.45057305693626404, "min": -1.0593223571777344, "p10": -0.39097595214843744, "median": 0.11204910278320312, "p90": 0.5021991729736329, "max": 1.6519927978515625, "pos_frac": 0.640625, "sample": [0.27875518798828125, -0.48919677734375, 0.5719203948974609, -0.143798828125, -0.3113422393798828, -0.9009628295898438, -0.17237091064453125, -0.346343994140625, -0.922760009765625, -0.951171875, 0.5133743286132812, 0.4524383544921875, 0.10953140258789062, 0.175933837890625, 0.025030136108398438, 0.11456680297851562, 0.3729133605957031, 0.4689483642578125, 0.0970611572265625, 0.47414398193359375, 0.1685333251953125, -0.403839111328125, 0.56854248046875, -0.1374969482421875, 0.12308883666992188, 0.486083984375, 0.6212005615234375, 0.0816192626953125, 0.380889892578125, 0.40850067138671875, 0.5091056823730469, -0.06939697265625, -0.6930313110351562, 0.38555145263671875, 0.15865707397460938, 1.1367340087890625, 0.045833587646484375, 0.21230316162109375, -0.2641143798828125, 0.17041778564453125, -1.0593223571777344, -0.060211181640625, -0.1609210968017578, -0.23580551147460938, 0.07515716552734375, 0.0701446533203125, -0.0814208984375, 0.28365325927734375, -0.3609619140625, 0.1449127197265625, -0.110015869140625, 0.3556098937988281, 0.43856048583984375, 0.09816741943359375, 0.12476348876953125, 0.21019744873046875, 1.6519927978515625, 0.07865524291992188, 0.22183799743652344, 0.2734260559082031, -0.3153533935546875, -0.03790283203125, 0.25928497314453125, -0.220550537109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000045.npy"}
|
|
{"epoch": 0.06802721088435375, "step": 46, "batch_size": 64, "mean": 0.12569761276245117, "std": 0.46151596307754517, "min": -1.0650711059570312, "p10": -0.40086593627929684, "median": 0.09792709350585938, "p90": 0.7270317077636719, "max": 0.986480712890625, "pos_frac": 0.5625, "sample": [-0.18859100341796875, -0.3618316650390625, -1.0650711059570312, 0.4071807861328125, 0.767181396484375, -0.09161376953125, -0.1916332244873047, 0.00366973876953125, -0.130767822265625, -0.047821044921875, 0.6799087524414062, -0.0184783935546875, 0.1396636962890625, 0.64410400390625, -0.40976715087890625, 0.21768569946289062, -0.84832763671875, 0.13571929931640625, 0.5695629119873047, 0.24460411071777344, 0.20903587341308594, 0.608154296875, -0.16671371459960938, 0.30445098876953125, 0.722412109375, -0.0053863525390625, 0.7290115356445312, 0.7117996215820312, -0.34395599365234375, -0.08119964599609375, -0.00588226318359375, 0.9707489013671875, 0.00498199462890625, 0.10631561279296875, -0.4123859405517578, 0.4559326171875, -0.07568550109863281, -0.06887626647949219, -0.8675956726074219, 0.54351806640625, 0.6172027587890625, -0.24481201171875, 0.65472412109375, 0.16984939575195312, 0.8439598083496094, 0.08953857421875, -0.482666015625, -0.3028984069824219, -0.14425277709960938, 0.6201152801513672, 0.03142738342285156, 0.1470947265625, -0.07014656066894531, 0.9099388122558594, 0.4142913818359375, 0.986480712890625, -0.4909515380859375, -0.380096435546875, -0.2359466552734375, 0.8481292724609375, 0.1307220458984375, 0.11185646057128906, -0.25098609924316406, 0.27801513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000046.npy"}
|
|
{"epoch": 0.06953892668178382, "step": 47, "batch_size": 64, "mean": 0.1746828556060791, "std": 0.38169217109680176, "min": -0.5625152587890625, "p10": -0.33772583007812496, "median": 0.1618671417236328, "p90": 0.6784248352050783, "max": 1.1985931396484375, "pos_frac": 0.6875, "sample": [0.41962432861328125, -0.0558624267578125, 0.2728691101074219, 0.12819671630859375, 0.7011566162109375, 0.2592811584472656, 0.107696533203125, 0.630279541015625, 0.09007072448730469, 0.3341827392578125, 0.07830047607421875, 0.05881500244140625, 0.16246795654296875, -0.49918365478515625, -0.1591796875, 0.09030532836914062, 0.41433143615722656, -0.3497467041015625, 0.5087127685546875, 0.4862823486328125, -0.17981719970703125, 0.315185546875, -0.3096771240234375, 0.2295074462890625, -0.19710159301757812, 0.49322509765625, 0.16126632690429688, 0.12163162231445312, -0.039539337158203125, 0.4866313934326172, -0.4745826721191406, 0.25814247131347656, -0.188262939453125, 0.28595733642578125, -0.5625152587890625, 0.269256591796875, 0.7238311767578125, 0.46569061279296875, 0.37096214294433594, 0.8054046630859375, 0.3413543701171875, 0.251678466796875, 0.10199737548828125, 0.34899139404296875, 0.6990585327148438, -0.21456527709960938, 1.1985931396484375, -0.041412353515625, 0.3207206726074219, 0.05810737609863281, -0.073638916015625, -0.17960357666015625, -0.417816162109375, -0.5243606567382812, -0.523101806640625, 0.7614822387695312, 0.00733184814453125, 0.5014419555664062, 0.5006370544433594, -0.18962860107421875, 1.0736503601074219, 0.46759033203125, 0.039520263671875, -0.04212188720703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000047.npy"}
|
|
{"epoch": 0.0710506424792139, "step": 48, "batch_size": 64, "mean": 0.08520197868347168, "std": 0.5004569888114929, "min": -0.9003372192382812, "p10": -0.514975357055664, "median": 0.03348827362060547, "p90": 0.6533195495605468, "max": 1.6955947875976562, "pos_frac": 0.53125, "sample": [0.23726844787597656, -0.4274444580078125, 0.2375659942626953, 0.011524200439453125, -0.13873291015625, 1.1500053405761719, -0.393402099609375, 0.36919403076171875, -0.3110237121582031, 0.2326526641845703, 0.46933746337890625, 0.18625450134277344, -0.38861846923828125, -0.9003372192382812, 0.6541824340820312, 0.65130615234375, 0.6359405517578125, 0.1590290069580078, -0.7745285034179688, 0.4846038818359375, -0.1104736328125, 0.33184814453125, 0.1739654541015625, 0.4504127502441406, 0.1862030029296875, 0.49920654296875, -0.0980377197265625, 0.19821929931640625, -0.07926177978515625, 1.1056594848632812, -0.46250152587890625, 0.44873046875, -0.06174278259277344, 0.22379302978515625, -0.12458038330078125, 0.2867279052734375, -0.5374641418457031, -0.258514404296875, 1.1284561157226562, -0.06342315673828125, -0.2599964141845703, -0.6460647583007812, -0.04598236083984375, 0.32338523864746094, -0.7131118774414062, 0.0010318756103515625, 0.05545234680175781, 0.19235992431640625, -0.1805419921875, -0.17546844482421875, 0.697235107421875, -0.1371307373046875, -0.6779003143310547, 0.2681694030761719, -0.1967620849609375, 0.3329925537109375, 1.6955947875976562, -0.05202293395996094, -0.043914794921875, -0.442047119140625, 0.11931419372558594, 0.9755401611328125, -0.4250965118408203, -0.5941085815429688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000048.npy"}
|
|
{"epoch": 0.07256235827664399, "step": 49, "batch_size": 64, "mean": 0.111835777759552, "std": 0.4490898549556732, "min": -1.1971435546875, "p10": -0.39778518676757807, "median": 0.08354568481445312, "p90": 0.6202651977539063, "max": 1.5863494873046875, "pos_frac": 0.625, "sample": [0.5179519653320312, -0.06917953491210938, 0.3322029113769531, 0.628875732421875, 0.5190353393554688, -0.03273773193359375, 0.08744430541992188, 0.1663970947265625, -0.1827411651611328, -0.044647216796875, -0.13398170471191406, -0.02562713623046875, -0.41544342041015625, 0.29692840576171875, -0.09002304077148438, -0.1514415740966797, -0.4436302185058594, 0.0531005859375, 0.13875579833984375, -1.1971435546875, 0.3038444519042969, 0.1529254913330078, 0.07291793823242188, 0.5178680419921875, 0.1694488525390625, 0.48229217529296875, 0.1384735107421875, -0.0009288787841796875, 0.7234058380126953, -0.7376708984375, 0.2320384979248047, 0.5358352661132812, -0.18432235717773438, 0.6878757476806641, 0.5171318054199219, 0.6001739501953125, -0.2338714599609375, 0.0863037109375, 1.0061492919921875, 0.72454833984375, 0.0971221923828125, 0.01784515380859375, 0.0039196014404296875, -0.8957061767578125, 0.08078765869140625, 0.35613250732421875, 0.0095367431640625, 0.4587841033935547, 0.11943626403808594, 0.00797271728515625, 0.6429214477539062, 0.5633621215820312, -0.533477783203125, -0.3565826416015625, -0.16607666015625, -0.2838783264160156, 0.431488037109375, 1.5863494873046875, -0.020427703857421875, 0.011850357055664062, -0.27910614013671875, -0.00658416748046875, 0.28942108154296875, -0.72613525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000049.npy"}
|
|
{"epoch": 0.07407407407407407, "step": 50, "batch_size": 64, "mean": 0.09780901670455933, "std": 0.3750927150249481, "min": -0.7951507568359375, "p10": -0.31090106964111325, "median": 0.09799003601074219, "p90": 0.5690189361572268, "max": 1.021484375, "pos_frac": 0.609375, "sample": [0.44663238525390625, 0.6557159423828125, 0.500213623046875, 0.09370040893554688, -0.10715103149414062, 0.20833587646484375, 0.3817405700683594, -0.45061492919921875, 1.021484375, 0.4341583251953125, 0.17574501037597656, 0.1022796630859375, -0.7951507568359375, 0.23318862915039062, -0.008312225341796875, -0.27850914001464844, 0.451629638671875, 0.27098846435546875, 0.5197868347167969, -0.23403167724609375, 0.3673858642578125, 0.2602195739746094, -0.07864761352539062, -0.7642974853515625, 0.17663192749023438, 0.105987548828125, 0.17851638793945312, 0.590118408203125, -0.00305938720703125, -0.0858001708984375, 0.4304084777832031, -0.07220458984375, 0.2931556701660156, 0.3902626037597656, -0.5643844604492188, -0.22731781005859375, 0.13356781005859375, 0.8732070922851562, -0.13677978515625, 0.0299072265625, 0.71728515625, 0.7919464111328125, 0.29156494140625, -0.26508331298828125, 0.103729248046875, 0.2408008575439453, -0.2192840576171875, -0.12172698974609375, -0.5260238647460938, 0.1997528076171875, 0.06459426879882812, -0.3247833251953125, 0.014675140380859375, 0.038158416748046875, 0.6549777984619141, -0.01586151123046875, -0.653076171875, 0.07692337036132812, -0.2019805908203125, -0.03543853759765625, 0.09197425842285156, -0.15586090087890625, 0.1734142303466797, -0.19960784912109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000050.npy"}
|
|
{"epoch": 0.07558578987150416, "step": 51, "batch_size": 64, "mean": 0.20231595635414124, "std": 0.5237143039703369, "min": -1.0470352172851562, "p10": -0.39807987213134766, "median": 0.16477012634277344, "p90": 0.8992256164550787, "max": 1.9341278076171875, "pos_frac": 0.671875, "sample": [0.25592041015625, 0.16053390502929688, -0.1817646026611328, -0.21123123168945312, 0.49222564697265625, 0.3147125244140625, 0.09015274047851562, 0.572906494140625, -0.4708232879638672, -0.070648193359375, 0.11063766479492188, 0.7438735961914062, -0.11815643310546875, 0.9991226196289062, -0.3621368408203125, 0.012603759765625, 0.09313201904296875, 0.663665771484375, 0.4091148376464844, 1.3939208984375, 0.020782470703125, 0.4535675048828125, 0.3058319091796875, 0.4647216796875, 0.0821685791015625, -0.2947540283203125, -0.5092697143554688, -0.4031352996826172, 0.07557296752929688, 0.40210723876953125, 0.417083740234375, 0.9658050537109375, -0.253814697265625, 0.379913330078125, -1.0470352172851562, -0.14842987060546875, 0.2867851257324219, -0.7629356384277344, -0.1639404296875, 0.39023590087890625, -0.38628387451171875, -0.4427337646484375, 0.6222267150878906, 1.194580078125, 0.23137283325195312, -0.24161529541015625, -0.3018646240234375, 1.2777099609375, 0.27874755859375, 0.9800491333007812, 0.1532878875732422, 0.601837158203125, 0.48314666748046875, -0.23384857177734375, 0.16900634765625, -0.09522628784179688, 0.2927894592285156, 1.9341278076171875, 0.15981101989746094, 0.21054840087890625, -0.4094810485839844, 0.7167816162109375, 0.18140411376953125, 0.01282501220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000051.npy"}
|
|
{"epoch": 0.07709750566893424, "step": 52, "batch_size": 64, "mean": 0.16292083263397217, "std": 0.5498591065406799, "min": -1.3981170654296875, "p10": -0.4603279113769531, "median": 0.15523147583007812, "p90": 0.8389831542968755, "max": 1.5461578369140625, "pos_frac": 0.609375, "sample": [0.9465560913085938, -0.660369873046875, 0.7001495361328125, 0.56658935546875, 0.6228866577148438, -0.21562957763671875, 0.3618450164794922, 0.0517730712890625, -0.00988006591796875, 0.8919830322265625, 1.0345001220703125, 0.7153167724609375, -0.7302322387695312, 0.6917266845703125, 0.6971435546875, -0.320465087890625, 0.46433258056640625, -0.02150726318359375, 0.038631439208984375, -0.4296302795410156, -0.2381439208984375, -0.23202133178710938, 0.5235061645507812, -0.04464149475097656, 0.4391937255859375, 1.0338668823242188, 0.200714111328125, 0.054779052734375, -0.0027618408203125, 0.20032119750976562, 0.18179702758789062, -1.3981170654296875, 0.4707794189453125, -0.3058929443359375, 0.18004798889160156, 1.1110610961914062, 0.377349853515625, 0.42485809326171875, 0.1358642578125, -0.75604248046875, -0.0403594970703125, 0.00685882568359375, 0.41233062744140625, 0.39534568786621094, -0.13072967529296875, 0.6440277099609375, -0.060482025146484375, -0.0069122314453125, 1.5461578369140625, -0.22081756591796875, 0.6460647583007812, -0.871978759765625, -0.3943138122558594, 0.17459869384765625, 0.41968536376953125, -0.9179916381835938, 1.2822799682617188, -0.4734840393066406, 0.37225341796875, 0.11333847045898438, 0.034381866455078125, 0.183349609375, -0.16363906860351562, -0.2752685546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000052.npy"}
|
|
{"epoch": 0.07860922146636433, "step": 53, "batch_size": 64, "mean": 0.17789193987846375, "std": 0.4355870187282562, "min": -1.2826309204101562, "p10": -0.35204010009765624, "median": 0.23389053344726562, "p90": 0.6690273284912109, "max": 1.2457656860351562, "pos_frac": 0.65625, "sample": [0.22264480590820312, 0.7072792053222656, 0.5788650512695312, 0.2684478759765625, 0.40628814697265625, -0.08965110778808594, 0.24513626098632812, 0.173187255859375, 0.5787143707275391, -0.127655029296875, 0.368072509765625, 0.3413543701171875, -0.33455657958984375, 0.6687202453613281, -0.08235931396484375, 0.21545791625976562, 0.18875694274902344, 0.981414794921875, -0.1199798583984375, -0.77880859375, 0.2974052429199219, 0.05448150634765625, 0.44893646240234375, 0.6881179809570312, -1.2826309204101562, 0.383148193359375, -0.34676361083984375, 0.1744842529296875, 0.5149097442626953, 0.2593421936035156, 0.5221576690673828, -0.4831390380859375, -0.24662399291992188, -0.22129058837890625, 0.53240966796875, 0.69915771484375, 0.41910552978515625, 1.2457656860351562, 0.16949462890625, -0.2726097106933594, -0.3550567626953125, -0.24639129638671875, 0.5708065032958984, -0.1124114990234375, 0.669158935546875, 0.331451416015625, 0.358184814453125, -0.35430145263671875, 0.4530372619628906, 0.612335205078125, -0.214111328125, 0.20539474487304688, -0.021902084350585938, 0.7949943542480469, 0.28865814208984375, -0.3386650085449219, 0.5909271240234375, -0.39287567138671875, 0.4307098388671875, 0.11472129821777344, -0.09793472290039062, 0.39818572998046875, -0.3694000244140625, 0.10238075256347656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000053.npy"}
|
|
{"epoch": 0.0801209372637944, "step": 54, "batch_size": 64, "mean": 0.1708942949771881, "std": 0.5429055690765381, "min": -0.9362258911132812, "p10": -0.40555686950683595, "median": 0.09276199340820312, "p90": 0.7495353698730469, "max": 2.624542236328125, "pos_frac": 0.609375, "sample": [0.03957366943359375, 1.6403961181640625, 0.0306243896484375, 0.20253753662109375, -0.051151275634765625, -0.5514144897460938, 0.793243408203125, 0.35269927978515625, 0.8846511840820312, -0.5200881958007812, -0.1828155517578125, -0.020544052124023438, 0.0506744384765625, -0.9362258911132812, -0.010356903076171875, -0.043369293212890625, 0.55029296875, 0.2867774963378906, 0.13922119140625, 0.3764495849609375, -0.2366790771484375, 2.624542236328125, 0.43236732482910156, 0.6300487518310547, -0.1876068115234375, 0.7383842468261719, 0.238922119140625, -0.13011932373046875, 0.165802001953125, 0.0372314453125, -0.3415985107421875, 0.007892608642578125, -0.33872222900390625, -0.4023265838623047, 0.3720989227294922, 0.3867034912109375, 0.8575515747070312, -0.193756103515625, 0.7543144226074219, 0.5320243835449219, 0.13144683837890625, -0.4068489074707031, -0.06468582153320312, 0.5399627685546875, -0.17547225952148438, 0.44510650634765625, 0.1812896728515625, 0.44072532653808594, -0.21540451049804688, 0.5067214965820312, 0.41722869873046875, -0.6192092895507812, 0.443572998046875, 0.355560302734375, 0.3668804168701172, -0.09206581115722656, 0.0115966796875, -0.4342060089111328, 1.0006752014160156, -0.4317893981933594, -0.26920318603515625, 0.0540771484375, 0.175567626953125, -0.4025421142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000054.npy"}
|
|
{"epoch": 0.08163265306122448, "step": 55, "batch_size": 64, "mean": 0.30910903215408325, "std": 0.6902233362197876, "min": -0.8693389892578125, "p10": -0.5267295837402344, "median": 0.1571340560913086, "p90": 1.3240077972412114, "max": 2.2113189697265625, "pos_frac": 0.671875, "sample": [1.7172927856445312, 0.220245361328125, 0.2985687255859375, 0.09027099609375, -0.2159290313720703, -0.530792236328125, 1.0661773681640625, -0.5309486389160156, -0.5172500610351562, -0.4102630615234375, 0.044391632080078125, 0.7241058349609375, -0.735626220703125, -0.05770111083984375, 0.1499786376953125, 1.9109954833984375, 1.3822822570800781, 0.8728485107421875, 0.45428466796875, -0.19945144653320312, 0.41204833984375, 0.029815673828125, 0.7170524597167969, 0.1542835235595703, 1.704559326171875, 0.4476280212402344, 0.5094928741455078, 1.1880340576171875, 0.09362030029296875, -0.0533905029296875, 0.3670063018798828, 1.6304473876953125, -0.00168609619140625, -0.14963722229003906, -0.16070556640625, -0.0798492431640625, 0.21520137786865234, 0.15354156494140625, 0.47686767578125, 0.07564926147460938, 0.027252197265625, 0.9804611206054688, -0.07068634033203125, 0.7847442626953125, 0.23740386962890625, 0.1961078643798828, 1.7286529541015625, -0.8066253662109375, -0.42059326171875, -0.6269454956054688, 0.15998458862304688, 0.7186565399169922, -0.20311737060546875, 0.00334930419921875, -0.3388252258300781, 0.6366291046142578, 0.6693572998046875, 0.28321266174316406, 0.7569713592529297, 2.2113189697265625, 0.0687713623046875, -0.6959304809570312, 0.8887062072753906, -0.8693389892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000055.npy"}
|
|
{"epoch": 0.08314436885865457, "step": 56, "batch_size": 64, "mean": 0.1584291160106659, "std": 0.6718885898590088, "min": -2.4834747314453125, "p10": -0.567216682434082, "median": 0.2303009033203125, "p90": 0.8770332336425785, "max": 1.7724075317382812, "pos_frac": 0.640625, "sample": [-0.14874267578125, 0.3388824462890625, -0.09173583984375, -0.5980682373046875, 0.2482147216796875, 0.3569183349609375, 0.7975387573242188, -1.4452056884765625, -0.4828300476074219, 0.13463973999023438, 0.25732421875, -0.8586578369140625, -0.6283416748046875, 0.07872772216796875, 0.09580230712890625, 1.3767013549804688, 0.7090072631835938, -0.4202251434326172, 0.407196044921875, 0.9795646667480469, -0.49522972106933594, 1.231292724609375, 0.5923080444335938, 0.5450668334960938, 0.2123870849609375, 0.6302947998046875, -0.21230697631835938, 0.14832305908203125, 0.07911872863769531, 0.7629146575927734, -0.6178855895996094, 0.280426025390625, 0.911102294921875, 0.6983165740966797, 0.6480255126953125, 0.3950347900390625, -2.4834747314453125, 0.49695396423339844, -0.12488746643066406, 0.01374053955078125, 0.5839691162109375, 1.3115386962890625, -0.9041595458984375, -0.3512992858886719, 0.09049034118652344, 0.2990608215332031, -0.10643768310546875, 0.3759002685546875, -0.1329631805419922, 0.2898521423339844, 0.6807518005371094, 0.5536651611328125, 0.9838409423828125, 0.7267951965332031, -0.3507080078125, -0.27756309509277344, -0.45563507080078125, 0.30704498291015625, 0.4504280090332031, -0.18090057373046875, -0.060321807861328125, 1.7724075317382812, 0.05988121032714844, -0.3444061279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000056.npy"}
|
|
{"epoch": 0.08465608465608465, "step": 57, "batch_size": 64, "mean": 0.3333667814731598, "std": 0.6822146773338318, "min": -0.8704910278320312, "p10": -0.6465179443359375, "median": 0.37102699279785156, "p90": 0.971013641357422, "max": 2.208709716796875, "pos_frac": 0.6875, "sample": [-0.6631622314453125, -0.24785232543945312, 0.3342475891113281, 2.208709716796875, -0.2400970458984375, 0.8202362060546875, 0.27646636962890625, -0.5694618225097656, 0.6084747314453125, 0.7144012451171875, 1.65277099609375, 0.8294754028320312, -0.7178573608398438, 0.1533794403076172, 0.30816650390625, -0.758819580078125, -0.13004302978515625, 0.6418609619140625, 1.8777542114257812, 0.33004188537597656, 0.7686614990234375, -0.2387237548828125, 0.74530029296875, 1.19781494140625, -0.030849456787109375, -0.7928352355957031, 0.14017486572265625, 0.11590576171875, 0.018276214599609375, -0.8704910278320312, 0.7252120971679688, 0.6187496185302734, 0.5395736694335938, 0.48476409912109375, -0.0707855224609375, 0.8686294555664062, -0.836700439453125, 0.7130661010742188, -0.6076812744140625, 0.915740966796875, 0.41677093505859375, -0.10064697265625, -0.8703842163085938, 0.7285747528076172, -0.40171051025390625, 0.53472900390625, 0.355682373046875, 0.9463882446289062, 0.9815673828125, 0.44267845153808594, 0.3863716125488281, 0.5737533569335938, -0.365386962890625, 1.678070068359375, 0.1322956085205078, 1.4560394287109375, 0.4713287353515625, 0.5292224884033203, 0.15148353576660156, 0.7438278198242188, 0.34490203857421875, -0.011510848999023438, 0.9302444458007812, -0.5513114929199219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000057.npy"}
|
|
{"epoch": 0.08616780045351474, "step": 58, "batch_size": 64, "mean": 0.22388166189193726, "std": 1.0813742876052856, "min": -1.2386283874511719, "p10": -0.7764579772949218, "median": 0.050843238830566406, "p90": 1.2144493103027358, "max": 5.444366455078125, "pos_frac": 0.53125, "sample": [-1.2386283874511719, 0.7657127380371094, 0.7612056732177734, -0.4496192932128906, 0.5341644287109375, -0.35132598876953125, 0.096343994140625, 0.48654937744140625, 0.0557403564453125, -0.4443359375, 0.7976264953613281, -0.5240898132324219, 0.3056640625, -0.31891632080078125, -0.7846450805664062, 0.6342697143554688, 1.3716201782226562, -0.78814697265625, 0.0924072265625, 1.7156982421875, 0.22267532348632812, -0.18969345092773438, 0.18520355224609375, 0.5378952026367188, -0.985687255859375, -0.238433837890625, -0.6446151733398438, -0.27231597900390625, -0.7204456329345703, 0.84771728515625, 0.23414230346679688, -1.1650772094726562, -0.004261016845703125, 0.29505157470703125, 5.444366455078125, -0.3205718994140625, -0.4814453125, 3.5441436767578125, 1.4654159545898438, 0.5389862060546875, -0.182525634765625, 0.0072269439697265625, 0.3651847839355469, -0.6104888916015625, -1.0231781005859375, 3.07373046875, -0.2671184539794922, -0.484710693359375, -0.2603340148925781, 0.70684814453125, 0.380279541015625, 0.8099212646484375, -0.052703857421875, -0.08199691772460938, 0.23523712158203125, 0.6306304931640625, 0.04594612121582031, -0.48488616943359375, -0.757354736328125, 1.4979400634765625, -0.7962570190429688, -0.0159912109375, 0.14472007751464844, 0.4379615783691406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000058.npy"}
|
|
{"epoch": 0.08767951625094482, "step": 59, "batch_size": 64, "mean": 0.2704203426837921, "std": 0.9029937386512756, "min": -1.231903076171875, "p10": -0.6663196563720702, "median": 0.14063739776611328, "p90": 1.2355018615722657, "max": 3.440216064453125, "pos_frac": 0.578125, "sample": [-0.2806968688964844, 0.7212448120117188, 0.162933349609375, -0.5314102172851562, -0.3469867706298828, -0.5356006622314453, 0.2144317626953125, -1.1793899536132812, 0.3785552978515625, 1.187713623046875, 2.311920166015625, 0.1694183349609375, 1.0616455078125, 1.2469711303710938, 0.7329940795898438, -0.1285552978515625, 1.713165283203125, -0.845123291015625, 0.11309814453125, 0.10114288330078125, 0.11834144592285156, -0.043857574462890625, -0.36060333251953125, 0.269287109375, -0.1316375732421875, 1.0713729858398438, -0.040462493896484375, 1.208740234375, 3.440216064453125, 0.78228759765625, 0.0427703857421875, 0.5617294311523438, 0.7906093597412109, -0.29158782958984375, -0.8223114013671875, 0.19942283630371094, -0.37932586669921875, 0.6183948516845703, -0.6414566040039062, -0.0196533203125, 3.0152130126953125, 0.6490631103515625, 1.2622833251953125, 0.20366859436035156, -0.5043754577636719, 0.3531341552734375, -0.4984283447265625, 0.7811470031738281, -0.11963844299316406, 1.8209075927734375, -1.1878623962402344, 0.64337158203125, -0.6769752502441406, -1.231903076171875, 0.32381439208984375, -0.51116943359375, -1.0517730712890625, 0.271453857421875, -0.11815261840820312, -0.09391021728515625, 0.07705497741699219, 0.585357666015625, 0.9037704467773438, -0.2288970947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000059.npy"}
|
|
{"epoch": 0.08919123204837491, "step": 60, "batch_size": 64, "mean": 0.0797492265701294, "std": 0.8014106154441833, "min": -1.921173095703125, "p10": -1.1309158325195312, "median": 0.17130470275878906, "p90": 0.9514749526977541, "max": 2.290740966796875, "pos_frac": 0.5625, "sample": [0.21702957153320312, 0.10758209228515625, 0.5089874267578125, -0.12731170654296875, -0.31676483154296875, 1.5204544067382812, -1.15765380859375, -0.6557464599609375, -0.05169677734375, -1.921173095703125, 0.14902877807617188, 0.25899505615234375, -0.34430694580078125, 0.22515296936035156, -1.3752975463867188, 0.4751167297363281, -0.3002948760986328, 0.9081001281738281, 1.3788089752197266, 0.37763214111328125, 0.0382080078125, 0.22903060913085938, -0.0423583984375, -1.175140380859375, 0.13741683959960938, 0.6585617065429688, 0.5831451416015625, 0.3464851379394531, 1.1880722045898438, -1.2087516784667969, 0.37849998474121094, -0.23587799072265625, 0.6050338745117188, 0.453887939453125, -0.18508148193359375, 0.29019927978515625, 0.6867790222167969, -0.5147247314453125, -0.43790435791015625, -0.9400691986083984, -0.8525161743164062, 0.8161754608154297, 0.6067428588867188, -1.4584159851074219, -0.10521888732910156, 1.5244293212890625, -0.229766845703125, -1.0203704833984375, 1.1064529418945312, -0.010339736938476562, 0.8858642578125, -0.014862060546875, 0.7108001708984375, 0.7338790893554688, 2.290740966796875, -0.4930248260498047, 0.3132362365722656, 0.19358062744140625, 0.43039703369140625, 0.3626842498779297, -1.2833099365234375, -1.0685272216796875, 0.9700641632080078, -0.03679847717285156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000060.npy"}
|
|
{"epoch": 0.09070294784580499, "step": 61, "batch_size": 64, "mean": 0.061431676149368286, "std": 1.0106133222579956, "min": -2.4408111572265625, "p10": -1.3611000061035152, "median": 0.0681924819946289, "p90": 1.3560516357421881, "max": 2.0169296264648438, "pos_frac": 0.53125, "sample": [0.24822998046875, 0.414947509765625, 1.8970069885253906, 2.0169296264648438, -0.09572982788085938, 0.24102401733398438, 0.5394973754882812, 2.0047225952148438, 0.6986846923828125, -0.23987197875976562, -0.0066375732421875, -0.23756980895996094, -0.5419769287109375, -0.1309356689453125, 0.6824722290039062, -0.7116546630859375, 0.35672760009765625, -0.1940631866455078, -0.15503692626953125, 0.4635028839111328, 0.013277053833007812, 0.6037349700927734, 0.40296173095703125, -0.11075592041015625, 0.13309097290039062, -0.17499160766601562, -0.76678466796875, -1.4931488037109375, 0.4317474365234375, -0.45795440673828125, 0.9805984497070312, -0.4820060729980469, -1.0529861450195312, 1.4257965087890625, -0.508575439453125, 0.6331710815429688, -0.023712158203125, -0.9437675476074219, -0.8351516723632812, 1.5865020751953125, 1.9286956787109375, -2.25335693359375, 0.40001678466796875, -1.7280349731445312, 0.5600395202636719, 1.1933135986328125, -0.030338287353515625, 0.3896465301513672, -0.49188995361328125, -0.16937255859375, 1.1001319885253906, -1.7768478393554688, 0.07660865783691406, 0.05977630615234375, 0.8385772705078125, -2.4408111572265625, 1.554534912109375, 0.8723297119140625, 0.248870849609375, -0.5312881469726562, 0.7752456665039062, 0.7336502075195312, -2.4295005798339844, -1.5596847534179688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000061.npy"}
|
|
{"epoch": 0.09221466364323508, "step": 62, "batch_size": 64, "mean": 0.2291584312915802, "std": 0.8918454647064209, "min": -2.10479736328125, "p10": -0.9267063140869141, "median": 0.2332143783569336, "p90": 1.1785194396972665, "max": 2.6465301513671875, "pos_frac": 0.640625, "sample": [0.8152923583984375, -1.6389427185058594, -0.7089500427246094, 0.2272777557373047, 0.2849884033203125, 0.09049224853515625, 0.7000312805175781, -0.4261035919189453, -1.3007278442382812, -0.04877471923828125, 2.090606689453125, -0.23872947692871094, 1.5754547119140625, -0.9920501708984375, 1.429046630859375, 0.776214599609375, 0.1612548828125, -0.0733184814453125, -0.04691314697265625, 0.5244102478027344, -0.461669921875, -0.9196701049804688, -0.972930908203125, 0.32814979553222656, 0.5681552886962891, -0.0816497802734375, -0.47684478759765625, 0.778289794921875, 0.6661758422851562, 1.2591934204101562, 0.384918212890625, -0.09067535400390625, 0.19263267517089844, 0.59942626953125, 0.9424819946289062, -2.10479736328125, -0.42948150634765625, 1.886270523071289, -0.021894454956054688, 0.0936279296875, 0.046722412109375, 0.8702468872070312, 0.43761444091796875, 0.6587486267089844, 0.8862018585205078, 0.42531585693359375, 0.090240478515625, -0.6582145690917969, 0.5331573486328125, 0.9902801513671875, -0.2082366943359375, 0.8439731597900391, 0.4949932098388672, 0.2391510009765625, -0.9297218322753906, 0.16092681884765625, 2.6465301513671875, 2.4291839599609375, 0.04556846618652344, 0.23992156982421875, 0.4363250732421875, -0.6519317626953125, -1.422607421875, 0.7214832305908203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000062.npy"}
|
|
{"epoch": 0.09372637944066516, "step": 63, "batch_size": 64, "mean": 0.5518038272857666, "std": 0.8451820611953735, "min": -2.2276763916015625, "p10": -0.2781715393066406, "median": 0.4958963394165039, "p90": 1.5527269363403324, "max": 3.070343017578125, "pos_frac": 0.78125, "sample": [0.6019058227539062, 0.10549163818359375, 0.0648040771484375, -0.18037033081054688, 3.070343017578125, -0.3154773712158203, 0.03226470947265625, 2.465576171875, 0.7100448608398438, 0.5395164489746094, -2.2276763916015625, -0.4050483703613281, 0.4119834899902344, -1.1610565185546875, 1.4309844970703125, 0.6826553344726562, 2.3731842041015625, -0.09371185302734375, 0.4879436492919922, 0.5038490295410156, -0.33309173583984375, 0.46668052673339844, 0.8606796264648438, 2.2823944091796875, 0.2831878662109375, 0.9952106475830078, 0.6621551513671875, 0.21240234375, 0.3898887634277344, 0.3556671142578125, 0.3395843505859375, 0.6128005981445312, 1.3376502990722656, 0.6752853393554688, 0.28333282470703125, -0.02832794189453125, 0.4627952575683594, -0.4880638122558594, 0.5657558441162109, 0.6022014617919922, 0.6288375854492188, 2.12432861328125, 0.08135414123535156, 1.6049022674560547, 0.44737815856933594, 1.300008773803711, 1.384246826171875, 1.2806320190429688, 0.734954833984375, 0.23150634765625, 0.15343856811523438, 0.6983356475830078, -0.022113800048828125, -0.25197601318359375, 0.5796432495117188, -0.215576171875, 0.6781425476074219, 0.6820926666259766, -0.289398193359375, 0.06627464294433594, 0.9492225646972656, 1.69818115234375, 1.311614990234375, -0.1759796142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000063.npy"}
|
|
{"epoch": 0.09523809523809523, "step": 64, "batch_size": 64, "mean": 0.09811821579933167, "std": 0.8610791563987732, "min": -2.5925750732421875, "p10": -0.7905258178710938, "median": 0.07754707336425781, "p90": 1.2185066223144534, "max": 2.596771240234375, "pos_frac": 0.546875, "sample": [1.660675048828125, 2.596771240234375, -1.0892162322998047, -0.7608108520507812, -0.55303955078125, 0.056396484375, 0.5919742584228516, -0.0056915283203125, 0.5123958587646484, 1.5904541015625, 0.48769187927246094, 0.23917388916015625, 1.3136348724365234, -0.32346343994140625, 0.5912628173828125, -0.12580108642578125, 0.5336456298828125, -0.2072906494140625, 0.09869766235351562, 0.6034145355224609, 1.4064483642578125, 0.21047210693359375, -0.1033172607421875, -0.10623359680175781, -0.8032608032226562, -0.2949333190917969, -0.19080734252929688, -2.5925750732421875, -0.46909332275390625, 0.6179008483886719, 0.024282455444335938, 0.1061248779296875, 0.4654388427734375, -0.41712188720703125, 0.20134544372558594, 0.05112457275390625, 0.5845603942871094, -2.5117645263671875, -1.0125312805175781, 0.7583541870117188, 1.2519073486328125, 0.3441314697265625, 0.7086029052734375, -0.3458595275878906, 0.3579387664794922, -0.7324981689453125, -0.3416290283203125, 0.2727203369140625, -0.33170318603515625, 1.5400772094726562, -0.6832962036132812, -0.28635597229003906, 1.1405715942382812, 0.1916046142578125, -0.030704498291015625, 0.4836902618408203, 0.9125633239746094, 0.10892105102539062, -1.1100387573242188, -0.5092010498046875, -0.1080474853515625, -0.8455886840820312, -0.13541221618652344, 0.6918830871582031], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000064.npy"}
|
|
{"epoch": 0.09674981103552532, "step": 65, "batch_size": 64, "mean": 0.5486023426055908, "std": 1.2344807386398315, "min": -3.0927352905273438, "p10": -0.8052139282226562, "median": 0.4346504211425781, "p90": 1.8035953521728518, "max": 5.21771240234375, "pos_frac": 0.71875, "sample": [0.23014068603515625, 0.8999557495117188, 1.420654296875, 0.21469879150390625, 0.03064727783203125, 1.0074615478515625, -1.4888763427734375, 0.501068115234375, -0.43601036071777344, 1.541107177734375, -0.501953125, 0.7691535949707031, 1.1529121398925781, 0.67132568359375, 0.8848876953125, 0.920501708984375, 0.7231311798095703, 0.2689647674560547, -0.668182373046875, 2.726531982421875, 0.071044921875, 0.40578460693359375, -1.29534912109375, 1.6677169799804688, 1.651754379272461, -0.10678863525390625, -3.0927352905273438, -0.03702545166015625, -0.811370849609375, 0.4635162353515625, 0.025098800659179688, -0.27032470703125, -0.43524169921875, -0.9794464111328125, 0.24444580078125, 3.11383056640625, 0.14281845092773438, 2.6495819091796875, 1.8429069519042969, 1.4703311920166016, 2.046907424926758, 0.08086013793945312, 1.7118682861328125, 1.1884384155273438, -0.8241462707519531, 0.14510345458984375, 0.5745754241943359, 0.17719268798828125, -1.0375251770019531, -0.3819313049316406, 0.23094558715820312, 0.7373046875, -0.31339263916015625, 1.0021953582763672, 0.6760787963867188, 0.8432121276855469, 2.7275772094726562, -0.26146697998046875, 1.4248046875, 1.4020462036132812, 0.7981719970703125, -0.7908477783203125, 5.21771240234375, 0.1461944580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000065.npy"}
|
|
{"epoch": 0.0982615268329554, "step": 66, "batch_size": 64, "mean": 0.32909053564071655, "std": 1.073946475982666, "min": -2.468353271484375, "p10": -1.0691009521484374, "median": 0.2823448181152344, "p90": 1.7189678192138675, "max": 3.222686767578125, "pos_frac": 0.65625, "sample": [0.10003471374511719, 2.744842529296875, 1.1094207763671875, 0.42354583740234375, 0.2476806640625, 0.3018951416015625, 0.30672454833984375, -0.137054443359375, -0.02736663818359375, 0.5626182556152344, -0.42704010009765625, 0.7023696899414062, 1.0110015869140625, 0.17565536499023438, -0.706390380859375, -1.7051239013671875, 0.05092620849609375, 0.8050575256347656, 1.9333953857421875, 1.7774200439453125, 0.4521636962890625, 1.6397590637207031, -1.2539901733398438, -0.07938003540039062, 0.554534912109375, 0.5601749420166016, 0.32076454162597656, 1.2754478454589844, 1.239471435546875, 0.05950927734375, 0.9387969970703125, 0.7338390350341797, -0.0582427978515625, -0.15513229370117188, -0.1421661376953125, 0.2350921630859375, 0.21587181091308594, 0.307037353515625, -0.937744140625, 0.4217720031738281, -2.468353271484375, -2.1860122680664062, 0.1063385009765625, -0.44211578369140625, 0.26279449462890625, -1.1650619506835938, 1.5924568176269531, -0.13426971435546875, 1.7529144287109375, 1.2732524871826172, -0.2679481506347656, -0.2986869812011719, 0.11645126342773438, 3.222686767578125, 2.4028549194335938, -1.125396728515625, 0.49273681640625, -1.4683456420898438, -0.26641845703125, 2.413482666015625, 1.0867538452148438, -0.35129547119140625, 0.6299972534179688, 0.3057861328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000066.npy"}
|
|
{"epoch": 0.09977324263038549, "step": 67, "batch_size": 64, "mean": 0.3329530954360962, "std": 0.8742587566375732, "min": -1.4413299560546875, "p10": -0.7620782852172852, "median": 0.3741426467895508, "p90": 1.5698348999023437, "max": 2.940765380859375, "pos_frac": 0.703125, "sample": [0.21002960205078125, -1.073150634765625, -0.3930530548095703, -0.48667144775390625, -0.7626419067382812, 0.5116386413574219, 1.850067138671875, 0.9510021209716797, 0.433929443359375, -0.2764472961425781, 1.8886184692382812, 0.09744071960449219, -1.2487106323242188, -1.1736373901367188, 1.3204193115234375, 0.8042278289794922, -0.412322998046875, 0.6319541931152344, 0.023792266845703125, 0.51446533203125, -0.3518524169921875, 1.1986846923828125, -1.4413299560546875, 2.940765380859375, 0.2622871398925781, 0.40264892578125, 0.38704681396484375, 0.7542953491210938, 1.574554443359375, -0.7607631683349609, 0.18936538696289062, 0.27570152282714844, 0.8436241149902344, 1.5588226318359375, 0.8332366943359375, 0.20343589782714844, 0.611602783203125, 0.37671852111816406, -0.4243335723876953, 0.7715911865234375, 2.4955902099609375, 0.0750579833984375, 0.6808395385742188, 0.7130813598632812, 0.05756378173828125, 0.3715667724609375, -0.16845703125, -0.1261310577392578, 0.5362091064453125, -0.47145843505859375, 0.4221382141113281, 1.624664306640625, 0.3991279602050781, 0.13171958923339844, -0.2678985595703125, -0.3583984375, 0.695037841796875, 0.2551765441894531, -0.9605693817138672, -1.188323974609375, 1.8053703308105469, 0.0344696044921875, 0.456390380859375, 0.47917938232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000067.npy"}
|
|
{"epoch": 0.10128495842781557, "step": 68, "batch_size": 64, "mean": 0.34818604588508606, "std": 1.1721558570861816, "min": -3.198883056640625, "p10": -1.0191184997558593, "median": 0.37759971618652344, "p90": 1.6210159301757816, "max": 3.6322021484375, "pos_frac": 0.6875, "sample": [3.6322021484375, 1.2307052612304688, 0.11640548706054688, 1.714874267578125, -0.6098098754882812, 2.2513885498046875, 0.83892822265625, 0.043361663818359375, 0.8236827850341797, -0.4734039306640625, 1.8992996215820312, -3.198883056640625, 1.0194091796875, 1.2247848510742188, 0.6575469970703125, 0.71490478515625, -0.334564208984375, -0.21668243408203125, 0.2475128173828125, 1.5177536010742188, 0.7888946533203125, 0.25071144104003906, 0.6019287109375, 0.98419189453125, 0.4466285705566406, 0.36193275451660156, 0.4931182861328125, -0.1425762176513672, 1.9924545288085938, -0.3274078369140625, -0.11952781677246094, 0.3925952911376953, 0.4223747253417969, 0.170074462890625, -1.2283515930175781, 0.37317657470703125, 3.5446319580078125, 0.3088874816894531, 0.09232330322265625, 0.11676788330078125, -1.823760986328125, -1.0667076110839844, 1.0220184326171875, 0.3820228576660156, -1.91943359375, -2.4598617553710938, 0.18544769287109375, 0.3469963073730469, -0.9080772399902344, -0.30774688720703125, 1.6572113037109375, 0.4181365966796875, -0.7013168334960938, 1.39801025390625, 0.9492263793945312, 0.623870849609375, -0.23140716552734375, 0.810821533203125, 1.53656005859375, 0.6468219757080078, 0.9008903503417969, -0.2206573486328125, -0.07278060913085938, -1.5046234130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000068.npy"}
|
|
{"epoch": 0.10279667422524566, "step": 69, "batch_size": 64, "mean": 0.6142208576202393, "std": 1.217162847518921, "min": -1.958343505859375, "p10": -0.7435968399047851, "median": 0.47939491271972656, "p90": 1.985781478881836, "max": 4.85882568359375, "pos_frac": 0.734375, "sample": [0.11916542053222656, 0.3873291015625, 2.7844467163085938, 0.0238189697265625, 0.33251953125, 0.2490997314453125, -0.2797508239746094, -0.11953926086425781, 1.5614013671875, 0.7711257934570312, 0.6185226440429688, 0.2501411437988281, 0.20328330993652344, 1.7321815490722656, 0.8087997436523438, 2.363739013671875, 0.4753131866455078, 1.52679443359375, 2.1620025634765625, -1.1344833374023438, 3.175811767578125, -1.958343505859375, 1.7628631591796875, -0.25867462158203125, 1.95074462890625, -0.49297332763671875, 0.29793548583984375, 0.74591064453125, 0.6927337646484375, -0.2544403076171875, 0.7542018890380859, -0.93914794921875, 0.6987590789794922, 0.499603271484375, 0.2772026062011719, -0.9800872802734375, 1.8953132629394531, 0.6710586547851562, -0.2615509033203125, 0.7974624633789062, 0.483123779296875, 0.49211883544921875, 0.4186248779296875, 0.22997283935546875, 1.4788055419921875, -0.6638259887695312, 1.4664840698242188, 0.8079261779785156, -0.15045928955078125, -1.4183578491210938, 0.19510269165039062, 1.2875347137451172, -1.9509849548339844, -0.7777843475341797, 1.4716072082519531, -0.4579925537109375, 1.4080963134765625, 0.4756660461425781, 2.0007972717285156, -0.49454689025878906, 4.85882568359375, 0.6960067749023438, 3.4617462158203125, 0.08135414123535156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000069.npy"}
|
|
{"epoch": 0.10430839002267574, "step": 70, "batch_size": 64, "mean": 0.31661567091941833, "std": 1.2026655673980713, "min": -3.3233795166015625, "p10": -1.312402534484863, "median": 0.386474609375, "p90": 1.6139308929443361, "max": 3.474517822265625, "pos_frac": 0.640625, "sample": [0.1383056640625, -1.3766593933105469, 1.7307586669921875, -1.828765869140625, 0.17738914489746094, -1.1624698638916016, -0.22826385498046875, 1.1333847045898438, 1.2167816162109375, -0.07337570190429688, 0.9946060180664062, 1.6224784851074219, -3.3233795166015625, 1.2185630798339844, 0.05827140808105469, -0.6349830627441406, 2.406585693359375, -1.952423095703125, -0.7041778564453125, -0.4652557373046875, 3.474517822265625, 0.4486236572265625, 1.041421890258789, 0.47116851806640625, -1.6696624755859375, 0.6672325134277344, 0.5421867370605469, 2.2614898681640625, -0.9238967895507812, 2.813568115234375, 0.09709358215332031, 0.2560577392578125, -0.461517333984375, -0.6365013122558594, 0.6463394165039062, 1.0263137817382812, 1.5939865112304688, 0.833526611328125, 0.5991287231445312, 1.5093154907226562, 0.6961097717285156, -0.6487026214599609, -0.21959495544433594, 0.5000839233398438, 1.3093643188476562, 0.7917594909667969, -0.36251068115234375, 1.58856201171875, 0.20043373107910156, 1.0708160400390625, 0.47032928466796875, 1.7821884155273438, 0.2743072509765625, -1.9129638671875, 0.21743011474609375, -0.2648124694824219, -0.10858154296875, -1.4216384887695312, 1.3097076416015625, 0.8166732788085938, -0.1663970947265625, 1.0533943176269531, -0.5746440887451172, 0.3243255615234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000070.npy"}
|
|
{"epoch": 0.10582010582010581, "step": 71, "batch_size": 64, "mean": 0.23174314200878143, "std": 1.4276823997497559, "min": -4.63153076171875, "p10": -1.4277153015136717, "median": 0.2457103729248047, "p90": 1.8188377380371097, "max": 2.879119873046875, "pos_frac": 0.59375, "sample": [-0.16541099548339844, 1.3693008422851562, 1.0878734588623047, 0.5032024383544922, -0.15082550048828125, -0.7097015380859375, 1.1682624816894531, -1.1104660034179688, 0.38048553466796875, 1.732635498046875, 0.23476791381835938, -1.6330032348632812, 0.3063507080078125, 0.9946060180664062, -2.7953033447265625, 0.0096588134765625, 0.21202850341796875, 2.879119873046875, 0.6058998107910156, 1.4894256591796875, -3.243377685546875, 2.6995391845703125, 0.18535995483398438, -0.3463706970214844, 0.9061622619628906, 2.8246612548828125, 0.64501953125, -0.3031005859375, 0.6829776763916016, -0.0562286376953125, -4.63153076171875, -0.4587860107421875, -1.55523681640625, -0.4765758514404297, 0.08988761901855469, 0.0775604248046875, 0.3550300598144531, 1.0110015869140625, 1.48370361328125, -1.1301651000976562, 0.25665283203125, -3.1048736572265625, -0.1052703857421875, 2.084197998046875, -0.35048675537109375, -0.26511096954345703, -1.7310333251953125, 1.5621337890625, -0.2503204345703125, -0.06402969360351562, 1.108327865600586, 2.3394126892089844, -0.6469345092773438, 1.8557815551757812, 1.43505859375, 1.0278244018554688, 0.7185935974121094, -0.3179779052734375, 0.3314323425292969, 0.40067100524902344, 1.3694992065429688, -0.036579132080078125, 2.8609695434570312, -0.8148155212402344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000071.npy"}
|
|
{"epoch": 0.1073318216175359, "step": 72, "batch_size": 64, "mean": 0.7345869541168213, "std": 1.3209006786346436, "min": -3.0293426513671875, "p10": -0.6596229553222654, "median": 0.6904449462890625, "p90": 2.1790483474731444, "max": 4.596824645996094, "pos_frac": 0.703125, "sample": [1.9061813354492188, 0.944854736328125, 0.49627685546875, 1.7614822387695312, -3.0293426513671875, 0.24718856811523438, 0.24929237365722656, 4.596824645996094, -1.4626712799072266, -0.4064197540283203, 1.2908515930175781, 1.3615951538085938, -0.053436279296875, 1.6982765197753906, -0.1194915771484375, -0.4686145782470703, -0.395416259765625, 0.8042526245117188, 1.6634674072265625, -0.0836944580078125, 3.36041259765625, 1.080862045288086, 2.1741390228271484, 0.31890106201171875, 0.5832748413085938, 0.9911575317382812, 0.07025909423828125, -0.7414836883544922, -1.0414199829101562, -0.36701202392578125, 2.6362075805664062, 1.5450401306152344, 1.2354965209960938, 2.18115234375, 1.2933349609375, 0.355987548828125, 0.6275215148925781, 1.4592170715332031, 0.6383209228515625, -0.20352935791015625, 1.5702400207519531, 0.8672313690185547, 1.5298538208007812, 0.19264793395996094, 0.7425689697265625, -0.274139404296875, 1.756317138671875, 2.2271575927734375, -1.1658668518066406, 1.5138587951660156, -1.3958740234375, 0.9035797119140625, 3.66241455078125, -0.3249797821044922, 3.4707717895507812, -0.21383094787597656, 1.5037727355957031, 0.9140548706054688, -1.8328094482421875, 0.14060211181640625, -0.18386077880859375, 1.0597305297851562, 0.6292953491210938, 0.5215301513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000072.npy"}
|
|
{"epoch": 0.10884353741496598, "step": 73, "batch_size": 64, "mean": 0.11641725897789001, "std": 1.6950359344482422, "min": -3.208587646484375, "p10": -1.527999114990234, "median": -0.18689918518066406, "p90": 1.4530879974365236, "max": 8.011505126953125, "pos_frac": 0.453125, "sample": [1.3849983215332031, 0.6375617980957031, -0.0526580810546875, -1.2337188720703125, -1.3051681518554688, -1.3405075073242188, -1.6083526611328125, -0.5871124267578125, -3.208587646484375, 1.3601036071777344, -1.207305908203125, 0.33690643310546875, 0.13834571838378906, -0.9399223327636719, -1.1000862121582031, 0.6964759826660156, -1.8000602722167969, -0.20648574829101562, -0.7945785522460938, 0.9389419555664062, -0.4895782470703125, -0.1673126220703125, -0.8270835876464844, -0.5756683349609375, 8.011505126953125, 2.1255645751953125, -0.38747406005859375, -0.8277759552001953, 0.06700897216796875, 1.0659904479980469, 0.6035652160644531, -0.3254413604736328, 0.9393711090087891, -0.46112060546875, -2.308135986328125, 0.5404415130615234, 1.482269287109375, 2.58184814453125, -1.680124282836914, -1.3163299560546875, -0.7518768310546875, 3.9468536376953125, -0.382598876953125, -1.8950653076171875, -0.46347808837890625, 1.2404098510742188, 0.38531494140625, -0.8451595306396484, -0.0327606201171875, -0.7701530456542969, -2.136199951171875, 1.27130126953125, 0.7516155242919922, 0.6853904724121094, 0.9637527465820312, 0.7202301025390625, 1.1134910583496094, 0.1792755126953125, -0.8080101013183594, -0.35294342041015625, 0.11195755004882812, 2.328533172607422, 4.606903076171875, -0.5763874053955078], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000073.npy"}
|
|
{"epoch": 0.11035525321239607, "step": 74, "batch_size": 64, "mean": 0.7280859351158142, "std": 1.3563203811645508, "min": -1.9503936767578125, "p10": -1.0065256118774413, "median": 0.6601438522338867, "p90": 2.1700759887695313, "max": 5.7889404296875, "pos_frac": 0.703125, "sample": [-0.6004486083984375, 2.173980712890625, -0.62152099609375, 0.49390411376953125, 1.1909751892089844, -0.07299041748046875, 2.2376861572265625, 5.7889404296875, 1.1797752380371094, 2.4030532836914062, 1.1204090118408203, 0.3413963317871094, 2.1609649658203125, 1.63909912109375, -0.22850799560546875, 1.0568580627441406, 0.604827880859375, 1.1439056396484375, -0.043857574462890625, -1.76446533203125, 0.6367874145507812, 2.02044677734375, 0.5367279052734375, 0.8779754638671875, 0.6181869506835938, 0.18531227111816406, -0.7078361511230469, 1.1515350341796875, 2.4169654846191406, -1.2733802795410156, 0.3162193298339844, 0.7104244232177734, 1.353616714477539, -1.1421737670898438, -1.0289726257324219, 1.225555419921875, -0.6795463562011719, 1.064605712890625, -1.1453475952148438, -0.6250152587890625, 0.6835002899169922, -0.9541492462158203, 1.4673995971679688, 0.7844390869140625, 0.7332572937011719, 3.32012939453125, 2.0946044921875, -0.658050537109375, 1.562042236328125, 1.4022064208984375, -1.9503936767578125, -0.06051826477050781, 0.5296249389648438, -0.01824188232421875, -1.2637710571289062, 0.4786262512207031, 0.530120849609375, 4.285064697265625, 1.8133697509765625, 1.6992721557617188, 1.5107421875, 0.948089599609375, 0.4485492706298828, 0.495513916015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000074.npy"}
|
|
{"epoch": 0.11186696900982615, "step": 75, "batch_size": 64, "mean": 0.7304142713546753, "std": 1.3417645692825317, "min": -2.7716026306152344, "p10": -0.6741661071777344, "median": 0.5019664764404297, "p90": 2.5375793457031257, "max": 4.59393310546875, "pos_frac": 0.75, "sample": [-0.65179443359375, -0.6837539672851562, 0.03636360168457031, 1.0049705505371094, 3.5473709106445312, -0.13997268676757812, 0.23015975952148438, 1.0678787231445312, 1.5088348388671875, 3.287120819091797, -0.2698974609375, 0.126495361328125, -1.1602458953857422, 0.41957855224609375, 0.1429462432861328, 1.7054252624511719, 1.5414505004882812, -1.3238048553466797, 0.487274169921875, 0.26429176330566406, 3.770904541015625, 1.1812820434570312, 1.802978515625, 2.1639556884765625, 1.0995445251464844, 0.07557296752929688, 1.4075431823730469, -1.27508544921875, 2.388294219970703, 0.07932281494140625, 0.5223884582519531, 2.6015586853027344, -0.187957763671875, 1.9673576354980469, 0.988128662109375, 0.4406623840332031, 4.59393310546875, 0.8927459716796875, 0.7267608642578125, 0.7765045166015625, -0.13161468505859375, 1.10491943359375, -2.7716026306152344, 0.5222148895263672, 1.4611282348632812, 0.329132080078125, -0.2797737121582031, 0.27509307861328125, 0.17790603637695312, 0.2972888946533203, 0.6516132354736328, 2.0051422119140625, -1.308004379272461, -0.5721778869628906, 0.7785224914550781, -0.520263671875, 2.913116455078125, -0.41462135314941406, -0.8244400024414062, 0.4613456726074219, 0.5166587829589844, 0.04281044006347656, 3.333282470703125, 1.541748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000075.npy"}
|
|
{"epoch": 0.11337868480725624, "step": 76, "batch_size": 64, "mean": 0.2583199441432953, "std": 1.5556631088256836, "min": -4.755348205566406, "p10": -1.4720039367675781, "median": 0.20407772064208984, "p90": 2.15294189453125, "max": 4.051216125488281, "pos_frac": 0.546875, "sample": [-4.755348205566406, 1.0160179138183594, -2.2104721069335938, 0.1798858642578125, 0.6811141967773438, -0.9652824401855469, -0.3046283721923828, 0.7951545715332031, 0.2282695770263672, 3.239227294921875, 2.1272964477539062, 4.051216125488281, -0.5487594604492188, -3.3363571166992188, 1.2472076416015625, -1.6516189575195312, 0.734161376953125, 0.924407958984375, 0.446563720703125, 0.24465560913085938, 0.7056427001953125, 2.1639328002929688, -1.5185699462890625, 2.9880294799804688, -1.4716033935546875, 1.6487312316894531, 0.7657127380371094, 1.4903793334960938, -0.5299053192138672, -0.5855445861816406, -0.33330535888671875, -0.8194236755371094, -1.4721755981445312, -0.489044189453125, -0.5026111602783203, 0.6761474609375, 2.2014617919921875, 0.6173095703125, -0.7915515899658203, 1.3164825439453125, -0.24678611755371094, 0.14817047119140625, 0.046085357666015625, -0.7017860412597656, 1.9256591796875, 0.7041378021240234, 0.43872833251953125, -1.3825092315673828, -0.013519287109375, -0.6122970581054688, -0.053558349609375, 0.9998703002929688, -0.0194854736328125, 2.1848602294921875, 3.5454864501953125, 0.6683864593505859, -0.5618667602539062, -1.0447349548339844, -2.581939697265625, 2.0965728759765625, 1.3732070922851562, -0.25756263732910156, 1.7110061645507812, -0.03645515441894531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000076.npy"}
|
|
{"epoch": 0.11489040060468632, "step": 77, "batch_size": 64, "mean": 0.8278361558914185, "std": 1.5796291828155518, "min": -2.863677978515625, "p10": -1.2423606872558592, "median": 0.6020793914794922, "p90": 2.8650028228759767, "max": 4.9727935791015625, "pos_frac": 0.703125, "sample": [-0.7425861358642578, 1.6243743896484375, -1.1597061157226562, -1.5345687866210938, 1.7838592529296875, 2.1352157592773438, -1.4861183166503906, 0.4425487518310547, -2.863677978515625, 0.5506515502929688, 1.714569091796875, 1.4847145080566406, -0.13168907165527344, -1.6520729064941406, 0.4273719787597656, 3.2132606506347656, 0.3707275390625, 1.3657608032226562, 0.8741836547851562, 3.4272384643554688, -0.15030288696289062, 0.22187232971191406, 2.823455810546875, 2.466655731201172, -0.0083465576171875, 0.5908889770507812, 0.32970619201660156, 2.8828086853027344, 0.49637413024902344, 0.6132698059082031, -0.4204978942871094, -1.187774658203125, -0.100250244140625, 0.2299041748046875, 1.408447265625, 2.327362060546875, 0.8101997375488281, 1.1646671295166016, 1.0229244232177734, 2.46954345703125, 1.7411956787109375, 0.14871978759765625, 0.8948707580566406, -0.4264106750488281, 0.14809799194335938, 4.9727935791015625, 0.23374557495117188, -0.15448951721191406, 3.4452972412109375, 1.0613594055175781, 4.826713562011719, -0.012115478515625, 0.2384014129638672, 1.8867416381835938, 2.3237075805664062, 0.6560783386230469, -1.9941787719726562, -1.424041748046875, 2.5810546875, -0.27692604064941406, 1.0885467529296875, 3.5158157348632812, 0.9373283386230469, -1.2657546997070312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000077.npy"}
|
|
{"epoch": 0.1164021164021164, "step": 78, "batch_size": 64, "mean": 0.4483870565891266, "std": 1.3830755949020386, "min": -3.3612823486328125, "p10": -1.2098121643066406, "median": 0.35007572174072266, "p90": 2.1644943237304695, "max": 3.68896484375, "pos_frac": 0.625, "sample": [2.5053977966308594, -2.4449462890625, 0.7242927551269531, 2.5018997192382812, 0.2949981689453125, 2.2375946044921875, 0.8197174072265625, 0.19118499755859375, 0.18690872192382812, 0.152801513671875, 1.37030029296875, -1.5545501708984375, -0.10299491882324219, 1.815216064453125, -0.161865234375, -0.36472320556640625, -1.511474609375, 3.3254318237304688, -1.2249069213867188, 0.6180267333984375, -0.471649169921875, -0.1790618896484375, -0.355621337890625, -0.6021804809570312, 1.5370330810546875, 0.6348953247070312, 0.267547607421875, -2.9109573364257812, 1.993927001953125, -3.3612823486328125, 1.2250900268554688, 1.7120437622070312, -0.21746063232421875, 1.8966407775878906, 0.17726898193359375, -0.2637176513671875, -1.238128662109375, 0.7360992431640625, 1.642404556274414, 1.9354629516601562, 0.7732353210449219, 0.4395256042480469, 0.406585693359375, 1.5418548583984375, -0.8385009765625, 1.812469482421875, -1.174591064453125, 0.2064075469970703, 0.4716339111328125, 1.5181350708007812, -0.503570556640625, 2.2567138671875, -0.32171630859375, -0.9792022705078125, 0.4310455322265625, -0.09497833251953125, 0.4051532745361328, -0.44393157958984375, 1.1980514526367188, 2.5135879516601562, -0.294830322265625, 0.23960113525390625, 3.68896484375, 1.9084644317626953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000078.npy"}
|
|
{"epoch": 0.11791383219954649, "step": 79, "batch_size": 64, "mean": 1.1612836122512817, "std": 2.0575318336486816, "min": -3.0214881896972656, "p10": -0.606903076171875, "median": 0.7740144729614258, "p90": 3.275159454345705, "max": 8.100372314453125, "pos_frac": 0.734375, "sample": [-0.03969764709472656, 0.188140869140625, -1.209360122680664, 1.6234016418457031, -0.33071136474609375, 1.1149978637695312, -3.0214881896972656, 0.2696247100830078, 0.8677902221679688, 0.47930908203125, -0.46996498107910156, 0.328826904296875, 1.9353065490722656, 8.052505493164062, -1.7491531372070312, 0.445587158203125, 2.0650672912597656, 0.6093959808349609, -0.32933807373046875, 0.54656982421875, 0.6939945220947266, 6.6591339111328125, 2.3352508544921875, 3.72308349609375, 0.7860450744628906, -1.6325206756591797, 4.595882415771484, 2.3175506591796875, 1.7064666748046875, 1.4218215942382812, -1.1964263916015625, -0.7047176361083984, 2.6055526733398438, 2.016448974609375, 2.006380081176758, 0.42073631286621094, 1.0760841369628906, 2.0090789794921875, -0.507049560546875, 2.784088134765625, -0.4987144470214844, 2.4693603515625, -0.006282806396484375, -0.4253711700439453, 0.7619838714599609, 0.9439620971679688, 2.56756591796875, 0.37795448303222656, 0.814056396484375, 0.10390472412109375, 1.0271663665771484, -0.5662078857421875, 5.66448974609375, -0.6243438720703125, 0.7050857543945312, 8.100372314453125, 0.8236236572265625, 0.8386764526367188, 0.39636993408203125, 2.2657737731933594, 0.00205230712890625, 0.9876899719238281, 3.4856185913085938, -0.3863258361816406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000079.npy"}
|
|
{"epoch": 0.11942554799697656, "step": 80, "batch_size": 64, "mean": 0.5516113042831421, "std": 1.9728193283081055, "min": -4.2127227783203125, "p10": -1.526697540283203, "median": 0.25531768798828125, "p90": 3.4276908874511722, "max": 5.253936767578125, "pos_frac": 0.546875, "sample": [-0.355804443359375, 3.0675392150878906, 0.5127410888671875, 2.1127166748046875, 1.0595989227294922, -0.634307861328125, -0.6459274291992188, -0.4549121856689453, -1.531982421875, -2.056427001953125, 2.061767578125, 3.3738632202148438, -0.7979736328125, -1.3065605163574219, -0.15662574768066406, 4.491813659667969, -1.4764328002929688, -0.4849090576171875, 0.8580474853515625, -0.10490608215332031, 0.26827239990234375, -1.7231979370117188, 0.19162750244140625, 1.6148757934570312, 4.5456085205078125, 3.117603302001953, 2.5296783447265625, -0.6005630493164062, 3.573028564453125, 4.300323486328125, -1.5143661499023438, 1.5518341064453125, 0.2921104431152344, 0.5219688415527344, 1.1223373413085938, 2.028228759765625, -0.8127288818359375, 0.24236297607421875, 4.100788116455078, -4.2127227783203125, 3.4507598876953125, -0.263885498046875, -3.286895751953125, 2.050018310546875, -0.3304767608642578, 0.4067230224609375, 1.7364959716796875, 0.4607219696044922, -1.4574432373046875, -1.5597648620605469, 2.4719085693359375, -0.32138824462890625, -0.36041259765625, -0.35059356689453125, -0.4572925567626953, 0.04150390625, 0.7121734619140625, -1.1900215148925781, 1.5167121887207031, -2.997386932373047, 0.3109588623046875, -0.4620819091796875, 5.253936767578125, 1.260467529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000080.npy"}
|
|
{"epoch": 0.12093726379440665, "step": 81, "batch_size": 64, "mean": 0.8343477845191956, "std": 1.8488696813583374, "min": -3.3653793334960938, "p10": -1.1912887573242186, "median": 0.7174568176269531, "p90": 2.9883571624755865, "max": 7.2932281494140625, "pos_frac": 0.640625, "sample": [0.876190185546875, 1.9023799896240234, 1.6126365661621094, -1.3876266479492188, 0.9485893249511719, -2.4580459594726562, 0.7573566436767578, 0.21790313720703125, -0.5170230865478516, -0.7861518859863281, 5.1016387939453125, -0.0615692138671875, 1.2204132080078125, 2.513214111328125, 0.6775569915771484, -0.1475982666015625, 0.49248313903808594, -0.0046558380126953125, 3.6652679443359375, 2.1682777404785156, 1.2332229614257812, -0.48797607421875, 1.1803665161132812, 1.295196533203125, 2.3853912353515625, 0.8885040283203125, 0.1912841796875, -0.5327796936035156, -0.5135898590087891, -1.4738616943359375, 0.9713516235351562, -1.8858833312988281, 1.3244247436523438, 0.37214088439941406, -0.7375640869140625, 4.0961151123046875, 0.5566558837890625, 7.2932281494140625, 2.261810302734375, 5.0495147705078125, -1.25244140625, 1.3002033233642578, 0.38149261474609375, -0.11418914794921875, 1.4939804077148438, -0.12381553649902344, 2.070432662963867, 0.81146240234375, 2.6833953857421875, 1.8392333984375, 3.283763885498047, -1.0485992431640625, 2.8146095275878906, -0.5215606689453125, -0.29290199279785156, 0.6069107055664062, -1.8569202423095703, 0.5454006195068359, 0.8852577209472656, 3.0628204345703125, -0.7287445068359375, 1.6855278015136719, -3.3653793334960938, -1.0204696655273438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000081.npy"}
|
|
{"epoch": 0.12244897959183673, "step": 82, "batch_size": 64, "mean": 1.01225745677948, "std": 1.5696290731430054, "min": -4.606201171875, "p10": -0.5162515640258789, "median": 0.8622531890869141, "p90": 3.0262817382812504, "max": 5.4623870849609375, "pos_frac": 0.78125, "sample": [0.7064380645751953, 1.3981208801269531, 0.5827178955078125, -0.6369667053222656, 2.2699661254882812, 0.08480453491210938, 0.7511482238769531, 2.1636199951171875, 3.6066131591796875, 1.2747421264648438, 1.215179443359375, 0.6958942413330078, 1.421234130859375, 2.6360244750976562, 0.2947731018066406, 1.3116378784179688, 0.6783294677734375, 0.5745201110839844, 1.2940540313720703, -0.11478424072265625, 0.3491058349609375, -0.30376434326171875, 0.9754905700683594, -0.07275581359863281, -0.46562957763671875, 3.9058074951171875, 1.4286003112792969, 1.5339508056640625, 1.8095111846923828, 1.4021625518798828, 0.2071990966796875, -0.4202842712402344, 5.4623870849609375, 0.4489898681640625, 1.4975738525390625, 3.7497100830078125, 2.960784912109375, 0.973358154296875, 1.4875030517578125, 3.5232467651367188, 0.3019752502441406, -1.9218597412109375, 2.8067550659179688, -1.3459701538085938, 1.6238861083984375, 0.08741378784179688, 2.4223709106445312, 0.11145401000976562, 3.054351806640625, 0.7371826171875, -0.09853363037109375, 0.1801166534423828, 1.7704620361328125, 3.9995803833007812, -0.2957344055175781, -0.7228031158447266, 1.6168632507324219, 1.2904167175292969, 1.6133155822753906, 0.6168632507324219, -1.29876708984375, -4.606201171875, -0.5379467010498047, 0.7182693481445312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000082.npy"}
|
|
{"epoch": 0.12396069538926682, "step": 83, "batch_size": 64, "mean": 1.1638286113739014, "std": 1.8501423597335815, "min": -3.1786956787109375, "p10": -0.7786952972412109, "median": 1.010171890258789, "p90": 3.0892593383789073, "max": 7.762516021728516, "pos_frac": 0.78125, "sample": [2.5461883544921875, 0.727752685546875, -0.8965377807617188, 0.418792724609375, 0.42131996154785156, -0.23728179931640625, 1.25140380859375, 1.252166748046875, 2.0689849853515625, 4.556549072265625, 0.557373046875, 1.6608161926269531, -1.8302803039550781, 1.5040855407714844, 1.81060791015625, 1.027008056640625, 2.0510940551757812, 2.1022109985351562, 2.3646583557128906, 1.4918708801269531, 2.251739501953125, -0.7974853515625, 4.265907287597656, 0.14850616455078125, -1.4022140502929688, 0.9322357177734375, 3.822174072265625, 0.3672027587890625, 0.46187591552734375, -0.1604938507080078, -0.3343963623046875, 0.4442577362060547, 2.8533554077148438, -3.1786956787109375, -0.8209896087646484, 2.2511444091796875, 1.4738712310791016, -0.0615234375, 1.3581714630126953, -2.9805526733398438, 0.2197284698486328, 0.39202880859375, 0.6497116088867188, 0.4623985290527344, -0.25382423400878906, -0.004421234130859375, 0.2700080871582031, 2.224395751953125, 3.2990264892578125, 1.7774658203125, 7.762516021728516, 1.7553882598876953, 3.1903610229492188, 0.9933357238769531, 1.2161636352539062, 1.7259464263916016, 1.2175178527832031, 1.946676254272461, 2.0278854370117188, 7.170166015625, 0.5455322265625, 0.45436859130859375, 0.4346332550048828, -0.7348518371582031], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000083.npy"}
|
|
{"epoch": 0.1254724111866969, "step": 84, "batch_size": 64, "mean": 0.9226758480072021, "std": 1.8515326976776123, "min": -3.4122314453125, "p10": -1.3027389526367186, "median": 0.6594390869140625, "p90": 3.5295570373535163, "max": 5.14862060546875, "pos_frac": 0.734375, "sample": [0.293548583984375, 2.14837646484375, 0.7405738830566406, 2.2731895446777344, 4.901458740234375, 2.009075164794922, 3.6114578247070312, -0.4705238342285156, -0.4564094543457031, -2.5476608276367188, 1.1448822021484375, 2.2087326049804688, 2.959056854248047, -0.06478118896484375, 0.5695228576660156, 0.16091156005859375, 0.3833808898925781, -0.49692535400390625, 4.408180236816406, -1.1576118469238281, 0.9955596923828125, 2.3809127807617188, 1.118988037109375, 2.1124191284179688, 0.4470024108886719, 0.677093505859375, -1.3307266235351562, 3.3384552001953125, 0.22655868530273438, 1.1471405029296875, 0.64178466796875, 0.5438079833984375, -0.2782630920410156, 5.14862060546875, -1.9559173583984375, 1.113189697265625, 1.6534748077392578, 2.241567611694336, 1.355316162109375, -2.5150985717773438, -3.4122314453125, 4.3827667236328125, 2.110015869140625, 0.18322372436523438, 3.6194534301757812, 0.48592185974121094, 0.01726531982421875, 2.9481372833251953, 1.2354469299316406, -2.2746734619140625, 0.5538101196289062, -0.891387939453125, 2.128591537475586, 1.4786834716796875, 1.0582542419433594, -1.2374343872070312, 4.1163330078125, 0.4287681579589844, 0.03559112548828125, -0.9016876220703125, 2.761871337890625, 0.48983001708984375, -1.7300262451171875, -0.2155914306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000084.npy"}
|
|
{"epoch": 0.12698412698412698, "step": 85, "batch_size": 64, "mean": 1.3195340633392334, "std": 2.0516273975372314, "min": -3.609508514404297, "p10": -1.5371662139892577, "median": 1.2301673889160156, "p90": 3.776522445678711, "max": 7.3465423583984375, "pos_frac": 0.78125, "sample": [2.663785934448242, -1.5815315246582031, 1.2161598205566406, 2.126039505004883, 0.972442626953125, 0.8705062866210938, 1.8934097290039062, 0.6702251434326172, -0.1542224884033203, 3.750591278076172, 3.48052978515625, 1.56890869140625, 2.1675758361816406, 0.7827911376953125, 1.330902099609375, 1.2779388427734375, 1.6839408874511719, 2.5996932983398438, 2.7676773071289062, 2.4142837524414062, -1.8639297485351562, 2.3012962341308594, -2.848236083984375, 2.6137237548828125, 0.5209541320800781, 4.2750244140625, 0.26535797119140625, 1.3325729370117188, 4.929355621337891, -0.0022678375244140625, -1.6322021484375, 1.8876972198486328, 2.7494354248046875, 1.2503204345703125, 1.1661243438720703, 0.7174606323242188, 3.7876358032226562, 2.422119140625, 0.07587051391601562, -0.19179534912109375, -3.609508514404297, 1.4661941528320312, -0.16637039184570312, 1.9197807312011719, 7.3465423583984375, 3.2347660064697266, 5.39886474609375, 0.2563037872314453, 1.2340888977050781, 1.2262458801269531, 1.1903610229492188, -2.0335235595703125, 1.0110321044921875, -0.20470428466796875, 0.6621246337890625, 0.916259765625, -2.167816162109375, -1.210723876953125, 4.491157531738281, 0.4439506530761719, 6.217185974121094, -1.4336471557617188, 0.5393142700195312, 1.464132308959961], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000085.npy"}
|
|
{"epoch": 0.12849584278155707, "step": 86, "batch_size": 64, "mean": 1.038106083869934, "std": 2.0192461013793945, "min": -3.7711181640625, "p10": -1.553772735595703, "median": 1.0022602081298828, "p90": 3.682967567443849, "max": 5.835044860839844, "pos_frac": 0.703125, "sample": [5.835044860839844, 0.132110595703125, -1.7483749389648438, 0.08764457702636719, 2.0572357177734375, 0.09215927124023438, 0.072265625, 0.7907638549804688, -3.7711181640625, 1.1647186279296875, -0.3173809051513672, -1.9977188110351562, 4.130706787109375, 5.66729736328125, -0.5094070434570312, 3.1366195678710938, 1.5808963775634766, -0.71368408203125, -3.2037124633789062, 0.50006103515625, 4.605682373046875, -0.135772705078125, -0.331329345703125, 2.5369491577148438, 3.9983749389648438, 0.23986244201660156, 1.2238006591796875, 1.1790428161621094, 3.1484375, 0.5203914642333984, 2.724689483642578, 4.350341796875, 2.7221832275390625, -1.8390655517578125, 3.025482177734375, 1.9070510864257812, -1.462432861328125, 1.0165290832519531, -0.9049453735351562, 0.540679931640625, -1.5929183959960938, 2.7580528259277344, 1.820556640625, -0.2318248748779297, 1.2196807861328125, 0.20553970336914062, 3.3967113494873047, -0.9846725463867188, 2.0938796997070312, -0.4801197052001953, 2.2386932373046875, 3.8056488037109375, 0.140716552734375, -0.13133621215820312, 2.190013885498047, 3.379364013671875, -2.424612045288086, 2.0572738647460938, 1.4210662841796875, -0.3836822509765625, 0.9879913330078125, 0.5007209777832031, 1.3492851257324219, 1.0506820678710938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000086.npy"}
|
|
{"epoch": 0.13000755857898716, "step": 87, "batch_size": 64, "mean": 1.0899386405944824, "std": 2.4459664821624756, "min": -6.966766357421875, "p10": -1.8580387115478512, "median": 1.110260009765625, "p90": 4.200419998168947, "max": 6.6611328125, "pos_frac": 0.75, "sample": [1.3811225891113281, 5.478851318359375, 6.6611328125, 0.17910385131835938, 1.5646820068359375, 0.58306884765625, 3.09765625, -1.4466705322265625, 3.3876495361328125, 2.7108154296875, -2.061290740966797, -1.4270915985107422, 1.4301376342773438, -2.320158004760742, -6.966766357421875, 5.5586700439453125, -2.5511932373046875, 3.9391937255859375, 1.2972736358642578, 1.0898246765136719, -1.229827880859375, 0.40100860595703125, 1.6742286682128906, -1.2552490234375, 3.571239471435547, -4.069664001464844, 0.1144256591796875, -0.32946205139160156, 1.8982696533203125, -1.2282085418701172, -1.0055313110351562, 0.7067756652832031, 4.312374114990234, -2.0188217163085938, 1.551666259765625, -1.5522079467773438, 0.6688003540039062, 1.6420269012451172, 2.4111251831054688, 0.7857208251953125, 1.2538986206054688, 2.8301773071289062, 0.9253616333007812, -1.9891090393066406, 1.11724853515625, 1.2806205749511719, 1.0263233184814453, 2.8615570068359375, -0.41408538818359375, 3.1478118896484375, 3.2375717163085938, 1.4231719970703125, 5.1263885498046875, 0.09100341796875, 0.2833061218261719, 1.5237503051757812, 0.5484199523925781, 0.215545654296875, 6.030242919921875, 5.428413391113281, 1.103271484375, 1.1334915161132812, 0.6686859130859375, 2.2683029174804688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000087.npy"}
|
|
{"epoch": 0.13151927437641722, "step": 88, "batch_size": 64, "mean": 0.8546925783157349, "std": 3.0108108520507812, "min": -8.05072021484375, "p10": -2.4435333251953124, "median": 0.5598869323730469, "p90": 3.896372222900391, "max": 11.003280639648438, "pos_frac": 0.703125, "sample": [6.1839447021484375, 0.8129520416259766, -1.8859519958496094, -1.079437255859375, 6.834663391113281, 3.5796127319335938, 6.489410400390625, 0.4958763122558594, 1.1400222778320312, 0.03353691101074219, 3.495635986328125, 3.9538650512695312, 1.4397449493408203, 1.7332324981689453, -0.6747970581054688, 2.1345901489257812, 4.469696044921875, -1.2881355285644531, -2.5008544921875, 1.3733291625976562, 2.5317115783691406, 0.9750766754150391, 0.4915885925292969, -0.5712261199951172, -2.594226837158203, 1.7563858032226562, 0.30426788330078125, 11.003280639648438, -1.00390625, -2.688882827758789, -7.650970458984375, -2.7693862915039062, 3.6214141845703125, 0.5522308349609375, 0.6545181274414062, 0.01619720458984375, -1.2417373657226562, 2.1661720275878906, 1.2313728332519531, 1.0983905792236328, 0.31935882568359375, 2.76055908203125, 2.0254287719726562, 0.9309272766113281, -2.309783935546875, 0.2661590576171875, 0.3694572448730469, 0.43811607360839844, 0.3590087890625, -0.05920600891113281, 2.1399574279785156, 0.5675430297851562, 6.274444580078125, 3.7622222900390625, 2.9611587524414062, -2.031545639038086, 1.000640869140625, 0.4684886932373047, 0.3048095703125, 3.006204605102539, -1.077260971069336, -8.05072021484375, -1.6433773040771484, -2.7054710388183594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000088.npy"}
|
|
{"epoch": 0.1330309901738473, "step": 89, "batch_size": 64, "mean": 1.3538047075271606, "std": 2.5874903202056885, "min": -4.135643005371094, "p10": -0.8479164123535154, "median": 1.1087093353271484, "p90": 4.036691665649414, "max": 10.947265625, "pos_frac": 0.703125, "sample": [0.49769020080566406, 1.7851104736328125, -0.4534950256347656, 1.8107490539550781, 3.4166259765625, 4.033153533935547, -0.6487884521484375, 0.48938751220703125, 1.7299480438232422, 4.712066650390625, 2.015594482421875, 3.9912872314453125, 2.0819549560546875, -0.4150238037109375, 0.4033203125, 0.4373626708984375, -0.7167892456054688, 1.602508544921875, -0.16088104248046875, -2.9555130004882812, 3.14044189453125, -1.0363235473632812, -0.01617431640625, 3.6383438110351562, -0.5532684326171875, 10.947265625, 2.168384552001953, 0.8687896728515625, -0.6836433410644531, 4.86041259765625, -0.12267303466796875, 1.2628936767578125, 1.1933479309082031, 0.740478515625, 3.7732391357421875, 0.0049304962158203125, 3.39892578125, 1.0240707397460938, -0.90411376953125, 1.950103759765625, 2.927854537963867, 2.9232635498046875, 7.0575714111328125, 7.039457321166992, 0.9172458648681641, -0.03325462341308594, 0.16130447387695312, 3.1319732666015625, 4.0382080078125, -3.1921005249023438, -0.55059814453125, 2.6665420532226562, 0.09135055541992188, 0.2508964538574219, 0.40655517578125, -4.028724670410156, 1.6862030029296875, -0.613494873046875, -3.1757125854492188, -4.135643005371094, 4.8274383544921875, 1.462738037109375, 1.5892086029052734, 1.883514404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000089.npy"}
|
|
{"epoch": 0.1345427059712774, "step": 90, "batch_size": 64, "mean": 1.3068695068359375, "std": 2.4874353408813477, "min": -4.511627197265625, "p10": -1.611101150512695, "median": 1.0958175659179688, "p90": 4.10790138244629, "max": 7.50531005859375, "pos_frac": 0.75, "sample": [-0.8596153259277344, 3.2152137756347656, 0.945892333984375, 2.7240028381347656, 3.4533119201660156, 0.32320213317871094, 3.235200881958008, 6.9359588623046875, -3.4300994873046875, -4.511627197265625, 3.776092529296875, 3.772266387939453, 3.1971073150634766, 0.9790859222412109, 2.7703933715820312, -0.9580268859863281, 1.5435981750488281, 5.331085205078125, -1.7009773254394531, 1.1109886169433594, 3.7883377075195312, -1.4013900756835938, 7.50531005859375, 4.170101165771484, 5.380157470703125, 1.0806465148925781, 1.3378372192382812, 4.294647216796875, 0.4692554473876953, 1.00299072265625, 0.32819366455078125, -0.15728759765625, 1.8008975982666016, 6.4797515869140625, 1.3687934875488281, 3.6951828002929688, -2.5477294921875, -1.2322921752929688, 3.9627685546875, 1.8247013092041016, -1.2625961303710938, 0.52752685546875, 1.8546142578125, -1.181610107421875, 3.878143310546875, 0.2498779296875, 0.9177665710449219, 0.3274402618408203, 0.29340553283691406, -0.4498481750488281, -3.144847869873047, 1.5620269775390625, 1.8457107543945312, 0.110382080078125, 0.3998241424560547, 1.1347770690917969, 0.38832855224609375, -3.1891841888427734, -2.1562423706054688, 2.6012916564941406, 2.1403579711914062, 1.7838134765625, -0.5888462066650391, 0.5936107635498047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000090.npy"}
|
|
{"epoch": 0.1360544217687075, "step": 91, "batch_size": 64, "mean": 1.1210139989852905, "std": 2.783144474029541, "min": -6.311534881591797, "p10": -1.8404340744018555, "median": 0.546051025390625, "p90": 4.471922302246094, "max": 11.360565185546875, "pos_frac": 0.65625, "sample": [3.193634033203125, 3.6080474853515625, 1.0667285919189453, -1.7743644714355469, 0.3947601318359375, 6.682830810546875, 0.38562774658203125, 1.304046630859375, 3.1212692260742188, 6.816215515136719, 11.360565185546875, -0.7186965942382812, 0.8436412811279297, -1.3442497253417969, 1.6509761810302734, -2.016237258911133, -0.8033294677734375, -0.16120147705078125, 3.0360488891601562, -2.859161376953125, 1.954620361328125, 2.9988136291503906, 0.5398406982421875, 0.48856163024902344, 1.2708053588867188, 2.0639266967773438, 3.1153488159179688, -1.8687496185302734, -3.3735809326171875, 3.5797805786132812, 0.7930679321289062, -1.7171134948730469, -2.1425628662109375, 1.300405502319336, 4.922727584838867, -0.6839103698730469, -0.344940185546875, 2.848358154296875, -0.8124637603759766, -6.311534881591797, 2.261383056640625, 0.537139892578125, -2.3675384521484375, -1.076568603515625, 0.3983612060546875, 0.5020904541015625, 0.5522613525390625, 5.286243438720703, 2.3029327392578125, 1.4656143188476562, -0.24781036376953125, -1.6588878631591797, -0.03475379943847656, -0.15546226501464844, 4.321136474609375, 3.2439842224121094, 5.9853515625, 0.04500007629394531, 4.5365447998046875, 2.619251251220703, -1.48773193359375, 0.25809478759765625, 1.7248077392578125, 0.32489776611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000091.npy"}
|
|
{"epoch": 0.13756613756613756, "step": 92, "batch_size": 64, "mean": 1.5593235492706299, "std": 3.159080982208252, "min": -5.130199432373047, "p10": -1.8206445693969726, "median": 0.7258205413818359, "p90": 5.610137939453125, "max": 11.733108520507812, "pos_frac": 0.671875, "sample": [3.35028076171875, -0.2794971466064453, -3.5656051635742188, 0.0073337554931640625, -0.8897933959960938, -0.29860687255859375, 0.207550048828125, 0.6177635192871094, 1.2782058715820312, -0.7464523315429688, -2.052947998046875, 1.3077564239501953, 3.3621368408203125, -1.387542724609375, 4.744573593139648, 2.4358158111572266, 2.2775001525878906, 0.029096603393554688, 5.367547988891602, 0.5707511901855469, 3.53680419921875, -0.47144317626953125, 1.5276565551757812, 0.5880813598632812, 4.465911865234375, -0.005756378173828125, -3.316558837890625, 1.6217117309570312, -4.137702941894531, 3.2281723022460938, -1.8775005340576172, 3.0369491577148438, 6.236534118652344, 0.7756233215332031, 3.7272186279296875, 0.6760177612304688, 0.2498016357421875, 3.880126953125, -3.4205856323242188, 2.4461593627929688, -0.5853500366210938, -0.18255615234375, 5.628082275390625, 2.5389404296875, 2.0287818908691406, -1.3857498168945312, 4.061714172363281, 5.6906890869140625, 3.790851593017578, -1.6879806518554688, -0.5498695373535156, 10.033905029296875, 1.3870010375976562, -0.211212158203125, 5.568267822265625, -5.130199432373047, 11.733108520507812, 4.474571228027344, 0.4523143768310547, 6.58758544921875, 6.026683807373047, 0.6121711730957031, 0.0703277587890625, -0.25846099853515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000092.npy"}
|
|
{"epoch": 0.13907785336356765, "step": 93, "batch_size": 64, "mean": 1.0133929252624512, "std": 3.214465856552124, "min": -5.3485870361328125, "p10": -2.615689277648926, "median": 0.5049276351928711, "p90": 5.03322982788086, "max": 14.030746459960938, "pos_frac": 0.625, "sample": [0.6728801727294922, 2.8576393127441406, 4.772369384765625, 3.953815460205078, -2.6558494567871094, 3.6246376037597656, -2.4549026489257812, -0.814727783203125, -0.20679855346679688, 2.58428955078125, 1.433084487915039, 2.1793270111083984, 14.030746459960938, 0.4248924255371094, -0.3736858367919922, 0.13680076599121094, 0.24607086181640625, 3.4001426696777344, -0.7588634490966797, -5.3485870361328125, 2.7511138916015625, 6.955780029296875, -2.2044906616210938, -0.45290374755859375, 2.440715789794922, -0.3806037902832031, 5.145027160644531, 0.04478263854980469, -1.6056480407714844, 1.4571990966796875, -0.556976318359375, -0.03291511535644531, -0.45125770568847656, 5.145561218261719, -1.309906005859375, -1.0116043090820312, 0.10703277587890625, 0.138153076171875, -5.188392639160156, 0.181304931640625, 1.0674018859863281, 0.4817543029785156, 1.6907634735107422, 1.9188957214355469, 1.8410224914550781, -3.4728240966796875, 0.7542724609375, 0.5281009674072266, 5.682647705078125, 1.2866897583007812, 4.190364837646484, 1.419900894165039, 4.745796203613281, 2.2841720581054688, -2.10076904296875, -2.521982192993164, 5.378032684326172, -4.954505920410156, -3.109466552734375, 5.485542297363281, 4.062339782714844, -3.2720489501953125, -0.5081100463867188, 3.103900909423828], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000093.npy"}
|
|
{"epoch": 0.14058956916099774, "step": 94, "batch_size": 64, "mean": 1.8918602466583252, "std": 4.034669876098633, "min": -10.473114013671875, "p10": -1.2660871505737301, "median": 1.7948017120361328, "p90": 5.791851043701173, "max": 14.29742431640625, "pos_frac": 0.765625, "sample": [3.738788604736328, 0.6126022338867188, 2.3691787719726562, 3.3294525146484375, 1.3003311157226562, -8.636287689208984, 0.07506179809570312, 0.7152557373046875, 1.1033554077148438, 2.7390060424804688, 0.6739749908447266, 5.5833282470703125, 1.547607421875, -8.480283737182617, 2.6300430297851562, 5.279541015625, -1.7436256408691406, -0.7914695739746094, 6.2585601806640625, 2.6682968139648438, 2.517078399658203, 2.3591537475585938, 3.4544830322265625, 1.6755027770996094, 2.3974456787109375, 1.2133369445800781, 0.25823020935058594, 0.6810073852539062, -3.2575225830078125, 3.9171676635742188, 3.6086864471435547, -0.9155063629150391, -0.03454399108886719, -0.432525634765625, 2.9241600036621094, 3.0230331420898438, 3.85662841796875, 2.6633148193359375, 4.213092803955078, 0.9495849609375, 5.881217956542969, 6.4208831787109375, 0.022861480712890625, 12.957550048828125, 4.368305206298828, 3.9804534912109375, -0.8533401489257812, 10.536407470703125, 3.7059326171875, 2.411012649536133, 14.29742431640625, 0.8153209686279297, -0.30957794189453125, -0.8363552093505859, 1.9811058044433594, 0.7944526672363281, -10.473114013671875, 1.9141006469726562, 9.397964477539062, 1.4928550720214844, 0.4279956817626953, -1.4163360595703125, -2.0784778594970703, -0.4041175842285156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000094.npy"}
|
|
{"epoch": 0.1421012849584278, "step": 95, "batch_size": 64, "mean": 1.5772008895874023, "std": 3.423574924468994, "min": -6.0201416015625, "p10": -2.592078399658203, "median": 1.1513099670410156, "p90": 6.392279052734375, "max": 9.210556030273438, "pos_frac": 0.65625, "sample": [5.7455902099609375, -3.0086517333984375, 1.2815628051757812, -0.7680397033691406, -2.4718246459960938, 0.020624160766601562, 9.210556030273438, 0.6338348388671875, 2.8812026977539062, 1.767547607421875, -1.2177619934082031, 1.3712005615234375, 1.0576019287109375, -0.3073616027832031, -0.6659164428710938, 1.1840286254882812, -4.0208587646484375, -1.92462158203125, -0.3896656036376953, 1.8811302185058594, 7.2850189208984375, 6.438385009765625, -0.466827392578125, 0.54949951171875, -5.718658447265625, -0.881988525390625, 7.57232666015625, 6.284698486328125, -3.258392333984375, -6.0201416015625, 2.6103515625, 6.970916748046875, 1.1078910827636719, 2.3919830322265625, -0.6469688415527344, 1.11859130859375, 5.676948547363281, -0.51678466796875, -3.5411605834960938, 2.677928924560547, 3.73876953125, 0.37183380126953125, 5.718475341796875, 2.7150840759277344, -2.64361572265625, 5.018287658691406, 5.259521484375, 0.06424331665039062, 1.7718544006347656, 5.550331115722656, -0.5297775268554688, 4.132225036621094, -0.6700897216796875, 3.2229156494140625, 5.296009063720703, 8.530685424804688, -1.1286087036132812, 0.237152099609375, 1.789510726928711, 3.4090805053710938, 0.5930252075195312, 2.1046276092529297, 6.479034423828125, -1.9835128784179688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000095.npy"}
|
|
{"epoch": 0.1436130007558579, "step": 96, "batch_size": 64, "mean": 1.8757414817810059, "std": 4.527604579925537, "min": -7.9029693603515625, "p10": -2.585195159912109, "median": 1.016179084777832, "p90": 7.89864959716797, "max": 15.742279052734375, "pos_frac": 0.6875, "sample": [3.2734451293945312, -2.6872787475585938, 7.62908935546875, -0.40161895751953125, 12.027069091796875, -7.9029693603515625, 3.309478759765625, 0.11998748779296875, 2.5748062133789062, 0.9480667114257812, 4.9520111083984375, -0.6299266815185547, 0.8670883178710938, 10.286483764648438, 3.1089935302734375, 1.7645416259765625, -5.183692932128906, 0.1077423095703125, 1.073171615600586, -0.4110755920410156, 2.249492645263672, 2.2954330444335938, 8.282371520996094, -5.2324066162109375, 0.49541473388671875, 0.8881931304931641, 5.496574401855469, 0.0454559326171875, -2.3470001220703125, -0.1338653564453125, 2.443035125732422, 8.014175415039062, 15.742279052734375, 5.374908447265625, -1.1601371765136719, -1.9221649169921875, -7.531486511230469, 0.5622539520263672, -3.1207199096679688, 1.7729454040527344, 3.511932373046875, 0.9591865539550781, 1.868631362915039, 0.23644638061523438, 13.233245849609375, -1.262664794921875, 3.8593711853027344, -0.4663848876953125, 1.8079261779785156, -0.782257080078125, 4.175621032714844, -1.5980644226074219, 0.8609809875488281, 7.017265319824219, 10.434402465820312, 2.69622802734375, -1.1575336456298828, 0.6748466491699219, 1.4374809265136719, 3.5618438720703125, 6.8872222900390625, 1.9160690307617188, -1.4359626770019531, -5.428539276123047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000096.npy"}
|
|
{"epoch": 0.14512471655328799, "step": 97, "batch_size": 64, "mean": 1.0323022603988647, "std": 5.044511318206787, "min": -16.7403564453125, "p10": -3.112880325317383, "median": 0.8853368759155273, "p90": 6.204517364501957, "max": 17.358306884765625, "pos_frac": 0.640625, "sample": [-10.762924194335938, -2.3804244995117188, 0.4189338684082031, 7.404886245727539, 0.7824974060058594, 13.444007873535156, -2.8603248596191406, -3.1400909423828125, -2.3444747924804688, -16.7403564453125, -3.049388885498047, -1.0990066528320312, 0.6163787841796875, 2.655252456665039, -4.0123443603515625, -9.021377563476562, 0.3076667785644531, 1.0927886962890625, 2.70977783203125, 0.4116325378417969, 0.9881763458251953, 3.3356094360351562, 1.6394767761230469, -2.4633026123046875, 2.936431884765625, -1.5806427001953125, -1.2165603637695312, 5.38433837890625, -1.5412445068359375, 11.293373107910156, 2.075714111328125, -2.5206756591796875, 3.2591094970703125, 0.5751991271972656, 0.7484779357910156, 4.4879913330078125, 7.842784881591797, 6.556022644042969, -1.0402545928955078, 4.1405029296875, 3.0087890625, -5.473789215087891, 17.358306884765625, 2.6823577880859375, 2.9209671020507812, -1.2540779113769531, 3.426790237426758, 3.7496337890625, 9.867279052734375, -3.1444625854492188, 0.23461151123046875, 2.2900390625, 1.5954742431640625, -1.6284942626953125, -1.4899749755859375, 2.998626708984375, 3.986879348754883, -2.1667404174804688, 2.780672073364258, 2.3156814575195312, 1.1053199768066406, 0.00595855712890625, 2.3677330017089844, -0.8038673400878906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000097.npy"}
|
|
{"epoch": 0.14663643235071808, "step": 98, "batch_size": 64, "mean": 0.6471325755119324, "std": 3.2106478214263916, "min": -8.37786865234375, "p10": -2.9695507049560548, "median": 0.9861698150634766, "p90": 5.0036445617675795, "max": 7.0745391845703125, "pos_frac": 0.640625, "sample": [0.2488555908203125, 1.4308109283447266, 5.154411315917969, 0.8095932006835938, 2.2534103393554688, 2.0934486389160156, 2.0744247436523438, 2.0604286193847656, -1.0920143127441406, 6.2307281494140625, -8.37786865234375, -3.1317977905273438, -1.9601593017578125, -1.1680221557617188, -2.7061691284179688, 0.03102874755859375, -0.2469482421875, 1.9723129272460938, 1.0237846374511719, -0.7208099365234375, 3.4484634399414062, 6.690773010253906, -2.97381591796875, 5.276702880859375, 0.6912345886230469, -1.981210708618164, 3.771434783935547, 0.9443855285644531, 0.9485549926757812, 1.2361412048339844, 1.8599700927734375, 1.0581130981445312, 5.6215362548828125, 1.854888916015625, -3.2909278869628906, 1.3175048828125, 1.510894775390625, 2.749114990234375, 0.8273639678955078, -0.3051643371582031, 2.4169464111328125, -0.07411956787109375, -8.023834228515625, 1.5258216857910156, -1.8507080078125, -3.8065013885498047, 2.02374267578125, 1.2598819732666016, -1.8990478515625, 5.448032379150391, -1.9271469116210938, 4.65185546875, -2.1035003662109375, -2.2263545989990234, -7.87298583984375, 3.2927703857421875, 2.1664657592773438, 0.2761688232421875, 0.6524696350097656, 7.0745391845703125, 2.695423126220703, 3.8226242065429688, -0.381866455078125, -2.9595985412597656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000098.npy"}
|
|
{"epoch": 0.14814814814814814, "step": 99, "batch_size": 64, "mean": 1.0689595937728882, "std": 4.5391154289245605, "min": -12.4384765625, "p10": -3.227545166015625, "median": 0.8946685791015625, "p90": 6.6735683441162115, "max": 14.17041015625, "pos_frac": 0.625, "sample": [8.240348815917969, 8.015010833740234, 6.806404113769531, 11.452232360839844, 0.17679977416992188, -2.8294830322265625, -0.3938255310058594, -1.6432437896728516, 0.726165771484375, 6.736598968505859, 2.784332275390625, 1.2653656005859375, 1.9843864440917969, 4.598102569580078, -0.4723052978515625, 1.8863372802734375, 1.0589752197265625, 0.1738910675048828, -1.1488227844238281, 1.7391910552978516, 6.201026916503906, -0.01435089111328125, -5.561882019042969, -2.478607177734375, -5.3759307861328125, -3.2799110412597656, 14.17041015625, -8.95745849609375, -2.272369384765625, 6.526496887207031, 1.802093505859375, -2.431732177734375, -0.5300140380859375, -0.0812835693359375, 0.7621116638183594, 1.2677764892578125, 1.4691085815429688, -3.105358123779297, 0.1049346923828125, 5.605621337890625, 1.0334968566894531, -12.4384765625, 5.652301788330078, 10.256072998046875, -0.5198974609375, 1.9859085083007812, 5.784461975097656, 0.6716136932373047, -8.28533935546875, 1.4520282745361328, 0.042636871337890625, 3.570758819580078, 2.167593002319336, 1.0272254943847656, 2.2999420166015625, -0.00756072998046875, 0.4038810729980469, 4.67620849609375, 1.5998668670654297, -0.8734817504882812, 1.7566757202148438, -2.4391746520996094, -2.0888137817382812, -4.291656494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000099.npy"}
|
|
{"epoch": 0.14965986394557823, "step": 100, "batch_size": 64, "mean": 1.468209981918335, "std": 6.080421447753906, "min": -16.160781860351562, "p10": -3.67082290649414, "median": 1.0627479553222656, "p90": 7.4635406494140675, "max": 25.6746826171875, "pos_frac": 0.65625, "sample": [2.082000732421875, -1.3315277099609375, -3.8725433349609375, 3.0995330810546875, 7.9433135986328125, -1.4788360595703125, 0.40740394592285156, 3.276121139526367, 5.787956237792969, 4.2590179443359375, 2.0007781982421875, 1.321502685546875, 4.184471130371094, 12.310905456542969, 1.4353561401367188, 4.2417449951171875, -1.4114227294921875, 1.5718612670898438, 0.5366096496582031, 0.9764862060546875, 0.7227115631103516, 0.250885009765625, -5.606586456298828, 0.3499736785888672, -4.072479248046875, 4.6602325439453125, -1.4458389282226562, 8.214462280273438, 1.1490097045898438, -14.06866455078125, -0.8387908935546875, -0.2784881591796875, -2.7646636962890625, 0.8574943542480469, 0.7318649291992188, -0.24734878540039062, 4.539337158203125, -0.294708251953125, -13.070976257324219, 5.534013748168945, 1.3306884765625, 17.573074340820312, -4.5768585205078125, 0.8112907409667969, -1.7636756896972656, -0.7890472412109375, 2.6970252990722656, 1.300323486328125, -1.0013809204101562, -0.7835159301757812, 6.3440704345703125, 8.554733276367188, -0.49524688720703125, 2.908405303955078, 2.4023590087890625, 1.4416961669921875, -3.2001419067382812, 11.86474609375, 4.482353210449219, 1.91204833984375, 25.6746826171875, 0.3143310546875, -16.160781860351562, 1.4620819091796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000100.npy"}
|
|
{"epoch": 0.15117157974300832, "step": 101, "batch_size": 64, "mean": 2.1045989990234375, "std": 4.245602130889893, "min": -11.719009399414062, "p10": -2.281474685668945, "median": 1.5141754150390625, "p90": 6.880474090576172, "max": 12.883819580078125, "pos_frac": 0.75, "sample": [-1.3664436340332031, -0.0325927734375, 3.2113265991210938, 0.2382659912109375, 6.9143829345703125, 6.390235900878906, 4.373054504394531, 1.0640029907226562, -0.29437255859375, -0.19517135620117188, 0.7279930114746094, 4.133735656738281, -11.719009399414062, 8.019569396972656, 4.358158111572266, 0.5371055603027344, 4.554084777832031, 2.3974857330322266, -1.0550117492675781, -3.1646862030029297, 4.184810638427734, 1.1866378784179688, 6.0592193603515625, -1.8011817932128906, 3.523700714111328, -1.8170928955078125, 4.2496185302734375, 3.7119903564453125, 1.0283870697021484, 6.253894805908203, 1.4455947875976562, 10.289886474609375, 6.039285659790039, 0.5829849243164062, -0.7298507690429688, -2.5252513885498047, 1.1633071899414062, -7.067695617675781, -2.4804954528808594, 4.044490814208984, -4.135078430175781, 4.1576995849609375, 1.1786994934082031, 1.5827560424804688, 6.801353454589844, 2.046112060546875, 0.5933074951171875, -8.864120483398438, 0.1853466033935547, 1.4000568389892578, 7.9037628173828125, 0.4634876251220703, 7.309486389160156, 3.030149459838867, 3.8533401489257812, 0.7395782470703125, 12.883819580078125, 2.9036178588867188, -0.4254188537597656, 2.744335174560547, 6.43402099609375, 3.029998779296875, 11.073410034179688, 1.3702583312988281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000101.npy"}
|
|
{"epoch": 0.15268329554043839, "step": 102, "batch_size": 64, "mean": 0.828357994556427, "std": 5.485098361968994, "min": -10.4454345703125, "p10": -4.105657958984375, "median": 0.3138160705566406, "p90": 6.807697296142578, "max": 21.6953125, "pos_frac": 0.53125, "sample": [7.288703918457031, -3.6568450927734375, -2.4221725463867188, 1.1713790893554688, -7.248054504394531, -1.3648872375488281, -4.016288757324219, 2.7036170959472656, -7.719184875488281, -0.137176513671875, 2.2822723388671875, 3.8085861206054688, 0.32143402099609375, 1.7174568176269531, 6.806121826171875, -3.4513015747070312, -2.3978214263916016, 16.731796264648438, -3.0962448120117188, 10.166236877441406, -2.846712112426758, 0.6090145111083984, -2.6973419189453125, -1.4832382202148438, 2.7783279418945312, 2.776762008666992, 0.6998443603515625, 6.507656097412109, -9.7113037109375, 1.1434345245361328, 6.808372497558594, 1.0712928771972656, -4.143959045410156, 0.9541740417480469, -3.9130096435546875, -1.6114654541015625, 3.53985595703125, 7.657810211181641, 10.790237426757812, -2.741413116455078, -0.10534477233886719, 2.3279266357421875, -10.4454345703125, -0.81060791015625, -1.1683998107910156, -0.16643524169921875, 0.3061981201171875, 5.402259826660156, 0.04337501525878906, -1.68426513671875, 2.136068344116211, 5.139793395996094, 1.7928009033203125, -4.630441665649414, -1.4153213500976562, -3.956775665283203, -1.4204139709472656, 21.6953125, -2.489532470703125, -7.1436309814453125, 3.961212158203125, 5.398284912109375, 5.500644683837891, 1.0716743469238281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000102.npy"}
|
|
{"epoch": 0.15419501133786848, "step": 103, "batch_size": 64, "mean": 1.5714372396469116, "std": 3.6922335624694824, "min": -6.695411682128906, "p10": -2.38699836730957, "median": 1.2182579040527344, "p90": 6.91147041320801, "max": 10.556549072265625, "pos_frac": 0.609375, "sample": [-3.1455001831054688, -0.8328399658203125, 7.736236572265625, -6.695411682128906, -0.21965980529785156, 3.5002288818359375, -1.088470458984375, -0.7550029754638672, 1.4285812377929688, 7.1498870849609375, -2.333141326904297, 2.0625534057617188, -1.7166976928710938, 1.4085731506347656, -1.8174591064453125, 5.1561126708984375, 2.6444568634033203, -0.5672683715820312, -3.954437255859375, 6.178409576416016, 1.59912109375, 4.518024444580078, -0.08228492736816406, 2.559192657470703, 2.190948486328125, -0.8010139465332031, -0.5976371765136719, 9.049827575683594, -2.4100799560546875, 2.790081024169922, 3.6323928833007812, 6.35125732421875, 8.8739013671875, -1.403818130493164, 0.9435501098632812, 3.9150772094726562, -5.93017578125, 3.65478515625, 1.2705764770507812, 0.6738166809082031, 5.197242736816406, 1.108184814453125, 0.8173770904541016, 2.7508010864257812, 3.406524658203125, -0.062774658203125, 6.362285614013672, -2.2177963256835938, 7.1468353271484375, -0.3141746520996094, 8.689056396484375, 2.9078903198242188, -4.738929748535156, 0.3720970153808594, -3.1722259521484375, 0.21862030029296875, -1.552703857421875, -1.4536781311035156, -1.1287689208984375, 1.1659393310546875, 3.18255615234375, 3.3316192626953125, 10.556549072265625, 3.0627670288085938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000103.npy"}
|
|
{"epoch": 0.15570672713529857, "step": 104, "batch_size": 64, "mean": 2.2744176387786865, "std": 6.480139255523682, "min": -11.953193664550781, "p10": -4.414015197753905, "median": 1.1356945037841797, "p90": 9.180105590820315, "max": 28.696075439453125, "pos_frac": 0.65625, "sample": [9.417282104492188, 4.9798583984375, 3.8005905151367188, -2.0803966522216797, -4.946197509765625, 2.7449569702148438, -1.481180191040039, 1.1550445556640625, 4.9397735595703125, -0.729888916015625, 0.7103424072265625, -11.953193664550781, 28.696075439453125, 7.4253692626953125, -6.420173645019531, -1.9979019165039062, 13.31695556640625, 2.1386451721191406, -0.7837791442871094, 7.803138732910156, -4.9176788330078125, -9.586944580078125, 5.78265380859375, 8.626693725585938, -0.4747467041015625, -0.4297618865966797, 0.969696044921875, -7.296287536621094, 4.356328964233398, -0.33868980407714844, 0.5016155242919922, 0.35548973083496094, -0.889007568359375, 2.4908485412597656, 8.291736602783203, 2.7830543518066406, 1.0716705322265625, -1.0039443969726562, 3.12066650390625, 2.3853073120117188, 1.1542282104492188, -2.7656116485595703, 18.687744140625, 0.6104984283447266, 3.4579124450683594, 11.564754486083984, 3.3016281127929688, 1.1171607971191406, -9.079048156738281, 0.250152587890625, 1.861673355102539, 3.1777076721191406, 6.431758880615234, 4.512611389160156, 1.1114215850830078, -3.238800048828125, -0.062255859375, -1.5681610107421875, 0.6872043609619141, 1.398773193359375, -2.81903076171875, 5.871635437011719, 11.623504638671875, 15.741249084472656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000104.npy"}
|
|
{"epoch": 0.15721844293272866, "step": 105, "batch_size": 64, "mean": 1.7572441101074219, "std": 5.000411033630371, "min": -8.029937744140625, "p10": -4.195858764648436, "median": 1.269333839416504, "p90": 7.673465728759767, "max": 18.936904907226562, "pos_frac": 0.625, "sample": [1.8557510375976562, 5.1786651611328125, 2.566864013671875, 3.5775184631347656, 1.2853336334228516, 4.878692626953125, -7.9842071533203125, 10.487823486328125, -3.3184661865234375, 2.366292953491211, -0.43701171875, 4.341005325317383, -1.6397705078125, 7.906227111816406, 0.4010581970214844, 8.2884521484375, -0.9990272521972656, -6.358856201171875, -0.5691661834716797, 4.2682342529296875, 3.1534194946289062, -1.5208892822265625, 1.2637252807617188, -4.5300445556640625, 0.7627487182617188, 4.054443359375, 0.5589942932128906, -5.3005523681640625, 3.00634765625, -6.6711578369140625, -2.1499481201171875, 0.5675449371337891, -1.931884765625, 1.7927932739257812, -8.029937744140625, 18.936904907226562, 1.274942398071289, -0.001678466796875, 2.3805999755859375, 0.1425037384033203, -0.5647430419921875, -1.4232597351074219, -0.2054290771484375, 1.925384521484375, -5.89996337890625, 5.072544097900391, 4.65130615234375, -0.8795166015625, 6.1104736328125, -2.42926025390625, -3.4160919189453125, 2.4731521606445312, 0.8111534118652344, 11.119766235351562, 13.759552001953125, 0.5851974487304688, 7.1303558349609375, 5.755340576171875, -0.28281211853027344, 2.8141326904296875, -1.2461700439453125, 10.455528259277344, 5.3418121337890625, 6.9508819580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000105.npy"}
|
|
{"epoch": 0.15873015873015872, "step": 106, "batch_size": 64, "mean": 1.586924433708191, "std": 5.317921161651611, "min": -16.47613525390625, "p10": -5.311459350585936, "median": 1.664907455444336, "p90": 6.148886108398439, "max": 15.689453125, "pos_frac": 0.6875, "sample": [-5.8488006591796875, 4.972103118896484, 3.4776172637939453, 5.246826171875, -6.4429931640625, -0.7500381469726562, 8.829025268554688, 2.9321823120117188, 1.2253952026367188, 0.8331375122070312, 4.851860046386719, 6.275238037109375, 10.505569458007812, 3.6493377685546875, 0.5940799713134766, -4.0576629638671875, 1.9796943664550781, 5.562095642089844, 3.2629165649414062, 0.9582061767578125, -0.7695999145507812, 0.3222160339355469, -7.753692626953125, 2.5191268920898438, -0.8462762832641602, -0.31850433349609375, -0.5277156829833984, -3.27947998046875, 1.786163330078125, 15.689453125, -0.6532363891601562, -0.18018722534179688, -1.7512893676757812, -2.6981124877929688, 5.819793701171875, 5.2315216064453125, 0.5668087005615234, -1.1439857482910156, 5.435760498046875, 15.438873291015625, 4.296241760253906, 2.5826454162597656, 4.768314361572266, 0.7612457275390625, -6.383110046386719, 1.1570892333984375, 1.6361560821533203, 1.6273078918457031, -16.47613525390625, 1.6936588287353516, -10.231170654296875, 1.8152351379394531, 4.8413848876953125, 4.57502555847168, 5.85406494140625, 8.715858459472656, 0.858123779296875, 2.5874252319335938, -0.7048797607421875, 10.251083374023438, -8.207015991210938, 1.886627197265625, 0.4686737060546875, 2.2458934783935547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000106.npy"}
|
|
{"epoch": 0.1602418745275888, "step": 107, "batch_size": 64, "mean": 0.8766939640045166, "std": 6.273787975311279, "min": -17.574951171875, "p10": -5.601045989990234, "median": 1.4294281005859375, "p90": 7.452977752685548, "max": 19.043655395507812, "pos_frac": 0.671875, "sample": [1.1355857849121094, -5.748008728027344, 2.4628143310546875, 0.3375816345214844, 7.089759826660156, 3.919403076171875, -17.02435302734375, 6.6340179443359375, -0.4076652526855469, 1.6619796752929688, -17.574951171875, 0.07750701904296875, 2.1093368530273438, 1.4679718017578125, 3.9437999725341797, -3.8937854766845703, -3.40020751953125, 4.6641082763671875, 3.7752609252929688, 0.3785247802734375, -1.5449752807617188, 10.859161376953125, 9.127906799316406, 0.6655635833740234, 4.173191070556641, -9.445945739746094, 3.8273391723632812, -2.5668869018554688, -1.4006881713867188, 1.3908843994140625, 2.9474754333496094, -7.112480163574219, 1.9044570922851562, 5.407718658447266, 1.0369949340820312, 5.7466888427734375, 6.141845703125, 1.196502685546875, 2.995758056640625, 9.387603759765625, 12.3135986328125, 7.608642578125, -4.744377136230469, -2.5045928955078125, 3.365192413330078, 0.7385177612304688, 0.11063766479492188, 3.1679840087890625, 2.549224853515625, -11.525146484375, 8.995986938476562, 3.0062713623046875, -2.425670623779297, -3.5410232543945312, -0.10941505432128906, 0.4857940673828125, 3.075408935546875, 19.043655395507812, 2.7293262481689453, -2.4135894775390625, -5.2581329345703125, -14.026443481445312, 2.2748947143554688, -3.1551284790039062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000107.npy"}
|
|
{"epoch": 0.1617535903250189, "step": 108, "batch_size": 64, "mean": 1.8981385231018066, "std": 5.445388317108154, "min": -11.7947998046875, "p10": -4.609212493896484, "median": 1.344919204711914, "p90": 8.047893524169925, "max": 18.75299072265625, "pos_frac": 0.609375, "sample": [2.4760208129882812, 5.203704833984375, 10.617660522460938, -0.8382034301757812, -0.31870269775390625, 5.358207702636719, 4.3056640625, 4.118688583374023, 5.868072509765625, -5.0432281494140625, 1.2586746215820312, 5.640380859375, 3.51434326171875, -1.3744239807128906, 5.823638916015625, 2.1222457885742188, -3.69476318359375, -0.3707714080810547, 0.782501220703125, 0.9280967712402344, 7.292388916015625, -0.20785140991210938, -0.2637748718261719, 4.859466552734375, 1.3127670288085938, -6.1300811767578125, 3.9507179260253906, -11.7947998046875, 12.77520751953125, 0.3823699951171875, -9.194610595703125, -2.0981063842773438, 7.245964050292969, -0.6733818054199219, 8.371681213378906, 5.498659133911133, 0.03270530700683594, -0.2464447021484375, -6.2541046142578125, 5.3333740234375, -3.7919540405273438, 8.836997985839844, 2.3484573364257812, -1.3744239807128906, -0.5568313598632812, 1.966766357421875, -1.4289302825927734, 2.969514846801758, 1.8449172973632812, -4.792686462402344, -7.355068206787109, 3.2077484130859375, -4.1811065673828125, 2.669281005859375, 11.420440673828125, 0.0857086181640625, 3.3470230102539062, 18.75299072265625, 1.3770713806152344, -0.36279296875, -0.35010719299316406, 5.450538635253906, -1.1502456665039062, 15.97760009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000108.npy"}
|
|
{"epoch": 0.16326530612244897, "step": 109, "batch_size": 64, "mean": 2.995046615600586, "std": 5.387655258178711, "min": -5.4530181884765625, "p10": -1.4162607192993164, "median": 1.6468009948730469, "p90": 8.130487823486328, "max": 24.655120849609375, "pos_frac": 0.6875, "sample": [-0.05052947998046875, 13.013771057128906, 15.927978515625, 0.6167182922363281, 4.37647819519043, 3.8683719635009766, 7.169826507568359, -0.1791534423828125, 2.0211944580078125, 4.691986083984375, -0.2863178253173828, 1.3270492553710938, 1.8787574768066406, 5.894718170166016, 1.5318145751953125, -3.72906494140625, 2.4396591186523438, 0.3746528625488281, -1.0322303771972656, 0.9407234191894531, 1.4169025421142578, 4.923557281494141, -3.48089599609375, -0.3608388900756836, -0.7274017333984375, 24.655120849609375, 0.3934898376464844, -0.24910545349121094, 3.8981189727783203, 19.818634033203125, 3.6093368530273438, 1.1904144287109375, 1.6524009704589844, 0.17953109741210938, 0.2749652862548828, 2.4645442962646484, -1.3555316925048828, 5.1974334716796875, -0.7316360473632812, 4.474189758300781, -1.3407249450683594, 4.088569641113281, 1.6412010192871094, 6.1553955078125, -0.6281604766845703, -1.4422874450683594, -0.4460182189941406, 1.6528491973876953, -5.4530181884765625, -3.5937957763671875, 1.3660736083984375, 3.5928802490234375, 5.591552734375, 8.091827392578125, 8.147056579589844, 14.173507690429688, 11.042108535766602, 1.8423004150390625, 5.5654296875, 3.312347412109375, -3.1188201904296875, -0.9286136627197266, -1.788818359375, 6.120513916015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000109.npy"}
|
|
{"epoch": 0.16477702191987906, "step": 110, "batch_size": 64, "mean": 0.24830776453018188, "std": 5.455912113189697, "min": -20.07232666015625, "p10": -5.314604949951171, "median": 0.6546039581298828, "p90": 6.540238189697267, "max": 14.07110595703125, "pos_frac": 0.59375, "sample": [1.941314697265625, 0.22237777709960938, -4.6003875732421875, 1.186248779296875, -0.6805038452148438, 6.665718078613281, 3.8082351684570312, -0.9458084106445312, -9.953125, 2.1810836791992188, 5.2205963134765625, 3.2297515869140625, 0.9892120361328125, 7.098899841308594, 0.30156707763671875, -2.590728759765625, 0.6891975402832031, 10.230758666992188, -1.0101146697998047, -3.9583492279052734, 2.2830123901367188, -2.8036651611328125, 1.9690990447998047, 6.2474517822265625, 0.4388141632080078, 2.6317596435546875, 0.9880695343017578, 0.7621383666992188, 0.4961051940917969, -5.607475280761719, -3.4458389282226562, 1.8074722290039062, 2.2773513793945312, -1.4392814636230469, -5.636970520019531, 1.0993423461914062, -4.6312408447265625, 5.433258056640625, -1.78240966796875, 1.8758258819580078, 9.188835144042969, 4.1827392578125, -1.5939483642578125, -2.8464813232421875, -1.3030204772949219, 2.4435348510742188, 4.655372619628906, -7.0200042724609375, -8.315582275390625, -16.458023071289062, -0.83612060546875, 0.6200103759765625, -0.4180450439453125, -0.13694000244140625, -0.8376312255859375, 14.07110595703125, 8.567047119140625, 3.568643569946289, 1.6347675323486328, 7.019989013671875, -4.286396026611328, 0.9412689208984375, 0.1341400146484375, -20.07232666015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000110.npy"}
|
|
{"epoch": 0.16628873771730915, "step": 111, "batch_size": 64, "mean": 1.9793009757995605, "std": 5.070337295532227, "min": -6.804252624511719, "p10": -3.151500129699707, "median": 1.1511383056640625, "p90": 9.241896820068359, "max": 18.482666015625, "pos_frac": 0.609375, "sample": [-0.7726974487304688, -2.9026546478271484, 0.06693267822265625, 12.927200317382812, 2.0118408203125, 7.287445068359375, -1.1012001037597656, 3.758209228515625, 2.5912132263183594, 0.6961708068847656, 5.228355407714844, -0.85577392578125, -3.7143020629882812, 2.2269363403320312, -6.56982421875, -0.11156082153320312, -0.36899375915527344, 0.7658863067626953, -2.439495086669922, 2.492940902709961, 8.975395202636719, 5.74029541015625, 10.84130859375, 9.670917510986328, 11.333160400390625, 4.1981658935546875, 3.6875534057617188, 6.2091522216796875, 2.0049362182617188, -6.804252624511719, -0.40759849548339844, -2.7216339111328125, -3.728363037109375, -6.456443786621094, 0.041782379150390625, -1.9355087280273438, 0.5037212371826172, 5.0282440185546875, -2.3284530639648438, 5.705101013183594, 9.212753295898438, 8.518447875976562, 10.128173828125, -1.608083724975586, -1.7741165161132812, -0.2848968505859375, 3.0077037811279297, 9.254386901855469, -2.535186767578125, 1.1407356262207031, 5.22698974609375, 3.303661346435547, 3.1230850219726562, -2.7102489471435547, -5.650413513183594, 1.1615409851074219, -3.258148193359375, 2.4651031494140625, -2.772655487060547, 1.3740081787109375, 18.482666015625, -2.2100677490234375, 2.0314903259277344, 0.27422332763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000111.npy"}
|
|
{"epoch": 0.16780045351473924, "step": 112, "batch_size": 64, "mean": 2.3736751079559326, "std": 4.60965633392334, "min": -10.08502197265625, "p10": -2.5338064193725582, "median": 2.4798355102539062, "p90": 7.959878158569337, "max": 12.32623291015625, "pos_frac": 0.734375, "sample": [12.0416259765625, 7.311737060546875, 8.548492431640625, 3.1042518615722656, 0.12062644958496094, 1.9289093017578125, 11.71929931640625, 3.334075927734375, 5.853446960449219, 3.0176620483398438, 0.37531280517578125, 6.6370849609375, -4.64678955078125, 2.646219253540039, 1.4459419250488281, -7.669891357421875, -1.224029541015625, 3.4937896728515625, -3.910797119140625, -0.24829483032226562, 3.0675506591796875, -2.6042327880859375, 3.461925506591797, 5.818382263183594, 2.088409423828125, 3.091205596923828, 5.678840637207031, 1.6589698791503906, 12.32623291015625, 3.199800491333008, 7.271942138671875, 1.8424072265625, -2.286172866821289, 0.9932460784912109, 5.824501037597656, -0.3486175537109375, 2.6778926849365234, -5.7721405029296875, 8.100421905517578, 2.9463844299316406, -1.7060718536376953, 2.4189453125, 4.825778961181641, 2.5407257080078125, 2.6727828979492188, 10.982254028320312, 5.4009857177734375, 5.489543914794922, -0.9177284240722656, 1.0270881652832031, -8.248466491699219, 2.242706298828125, -0.7884368896484375, 1.5960617065429688, 7.6319427490234375, 1.1318511962890625, 1.8167953491210938, -2.369478225708008, -2.1111297607421875, 4.8013916015625, -0.6104354858398438, -10.08502197265625, 1.6611785888671875, 9.596321105957031], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000112.npy"}
|
|
{"epoch": 0.1693121693121693, "step": 113, "batch_size": 64, "mean": 1.3697593212127686, "std": 4.568942070007324, "min": -14.588760375976562, "p10": -3.5239101409912106, "median": 1.074380874633789, "p90": 6.588486862182617, "max": 12.972320556640625, "pos_frac": 0.625, "sample": [-2.9367828369140625, -5.861442565917969, -2.071746826171875, 1.0558052062988281, -1.0659637451171875, 6.010295867919922, 2.7761077880859375, -8.678237915039062, 0.41854095458984375, 3.903871536254883, 1.2753257751464844, -1.1739654541015625, -3.3155288696289062, 2.5813217163085938, -3.6132164001464844, -0.01325225830078125, 1.5901908874511719, 8.734870910644531, -0.5455970764160156, -3.629873275756836, -2.869964599609375, 3.2390213012695312, 6.631275177001953, -5.283660888671875, -0.3667564392089844, 3.13616943359375, 4.620429992675781, 2.9711074829101562, 6.657161712646484, -0.6144943237304688, 3.9484901428222656, 12.972320556640625, 0.7240524291992188, -0.934783935546875, 3.97406005859375, 7.347816467285156, -14.588760375976562, 3.2485122680664062, 5.93890380859375, -0.8426513671875, 1.5090694427490234, 5.438512802124023, -4.4127044677734375, 6.4886474609375, -2.7704010009765625, 8.230850219726562, 1.0518875122070312, 0.2805461883544922, -0.4152336120605469, 4.257053375244141, 2.910430908203125, -2.7166290283203125, 1.727813720703125, 12.011795043945312, 5.54339599609375, 0.3601818084716797, 3.4066238403320312, 0.3850059509277344, 4.666801452636719, 0.42816162109375, 1.09295654296875, -0.10384368896484375, -2.68402099609375, 5.62872314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000113.npy"}
|
|
{"epoch": 0.1708238851095994, "step": 114, "batch_size": 64, "mean": 2.2689995765686035, "std": 4.578832626342773, "min": -5.716888427734375, "p10": -3.359594917297363, "median": 1.5232152938842773, "p90": 7.950549507141116, "max": 15.519134521484375, "pos_frac": 0.6875, "sample": [-5.716888427734375, -4.172325134277344, 0.6442375183105469, 6.270885467529297, 4.380836486816406, 1.8420867919921875, 3.51190185546875, 7.0519561767578125, 4.4999847412109375, -2.4329986572265625, -3.51348876953125, 1.0404167175292969, 3.3347854614257812, -2.64129638671875, -0.12104034423828125, 1.1743011474609375, 3.4069595336914062, -0.18609619140625, 0.2923927307128906, -0.8108444213867188, 2.2323455810546875, -1.9287872314453125, 15.310455322265625, -3.000509262084961, 0.4348640441894531, 7.064689636230469, 4.449836730957031, 5.136016845703125, 0.761871337890625, 6.3874664306640625, -2.0016841888427734, 1.5551319122314453, 8.308944702148438, 3.2889556884765625, 2.3520355224609375, 0.9798355102539062, 6.8245697021484375, 2.537567138671875, 5.430732727050781, -4.961463928222656, 15.519134521484375, 7.114294052124023, 5.4087066650390625, 0.07009696960449219, 3.6405029296875, 1.0482006072998047, 10.594818115234375, -2.0621414184570312, -1.8533935546875, -3.8101730346679688, 0.5555877685546875, -0.468048095703125, 3.6527767181396484, 9.012451171875, 8.780040740966797, 10.2037353515625, 2.003213882446289, -0.7090282440185547, -5.222328186035156, -3.9537696838378906, 6.038581848144531, -1.728271484375, 1.4912986755371094, 0.8710498809814453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000114.npy"}
|
|
{"epoch": 0.17233560090702948, "step": 115, "batch_size": 64, "mean": 1.8141748905181885, "std": 4.275871276855469, "min": -7.422630310058594, "p10": -3.0104248046874997, "median": 1.718435287475586, "p90": 7.089870452880861, "max": 14.52294921875, "pos_frac": 0.671875, "sample": [-1.3196220397949219, -1.9357070922851562, 3.8507614135742188, -7.1148223876953125, 14.52294921875, 2.1332836151123047, 3.6970138549804688, 4.571765899658203, 0.5780410766601562, 12.917526245117188, 1.8104400634765625, -1.4631271362304688, 10.16244125366211, -5.060340881347656, -2.770782470703125, -3.1333694458007812, 0.08465766906738281, 6.17974853515625, 1.6651763916015625, 2.7533035278320312, 4.2423095703125, 3.3203277587890625, 1.8084335327148438, -0.290771484375, -3.113128662109375, 1.8568534851074219, 3.5621337890625, 1.5104522705078125, -7.422630310058594, 7.241687774658203, -3.1985015869140625, -0.3989372253417969, 0.34168434143066406, 7.406219482421875, 1.7716941833496094, 3.8112258911132812, -0.42665863037109375, 3.2673168182373047, -1.9273681640625, 2.6490325927734375, 2.7508316040039062, -6.177799224853516, 6.017601013183594, 3.5143585205078125, 10.750823974609375, 2.477914810180664, 1.8133621215820312, 1.6264705657958984, -1.5286140441894531, 5.040252685546875, 1.055105209350586, -0.1374225616455078, 6.735630035400391, -0.816253662109375, 0.5776691436767578, -0.9375629425048828, -2.0070724487304688, -0.49440574645996094, 0.1072998046875, 1.2769508361816406, 4.104764938354492, 0.20945358276367188, 2.3538284301757812, 9.653289794921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000115.npy"}
|
|
{"epoch": 0.17384731670445955, "step": 116, "batch_size": 64, "mean": 2.735952854156494, "std": 3.860740900039673, "min": -5.147254943847656, "p10": -2.0082180023193357, "median": 2.968873977661133, "p90": 6.656412506103516, "max": 12.281097412109375, "pos_frac": 0.71875, "sample": [6.373558044433594, -2.5232696533203125, 0.9002189636230469, 8.391273498535156, 5.442695617675781, 4.5286865234375, 2.9622573852539062, 1.3781204223632812, 4.988649368286133, 6.72198486328125, 5.81451416015625, -2.4320602416992188, -4.6296539306640625, 9.67132568359375, 0.46508216857910156, -1.1585922241210938, 0.5584716796875, 0.034717559814453125, 3.6858863830566406, 5.210784912109375, -4.648674011230469, 2.3158493041992188, 5.209514617919922, 3.154064178466797, -0.2862262725830078, 5.952831268310547, 4.8972320556640625, 6.503410339355469, 5.3776092529296875, 1.6717453002929688, 1.3783454895019531, -0.9351215362548828, -2.030384063720703, 0.6995773315429688, 10.134231567382812, -0.37277984619140625, 4.354766845703125, 5.2508087158203125, -0.22157669067382812, -1.1406326293945312, 5.182289123535156, 4.398265838623047, 4.077144622802734, 1.5560760498046875, 12.281097412109375, 4.39794921875, 9.983016967773438, -3.4776382446289062, 0.5032577514648438, 2.9754905700683594, -0.18122482299804688, 6.0615081787109375, 0.7937088012695312, -5.147254943847656, -0.7582511901855469, 5.098335266113281, -0.9319267272949219, 3.5367965698242188, 1.260385513305664, 9.254928588867188, 6.234825134277344, -0.08315658569335938, 6.392614364624023, -1.9564971923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000116.npy"}
|
|
{"epoch": 0.17535903250188964, "step": 117, "batch_size": 64, "mean": 1.8007094860076904, "std": 4.414554119110107, "min": -11.670501708984375, "p10": -2.8640159606933593, "median": 1.7501239776611328, "p90": 6.790242004394533, "max": 15.304397583007812, "pos_frac": 0.71875, "sample": [0.8586578369140625, 4.722841262817383, 3.217864990234375, 0.6223106384277344, 8.015396118164062, 3.423707962036133, 4.932895660400391, 1.6767616271972656, -2.2163467407226562, 3.8254566192626953, 0.7723617553710938, -1.6912269592285156, 6.04638671875, -3.6231918334960938, 0.4538116455078125, 3.7768630981445312, -2.8391876220703125, 2.179229736328125, 3.2017745971679688, 5.251029968261719, 1.823486328125, 3.5455360412597656, 4.248809814453125, 2.5682296752929688, -1.1697120666503906, -4.625801086425781, 1.169464111328125, -0.39554595947265625, 3.2466659545898438, 10.301937103271484, -2.16912841796875, 9.32220458984375, 6.999824523925781, -0.025753021240234375, 8.282440185546875, 2.071422576904297, 3.9314517974853516, 8.727584838867188, 1.0434417724609375, 6.301216125488281, 2.2874374389648438, -1.0539627075195312, 1.8591842651367188, 2.886810302734375, -1.9765548706054688, -10.317955017089844, 1.2974605560302734, 0.7845726013183594, 2.887664794921875, 4.284461975097656, 1.163726806640625, 0.01088714599609375, 6.011932373046875, -2.8746566772460938, 15.304397583007812, -6.2155303955078125, -3.4391117095947266, 1.5710182189941406, -11.670501708984375, -0.9526443481445312, 4.964015960693359, 0.22533416748046875, 0.6303367614746094, -0.2280864715576172], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000117.npy"}
|
|
{"epoch": 0.17687074829931973, "step": 118, "batch_size": 64, "mean": 2.488833427429199, "std": 3.7283823490142822, "min": -7.514881134033203, "p10": -1.7973360061645507, "median": 2.3010215759277344, "p90": 7.911499023437501, "max": 10.521881103515625, "pos_frac": 0.6875, "sample": [2.2554168701171875, 2.1347122192382812, 4.5594024658203125, -2.766265869140625, -7.514881134033203, -1.313262939453125, 5.5262908935546875, 2.3466262817382812, 4.7830657958984375, -1.5022621154785156, 2.8846893310546875, 3.471282958984375, 0.9858932495117188, 8.135082244873047, 0.9663906097412109, -0.0378265380859375, -1.8305740356445312, 5.592952728271484, 0.05400848388671875, 0.8078136444091797, -1.6799507141113281, 7.68505859375, 3.9380531311035156, 4.844757080078125, 3.442310333251953, -1.8723258972167969, -0.46088409423828125, -2.1778793334960938, 4.623205184936523, -0.19267845153808594, 3.538501739501953, 2.8898582458496094, 8.008544921875, -1.8066749572753906, 1.0487899780273438, 4.20367431640625, 1.7651901245117188, 10.521881103515625, 8.404220581054688, -3.4273452758789062, -1.642496109008789, 6.9845123291015625, -0.04656982421875, 5.484893798828125, -1.256378173828125, 7.032371520996094, -1.3603363037109375, 8.25262451171875, 3.7268829345703125, 8.206680297851562, 4.398918151855469, 0.9704551696777344, 2.8547821044921875, 9.160774230957031, 6.595855712890625, 1.415618896484375, -1.42144775390625, 4.191347122192383, 5.537055969238281, 7.517732620239258, 2.1190719604492188, -1.7755451202392578, 0.9457988739013672, -1.4421310424804688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000118.npy"}
|
|
{"epoch": 0.17838246409674982, "step": 119, "batch_size": 64, "mean": 1.7794040441513062, "std": 5.0541558265686035, "min": -11.556182861328125, "p10": -3.8944381713867187, "median": 1.6422348022460938, "p90": 8.279048919677734, "max": 19.735809326171875, "pos_frac": 0.625, "sample": [-1.6187419891357422, 4.151268005371094, -3.513334274291992, 0.6654891967773438, -3.9540023803710938, -0.9628944396972656, 4.561712265014648, 3.29241943359375, -1.7294425964355469, 1.1065025329589844, -4.023040771484375, -0.4609870910644531, 2.3030853271484375, 8.787406921386719, 6.976245880126953, -11.556182861328125, -0.12333297729492188, -3.7554550170898438, 0.7249584197998047, 3.263916015625, -1.6296157836914062, 5.722816467285156, -1.7664947509765625, 2.709024429321289, 8.775260925292969, 4.0370941162109375, -4.520450592041016, 2.6947097778320312, 8.356124877929688, 10.013717651367188, 6.404571533203125, -2.4052963256835938, 3.420713424682617, 4.097843170166016, 1.2449951171875, 19.735809326171875, 1.6832656860351562, -0.5249099731445312, 0.9835891723632812, 8.099205017089844, 2.8239288330078125, 7.024147033691406, -2.238933563232422, -0.00046539306640625, -5.485443115234375, -4.548675537109375, 10.363265991210938, 5.019199371337891, 2.2358322143554688, 4.233329772949219, 1.9350128173828125, -10.918731689453125, -3.046295166015625, -1.2940120697021484, 1.278146743774414, 1.6012039184570312, 4.710161209106445, 9.775619506835938, 5.769477844238281, 1.4285507202148438, 2.3146419525146484, 2.1450958251953125, -1.4640922546386719, -1.0466690063476562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000119.npy"}
|
|
{"epoch": 0.17989417989417988, "step": 120, "batch_size": 64, "mean": 2.6676409244537354, "std": 3.589154005050659, "min": -4.466548919677734, "p10": -1.230725860595703, "median": 2.030454635620117, "p90": 6.559855079650879, "max": 13.21527099609375, "pos_frac": 0.8125, "sample": [1.7006988525390625, 0.8244056701660156, -2.89434814453125, 6.630132675170898, 4.908538818359375, 0.7805862426757812, 5.631401062011719, 1.3586082458496094, 0.43402862548828125, -3.5297164916992188, 6.928802490234375, 1.69256591796875, 2.8136844635009766, 3.0958404541015625, 3.4324569702148438, 11.781341552734375, 4.3543853759765625, 0.0822601318359375, 1.9829063415527344, 4.999366760253906, 5.52349853515625, -0.9958267211914062, 4.621166229248047, 2.687620162963867, 1.035552978515625, 1.8588809967041016, 6.3958740234375, 1.0100250244140625, 4.945777893066406, 4.4630584716796875, 13.21527099609375, -2.2151718139648438, 8.812858581542969, -0.9198036193847656, 7.896720886230469, 0.4906425476074219, 2.60504150390625, 0.22261619567871094, 5.751251220703125, -0.8584213256835938, -0.26568603515625, 2.1832008361816406, 1.6516952514648438, 11.802093505859375, -1.3049087524414062, -1.472137451171875, -4.466548919677734, 6.1573333740234375, 3.8360595703125, 1.2949066162109375, 0.22154808044433594, 0.23435211181640625, 5.912628173828125, 0.6676864624023438, 3.239025115966797, 0.027923583984375, 2.0780029296875, 4.1927032470703125, -1.0576324462890625, -3.1409378051757812, 5.4631805419921875, 1.6460037231445312, 5.432394027709961, 2.8415603637695312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000120.npy"}
|
|
{"epoch": 0.18140589569160998, "step": 121, "batch_size": 64, "mean": 2.6989834308624268, "std": 4.433727264404297, "min": -3.6571502685546875, "p10": -1.7152648925781249, "median": 1.5043668746948242, "p90": 7.974171829223635, "max": 19.6925048828125, "pos_frac": 0.796875, "sample": [2.5637054443359375, 11.585784912109375, 0.6440773010253906, -1.8597259521484375, 5.793416976928711, 12.18133544921875, 7.548931121826172, -2.083444595336914, 2.3808746337890625, 3.8564224243164062, -0.700164794921875, -1.7350082397460938, 11.741500854492188, 3.070852279663086, 0.4537811279296875, 0.6008129119873047, 4.692583084106445, -0.25838470458984375, 4.6711273193359375, -3.6571502685546875, 8.156417846679688, 0.07930374145507812, 5.831401824951172, 2.8954925537109375, 3.8202667236328125, 2.7530975341796875, 1.9445075988769531, -2.1732826232910156, 2.7964935302734375, 1.4014205932617188, -0.7107620239257812, 19.6925048828125, 2.44189453125, 2.1032028198242188, 0.6232032775878906, -0.0545806884765625, -0.5107421875, 1.4068450927734375, 2.2518768310546875, 0.24785995483398438, 0.10556411743164062, 0.08816719055175781, 2.3164901733398438, 4.735237121582031, 1.4098968505859375, 1.34716796875, -3.5006103515625, 14.407806396484375, 0.2458953857421875, 2.392059326171875, 1.5456485748291016, 1.4630851745605469, 2.2156105041503906, 3.861227035522461, 0.9574203491210938, 14.245819091796875, 1.282449722290039, -1.6691970825195312, 1.1845378875732422, 0.01662445068359375, 0.0209503173828125, 2.934368133544922, 6.831207275390625, -2.1902294158935547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000121.npy"}
|
|
{"epoch": 0.18291761148904007, "step": 122, "batch_size": 64, "mean": 2.5136897563934326, "std": 5.020443439483643, "min": -5.1788482666015625, "p10": -3.4210330963134763, "median": 1.8117189407348633, "p90": 8.102792739868168, "max": 20.15496826171875, "pos_frac": 0.65625, "sample": [20.15496826171875, -3.3445587158203125, 0.4728546142578125, 6.98468017578125, -3.8271484375, -2.183696746826172, 11.3450927734375, 8.383522033691406, -0.5930004119873047, 5.371063232421875, 4.441291809082031, 1.4705047607421875, 0.7055225372314453, 2.094707489013672, 0.5589466094970703, 13.672760009765625, -2.2153778076171875, 2.9844207763671875, -4.721126556396484, 6.077051162719727, 1.166534423828125, 11.110572814941406, 4.040843963623047, 5.376707077026367, 7.447757720947266, 5.0039520263671875, -1.7055034637451172, -3.453807830810547, 3.8618316650390625, 4.494697570800781, -5.1788482666015625, 5.256813049316406, -0.28579139709472656, -1.314971923828125, -4.293800354003906, 3.8086318969726562, 2.094839096069336, -1.0791015625, 3.1220741271972656, -0.2451019287109375, -0.7329349517822266, 0.6550140380859375, -3.8298683166503906, 10.625442504882812, -1.9939422607421875, 5.421304702758789, 1.9496345520019531, 2.0004005432128906, -0.2197265625, 1.740570068359375, 1.9004592895507812, 1.8828678131103516, 0.1994171142578125, 0.795928955078125, -3.5466995239257812, 2.8163013458251953, -0.3142509460449219, -0.2581024169921875, -0.1812286376953125, 6.539287567138672, 5.090572357177734, 4.278255462646484, 1.0543212890625, 17.942306518554688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000122.npy"}
|
|
{"epoch": 0.18442932728647016, "step": 123, "batch_size": 64, "mean": 2.486788272857666, "std": 3.9826345443725586, "min": -7.3834228515625, "p10": -1.6426376342773437, "median": 2.0495662689208984, "p90": 7.236998748779297, "max": 15.383209228515625, "pos_frac": 0.703125, "sample": [7.1119384765625, 5.400398254394531, 1.8084564208984375, 15.383209228515625, -4.200935363769531, -1.0522117614746094, 2.3982696533203125, 0.6925582885742188, 11.159049987792969, 5.668516159057617, 1.2378768920898438, 7.9991302490234375, 0.1637725830078125, -1.7271728515625, 5.1370849609375, 5.779415130615234, 2.5237674713134766, 3.795034408569336, -1.4453887939453125, 7.290596008300781, 2.0572662353515625, 6.674896240234375, 1.7029571533203125, 6.863883972167969, 4.2390594482421875, 9.341400146484375, 2.8599605560302734, -2.1287784576416016, 5.9957275390625, 2.0418663024902344, 0.8660373687744141, 10.367469787597656, -2.349689483642578, 5.96998405456543, -0.7651939392089844, -4.442420959472656, 2.0138092041015625, -1.9986209869384766, -0.027647018432617188, 1.561746597290039, 2.1405258178710938, -1.31964111328125, 2.1451778411865234, -0.1636962890625, 1.8275413513183594, 5.589080810546875, -0.4322376251220703, -1.1959190368652344, 1.2090702056884766, 5.930471420288086, -0.5861625671386719, 2.735260009765625, 1.3889694213867188, 3.7083587646484375, -7.3834228515625, 7.6220703125, -0.7502918243408203, 2.131855010986328, -1.3619308471679688, 2.484619140625, 1.4178180694580078, 2.440074920654297, 5.0220184326171875, -1.4122314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000123.npy"}
|
|
{"epoch": 0.18594104308390022, "step": 124, "batch_size": 64, "mean": 3.2994606494903564, "std": 4.136878967285156, "min": -4.728828430175781, "p10": -1.7386470794677733, "median": 3.1839981079101562, "p90": 8.777828979492188, "max": 14.923553466796875, "pos_frac": 0.78125, "sample": [-1.0328369140625, 14.708999633789062, 4.451072692871094, -4.728828430175781, 3.3350048065185547, 3.5544815063476562, 14.923553466796875, 5.150447845458984, 10.179962158203125, 9.813766479492188, 6.369956970214844, 5.827640533447266, 1.4014549255371094, 11.25244140625, 0.9967746734619141, 2.15435791015625, 1.76641845703125, 5.531316757202148, 1.3110160827636719, -1.8318901062011719, -0.707489013671875, 3.3014259338378906, 3.6446304321289062, 1.6552047729492188, 1.9419441223144531, 3.9135513305664062, 1.4612483978271484, 0.528076171875, -1.8810272216796875, -2.5677947998046875, 4.193309783935547, 4.57080078125, -1.2166748046875, 8.611801147460938, 4.206012725830078, 7.23675537109375, 10.461109161376953, -2.859649658203125, 7.689338684082031, 3.0843353271484375, 0.5364761352539062, 7.078025817871094, -1.5210800170898438, 4.683937072753906, 3.283660888671875, 3.3820953369140625, 1.7934341430664062, 5.747688293457031, -2.193460464477539, 4.199831008911133, 8.848983764648438, 1.4667167663574219, 0.17746543884277344, 2.8225440979003906, 3.05072021484375, 2.727020263671875, 2.8289718627929688, -1.4150009155273438, -3.8784751892089844, 4.1260223388671875, -0.7534561157226562, 4.080230712890625, -0.20557785034179688, 7.8966827392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000124.npy"}
|
|
{"epoch": 0.1874527588813303, "step": 125, "batch_size": 64, "mean": 2.776418685913086, "std": 4.361453533172607, "min": -5.876152038574219, "p10": -2.5370590209960935, "median": 2.1378231048583984, "p90": 8.215481185913086, "max": 17.042510986328125, "pos_frac": 0.734375, "sample": [1.318756103515625, 7.202629089355469, 4.746299743652344, 2.112762451171875, 0.96942138671875, -1.0667610168457031, 11.088115692138672, -0.01551055908203125, 4.483001708984375, 3.167144775390625, 17.042510986328125, -2.4138412475585938, -0.27620697021484375, 3.2340011596679688, 1.7877788543701172, 8.39370346069336, 7.7697906494140625, 3.118194580078125, 2.4870834350585938, -1.0105667114257812, 8.333232879638672, 0.06958770751953125, 3.4658126831054688, 3.9743690490722656, -1.1917228698730469, 1.9926929473876953, 7.595184326171875, -3.3101024627685547, -1.8731460571289062, -0.302642822265625, 7.03643798828125, 3.99017333984375, 0.6404342651367188, 10.294281005859375, 5.539678573608398, -2.7860641479492188, 5.194122314453125, 1.3716392517089844, 10.131752014160156, 13.6209716796875, 7.563201904296875, 2.189910888671875, 4.544044494628906, 4.623577117919922, 2.8912925720214844, -5.876152038574219, 0.44931793212890625, 7.940727233886719, 0.5969028472900391, 0.6578350067138672, 1.869140625, 3.118532180786133, 3.3288497924804688, 2.10870361328125, -2.5898666381835938, 3.6468734741210938, 0.6003646850585938, -2.68310546875, 2.126720428466797, -0.4766731262207031, 2.14892578125, -2.8138809204101562, -5.474281311035156, -0.72515869140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000125.npy"}
|
|
{"epoch": 0.1889644746787604, "step": 126, "batch_size": 64, "mean": 1.6986485719680786, "std": 4.268459320068359, "min": -9.105865478515625, "p10": -2.7435916900634765, "median": 1.7237625122070312, "p90": 5.919696044921875, "max": 16.560134887695312, "pos_frac": 0.65625, "sample": [2.8725109100341797, 0.3854084014892578, 5.863256454467773, 2.531522750854492, -0.04685020446777344, 6.028423309326172, 6.271514892578125, 0.11499214172363281, 2.72412109375, -2.208009719848633, 5.7272186279296875, 5.7760467529296875, -7.8690948486328125, 1.7056884765625, 5.930938720703125, 2.1282501220703125, -2.2022628784179688, -1.7702980041503906, -0.28081321716308594, -3.4254379272460938, 1.8797760009765625, 5.893463134765625, 1.6519927978515625, 2.740248680114746, -1.0067272186279297, -0.19223785400390625, 4.2087554931640625, -4.3110198974609375, -1.0511016845703125, -2.0028533935546875, 0.13810348510742188, -1.9611892700195312, -2.5775718688964844, 2.9195690155029297, -3.5078601837158203, -0.6923904418945312, -2.121248245239258, 16.560134887695312, 4.119945526123047, -9.105865478515625, 2.859966278076172, 3.9035186767578125, 2.9464778900146484, 4.389797210693359, 3.987274169921875, 4.38238525390625, 5.5872802734375, 12.313400268554688, 0.134918212890625, 2.7137680053710938, 0.8512153625488281, 7.7655792236328125, 3.041637420654297, 3.417926788330078, 1.3843765258789062, 0.0074462890625, -2.8147430419921875, 5.035139083862305, -1.1154937744140625, -0.45587921142578125, 1.2511138916015625, 9.370796203613281, -5.825281143188477, 1.7418365478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000126.npy"}
|
|
{"epoch": 0.19047619047619047, "step": 127, "batch_size": 64, "mean": 3.0703744888305664, "std": 3.7720015048980713, "min": -6.08953857421875, "p10": -0.8267501831054684, "median": 2.7255935668945312, "p90": 8.055325317382813, "max": 12.969734191894531, "pos_frac": 0.765625, "sample": [-1.5623779296875, 12.007156372070312, 5.795310974121094, 3.868480682373047, -0.3333282470703125, 7.570037841796875, 3.5118026733398438, 8.072967529296875, 0.6914520263671875, 9.588722229003906, 6.400665283203125, 3.811309814453125, 4.271942138671875, 8.099411010742188, 2.5329132080078125, 1.6195526123046875, 5.0654449462890625, 3.5901412963867188, -0.27201080322265625, 1.8760108947753906, 6.068260192871094, -6.08953857421875, 4.1250457763671875, 6.689765930175781, 3.599285125732422, -4.262153625488281, 3.0013809204101562, 1.4156417846679688, 1.0245819091796875, 1.4686431884765625, 0.3627586364746094, 2.7237167358398438, -0.0002727508544921875, 7.53662109375, -0.07535171508789062, 6.986236572265625, 1.0952415466308594, 8.01416015625, 8.50887680053711, 12.969734191894531, -1.1971054077148438, 5.6382904052734375, 1.8380050659179688, -2.100301742553711, 3.1801528930664062, 2.7518234252929688, -0.5468292236328125, -0.94671630859375, 6.7787017822265625, -3.0917510986328125, 0.813995361328125, 7.79931640625, 2.283588409423828, 1.2488784790039062, 1.0919685363769531, -0.01576995849609375, -0.43695068359375, 1.6865081787109375, 2.7274703979492188, 3.0935840606689453, 3.734149932861328, 0.2500038146972656, 9.029975891113281, -0.4752655029296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000127.npy"}
|
|
{"epoch": 0.19198790627362056, "step": 128, "batch_size": 64, "mean": 3.0327281951904297, "std": 5.406322002410889, "min": -9.797195434570312, "p10": -3.2330801010131833, "median": 2.440882682800293, "p90": 10.422732162475588, "max": 15.552642822265625, "pos_frac": 0.75, "sample": [1.933929443359375, 10.720443725585938, 2.5897960662841797, 3.5638427734375, 2.750438690185547, 1.9114227294921875, 4.0938873291015625, 8.457000732421875, 1.4268875122070312, 12.442062377929688, 1.37689208984375, -9.797195434570312, 10.014842987060547, -1.3108901977539062, 3.3460617065429688, -5.4642791748046875, 0.8454647064208984, -1.5474090576171875, -5.779289245605469, 10.597541809082031, 12.922393798828125, -4.8597564697265625, -2.6065444946289062, 15.430267333984375, 15.552642822265625, 6.111572265625, 0.7761154174804688, 9.939308166503906, -1.3554859161376953, 6.508356094360352, 1.6295166015625, 5.140052795410156, 2.4523468017578125, 7.2757568359375, 5.948558807373047, 5.3013916015625, 0.2875823974609375, 9.265216827392578, -4.938697814941406, 0.8750457763671875, -2.7683792114257812, 2.4294185638427734, 2.28936767578125, -0.01744842529296875, -3.4322376251220703, 0.915130615234375, 2.207212448120117, 3.487262725830078, 3.5768280029296875, -1.9631996154785156, 3.4494094848632812, 3.0056610107421875, 9.176544189453125, -1.0480499267578125, 5.141010284423828, 1.4332561492919922, 0.6777191162109375, 5.5975341796875, 14.341476440429688, -1.3529415130615234, -7.819210052490234, 0.6004791259765625, 5.331996917724609, 5.0086669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000128.npy"}
|
|
{"epoch": 0.19349962207105065, "step": 129, "batch_size": 64, "mean": 1.9325857162475586, "std": 4.8436808586120605, "min": -11.933349609375, "p10": -2.5737817764282225, "median": 1.5204658508300781, "p90": 7.525976371765137, "max": 14.732799530029297, "pos_frac": 0.625, "sample": [0.7611160278320312, -0.9459095001220703, 4.874156951904297, -3.5283737182617188, -0.6138248443603516, 6.577503204345703, -1.686187744140625, -1.9614028930664062, 1.2855377197265625, 8.491386413574219, 6.291263580322266, 2.4205093383789062, 4.889495849609375, -2.0636444091796875, -2.3190155029296875, -2.086029052734375, 6.9829559326171875, 0.7604866027832031, 9.387161254882812, 7.4633941650390625, 1.450225830078125, -9.578460693359375, 1.5907058715820312, 3.0497283935546875, -2.5866947174072266, 14.527618408203125, 4.396516799926758, -0.040699005126953125, 3.6041088104248047, 4.211883544921875, 4.593780517578125, -4.2414398193359375, -4.1890869140625, 7.552797317504883, 0.466827392578125, 5.787506103515625, -6.233978271484375, -0.25725555419921875, -1.9944438934326172, 9.005142211914062, 4.063812255859375, -0.633575439453125, -1.7179489135742188, 1.127523422241211, 4.372333526611328, 3.2746410369873047, -0.6386947631835938, 14.732799530029297, 6.011878967285156, 7.252166748046875, 2.501575469970703, 6.948310852050781, 7.857170104980469, -1.0715789794921875, -0.8528842926025391, -2.543651580810547, 3.7459259033203125, -11.933349609375, 1.8861083984375, 0.249542236328125, 1.6524620056152344, -1.82110595703125, 0.04369163513183594, 3.0829696655273438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000129.npy"}
|
|
{"epoch": 0.19501133786848074, "step": 130, "batch_size": 64, "mean": 2.8034071922302246, "std": 3.9130196571350098, "min": -4.624126434326172, "p10": -2.874742889404297, "median": 2.6324291229248047, "p90": 7.132591247558594, "max": 12.857807159423828, "pos_frac": 0.78125, "sample": [2.6356849670410156, 1.1395225524902344, 9.229782104492188, 6.2650299072265625, 5.825401306152344, 1.78265380859375, -1.989776611328125, 7.1541748046875, 2.5988025665283203, 0.6712188720703125, -3.7503204345703125, 5.6123809814453125, 5.923896789550781, 1.8119468688964844, -3.6542282104492188, 5.007652282714844, 2.705961227416992, -2.9591751098632812, 1.4647216796875, 9.214164733886719, 3.599332809448242, 4.233970642089844, 6.812797546386719, 6.428619384765625, -2.0400848388671875, -3.2609634399414062, -2.677734375, 0.926239013671875, -2.512289047241211, 0.6057662963867188, 5.846710205078125, 7.603172302246094, 5.821098327636719, 4.192104339599609, 1.1121330261230469, 12.857807159423828, 6.12005615234375, 3.0972423553466797, 12.856521606445312, 1.9386634826660156, -0.9706687927246094, 4.481441497802734, -1.0162124633789062, -4.624126434326172, 2.326200485229492, 3.55706787109375, 4.398715972900391, 1.1317253112792969, 4.6035003662109375, -0.846710205078125, -4.1540679931640625, 2.307720184326172, 1.957000732421875, 4.808027267456055, -3.8378524780273438, 3.8495521545410156, 2.6291732788085938, 2.2131385803222656, 2.5546035766601562, 7.0822296142578125, 8.727645874023438, 3.371335983276367, 1.5199298858642578, 3.0980224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000130.npy"}
|
|
{"epoch": 0.1965230536659108, "step": 131, "batch_size": 64, "mean": 3.5129053592681885, "std": 5.237935543060303, "min": -7.975624084472656, "p10": -3.4872562408447263, "median": 2.5492801666259766, "p90": 10.86331024169922, "max": 15.016647338867188, "pos_frac": 0.765625, "sample": [2.2603111267089844, 5.647487640380859, -0.4990825653076172, -1.194314956665039, -1.9309673309326172, 15.016647338867188, 3.2114124298095703, 4.303037643432617, -1.4666805267333984, 10.304161071777344, 3.14300537109375, 1.528066635131836, -3.124553680419922, -3.7491378784179688, 1.1132698059082031, 5.21766471862793, -4.26690673828125, 2.7751693725585938, 7.9353179931640625, -4.115081787109375, -3.6427001953125, -2.669097900390625, 11.862396240234375, 5.281639099121094, 2.80743408203125, 0.11803436279296875, 2.14910888671875, -4.7105865478515625, -0.4897880554199219, 0.4947357177734375, 5.616844177246094, 11.0235595703125, 10.489395141601562, 6.3826904296875, 10.019287109375, 12.892303466796875, 1.9310722351074219, 11.56414794921875, 3.795888900756836, 10.4046630859375, 2.642974853515625, 13.199676513671875, -1.9826393127441406, 6.7935333251953125, 2.455585479736328, 0.8017234802246094, 0.7628879547119141, 5.728363037109375, -4.526908874511719, 6.61029052734375, 0.07177734375, 0.26364898681640625, 6.454486846923828, 0.5672340393066406, -7.975624084472656, 8.667457580566406, 2.0699462890625, 1.930999755859375, 10.4173583984375, 2.1531639099121094, 2.3721160888671875, 8.051979064941406, 12.198217391967773, 7.667842864990234], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000131.npy"}
|
|
{"epoch": 0.1980347694633409, "step": 132, "batch_size": 64, "mean": 2.9671125411987305, "std": 4.425008773803711, "min": -10.92486572265625, "p10": -1.8481172561645507, "median": 3.273137092590332, "p90": 8.043568420410157, "max": 13.760574340820312, "pos_frac": 0.75, "sample": [9.137229919433594, 2.511941909790039, 4.868751525878906, 4.982501983642578, -1.529083251953125, -1.8497142791748047, 5.9820098876953125, 0.1674346923828125, -1.844390869140625, 5.609832763671875, 5.591682434082031, 7.767168045043945, 3.593036651611328, 3.6222305297851562, -1.4119911193847656, 1.8358535766601562, -0.7612113952636719, 10.838577270507812, 3.9675140380859375, -0.7357864379882812, -1.3065414428710938, -10.92486572265625, 5.983697891235352, -0.05304718017578125, -1.9434623718261719, -6.1807708740234375, 5.0042724609375, 8.29079818725586, 8.906064987182617, -1.2802257537841797, 2.4915542602539062, 0.832489013671875, -9.835550308227539, 3.0792312622070312, 2.2726593017578125, 2.86346435546875, 13.760574340820312, 5.716424942016602, 7.773185729980469, 5.064208984375, -0.36489105224609375, 10.014373779296875, 2.0305557250976562, 8.159446716308594, 6.904520034790039, 3.9818458557128906, 2.8136138916015625, 1.9001045227050781, 2.2589645385742188, 3.467042922973633, 4.873252868652344, 2.72772216796875, 5.170196533203125, 4.7133331298828125, 7.688030242919922, 5.1229705810546875, 6.0177459716796875, -2.007965087890625, 1.043426513671875, 2.6878204345703125, -3.388916015625, 4.2914886474609375, 5.0572662353515625, 1.875518798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000132.npy"}
|
|
{"epoch": 0.19954648526077098, "step": 133, "batch_size": 64, "mean": 2.466233253479004, "std": 4.520086288452148, "min": -11.44354248046875, "p10": -2.1949378967285154, "median": 2.5323848724365234, "p90": 8.183798408508302, "max": 14.591117858886719, "pos_frac": 0.703125, "sample": [4.838756561279297, -3.29925537109375, 2.793994903564453, 9.126678466796875, 0.18343162536621094, 0.1006927490234375, 2.4356956481933594, 8.344879150390625, 0.6341094970703125, 0.8832931518554688, 2.786834716796875, 10.679458618164062, 6.948253631591797, 1.7864570617675781, -4.712150573730469, 3.3589248657226562, 3.486846923828125, -11.44354248046875, 5.756694793701172, 0.606719970703125, -0.7848739624023438, 5.454498291015625, -0.5680885314941406, 6.003013610839844, 4.090476989746094, -0.105743408203125, 5.3223724365234375, -0.349700927734375, 2.6872024536132812, -0.4914588928222656, 2.7672882080078125, 1.2126598358154297, -7.1124114990234375, 7.807943344116211, -2.3333587646484375, 2.8670501708984375, -0.801910400390625, -1.0672760009765625, 7.47743034362793, 6.44952392578125, -0.34095001220703125, 1.8468856811523438, 4.117698669433594, -6.086582183837891, 2.6290740966796875, 11.453475952148438, 5.264930725097656, 6.767913818359375, 2.05731201171875, 14.591117858886719, 0.9675445556640625, -0.02550506591796875, -1.8719558715820312, 10.140533447265625, 4.914985656738281, 0.8571357727050781, 4.3596038818359375, 8.711141586303711, 3.980438232421875, -1.0446300506591797, 3.4198455810546875, -0.61181640625, -3.662609100341797, 1.5819168090820312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000133.npy"}
|
|
{"epoch": 0.20105820105820105, "step": 134, "batch_size": 64, "mean": 3.875655174255371, "std": 4.888887405395508, "min": -7.8152008056640625, "p10": -1.7655744552612298, "median": 3.37265682220459, "p90": 9.83406677246094, "max": 17.577774047851562, "pos_frac": 0.796875, "sample": [0.2302093505859375, -0.6558074951171875, 3.8731536865234375, 3.4880905151367188, 3.4458656311035156, 3.299448013305664, 1.9081611633300781, 5.500556945800781, 11.093696594238281, 8.90478515625, 3.6163253784179688, 7.441413879394531, 4.4886322021484375, 2.906475067138672, 3.5478038787841797, -0.1740398406982422, 7.409854888916016, 5.692607879638672, -2.094585418701172, 4.5696563720703125, 6.665271759033203, 3.2899017333984375, 8.269729614257812, 0.18002700805664062, 2.853374481201172, 8.194221496582031, 0.6934356689453125, 1.4924468994140625, -2.0059375762939453, 5.993677139282227, -3.1395530700683594, -0.29375457763671875, 9.390457153320312, -0.12127685546875, 0.8208980560302734, -5.769041061401367, 3.6374130249023438, 2.5716686248779297, 2.9297637939453125, 4.0649566650390625, 16.745834350585938, 7.700592041015625, -7.8152008056640625, -1.2047271728515625, 5.627784729003906, 0.81427001953125, 11.466705322265625, -2.025970458984375, 14.023948669433594, 2.5919342041015625, 0.7017402648925781, 6.015384674072266, 5.077861785888672, 6.4457244873046875, 14.053970336914062, 1.0510940551757812, -2.7356224060058594, 3.1588897705078125, 2.848112106323242, 7.060295104980469, 10.024185180664062, 17.577774047851562, 1.350515365600586, -0.72314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000134.npy"}
|
|
{"epoch": 0.20256991685563114, "step": 135, "batch_size": 64, "mean": 2.561357021331787, "std": 4.629579544067383, "min": -10.90377426147461, "p10": -2.2676517486572263, "median": 2.2060537338256836, "p90": 7.655521392822266, "max": 13.473953247070312, "pos_frac": 0.765625, "sample": [5.090538024902344, -1.3991279602050781, 6.630645751953125, 1.8060264587402344, 2.1456336975097656, -2.410736083984375, 0.6476211547851562, -3.9998092651367188, -0.5296916961669922, 10.927017211914062, 6.1450042724609375, -0.8546829223632812, 5.2422637939453125, -0.7996406555175781, 0.32981109619140625, -4.517726898193359, 8.570098876953125, 3.9068069458007812, -1.201070785522461, 2.031461715698242, 2.2664737701416016, 0.9851760864257812, -10.90377426147461, 0.9248428344726562, 2.986675262451172, 7.682014465332031, 1.9110183715820312, 2.351104736328125, 0.9484481811523438, 3.064830780029297, 2.7895355224609375, 1.4712600708007812, 6.752952575683594, 1.6918277740478516, 4.93310546875, 1.7114295959472656, 2.08819580078125, 2.9032516479492188, 5.791893005371094, -5.76060676574707, 0.027584075927734375, 13.473953247070312, 0.1495075225830078, 2.983795166015625, 2.8963165283203125, 2.2973861694335938, 0.1950225830078125, -4.740324020385742, 6.150718688964844, 6.0493621826171875, 5.327678680419922, -7.426177978515625, 0.9856719970703125, 6.33265495300293, 11.849128723144531, 13.190452575683594, -1.9337882995605469, 5.403772354125977, -1.4171829223632812, 5.281108856201172, 3.1491851806640625, 7.5937042236328125, 11.815837860107422, -0.058624267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000135.npy"}
|
|
{"epoch": 0.20408163265306123, "step": 136, "batch_size": 64, "mean": 3.1969728469848633, "std": 6.324760437011719, "min": -10.489675521850586, "p10": -3.9111244201660154, "median": 2.9109249114990234, "p90": 12.770866584777837, "max": 18.289886474609375, "pos_frac": 0.703125, "sample": [14.76641845703125, 8.456676483154297, -7.506673812866211, -1.4814796447753906, -4.33984375, 6.33123779296875, -1.0253829956054688, 1.1344375610351562, -3.4971389770507812, 5.110944747924805, 3.139984130859375, -9.506011962890625, -0.6444797515869141, 4.7601165771484375, -1.7537460327148438, 6.052116394042969, 13.168848037719727, 2.465808868408203, 1.341562271118164, -3.4410171508789062, 7.7866668701171875, 4.123584747314453, 7.30914306640625, 14.313636779785156, -1.2249526977539062, 0.02333831787109375, 13.267807006835938, 3.5356101989746094, 2.001373291015625, 18.289886474609375, 7.4302215576171875, 0.01204681396484375, 9.637237548828125, 9.911178588867188, 0.125762939453125, 3.3388671875, -1.7372512817382812, 4.477073669433594, 4.5506591796875, 1.35040283203125, -10.489675521850586, 4.0759124755859375, 1.7988433837890625, 8.414779663085938, 2.738208770751953, 6.914356231689453, -2.0596160888671875, 5.0196380615234375, -4.0885467529296875, 16.799644470214844, 3.0836410522460938, -6.346107482910156, -2.65570068359375, 5.3025360107421875, 1.421234130859375, 9.456695556640625, 0.20610809326171875, -0.100860595703125, 11.842243194580078, 14.248367309570312, 6.192352294921875, -0.394439697265625, 1.1221504211425781, -9.950172424316406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000136.npy"}
|
|
{"epoch": 0.20559334845049132, "step": 137, "batch_size": 64, "mean": 3.24276065826416, "std": 4.985215663909912, "min": -8.865386962890625, "p10": -2.1867015838623045, "median": 2.6847305297851562, "p90": 9.352861404418949, "max": 18.520484924316406, "pos_frac": 0.6875, "sample": [-0.8982925415039062, -8.865386962890625, 5.316673278808594, 7.26409912109375, 5.583972930908203, 9.746353149414062, 8.57675552368164, 6.8225250244140625, -0.6117439270019531, 14.170639038085938, 7.487752914428711, 2.5375213623046875, 10.156322479248047, 3.830169677734375, 9.685478210449219, 0.7829360961914062, 2.831939697265625, 9.856285095214844, -0.3621063232421875, 1.3805274963378906, -1.9346752166748047, 4.40924072265625, 6.629974365234375, -5.209648132324219, 1.5424633026123047, -0.2006378173828125, 6.862648010253906, 18.520484924316406, -2.0310611724853516, 0.27768707275390625, -2.6880645751953125, -0.44005584716796875, -0.5921783447265625, 5.3390655517578125, 13.957176208496094, 1.6846237182617188, 0.079620361328125, -3.947551727294922, 4.842277526855469, -1.974386215209961, -0.018825531005859375, 1.7437152862548828, 4.435756683349609, 7.253875732421875, 5.286411285400391, -2.2534046173095703, 2.082721710205078, 6.944841384887695, 4.667724609375, 8.439437866210938, 1.5368576049804688, -3.0899429321289062, -0.3164653778076172, -4.670654296875, 7.798887252807617, 3.4171485900878906, 1.4791584014892578, -1.2741775512695312, -0.6234130859375, 5.683837890625, 1.2569541931152344, 7.984428405761719, 4.64862060546875, 4.7037506103515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000137.npy"}
|
|
{"epoch": 0.20710506424792138, "step": 138, "batch_size": 64, "mean": 2.6813035011291504, "std": 5.273225784301758, "min": -12.007640838623047, "p10": -3.2415185928344723, "median": 3.2310333251953125, "p90": 8.234019470214845, "max": 16.472610473632812, "pos_frac": 0.671875, "sample": [-0.042873382568359375, 6.565452575683594, 3.94525146484375, 7.9965667724609375, 4.790248870849609, 3.171245574951172, 1.3328094482421875, 4.319969177246094, -4.6803436279296875, 4.019542694091797, 2.633943557739258, -9.677139282226562, 8.335784912109375, -0.6789627075195312, 16.472610473632812, -5.2221832275390625, -3.7764053344726562, 7.165182113647461, 7.588829040527344, 5.603603363037109, 4.091022491455078, 1.7590560913085938, 12.340988159179688, 6.575124740600586, 1.0363731384277344, 5.532354354858398, 7.6158294677734375, 4.58905029296875, 0.8903350830078125, 1.6123905181884766, 6.965082168579102, -0.10416984558105469, 9.263376235961914, -0.4023895263671875, -3.3264217376708984, -1.2812347412109375, 0.2304210662841797, -9.330001831054688, -0.5565242767333984, 5.145778656005859, 3.3673095703125, 7.577384948730469, -0.05161285400390625, 3.4248046875, -1.8981666564941406, -12.007640838623047, 13.622438430786133, 2.7368316650390625, 3.5923538208007812, -0.3611297607421875, 2.62261962890625, 3.9373779296875, 6.4524993896484375, -2.5475502014160156, -1.923828125, -0.8806552886962891, -3.0434112548828125, 4.648111343383789, -0.7063179016113281, 4.22344970703125, 12.591606140136719, 3.290821075439453, 9.022506713867188, 1.4040603637695312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000138.npy"}
|
|
{"epoch": 0.20861678004535147, "step": 139, "batch_size": 64, "mean": 3.3494184017181396, "std": 4.673854351043701, "min": -9.7713623046875, "p10": -2.2339813232421872, "median": 3.0401487350463867, "p90": 9.65934982299805, "max": 14.700851440429688, "pos_frac": 0.75, "sample": [3.054494857788086, 8.896614074707031, 2.1313705444335938, 2.5309829711914062, -4.339908599853516, 5.962863922119141, 3.4741363525390625, 8.520854949951172, 6.804386138916016, -0.22802162170410156, -0.27594757080078125, 14.375177383422852, 6.668388366699219, -1.6653614044189453, 3.7228546142578125, 10.18206787109375, 14.700851440429688, 2.022939682006836, -0.04923248291015625, 8.35400390625, -2.35101318359375, -9.7713623046875, -0.9662704467773438, 2.5631256103515625, 5.8676605224609375, 2.470165252685547, 4.5708465576171875, 3.2679367065429688, 6.562702178955078, 0.75433349609375, 4.6014862060546875, 5.823272705078125, 3.3768539428710938, -0.455657958984375, -0.4510040283203125, 5.3714447021484375, 5.069644927978516, 2.7532958984375, 3.36468505859375, 0.3377952575683594, 8.383995056152344, 0.5436973571777344, 10.442024230957031, -0.6217765808105469, -2.388826370239258, -2.950031280517578, 3.0258026123046875, 4.976432800292969, 14.178619384765625, 0.9760780334472656, 9.986236572265625, 2.7236251831054688, 2.553863525390625, -4.789398193359375, -1.960906982421875, 5.238838195800781, 0.8611297607421875, 2.2850875854492188, -3.5919570922851562, 5.5337677001953125, 5.783454895019531, 10.505210876464844, 3.1092796325683594, 1.955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000139.npy"}
|
|
{"epoch": 0.21012849584278157, "step": 140, "batch_size": 64, "mean": 4.307365894317627, "std": 6.744582176208496, "min": -18.993972778320312, "p10": -4.345431518554687, "median": 4.7760162353515625, "p90": 13.251701354980469, "max": 15.445337295532227, "pos_frac": 0.75, "sample": [13.255020141601562, 7.436309814453125, 0.11301803588867188, 0.21231842041015625, 0.25893402099609375, 7.855815887451172, -6.44207763671875, -4.5072784423828125, 8.996238708496094, 8.887367248535156, -18.993972778320312, 6.381797790527344, 4.542808532714844, 1.952301025390625, 13.241806030273438, 5.483207702636719, -4.705837249755859, 1.739593505859375, -1.2912216186523438, 11.532585144042969, 3.7392520904541016, 13.24395751953125, 7.319240570068359, 6.6672515869140625, 0.7395515441894531, 6.9658203125, 3.0636024475097656, 15.445337295532227, 1.51385498046875, -3.9677886962890625, -5.035125732421875, -3.1656723022460938, 7.509067535400391, -2.5840911865234375, -1.8149471282958984, 2.2349090576171875, 10.091789245605469, 14.143402099609375, 12.000175476074219, 13.95184326171875, 5.535163879394531, 11.326301574707031, 14.305435180664062, 12.25514030456543, 3.8124160766601562, -5.358837127685547, 13.4483642578125, 0.4924354553222656, 14.292770385742188, -2.1356868743896484, -6.667842864990234, 8.00146484375, 2.8501548767089844, 5.503181457519531, 9.09527587890625, 2.0127010345458984, 7.893684387207031, -0.6684951782226562, 5.515411376953125, 4.010643005371094, -0.349273681640625, 10.817840576171875, -3.3362293243408203, 5.009223937988281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000140.npy"}
|
|
{"epoch": 0.21164021164021163, "step": 141, "batch_size": 64, "mean": 4.188300609588623, "std": 6.507133483886719, "min": -9.56134033203125, "p10": -3.6583736419677724, "median": 3.013490676879883, "p90": 12.375619125366212, "max": 20.332489013671875, "pos_frac": 0.734375, "sample": [-9.56134033203125, 1.7741241455078125, 11.58013916015625, 1.9096031188964844, 6.2642974853515625, -0.01174163818359375, 1.9725570678710938, 8.707935333251953, 2.2947254180908203, -6.8741455078125, 4.772918701171875, 2.915699005126953, 8.460519790649414, -4.128406524658203, 8.920047760009766, 5.074932098388672, 6.291351318359375, 0.8678741455078125, -0.2409191131591797, -7.856475830078125, 4.104377746582031, 0.6496410369873047, 1.0698394775390625, 6.392782211303711, 5.538673400878906, 1.0090484619140625, 2.2268524169921875, 9.306838989257812, -2.014892578125, -4.6331329345703125, -1.3605575561523438, -5.199348449707031, 5.971347808837891, 6.205436706542969, 9.430835723876953, -2.3713321685791016, 9.462329864501953, 2.4928512573242188, 18.574703216552734, 14.6275634765625, 11.698570251464844, 9.179874420166016, 9.059619903564453, 5.970069885253906, 15.002662658691406, 0.9418411254882812, -0.4877605438232422, 3.3662796020507812, 14.871479034423828, 12.027870178222656, -6.774627685546875, 2.111896514892578, 2.9391441345214844, 18.7786865234375, 6.518035888671875, 20.332489013671875, -2.5616302490234375, 12.524654388427734, -0.179412841796875, 3.0878372192382812, 6.696952819824219, 2.3328685760498047, -2.0792388916015625, -1.924468994140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000141.npy"}
|
|
{"epoch": 0.21315192743764172, "step": 142, "batch_size": 64, "mean": 3.2472176551818848, "std": 5.049487590789795, "min": -8.830005645751953, "p10": -2.2947154998779298, "median": 2.800783157348633, "p90": 9.435478210449219, "max": 15.829986572265625, "pos_frac": 0.765625, "sample": [3.2832412719726562, -7.065757751464844, 1.4883899688720703, 6.604576110839844, 9.8780517578125, 14.767356872558594, 6.508636474609375, -2.2517051696777344, 2.694000244140625, 7.0799713134765625, 1.0065155029296875, -0.3465728759765625, 5.488149642944336, -8.830005645751953, 1.1177406311035156, 6.032691955566406, 4.39581298828125, -6.592702865600586, 1.4469413757324219, 2.8980274200439453, 1.810028076171875, 7.8090667724609375, 4.56428337097168, 5.022541046142578, 2.990203857421875, 9.290313720703125, -0.6173782348632812, -0.5156173706054688, 11.693412780761719, -2.3131484985351562, 2.9562149047851562, 0.0980072021484375, 10.815872192382812, 15.829986572265625, 3.1417007446289062, -3.56689453125, 2.051715850830078, -4.798885345458984, 0.396209716796875, 2.737255096435547, 13.2203369140625, 9.297531127929688, 1.1872749328613281, 1.5928688049316406, 0.509552001953125, 8.82742691040039, -3.6798458099365234, 2.8643112182617188, -0.7616195678710938, 6.5217742919921875, 1.113677978515625, -1.8800907135009766, 1.2166748046875, 8.55950927734375, 2.607807159423828, 4.7501068115234375, 9.494598388671875, 9.19975471496582, 4.4530029296875, -0.23043251037597656, 1.4018402099609375, -1.7810249328613281, 7.0503387451171875, 3.2883148193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000142.npy"}
|
|
{"epoch": 0.2146636432350718, "step": 143, "batch_size": 64, "mean": 2.783491849899292, "std": 6.326268672943115, "min": -17.24554443359375, "p10": -3.628653717041015, "median": 2.9122695922851562, "p90": 9.8715633392334, "max": 15.830764770507812, "pos_frac": 0.75, "sample": [2.22540283203125, 6.683868408203125, 8.379253387451172, 6.114315032958984, 4.080619812011719, -1.8584747314453125, 1.968109130859375, -0.7867851257324219, 13.299184799194336, -7.372097015380859, 8.109731674194336, 0.364349365234375, 13.522537231445312, 3.3274307250976562, 5.9227447509765625, 4.652076721191406, -11.216262817382812, 2.894805908203125, 12.626983642578125, 1.3160209655761719, 2.9297332763671875, -2.051288604736328, 15.830764770507812, 9.36526870727539, 6.5051727294921875, 2.192575454711914, 3.5050811767578125, -10.02435302734375, 2.2380332946777344, -0.6308097839355469, 1.0903701782226562, 2.399168014526367, -7.0105133056640625, 6.9237213134765625, 12.608963012695312, -2.674755096435547, -3.8966903686523438, 0.14540863037109375, 0.01202392578125, 4.9257965087890625, 1.8497085571289062, -17.24554443359375, 9.150333404541016, 1.9842529296875, -1.4780197143554688, 6.474884033203125, 6.721435546875, 8.307754516601562, 0.22039794921875, -1.5912551879882812, -3.00323486328125, 4.016408920288086, 10.088546752929688, 7.9628753662109375, -12.742622375488281, 12.092342376708984, 5.911033630371094, 3.1316986083984375, 7.4876556396484375, 6.575521469116211, 0.5801372528076172, -1.3800067901611328, 3.1939125061035156, 1.1977710723876953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000143.npy"}
|
|
{"epoch": 0.2161753590325019, "step": 144, "batch_size": 64, "mean": 3.5666003227233887, "std": 6.268381595611572, "min": -15.043556213378906, "p10": -3.7017713546752926, "median": 2.8138256072998047, "p90": 12.781754112243656, "max": 19.414657592773438, "pos_frac": 0.71875, "sample": [13.237590789794922, 6.487545013427734, -3.2615890502929688, 0.6662731170654297, -1.6734733581542969, 2.398914337158203, -0.9237651824951172, 7.612602233886719, 1.2573051452636719, 0.14542198181152344, -4.816017150878906, 3.3399810791015625, 1.0975990295410156, 3.916830062866211, 7.997406005859375, 4.187904357910156, 12.001937866210938, 0.38520050048828125, -2.8291168212890625, 6.860084533691406, 8.082427978515625, 4.292011260986328, 1.8367919921875, -1.504638671875, -1.3759651184082031, 2.2045974731445312, 6.1976776123046875, 8.819053649902344, -0.9446697235107422, -1.0654220581054688, 6.5670013427734375, 2.3336410522460938, -5.172504425048828, -1.029815673828125, 9.341209411621094, 0.19023895263671875, -1.200845718383789, 11.93333625793457, 13.352996826171875, 13.779975891113281, 5.9969482421875, 6.095359802246094, -6.044654846191406, 9.684707641601562, -4.9589996337890625, 4.7935638427734375, 3.8006134033203125, -3.890420913696289, 10.122051239013672, 1.3958358764648438, 0.65252685546875, -2.4713287353515625, 3.2287368774414062, 15.543193817138672, -4.4380950927734375, 13.115961074829102, 15.674819946289062, 19.414657592773438, 0.7358169555664062, -15.043556213378906, 3.40936279296875, 1.9102535247802734, 7.214670181274414, 7.594673156738281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000144.npy"}
|
|
{"epoch": 0.21768707482993196, "step": 145, "batch_size": 64, "mean": 3.6623246669769287, "std": 6.387692928314209, "min": -15.736846923828125, "p10": -2.8364974975585935, "median": 3.6097583770751953, "p90": 12.606954193115236, "max": 19.879070281982422, "pos_frac": 0.734375, "sample": [4.5035552978515625, 4.822969436645508, -2.9109344482421875, 1.997650146484375, 0.30974578857421875, -15.736846923828125, 5.2784576416015625, -0.6760482788085938, 19.879070281982422, 4.7878265380859375, 4.949708938598633, 6.082923889160156, 2.0778465270996094, 5.925205230712891, 1.995361328125, 6.017692565917969, -1.0867652893066406, 12.293937683105469, -2.662811279296875, 5.4912872314453125, 1.3325042724609375, 2.439533233642578, -3.738056182861328, 6.261016845703125, 14.440290451049805, -0.8903884887695312, 0.3309173583984375, 6.901641845703125, 6.9941558837890625, -0.2915840148925781, 16.372623443603516, 9.4398193359375, 3.0518341064453125, 16.445091247558594, 0.573211669921875, 13.620851516723633, 2.7693424224853516, 8.167545318603516, 4.597507476806641, 9.40376091003418, -9.062820434570312, 5.630603790283203, 12.741104125976562, 4.999031066894531, 0.3611106872558594, -0.753509521484375, 2.0856456756591797, 4.167682647705078, -2.386066436767578, 6.6878204345703125, 7.815601348876953, 6.021249771118164, 9.926063537597656, -1.045440673828125, 1.8296737670898438, 13.400146484375, 1.5813140869140625, -4.015613555908203, -10.5025634765625, 2.5345001220703125, 9.4365234375, -5.613170623779297, -2.1801910400390625, -0.8313674926757812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000145.npy"}
|
|
{"epoch": 0.21919879062736206, "step": 146, "batch_size": 64, "mean": 3.874772310256958, "std": 5.866604328155518, "min": -8.412391662597656, "p10": -4.293555450439452, "median": 3.614114761352539, "p90": 11.261217498779299, "max": 19.411884307861328, "pos_frac": 0.734375, "sample": [9.481399536132812, 7.293190002441406, -3.5775527954101562, 3.7221450805664062, 1.3972625732421875, -4.465919494628906, -2.430084228515625, 6.503028869628906, 6.319324493408203, 5.0987548828125, -5.66143798828125, -3.6522445678710938, 8.348922729492188, 5.9261627197265625, 3.506084442138672, -4.836265563964844, 8.18269157409668, -0.48828125, 9.26310920715332, -8.412391662597656, 4.8318328857421875, 12.246414184570312, 2.0554962158203125, -0.17523956298828125, -5.954814910888672, 0.3034629821777344, 2.5691757202148438, 10.309226989746094, -4.678417205810547, 2.6328201293945312, 5.664407730102539, -5.002857208251953, 16.716171264648438, 3.8362808227539062, 3.98895263671875, -1.1859970092773438, 2.8445510864257812, 0.6572494506835938, 7.002960205078125, -0.08708763122558594, 10.895339965820312, 5.109737396240234, 2.2303218841552734, 7.816253662109375, 9.363800048828125, 14.558975219726562, 8.724761962890625, 5.128786087036133, 9.996940612792969, 2.7146530151367188, 5.5956268310546875, 0.986968994140625, 12.345865249633789, 3.237873077392578, -3.6118030548095703, -3.8913726806640625, 19.411884307861328, -1.504241943359375, 3.0406551361083984, 2.933889389038086, 11.834892272949219, 7.069164276123047, 2.4859466552734375, 11.418022155761719], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000146.npy"}
|
|
{"epoch": 0.22071050642479215, "step": 147, "batch_size": 64, "mean": 3.096158504486084, "std": 6.243993282318115, "min": -15.216796875, "p10": -2.3356498718261713, "median": 2.6982650756835938, "p90": 10.74941024780274, "max": 16.96502685546875, "pos_frac": 0.6875, "sample": [4.288820266723633, 6.044319152832031, 9.235282897949219, -1.5304374694824219, 2.7464447021484375, 6.1262054443359375, -5.753578186035156, 0.9262847900390625, -1.710235595703125, -1.9714813232421875, 6.638694763183594, 4.491809844970703, 7.347625732421875, 6.941253662109375, -1.4470672607421875, 9.738380432128906, 15.095771789550781, -15.216796875, 7.623226165771484, -1.0184078216552734, 3.7830982208251953, 1.5459823608398438, 0.03833961486816406, 2.1336669921875, 0.4632568359375, 12.752891540527344, -2.4917221069335938, 7.360660552978516, 3.4280776977539062, 2.65008544921875, -0.6766376495361328, 3.5233230590820312, 3.90911865234375, -6.158031463623047, -1.6522445678710938, 7.386375427246094, 9.499723434448242, 16.96502685546875, 14.586540222167969, -0.4298248291015625, 2.91864013671875, 0.2608642578125, 0.10842132568359375, 7.575418472290039, 7.952537536621094, -3.5206031799316406, 16.767478942871094, -1.2413520812988281, 9.492866516113281, -0.19435882568359375, 1.214111328125, 4.602935791015625, 5.5289764404296875, -1.2220172882080078, -1.69403076171875, 0.35732269287109375, 2.5771312713623047, 0.3755664825439453, 11.182708740234375, -11.06805419921875, -9.095588684082031, 4.41363525390625, 13.854909896850586, -0.2071990966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000147.npy"}
|
|
{"epoch": 0.2222222222222222, "step": 148, "batch_size": 64, "mean": 3.770010232925415, "std": 5.9365973472595215, "min": -10.474945068359375, "p10": -4.078484344482421, "median": 4.406822204589844, "p90": 10.344964599609376, "max": 16.8665771484375, "pos_frac": 0.734375, "sample": [12.3837890625, -7.012979507446289, 4.930612564086914, -5.812206268310547, 10.4765625, 15.085418701171875, 6.006980895996094, -8.230781555175781, 9.44915771484375, 11.724853515625, -1.4876060485839844, 8.241752624511719, 2.487773895263672, 2.7672195434570312, 7.360919952392578, -2.8062362670898438, -2.035482406616211, 10.03790283203125, 6.1539459228515625, -2.0557518005371094, 6.779994964599609, -4.868804931640625, 9.497817993164062, 1.6416015625, 7.5684051513671875, 14.912918090820312, 8.027732849121094, -0.71905517578125, 15.428985595703125, 2.4407806396484375, 4.482494354248047, -1.69561767578125, 8.908987045288086, -4.6237335205078125, 5.0633392333984375, 8.460311889648438, 0.911895751953125, 9.720619201660156, -10.474945068359375, -0.011362075805664062, 4.001792907714844, 0.05106353759765625, 8.601493835449219, 16.8665771484375, -7.144689559936523, 5.501142501831055, 4.444175720214844, 3.4130725860595703, 8.026802062988281, 1.1387176513671875, 6.846757888793945, 0.3926849365234375, 2.616273880004883, 0.7550315856933594, 0.11503791809082031, 7.47589111328125, 5.894622802734375, 5.142673492431641, 4.404212951660156, -0.6500320434570312, 1.26397705078125, 4.409431457519531, -0.9133987426757812, -0.4908599853515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000148.npy"}
|
|
{"epoch": 0.2237339380196523, "step": 149, "batch_size": 64, "mean": 4.3560028076171875, "std": 5.44849157333374, "min": -10.226810455322266, "p10": -1.5523681640625, "median": 3.988234519958496, "p90": 10.42547607421875, "max": 17.852813720703125, "pos_frac": 0.796875, "sample": [8.093055725097656, 9.877471923828125, 9.728252410888672, -1.38433837890625, 4.377166748046875, 0.5614662170410156, -1.5734138488769531, 8.154775619506836, -0.8810234069824219, 9.530597686767578, -5.87896728515625, 3.692045211791992, 0.5799942016601562, 3.265186309814453, 7.7855987548828125, 0.08916473388671875, 2.310771942138672, 3.0408782958984375, 0.9529609680175781, 5.24725341796875, 1.966796875, 10.230148315429688, 12.504426956176758, -1.4704475402832031, 17.852813720703125, 4.447626113891602, 3.3113136291503906, -1.725881576538086, 5.107419967651367, -10.226810455322266, -1.5032615661621094, 8.978523254394531, -6.6089935302734375, 15.103490829467773, 1.506683349609375, 12.757600784301758, 8.204385757446289, -2.7596359252929688, 8.973114013671875, -2.6605911254882812, -0.0063629150390625, 5.1343841552734375, 1.2803115844726562, 8.876876831054688, 10.364139556884766, 10.307134628295898, 5.579109191894531, 6.383243560791016, 3.371458053588867, 3.6252593994140625, 6.040153503417969, 3.3656997680664062, 1.2976531982421875, -1.159332275390625, 0.02597808837890625, 7.654510498046875, 4.284423828125, 15.72528076171875, 10.451763153076172, 3.604299545288086, 4.507011413574219, 10.815587997436523, 0.42755126953125, 5.270408630371094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000149.npy"}
|
|
{"epoch": 0.2252456538170824, "step": 150, "batch_size": 64, "mean": 5.431859493255615, "std": 6.133857250213623, "min": -5.7869720458984375, "p10": -2.2284606933593745, "median": 4.453893661499023, "p90": 14.198203277587892, "max": 20.257080078125, "pos_frac": 0.84375, "sample": [4.486202239990234, 2.0706539154052734, 14.58917236328125, 2.3949661254882812, -4.985755920410156, -1.3805351257324219, 3.8515548706054688, 8.693832397460938, 2.79547119140625, 7.983835220336914, 1.3028640747070312, 7.059745788574219, 0.348358154296875, 5.425880432128906, 0.4964122772216797, 4.21490478515625, 3.2573623657226562, 19.28509521484375, 1.73529052734375, 6.1748809814453125, 5.296234130859375, 0.9540443420410156, 13.832565307617188, -5.7869720458984375, 0.93023681640625, 9.054605484008789, 2.1269702911376953, 6.299564361572266, 2.609344482421875, 14.343032836914062, 7.390594482421875, 9.089080810546875, 12.470260620117188, 3.8236465454101562, 17.8175048828125, -4.852622985839844, 1.936574935913086, 7.258945465087891, 8.00146484375, 15.89959716796875, 6.470672607421875, 12.605323791503906, -1.8987655639648438, 5.425025939941406, 13.860267639160156, 2.886322021484375, 10.057283401489258, 11.310611724853516, -5.417549133300781, 20.257080078125, 3.3389892578125, -0.5656585693359375, 6.2042083740234375, -2.5743141174316406, 1.9543285369873047, 14.595817565917969, -4.7912750244140625, -2.3697586059570312, 10.801826477050781, 0.34650421142578125, 4.053920745849609, 9.482513427734375, 4.4215850830078125, 8.889183044433594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000150.npy"}
|
|
{"epoch": 0.22675736961451248, "step": 151, "batch_size": 64, "mean": 5.999746799468994, "std": 6.515707015991211, "min": -6.600231170654297, "p10": -1.661405372619629, "median": 6.318927764892578, "p90": 13.489552307128907, "max": 24.77197265625, "pos_frac": 0.765625, "sample": [6.2671356201171875, 8.64719009399414, 11.228874206542969, -2.5915603637695312, 1.4687042236328125, 2.2943878173828125, 24.77197265625, 1.2439537048339844, 13.2447509765625, 1.1024665832519531, 5.748809814453125, 10.403499603271484, 4.9931640625, -0.5079193115234375, 6.912616729736328, 8.483705520629883, 10.834428787231445, -1.1626758575439453, 5.537054061889648, -0.8990249633789062, -2.8012619018554688, -1.643117904663086, 8.390403747558594, -2.9948997497558594, 21.924301147460938, -5.2835235595703125, 8.444662094116211, 6.520938873291016, -1.6901378631591797, -0.5884780883789062, 5.102840423583984, -1.1314620971679688, 9.682327270507812, 6.549774169921875, 2.733123779296875, -0.860931396484375, 6.969413757324219, 7.6460418701171875, 3.8147354125976562, 20.42926025390625, 0.6626052856445312, 6.7400054931640625, 13.770748138427734, 4.014913558959961, 17.73113250732422, 12.309326171875, 13.594467163085938, 10.635757446289062, 12.856369018554688, 4.943450927734375, 6.370719909667969, 8.276168823242188, 2.7016983032226562, 15.589332580566406, -6.600231170654297, 9.183769226074219, 8.722597122192383, 4.275661468505859, -1.6692428588867188, -0.09027099609375, 7.3626708984375, 12.35745620727539, 0.28402137756347656, 10.725135803222656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000151.npy"}
|
|
{"epoch": 0.22826908541194255, "step": 152, "batch_size": 64, "mean": 4.019511699676514, "std": 7.0197529792785645, "min": -10.723739624023438, "p10": -4.453387451171874, "median": 3.131275177001953, "p90": 14.405239295959475, "max": 17.236801147460938, "pos_frac": 0.703125, "sample": [-1.4243030548095703, 1.27191162109375, 10.053014755249023, 2.1524887084960938, -10.723739624023438, 1.4922409057617188, 7.4585723876953125, 6.8300628662109375, 3.9814910888671875, -5.2895355224609375, 3.5554752349853516, -7.3680877685546875, 6.065952301025391, -0.38634490966796875, 13.183998107910156, 13.024131774902344, 1.4001350402832031, 4.780677795410156, 1.3007049560546875, -2.485424041748047, -10.400932312011719, 15.15155029296875, 7.009773254394531, 7.460845947265625, -1.9299507141113281, 1.2530994415283203, 10.230491638183594, 3.9710540771484375, 1.6425819396972656, 4.696601867675781, 1.7835063934326172, 15.902341842651367, 0.9478168487548828, 2.7257843017578125, -0.8407821655273438, -4.0360107421875, 1.7981853485107422, -2.1728134155273438, 9.329307556152344, 2.3696136474609375, -4.63226318359375, 14.781797409057617, 9.015289306640625, -3.5422592163085938, 7.329856872558594, 17.236801147460938, 9.908004760742188, -0.38646697998046875, 2.27093505859375, 4.4323272705078125, -9.613006591796875, 12.730636596679688, -0.2402172088623047, 16.979820251464844, 10.528553009033203, 5.054054260253906, 16.060935974121094, 13.526603698730469, -2.2321319580078125, 15.725547790527344, -1.1043167114257812, -6.820213317871094, 3.5367660522460938, 10.936203002929688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000152.npy"}
|
|
{"epoch": 0.22978080120937264, "step": 153, "batch_size": 64, "mean": 4.523066997528076, "std": 5.772233963012695, "min": -7.8549346923828125, "p10": -3.3502857208251955, "median": 3.8573341369628906, "p90": 12.075769805908203, "max": 16.166351318359375, "pos_frac": 0.78125, "sample": [7.513040542602539, 2.6529464721679688, 5.196998596191406, 11.533920288085938, 9.890130996704102, 14.861080169677734, 6.8976287841796875, 2.794506072998047, 2.7798995971679688, 3.2643966674804688, 4.644388198852539, 12.122058868408203, 2.115203857421875, 10.039539337158203, 3.7136383056640625, -4.0602569580078125, 6.1892242431640625, 0.5965576171875, 2.2971649169921875, 13.374099731445312, 16.166351318359375, 10.215370178222656, -3.333972930908203, 4.741569519042969, 15.986248016357422, -5.91743278503418, 11.124134063720703, 1.9773902893066406, 0.2987213134765625, 11.967761993408203, -4.803672790527344, 2.8324508666992188, -7.8549346923828125, 0.6222801208496094, 6.519805908203125, 2.9636402130126953, -0.8916397094726562, 11.332618713378906, 4.41424560546875, 3.6436004638671875, -0.152557373046875, -1.1715621948242188, 13.39208984375, 6.517749786376953, 6.251285552978516, -0.5742988586425781, 6.876335144042969, 4.001029968261719, 7.259895324707031, -1.2685222625732422, 0.314788818359375, 9.448760986328125, 2.7233448028564453, 0.6580047607421875, 5.11859130859375, -4.395698547363281, -5.0399017333984375, 11.626005172729492, -0.964935302734375, 12.515914916992188, 11.163352966308594, 0.7671604156494141, -3.3572769165039062, 7.3460235595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000153.npy"}
|
|
{"epoch": 0.23129251700680273, "step": 154, "batch_size": 64, "mean": 3.3994009494781494, "std": 6.537519454956055, "min": -13.441333770751953, "p10": -6.151721572875976, "median": 4.129621505737305, "p90": 11.24611053466797, "max": 17.21354866027832, "pos_frac": 0.71875, "sample": [-6.9457855224609375, 3.00390625, -9.044509887695312, -13.441333770751953, 13.326751708984375, 8.761327743530273, -7.253194808959961, 4.245731353759766, 3.2353363037109375, -6.259773254394531, 3.9959983825683594, -0.5127296447753906, 8.522397994995117, 2.23309326171875, -5.138664245605469, -5.03521728515625, -1.8392677307128906, 4.833576202392578, 0.4318828582763672, -0.0826873779296875, 9.80133056640625, 17.21354866027832, 15.88247299194336, 4.013511657714844, -7.122749328613281, 5.743673324584961, 11.065834045410156, 0.01258087158203125, 12.435546875, -9.766098022460938, 5.899225234985352, 3.329854965209961, 10.885223388671875, 9.265029907226562, 5.658672332763672, 0.06250762939453125, 4.902524948120117, 9.759841918945312, 6.237360000610352, 5.269115447998047, -1.34765625, -1.0512237548828125, 7.759466171264648, -0.9266071319580078, 1.392364501953125, 15.206418991088867, 5.946069717407227, 4.654563903808594, 7.507568359375, -5.899600982666016, 0.5292816162109375, -2.0836715698242188, 2.2634658813476562, 1.0682907104492188, 6.310546875, 4.385002136230469, 11.96917724609375, -0.3955860137939453, 10.830644607543945, 11.323371887207031, 6.5521697998046875, 5.679975509643555, 2.037149429321289, 6.264621734619141], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000154.npy"}
|
|
{"epoch": 0.2328042328042328, "step": 155, "batch_size": 64, "mean": 4.569487571716309, "std": 7.925663948059082, "min": -15.098331451416016, "p10": -6.304520416259765, "median": 4.319957733154297, "p90": 16.129763412475587, "max": 19.61009979248047, "pos_frac": 0.671875, "sample": [4.117240905761719, -0.303497314453125, 19.121185302734375, 5.005195617675781, 1.6978225708007812, 4.21881103515625, -0.14652633666992188, 12.003273010253906, -1.9315681457519531, 10.495075225830078, 8.804412841796875, 11.082807540893555, -0.1956939697265625, -7.4390106201171875, -0.154266357421875, -3.389801025390625, 6.63385009765625, 17.800506591796875, 6.758537292480469, 17.563236236572266, 15.557456970214844, -0.8637218475341797, 9.077674865722656, -8.280477523803711, 8.459716796875, 17.864501953125, -0.080352783203125, 10.799163818359375, -2.6112518310546875, 10.715034484863281, -10.364288330078125, 5.65031623840332, 16.257549285888672, 8.475166320800781, 19.61009979248047, 7.2664337158203125, -0.4697761535644531, -1.462005615234375, -9.748641967773438, 11.977836608886719, -5.419891357421875, 1.1620521545410156, 9.022613525390625, 0.8352088928222656, 17.297027587890625, 5.593070983886719, 4.279052734375, 13.011848449707031, -6.683647155761719, 4.1020355224609375, 5.741111755371094, 5.30169677734375, 0.8181743621826172, 1.8718719482421875, 2.5695743560791016, 0.22727203369140625, 6.232460021972656, -7.18939208984375, -15.098331451416016, -0.600830078125, 4.360862731933594, 15.831596374511719, 12.305923461914062, -2.6961898803710938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000155.npy"}
|
|
{"epoch": 0.23431594860166288, "step": 156, "batch_size": 64, "mean": 5.373299598693848, "std": 6.523813724517822, "min": -7.8274078369140625, "p10": -2.809964179992675, "median": 4.491199493408203, "p90": 14.856770324707034, "max": 21.84231948852539, "pos_frac": 0.8125, "sample": [4.706451416015625, -1.8594417572021484, -3.234651565551758, -7.8274078369140625, 8.491683959960938, 17.402236938476562, 21.84231948852539, 7.169990539550781, 8.428939819335938, 15.10184097290039, 2.1677169799804688, 13.029827117919922, -3.2931060791015625, 9.587127685546875, 11.71343994140625, 11.833721160888672, 14.289596557617188, 10.413261413574219, -1.219278335571289, 8.429691314697266, 2.6454715728759766, 1.6718502044677734, 6.4764404296875, 10.150321960449219, 0.02730560302734375, 2.7343063354492188, 0.1224212646484375, 9.971817016601562, 4.275947570800781, 9.620208740234375, 5.1228790283203125, 3.5959625244140625, -3.2173309326171875, 8.8277587890625, 5.70025634765625, 18.509666442871094, -6.67071533203125, 6.8587799072265625, -4.502647399902344, 16.252288818359375, 8.282806396484375, 16.890289306640625, 2.152080535888672, 1.4723396301269531, 0.44602394104003906, 2.210569381713867, 0.11771392822265625, -1.4029541015625, 5.1058807373046875, -0.6386871337890625, 4.153467178344727, 1.29205322265625, 6.770174026489258, 0.48821258544921875, 5.124931335449219, 3.1063079833984375, -1.8047714233398438, 10.234397888183594, 15.093017578125, 1.6443614959716797, 4.16670036315918, 14.305526733398438, 3.218273162841797, -3.8844833374023438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000156.npy"}
|
|
{"epoch": 0.23582766439909297, "step": 157, "batch_size": 64, "mean": 4.330805778503418, "std": 6.7083258628845215, "min": -12.165054321289062, "p10": -3.206073379516601, "median": 4.699699401855469, "p90": 12.921266937255863, "max": 22.37603759765625, "pos_frac": 0.765625, "sample": [16.288467407226562, 17.056903839111328, -1.8028640747070312, 7.899482727050781, 10.254188537597656, -11.207160949707031, 3.0706787109375, -2.752105712890625, 9.878910064697266, -2.3029403686523438, 0.6486854553222656, 4.885711669921875, 11.755157470703125, 3.2270450592041016, -12.165054321289062, 0.10830879211425781, 1.2562637329101562, 3.27099609375, 9.059572219848633, 13.947420120239258, 6.39990234375, 11.329952239990234, 22.37603759765625, 7.08357048034668, 11.191858291625977, 6.494224548339844, -3.653472900390625, -0.39898109436035156, 5.604358673095703, -1.598184585571289, 10.888038635253906, 0.09539794921875, 6.200738906860352, 4.91546630859375, 7.3645172119140625, 13.421028137207031, -5.144172668457031, 14.97957992553711, 15.088516235351562, 1.0457000732421875, -3.4006309509277344, 3.6624374389648438, -0.5912704467773438, 9.63873291015625, 8.312164306640625, 2.0418643951416016, 1.04693603515625, 6.481468200683594, 5.210508346557617, 1.0913028717041016, 2.28985595703125, -2.0178604125976562, 5.277168273925781, 1.3023910522460938, 5.221668243408203, 8.819215774536133, 1.63140869140625, 6.6740264892578125, -5.677154541015625, 9.094017028808594, 0.0463104248046875, -0.7240447998046875, -8.834381103515625, 4.5136871337890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000157.npy"}
|
|
{"epoch": 0.23733938019652306, "step": 158, "batch_size": 64, "mean": 4.462376594543457, "std": 6.946750640869141, "min": -12.583587646484375, "p10": -3.1464441299438475, "median": 3.408672332763672, "p90": 14.386943817138672, "max": 19.214447021484375, "pos_frac": 0.71875, "sample": [0.9640274047851562, 8.337814331054688, 3.107177734375, 5.193941116333008, -12.583587646484375, 9.826692581176758, 15.595611572265625, 8.428108215332031, -0.35308837890625, -0.07421112060546875, 14.512939453125, 5.74725341796875, 6.9984130859375, 9.230194091796875, -10.053276062011719, 8.342370986938477, 19.214447021484375, 3.3519649505615234, 2.4344024658203125, 5.065296173095703, -1.8400382995605469, 1.7991485595703125, 8.41629409790039, 0.548431396484375, 14.255805969238281, 8.580841064453125, -2.118621826171875, 15.688697814941406, 9.970321655273438, 7.837066650390625, -2.7286224365234375, 7.1389617919921875, 14.202430725097656, 2.7058181762695312, 14.159280776977539, 14.611791610717773, 10.177467346191406, -3.23443603515625, 2.249309539794922, -5.29229736328125, 12.9017333984375, 4.260841369628906, 3.42132568359375, 14.443145751953125, 7.3720245361328125, 8.463394165039062, -2.2929553985595703, 0.7090225219726562, 18.539886474609375, -2.941129684448242, 3.3960189819335938, -2.371337890625, -4.379383087158203, 1.4686393737792969, 0.9203815460205078, -0.6432094573974609, -2.884256362915039, 6.126934051513672, 7.6696624755859375, 2.6773643493652344, -6.913379669189453, 1.6867485046386719, -0.12060546875, -6.332916259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000158.npy"}
|
|
{"epoch": 0.23885109599395313, "step": 159, "batch_size": 64, "mean": 5.16028356552124, "std": 7.447564125061035, "min": -20.690048217773438, "p10": -2.7087537765502927, "median": 4.394725799560547, "p90": 15.52455863952637, "max": 21.270477294921875, "pos_frac": 0.8125, "sample": [10.904163360595703, 9.67149543762207, 1.1646804809570312, 12.631538391113281, 5.608680725097656, -5.457405090332031, -5.1986083984375, 16.252986907958984, 1.0673904418945312, 21.270477294921875, 19.310134887695312, 3.7506179809570312, 13.050310134887695, 4.089729309082031, -2.184284210205078, 3.474132537841797, 4.949180603027344, 20.827301025390625, 11.208393096923828, -0.04087257385253906, 1.3865509033203125, -6.8799285888671875, 3.5973453521728516, 2.6535263061523438, 4.016761779785156, 3.1066951751708984, 2.1776981353759766, -5.061408996582031, 12.166519165039062, 16.563400268554688, -2.732433319091797, 8.05487060546875, 7.299896240234375, 0.7301101684570312, 2.5009765625, 6.456010818481445, 14.678394317626953, 7.3377838134765625, 2.1988525390625, 0.01149749755859375, 3.7585983276367188, -6.27203369140625, 16.671173095703125, 0.808349609375, 6.636070251464844, 15.871379852294922, -2.653501510620117, 8.525224685668945, -0.46916961669921875, 4.6997222900390625, 0.2833213806152344, 14.715309143066406, 5.702812194824219, -1.299224853515625, 0.07853317260742188, -20.690048217773438, 11.475580215454102, 8.050445556640625, 0.1632843017578125, 5.286542892456055, 8.756576538085938, 5.5589141845703125, 9.783416748046875, 8.203704833984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000159.npy"}
|
|
{"epoch": 0.24036281179138322, "step": 160, "batch_size": 64, "mean": 4.414180755615234, "std": 7.78603982925415, "min": -11.91461181640625, "p10": -3.721882057189941, "median": 2.984701156616211, "p90": 14.499375534057616, "max": 24.976882934570312, "pos_frac": 0.71875, "sample": [1.6139411926269531, -2.192432403564453, 8.524337768554688, 20.268882751464844, 1.5941696166992188, 15.153633117675781, 1.462982177734375, -1.9869766235351562, 4.991336822509766, 8.468902587890625, 23.272476196289062, -2.7215003967285156, 13.026029586791992, 11.953147888183594, 5.953639984130859, 14.461326599121094, 0.10608673095703125, -8.747047424316406, 8.048858642578125, 8.225954055786133, -2.4083023071289062, -5.094062805175781, 2.0382232666015625, -1.3428955078125, 5.342809677124023, 2.114421844482422, 7.785198211669922, 1.505990982055664, 3.9925613403320312, 0.9097404479980469, 6.635076522827148, 5.667116165161133, 24.976882934570312, 18.704769134521484, 1.9006175994873047, -3.6533851623535156, 9.669013977050781, 1.0786399841308594, -0.3528099060058594, 14.515682220458984, 0.9267082214355469, 9.541221618652344, 1.4488601684570312, -6.579132080078125, -2.7855300903320312, 6.0737457275390625, -8.910301208496094, -3.7512378692626953, 8.858985900878906, 6.912178039550781, -6.446037292480469, -2.5902271270751953, -2.765380859375, 8.362430572509766, 9.290351867675781, 1.6288299560546875, 1.1167831420898438, 17.361495971679688, -11.91461181640625, 3.85498046875, 12.375602722167969, 14.385948181152344, 4.00042724609375, -3.3515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000160.npy"}
|
|
{"epoch": 0.2418745275888133, "step": 161, "batch_size": 64, "mean": 4.799029350280762, "std": 7.4354729652404785, "min": -17.490257263183594, "p10": -3.0002840995788573, "median": 4.564769744873047, "p90": 13.751959991455077, "max": 23.848190307617188, "pos_frac": 0.75, "sample": [4.837745666503906, -10.796321868896484, 2.452392578125, 5.5432281494140625, 1.6105194091796875, -0.6757888793945312, 10.410400390625, 18.426734924316406, -4.265850067138672, -2.9530153274536133, -7.2281951904296875, 6.418235778808594, 6.930601119995117, 8.051132202148438, 16.435646057128906, 5.933320999145508, 5.310253143310547, -17.490257263183594, 9.61322021484375, -3.6199264526367188, -2.8435134887695312, 1.3878021240234375, 3.7651519775390625, 1.5479965209960938, 9.126113891601562, -2.597858428955078, 2.5570335388183594, 8.482950210571289, 8.006965637207031, -0.9794235229492188, 12.805631637573242, 11.915306091308594, 7.342607498168945, 13.719718933105469, 13.765777587890625, -2.2217864990234375, -5.032417297363281, 23.848190307617188, 15.750694274902344, 13.649188995361328, 3.083141326904297, 11.816757202148438, -0.016201019287109375, 3.6673736572265625, 5.768898010253906, 1.5862274169921875, 3.556964874267578, 6.216846466064453, 4.2917938232421875, 11.304100036621094, -3.0205421447753906, 5.080146789550781, 17.5327091217041, 3.287120819091797, 0.17595672607421875, 0.4238014221191406, -2.283966064453125, 2.4693260192871094, 8.140274047851562, 6.784912109375, 21.206787109375, -0.7556076049804688, 0.6952018737792969, 7.185657501220703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000161.npy"}
|
|
{"epoch": 0.24338624338624337, "step": 162, "batch_size": 64, "mean": 6.362528324127197, "std": 7.0631842613220215, "min": -8.38995361328125, "p10": -4.584012603759764, "median": 6.411273956298828, "p90": 15.155873870849609, "max": 25.661407470703125, "pos_frac": 0.8125, "sample": [9.235305786132812, 8.271209716796875, -8.38995361328125, 3.3110809326171875, 1.7793045043945312, 12.185237884521484, 7.871978759765625, 4.370719909667969, 14.203445434570312, 16.77237892150879, 7.296272277832031, 10.55172348022461, 4.616872787475586, 25.661407470703125, 11.698760986328125, 5.531454086303711, 6.40142822265625, 18.00031280517578, 9.534637451171875, 17.10128402709961, 15.169082641601562, 12.326961517333984, 3.1809158325195312, -3.379547119140625, 7.4152984619140625, -2.601531982421875, 8.298126220703125, 1.9590835571289062, -6.973541259765625, 9.569427490234375, 3.364175796508789, 4.0192718505859375, 1.0089569091796875, 12.072484970092773, 12.11517333984375, 6.048126220703125, 10.9542236328125, 10.955755233764648, -0.021394729614257812, -5.688852310180664, -5.824470520019531, -0.13623809814453125, 15.125053405761719, 13.807151794433594, 9.706186294555664, 4.370758056640625, -6.802398681640625, 10.013381958007812, -5.4407958984375, -0.03719329833984375, 4.971416473388672, 0.10602569580078125, 6.421119689941406, 11.491281509399414, 15.183570861816406, 8.729705810546875, 1.6067752838134766, -5.100212097167969, 3.1494903564453125, 3.157318115234375, 3.7155399322509766, 14.42538070678711, 15.316682815551758, 3.44921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000162.npy"}
|
|
{"epoch": 0.24489795918367346, "step": 163, "batch_size": 64, "mean": 6.252363204956055, "std": 7.944355487823486, "min": -10.818109512329102, "p10": -5.495306777954101, "median": 5.857735633850098, "p90": 16.70326881408692, "max": 21.19704818725586, "pos_frac": 0.78125, "sample": [-1.327988624572754, 7.630302429199219, -7.8087158203125, 12.70550537109375, 5.839134216308594, 1.7434120178222656, -10.818109512329102, 2.8800048828125, 17.72894287109375, -7.040763854980469, 6.595237731933594, 11.068557739257812, 18.751792907714844, 3.8275527954101562, 3.8461532592773438, 10.06832504272461, 12.887714385986328, -0.4955291748046875, 20.956268310546875, 7.500633239746094, 1.4110641479492188, 12.835990905761719, 14.08163070678711, -5.450469970703125, -5.733600616455078, 17.286972045898438, 6.096508026123047, 14.302177429199219, 5.876337051391602, 4.196510314941406, 9.147830963134766, 3.655303955078125, 10.375457763671875, -0.8119640350341797, 12.055526733398438, 13.218910217285156, -5.514522552490234, 13.921722412109375, -5.365989685058594, 2.212839126586914, 10.862926483154297, 21.19704818725586, 8.308387756347656, 14.90028190612793, 2.8895263671875, -2.1186981201171875, -0.4853324890136719, 5.158367156982422, 3.0937728881835938, 0.5798187255859375, -9.369140625, 15.264549255371094, 12.046321868896484, 0.5153217315673828, 2.7308311462402344, -5.894500732421875, 15.987869262695312, 18.190093994140625, 17.009868621826172, 14.060966491699219, 8.785120010375977, 2.7618579864501953, 4.35413932800293, 4.985170364379883], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000163.npy"}
|
|
{"epoch": 0.24640967498110355, "step": 164, "batch_size": 64, "mean": 6.10004186630249, "std": 8.122475624084473, "min": -16.216007232666016, "p10": -2.9386302947998044, "median": 5.280632019042969, "p90": 16.513520812988283, "max": 25.40472412109375, "pos_frac": 0.796875, "sample": [-1.3688888549804688, 0.9143543243408203, -0.5706558227539062, 4.438816070556641, 4.680397033691406, 5.015419006347656, 22.46136474609375, 13.892524719238281, -0.46175384521484375, 2.6291770935058594, 10.8299560546875, -2.5217971801757812, 12.422372817993164, -8.374433517456055, 7.548679351806641, 10.105224609375, 13.434516906738281, 5.319053649902344, 20.759063720703125, -2.991220474243164, 4.467741012573242, -6.638572692871094, 6.093730926513672, 6.018585205078125, 0.6240043640136719, 5.242210388183594, 16.274124145507812, 2.3737411499023438, 8.327499389648438, 2.0222015380859375, 15.275510787963867, 18.55329132080078, -2.815919876098633, 6.964931488037109, 4.881813049316406, -6.634803771972656, 16.616119384765625, 25.40472412109375, 19.127681732177734, -7.571281433105469, 4.429147720336914, 4.822959899902344, 1.7100868225097656, 11.21986198425293, 6.142711639404297, 7.440071105957031, 14.223403930664062, 1.2340335845947266, 12.940582275390625, 1.3805122375488281, 3.5986328125, 1.8766860961914062, 16.68183135986328, -6.534746170043945, -0.48416709899902344, -16.216007232666016, 6.0781402587890625, 14.945791244506836, 15.262191772460938, 0.639801025390625, 15.553413391113281, 6.173431396484375, 6.954174041748047, 7.560646057128906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000164.npy"}
|
|
{"epoch": 0.24792139077853365, "step": 165, "batch_size": 64, "mean": 5.55222749710083, "std": 7.841665744781494, "min": -13.308097839355469, "p10": -3.72901439666748, "median": 5.212632179260254, "p90": 15.382766342163087, "max": 20.69378662109375, "pos_frac": 0.734375, "sample": [20.69378662109375, 7.429868698120117, 7.765447616577148, 13.077442169189453, 5.017784118652344, -0.32558441162109375, 3.681720733642578, -0.6808319091796875, 8.450031280517578, 3.730396270751953, 1.7764205932617188, 7.142986297607422, 16.700958251953125, 1.6273670196533203, -0.5694580078125, 4.576507568359375, 5.182281494140625, 8.249687194824219, -11.015869140625, -2.0171966552734375, -11.202438354492188, -1.1795768737792969, 14.562957763671875, -1.8862190246582031, 15.584945678710938, 16.68753433227539, 5.242982864379883, 8.433740615844727, 18.237642288208008, 18.55651092529297, 4.978477478027344, 0.6719417572021484, 17.642547607421875, 14.911014556884766, 9.960273742675781, -13.308097839355469, 1.4044647216796875, 10.641864776611328, 10.75680923461914, 12.570419311523438, -3.2408390045166016, -0.33278656005859375, -0.4222869873046875, 2.5683841705322266, 14.463386535644531, -4.7189788818359375, 2.666179656982422, 1.5819549560546875, 9.357641220092773, -0.5673065185546875, -3.938232421875, 8.946022033691406, -12.121223449707031, 13.428077697753906, 3.9881839752197266, -5.066478729248047, 5.569328308105469, 3.621187210083008, 10.714994430541992, 5.331352233886719, 12.500411987304688, 14.754241943359375, 10.015113830566406, 12.482681274414062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000165.npy"}
|
|
{"epoch": 0.2494331065759637, "step": 166, "batch_size": 64, "mean": 4.166682243347168, "std": 7.009352684020996, "min": -6.441707611083984, "p10": -3.9887081146240235, "median": 3.0384254455566406, "p90": 13.649235916137698, "max": 22.217327117919922, "pos_frac": 0.625, "sample": [10.341659545898438, -1.2376689910888672, -3.65472412109375, 11.282920837402344, 12.683822631835938, 0.8313674926757812, -3.8802490234375, 14.028091430664062, 2.612060546875, 4.150970458984375, -4.550464630126953, 5.141559600830078, 16.57152557373047, -3.026519775390625, 10.96893310546875, 2.6723175048828125, -0.5888214111328125, 9.528043746948242, -1.0575942993164062, 1.3594551086425781, -2.7099456787109375, 12.765239715576172, -5.5100555419921875, 4.584831237792969, 14.381332397460938, 5.852409362792969, -1.6510162353515625, 9.84100341796875, 8.997177124023438, 4.493217468261719, -3.8382568359375, -6.441707611083984, 2.604053497314453, 16.527114868164062, 9.970771789550781, -1.3705329895019531, 12.709781646728516, 3.1264114379882812, 7.5432891845703125, 7.32148551940918, 2.950439453125, 19.291656494140625, -3.47381591796875, -5.4081878662109375, 5.68133544921875, -4.9030303955078125, -0.8609085083007812, -0.4125556945800781, 5.678539276123047, 22.217327117919922, 5.533393859863281, -5.0403289794921875, -0.1688079833984375, 8.869892120361328, -4.035190582275391, -1.6861724853515625, 8.225433349609375, -2.7092247009277344, 17.608627319335938, -2.134868621826172, 6.08831787109375, 9.155471801757812, 1.5909481048583984, 1.2361183166503906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000166.npy"}
|
|
{"epoch": 0.2509448223733938, "step": 167, "batch_size": 64, "mean": 6.430629730224609, "std": 7.281737804412842, "min": -9.361183166503906, "p10": -1.6907096862792965, "median": 4.871953010559082, "p90": 15.774962615966798, "max": 21.837249755859375, "pos_frac": 0.8125, "sample": [3.4174652099609375, 4.477130889892578, 2.2512760162353516, 6.086544036865234, -7.9036102294921875, 13.302055358886719, 10.101982116699219, 2.5741043090820312, 5.184789657592773, 10.244548797607422, 11.650690078735352, 4.559116363525391, 4.418849945068359, 0.40622711181640625, -0.16788673400878906, 0.7104454040527344, 15.713775634765625, 11.873214721679688, 13.981796264648438, -0.4160308837890625, -9.361183166503906, 2.0295677185058594, 3.740274429321289, 2.3902816772460938, 5.251350402832031, 1.6790313720703125, 3.4272232055664062, -0.1363239288330078, 4.478050231933594, 10.407379150390625, 3.5051040649414062, 11.257213592529297, 1.7870941162109375, 8.210853576660156, 8.832611083984375, 8.267257690429688, 20.100486755371094, 21.837249755859375, 13.690460205078125, 12.786552429199219, 1.6781883239746094, 12.306411743164062, -1.8565139770507812, 5.694610595703125, 12.046682357788086, -4.303253173828125, -6.7878570556640625, 20.13910675048828, -1.3038330078125, 11.282875061035156, 12.347156524658203, -0.14229965209960938, 1.0206184387207031, 20.835437774658203, 11.801168441772461, 16.998554229736328, -2.4642105102539062, 21.545486450195312, 11.242973327636719, 15.801185607910156, 6.9013519287109375, 1.3198909759521484, 1.5872344970703125, -2.7776718139648438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000167.npy"}
|
|
{"epoch": 0.25245653817082386, "step": 168, "batch_size": 64, "mean": 4.702498435974121, "std": 7.06962776184082, "min": -7.028144836425781, "p10": -3.002082061767578, "median": 3.0868072509765625, "p90": 15.706799125671388, "max": 19.54674530029297, "pos_frac": 0.703125, "sample": [6.709869384765625, -4.750022888183594, 16.08182144165039, -2.4849929809570312, 3.255746841430664, 14.185127258300781, 9.15185546875, -1.12054443359375, -7.028144836425781, -6.3484344482421875, 4.978660583496094, -2.5184974670410156, 3.0699462890625, 17.958301544189453, 6.6552886962890625, 2.1855125427246094, 9.794361114501953, 3.8940200805664062, -0.29137420654296875, 8.524856567382812, 7.728706359863281, -1.8185386657714844, 1.6739578247070312, 4.183347702026367, -0.3908863067626953, 5.958255767822266, 17.189563751220703, 1.8228492736816406, -5.363227844238281, 7.12451171875, 0.6558761596679688, 1.291748046875, 0.1261138916015625, 13.885093688964844, 14.217273712158203, -1.327117919921875, 12.084442138671875, 1.2664642333984375, -1.3377456665039062, 0.7115020751953125, 1.4966297149658203, 19.54674530029297, -2.8863143920898438, 15.526782989501953, 13.289054870605469, -0.5132293701171875, 9.947235107421875, -2.468994140625, 2.1877822875976562, -3.05169677734375, 8.528167724609375, 4.4746856689453125, 13.096141815185547, 3.103668212890625, -2.5426101684570312, 15.78394889831543, -3.1996612548828125, -5.451953887939453, 0.4103355407714844, 0.9397907257080078, 12.987058639526367, 17.57587432861328, 15.915542602539062, 4.679359436035156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000168.npy"}
|
|
{"epoch": 0.25396825396825395, "step": 169, "batch_size": 64, "mean": 5.947606563568115, "std": 9.286237716674805, "min": -15.936454772949219, "p10": -3.9628841400146473, "median": 4.799613952636719, "p90": 17.801773834228516, "max": 35.86383056640625, "pos_frac": 0.765625, "sample": [6.799171447753906, 15.238670349121094, 13.289398193359375, -6.858684539794922, 6.06450080871582, 2.2000370025634766, -4.409717559814453, 0.18051910400390625, 8.011783599853516, 8.69720458984375, 4.5666351318359375, 0.31664276123046875, 12.742828369140625, 17.245376586914062, 3.6066055297851562, -2.4337005615234375, 4.381755828857422, 8.001373291015625, 9.254257202148438, -1.0498733520507812, -2.9202728271484375, 16.44598388671875, 18.04022979736328, 6.533477783203125, -2.5504302978515625, 31.46587371826172, 1.722686767578125, 6.310302734375, -0.7097129821777344, 7.866451263427734, 3.391834259033203, 3.6691665649414062, 13.293453216552734, -2.374330520629883, 13.013551712036133, 3.286958694458008, 11.944475173950195, -15.936454772949219, 6.149099349975586, 4.285179138183594, 0.46446990966796875, 5.0325927734375, 2.018096923828125, 0.3886756896972656, 7.169374465942383, 4.181953430175781, 6.594261169433594, 2.22076416015625, 19.272811889648438, 9.488265991210938, -0.6480255126953125, 8.297050476074219, 6.467308044433594, -9.677154541015625, -5.29473876953125, 35.86383056640625, 19.150375366210938, 20.816993713378906, -10.382280349731445, 23.56381607055664, 3.4356613159179688, 12.474210739135742, -7.850372314453125, -1.1734199523925781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000169.npy"}
|
|
{"epoch": 0.25547996976568405, "step": 170, "batch_size": 64, "mean": 5.716863632202148, "std": 9.148222923278809, "min": -20.83148765563965, "p10": -7.152346038818359, "median": 6.399927139282227, "p90": 16.520081329345704, "max": 25.836929321289062, "pos_frac": 0.78125, "sample": [9.603271484375, 1.3491172790527344, 5.958333969116211, 6.8998260498046875, 3.18914794921875, 23.256912231445312, 4.981744766235352, 6.334707260131836, 5.097236633300781, 20.19908905029297, -9.383918762207031, 2.180156707763672, 11.326789855957031, 23.147125244140625, 18.313003540039062, -4.269021987915039, 2.872507095336914, 8.996818542480469, 13.157371520996094, -1.8217849731445312, 1.7398872375488281, 16.40679168701172, 8.654924392700195, 1.052194595336914, -7.344261169433594, 7.320037841796875, -9.549613952636719, -3.6327285766601562, -20.83148765563965, 9.917205810546875, 13.505664825439453, 8.200096130371094, 13.863082885742188, 6.465147018432617, -10.801982879638672, 6.204494476318359, -6.7045440673828125, 9.956356048583984, 15.227409362792969, -0.8098526000976562, 7.6795501708984375, -0.2921409606933594, 0.031917572021484375, 10.167259216308594, 7.043037414550781, 25.836929321289062, 6.73583984375, 16.568634033203125, 5.842411041259766, 0.6910858154296875, 0.5163192749023438, -3.0154666900634766, 1.530914306640625, 20.18658447265625, -8.08331298828125, 2.3685340881347656, 11.517097473144531, 16.233741760253906, 0.6339988708496094, 6.907962799072266, 13.676136016845703, 14.143951416015625, 8.860176086425781, -10.129127502441406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000170.npy"}
|
|
{"epoch": 0.25699168556311414, "step": 171, "batch_size": 64, "mean": 6.47409200668335, "std": 9.917036056518555, "min": -21.931488037109375, "p10": -2.3288246154785153, "median": 5.391374588012695, "p90": 20.187104034423832, "max": 23.926860809326172, "pos_frac": 0.71875, "sample": [6.02056884765625, 6.681713104248047, 18.129104614257812, 7.453578948974609, 12.827163696289062, 22.025127410888672, 10.856563568115234, 17.322982788085938, 0.44634437561035156, 3.627023696899414, -0.7989788055419922, 1.3789520263671875, 20.791358947753906, 9.615074157714844, 23.926860809326172, -1.5530624389648438, 13.604711532592773, 3.5415878295898438, 14.039712905883789, -0.9784164428710938, -0.6919097900390625, 4.158819198608398, 9.121387481689453, -3.71771240234375, 16.12018394470215, 0.3690681457519531, 7.9142913818359375, 1.2294921875, 21.724029541015625, -17.13982582092285, 8.664154052734375, 10.993675231933594, 23.136383056640625, -3.334613800048828, 9.042957305908203, 17.4217529296875, -2.4533157348632812, -7.575599670410156, 23.48861312866211, 18.777175903320312, -21.931488037109375, 2.2442703247070312, 11.933185577392578, 2.7668609619140625, 1.8678207397460938, 17.98717498779297, 14.08123779296875, -0.8843154907226562, -2.0383453369140625, 9.8331298828125, 4.762180328369141, 22.552474975585938, 14.96636962890625, 1.7012481689453125, -0.9065666198730469, -1.8740234375, 9.287582397460938, -0.18798446655273438, -1.0944938659667969, -16.447906494140625, 15.29925537109375, 1.0915870666503906, 3.69677734375, -0.5711135864257812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000171.npy"}
|
|
{"epoch": 0.2585034013605442, "step": 172, "batch_size": 64, "mean": 6.141225814819336, "std": 8.880919456481934, "min": -20.073989868164062, "p10": -2.7750534057617178, "median": 5.958779335021973, "p90": 17.958266067504884, "max": 26.468536376953125, "pos_frac": 0.78125, "sample": [-7.365776062011719, 6.582481384277344, 1.8140716552734375, 4.820884704589844, 1.9116744995117188, 1.6496829986572266, 0.017213821411132812, 9.924388885498047, 12.253738403320312, 18.564098358154297, 26.468536376953125, 6.9856719970703125, 5.9896697998046875, -1.0667304992675781, 10.569305419921875, 26.176000595092773, 9.567901611328125, -8.261138916015625, -0.5370254516601562, 6.589057922363281, 7.650970458984375, 2.0618667602539062, -10.486103057861328, 0.120635986328125, 6.190834045410156, 5.927888870239258, 4.1061248779296875, 11.929061889648438, 11.1455078125, -0.406280517578125, 8.047027587890625, 11.028205871582031, 7.6440887451171875, 14.029434204101562, 1.3956108093261719, -0.5738487243652344, 18.05817413330078, 10.155845642089844, 17.725147247314453, 2.012725830078125, -0.39653587341308594, 7.03216552734375, 5.867347717285156, 15.711675643920898, 14.958892822265625, -11.268543243408203, 1.5232772827148438, -1.8640899658203125, 3.38262939453125, 9.550588607788086, -5.456092834472656, 16.630706787109375, 19.657447814941406, -20.073989868164062, 2.6535415649414062, 13.904296875, 0.9519882202148438, 4.472267150878906, 1.9270477294921875, 19.814374923706055, -3.16546630859375, -0.6876316070556641, 21.284523010253906, 16.211380004882812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000172.npy"}
|
|
{"epoch": 0.2600151171579743, "step": 173, "batch_size": 64, "mean": 6.471645832061768, "std": 11.461297988891602, "min": -17.835676193237305, "p10": -6.952143669128417, "median": 7.485590934753418, "p90": 22.733749389648438, "max": 30.020050048828125, "pos_frac": 0.6875, "sample": [-10.103723526000977, 3.3620452880859375, -1.626983642578125, 8.114215850830078, 9.673088073730469, 17.88286590576172, 9.276384353637695, -1.4669952392578125, -1.5823192596435547, 12.435211181640625, 2.1621246337890625, 8.789325714111328, 19.046279907226562, -10.879056930541992, -5.707609176635742, 22.422393798828125, 25.490859985351562, 2.31707763671875, 7.645471572875977, 2.3918380737304688, 10.609466552734375, 25.0474853515625, -3.0925064086914062, 22.3154296875, -17.835676193237305, -16.010498046875, 10.283348083496094, 15.1905517578125, 3.3185081481933594, 1.6212997436523438, -5.065723419189453, 23.063522338867188, -5.44426155090332, 22.8671875, 8.926620483398438, 2.990978240966797, 17.30419158935547, -5.229240417480469, 14.806671142578125, 15.317230224609375, 7.979133605957031, 1.27496337890625, 3.464090347290039, 18.81188201904297, 8.513053894042969, -16.504730224609375, -3.1394805908203125, -4.424896240234375, 30.020050048828125, 17.067413330078125, 24.308820724487305, -10.520719528198242, 7.6908721923828125, 11.602371215820312, 20.099491119384766, -4.364662170410156, -0.02225494384765625, -7.485515594482422, 7.325710296630859, 25.860626220703125, 1.3816909790039062, 13.288097381591797, -0.6721534729003906, 2.00439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000173.npy"}
|
|
{"epoch": 0.2615268329554044, "step": 174, "batch_size": 64, "mean": 6.794708728790283, "std": 9.845768928527832, "min": -18.5531005859375, "p10": -4.0948230743408205, "median": 5.969699859619141, "p90": 20.589328002929694, "max": 29.376113891601562, "pos_frac": 0.75, "sample": [4.7608795166015625, 9.175506591796875, 15.111228942871094, 4.351531982421875, 8.197416305541992, -3.0440521240234375, 3.3458480834960938, 15.090896606445312, 1.1106033325195312, 6.024787902832031, 21.1800537109375, 17.275413513183594, 2.4888916015625, -1.1243858337402344, 13.11773681640625, -3.9119911193847656, 16.530521392822266, 13.14544677734375, 11.025901794433594, -15.775588989257812, 21.301406860351562, 8.336654663085938, -2.7766036987304688, 10.982688903808594, 27.806774139404297, -1.4257698059082031, 7.4425048828125, -2.5943069458007812, 25.559913635253906, 3.8675003051757812, 0.46160888671875, 29.376113891601562, 25.39378547668457, -4.173179626464844, 9.729412078857422, 11.078643798828125, 3.5293731689453125, 9.634033203125, 1.6649551391601562, 19.210968017578125, -4.722984313964844, 18.743896484375, 6.412578582763672, -0.01903533935546875, -0.181488037109375, 14.542098999023438, 7.969871520996094, -6.120643615722656, 21.71588134765625, 4.609687805175781, -18.5531005859375, 13.468460083007812, 1.1981430053710938, 1.09466552734375, 12.62274169921875, -3.562347412109375, 8.585395812988281, -4.769401550292969, 13.028690338134766, -8.181205749511719, 3.291780471801758, 3.8653106689453125, 5.91461181640625, 1.42462158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000174.npy"}
|
|
{"epoch": 0.26303854875283444, "step": 175, "batch_size": 64, "mean": 7.589766502380371, "std": 10.620962142944336, "min": -16.084869384765625, "p10": -4.704790878295898, "median": 6.376428604125977, "p90": 22.254410934448245, "max": 29.573516845703125, "pos_frac": 0.75, "sample": [-8.223100662231445, 12.372316360473633, -2.9270477294921875, 2.220773696899414, -3.3869476318359375, 15.986442565917969, 0.028875350952148438, 14.304443359375, 6.7205352783203125, -1.4708213806152344, 3.0667190551757812, 21.161270141601562, 8.058387756347656, 10.836189270019531, -3.9619140625, 22.66871452331543, 8.648208618164062, 19.421363830566406, 17.03557586669922, 1.28082275390625, 4.380973815917969, 4.534381866455078, 4.3735504150390625, 16.865631103515625, 15.980941772460938, 7.688507080078125, 0.5658645629882812, -2.346710205078125, 24.73923683166504, -16.084869384765625, 20.49742889404297, 0.5927963256835938, 7.599983215332031, 4.740081787109375, 0.5631256103515625, -0.7188358306884766, 7.6866302490234375, 27.16094970703125, -1.8345928192138672, -10.354728698730469, 6.843879699707031, -4.6586456298828125, 1.3328018188476562, 6.032321929931641, -5.201324462890625, -2.9617156982421875, 18.266372680664062, -6.184246063232422, 18.76673126220703, 20.855445861816406, 8.960794448852539, 21.287702560424805, 2.4403152465820312, -7.288154602050781, 2.54022216796875, 18.254432678222656, -4.724567413330078, 25.895225524902344, 13.204681396484375, 29.573516845703125, 1.371002197265625, 23.427078247070312, 12.417121887207031, 24.82290267944336], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000175.npy"}
|
|
{"epoch": 0.26455026455026454, "step": 176, "batch_size": 64, "mean": 4.949705123901367, "std": 10.626727104187012, "min": -19.441085815429688, "p10": -6.287376403808593, "median": 3.968057632446289, "p90": 18.992378807067873, "max": 34.683837890625, "pos_frac": 0.640625, "sample": [25.64398193359375, -15.317543029785156, 10.944175720214844, 13.677997589111328, 0.54937744140625, 3.7115859985351562, -2.6582717895507812, 10.871925354003906, 4.990812301635742, 2.5081787109375, -15.079544067382812, -6.555259704589844, -5.400333404541016, 8.848289489746094, -11.725975036621094, 18.836889266967773, -0.4431571960449219, 14.48394775390625, -3.559417724609375, -1.2011032104492188, -2.4473495483398438, -2.424346923828125, 4.137607574462891, 27.61313247680664, 10.160507202148438, 6.4886627197265625, -5.662315368652344, 11.245048522949219, -0.3981475830078125, 13.942028045654297, 10.36761474609375, -1.4451236724853516, -2.5080490112304688, 14.544151306152344, 9.896385192871094, 24.653507232666016, 0.8621978759765625, 9.2117919921875, 1.1502685546875, -7.3840484619140625, 2.1822738647460938, 3.7985076904296875, 8.754411697387695, 19.059017181396484, 6.174812316894531, 10.616874694824219, -3.1328887939453125, -3.9426498413085938, 12.028549194335938, 21.606002807617188, -1.5006256103515625, 14.326507568359375, 1.9039077758789062, 10.288955688476562, 1.8513336181640625, 6.210254669189453, 34.683837890625, -0.34058380126953125, -1.4979019165039062, 19.764373779296875, 7.076072692871094, -19.441085815429688, 5.549232482910156, -14.368148803710938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000176.npy"}
|
|
{"epoch": 0.2660619803476946, "step": 177, "batch_size": 64, "mean": 9.803640365600586, "std": 10.055026054382324, "min": -17.439453125, "p10": -0.6760223388671873, "median": 8.096147537231445, "p90": 23.95857372283936, "max": 31.674049377441406, "pos_frac": 0.859375, "sample": [4.0940399169921875, 1.2146472930908203, 14.020063400268555, 31.674049377441406, 16.23113250732422, -0.77130126953125, 4.4964752197265625, 9.55255126953125, 27.005294799804688, 9.106040954589844, 30.926612854003906, 21.702789306640625, 19.61476707458496, -0.41024017333984375, 7.658100128173828, -6.931127548217773, 25.384048461914062, 3.5929183959960938, -3.9316844940185547, 11.438953399658203, 9.785247802734375, 8.534194946289062, 3.4567794799804688, 20.353487014770508, 13.041427612304688, 2.3589553833007812, 5.17609977722168, 18.487045288085938, -3.4218883514404297, 3.6463165283203125, 10.155746459960938, 5.268890380859375, 24.633163452148438, 10.3648681640625, 2.1069278717041016, 19.115917205810547, 22.384531021118164, 5.6065826416015625, 6.8779296875, 14.659355163574219, 2.3370361328125, 6.413347244262695, 15.64797592163086, 4.640167236328125, 15.35137939453125, -8.930740356445312, 3.7581329345703125, -1.374755859375, 21.215789794921875, 0.20778465270996094, 20.330810546875, 0.04263877868652344, -17.439453125, 28.75475311279297, 7.593994140625, 12.903480529785156, 9.945354461669922, 25.940399169921875, 13.995330810546875, 7.050144195556641, 5.6204681396484375, 19.10285758972168, -0.453704833984375, 6.520057678222656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000177.npy"}
|
|
{"epoch": 0.2675736961451247, "step": 178, "batch_size": 64, "mean": 5.828727722167969, "std": 9.251792907714844, "min": -14.7769775390625, "p10": -5.428866958618163, "median": 4.690090179443359, "p90": 19.697300720214844, "max": 25.314716339111328, "pos_frac": 0.765625, "sample": [4.397817611694336, 16.274383544921875, 6.341724395751953, -0.0650787353515625, 14.46823501586914, 0.8913478851318359, -1.5609130859375, 10.422611236572266, 14.37420654296875, 8.09286880493164, -5.990596771240234, 3.30157470703125, 4.023101806640625, -11.274015426635742, 6.481315612792969, 5.17620849609375, 21.694120407104492, -0.8047218322753906, -4.1181640625, 0.17889404296875, 22.508159637451172, 2.7430763244628906, 6.673095703125, -13.502685546875, 4.369297027587891, 2.951812744140625, 4.380805969238281, 7.0149993896484375, 4.324394226074219, 3.911489486694336, 12.983154296875, 0.05784797668457031, 4.190773010253906, 25.314716339111328, 4.9748687744140625, 10.351585388183594, 5.1671142578125, 4.05072021484375, 24.063735961914062, 3.3418502807617188, -9.483627319335938, 16.773765563964844, 19.78978729248047, 6.310417175292969, -14.7769775390625, 19.48149871826172, -1.3001556396484375, -7.993766784667969, 20.958816528320312, 5.500537872314453, 16.870986938476562, 20.7891845703125, 7.030712127685547, 13.371955871582031, 4.405311584472656, 1.6036376953125, 12.430404663085938, -1.75177001953125, -12.379508972167969, -2.7769012451171875, 8.896408081054688, 9.564399719238281, 10.975502014160156, -3.427785873413086], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000178.npy"}
|
|
{"epoch": 0.2690854119425548, "step": 179, "batch_size": 64, "mean": 6.869441509246826, "std": 11.100870132446289, "min": -23.659584045410156, "p10": -6.37825527191162, "median": 6.534486770629883, "p90": 19.466738700866706, "max": 34.89295959472656, "pos_frac": 0.71875, "sample": [9.532768249511719, -2.6785888671875, -1.1733856201171875, 9.50469970703125, -0.5719947814941406, 9.630098342895508, 8.786821365356445, 10.448776245117188, 27.383056640625, 11.173606872558594, 17.64761734008789, 20.142778396606445, 15.394046783447266, -13.14382553100586, 7.876611709594727, -7.061759948730469, -23.659584045410156, -6.986749649047852, 3.1391029357910156, 5.776466369628906, -2.376312255859375, 16.70018768310547, 4.6038665771484375, 15.093276977539062, 22.016029357910156, -12.426559448242188, -2.1155242919921875, 16.069618225097656, 4.020652770996094, 5.285730361938477, 7.292507171630859, 15.501869201660156, 3.2813339233398438, -4.95843505859375, 1.8086700439453125, -8.833572387695312, 3.2646026611328125, 33.35194396972656, 17.125625610351562, -0.5437698364257812, 13.125211715698242, 13.020492553710938, 12.864652633666992, 9.866344451904297, -3.8083877563476562, 15.121994018554688, 23.088783264160156, 5.4493408203125, 23.343833923339844, 34.89295959472656, 12.188098907470703, 11.257013320922852, 17.889312744140625, 2.398639678955078, -1.1853523254394531, 12.762863159179688, 4.353340148925781, 0.918121337890625, 5.132219314575195, -2.8739776611328125, -13.847112655639648, 0.010379791259765625, 12.738166809082031, -4.384998321533203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000179.npy"}
|
|
{"epoch": 0.2705971277399849, "step": 180, "batch_size": 64, "mean": 5.987348556518555, "std": 11.041803359985352, "min": -18.73876190185547, "p10": -8.054885101318359, "median": 4.604708671569824, "p90": 23.286320495605473, "max": 29.379005432128906, "pos_frac": 0.765625, "sample": [14.850265502929688, 0.7688217163085938, 7.645267486572266, 3.2940826416015625, 3.8614654541015625, 10.223037719726562, -10.965179443359375, 7.772083282470703, 0.5474300384521484, -2.672700881958008, 21.445756912231445, 10.379325866699219, -6.0162353515625, -0.4649162292480469, 7.048431396484375, 1.0199756622314453, -8.559356689453125, 19.912872314453125, 0.42917633056640625, -16.857681274414062, 24.233779907226562, 7.091209411621094, 6.922693252563477, 12.905014038085938, 9.731441497802734, 14.69266128540039, 0.7632369995117188, 1.8257980346679688, 5.347951889038086, 7.153007507324219, 8.77880859375, 19.597103118896484, -10.350532531738281, 1.417877197265625, 7.611705780029297, 12.478057861328125, 3.734537124633789, 24.866477966308594, -0.0968017578125, -2.091085433959961, 3.561351776123047, 14.361021041870117, 24.161130905151367, 0.06691169738769531, -3.5576820373535156, 2.6582813262939453, 27.74786376953125, -2.1303939819335938, 29.379005432128906, 0.79119873046875, 22.2110595703125, -13.248035430908203, 3.3618087768554688, -6.877784729003906, 3.3128890991210938, 6.445659637451172, 27.718883514404297, -18.73876190185547, 9.598712921142578, -13.868324279785156, 12.738082885742188, 6.7275238037109375, 23.747146606445312, 2.7478866577148438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000180.npy"}
|
|
{"epoch": 0.272108843537415, "step": 181, "batch_size": 64, "mean": 8.793804168701172, "std": 9.283597946166992, "min": -9.479347229003906, "p10": -2.105800437927245, "median": 7.949489593505859, "p90": 20.856153488159187, "max": 31.934768676757812, "pos_frac": 0.84375, "sample": [9.217582702636719, 13.662487030029297, 0.96356201171875, -0.2151203155517578, 4.848171234130859, 2.6539173126220703, 8.317825317382812, -8.156814575195312, 2.6120567321777344, 1.9760093688964844, 15.592933654785156, 4.9579925537109375, 9.45443344116211, -3.6571197509765625, 13.226625442504883, 7.915016174316406, 13.953926086425781, 7.388885498046875, 18.909385681152344, -4.999603271484375, 12.751419067382812, 23.346694946289062, 16.200481414794922, 11.569564819335938, 28.02098846435547, 14.384292602539062, 5.96014404296875, 31.934768676757812, 2.7321929931640625, -9.479347229003906, 17.83282470703125, 4.174022674560547, 25.59295654296875, -6.840753555297852, 15.693267822265625, 1.429046630859375, -0.8591403961181641, 23.983963012695312, 1.7815399169921875, 2.66412353515625, 16.36101531982422, 4.028720855712891, 12.907821655273438, 19.506256103515625, 18.98281478881836, 1.5910663604736328, 1.2100505828857422, 7.781257629394531, 10.523136138916016, 0.5961093902587891, 3.3133544921875, 21.434680938720703, 28.025421142578125, 4.3756103515625, 7.9839630126953125, 18.226268768310547, 3.3061447143554688, 9.184158325195312, 9.853893280029297, -2.6400833129882812, 10.423408508300781, -0.3232555389404297, 17.366371154785156, -2.709930419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000181.npy"}
|
|
{"epoch": 0.273620559334845, "step": 182, "batch_size": 64, "mean": 9.25307846069336, "std": 12.289328575134277, "min": -24.526329040527344, "p10": -5.069400787353516, "median": 7.86555290222168, "p90": 25.42343521118164, "max": 30.569732666015625, "pos_frac": 0.75, "sample": [28.39508056640625, -7.579801559448242, -6.448627471923828, 10.708740234375, 17.4100341796875, 15.657588958740234, 13.160987854003906, -6.374294281005859, -5.173553466796875, 27.846847534179688, 12.883731842041016, 5.4845123291015625, 30.569732666015625, 0.0456390380859375, 21.483139038085938, 13.089508056640625, 8.178077697753906, 6.739967346191406, 24.78752899169922, 7.904914855957031, 12.512578964233398, 0.7290840148925781, 1.606201171875, -2.0693511962890625, 15.530323028564453, -1.4417877197265625, 22.658979415893555, 7.826190948486328, 23.59124755859375, 24.438501358032227, 29.347747802734375, 7.664348602294922, 9.461944580078125, 5.758655548095703, 25.193443298339844, -0.8681182861328125, 1.278839111328125, 3.7569427490234375, 23.545440673828125, -0.3348503112792969, 7.07476806640625, -15.260229110717773, 26.905784606933594, -3.47430419921875, 22.929302215576172, 17.127662658691406, 4.5428314208984375, -2.648712158203125, 20.16022491455078, 4.336395263671875, 15.301700592041016, 25.522003173828125, -4.826377868652344, 22.143999099731445, 27.32049560546875, 4.500005722045898, -3.422210693359375, 8.418615341186523, 17.985126495361328, 2.498933792114258, 3.468841552734375, -11.863327026367188, -0.97430419921875, -24.526329040527344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000182.npy"}
|
|
{"epoch": 0.2751322751322751, "step": 183, "batch_size": 64, "mean": 3.769627571105957, "std": 10.002243995666504, "min": -26.087406158447266, "p10": -8.504594421386718, "median": 1.9445600509643555, "p90": 17.54408969879151, "max": 23.598480224609375, "pos_frac": 0.671875, "sample": [2.8936386108398438, 19.438552856445312, -0.3910064697265625, 0.0672760009765625, 3.7788562774658203, -2.9848785400390625, -9.471908569335938, 21.02294158935547, -9.169239044189453, -1.7311553955078125, -26.087406158447266, -0.4526214599609375, 13.550020217895508, -3.87261962890625, 10.336551666259766, 16.533384323120117, -0.6786308288574219, 9.685401916503906, 14.467727661132812, 10.890905380249023, 11.338859558105469, -3.4788818359375, 14.5667724609375, -6.555776596069336, 1.7609939575195312, -3.7666473388671875, 6.995208740234375, 0.9537734985351562, -13.509033203125, -8.178573608398438, 23.598480224609375, 8.515518188476562, 1.7339935302734375, 1.1030960083007812, -9.160102844238281, 10.239784240722656, 8.66537857055664, 21.05187225341797, -17.896865844726562, 21.183582305908203, -7.21624755859375, 0.5865020751953125, 20.150253295898438, -0.8188056945800781, 1.6560955047607422, 6.606695175170898, 12.831727981567383, 2.5815601348876953, -7.553031921386719, -0.7359485626220703, 4.002250671386719, 6.599067687988281, 0.7157249450683594, 1.5032424926757812, 15.816436767578125, 0.15897178649902344, 8.787849426269531, -8.644317626953125, 2.1281261444091797, 0.688568115234375, 5.79595947265625, 17.977249145507812, 12.097602844238281, 8.55340576171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000183.npy"}
|
|
{"epoch": 0.2766439909297052, "step": 184, "batch_size": 64, "mean": 6.898704528808594, "std": 9.136359214782715, "min": -7.362094879150391, "p10": -2.2677307128906246, "median": 4.5305328369140625, "p90": 20.843729782104493, "max": 26.32855224609375, "pos_frac": 0.734375, "sample": [1.6899566650390625, 17.701370239257812, 20.951946258544922, 1.2685050964355469, 16.817420959472656, 3.56024169921875, -0.7017097473144531, -0.9772415161132812, -1.8541831970214844, 3.971099853515625, 0.2819252014160156, 23.210906982421875, -0.25527000427246094, 5.278675079345703, 16.011592864990234, -6.0386199951171875, 18.458999633789062, 12.098564147949219, 6.266185760498047, 26.32855224609375, -5.142169952392578, 26.27083396911621, 11.312152862548828, 26.286659240722656, -6.269344329833984, 18.95117950439453, 2.8352108001708984, 10.366897583007812, 22.35308074951172, 2.004375457763672, 10.425559997558594, -0.43912506103515625, 4.943229675292969, -7.362094879150391, 0.016363143920898438, 6.1541748046875, 24.606842041015625, 7.033357620239258, 1.7752571105957031, 17.877830505371094, 7.789861679077148, -0.31169891357421875, 14.655838012695312, -0.20905113220214844, 13.727027893066406, 6.431144714355469, 7.8800506591796875, 5.220207214355469, 0.6301364898681641, 20.591224670410156, -6.196664810180664, -1.2441024780273438, 4.141029357910156, 4.4942626953125, 12.441131591796875, 12.45574951171875, -4.624994277954102, 4.566802978515625, -0.8968753814697266, 4.274131774902344, -2.444965362548828, 1.679159164428711, -1.7958202362060547, 0.1943206787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000184.npy"}
|
|
{"epoch": 0.2781557067271353, "step": 185, "batch_size": 64, "mean": 7.422008514404297, "std": 12.94803524017334, "min": -19.82798957824707, "p10": -7.511771774291992, "median": 7.284778594970703, "p90": 23.363242149353027, "max": 32.5455322265625, "pos_frac": 0.671875, "sample": [4.4932098388671875, 29.55408477783203, -2.4735794067382812, 15.471254348754883, 29.526229858398438, -17.88565444946289, 10.081180572509766, 6.347263336181641, -5.870758056640625, 1.956817626953125, 27.30632781982422, 27.97272491455078, -0.0055999755859375, -6.868072509765625, 29.296737670898438, 4.505544662475586, 32.5455322265625, 5.653663635253906, 21.056171417236328, 22.558202743530273, -4.800071716308594, 23.377052307128906, 12.908626556396484, 6.4562835693359375, 1.9509944915771484, 17.32302474975586, 9.687532424926758, 11.219226837158203, 6.359169006347656, 15.55908203125, -7.276594161987305, -0.8877639770507812, 12.1181640625, 19.229461669921875, -19.82798957824707, 16.458778381347656, 17.351974487304688, 12.522775650024414, 10.48397445678711, -2.646383285522461, -17.795433044433594, 8.113273620605469, -4.613994598388672, 10.98126220703125, 18.473739624023438, -14.781326293945312, 20.729219436645508, -2.3807239532470703, 21.493804931640625, -3.604827880859375, 4.6480255126953125, -11.660177230834961, 9.765274047851562, -5.053211212158203, 0.9390640258789062, 9.20716667175293, -13.02254867553711, -7.61256217956543, 13.423316955566406, -1.739593505859375, -3.5286827087402344, 4.356021881103516, 22.551851272583008, 23.331018447875977], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000185.npy"}
|
|
{"epoch": 0.2796674225245654, "step": 186, "batch_size": 64, "mean": 8.341560363769531, "std": 12.806977272033691, "min": -23.66332244873047, "p10": -5.45612564086914, "median": 9.009025573730469, "p90": 25.408754158020024, "max": 35.30859375, "pos_frac": 0.71875, "sample": [17.100845336914062, 30.761520385742188, -14.938980102539062, 9.132003784179688, 13.100700378417969, -4.093475341796875, 9.108070373535156, -11.262210845947266, -5.61663818359375, 18.02727508544922, 11.492286682128906, -3.813396453857422, 12.146713256835938, 18.996196746826172, 35.06304931640625, 20.908573150634766, 5.845008850097656, 7.525794982910156, 4.906658172607422, -18.096542358398438, 3.821500778198242, 14.10777473449707, 18.766576766967773, 30.080089569091797, -23.66332244873047, -5.079948425292969, -5.081596374511719, -4.381931304931641, -1.677581787109375, 1.8928413391113281, 5.780521392822266, 21.38715934753418, 2.353790283203125, -2.60723876953125, -3.6688766479492188, -1.4338722229003906, 9.794349670410156, 24.0987548828125, 20.51464080810547, 3.6708641052246094, -2.9564895629882812, -8.677482604980469, 3.3018531799316406, 4.058341979980469, 11.6201171875, 7.052059173583984, 25.970182418823242, 15.57208251953125, -7.1102294921875, 9.111724853515625, 32.84883117675781, 9.427103042602539, 8.909980773925781, 35.30859375, 9.870796203613281, 9.671173095703125, 32.849639892578125, 18.43804359436035, 15.701690673828125, 12.377653121948242, 6.552093505859375, 13.8638916015625, -2.4457855224609375, 7.57611083984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000186.npy"}
|
|
{"epoch": 0.2811791383219955, "step": 187, "batch_size": 64, "mean": 7.580622673034668, "std": 12.523482322692871, "min": -22.697473526000977, "p10": -8.127715682983398, "median": 7.824917793273926, "p90": 24.290243911743165, "max": 34.06463623046875, "pos_frac": 0.71875, "sample": [10.348304748535156, -4.233673095703125, 20.260513305664062, 29.03290557861328, 34.06463623046875, 14.968788146972656, 11.610374450683594, 4.9484405517578125, 6.982330322265625, 22.306289672851562, -18.072036743164062, 10.10760498046875, -0.7256698608398438, 10.589248657226562, 11.919967651367188, 1.7182483673095703, 1.8228740692138672, 21.52942657470703, 14.589225769042969, 7.991395950317383, -11.722076416015625, -22.697473526000977, 23.890464782714844, -0.5153732299804688, 3.9854812622070312, -2.0707015991210938, 1.1310501098632812, 4.2008209228515625, 2.4916839599609375, 21.216228485107422, 19.30200958251953, 24.5625, -11.880996704101562, -1.5536537170410156, 10.051063537597656, 8.458259582519531, 5.0420684814453125, 0.12572479248046875, -5.586204528808594, 14.005378723144531, -2.425783157348633, -8.196605682373047, 21.19900131225586, 0.029947280883789062, 7.658439636230469, 12.750282287597656, 25.44080352783203, -4.808773040771484, 9.926681518554688, 17.313858032226562, 23.93004608154297, -7.966972351074219, 15.755538940429688, 30.407958984375, 3.3946609497070312, -4.378990173339844, 27.566818237304688, 24.44461441040039, 2.69659423828125, 8.922325134277344, -2.0374069213867188, -12.407472610473633, -11.312591552734375, 13.061431884765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000187.npy"}
|
|
{"epoch": 0.28269085411942557, "step": 188, "batch_size": 64, "mean": 8.116436004638672, "std": 12.007192611694336, "min": -26.239097595214844, "p10": -4.0358837127685545, "median": 5.783147811889648, "p90": 25.78876914978028, "max": 32.79755401611328, "pos_frac": 0.78125, "sample": [4.815177917480469, -2.9144725799560547, 28.91207504272461, -6.854272842407227, -0.2323455810546875, 21.186386108398438, 18.61431884765625, 5.485376358032227, 19.388763427734375, 5.820526123046875, 3.2666664123535156, 32.02745056152344, 22.281612396240234, -0.40178680419921875, 0.6180744171142578, 1.2394523620605469, 13.976531982421875, 27.167449951171875, 0.8428153991699219, -4.282779693603516, 8.85932731628418, 1.8162841796875, 2.1605453491210938, 0.2869110107421875, 8.875213623046875, -0.6517791748046875, -5.076360702514648, 19.073104858398438, 6.468006134033203, 18.97878646850586, 13.422796249389648, 2.1348876953125, 7.71551513671875, 32.79755401611328, 20.838096618652344, 0.8397083282470703, 24.307418823242188, 5.6468353271484375, 0.8500289916992188, 7.033226013183594, 30.713886260986328, 1.8629150390625, 5.745769500732422, 9.778610229492188, 0.1842498779296875, -2.2194976806640625, 4.344490051269531, 23.683509826660156, 2.7983474731445312, -0.10771369934082031, 27.572715759277344, -15.27191162109375, -3.4597930908203125, 8.084392547607422, 7.172637939453125, 6.0549774169921875, 7.819425582885742, -9.668600082397461, 21.277488708496094, 26.423633575439453, 23.138330459594727, 6.754783630371094, -4.324743270874023, -26.239097595214844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000188.npy"}
|
|
{"epoch": 0.2842025699168556, "step": 189, "batch_size": 64, "mean": 7.767352104187012, "std": 11.796666145324707, "min": -20.532817840576172, "p10": -8.461125946044918, "median": 7.056486129760742, "p90": 25.156742858886723, "max": 32.227622985839844, "pos_frac": 0.796875, "sample": [1.9288864135742188, 9.734386444091797, 8.155498504638672, 11.608383178710938, 11.3719482421875, 25.497650146484375, 24.439781188964844, 12.627973556518555, 0.9436435699462891, 7.5009002685546875, -15.925125122070312, 24.06214141845703, -13.3828125, 2.6923675537109375, -0.6717758178710938, -1.628122329711914, 26.754798889160156, 17.486000061035156, -12.028865814208984, 10.29996109008789, 6.030254364013672, 16.41103744506836, 5.003181457519531, 8.246047973632812, -0.29854583740234375, 2.8723411560058594, 15.213462829589844, 10.06845474243164, 4.429866790771484, -0.15668487548828125, 26.41423797607422, -3.729095458984375, 3.439453125, 27.47185516357422, 25.464012145996094, 7.299652099609375, 0.8818321228027344, 2.6746063232421875, 3.2828407287597656, 14.655685424804688, 24.043079376220703, 8.498504638671875, 11.640844345092773, 5.902759552001953, 13.309745788574219, 7.585117340087891, -4.571044921875, -15.468193054199219, 6.25830078125, 3.3124313354492188, 3.1878280639648438, 19.10111427307129, -13.661178588867188, 32.227622985839844, 18.0860595703125, 4.8655548095703125, -10.128303527832031, -20.532817840576172, 9.223512649536133, 5.497314453125, 6.813320159912109, 2.015045166015625, 28.45775604248047, 24.30406379699707], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000189.npy"}
|
|
{"epoch": 0.2857142857142857, "step": 190, "batch_size": 64, "mean": 10.017097473144531, "std": 11.790956497192383, "min": -13.154769897460938, "p10": -4.382789421081542, "median": 8.445439338684082, "p90": 27.48520202636719, "max": 37.324676513671875, "pos_frac": 0.796875, "sample": [6.348918914794922, 17.748016357421875, 34.63481140136719, -6.1439361572265625, 13.597652435302734, 16.209945678710938, 0.39632415771484375, 2.4404449462890625, -8.843233108520508, 29.386688232421875, 6.378120422363281, 16.320022583007812, 8.22296142578125, 0.6829910278320312, 6.3841400146484375, 1.4598159790039062, -2.8403053283691406, 27.136932373046875, 5.6402740478515625, 9.320335388183594, 20.43082046508789, 29.65311050415039, 1.507965087890625, 2.6662864685058594, -5.091741561889648, 9.352973937988281, -13.154769897460938, 4.966876983642578, 5.5963897705078125, 12.925556182861328, 12.041641235351562, -1.6397438049316406, 7.471435546875, 10.8868408203125, 18.37065887451172, 1.4647369384765625, 27.63446044921875, -4.693918228149414, -2.4315719604492188, 11.674285888671875, -5.0130462646484375, 18.098968505859375, 19.761978149414062, 37.324676513671875, 5.64671516418457, 2.892486572265625, -2.863515853881836, 16.845352172851562, -10.603302001953125, -3.6568222045898438, 8.667917251586914, 10.143257141113281, 3.4168777465820312, 20.22229766845703, 14.960891723632812, 27.996238708496094, 32.87409973144531, 4.757621765136719, 21.55453872680664, 26.106239318847656, -0.6241569519042969, 25.47088623046875, 17.3070068359375, 15.693775177001953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000190.npy"}
|
|
{"epoch": 0.2872260015117158, "step": 191, "batch_size": 64, "mean": 7.888982772827148, "std": 12.497056007385254, "min": -23.527267456054688, "p10": -3.8195678710937497, "median": 7.663163185119629, "p90": 22.683739471435548, "max": 34.18968963623047, "pos_frac": 0.765625, "sample": [22.22113037109375, -2.5489654541015625, 7.4794158935546875, 15.248332977294922, 32.63723373413086, -0.17890167236328125, 4.303863525390625, 26.823806762695312, 7.524312973022461, 3.7954025268554688, 3.7008514404296875, 34.18968963623047, 5.2469024658203125, -0.0355072021484375, 9.376014709472656, 2.821186065673828, -23.339553833007812, 4.7278289794921875, 26.720184326171875, 10.987737655639648, 10.568313598632812, 20.176666259765625, 12.589859008789062, 22.874794006347656, -13.608024597167969, 21.87773895263672, 10.037155151367188, 11.574981689453125, -3.482940673828125, 0.3265972137451172, 10.623872756958008, -0.64996337890625, -1.2611026763916016, 1.8534507751464844, 29.289474487304688, -23.527267456054688, 0.3341941833496094, -11.285858154296875, 10.006214141845703, 18.92877960205078, 12.37655258178711, 10.823783874511719, 7.802013397216797, 13.464035034179688, 15.828073501586914, 2.4593753814697266, 21.794700622558594, 26.72262954711914, -3.963836669921875, 2.392822265625, -3.208883285522461, 6.669797897338867, 22.237945556640625, 8.145027160644531, -9.447349548339844, 12.220428466796875, 21.49212646484375, -21.510574340820312, 11.978776931762695, -0.10569000244140625, 1.3749465942382812, 21.534942626953125, 1.6671371459960938, 3.1982345581054688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000191.npy"}
|
|
{"epoch": 0.2887377173091459, "step": 192, "batch_size": 64, "mean": 5.161694526672363, "std": 10.434894561767578, "min": -27.811904907226562, "p10": -4.658485412597655, "median": 3.7678871154785156, "p90": 18.95935115814209, "max": 31.226905822753906, "pos_frac": 0.65625, "sample": [-27.811904907226562, -9.976249694824219, 1.6767997741699219, 11.489551544189453, 15.496273040771484, 2.2923851013183594, -4.187873840332031, 10.994400024414062, 27.0555419921875, -4.860176086425781, -7.7274932861328125, -4.010505676269531, -2.5349502563476562, 0.7127151489257812, 1.9727249145507812, 9.7337646484375, 18.987628936767578, 6.1590423583984375, -0.2753028869628906, 0.29340362548828125, 18.893369674682617, 20.229644775390625, -3.263368606567383, -1.9272003173828125, 23.576141357421875, -0.02930450439453125, 2.820220947265625, 14.609466552734375, 4.012794494628906, 3.522979736328125, 10.42236328125, 13.849441528320312, 10.701791763305664, 14.830080032348633, -2.157938003540039, -3.390972137451172, 4.183189392089844, -0.3087272644042969, -2.5733909606933594, 4.900047302246094, 5.094209671020508, -0.9490776062011719, -16.849510192871094, 31.226905822753906, 13.428131103515625, 6.577749252319336, -3.52789306640625, 2.998147964477539, 5.8047027587890625, -11.540557861328125, 8.32061767578125, 21.580459594726562, 2.002361297607422, 13.760490417480469, -0.8050594329833984, 13.027717590332031, 12.632831573486328, -1.5398788452148438, 15.537109375, 21.422340393066406, 11.481569290161133, -5.711917877197266, 2.277454376220703, 5.719144821166992], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000192.npy"}
|
|
{"epoch": 0.29024943310657597, "step": 193, "batch_size": 64, "mean": 9.353704452514648, "std": 11.57327651977539, "min": -22.159645080566406, "p10": -2.0671407699584954, "median": 6.055502891540527, "p90": 25.068632888793946, "max": 38.218841552734375, "pos_frac": 0.78125, "sample": [1.0973625183105469, 5.264936447143555, 17.41057586669922, 5.870353698730469, 6.240652084350586, -6.852226257324219, -0.5469284057617188, 24.32018280029297, -1.2499847412109375, -0.9170665740966797, 22.367557525634766, 2.0375194549560547, 14.861343383789062, 10.88116455078125, 21.103050231933594, -0.06060791015625, 11.793182373046875, -0.5896873474121094, 14.30146598815918, 19.002349853515625, -2.313173294067383, 26.167503356933594, 38.218841552734375, -4.401050567626953, 14.488006591796875, 24.5369873046875, 2.1147098541259766, 5.121833801269531, -1.4930648803710938, 10.071115493774414, 24.987266540527344, 12.782379150390625, 4.574394226074219, 11.221176147460938, 10.358612060546875, 4.086250305175781, -2.3168869018554688, 2.401947021484375, 11.682350158691406, 9.177650451660156, -8.314468383789062, 5.086433410644531, 22.52042007446289, 32.38404083251953, -2.718353271484375, 3.8069915771484375, 9.632484436035156, -1.3379497528076172, 5.405216217041016, 27.873077392578125, 18.678085327148438, 0.48999786376953125, 29.771438598632812, 31.63785743713379, 25.103504180908203, 4.187675476074219, 11.209182739257812, -22.159645080566406, 7.607765197753906, 24.10053062438965, 2.165454864501953, 2.208538055419922, 1.1859130859375, 0.31084251403808594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000193.npy"}
|
|
{"epoch": 0.29176114890400606, "step": 194, "batch_size": 64, "mean": 7.695527076721191, "std": 13.4341459274292, "min": -24.36819839477539, "p10": -9.173716354370114, "median": 8.043633460998535, "p90": 25.59816856384278, "max": 33.32362365722656, "pos_frac": 0.75, "sample": [29.735801696777344, 6.4528961181640625, 14.935066223144531, -10.191211700439453, 21.39550018310547, 3.4343032836914062, -16.714574813842773, 2.8054847717285156, 1.7890090942382812, 8.7738037109375, 7.186588287353516, 13.166915893554688, 4.988006591796875, 20.3963623046875, -13.526802062988281, -5.747245788574219, 15.358100891113281, -6.799560546875, 9.634906768798828, -2.4833908081054688, 16.186553955078125, 3.315906524658203, 17.0050048828125, 21.325481414794922, -4.249156951904297, 11.404495239257812, -4.7339935302734375, 16.160736083984375, 0.08005523681640625, 11.177589416503906, -0.598846435546875, 24.12624168395996, -12.723609924316406, -6.1020660400390625, -19.997669219970703, -0.8757095336914062, 33.32362365722656, 14.17767333984375, 13.14022445678711, 28.37015724182129, -24.36819839477539, 9.459732055664062, 5.319307327270508, 1.9703254699707031, 16.047515869140625, 32.45343017578125, 10.752645492553711, -23.19873809814453, 31.491432189941406, 17.724578857421875, 14.394489288330078, 0.135009765625, 14.152702331542969, 26.228994369506836, 23.506698608398438, 29.19641876220703, 9.95439338684082, 0.7018318176269531, 2.2751235961914062, 16.139739990234375, 4.890684127807617, 3.045482635498047, -2.17596435546875, 7.31346321105957], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000194.npy"}
|
|
{"epoch": 0.29327286470143615, "step": 195, "batch_size": 64, "mean": 9.208534240722656, "std": 12.571380615234375, "min": -18.30438232421875, "p10": -7.5155609130859355, "median": 8.992263793945312, "p90": 26.615843200683596, "max": 34.56788635253906, "pos_frac": 0.6875, "sample": [4.646537780761719, 8.84182357788086, -1.3885650634765625, 8.657073974609375, 32.77613067626953, 19.292709350585938, -0.5598220825195312, 10.305526733398438, 11.40277099609375, 26.670303344726562, 16.282196044921875, 19.486520767211914, -9.060661315917969, -9.502670288085938, -8.501968383789062, -18.30438232421875, -1.1635894775390625, 9.098892211914062, 9.799257278442383, 23.462038040161133, -5.2139434814453125, 7.603202819824219, 11.81026840209961, 14.990570068359375, 9.104118347167969, 10.38241958618164, 29.21125602722168, 30.109970092773438, 5.138862609863281, 14.542823791503906, 4.712446212768555, -1.4370193481445312, 30.536067962646484, 5.823204040527344, 26.48876953125, -0.43468475341796875, -2.7032527923583984, 22.90540313720703, 18.39496612548828, 19.720306396484375, 8.99929428100586, -4.62200927734375, -2.4168243408203125, -0.4060821533203125, 0.6771869659423828, 3.207609176635742, -0.04144287109375, 25.78136444091797, -16.901260375976562, -0.26910972595214844, -9.133567810058594, -0.0070953369140625, 34.56788635253906, 21.548019409179688, 17.773574829101562, 16.502798080444336, 21.544708251953125, 19.497657775878906, -9.022186279296875, 27.521774291992188, 8.985233306884766, 0.7946739196777344, 5.7339935302734375, 15.104108810424805], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000195.npy"}
|
|
{"epoch": 0.2947845804988662, "step": 196, "batch_size": 64, "mean": 7.773179054260254, "std": 11.342694282531738, "min": -32.52987289428711, "p10": -3.669218063354492, "median": 7.285453796386719, "p90": 22.67310333251953, "max": 31.094146728515625, "pos_frac": 0.78125, "sample": [5.0463409423828125, -32.52987289428711, -3.468425750732422, 3.91046142578125, 7.821722030639648, -7.302345275878906, 2.81427001953125, 4.5798492431640625, 7.2977752685546875, -3.1418190002441406, 6.037322998046875, 22.793861389160156, 24.13916015625, 9.567497253417969, 8.730567932128906, 5.976654052734375, -17.67068099975586, 20.379783630371094, -1.8403854370117188, 19.663719177246094, -1.9071292877197266, 4.447322845458984, 26.821949005126953, 0.5619792938232422, -5.107471466064453, 4.459503173828125, 11.193450927734375, 8.994316101074219, 8.716232299804688, 31.094146728515625, 8.54317855834961, 16.694568634033203, 2.2085723876953125, -3.7552719116210938, -3.3442306518554688, 10.2281494140625, 2.985200881958008, 12.274232864379883, 8.733863830566406, 27.789962768554688, 11.139503479003906, 7.0420379638671875, 19.80117416381836, 3.1632957458496094, 18.185264587402344, -6.998481750488281, 0.5465297698974609, 7.27313232421875, 10.965232849121094, 24.856460571289062, 22.391334533691406, 21.543834686279297, -1.8057479858398438, -1.0349884033203125, 8.808895111083984, 22.2181396484375, 15.538381576538086, 15.920204162597656, 2.8785057067871094, 2.4864044189453125, 3.2722396850585938, -5.080478668212891, 11.025611877441406, 28.908950805664062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000196.npy"}
|
|
{"epoch": 0.2962962962962963, "step": 197, "batch_size": 64, "mean": 9.469017028808594, "std": 15.16988754272461, "min": -29.13399887084961, "p10": -11.388403892517088, "median": 7.765323638916016, "p90": 28.203494262695312, "max": 36.67913055419922, "pos_frac": 0.75, "sample": [-3.7317276000976562, 33.617408752441406, 6.779155731201172, 1.5075225830078125, 10.243255615234375, 16.480201721191406, -12.021888732910156, -25.845138549804688, 1.8486480712890625, -2.0057449340820312, 23.07394027709961, 27.23024559020996, -12.475231170654297, -15.13613510131836, 7.416996002197266, 29.983612060546875, 10.844367980957031, 3.1730995178222656, 36.67913055419922, 29.018997192382812, 5.7085113525390625, 7.415008544921875, 21.414813995361328, 15.453971862792969, 10.799365997314453, 2.1483993530273438, 4.184043884277344, -9.910272598266602, 26.439035415649414, 13.831878662109375, 27.499412536621094, 28.207000732421875, -18.900421142578125, 23.743423461914062, -3.4518775939941406, -29.13399887084961, 4.2104644775390625, 8.811676025390625, 23.293914794921875, 13.488632202148438, -0.0928802490234375, 25.511489868164062, 28.1953125, 2.236419677734375, -1.7882537841796875, 31.133621215820312, 4.75152587890625, -0.09549713134765625, 6.785396575927734, 22.575105667114258, -8.170377731323242, -3.5016326904296875, 18.650070190429688, 19.272342681884766, 3.3429384231567383, -16.642181396484375, 34.53907012939453, 18.195907592773438, 1.9823932647705078, 19.137523651123047, 5.108802795410156, 20.90948486328125, 8.113651275634766, 23.933198928833008], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000197.npy"}
|
|
{"epoch": 0.29780801209372637, "step": 198, "batch_size": 64, "mean": 10.072783470153809, "std": 11.70544147491455, "min": -17.596206665039062, "p10": -5.654905891418455, "median": 10.368789672851562, "p90": 24.246012115478518, "max": 39.23320770263672, "pos_frac": 0.8125, "sample": [-2.36456298828125, 20.132972717285156, 0.7651462554931641, 13.102615356445312, 8.587390899658203, 11.177543640136719, -9.806634902954102, 3.7252883911132812, 8.055173873901367, 5.7490234375, 22.805259704589844, 8.306808471679688, 0.8880119323730469, -13.308374404907227, 39.23320770263672, 29.55327606201172, 14.42364501953125, 12.293670654296875, 14.000595092773438, -3.771352767944336, 17.728363037109375, 17.63096046447754, 25.702913284301758, -6.814985275268555, 27.47113037109375, -13.277938842773438, -6.8708343505859375, -0.153900146484375, 20.572265625, 17.4135799407959, 19.884368896484375, 15.195526123046875, 10.754196166992188, -17.596206665039062, 7.021354675292969, -1.9139213562011719, 1.2871665954589844, 3.077983856201172, -6.4621429443359375, 10.241310119628906, 5.635520935058594, 24.411033630371094, 23.8609619140625, 10.526222229003906, 14.235647201538086, 9.561492919921875, 0.7561454772949219, 28.850099563598633, 36.263328552246094, 9.540977478027344, 5.4918975830078125, 1.2120742797851562, 17.904075622558594, 23.471546173095703, 7.527381896972656, 10.496269226074219, 11.049362182617188, 11.637990951538086, -0.7489662170410156, 9.312328338623047, 12.27481460571289, 9.422801971435547, 22.327163696289062, 15.198074340820312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000198.npy"}
|
|
{"epoch": 0.29931972789115646, "step": 199, "batch_size": 64, "mean": 7.822010040283203, "std": 14.300711631774902, "min": -26.677852630615234, "p10": -10.319304466247559, "median": 9.700084686279297, "p90": 24.88787422180176, "max": 36.29959487915039, "pos_frac": 0.6875, "sample": [16.36855697631836, 20.01451873779297, 3.2342681884765625, 8.912654876708984, -10.152551651000977, -17.346790313720703, 9.591514587402344, 26.90125274658203, 14.012039184570312, 9.80865478515625, 14.488250732421875, -3.5060195922851562, 36.29959487915039, 19.39776611328125, 33.82558822631836, 25.557388305664062, 17.91434097290039, 22.54412841796875, 15.941390991210938, 24.948787689208984, -4.2165069580078125, -16.003894805908203, 17.59634780883789, 22.235809326171875, -4.050630569458008, -0.33272552490234375, 12.903976440429688, -5.2478485107421875, 12.174602508544922, 16.590782165527344, 14.559104919433594, 3.8645992279052734, 15.97439956665039, -2.794097900390625, -1.0253667831420898, 7.3282012939453125, 7.800689697265625, -10.390769958496094, -15.747856140136719, 12.768936157226562, -26.677852630615234, 16.384307861328125, 3.597423553466797, 22.019493103027344, -20.872901916503906, 10.561286926269531, 9.443660736083984, 5.776493072509766, -6.300483703613281, 3.1935195922851562, 0.3718223571777344, -23.295089721679688, -6.4791412353515625, 17.103656768798828, -9.117767333984375, -1.657989501953125, 24.745742797851562, -4.604759216308594, 22.041580200195312, 13.474624633789062, 29.735137939453125, 4.884063720703125, 30.726829528808594, 12.811893463134766], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000199.npy"}
|
|
{"epoch": 0.30083144368858655, "step": 200, "batch_size": 64, "mean": 11.200349807739258, "std": 15.268163681030273, "min": -31.901870727539062, "p10": -6.465955066680908, "median": 8.739452362060547, "p90": 34.527139282226564, "max": 37.22908020019531, "pos_frac": 0.734375, "sample": [5.9538116455078125, 22.167224884033203, 35.58978271484375, 4.731536865234375, 8.732391357421875, 2.1946983337402344, 29.158935546875, 6.371406555175781, 7.704429626464844, -1.8512344360351562, 2.034698486328125, 21.88250732421875, -10.07442855834961, 35.03144836425781, 2.069488525390625, -0.9829864501953125, 7.6933441162109375, -1.3439407348632812, 33.35041809082031, -6.62890625, -1.473114013671875, -0.192138671875, 18.171585083007812, 27.59006118774414, 2.008533477783203, 17.00518035888672, 24.65639305114746, -6.6249542236328125, 22.546695709228516, -1.9476509094238281, 28.700965881347656, 17.963241577148438, 12.28338623046875, 35.60980224609375, 35.93775939941406, -2.522794723510742, -31.901870727539062, 20.90776824951172, 35.46022033691406, 27.756309509277344, 28.47222137451172, 14.667144775390625, 37.22908020019531, 9.619598388671875, 9.66082763671875, -20.163848876953125, 3.0506439208984375, 8.677154541015625, 7.787275314331055, 36.24763488769531, -3.044219970703125, 11.869125366210938, 10.759481430053711, -6.6244354248046875, 8.746513366699219, 12.761505126953125, -5.89497184753418, -7.156024932861328, 12.652767181396484, 7.012966156005859, 0.7277488708496094, -6.09616756439209, 27.537826538085938, 32.60249328613281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000200.npy"}
|
|
{"epoch": 0.30234315948601664, "step": 201, "batch_size": 64, "mean": 8.794891357421875, "std": 14.263848304748535, "min": -22.29247283935547, "p10": -11.080042839050291, "median": 6.8618574142456055, "p90": 29.798571014404306, "max": 41.03565216064453, "pos_frac": 0.78125, "sample": [5.1917724609375, 16.01971435546875, 6.303619384765625, 8.122451782226562, 2.6879100799560547, 8.578449249267578, 24.608341217041016, 4.93647575378418, 12.881317138671875, 0.84442138671875, 4.017845153808594, 18.75987434387207, 20.16567039489746, 17.977081298828125, 1.0266590118408203, -14.6466064453125, 41.03565216064453, 27.833351135253906, 0.290618896484375, 35.99333953857422, -11.666664123535156, 2.5077743530273438, 5.6734771728515625, -12.127765655517578, 2.3521671295166016, 31.561893463134766, 15.331138610839844, 12.371391296386719, -9.711259841918945, 25.361473083496094, 1.8331680297851562, 30.64080810546875, 2.2757701873779297, -4.6122283935546875, 21.449901580810547, 21.238941192626953, -0.948333740234375, -13.85736083984375, 0.687286376953125, 5.7421875, 21.796722412109375, 8.74298095703125, -2.6553306579589844, 17.97894287109375, -4.917600631713867, 16.299423217773438, 8.197296142578125, -16.999664306640625, 7.420095443725586, 33.72705078125, -2.747446060180664, 8.434738159179688, 12.953327178955078, 22.73709487915039, 14.05306625366211, 34.479454040527344, 20.220706939697266, -7.49554443359375, 1.4115219116210938, -12.002716064453125, 33.20073318481445, 0.4644012451171875, 1.134552001953125, -22.29247283935547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000201.npy"}
|
|
{"epoch": 0.30385487528344673, "step": 202, "batch_size": 64, "mean": 10.093530654907227, "std": 13.857715606689453, "min": -29.961395263671875, "p10": -4.939467620849609, "median": 11.62735652923584, "p90": 24.810955619812017, "max": 45.480987548828125, "pos_frac": 0.796875, "sample": [-4.6597442626953125, 7.4306793212890625, 37.98320388793945, 45.480987548828125, 23.5989933013916, -2.9740524291992188, 17.180442810058594, 16.815689086914062, 19.751129150390625, 13.231842041015625, 3.5451412200927734, 15.18698501586914, 3.4497947692871094, 6.451290130615234, -3.7294921875, 18.27259063720703, 21.80328369140625, 2.1231155395507812, 16.25640869140625, -6.737056732177734, 7.5717620849609375, 10.485572814941406, 20.178070068359375, 3.381195068359375, -5.059349060058594, 15.690628051757812, 4.907947540283203, 38.89763259887695, 0.0956268310546875, 19.507463455200195, -2.10137939453125, 9.90872573852539, 13.212066650390625, 16.404953002929688, 5.3974456787109375, -29.961395263671875, 15.390068054199219, 12.180627822875977, 17.963226318359375, -17.58831024169922, 14.515167236328125, 13.7342529296875, 15.798858642578125, 7.5243377685546875, 25.330368041992188, 8.377975463867188, 22.692550659179688, 5.004655838012695, 27.617813110351562, -3.5732040405273438, 0.2475414276123047, -20.789684295654297, -0.17882156372070312, -6.531803131103516, 4.050994873046875, 18.702957153320312, 30.896682739257812, 3.3027820587158203, 11.074085235595703, 18.259145736694336, -19.978717803955078, 23.342201232910156, 13.170644760131836, 26.471378326416016], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000202.npy"}
|
|
{"epoch": 0.30536659108087677, "step": 203, "batch_size": 64, "mean": 9.159772872924805, "std": 14.037582397460938, "min": -21.411712646484375, "p10": -6.951947784423828, "median": 7.065826416015625, "p90": 27.60349731445313, "max": 43.6104736328125, "pos_frac": 0.75, "sample": [8.822097778320312, 2.831684112548828, -18.123214721679688, -2.9677734375, 4.9477081298828125, 15.076770782470703, 1.2113037109375, 20.8045654296875, 5.878791809082031, -6.8961334228515625, 22.60595703125, 9.944496154785156, 11.889532089233398, 35.71635437011719, 3.923147201538086, 43.6104736328125, 25.276351928710938, 5.2975311279296875, 24.866596221923828, 10.010574340820312, -21.411712646484375, -1.138031005859375, 12.772392272949219, -3.9699172973632812, -0.5054607391357422, -10.446640014648438, 5.5016021728515625, 34.01588439941406, 4.536170959472656, 20.860145568847656, 8.252861022949219, 3.7696456909179688, 28.52178955078125, 8.48873519897461, 16.935165405273438, 1.3392486572265625, 27.893051147460938, 17.119335174560547, 26.927871704101562, 9.939393997192383, 4.5509796142578125, -2.530303955078125, 4.243232727050781, 11.087554931640625, 20.619956970214844, -3.0474853515625, 28.724319458007812, 26.784896850585938, -6.975868225097656, 39.802337646484375, 15.812610626220703, 1.1056175231933594, 1.5155715942382812, 5.438629150390625, 9.924087524414062, -17.42228126525879, -0.7427215576171875, 5.821937561035156, 23.59510040283203, 17.304025650024414, 14.185760498046875, -13.887557983398438, -13.003032684326172, -0.8102207183837891], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000203.npy"}
|
|
{"epoch": 0.30687830687830686, "step": 204, "batch_size": 64, "mean": 9.871435165405273, "std": 14.24223804473877, "min": -26.592742919921875, "p10": -5.383401870727538, "median": 8.316764831542969, "p90": 32.01246967315674, "max": 40.37287902832031, "pos_frac": 0.75, "sample": [37.46662902832031, 27.511550903320312, -6.014286041259766, 7.701457977294922, 8.201751708984375, 40.37287902832031, 1.990966796875, 12.214456558227539, -11.417472839355469, 8.068389892578125, 35.248382568359375, -15.401268005371094, -4.298944473266602, 7.842174530029297, 8.370506286621094, 13.2952880859375, 17.55286407470703, 2.9232864379882812, -13.815757751464844, 0.498138427734375, 19.145654678344727, 8.403694152832031, 19.576927185058594, 29.14397430419922, -1.838693618774414, -3.1964492797851562, 12.112152099609375, 14.789600372314453, 2.876983642578125, 6.032093048095703, 31.831342697143555, -4.914726257324219, 4.430246353149414, 13.538772583007812, 17.242502212524414, -2.8938674926757812, 12.262710571289062, 36.26566696166992, -4.288492202758789, 8.263023376464844, 4.180992126464844, 6.44671630859375, 26.17456817626953, 26.876380920410156, -8.349727630615234, 10.194656372070312, 13.7186279296875, -1.4878997802734375, 19.628524780273438, -1.1634559631347656, 32.09009552001953, 33.96143341064453, 0.42517852783203125, 21.41229248046875, 14.032928466796875, 33.36528396606445, 14.670974731445312, 1.0337905883789062, 1.2879295349121094, 10.872978210449219, -26.592742919921875, -5.584262847900391, 12.068731307983398, -4.586273193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000204.npy"}
|
|
{"epoch": 0.30839002267573695, "step": 205, "batch_size": 64, "mean": 9.760737419128418, "std": 16.018522262573242, "min": -33.972930908203125, "p10": -7.067110633850096, "median": 9.600848197937012, "p90": 30.11375579833985, "max": 47.7061767578125, "pos_frac": 0.71875, "sample": [47.7061767578125, 3.3589305877685547, -0.2235565185546875, 8.004405975341797, 10.486089706420898, 41.94036865234375, 9.354742050170898, 14.037727355957031, -33.972930908203125, 26.339019775390625, 3.4952754974365234, 3.389019012451172, 45.28192901611328, 12.077085494995117, 10.78298568725586, 1.6071853637695312, 7.464447021484375, -4.4629058837890625, -4.785247802734375, 15.826789855957031, 7.17839241027832, 17.119544982910156, 2.9137420654296875, 10.566360473632812, 23.033851623535156, -1.8110427856445312, -7.485420227050781, -3.2676315307617188, 5.300910949707031, 38.700294494628906, -2.357666015625, -6.091054916381836, 17.389862060546875, 9.846954345703125, -1.552703857421875, 14.619873046875, 14.54161262512207, 16.707855224609375, 0.6742401123046875, 18.627342224121094, -1.9189453125, 26.61139678955078, 20.17502212524414, 14.601669311523438, 35.601844787597656, 30.570068359375, 1.263275146484375, 12.260848999023438, 25.923080444335938, -7.848793029785156, 9.02114486694336, 8.900321960449219, 21.94234848022461, 14.791175842285156, -23.32440185546875, 29.049026489257812, 19.93471908569336, -9.868261337280273, 32.50408172607422, -7.871974945068359, -2.25872802734375, -30.01458740234375, -1.8738479614257812, 14.153858184814453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000205.npy"}
|
|
{"epoch": 0.30990173847316704, "step": 206, "batch_size": 64, "mean": 7.38308048248291, "std": 16.126510620117188, "min": -28.827072143554688, "p10": -14.388344573974605, "median": 5.456782341003418, "p90": 29.92555313110352, "max": 37.06981658935547, "pos_frac": 0.6875, "sample": [-9.284852981567383, -2.16668701171875, 3.4321212768554688, -20.986045837402344, -9.697540283203125, -2.6197586059570312, 4.4488372802734375, 7.851757049560547, 23.09253692626953, 9.849668502807617, -1.2425289154052734, 36.48725128173828, 15.873790740966797, -4.178747177124023, -6.7871246337890625, 21.696044921875, -1.930389404296875, 36.06847381591797, -17.834716796875, 6.332822799682617, 37.06981658935547, 15.079532623291016, 3.4408035278320312, 7.1112823486328125, -6.031305313110352, 30.526626586914062, 6.737424850463867, -24.114660263061523, 18.310556411743164, 21.471267700195312, 25.05559539794922, 32.11517333984375, 3.450347900390625, 5.966817855834961, 27.269775390625, 4.946746826171875, -28.827072143554688, -16.39868927001953, -4.066183090209961, 1.6175918579101562, 16.571556091308594, 1.0413322448730469, 20.97210693359375, -6.6074676513671875, -1.6584205627441406, -17.471500396728516, 33.003562927246094, 14.916168212890625, 4.5851287841796875, 6.26507568359375, 21.3358154296875, 27.452606201171875, 27.155197143554688, -7.045646667480469, 14.722198486328125, 28.523048400878906, 4.790302276611328, 1.322092056274414, 7.818550109863281, 30.600067138671875, -24.233694076538086, 1.2021713256835938, 1.8034210205078125, 16.317115783691406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000206.npy"}
|
|
{"epoch": 0.31141345427059713, "step": 207, "batch_size": 64, "mean": 10.263675689697266, "std": 14.74782657623291, "min": -18.497726440429688, "p10": -6.627062606811523, "median": 7.402593612670898, "p90": 32.92481307983399, "max": 43.8516845703125, "pos_frac": 0.75, "sample": [43.8516845703125, 2.067514419555664, -4.950782775878906, 3.0129318237304688, 11.59375, 11.931221008300781, 2.459381103515625, 5.791831970214844, 2.1690139770507812, 30.339401245117188, 9.079429626464844, 4.812812805175781, -2.058748245239258, 33.44512176513672, -2.2274646759033203, -6.847293853759766, 8.343780517578125, 19.564483642578125, 7.092761993408203, 27.53519058227539, 29.33188247680664, 8.21160888671875, 12.839508056640625, 4.44891357421875, -8.080465316772461, -6.113189697265625, -3.7686691284179688, 33.384666442871094, -9.0723876953125, 7.712425231933594, 31.851821899414062, 5.982658386230469, 39.89360046386719, -4.4064483642578125, 6.009857177734375, -3.3082923889160156, -13.646974563598633, 13.630836486816406, 8.49786376953125, 2.2141475677490234, 23.597366333007812, 26.92499542236328, 3.7825164794921875, 13.827198028564453, -0.08962631225585938, 18.310165405273438, 19.488670349121094, -18.497726440429688, 26.594087600708008, 16.784652709960938, -17.778038024902344, 22.002683639526367, 2.186004638671875, 13.105712890625, 5.2137298583984375, 38.47608184814453, 35.3160400390625, 6.140266418457031, 12.820999145507812, 15.795703887939453, -8.981719970703125, -4.0839080810546875, 5.489753723144531, 37.83024215698242], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000207.npy"}
|
|
{"epoch": 0.3129251700680272, "step": 208, "batch_size": 64, "mean": 5.818046569824219, "std": 14.487380027770996, "min": -28.85187530517578, "p10": -10.383737182617187, "median": 6.045205116271973, "p90": 26.610861396789556, "max": 35.45391845703125, "pos_frac": 0.625, "sample": [8.16569709777832, 1.1390933990478516, 23.9660701751709, -6.430713653564453, 28.469207763671875, 28.415058135986328, 2.905557632446289, -6.2259979248046875, -9.841583251953125, -24.475791931152344, -3.387542724609375, -10.6160888671875, 24.387596130371094, 14.737953186035156, 21.900318145751953, 4.993904113769531, 6.716556549072266, 12.0235595703125, -3.7818851470947266, -10.764022827148438, 19.24536895751953, -6.2254638671875, 2.7683448791503906, -7.668647766113281, 15.652582168579102, 27.004125595092773, 22.400474548339844, -6.917102813720703, 8.317207336425781, 17.347341537475586, -6.576946258544922, 0.5605087280273438, -8.84234619140625, 8.800094604492188, -19.76715087890625, 14.715911865234375, 29.372188568115234, 19.568984985351562, 6.2325439453125, 1.18939208984375, 35.45391845703125, 9.0784912109375, 4.038749694824219, -4.833770751953125, 25.69324493408203, -6.231758117675781, 29.662109375, -3.381654739379883, 14.773721694946289, 6.221406936645508, -11.719917297363281, 16.900299072265625, -1.0926513671875, 17.283889770507812, -8.57973861694336, 12.626091003417969, 8.0341796875, -5.4764251708984375, -18.1417179107666, -0.0733795166015625, -28.85187530517578, 7.226226806640625, 5.8690032958984375, 28.40221405029297], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000208.npy"}
|
|
{"epoch": 0.3144368858654573, "step": 209, "batch_size": 64, "mean": 11.871179580688477, "std": 17.419673919677734, "min": -40.22966766357422, "p10": -5.348920440673828, "median": 10.37185287475586, "p90": 32.68497848510742, "max": 47.51111602783203, "pos_frac": 0.78125, "sample": [30.208499908447266, 32.475120544433594, 4.78515625, 9.517841339111328, -7.369098663330078, 37.854278564453125, 11.32746696472168, 29.452590942382812, 6.137779235839844, 40.17633056640625, -4.558616638183594, 4.056816101074219, -12.468521118164062, 29.702674865722656, 3.240509033203125, 10.071983337402344, -3.0615997314453125, -1.9709129333496094, -2.3247756958007812, 44.16087341308594, 2.3460464477539062, -5.6876220703125, 0.39801788330078125, 23.09856414794922, 3.99200439453125, 8.176620483398438, 23.5673828125, 16.439361572265625, 36.95164489746094, 12.615585327148438, 18.1546688079834, 19.283538818359375, 6.020862579345703, 30.050514221191406, 32.77491760253906, -1.1715087890625, 32.47068405151367, 14.860176086425781, -40.22966766357422, 4.597015380859375, -6.770622253417969, 3.1715011596679688, -3.4706039428710938, -3.1355438232421875, 29.295578002929688, 11.876585006713867, 47.51111602783203, 29.751312255859375, 10.671722412109375, 27.363117218017578, 6.267402648925781, -16.811704635620117, 3.881662368774414, 15.359466552734375, 6.8028564453125, 2.772796630859375, 21.939571380615234, -36.687164306640625, 11.939155578613281, 15.490453720092773, 31.37714385986328, 36.50324630737305, 14.026935577392578, 0.5063629150390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000209.npy"}
|
|
{"epoch": 0.31594860166288735, "step": 210, "batch_size": 64, "mean": 11.528844833374023, "std": 16.586437225341797, "min": -20.77465057373047, "p10": -11.15234565734863, "median": 9.125539779663086, "p90": 35.09998016357422, "max": 45.32064437866211, "pos_frac": 0.78125, "sample": [39.045494079589844, 10.113582611083984, 9.369720458984375, 14.71368408203125, -2.647045135498047, 19.84954071044922, 45.32064437866211, 17.56005859375, 37.53099822998047, 37.97119140625, 3.133892059326172, 35.3438720703125, 33.449676513671875, 2.76953125, 36.83088684082031, 2.594329833984375, -17.644624710083008, -18.514690399169922, 33.863311767578125, -12.98004150390625, 34.53089904785156, 11.512340545654297, 6.607889175415039, -3.0525970458984375, 30.15752410888672, 0.989288330078125, -1.0493602752685547, 6.948619842529297, 33.58781433105469, -3.9057464599609375, 32.273704528808594, 3.7959136962890625, 12.322311401367188, 6.0994415283203125, 10.88546371459961, -2.0829849243164062, 1.6142807006835938, -6.887722015380859, 0.5193023681640625, 31.989919662475586, -16.633665084838867, 17.19683837890625, 10.563652038574219, 5.20147705078125, 23.99591064453125, -0.4907569885253906, 13.594316482543945, 4.771062850952148, 8.932151794433594, 4.789154052734375, 8.527835845947266, 23.36686134338379, 9.318927764892578, -17.538827896118164, 4.1031646728515625, 41.02568817138672, 22.065383911132812, 19.836711883544922, 24.23421287536621, -20.77465057373047, 1.8807258605957031, -14.079750061035156, 8.2611083984375, 21.16822052001953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000210.npy"}
|
|
{"epoch": 0.31746031746031744, "step": 211, "batch_size": 64, "mean": 10.516796112060547, "std": 15.040623664855957, "min": -19.8148193359375, "p10": -5.675372123718262, "median": 7.6059370040893555, "p90": 33.17642211914063, "max": 53.432586669921875, "pos_frac": 0.765625, "sample": [1.2818527221679688, -19.8148193359375, 14.84893798828125, 4.012636184692383, -0.36827850341796875, 1.7956657409667969, -5.3430023193359375, -1.233673095703125, 12.684661865234375, 3.551694869995117, 15.16943359375, -1.7961006164550781, 14.386825561523438, 31.935272216796875, 3.8123435974121094, 18.307445526123047, 1.2360076904296875, -0.18365859985351562, 28.907752990722656, -7.515964508056641, 6.653890609741211, -13.474212646484375, 16.012943267822266, -10.50982666015625, 4.070514678955078, -13.406229019165039, 0.813720703125, 10.877700805664062, 38.56447982788086, 15.507570266723633, 17.288192749023438, 8.5579833984375, 10.606447219848633, 2.2190017700195312, 33.708343505859375, 13.272693634033203, 10.157241821289062, 3.716949462890625, -5.743152618408203, 16.790111541748047, 6.406162261962891, 53.432586669921875, 41.464630126953125, 1.731161117553711, -1.7186508178710938, 21.636165618896484, 34.38441848754883, 14.284500122070312, 14.207481384277344, 35.61686706542969, -10.652519226074219, 11.501705169677734, 31.352310180664062, 29.326522827148438, 24.703407287597656, 29.207611083984375, 4.968452453613281, 19.60962677001953, -5.517217636108398, 34.20164489746094, 1.1024818420410156, -4.7804718017578125, 2.310699462890625, 2.936004638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000211.npy"}
|
|
{"epoch": 0.31897203325774753, "step": 212, "batch_size": 64, "mean": 11.417258262634277, "std": 16.215335845947266, "min": -38.18583679199219, "p10": -6.6039794921874995, "median": 11.10525894165039, "p90": 34.01161117553712, "max": 45.719512939453125, "pos_frac": 0.78125, "sample": [35.072914123535156, 21.053447723388672, 23.64993667602539, 32.2400016784668, 0.8966960906982422, 5.224803924560547, 43.345455169677734, 17.48287010192871, -4.482307434082031, 26.910049438476562, -2.9686546325683594, -38.18583679199219, 28.863557815551758, 45.719512939453125, 6.302032470703125, 26.298667907714844, 26.03545379638672, 15.831039428710938, -10.53802490234375, 17.48853874206543, 12.611129760742188, 2.0700740814208984, 0.3944816589355469, 10.888702392578125, -22.46331787109375, 0.72119140625, 14.047332763671875, 0.8636455535888672, 32.29559326171875, 36.9149169921875, 18.228073120117188, 18.58572006225586, 19.31452178955078, 9.276481628417969, 11.321815490722656, 7.068382263183594, -4.593120574951172, 14.310081481933594, 0.8625717163085938, 9.710689544677734, -2.3824386596679688, 5.075096130371094, 1.5581092834472656, -5.792083740234375, 13.302133560180664, 21.46851348876953, 13.80038070678711, 38.93486785888672, -11.424800872802734, -6.951934814453125, -1.4029369354248047, 3.31256103515625, 18.995620727539062, -7.867576599121094, 18.698150634765625, 35.12230682373047, 26.889114379882812, 34.747047424316406, 24.81023406982422, 1.328857421875, 6.8880157470703125, 7.8082122802734375, -13.42578125, -1.4562759399414062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000212.npy"}
|
|
{"epoch": 0.3204837490551776, "step": 213, "batch_size": 64, "mean": 13.19664192199707, "std": 17.5982723236084, "min": -40.52397155761719, "p10": -7.262454223632812, "median": 10.164331436157227, "p90": 36.77757987976075, "max": 58.149566650390625, "pos_frac": 0.796875, "sample": [25.549949645996094, 8.701927185058594, 5.3665618896484375, 30.103851318359375, 13.640846252441406, 8.036384582519531, -9.77984619140625, -3.9150009155273438, -6.507659912109375, 29.133159637451172, 34.41705322265625, 2.76922607421875, 4.23028564453125, -7.5859375, -40.52397155761719, -11.773605346679688, 23.006200790405273, 58.149566650390625, -0.9403800964355469, 19.7313232421875, 28.72158432006836, -19.27669906616211, 32.88195037841797, 44.44109344482422, 30.252975463867188, 8.707561492919922, -2.2644786834716797, -15.853515625, 4.224021911621094, 6.538852691650391, -9.263477325439453, 12.972259521484375, 15.592666625976562, 10.809864044189453, 17.353439331054688, 16.14202880859375, 3.642110824584961, 23.41891860961914, 34.40428161621094, 5.599342346191406, 5.515987396240234, 11.835975646972656, 39.158626556396484, 1.7830829620361328, 28.14935302734375, 15.77264404296875, 3.6609458923339844, 23.39340591430664, 38.35053253173828, 3.1204299926757812, -0.055694580078125, 19.860733032226562, -2.6860809326171875, 0.48980712890625, 4.785247802734375, 15.281005859375, 7.0770263671875, 39.511688232421875, 9.518798828125, 8.960617065429688, 34.20918273925781, 38.02207565307617, 30.225723266601562, 37.78923416137695], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000213.npy"}
|
|
{"epoch": 0.3219954648526077, "step": 214, "batch_size": 64, "mean": 12.956791877746582, "std": 17.630815505981445, "min": -23.82788848876953, "p10": -9.339817428588866, "median": 12.39657211303711, "p90": 36.932698059082036, "max": 52.04954528808594, "pos_frac": 0.6875, "sample": [22.205617904663086, 16.003395080566406, -2.206634521484375, 9.676006317138672, 12.701622009277344, 29.115312576293945, 2.8931007385253906, 7.600347518920898, 22.479896545410156, 32.799766540527344, 37.01985168457031, 26.28801918029785, 15.877893447875977, 2.4842529296875, 24.801956176757812, 30.777511596679688, 23.294322967529297, -22.918731689453125, 21.462875366210938, 22.013168334960938, -1.6811141967773438, 19.08470916748047, 6.939613342285156, -2.1020050048828125, -0.6392498016357422, 29.705717086791992, -11.980728149414062, 18.21861457824707, 40.82868576049805, 28.46076202392578, 12.091522216796875, -3.4903030395507812, -8.467586517333984, 2.54608154296875, 20.961952209472656, 28.798606872558594, -15.362640380859375, -3.8734893798828125, 36.729339599609375, 52.04954528808594, -0.0570526123046875, -9.713630676269531, 33.58181381225586, 23.633960723876953, 41.75498580932617, 42.73139953613281, 2.528858184814453, -0.3608283996582031, 8.894065856933594, -7.89080810546875, -10.59814453125, 41.279144287109375, 7.458795547485352, -3.682645797729492, 5.072301864624023, -1.0429611206054688, 39.939697265625, 10.201072692871094, -23.82788848876953, 22.19366455078125, 18.853160858154297, 21.518531799316406, -3.3968238830566406, -13.023565292358398], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000214.npy"}
|
|
{"epoch": 0.3235071806500378, "step": 215, "batch_size": 64, "mean": 13.223045349121094, "std": 18.13860321044922, "min": -26.60675811767578, "p10": -10.369454383850096, "median": 8.740869522094727, "p90": 36.30761222839356, "max": 46.87152099609375, "pos_frac": 0.703125, "sample": [13.944942474365234, -9.373123168945312, -26.60675811767578, 17.428924560546875, 12.739753723144531, -16.29175567626953, 6.99346923828125, -2.1632766723632812, 34.95027160644531, 36.76274490356445, 5.613151550292969, -1.6395072937011719, -13.071846008300781, 8.699237823486328, 15.937332153320312, 40.003753662109375, 1.575897216796875, 32.84901428222656, 2.46783447265625, -0.3338775634765625, 8.782501220703125, 5.663505554199219, -1.3495063781738281, 25.810699462890625, 33.569007873535156, 1.5808525085449219, -2.002166748046875, 28.47002410888672, 28.486061096191406, -10.796453475952148, -2.346271514892578, 33.248252868652344, 2.7769813537597656, -8.006317138671875, -6.58953857421875, -14.845413208007812, 26.842344284057617, 21.013450622558594, 27.81810760498047, 40.775177001953125, 8.65473747253418, 7.6833343505859375, -5.2772369384765625, -0.8535232543945312, 29.353485107421875, 35.245635986328125, 41.22650909423828, 42.47832107543945, 31.291664123535156, 46.87152099609375, 19.36941909790039, 34.78046417236328, 26.574569702148438, -12.462356567382812, 6.986026763916016, 2.4099769592285156, 12.81561279296875, 40.651214599609375, 27.658920288085938, -0.5016212463378906, 32.055458068847656, 3.026874542236328, 28.08757781982422, -11.239177703857422], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000215.npy"}
|
|
{"epoch": 0.3250188964474679, "step": 216, "batch_size": 64, "mean": 13.901348114013672, "std": 17.800413131713867, "min": -31.92864990234375, "p10": -3.8316648483276365, "median": 13.824542999267578, "p90": 36.823696136474624, "max": 74.06265258789062, "pos_frac": 0.78125, "sample": [1.428955078125, 32.198455810546875, 25.82843017578125, 33.30162048339844, 20.15378189086914, 7.395851135253906, 12.817802429199219, 5.500640869140625, 23.110519409179688, 33.97441864013672, 23.559494018554688, 19.524940490722656, 2.3036117553710938, 5.079742431640625, 15.903587341308594, -5.40594482421875, 5.789512634277344, 12.342979431152344, -2.2969894409179688, 41.892578125, 21.959686279296875, 38.498565673828125, 17.51398468017578, 28.050643920898438, 19.635417938232422, 9.504852294921875, 39.13642120361328, 1.0406684875488281, 3.5698699951171875, 41.433502197265625, 26.446449279785156, -31.92864990234375, 8.177654266357422, 15.570659637451172, 5.42279052734375, 17.4896240234375, -15.445037841796875, 1.2704353332519531, 17.311100006103516, -3.6160831451416016, -2.0250205993652344, 74.06265258789062, 38.04481506347656, -7.821815490722656, 7.0604400634765625, -12.632484436035156, 14.831283569335938, 1.713409423828125, -3.5754776000976562, 27.04248046875, 45.70448684692383, -16.81283187866211, -3.4957275390625, 0.09358406066894531, 32.56713104248047, -1.1625995635986328, 10.939691543579102, 20.11608123779297, -3.9240570068359375, 32.015960693359375, -1.7455787658691406, 27.3017578125, 18.488487243652344, 17.453033447265625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000216.npy"}
|
|
{"epoch": 0.32653061224489793, "step": 217, "batch_size": 64, "mean": 14.666132926940918, "std": 17.786197662353516, "min": -23.56414031982422, "p10": -4.510900115966796, "median": 11.50804328918457, "p90": 37.87832756042481, "max": 57.00163269042969, "pos_frac": 0.78125, "sample": [24.262542724609375, 36.32605743408203, -6.1528472900390625, 0.6242752075195312, -0.5686283111572266, -5.094356536865234, 3.7685375213623047, 57.00163269042969, -2.0826568603515625, 6.3101806640625, 47.028564453125, 47.82429504394531, 5.295921325683594, -3.1495018005371094, 9.451606750488281, 3.8891983032226562, 32.14838409423828, 18.19715118408203, 6.535003662109375, 35.44713592529297, 7.869516372680664, 23.019556045532227, 3.99346923828125, 32.029937744140625, 10.231452941894531, 35.980262756347656, -23.56414031982422, 0.46326446533203125, 3.3122177124023438, 12.78463363647461, 37.012088775634766, 16.3255615234375, 13.326629638671875, -7.0402984619140625, 21.990509033203125, -12.511404037475586, 6.009925842285156, 23.750539779663086, -0.48321533203125, -11.331260681152344, -22.2965087890625, 32.15620422363281, 17.521343231201172, -2.3052330017089844, -0.9130783081054688, 4.975502014160156, 8.401374816894531, 8.71405029296875, 18.676116943359375, 51.66331481933594, 15.407569885253906, 31.532791137695312, 9.15726089477539, 1.7710227966308594, 41.323158264160156, 22.8994140625, 41.45977020263672, 19.873512268066406, 18.202499389648438, 34.29665756225586, 22.296142578125, 38.24957275390625, 16.27676773071289, -0.9384918212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000217.npy"}
|
|
{"epoch": 0.328042328042328, "step": 218, "batch_size": 64, "mean": 9.199453353881836, "std": 19.86527442932129, "min": -46.792457580566406, "p10": -11.933803749084472, "median": 8.825904846191406, "p90": 34.95010986328125, "max": 55.50526428222656, "pos_frac": 0.703125, "sample": [7.6586456298828125, 1.2924785614013672, 10.963325500488281, 18.754587173461914, 5.378143310546875, -5.7691802978515625, 6.389106750488281, -0.798370361328125, 0.034393310546875, 35.007225036621094, 10.125839233398438, 9.9931640625, 35.14472198486328, -30.680152893066406, -12.24029541015625, -21.918733596801758, 7.054023742675781, 38.4888916015625, -25.749542236328125, -2.9095916748046875, 20.177352905273438, 11.432815551757812, 5.875762939453125, 10.944389343261719, 23.226238250732422, 55.50526428222656, 34.81684112548828, 45.48902893066406, 18.30409812927246, 20.330062866210938, -0.9561653137207031, 22.908050537109375, 18.959930419921875, -10.086071014404297, -7.975688934326172, -9.560348510742188, 4.3003387451171875, 14.080648422241211, -0.9119453430175781, 29.463951110839844, 34.2726936340332, 21.131378173828125, -2.2441635131835938, 3.7098159790039062, 10.276466369628906, -11.218656539916992, 23.272871017456055, 0.6706867218017578, -21.426258087158203, -9.826454162597656, 29.149883270263672, -28.776710510253906, 13.49420166015625, 0.2214202880859375, 10.579627990722656, 29.675537109375, 5.588188171386719, 43.31250762939453, -46.792457580566406, -4.2524566650390625, 15.809566497802734, 6.9544525146484375, 27.47393798828125, 45.16572570800781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000218.npy"}
|
|
{"epoch": 0.3295540438397581, "step": 219, "batch_size": 64, "mean": 9.416379928588867, "std": 15.713883399963379, "min": -24.520793914794922, "p10": -11.101161193847656, "median": 8.33357048034668, "p90": 31.33138542175293, "max": 43.014869689941406, "pos_frac": 0.71875, "sample": [9.852500915527344, 34.80788040161133, 24.18053436279297, 7.698492050170898, 15.151176452636719, 0.19976043701171875, 1.9413185119628906, 1.870107650756836, 11.292823791503906, -11.350189208984375, -2.1943931579589844, 41.693687438964844, 19.59124755859375, -13.779541015625, -17.019332885742188, 17.654220581054688, 14.691055297851562, 2.9855918884277344, 12.518733978271484, 1.3537712097167969, 2.279043197631836, -8.450820922851562, -11.3780517578125, 8.20479965209961, -10.520095825195312, -0.724334716796875, 6.034706115722656, -1.7026290893554688, 14.207260131835938, 8.46234130859375, 26.905006408691406, 33.995697021484375, 35.99608612060547, 38.282470703125, 19.48310089111328, 31.450759887695312, -3.5483245849609375, 3.655641555786133, -3.3991546630859375, 13.352397918701172, 13.412094116210938, 8.505996704101562, -23.81354522705078, 15.594612121582031, -2.2672348022460938, -2.819183349609375, 22.566333770751953, 1.1429290771484375, 14.600677490234375, 43.014869689941406, 24.237472534179688, 29.014846801757812, 9.45145034790039, -1.7180252075195312, -24.520793914794922, 3.680706024169922, 6.855430603027344, 6.914131164550781, 26.313507080078125, -1.8492889404296875, 31.052845001220703, -15.381072998046875, 28.84195327758789, 14.092269897460938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000219.npy"}
|
|
{"epoch": 0.3310657596371882, "step": 220, "batch_size": 64, "mean": 11.239852905273438, "std": 18.789094924926758, "min": -30.87594985961914, "p10": -7.825466918945312, "median": 8.47935962677002, "p90": 36.82226562500001, "max": 52.72007751464844, "pos_frac": 0.6875, "sample": [7.300119400024414, -7.918914794921875, 17.048755645751953, 6.388053894042969, 2.3166961669921875, 15.812255859375, 46.630126953125, 10.070953369140625, 8.840608596801758, 6.721321105957031, -4.734100341796875, 45.24342346191406, 4.523738861083984, -22.627498626708984, -7.607421875, -5.870861053466797, 37.345558166503906, 0.8134231567382812, -3.551250457763672, 1.1746826171875, 29.016555786132812, 11.398643493652344, -3.2077808380126953, -24.56419563293457, -16.562759399414062, -10.721548080444336, -5.447925567626953, -9.390792846679688, -1.599609375, 24.394798278808594, 25.61614227294922, 17.176132202148438, 18.370452880859375, 34.101226806640625, 37.94677734375, 4.267093658447266, 28.580352783203125, 22.58489227294922, 30.530622482299805, -3.404632568359375, 2.5546417236328125, 19.83173179626465, -5.033393859863281, 17.608070373535156, 52.72007751464844, -3.0583648681640625, 38.082313537597656, 0.019901275634765625, 13.929445266723633, 35.60124969482422, 30.575868606567383, -6.391324996948242, 34.07904815673828, -1.4849357604980469, 11.841522216796875, 8.118110656738281, -30.87594985961914, 13.383350372314453, 27.606643676757812, 0.849853515625, 51.183441162109375, -6.463687896728516, 31.86349105834961, 15.805366516113281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000220.npy"}
|
|
{"epoch": 0.3325774754346183, "step": 221, "batch_size": 64, "mean": 12.249614715576172, "std": 18.936140060424805, "min": -29.71125030517578, "p10": -10.140297317504883, "median": 9.022319793701172, "p90": 39.218944549560575, "max": 54.236297607421875, "pos_frac": 0.765625, "sample": [5.673734664916992, -10.14773941040039, 10.496437072753906, 12.688621520996094, -1.3305797576904297, 32.49107360839844, 1.1661453247070312, 6.369659423828125, 17.998184204101562, -16.563560485839844, -2.906085968017578, 8.878219604492188, 9.620784759521484, 11.213775634765625, -10.969032287597656, 6.375389099121094, 0.7846221923828125, -1.0631160736083984, 18.448017120361328, 18.683837890625, 30.707794189453125, 31.245956420898438, 11.031513214111328, 31.278213500976562, 54.236297607421875, 5.156917572021484, 28.825119018554688, 50.48397445678711, -2.924346923828125, 3.9327850341796875, 20.679794311523438, 42.102317810058594, 1.0217742919921875, 16.18350601196289, 52.20850372314453, 7.7714691162109375, 24.8929443359375, -10.122932434082031, 18.63176727294922, 26.83355712890625, 22.96358871459961, -13.010059356689453, 3.5531997680664062, -0.5817852020263672, 0.19710731506347656, -1.7388916015625, 0.9630947113037109, 46.32342529296875, 2.4870376586914062, 3.269367218017578, -19.994491577148438, -29.71125030517578, 1.9001960754394531, 52.725616455078125, 29.006271362304688, -19.54413414001465, 9.357536315917969, 14.966789245605469, 50.936767578125, 30.185470581054688, -0.0034332275390625, 0.2970123291015625, 28.1751708984375, 9.166419982910156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000221.npy"}
|
|
{"epoch": 0.3340891912320484, "step": 222, "batch_size": 64, "mean": 16.3406982421875, "std": 19.27525520324707, "min": -26.10077667236328, "p10": -5.0058702468872065, "median": 16.002638816833496, "p90": 38.66836357116699, "max": 57.902984619140625, "pos_frac": 0.765625, "sample": [-2.210721969604492, -20.56523895263672, 21.20745849609375, 11.150436401367188, 33.08529281616211, 34.50246810913086, 19.88058853149414, -5.913349151611328, 36.763702392578125, 45.12103271484375, 19.026824951171875, 5.097370147705078, -20.103187561035156, 2.532001495361328, 29.435895919799805, 4.7576446533203125, 7.212608337402344, 7.109935760498047, 5.375377655029297, 28.682327270507812, 21.371788024902344, -0.16747474670410156, 28.764923095703125, 36.524070739746094, -2.5411300659179688, 38.73912811279297, -3.9512557983398438, 56.562225341796875, -2.829183578491211, 27.80219268798828, 13.47248649597168, 10.827255249023438, 39.85546875, 34.75269317626953, 10.111129760742188, 22.091293334960938, 21.07720947265625, -4.1058349609375, -5.391599655151367, -2.3987884521484375, -26.10077667236328, 18.532791137695312, 12.423736572265625, 12.605682373046875, 1.6956825256347656, 57.902984619140625, 8.375526428222656, -23.25975799560547, 0.072906494140625, 30.656082153320312, 31.57379913330078, -1.810516357421875, 36.15788269042969, 46.915550231933594, 24.094512939453125, 8.564674377441406, 38.50324630737305, 32.896484375, 11.755615234375, 36.282135009765625, 24.621383666992188, 23.187843322753906, 46.536468505859375, -9.094314575195312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000222.npy"}
|
|
{"epoch": 0.3356009070294785, "step": 223, "batch_size": 64, "mean": 14.373746871948242, "std": 17.660062789916992, "min": -16.4720458984375, "p10": -6.538718605041503, "median": 12.863533020019531, "p90": 38.780368804931655, "max": 55.384178161621094, "pos_frac": 0.71875, "sample": [25.269287109375, 7.031791687011719, 28.4852294921875, 7.0881500244140625, -3.235137939453125, 25.515338897705078, -0.76483154296875, 18.088180541992188, 10.603752136230469, 22.00159454345703, 2.018596649169922, 44.30799865722656, 26.680564880371094, 29.455963134765625, 1.4766960144042969, -9.487422943115234, -0.7802715301513672, 27.57354736328125, 9.483978271484375, -5.006481170654297, 35.10578536987305, 18.27117919921875, -8.80337142944336, 23.356409072875977, -7.5791168212890625, 4.68603515625, -1.6230030059814453, 27.834884643554688, -4.486291885375977, -11.489845275878906, 28.27044677734375, 9.017841339111328, -16.4720458984375, 5.272514343261719, 2.2861328125, 47.43757247924805, 33.84502410888672, 22.065967559814453, -3.8448867797851562, -14.478038787841797, -4.69354248046875, -6.685850143432617, 21.187767028808594, 23.361618041992188, 41.948333740234375, 8.807315826416016, 17.667144775390625, 28.1545467376709, 11.647674560546875, 40.35519027709961, 55.384178161621094, -6.195411682128906, 52.53861999511719, 43.04686737060547, 32.56500244140625, 16.7955322265625, 28.15416717529297, 29.808067321777344, 14.079391479492188, -2.626893997192383, 1.338165283203125, 17.851531982421875, 2.169830322265625, -1.2192230224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000223.npy"}
|
|
{"epoch": 0.3371126228269085, "step": 224, "batch_size": 64, "mean": 8.115743637084961, "std": 17.603456497192383, "min": -26.023235321044922, "p10": -15.221031761169431, "median": 5.008315086364746, "p90": 33.47135429382325, "max": 52.220245361328125, "pos_frac": 0.6875, "sample": [9.776668548583984, 52.220245361328125, 14.769935607910156, 13.4429931640625, 5.015045166015625, 38.454898834228516, 19.929420471191406, 21.647804260253906, -1.1826114654541016, 3.728118896484375, 33.58712387084961, 5.933115005493164, 28.406021118164062, 35.40281295776367, -16.518787384033203, -0.6290988922119141, 5.001585006713867, 10.839332580566406, 17.26941680908203, 0.1244964599609375, 2.9405517578125, 19.479537963867188, 1.2951812744140625, -24.859506607055664, -4.922271728515625, -1.0149116516113281, -8.248279571533203, -11.283660888671875, 11.778533935546875, 5.7100067138671875, 24.388587951660156, -26.023235321044922, -15.863016128540039, -7.3415069580078125, -6.77325439453125, 0.08473777770996094, -0.5327548980712891, -22.674468994140625, 5.024787902832031, 13.845466613769531, -17.17510223388672, 17.029617309570312, 27.223939895629883, 1.3759613037109375, -16.2247314453125, 3.79803466796875, 12.941566467285156, 22.254215240478516, -13.723068237304688, -2.4289093017578125, -2.5599746704101562, 16.197294235229492, 26.27769660949707, 33.20122528076172, 51.91832733154297, 12.686203002929688, 3.9434661865234375, 9.21258544921875, 0.4439563751220703, 40.7406120300293, -4.851776123046875, 39.61822509765625, 3.3228988647460938, 1.9562568664550781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000224.npy"}
|
|
{"epoch": 0.3386243386243386, "step": 225, "batch_size": 64, "mean": 14.582175254821777, "std": 19.1368465423584, "min": -24.179122924804688, "p10": -5.768703842163084, "median": 11.522954940795898, "p90": 42.414061355590825, "max": 56.954864501953125, "pos_frac": 0.78125, "sample": [2.0765438079833984, 14.268096923828125, 2.348308563232422, -0.1685333251953125, -1.186065673828125, 28.363922119140625, -3.855060577392578, -0.0946502685546875, 20.942602157592773, 11.670318603515625, 5.3965911865234375, 35.694488525390625, 4.514320373535156, 42.11767578125, -7.809131622314453, 2.062580108642578, 11.548507690429688, 7.015724182128906, 28.63867950439453, 42.2164306640625, 44.95801544189453, 19.761695861816406, 52.69989776611328, 19.657752990722656, 56.954864501953125, 11.49740219116211, 2.497833251953125, 31.486900329589844, 42.49876022338867, -3.406147003173828, 0.7052001953125, -18.09271240234375, 23.69414520263672, -7.454017639160156, 48.75584411621094, 22.202713012695312, 1.5906181335449219, 19.5753173828125, 19.9959716796875, 7.585773468017578, 2.5250625610351562, 6.8644561767578125, 40.227088928222656, 0.33450889587402344, 27.783706665039062, -22.221893310546875, -15.80377197265625, -2.5070228576660156, 19.481857299804688, 7.260108947753906, 27.1607666015625, 0.18240928649902344, 5.275993347167969, 47.94481658935547, 34.36247634887695, -1.2467632293701172, 4.773284912109375, 15.191490173339844, 46.31193161010742, 19.808191299438477, -24.179122924804688, 20.721893310546875, 36.66936492919922, -6.588836669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000225.npy"}
|
|
{"epoch": 0.3401360544217687, "step": 226, "batch_size": 64, "mean": 11.677728652954102, "std": 20.608911514282227, "min": -34.181053161621094, "p10": -13.471802902221677, "median": 10.571100234985352, "p90": 39.10453834533692, "max": 48.37884521484375, "pos_frac": 0.75, "sample": [5.683204650878906, -14.696365356445312, 17.815185546875, 20.04017448425293, 14.380615234375, 13.626296997070312, 19.12725067138672, 3.2567272186279297, 22.863248825073242, -10.614490509033203, 31.14872169494629, 6.732255935668945, 23.171615600585938, 36.79936599731445, -24.207138061523438, 0.2413177490234375, 4.775472640991211, -34.181053161621094, 39.843990325927734, 23.568191528320312, 43.192649841308594, -4.30145263671875, 33.49609375, 3.6804656982421875, 19.145339965820312, 0.5633735656738281, 5.208026885986328, 40.23754119873047, 34.969886779785156, 28.515342712402344, -7.050313949584961, 37.379150390625, 9.233333587646484, 13.380577087402344, 45.887413024902344, 7.3042449951171875, 17.743003845214844, -26.520843505859375, -3.8539810180664062, 2.6300411224365234, 1.8555908203125, 11.167800903320312, -9.167152404785156, -31.90624237060547, 44.572052001953125, 48.37884521484375, -25.465171813964844, 10.752296447753906, -26.502586364746094, 18.60948944091797, 8.768909454345703, 4.100444793701172, 10.389904022216797, 14.407068252563477, 0.4171600341796875, 46.27970886230469, 36.0194091796875, 34.54479217529297, -7.4981231689453125, -2.6797046661376953, 34.265464782714844, -1.4696922302246094, -5.152997970581055, 32.47289276123047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000226.npy"}
|
|
{"epoch": 0.3416477702191988, "step": 227, "batch_size": 64, "mean": 12.463022232055664, "std": 21.39310646057129, "min": -37.19629669189453, "p10": -11.00160903930664, "median": 6.375667572021484, "p90": 42.51992568969727, "max": 50.92134094238281, "pos_frac": 0.734375, "sample": [30.35980224609375, 11.113815307617188, 0.179229736328125, 37.18524169921875, 1.9888153076171875, -1.317962646484375, 26.738677978515625, 47.80335235595703, -37.19629669189453, -4.743783950805664, 50.603981018066406, -9.799140930175781, -2.4246158599853516, 44.26805877685547, 15.061431884765625, 7.502227783203125, 1.3566551208496094, -1.4983139038085938, 41.07020568847656, -4.311866760253906, 19.988616943359375, 34.6806640625, 6.367713928222656, 5.1702880859375, 50.92134094238281, 17.791336059570312, -1.6921844482421875, 30.32537841796875, -2.240297317504883, -2.1855010986328125, 25.338088989257812, 50.254512786865234, 20.67657470703125, 2.668548583984375, 1.6612548828125, 29.216068267822266, 49.53544998168945, 1.9204788208007812, 20.653255462646484, 42.684654235839844, 5.048118591308594, 9.730712890625, -11.516952514648438, 10.306098937988281, 31.645431518554688, 3.3641815185546875, 3.4074935913085938, 4.811491012573242, -18.007728576660156, 0.23512649536132812, -27.251850128173828, 25.150650024414062, 23.437789916992188, 28.335617065429688, 1.9332332611083984, -30.87872314453125, 6.3836212158203125, 0.21022415161132812, 42.13555908203125, -7.3453521728515625, -25.029083251953125, -13.624969482421875, 37.82313537597656, 39.65381622314453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000227.npy"}
|
|
{"epoch": 0.3431594860166289, "step": 228, "batch_size": 64, "mean": 13.135068893432617, "std": 19.59518814086914, "min": -32.7256965637207, "p10": -9.771151733398437, "median": 13.305561065673828, "p90": 43.17589874267579, "max": 57.272003173828125, "pos_frac": 0.71875, "sample": [-0.5237102508544922, 36.240211486816406, 2.3992996215820312, 44.56727600097656, 13.234237670898438, 18.83721160888672, 11.504913330078125, 45.93754577636719, 7.311004638671875, -23.47443389892578, 19.536691665649414, -0.071929931640625, 5.949005126953125, -32.7256965637207, 14.199874877929688, 2.5624427795410156, 32.83839797973633, -9.063034057617188, 12.661579132080078, 14.504751205444336, -0.2457427978515625, 47.25457763671875, -22.090492248535156, 15.458305358886719, -6.232818603515625, 31.289546966552734, 27.26017189025879, 57.272003173828125, 8.508707046508789, 28.981239318847656, -5.0858001708984375, -10.074630737304688, 28.802806854248047, 0.7460479736328125, 22.330289840698242, -5.572858810424805, 21.556896209716797, -1.9972820281982422, -14.367202758789062, -7.916891098022461, 11.877578735351562, 24.305458068847656, 34.0841064453125, -19.276155471801758, 14.109130859375, 2.955923080444336, 45.56714630126953, 47.17896270751953, -14.881492614746094, 6.418914794921875, -3.1953201293945312, 11.741436004638672, 25.771682739257812, 21.014423370361328, -3.6709060668945312, 16.755538940429688, 39.929351806640625, 44.64519500732422, 33.068084716796875, 16.71796417236328, 0.03783416748046875, 14.919593811035156, 24.890546798706055, 13.376884460449219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000228.npy"}
|
|
{"epoch": 0.34467120181405897, "step": 229, "batch_size": 64, "mean": 11.799358367919922, "std": 21.295024871826172, "min": -42.27473068237305, "p10": -10.426415252685546, "median": 10.024227142333984, "p90": 38.2568717956543, "max": 52.466339111328125, "pos_frac": 0.78125, "sample": [20.123321533203125, 37.81520080566406, 24.302886962890625, 18.538070678710938, 9.433036804199219, 15.568946838378906, 28.75098419189453, 15.710866928100586, 15.356769561767578, -19.511642456054688, 31.282127380371094, 20.274658203125, 35.522804260253906, -4.244834899902344, 16.552734375, 5.029701232910156, -38.417877197265625, -7.639430999755859, 36.86259460449219, 35.20571517944336, 11.660110473632812, 2.050016403198242, 10.61541748046875, 20.3621826171875, -19.65179443359375, 8.468696594238281, -10.072212219238281, -0.9323158264160156, 3.408538818359375, -8.484306335449219, 1.2744598388671875, 52.466339111328125, -42.27473068237305, 38.44615936279297, -10.578216552734375, 3.773275375366211, 34.20469665527344, -26.53182601928711, 5.381004333496094, 15.6390380859375, 42.56044006347656, 21.79031753540039, 0.25341796875, -2.380281448364258, 34.0564079284668, 0.5258445739746094, 42.57720184326172, 15.832326889038086, 36.71092224121094, 7.600767135620117, 2.1100234985351562, -1.6067180633544922, 21.258193969726562, 1.1926116943359375, 45.868614196777344, 5.1950225830078125, 0.9596099853515625, 1.6678409576416016, 39.29292678833008, 50.50506591796875, 36.12287902832031, 3.9219741821289062, 2.1451148986816406, -38.74280548095703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000229.npy"}
|
|
{"epoch": 0.34618291761148906, "step": 230, "batch_size": 64, "mean": 13.964266777038574, "std": 19.80023765563965, "min": -39.75444793701172, "p10": -3.342880821228026, "median": 7.911427021026611, "p90": 40.85315628051758, "max": 68.4873046875, "pos_frac": 0.84375, "sample": [-4.1668701171875, 8.862945556640625, 31.083826065063477, 1.8019142150878906, 14.819923400878906, -27.857696533203125, 3.0762577056884766, 1.20489501953125, 9.202560424804688, 44.67786407470703, 16.40319061279297, 48.552757263183594, 38.135704040527344, 43.26258087158203, 12.576644897460938, 18.716285705566406, 19.908100128173828, 38.87773895263672, -12.976722717285156, 3.4101409912109375, -39.75444793701172, 5.62457275390625, 13.62847900390625, 21.695518493652344, 21.275489807128906, -1.0047760009765625, 41.294776916503906, 5.500274658203125, 38.59712219238281, 32.861045837402344, 22.178028106689453, 4.354526519775391, -2.114389419555664, 3.707225799560547, 4.733146667480469, 5.614067077636719, 3.4043121337890625, 5.065998077392578, 68.4873046875, 8.144943237304688, 7.402050018310547, 7.101661682128906, 1.4886817932128906, 15.725784301757812, 57.153343200683594, 29.828773498535156, -13.548891067504883, 45.97516632080078, 5.0272979736328125, 2.0030860900878906, 39.82270812988281, -18.62640380859375, 36.275054931640625, 30.380258560180664, 12.728691101074219, 9.295085906982422, 2.8706607818603516, -1.2460498809814453, 6.338094711303711, 2.5101470947265625, 7.677910804748535, 32.21257019042969, -3.8693771362304688, 6.321468353271484], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000230.npy"}
|
|
{"epoch": 0.3476946334089191, "step": 231, "batch_size": 64, "mean": 20.25579833984375, "std": 20.50758171081543, "min": -38.465911865234375, "p10": -4.791854476928711, "median": 24.08773422241211, "p90": 45.26665420532227, "max": 59.111846923828125, "pos_frac": 0.828125, "sample": [39.69230651855469, 10.050247192382812, 30.815631866455078, 38.34381103515625, 2.2908477783203125, 16.750991821289062, 2.3287315368652344, 43.411468505859375, 8.446784973144531, 7.993780136108398, 26.845273971557617, 1.38250732421875, 25.332107543945312, -4.056133270263672, -5.029693603515625, 26.291824340820312, 13.915252685546875, 44.593231201171875, 13.905197143554688, -6.516868591308594, 33.67694091796875, -4.236896514892578, 25.226959228515625, 33.95429611206055, 5.584377288818359, 50.271636962890625, -8.8548583984375, -16.736366271972656, 50.27809143066406, 26.644203186035156, 45.019134521484375, -27.166278839111328, 24.686782836914062, 45.75209045410156, 34.943973541259766, 6.484161376953125, 3.7410144805908203, 20.665050506591797, 4.55560302734375, -2.608612060546875, 59.111846923828125, 6.0929412841796875, 21.2095947265625, 23.68169403076172, 25.434974670410156, 10.582601547241211, 32.21092224121094, -7.11187744140625, 9.927967071533203, -2.511402130126953, 38.43549728393555, 44.06439208984375, 23.161827087402344, 37.244041442871094, 36.065757751464844, 49.269447326660156, 29.14288330078125, 33.06999969482422, 24.4937744140625, 37.76078796386719, -38.465911865234375, 19.350778579711914, 45.37273406982422, 50.10713195800781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000231.npy"}
|
|
{"epoch": 0.3492063492063492, "step": 232, "batch_size": 64, "mean": 13.691271781921387, "std": 24.459745407104492, "min": -59.74822998046875, "p10": -12.413922691345215, "median": 10.716134071350098, "p90": 45.489205551147464, "max": 80.4696044921875, "pos_frac": 0.734375, "sample": [1.6231918334960938, 45.25395584106445, 31.422157287597656, 52.38886260986328, 6.9263458251953125, 37.37474060058594, 14.954345703125, 42.21525573730469, -0.07440185546875, -14.223197937011719, -13.039249420166016, 4.2909393310546875, -5.877098083496094, 11.024158477783203, 12.987957000732422, 40.6004753112793, 1.1217460632324219, 5.276115417480469, 28.46771240234375, 2.6101722717285156, 0.7942771911621094, 26.65967559814453, 30.983795166015625, 17.89288330078125, 8.139739990234375, 25.392536163330078, -7.523435592651367, 50.66392517089844, -7.792926788330078, 14.843818664550781, 15.889631271362305, 43.60014343261719, -12.349420547485352, -2.1851577758789062, 13.061798095703125, 1.37091064453125, 1.112936019897461, -28.591060638427734, 55.676605224609375, 13.143547058105469, -0.02660369873046875, 45.59002685546875, -7.363059997558594, 68.53109741210938, 52.4085693359375, 80.4696044921875, -12.561904907226562, -12.441566467285156, -5.0067596435546875, 7.774709701538086, 12.448799133300781, 39.636695861816406, 29.534698486328125, 11.158973693847656, 4.543247222900391, -23.666641235351562, 0.6112461090087891, 40.55033874511719, 9.657119750976562, 14.440326690673828, -10.221931457519531, -59.74822998046875, 13.406097412109375, 10.408109664916992], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000232.npy"}
|
|
{"epoch": 0.3507180650037793, "step": 233, "batch_size": 64, "mean": 10.83167552947998, "std": 20.816823959350586, "min": -35.453697204589844, "p10": -12.152743530273437, "median": 6.215507507324219, "p90": 36.00264053344727, "max": 59.2398681640625, "pos_frac": 0.65625, "sample": [58.16272735595703, 5.794647216796875, 41.45134735107422, -12.538963317871094, 7.552936553955078, 19.67300796508789, 18.453048706054688, 17.718441009521484, -4.030979156494141, -4.732505798339844, 7.8162384033203125, 59.2398681640625, -3.3605003356933594, -0.4374580383300781, -4.477119445800781, 18.225128173828125, 33.691585540771484, 4.65928840637207, -3.0145950317382812, 1.761423110961914, 40.0482177734375, 5.432342529296875, 31.039390563964844, 35.55262756347656, -16.215789794921875, 14.18505859375, -11.251564025878906, 20.713211059570312, -20.298133850097656, -15.873260498046875, 31.503211975097656, 3.6717567443847656, 33.94633483886719, 17.244792938232422, 36.19550323486328, -35.453697204589844, 11.999603271484375, 2.3570098876953125, -2.5692138671875, 1.1200828552246094, 6.6363677978515625, -27.91570281982422, 28.255569458007812, -0.5121002197265625, 21.654685974121094, -9.963737487792969, -4.3808135986328125, 18.094406127929688, 32.18926239013672, 28.921585083007812, -9.64011001586914, -9.247613906860352, 2.6367149353027344, 56.229576110839844, 31.16313934326172, 53.03221130371094, 3.8311100006103516, -7.719104766845703, 30.753631591796875, 0.3916168212890625, 12.714897155761719, -4.980926513671875, -12.690645217895508, 8.818138122558594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000233.npy"}
|
|
{"epoch": 0.35222978080120937, "step": 234, "batch_size": 64, "mean": 10.373629570007324, "std": 22.176694869995117, "min": -31.154539108276367, "p10": -16.69345245361328, "median": 5.06169319152832, "p90": 42.5477481842041, "max": 54.40357208251953, "pos_frac": 0.65625, "sample": [6.814170837402344, -25.61602783203125, 4.367012023925781, -14.856903076171875, 18.77875518798828, -2.2815589904785156, 2.930837631225586, -19.60222625732422, 30.845947265625, -12.750411987304688, 14.523231506347656, 16.880672454833984, 37.936241149902344, 15.671001434326172, 45.750244140625, 29.45382308959961, -15.245002746582031, -30.595890045166016, 27.65399169921875, -3.0526905059814453, 5.142974853515625, 2.6579113006591797, 43.14216232299805, 54.40357208251953, -0.7103023529052734, -7.3580780029296875, 42.550228118896484, -1.0345783233642578, -27.75713348388672, 6.03082275390625, 35.39331817626953, 47.0234489440918, 46.16819763183594, 38.35325622558594, -2.179471969604492, -31.154539108276367, -0.47308349609375, -0.511810302734375, 4.967802047729492, -14.72650146484375, 35.332279205322266, 27.105667114257812, 4.971363067626953, 28.967857360839844, 36.64324188232422, 3.3889808654785156, 4.173366546630859, 23.148529052734375, 42.541961669921875, -12.620031356811523, -6.147918701171875, 10.347444534301758, -17.31421661376953, 22.59176254272461, -24.185157775878906, 0.5066566467285156, 28.214599609375, 5.474479675292969, 1.14044189453125, 4.980411529541016, 11.397964477539062, 49.50151824951172, 28.51447296142578, -12.296806335449219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000234.npy"}
|
|
{"epoch": 0.35374149659863946, "step": 235, "batch_size": 64, "mean": 18.569324493408203, "std": 19.744070053100586, "min": -26.890422821044922, "p10": -3.406922912597656, "median": 17.640417098999023, "p90": 41.927540969848636, "max": 60.609771728515625, "pos_frac": 0.84375, "sample": [3.347614288330078, 19.8515625, 10.571464538574219, 19.278385162353516, 22.476715087890625, 14.500495910644531, 12.746559143066406, 60.609771728515625, 55.17485046386719, 33.36891174316406, 12.678741455078125, 31.40624237060547, 13.255241394042969, 22.59130859375, 40.824005126953125, 36.9490966796875, 23.84629249572754, 4.858303070068359, 33.78417205810547, 40.92428970336914, 6.4320220947265625, 7.854633331298828, 8.164962768554688, 38.74261474609375, 40.422630310058594, 29.162246704101562, -3.5625534057617188, -5.668254852294922, 46.39352798461914, -26.890422821044922, 31.615234375, 2.1959667205810547, 7.291788101196289, 19.888397216796875, 42.47974395751953, 8.210716247558594, 21.531070709228516, 4.661464691162109, 31.443695068359375, -23.157997131347656, -6.652864456176758, 38.7681884765625, 19.961883544921875, 8.183704376220703, 37.015220642089844, 3.368816375732422, -1.9939804077148438, 38.95905685424805, 58.079864501953125, 7.237527847290039, 10.245536804199219, -5.22381591796875, 48.637542724609375, 16.00244903564453, 2.8023300170898438, -25.239669799804688, 33.755409240722656, 36.999298095703125, 42.357505798339844, 1.7117843627929688, 25.91657257080078, -1.8929252624511719, 2.2255287170410156, -3.0437850952148438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000235.npy"}
|
|
{"epoch": 0.35525321239606955, "step": 236, "batch_size": 64, "mean": 14.157608032226562, "std": 23.216150283813477, "min": -50.2593994140625, "p10": -7.996516609191894, "median": 10.585113525390625, "p90": 46.60598602294922, "max": 66.44795227050781, "pos_frac": 0.75, "sample": [35.32591247558594, -4.313201904296875, 3.050079345703125, 34.21518325805664, 10.764551162719727, 4.340057373046875, 35.80003356933594, 23.884933471679688, -4.868438720703125, 3.9613990783691406, 22.463058471679688, 6.832794189453125, -6.498414993286133, 5.740692138671875, -8.336400985717773, 5.208686828613281, 31.859907150268555, 39.721160888671875, 23.269119262695312, 46.097755432128906, 0.15207481384277344, -1.6556777954101562, 5.339691162109375, -1.5406379699707031, 15.831756591796875, 11.420974731445312, 23.509445190429688, 35.829437255859375, 49.953983306884766, 29.632911682128906, 46.82379913330078, 0.4659881591796875, 3.0653533935546875, 11.182292938232422, 50.00877380371094, -13.962173461914062, 1.9592437744140625, 66.44795227050781, 6.253246307373047, 11.444887161254883, -26.90352439880371, 14.68850326538086, 36.966278076171875, 44.912689208984375, 20.609329223632812, 16.813091278076172, -1.0997772216796875, -9.260749816894531, 48.857120513916016, 6.000324249267578, 0.13341140747070312, -50.2593994140625, -46.5322265625, 50.47798538208008, 48.78529357910156, 42.382423400878906, 13.327499389648438, -13.149297714233398, -7.203453063964844, -6.503971099853516, 45.43611526489258, -1.00897216796875, 10.405675888061523, 7.530326843261719], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000236.npy"}
|
|
{"epoch": 0.35676492819349964, "step": 237, "batch_size": 64, "mean": 16.323091506958008, "std": 24.00326919555664, "min": -45.12897491455078, "p10": -9.787564086914061, "median": 15.48147201538086, "p90": 43.92091522216798, "max": 69.02294921875, "pos_frac": 0.78125, "sample": [35.42961120605469, 39.14155578613281, 10.707931518554688, 50.753273010253906, -1.676788330078125, -9.951507568359375, 11.388885498046875, 16.098007202148438, 9.985591888427734, 34.700775146484375, 22.221176147460938, 29.950149536132812, -9.405029296875, 5.3835296630859375, 35.0570068359375, 16.771949768066406, -2.346139907836914, 32.27936553955078, -22.164077758789062, 40.514137268066406, -45.12897491455078, 36.195220947265625, 18.015457153320312, 32.22700119018555, 58.06170654296875, -3.4466476440429688, 29.345062255859375, 37.87061309814453, -6.373382568359375, 10.936929702758789, 39.177284240722656, 9.307086944580078, 12.485916137695312, -43.701263427734375, 20.82642936706543, 6.004367828369141, 6.236000061035156, -42.80644989013672, 62.113006591796875, 4.338682174682617, 24.889888763427734, 17.33551788330078, 7.0754547119140625, 14.864936828613281, 6.775352478027344, 0.1732463836669922, 27.22981834411621, 27.4437255859375, 29.3165283203125, 44.788665771484375, 41.89616394042969, 10.767257690429688, 50.03455352783203, 69.02294921875, -25.544342041015625, 9.434818267822266, -2.0681705474853516, 37.55524444580078, 4.7295684814453125, 20.08818817138672, -3.418659210205078, 10.539634704589844, -13.537818908691406, 48.761898040771484], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000237.npy"}
|
|
{"epoch": 0.35827664399092973, "step": 238, "batch_size": 64, "mean": 16.444175720214844, "std": 22.955997467041016, "min": -34.02244567871094, "p10": -12.460615348815917, "median": 16.443819046020508, "p90": 47.09164924621583, "max": 56.24006652832031, "pos_frac": 0.734375, "sample": [-8.905864715576172, 6.761653900146484, 45.33281326293945, -12.926963806152344, 43.61248016357422, 54.96307373046875, 0.9238090515136719, 10.19183349609375, 44.20189666748047, -11.372468948364258, 15.526603698730469, -10.883834838867188, 47.94647979736328, 10.1884765625, 24.446496963500977, 17.361034393310547, -1.9034366607666016, -7.421232223510742, 28.085952758789062, 49.14014434814453, 35.008689880371094, 50.93879699707031, 35.81642150878906, 30.18643569946289, 5.408992767333984, 40.710426330566406, 23.30236053466797, -15.294868469238281, -20.003822326660156, 34.72966003417969, 37.47905349731445, 32.5731086730957, 3.0307674407958984, 40.383689880371094, -1.8871345520019531, 7.446868896484375, 27.655181884765625, -3.563631057739258, 44.8404541015625, 47.845436096191406, 4.808475494384766, -7.097496032714844, 0.7026634216308594, 21.4439697265625, 37.779273986816406, 13.603790283203125, 56.24006652832031, 31.696327209472656, 17.63149070739746, 19.890625, 6.23187255859375, 30.877700805664062, 43.438385009765625, -14.63254165649414, -9.579442977905273, 0.7661018371582031, -34.02244567871094, 6.158233642578125, 51.115501403808594, -9.088226318359375, 20.14312744140625, -15.837066650390625, -23.817333221435547, 2.098329544067383], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000238.npy"}
|
|
{"epoch": 0.35978835978835977, "step": 239, "batch_size": 64, "mean": 16.72553062438965, "std": 24.22995948791504, "min": -40.46702575683594, "p10": -11.335575103759766, "median": 16.182552337646484, "p90": 52.54068984985352, "max": 69.07136535644531, "pos_frac": 0.765625, "sample": [10.411827087402344, 8.906143188476562, 24.127227783203125, 0.4365425109863281, 55.05735778808594, 14.68585205078125, -7.132293701171875, 2.8184127807617188, -6.648994445800781, 13.113609313964844, -10.804801940917969, 16.82464599609375, -0.13821792602539062, 52.80854797363281, 38.70198059082031, 4.783203125, -14.789321899414062, -19.517908096313477, 16.2728271484375, 43.63287353515625, 3.6332168579101562, 14.256851196289062, 41.27111053466797, 53.43231964111328, 53.52372360229492, 12.464509963989258, 25.55451202392578, 31.98928451538086, 20.503406524658203, 12.771369934082031, 43.27846908569336, 17.59923553466797, -20.56462860107422, -10.242555618286133, 16.5040283203125, 43.810638427734375, -2.2957916259765625, 7.511930465698242, -11.56304931640625, 19.10626220703125, -2.2474536895751953, 16.352920532226562, 16.70232391357422, 14.587921142578125, -34.882362365722656, 25.798484802246094, 63.91155242919922, 65.3328857421875, 69.07136535644531, -28.96683120727539, 0.10367393493652344, 5.8981781005859375, -40.46702575683594, 43.43095397949219, 27.344879150390625, 28.810012817382812, 51.915687561035156, -2.0815391540527344, 16.846649169921875, 38.37727355957031, 3.912332534790039, 28.037887573242188, 30.45757293701172, 16.09227752685547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000239.npy"}
|
|
{"epoch": 0.36130007558578986, "step": 240, "batch_size": 64, "mean": 21.355527877807617, "std": 23.79172706604004, "min": -45.66255187988281, "p10": -6.707893753051757, "median": 22.27164649963379, "p90": 53.299285125732425, "max": 67.64164733886719, "pos_frac": 0.8125, "sample": [25.274925231933594, -8.112052917480469, 43.065826416015625, 2.7025146484375, 28.388694763183594, 67.64164733886719, 26.20465087890625, 34.02747344970703, -18.80596351623535, -11.887462615966797, 7.317741394042969, -2.604024887084961, 28.987144470214844, 26.918424606323242, 44.4366569519043, 56.72728729248047, 36.67597961425781, -12.159683227539062, 50.96404266357422, 28.633468627929688, -6.560005187988281, 37.13407897949219, 19.26799774169922, -21.761505126953125, 22.06268310546875, 44.57136535644531, 18.395736694335938, 41.31553649902344, 64.59849548339844, 23.431495666503906, 43.168312072753906, 60.7093505859375, 5.314674377441406, 11.45062255859375, 2.2698440551757812, 39.2921028137207, 12.20969009399414, 33.81121826171875, 47.54376220703125, 1.2133350372314453, 41.187408447265625, 7.794734954833984, -6.771274566650391, 31.177818298339844, 58.469398498535156, -45.66255187988281, -0.8018054962158203, 0.9629631042480469, 21.21257209777832, 4.9775390625, 22.480609893798828, 52.874725341796875, -1.182220458984375, 10.865501403808594, 0.9087409973144531, 8.759056091308594, 53.481239318847656, 56.3934326171875, 29.279464721679688, 8.882883071899414, 39.036460876464844, 3.9653396606445312, -2.7782745361328125, 17.403884887695312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000240.npy"}
|
|
{"epoch": 0.36281179138321995, "step": 241, "batch_size": 64, "mean": 13.068241119384766, "std": 23.529926300048828, "min": -39.170570373535156, "p10": -13.806573486328125, "median": 8.879974365234375, "p90": 43.61279296875, "max": 67.4107437133789, "pos_frac": 0.671875, "sample": [-14.888427734375, -14.262954711914062, 36.00102233886719, -15.259099960327148, 28.121421813964844, -12.741683959960938, 10.586570739746094, 36.17742156982422, 37.37701416015625, 36.89656066894531, -6.50244140625, 43.532684326171875, -8.58487319946289, 13.812784194946289, 3.2740249633789062, 47.490814208984375, 1.6294174194335938, 3.096792221069336, 0.6604766845703125, 41.16499328613281, 29.136093139648438, 22.722991943359375, -5.0404052734375, 7.511741638183594, 11.661338806152344, -3.631296157836914, 38.41561508178711, -0.8478469848632812, -4.588203430175781, -12.607513427734375, 44.09095001220703, 62.026214599609375, 59.57243347167969, 8.818695068359375, -3.4794883728027344, -25.32720947265625, -4.925971984863281, 24.405364990234375, 10.766021728515625, 5.212596893310547, -39.170570373535156, 1.5753097534179688, 0.11192131042480469, 44.94429016113281, 10.82213020324707, 18.500762939453125, -34.60337829589844, 67.4107437133789, 21.545989990234375, -20.608200073242188, 43.647125244140625, 14.954833984375, 33.60894012451172, 7.8387451171875, 42.824623107910156, -4.290920257568359, -1.0298080444335938, 37.900657653808594, 32.683555603027344, 23.507381439208984, 2.521089553833008, 8.941253662109375, -6.96031379699707, -1.7834148406982422], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000241.npy"}
|
|
{"epoch": 0.36432350718065004, "step": 242, "batch_size": 64, "mean": 19.037189483642578, "std": 26.442964553833008, "min": -49.123374938964844, "p10": -4.141569328308105, "median": 13.441652297973633, "p90": 55.34666442871095, "max": 88.99432373046875, "pos_frac": 0.8125, "sample": [-33.96868133544922, 13.60561752319336, 17.99927520751953, 4.435699462890625, 31.120018005371094, -23.906085968017578, 51.472129821777344, 36.383331298828125, 21.498245239257812, 37.56812286376953, 26.334121704101562, 35.494937896728516, -0.34956932067871094, 73.2744369506836, 51.29025650024414, -5.8477020263671875, 59.03634262084961, -1.2363662719726562, 5.385215759277344, 3.6963844299316406, 10.128303527832031, 32.41614532470703, 57.007179260253906, 3.953723907470703, -1.0507431030273438, 17.783966064453125, 8.64617919921875, 2.250274658203125, 39.04833984375, 30.647720336914062, 31.999130249023438, 0.6920261383056641, 69.75787353515625, 9.912368774414062, 36.82719039916992, 13.277687072753906, 0.029886245727539062, -0.4630928039550781, 88.99432373046875, 6.66253662109375, -24.24517822265625, 26.537826538085938, 28.79247283935547, 2.024829864501953, 25.49687957763672, 0.7867889404296875, 34.45018005371094, 12.633567810058594, 41.294647216796875, 47.57630157470703, 62.7320442199707, 28.13836097717285, 67.82528686523438, -4.638387680053711, 0.07292366027832031, 23.23871612548828, 35.60484313964844, 2.2258682250976562, 9.498908996582031, -49.123374938964844, 3.4960098266601562, -2.9823265075683594, -26.21210289001465, 11.348173141479492], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000242.npy"}
|
|
{"epoch": 0.36583522297808013, "step": 243, "batch_size": 64, "mean": 16.858591079711914, "std": 24.61781120300293, "min": -40.08001708984375, "p10": -14.716612243652339, "median": 13.708335876464844, "p90": 50.65325813293457, "max": 60.50007629394531, "pos_frac": 0.765625, "sample": [2.1411209106445312, 47.85594177246094, 38.86803436279297, -22.05072021484375, 47.50798034667969, 3.8932857513427734, -40.08001708984375, 1.6065902709960938, 37.8326416015625, -1.8699264526367188, 38.31224060058594, 18.311996459960938, 3.644927978515625, 10.39202880859375, 11.245613098144531, 59.962860107421875, -0.969329833984375, 3.2942123413085938, 50.68101119995117, -6.4309844970703125, 2.8850250244140625, 60.50007629394531, 22.09393310546875, -4.221454620361328, 30.902565002441406, 25.666282653808594, 4.560157775878906, 28.044164657592773, 47.91114807128906, 24.23493194580078, -26.487213134765625, -23.027454376220703, 29.281692504882812, -5.29747200012207, 46.283905029296875, 23.515884399414062, 42.65686798095703, 48.29846954345703, -22.4842529296875, 15.44879150390625, 58.59916687011719, 21.793601989746094, 20.871482849121094, 59.32195281982422, 10.548797607421875, -0.13977813720703125, 1.7272720336914062, 10.553152084350586, 50.92053985595703, 26.497962951660156, 1.5481414794921875, 11.967880249023438, 4.053489685058594, 50.5885009765625, -16.422706604003906, 30.044967651367188, 15.838584899902344, 5.979351043701172, -10.735725402832031, -7.555511474609375, -23.806617736816406, 56.81846618652344, 6.6936492919921875, 18.327667236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000243.npy"}
|
|
{"epoch": 0.3673469387755102, "step": 244, "batch_size": 64, "mean": 22.894798278808594, "std": 24.179794311523438, "min": -45.665077209472656, "p10": -0.19172973632812448, "median": 18.413341522216797, "p90": 55.96593246459961, "max": 68.21266174316406, "pos_frac": 0.890625, "sample": [8.568626403808594, 22.483924865722656, 18.320327758789062, 18.50635528564453, -3.5578155517578125, 17.22771644592285, -2.4009227752685547, 53.45586395263672, 24.3601016998291, -15.20806884765625, 11.419925689697266, 44.26823043823242, 7.38226318359375, 48.15516662597656, 20.688194274902344, 5.0479736328125, 33.917449951171875, 57.74645233154297, -45.665077209472656, 49.66832733154297, 17.081192016601562, 11.452522277832031, -31.655189514160156, 37.36622619628906, 9.049812316894531, 48.193504333496094, 33.27410888671875, 49.57301330566406, 55.98992156982422, 10.656923294067383, 62.66685485839844, 19.443458557128906, 0.32717132568359375, -0.41411590576171875, 3.5438461303710938, 9.59282112121582, 0.4356250762939453, 42.68053436279297, 55.10455322265625, 68.21266174316406, 2.6143245697021484, 56.4161376953125, 1.1200084686279297, 8.972366333007812, 3.1265945434570312, 60.14475631713867, 22.976181030273438, 44.30253601074219, 3.2783584594726562, 15.08914566040039, 2.5295066833496094, 12.390541076660156, 53.359657287597656, 19.915115356445312, 59.17173767089844, 26.76923370361328, -1.1505355834960938, 18.704635620117188, 6.123912811279297, 6.048332214355469, 41.714012145996094, 13.288381576538086, 55.49183654785156, 55.90995788574219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000244.npy"}
|
|
{"epoch": 0.3688586545729403, "step": 245, "batch_size": 64, "mean": 25.920265197753906, "std": 25.090444564819336, "min": -39.62260437011719, "p10": -1.8270004272460934, "median": 26.357423782348633, "p90": 59.70992660522462, "max": 74.52339935302734, "pos_frac": 0.875, "sample": [33.045326232910156, 32.92626953125, 32.9984016418457, 61.464569091796875, 17.262449264526367, 4.282745361328125, 2.9395179748535156, 53.910133361816406, 40.890464782714844, 33.649330139160156, 2.006336212158203, 15.865325927734375, 48.89012908935547, 29.856327056884766, 28.279029846191406, 55.719268798828125, 4.549385070800781, 47.33721160888672, -19.388442993164062, 74.52339935302734, 68.24420928955078, 3.0939865112304688, -1.985565185546875, 30.2337589263916, 46.43305587768555, 25.11212158203125, 20.84765625, -2.24786376953125, 44.69811248779297, 0.4587898254394531, 50.498748779296875, -25.127458572387695, 0.9168014526367188, 32.17976379394531, 7.955802917480469, 39.39160919189453, -1.4570159912109375, 16.011110305786133, 39.636722564697266, 69.21935272216797, 24.456459045410156, 1.7087249755859375, 65.7560806274414, 6.480339050292969, -3.2362308502197266, 23.460594177246094, 57.817604064941406, 60.263648986816406, 1.9896392822265625, 6.5603790283203125, 36.17573547363281, 45.811283111572266, -11.391082763671875, 41.26306915283203, 63.23236083984375, 23.134754180908203, -39.62260437011719, 58.41790771484375, 21.189929962158203, 9.232709884643555, 7.8539886474609375, 48.59771728515625, 17.02042007446289, 27.602725982666016], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000245.npy"}
|
|
{"epoch": 0.37037037037037035, "step": 246, "batch_size": 64, "mean": 20.76578140258789, "std": 32.07914352416992, "min": -52.30419921875, "p10": -14.194102287292477, "median": 14.460174560546875, "p90": 59.20586051940918, "max": 91.1109619140625, "pos_frac": 0.765625, "sample": [46.789180755615234, 53.35566711425781, 53.993438720703125, 55.88899612426758, 9.878555297851562, 4.1727752685546875, -47.82339096069336, 57.11824035644531, 19.70661163330078, -10.877985000610352, 49.69914245605469, 59.209320068359375, 28.748092651367188, -4.724973678588867, 40.575439453125, 41.39010238647461, 14.951896667480469, -5.083873748779297, -7.994842529296875, -2.2466278076171875, 11.4158935546875, 41.1400146484375, 1.4355087280273438, 59.19778823852539, 72.5410385131836, -31.807052612304688, 17.886062622070312, 91.1109619140625, -15.824752807617188, 7.89385986328125, 7.288515090942383, 56.319358825683594, 13.374160766601562, 7.382345199584961, -15.61529541015625, 32.18339538574219, -50.478363037109375, 11.619792938232422, 65.6952896118164, -0.4388580322265625, 30.416259765625, -40.25542449951172, -10.350990295410156, 59.250892639160156, 26.22885513305664, 0.938232421875, 37.34181213378906, 52.45909881591797, 61.70074462890625, -52.30419921875, 3.6399688720703125, 13.968452453613281, 37.53941345214844, 78.25945281982422, 8.52994155883789, 0.57720947265625, 51.105308532714844, 54.26002502441406, -3.091888427734375, 0.34381103515625, 24.466785430908203, 12.164291381835938, 42.02959442138672, 0.7469482421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000246.npy"}
|
|
{"epoch": 0.37188208616780044, "step": 247, "batch_size": 64, "mean": 12.279106140136719, "std": 26.714599609375, "min": -57.255714416503906, "p10": -24.157194519042964, "median": 12.306970596313477, "p90": 43.75331115722657, "max": 88.59149169921875, "pos_frac": 0.71875, "sample": [-28.683624267578125, 18.250568389892578, 22.080215454101562, 11.627632141113281, 3.8651809692382812, -10.809112548828125, 1.988739013671875, 57.89884948730469, 61.78086471557617, -4.521514892578125, 35.51130676269531, 22.927764892578125, 44.5592041015625, 25.43169593811035, 10.289443969726562, 41.23773193359375, 39.672882080078125, 19.156917572021484, -5.234657287597656, -35.2718505859375, 88.59149169921875, 50.38804626464844, -40.5196533203125, -9.280923843383789, 3.4939422607421875, 8.745323181152344, 29.063098907470703, -3.16064453125, -26.2523193359375, 32.57289123535156, 22.208404541015625, 14.339614868164062, 17.057403564453125, 3.967742919921875, 9.7254638671875, 9.152603149414062, 37.10102844238281, -57.255714416503906, -19.268569946289062, 0.27372169494628906, -1.38189697265625, 8.356729507446289, 0.56524658203125, 49.726539611816406, 41.872894287109375, 14.099079132080078, 2.4596214294433594, -14.218162536621094, 12.986309051513672, 31.02337646484375, 15.726669311523438, 29.999372482299805, -13.047210693359375, 13.671609878540039, -42.20555114746094, 29.020780563354492, 36.14985656738281, 46.3150634765625, -3.292999267578125, -26.610366821289062, -9.953170776367188, 32.92926025390625, 18.08441162109375, 10.884140014648438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000247.npy"}
|
|
{"epoch": 0.37339380196523053, "step": 248, "batch_size": 64, "mean": 21.386653900146484, "std": 32.434974670410156, "min": -38.772647857666016, "p10": -21.44198360443115, "median": 14.19005298614502, "p90": 62.43144836425782, "max": 88.57694244384766, "pos_frac": 0.71875, "sample": [5.9568328857421875, 42.42463684082031, -33.23313903808594, -32.50529479980469, 76.78675079345703, 57.46852111816406, 56.68006896972656, 42.958030700683594, -1.4273929595947266, -13.013280868530273, 1.1965103149414062, 58.47520446777344, 20.79198455810547, -21.465478897094727, 61.88134765625, 27.468185424804688, 67.54400634765625, -10.036033630371094, -35.32025909423828, 17.33246612548828, -23.893447875976562, 4.974723815917969, 11.394889831542969, 7.571691513061523, 59.53401184082031, 38.09196090698242, 52.49220275878906, 15.166969299316406, 30.679981231689453, 62.667205810546875, -29.24384307861328, 59.39952087402344, -0.8306217193603516, -7.82581901550293, 61.77963638305664, 56.30879211425781, 8.592605590820312, 13.224052429199219, 15.15605354309082, 22.520355224609375, 43.074859619140625, 49.93638610839844, 10.833965301513672, 73.94868469238281, 72.13580322265625, 21.098834991455078, 3.811094284057617, -1.1844444274902344, 6.085676193237305, 44.369415283203125, 88.57694244384766, 6.938220977783203, 37.97819519042969, 5.709480285644531, -38.772647857666016, 8.929452896118164, -8.048553466796875, 66.44832611083984, 58.659523010253906, -3.4309005737304688, -1.8058452606201172, 4.2228240966796875, -21.387161254882812, -7.106788635253906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000248.npy"}
|
|
{"epoch": 0.3749055177626606, "step": 249, "batch_size": 64, "mean": 8.998100280761719, "std": 29.92830467224121, "min": -69.69783020019531, "p10": -30.601294708251952, "median": 10.096494674682617, "p90": 48.2038387298584, "max": 65.43171691894531, "pos_frac": 0.703125, "sample": [17.35357666015625, -25.0118465423584, 6.418731689453125, 30.611175537109375, 9.863548278808594, -19.936050415039062, -38.646392822265625, 23.98223876953125, 20.167999267578125, 33.11273193359375, 19.81036376953125, 43.184173583984375, 1.0315704345703125, 46.03404998779297, -50.873207092285156, -17.113128662109375, 18.074478149414062, -30.219139099121094, -35.59942626953125, -21.63079833984375, 5.012643814086914, -3.5285797119140625, -27.245361328125, 9.404455184936523, 23.831809997558594, 13.361701965332031, 0.36413002014160156, -8.626144409179688, 65.43171691894531, -16.98358154296875, -12.492080688476562, 49.191627502441406, 7.488910675048828, 19.152536392211914, 41.74982833862305, -45.27471923828125, 10.506813049316406, 2.294708251953125, 12.210973739624023, -30.76507568359375, 6.29150390625, 57.302433013916016, 48.31201171875, 59.322174072265625, 20.919357299804688, 10.32944107055664, -35.498382568359375, 34.09120559692383, 23.022796630859375, 8.623626708984375, 21.785608291625977, 47.95143508911133, 30.442825317382812, 1.4287109375, 16.74906349182129, 12.894012451171875, -12.00444221496582, 45.35497283935547, 64.85215759277344, -69.69783020019531, 55.9399299621582, -26.413721084594727, 1.127349853515625, 7.051246643066406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000249.npy"}
|
|
{"epoch": 0.3764172335600907, "step": 250, "batch_size": 64, "mean": 22.645402908325195, "std": 30.161701202392578, "min": -76.26980590820312, "p10": -5.193807220458983, "median": 18.29093074798584, "p90": 61.55714645385742, "max": 83.5948486328125, "pos_frac": 0.8125, "sample": [30.49753761291504, 16.519575119018555, -3.512500762939453, 12.182769775390625, 1.92755126953125, 33.39772033691406, 34.706336975097656, 74.31358337402344, 61.25402069091797, 2.3848876953125, 42.45485305786133, -16.594375610351562, 43.071380615234375, 10.951560974121094, 39.1236572265625, 1.2066879272460938, 1.1322021484375, -76.26980590820312, 51.136016845703125, 7.580787658691406, 27.81007194519043, -7.916784286499023, 14.773307800292969, -34.935272216796875, 50.2740478515625, 1.341339111328125, 65.78752136230469, 26.89252471923828, 4.7204437255859375, 28.999801635742188, 0.5600700378417969, 34.173858642578125, 61.68705749511719, -0.03241539001464844, 28.33356475830078, 6.6327972412109375, 0.016126632690429688, 45.17844009399414, 80.1082534790039, 26.243820190429688, -37.08165740966797, -3.803569793701172, 30.552082061767578, 2.2705154418945312, 76.07139587402344, 39.565250396728516, 51.877197265625, 45.41447067260742, 61.92644119262695, 51.50442123413086, -3.2738800048828125, -9.463996887207031, 53.8984375, 3.72088623046875, 4.742034912109375, 11.403572082519531, 83.5948486328125, 53.60332489013672, 16.342395782470703, -0.19621658325195312, -5.789623260498047, 58.324851989746094, 15.927278518676758, 20.062286376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000250.npy"}
|
|
{"epoch": 0.3779289493575208, "step": 251, "batch_size": 64, "mean": 14.196611404418945, "std": 30.2620849609375, "min": -59.377410888671875, "p10": -25.95013732910156, "median": 14.816108703613281, "p90": 49.76530075073243, "max": 81.73726654052734, "pos_frac": 0.71875, "sample": [2.359243392944336, 72.67818450927734, 73.961181640625, -0.9703502655029297, 24.559677124023438, 30.220016479492188, -59.377410888671875, 18.322826385498047, 35.99122619628906, 81.73726654052734, 1.8560714721679688, -45.05529022216797, 14.319564819335938, -14.700981140136719, 7.639148712158203, -16.798240661621094, 24.95416259765625, 65.20004272460938, 8.753423690795898, -48.86012268066406, -17.71862030029297, 67.96121978759766, 0.020914077758789062, -31.17669677734375, 14.259269714355469, -8.954200744628906, 20.895118713378906, 41.085174560546875, -10.897109985351562, 2.551025390625, 48.31568145751953, 40.81431579589844, 7.097553253173828, 15.669538497924805, 29.87029266357422, 16.901363372802734, 33.84575653076172, 24.950668334960938, -28.03356170654297, -1.154825210571289, 43.174556732177734, 1.579498291015625, 46.27378845214844, 18.861858367919922, 11.386802673339844, 24.688873291015625, 15.312652587890625, 12.776473999023438, 21.570022583007812, 38.526824951171875, 52.447662353515625, 5.1988677978515625, 44.702144622802734, -17.08924102783203, -6.689506530761719, -1.4200592041015625, 36.3968620300293, -33.693695068359375, 17.088735580444336, -37.88172149658203, -21.08881378173828, 29.772308349609375, 50.386566162109375, 13.209165573120117], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000251.npy"}
|
|
{"epoch": 0.3794406651549509, "step": 252, "batch_size": 64, "mean": 13.216939926147461, "std": 33.249908447265625, "min": -76.98199462890625, "p10": -27.715672874450682, "median": 10.951376914978027, "p90": 56.35322837829591, "max": 84.32138061523438, "pos_frac": 0.703125, "sample": [25.394393920898438, -63.80290222167969, -9.109186172485352, -56.81134033203125, 41.4652099609375, -6.184242248535156, 10.75424575805664, 44.86993408203125, 1.757986068725586, 65.5831527709961, 58.78816223144531, 30.833690643310547, 13.564140319824219, 9.019584655761719, 75.40715789794922, -8.075454711914062, 70.52632141113281, -41.81300354003906, 14.107856750488281, -0.6852874755859375, -34.86005401611328, 11.148508071899414, 29.754348754882812, 11.156156539916992, 19.893447875976562, 6.598670959472656, 9.423309326171875, -9.590003967285156, 53.13480758666992, -0.4671745300292969, 30.955747604370117, 82.25624084472656, 47.83308029174805, 9.91033935546875, 26.61322021484375, 5.273876190185547, 17.087677001953125, 15.217977523803711, 23.965707778930664, -17.72121810913086, 24.714893341064453, -13.630874633789062, -0.0711517333984375, 39.9015998840332, 6.612518310546875, 84.32138061523438, -28.439598083496094, 1.2470016479492188, -76.98199462890625, -45.59092330932617, -10.547351837158203, 9.255828857421875, 38.05780792236328, -26.026514053344727, 10.675483703613281, 57.73255157470703, 8.288375854492188, 50.05860137939453, 14.850608825683594, 26.320629119873047, 43.01397705078125, 25.274063110351562, -13.100202560424805, 6.7723388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000252.npy"}
|
|
{"epoch": 0.38095238095238093, "step": 253, "batch_size": 64, "mean": 16.497447967529297, "std": 28.568967819213867, "min": -44.92769241333008, "p10": -17.59422149658203, "median": 14.914116859436035, "p90": 55.10547828674317, "max": 77.01600646972656, "pos_frac": 0.703125, "sample": [38.21239471435547, -2.2344284057617188, 67.25994873046875, 2.885467529296875, -1.1626472473144531, 19.83572006225586, 18.656051635742188, 49.410884857177734, 40.96612548828125, 21.001707077026367, 38.27013397216797, -5.907524108886719, 41.49578094482422, 4.0361480712890625, -1.5599117279052734, -42.66775131225586, -16.967445373535156, 23.97904396057129, -17.106658935546875, 37.34712219238281, 41.482025146484375, 34.51605224609375, -44.92769241333008, 37.42289733886719, 17.47920036315918, 8.5841064453125, 55.80250930786133, 22.60894012451172, 0.5768661499023438, 29.740676879882812, 8.385360717773438, 7.379524230957031, 47.711647033691406, -3.6556663513183594, 7.9259033203125, 77.01600646972656, 26.767196655273438, -44.21159362792969, 14.008926391601562, 26.68242835998535, 7.4909515380859375, -8.538604736328125, 17.266754150390625, -7.933694839477539, 60.52960968017578, -16.294544219970703, -8.733970642089844, 24.703659057617188, 69.96228790283203, 15.624456405639648, 13.976852416992188, 63.91728210449219, 75.33504486083984, 6.614250183105469, 53.47907257080078, -18.212966918945312, 25.946311950683594, -22.698923110961914, 39.498252868652344, -26.802947998046875, -17.803176879882812, -2.9129085540771484, 10.174365997314453, 14.203777313232422], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000253.npy"}
|
|
{"epoch": 0.382464096749811, "step": 254, "batch_size": 64, "mean": 17.936750411987305, "std": 29.45363426208496, "min": -49.81974792480469, "p10": -11.951998329162597, "median": 11.760709762573242, "p90": 62.62532196044923, "max": 76.6537094116211, "pos_frac": 0.65625, "sample": [-6.025390625, 17.999008178710938, 68.45962524414062, 14.044769287109375, -4.6153411865234375, 38.933990478515625, 43.47342300415039, 53.29985046386719, 2.7688465118408203, 0.064208984375, -6.517242431640625, 10.854259490966797, 37.33006286621094, 45.003021240234375, 51.15751647949219, 47.68648147583008, 67.47689819335938, 18.604026794433594, -14.833824157714844, 39.246337890625, -2.7789688110351562, 76.6537094116211, -18.1181640625, 64.13607025146484, -7.746376037597656, -49.81974792480469, -13.5714111328125, 63.508934020996094, 22.991727828979492, -1.8138961791992188, 74.16152954101562, -6.15313720703125, 8.562973022460938, -1.0539894104003906, -3.5080299377441406, 13.162139892578125, 3.658672332763672, 47.293121337890625, 0.8276710510253906, 27.384300231933594, 9.004837036132812, -5.776294708251953, 17.136058807373047, -20.906593322753906, -12.180959701538086, 68.75142669677734, -0.5704536437988281, 60.563560485839844, 28.1268310546875, 16.002490997314453, 12.667160034179688, -8.12109375, -8.777961730957031, 43.2476806640625, 59.136962890625, -8.222679138183594, 23.239059448242188, 3.5490665435791016, 51.61181640625, 42.0755615234375, 6.311479568481445, -11.417755126953125, 1.6977157592773438, -41.38356018066406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000254.npy"}
|
|
{"epoch": 0.3839758125472411, "step": 255, "batch_size": 64, "mean": 22.800310134887695, "std": 32.863258361816406, "min": -50.975399017333984, "p10": -10.750941467285156, "median": 13.621794700622559, "p90": 73.28849563598634, "max": 91.72244262695312, "pos_frac": 0.75, "sample": [68.40701293945312, 25.556365966796875, 91.72244262695312, 30.998992919921875, 52.585784912109375, 16.5433349609375, -4.688362121582031, -7.089679718017578, 8.565288543701172, 48.69915771484375, -32.217559814453125, -0.8736553192138672, 15.48263931274414, 52.29736328125, -17.149513244628906, 8.020748138427734, 35.39575958251953, 12.675483703613281, 8.487167358398438, -49.79477310180664, 62.202388763427734, 2.7765426635742188, 14.326675415039062, 10.684822082519531, 73.47311401367188, 52.59058380126953, -18.912086486816406, 10.970409393310547, 76.52396392822266, -8.22674560546875, 72.85771942138672, 46.27558135986328, -10.9669189453125, 21.733089447021484, 12.73297119140625, 12.916913986206055, 10.126970291137695, 49.00787353515625, 77.29895782470703, 12.519887924194336, 3.2134532928466797, 45.454254150390625, 61.475830078125, -10.246994018554688, 9.36955451965332, -2.972391128540039, 17.92406463623047, 84.42178344726562, -29.60199737548828, 25.12889862060547, 1.8014755249023438, 74.3032455444336, 12.369590759277344, 77.75021362304688, 25.901382446289062, 35.92889404296875, 2.7206649780273438, 40.85398483276367, 60.765869140625, -50.975399017333984, -2.08544921875, -3.9827651977539062, 36.526336669921875, -1.3613662719726562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000255.npy"}
|
|
{"epoch": 0.3854875283446712, "step": 256, "batch_size": 64, "mean": 20.204849243164062, "std": 33.94083786010742, "min": -65.42902374267578, "p10": -15.927235412597655, "median": 16.782726287841797, "p90": 67.03950119018555, "max": 90.06939697265625, "pos_frac": 0.6875, "sample": [-14.322319030761719, 14.640884399414062, 40.96416473388672, 26.93552017211914, 50.832855224609375, -18.83971405029297, -13.182188034057617, 70.47322082519531, 25.163976669311523, 2.704660415649414, -26.7203369140625, 3.251758575439453, -16.4451904296875, 81.45248413085938, 12.100799560546875, 75.52165985107422, -0.91357421875, 8.8643798828125, -65.42902374267578, 42.474388122558594, 1.2630271911621094, 47.14764404296875, -10.354291915893555, 14.3780517578125, -19.738540649414062, -50.707672119140625, 21.58446502685547, -7.484443664550781, 18.92456817626953, 44.62982940673828, 0.2541351318359375, 31.591796875, 27.78015899658203, 62.26632308959961, 61.438270568847656, 33.05931854248047, 67.06361389160156, 52.2449951171875, -7.1346435546875, 70.35182189941406, 13.656990051269531, 23.18424415588379, 80.38189697265625, -10.026031494140625, 64.49878692626953, 10.452346801757812, 52.13787841796875, 3.17034912109375, 27.029516220092773, 90.06939697265625, -13.078689575195312, 54.495323181152344, -0.70123291015625, 5.3934783935546875, 43.86700439453125, -11.336658477783203, 38.58453369140625, -7.696586608886719, -1.7465457916259766, 66.98323822021484, -35.71122741699219, 21.030471801757812, -14.718673706054688, 35.103729248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000256.npy"}
|
|
{"epoch": 0.3869992441421013, "step": 257, "batch_size": 64, "mean": 23.807159423828125, "std": 35.53892135620117, "min": -63.551795959472656, "p10": -9.928017425537107, "median": 12.835681915283203, "p90": 76.29552536010742, "max": 111.9228515625, "pos_frac": 0.734375, "sample": [-0.93231201171875, 7.572845458984375, -11.248443603515625, 111.9228515625, 81.18328857421875, 30.836437225341797, 1.3307743072509766, 45.630126953125, 14.723167419433594, 47.07341384887695, 76.99019622802734, 9.176795959472656, 4.428400039672852, -5.098548889160156, 54.864410400390625, 57.63714599609375, 72.27058410644531, -0.583709716796875, -25.68646240234375, -4.291435241699219, -0.08758354187011719, 28.509780883789062, 31.0452880859375, 7.1333770751953125, 22.552284240722656, 0.119415283203125, -0.6280250549316406, 74.61334991455078, 51.59381866455078, -1.598175048828125, 57.28438949584961, 76.21524047851562, 21.202957153320312, 4.274147033691406, 24.88764190673828, 14.460258483886719, -3.73046875, 67.53768157958984, 90.50665283203125, -41.25425338745117, -63.551795959472656, 89.850830078125, 19.90509796142578, 21.76793670654297, 3.078033447265625, -15.187999725341797, 9.385704040527344, 2.6646575927734375, 28.648780822753906, 36.72692108154297, 50.37513732910156, 5.928962707519531, 76.3299331665039, 8.91824722290039, -32.19189453125, 79.83792114257812, -3.3236541748046875, -20.71691131591797, 72.16845703125, 11.036300659179688, 34.480587005615234, 10.72552490234375, 11.211105346679688, -6.847023010253906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000257.npy"}
|
|
{"epoch": 0.3885109599395314, "step": 258, "batch_size": 64, "mean": 19.57223892211914, "std": 25.227291107177734, "min": -44.25098419189453, "p10": -1.4042623519897437, "median": 18.153793334960938, "p90": 52.76585235595705, "max": 81.488525390625, "pos_frac": 0.890625, "sample": [25.35211181640625, 4.698219299316406, 22.128341674804688, -2.407163619995117, 7.055816650390625, 66.17491149902344, 79.45222473144531, 28.2994384765625, 40.59562683105469, 3.830533981323242, 13.679571151733398, 12.854057312011719, -35.41435241699219, 21.711267471313477, 19.319259643554688, 46.950531005859375, 8.119354248046875, -25.80398941040039, 27.130577087402344, -44.25098419189453, 0.9358406066894531, 30.620506286621094, -31.459762573242188, 18.761634826660156, 12.02325439453125, 27.308753967285156, 18.888763427734375, 30.835922241210938, 10.314769744873047, 12.072257995605469, 16.54913330078125, 5.385746002197266, 35.721519470214844, 41.562599182128906, 15.265552520751953, 6.393608093261719, 7.912147521972656, 10.360054016113281, 17.547515869140625, 81.488525390625, 29.671295166015625, -8.491119384765625, 69.51585388183594, 33.183349609375, 27.31109619140625, 64.32670593261719, 40.122955322265625, 34.415096282958984, 3.6669998168945312, 55.25813293457031, 2.0929336547851562, 9.118927001953125, 27.142906188964844, 18.76007080078125, 10.814277648925781, 6.733968734741211, 26.492218017578125, -37.85395812988281, 8.17120361328125, 2.086984634399414, 28.422134399414062, 41.97649002075195, 58.821197509765625, 12.899785995483398], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000258.npy"}
|
|
{"epoch": 0.3900226757369615, "step": 259, "batch_size": 64, "mean": 29.120864868164062, "std": 33.70142364501953, "min": -44.3692626953125, "p10": -21.223809432983384, "median": 34.6446590423584, "p90": 75.83728790283203, "max": 91.77462768554688, "pos_frac": 0.8125, "sample": [-0.65081787109375, -35.567474365234375, -44.3692626953125, 52.817989349365234, 27.60894775390625, 55.58723449707031, 77.07527923583984, 62.10420227050781, 36.920692443847656, -42.44709777832031, 46.904197692871094, 0.6484832763671875, 20.035661697387695, 56.85272216796875, 45.70009994506836, -1.871673583984375, -28.72715187072754, 1.2554779052734375, 76.11212158203125, 27.799530029296875, 36.02504348754883, 54.26495361328125, 60.684078216552734, 71.248046875, 42.82416534423828, -6.551807403564453, -27.511810302734375, 14.719051361083984, 7.8079681396484375, 13.004793167114258, 53.54130172729492, 46.004638671875, -5.737331390380859, 19.35407257080078, 37.31568908691406, 31.928932189941406, 87.52812957763672, -3.197601318359375, 57.14642333984375, 51.04252624511719, 75.49853515625, 34.93632125854492, 7.815757751464844, 91.77462768554688, 7.461174011230469, 7.25103759765625, -29.984840393066406, 0.2541675567626953, -34.840293884277344, 1.6394767761230469, 80.04833984375, 75.98246765136719, 46.37031555175781, 34.352996826171875, 42.71058654785156, 52.36815643310547, 23.092254638671875, 25.370628356933594, 10.997451782226562, 30.563438415527344, 45.79130554199219, 77.65962982177734, 41.57268524169922, 39.81878662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000259.npy"}
|
|
{"epoch": 0.3915343915343915, "step": 260, "batch_size": 64, "mean": 23.009288787841797, "std": 35.38461685180664, "min": -54.62413787841797, "p10": -21.08153190612793, "median": 20.816265106201172, "p90": 70.57332077026368, "max": 83.50257110595703, "pos_frac": 0.703125, "sample": [70.98331451416016, -32.039772033691406, 53.85075759887695, -14.27457046508789, -3.5628089904785156, 81.60597229003906, 28.79729461669922, 19.797752380371094, 73.22848510742188, -35.212982177734375, 61.6339111328125, 38.941619873046875, 60.328826904296875, 47.10585021972656, 52.84922790527344, -22.78424835205078, 69.61666870117188, 45.221839904785156, 15.572935104370117, 83.50257110595703, 54.42725372314453, 4.670379638671875, 72.39825439453125, 74.6930160522461, -4.168983459472656, 10.536754608154297, 66.41578674316406, 67.03057861328125, 1.9278945922851562, -4.6603240966796875, 13.130775451660156, 14.154998779296875, 60.04943084716797, 21.83477783203125, 1.9834461212158203, 34.238773345947266, 11.798530578613281, -15.718414306640625, -54.62413787841797, 22.493316650390625, 2.6645851135253906, 39.2252197265625, -0.4118366241455078, -3.6087493896484375, 16.983840942382812, 42.277557373046875, 80.5930404663086, -13.263671875, 65.65008544921875, -17.469375610351562, -36.81094741821289, 31.646059036254883, 29.149864196777344, 9.3240966796875, 46.342384338378906, 31.751365661621094, -16.766357421875, 54.48167419433594, 46.147216796875, -22.12010955810547, -0.028759002685546875, -42.44884490966797, -18.658184051513672, 0.16968536376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000260.npy"}
|
|
{"epoch": 0.3930461073318216, "step": 261, "batch_size": 64, "mean": 19.59906768798828, "std": 31.07403564453125, "min": -39.674400329589844, "p10": -21.00532913208007, "median": 15.882806777954102, "p90": 68.05322647094727, "max": 89.27574920654297, "pos_frac": 0.71875, "sample": [1.0931205749511719, 14.521759033203125, 26.166860580444336, 0.8788509368896484, -12.587966918945312, 46.35206604003906, -1.6391868591308594, 78.44232177734375, 23.703449249267578, -4.107736587524414, -23.4061279296875, -10.412948608398438, -4.671379089355469, 70.71705627441406, 41.862510681152344, -8.711593627929688, -0.745941162109375, 81.82760620117188, 7.867181777954102, 68.99874877929688, 5.574127197265625, 21.121864318847656, -9.925304412841797, -24.538360595703125, 59.909515380859375, 30.17749786376953, 89.27574920654297, 17.243854522705078, -0.32083892822265625, 25.941190719604492, 34.99112319946289, 42.38945770263672, 32.337684631347656, 65.27861022949219, 23.8841609954834, 4.887781143188477, 69.2156982421875, -32.62837219238281, -24.97003173828125, 46.311492919921875, 26.19745635986328, 6.287542343139648, 47.29418182373047, 43.87353515625, 8.708843231201172, -39.674400329589844, 2.56378173828125, -26.883865356445312, -29.304908752441406, 8.26449966430664, 18.653724670410156, -3.1999664306640625, 54.29376220703125, 67.5803451538086, 13.854652404785156, -15.403465270996094, 27.856204986572266, 17.35137176513672, 1.6849365234375, 68.25588989257812, 11.696243286132812, 50.28760528564453, 18.445663452148438, 3.3511219024658203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000261.npy"}
|
|
{"epoch": 0.3945578231292517, "step": 262, "batch_size": 64, "mean": 21.239696502685547, "std": 30.793964385986328, "min": -41.60993194580078, "p10": -18.405171203613282, "median": 17.094959259033203, "p90": 60.282799911499026, "max": 84.91497802734375, "pos_frac": 0.78125, "sample": [-30.70520782470703, 37.249244689941406, 14.232192993164062, 1.7847576141357422, -4.130828857421875, 57.76668167114258, 9.384719848632812, 3.58856201171875, 17.811729431152344, 72.40460205078125, -7.437660217285156, 55.09309387207031, -23.435287475585938, 3.4997081756591797, 47.42703628540039, -29.867610931396484, 24.714609146118164, 69.91058349609375, 31.552032470703125, 39.9252815246582, -41.60993194580078, -1.8650283813476562, 5.722358703613281, 6.411376953125, 26.33051300048828, 59.55952835083008, 6.862579345703125, 84.91497802734375, -28.242645263671875, 6.341152191162109, 59.26294708251953, 41.275535583496094, 29.86780548095703, 49.925437927246094, -13.304313659667969, 68.50102996826172, 6.33159065246582, 33.39067459106445, 76.56143188476562, 3.218017578125, -12.388387680053711, -12.76129150390625, -18.40777587890625, 47.71009826660156, 55.6312255859375, 26.8619441986084, 31.02373504638672, 5.676233291625977, -27.936874389648438, 19.691532135009766, 42.43492126464844, 4.12286376953125, 58.12554931640625, 1.3040485382080078, 32.27008056640625, 16.378189086914062, 60.5927734375, 60.99513244628906, 2.8202972412109375, 47.82923889160156, 9.282943725585938, 1.9162673950195312, 54.34368133544922, -18.399093627929688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000262.npy"}
|
|
{"epoch": 0.3960695389266818, "step": 263, "batch_size": 64, "mean": 18.507747650146484, "std": 31.24783706665039, "min": -61.93293762207031, "p10": -15.117512512207028, "median": 15.015304565429688, "p90": 57.320220947265625, "max": 78.21243286132812, "pos_frac": 0.703125, "sample": [25.418209075927734, 44.9893684387207, -31.312896728515625, -8.291877746582031, -7.14109992980957, -5.356880187988281, 47.259002685546875, -2.993743896484375, 48.554290771484375, 57.50074005126953, 32.10308837890625, 2.3964481353759766, -11.87021255493164, 41.608272552490234, 4.2184295654296875, -40.53124237060547, 56.46098327636719, -0.771270751953125, 12.231521606445312, -10.659126281738281, 38.04083251953125, 49.49195861816406, 77.06367492675781, 12.647083282470703, 1.7316131591796875, 2.0865726470947266, 63.545875549316406, 33.76029968261719, 56.899009704589844, 51.286354064941406, 76.7850341796875, 15.382568359375, -18.230178833007812, 14.648040771484375, -8.538970947265625, 22.62128448486328, -54.19474792480469, 51.85002517700195, 21.126968383789062, 54.33576965332031, 16.61749267578125, -7.442420959472656, -10.877792358398438, 59.741859436035156, -17.723350524902344, 5.18511962890625, 28.913970947265625, 47.598289489746094, 40.94806671142578, 14.23614501953125, 2.9023895263671875, 78.21243286132812, -1.366607666015625, 1.0594215393066406, 20.576629638671875, -4.68061637878418, 22.504417419433594, 47.779823303222656, 12.187881469726562, -16.509212493896484, 5.73907470703125, -61.93293762207031, 61.98635482788086, 22.68838119506836], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000263.npy"}
|
|
{"epoch": 0.3975812547241119, "step": 264, "batch_size": 64, "mean": 21.974641799926758, "std": 35.12774658203125, "min": -64.21449279785156, "p10": -19.78705940246582, "median": 18.47644805908203, "p90": 70.90829315185549, "max": 96.96377563476562, "pos_frac": 0.71875, "sample": [41.63929748535156, -15.892929077148438, 10.143043518066406, 0.7991161346435547, 1.835073471069336, -9.935022354125977, 2.041656494140625, 0.26604461669921875, 54.16459274291992, 18.669937133789062, -25.10137367248535, 5.586769104003906, -64.21449279785156, 39.22450256347656, -21.311080932617188, 29.175613403320312, 46.18596649169922, 38.41300964355469, 5.643928527832031, 72.54464721679688, 51.47578430175781, 34.75789260864258, 58.80464172363281, 39.05523681640625, 45.75311279296875, 90.24937438964844, -0.5576858520507812, 20.091262817382812, -7.958251953125, 20.24957847595215, 13.490455627441406, -1.9282608032226562, 63.039878845214844, 2.7229537963867188, -10.75503921508789, 19.587265014648438, 79.96131896972656, -17.44452667236328, 44.220130920410156, 78.09725952148438, -19.9378662109375, -17.546974182128906, 15.420906066894531, 78.95228576660156, 56.135841369628906, 54.18618392944336, -38.027679443359375, 47.64994812011719, -32.255096435546875, 7.7778472900390625, -30.996627807617188, 5.49462890625, 85.51599884033203, 30.029186248779297, 33.6136589050293, -19.435176849365234, 15.445487976074219, 51.13671875, 18.282958984375, -0.3752593994140625, -2.5159149169921875, 96.96377563476562, 50.981414794921875, 67.09013366699219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000264.npy"}
|
|
{"epoch": 0.39909297052154197, "step": 265, "batch_size": 64, "mean": 23.634689331054688, "std": 39.39879608154297, "min": -67.57553100585938, "p10": -28.182225036621084, "median": 19.48969554901123, "p90": 75.6153175354004, "max": 83.00022888183594, "pos_frac": 0.765625, "sample": [11.541122436523438, 3.164632797241211, 1.4755859375, -32.51641845703125, 10.747013092041016, -1.6487655639648438, 27.66155433654785, 20.950172424316406, -42.45741271972656, 76.51918029785156, 19.578094482421875, 4.383823394775391, 36.700050354003906, 37.806488037109375, -67.57553100585938, -12.306968688964844, 30.423254013061523, 4.873512268066406, 19.401296615600586, -64.6113510131836, 10.97358512878418, 67.57904052734375, 66.459716796875, 19.14497184753418, 3.1684131622314453, -1.7821311950683594, 60.826385498046875, 1.2721633911132812, 22.547897338867188, 5.382383346557617, 83.00022888183594, 81.43075561523438, -12.765068054199219, 75.50942993164062, 56.51763916015625, 79.03500366210938, 5.208280563354492, 12.168231964111328, -3.9703826904296875, -18.069107055664062, 26.56927490234375, 62.862693786621094, 65.71796417236328, 75.66069793701172, 78.53768920898438, 55.259544372558594, 82.29566955566406, 37.321815490722656, 13.300270080566406, -38.73495101928711, 17.768936157226562, 29.13799285888672, -12.725231170654297, 74.01226806640625, 68.11671447753906, 72.18730926513672, 74.8065185546875, -53.18634796142578, 44.400657653808594, -7.725940704345703, 50.509521484375, 0.3856353759765625, 49.298240661621094, -50.90354919433594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000265.npy"}
|
|
{"epoch": 0.40060468631897206, "step": 266, "batch_size": 64, "mean": 23.04697036743164, "std": 33.32042694091797, "min": -67.15645599365234, "p10": -9.660569381713866, "median": 16.098182678222656, "p90": 70.07636947631836, "max": 96.59603881835938, "pos_frac": 0.765625, "sample": [-6.329925537109375, 66.20329284667969, 72.68374633789062, 6.086215972900391, 42.61780548095703, -1.1850547790527344, 0.34503936767578125, -28.760589599609375, 6.248416900634766, 23.083213806152344, 25.09206771850586, 5.9468994140625, 8.110115051269531, 36.913780212402344, 48.346961975097656, 46.17293167114258, 5.768463134765625, 96.59603881835938, -9.830570220947266, -6.1214141845703125, 79.61804962158203, 17.022415161132812, -0.30507659912109375, 77.05435943603516, 36.366943359375, 83.29136657714844, 2.4189453125, 71.0738525390625, 37.492591857910156, -9.263900756835938, 61.59861755371094, 20.095382690429688, 67.74890899658203, -5.833183288574219, 31.241432189941406, -9.941993713378906, 32.5352783203125, 49.50181579589844, 53.491153717041016, 0.7341804504394531, 0.872283935546875, -23.814437866210938, -20.46045684814453, 8.234210968017578, -67.15645599365234, 8.861549377441406, 15.1739501953125, 14.480339050292969, 28.761512756347656, 43.50133514404297, -41.98228454589844, 4.670957565307617, -6.1317901611328125, 19.143096923828125, 64.91191864013672, 39.19363784790039, 81.97896575927734, -5.789823532104492, 44.672393798828125, 7.1181182861328125, 47.32246017456055, 8.026983261108398, 67.36201477050781, 2.1270179748535156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000266.npy"}
|
|
{"epoch": 0.4021164021164021, "step": 267, "batch_size": 64, "mean": 18.101757049560547, "std": 35.361305236816406, "min": -82.36337280273438, "p10": -17.435389137268064, "median": 12.209686279296875, "p90": 64.41185455322267, "max": 87.31531524658203, "pos_frac": 0.703125, "sample": [40.271148681640625, 2.3361740112304688, 4.218910217285156, -52.13808059692383, 6.355224609375, -15.171688079833984, 52.8227653503418, 74.77888488769531, 53.92982482910156, 4.419795989990234, -9.165243148803711, 35.00977325439453, 1.6074714660644531, 85.77397918701172, 5.431987762451172, -35.2313232421875, 21.295608520507812, -40.281124114990234, -9.936721801757812, -57.01287078857422, -16.974485397338867, -17.632919311523438, 0.7702713012695312, 42.19327163696289, 58.032470703125, 74.63363647460938, -0.006099700927734375, 8.395023345947266, 51.15376281738281, -13.784927368164062, 26.68132781982422, 1.8062324523925781, 61.397979736328125, 43.31938552856445, -10.395851135253906, 18.420352935791016, 32.48455047607422, -4.6537628173828125, 5.1804962158203125, 87.31531524658203, 24.350723266601562, -0.1680126190185547, 80.30184936523438, -5.041778564453125, 65.24411010742188, 38.22736740112305, 62.46992492675781, 20.22270965576172, -82.36337280273438, 60.573463439941406, 26.496795654296875, -19.465911865234375, -3.150012969970703, 13.752147674560547, 9.932441711425781, 2.886432647705078, 41.78639221191406, 38.32415771484375, 66.75665283203125, 25.830215454101562, 10.667224884033203, 47.594478607177734, 17.32970428466797, -1.695913314819336], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000267.npy"}
|
|
{"epoch": 0.4036281179138322, "step": 268, "batch_size": 64, "mean": 21.290382385253906, "std": 29.448326110839844, "min": -36.78025817871094, "p10": -8.377500724792478, "median": 12.843878746032715, "p90": 63.4060962677002, "max": 85.50640869140625, "pos_frac": 0.75, "sample": [8.175445556640625, 84.08097076416016, 1.483480453491211, -4.9370574951171875, 41.46797180175781, 69.45967864990234, 21.208587646484375, -2.6946792602539062, -10.928829193115234, 28.90090560913086, 62.9725341796875, 0.195556640625, -0.5513114929199219, 4.759773254394531, 9.951553344726562, 6.611248016357422, 27.76120948791504, -11.348098754882812, -1.5977954864501953, 63.18766403198242, 49.53907775878906, 59.77995300292969, -1.888742446899414, 2.128570556640625, -11.871910095214844, 19.317138671875, 9.87673568725586, 42.67223358154297, 19.695968627929688, 6.110420227050781, 55.999916076660156, 7.311038970947266, 16.676509857177734, 20.74297332763672, -9.319992065429688, 60.188629150390625, -3.2957687377929688, -15.951885223388672, -33.299774169921875, 13.561216354370117, 42.68354034423828, 5.29493522644043, 1.9528121948242188, 63.49971008300781, 53.12293243408203, 27.277236938476562, 70.14166259765625, -6.178354263305664, 12.126541137695312, -36.78025817871094, -4.458457946777344, -5.0332794189453125, 43.54637145996094, 0.9005584716796875, 85.50640869140625, 35.96910858154297, 16.819984436035156, 25.593170166015625, 77.29769897460938, 28.027109146118164, 0.6587142944335938, 84.54693603515625, 31.57394790649414, 2.3643417358398438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000268.npy"}
|
|
{"epoch": 0.4051398337112623, "step": 269, "batch_size": 64, "mean": 24.284420013427734, "std": 32.21326446533203, "min": -65.88279724121094, "p10": -5.166559219360352, "median": 15.523634910583496, "p90": 69.72800140380859, "max": 87.97753143310547, "pos_frac": 0.75, "sample": [-1.603851318359375, -22.835922241210938, 10.415313720703125, 62.71084213256836, 32.300689697265625, -4.4482421875, 19.423431396484375, 62.405975341796875, 13.534980773925781, -3.7648391723632812, -9.337692260742188, 21.897056579589844, 46.74378967285156, 28.816314697265625, 76.57974243164062, 68.43690490722656, -1.0057258605957031, 27.4434814453125, 87.97753143310547, -44.94738006591797, 15.519073486328125, -5.191650390625, 64.61345672607422, 8.373123168945312, 5.252655029296875, 72.79776763916016, -7.877351760864258, 86.35037231445312, 35.58984375, 68.82011413574219, 15.528196334838867, 9.822322845458984, -5.108013153076172, 44.26885986328125, -1.1388092041015625, 4.716650009155273, 4.503501892089844, 26.725753784179688, -1.7535171508789062, 66.84940338134766, 70.8075942993164, 38.49252700805664, 13.303215026855469, 9.555652618408203, 61.91350555419922, -15.36856460571289, 34.89844512939453, 13.235008239746094, -1.6050071716308594, 3.06500244140625, 70.11709594726562, -0.6100330352783203, 77.6932373046875, 7.3243865966796875, 1.9449386596679688, -65.88279724121094, 9.48382568359375, 48.613372802734375, 16.551544189453125, 38.0357666015625, 39.978187561035156, 24.06365966796875, 66.26753234863281, 12.920574188232422], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000269.npy"}
|
|
{"epoch": 0.40665154950869237, "step": 270, "batch_size": 64, "mean": 19.895034790039062, "std": 32.49760437011719, "min": -65.74136352539062, "p10": -10.171727752685545, "median": 15.148857116699219, "p90": 66.58075408935548, "max": 90.10505676269531, "pos_frac": 0.796875, "sample": [33.41650390625, 79.86249542236328, -10.691841125488281, 82.31588745117188, 25.237674713134766, 76.79570007324219, 11.761356353759766, 18.60428237915039, 39.361968994140625, 14.478042602539062, 90.10505676269531, 44.470703125, 20.468299865722656, 52.06134796142578, 3.3475570678710938, -21.928085327148438, 20.954147338867188, 44.978477478027344, 9.31502914428711, 48.152587890625, 16.261489868164062, 43.674102783203125, 9.217498779296875, -26.980213165283203, 2.50048828125, 10.004837036132812, 55.767555236816406, -62.878173828125, 12.606365203857422, 7.886072158813477, 17.446075439453125, 12.3922119140625, 3.1998748779296875, 37.524269104003906, -1.808013916015625, 8.449028015136719, 75.44693756103516, -8.9581298828125, 17.499080657958984, -65.74136352539062, 67.99166870117188, 56.482154846191406, 49.953948974609375, -6.255464553833008, 12.408611297607422, 26.81041717529297, -4.4560699462890625, 7.86224365234375, 63.28861999511719, 31.594091415405273, 68.47312927246094, 0.36835479736328125, -15.915023803710938, 39.590789794921875, 3.895294189453125, -5.189384460449219, 3.7653121948242188, -59.20439910888672, 35.16841125488281, 9.735122680664062, 21.375106811523438, -2.031513214111328, 5.174034118652344, 15.819671630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000270.npy"}
|
|
{"epoch": 0.40816326530612246, "step": 271, "batch_size": 64, "mean": 16.715612411499023, "std": 38.459320068359375, "min": -56.81487274169922, "p10": -21.321155929565425, "median": 5.089056015014648, "p90": 73.2817352294922, "max": 115.12918090820312, "pos_frac": 0.640625, "sample": [17.40155029296875, 29.971275329589844, -6.69000244140625, 5.602836608886719, 82.00963592529297, 67.48921966552734, 11.90310287475586, 2.1699905395507812, 67.30769348144531, -22.836002349853516, 48.9925537109375, -10.041177749633789, 96.878662109375, -12.454330444335938, 36.474273681640625, 71.90264129638672, 2.6507740020751953, 4.8571014404296875, 1.9669113159179688, 83.05305480957031, 9.72491455078125, 6.649242401123047, 62.14491271972656, -10.572097778320312, -7.844818115234375, 0.7825546264648438, -10.067703247070312, 39.35906982421875, -4.5934295654296875, -16.316484451293945, 3.628877639770508, 47.74644470214844, 8.347625732421875, -42.34571838378906, -6.135284423828125, 71.91941833496094, 12.980323791503906, 2.9075393676757812, 48.0211181640625, 71.57157897949219, -0.696563720703125, -39.100067138671875, 115.12918090820312, -30.76966094970703, -7.107063293457031, 10.10015869140625, -2.8714561462402344, -56.81487274169922, -56.616729736328125, 5.204841613769531, 4.973270416259766, -15.461883544921875, 74.38574981689453, 73.86558532714844, 39.15570831298828, 2.771484375, -25.226806640625, 34.86009979248047, -2.60784912109375, 5.735553741455078, -9.432708740234375, -17.786514282226562, 93.39226531982422, 8.199615478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000271.npy"}
|
|
{"epoch": 0.40967498110355255, "step": 272, "batch_size": 64, "mean": 30.634031295776367, "std": 30.88543128967285, "min": -38.82512283325195, "p10": -1.510014533996582, "median": 32.44437789916992, "p90": 81.81999816894532, "max": 90.0846176147461, "pos_frac": 0.859375, "sample": [9.912429809570312, -1.9994049072265625, -11.693611145019531, 35.28676223754883, 31.344581604003906, 84.52092742919922, 29.850311279296875, 65.14622497558594, 82.22180938720703, 10.874320983886719, 50.04853057861328, 33.92068862915039, 9.862709045410156, -1.469512939453125, 40.65857696533203, 34.26280212402344, 42.711273193359375, 2.5136566162109375, 52.77374267578125, 50.6565055847168, 64.3874282836914, 10.569744110107422, 53.60378646850586, 42.03742218017578, 87.63909912109375, 10.708213806152344, 33.847633361816406, -24.53227996826172, 8.508995056152344, 80.88243865966797, 33.54417419433594, 87.97069549560547, 46.086814880371094, 0.5665969848632812, 40.55619812011719, 8.645381927490234, 4.765508651733398, 1.0994949340820312, 24.212730407714844, -38.82512283325195, 61.27703094482422, 85.93794250488281, 90.0846176147461, 23.80063247680664, 1.954559326171875, 9.615760803222656, 77.04692077636719, 34.24102020263672, 40.571537017822266, 46.20660400390625, 24.89320182800293, 56.21345520019531, -14.323799133300781, -0.3204193115234375, 9.931243896484375, 19.827117919921875, 37.54651641845703, 41.55906677246094, -12.30411148071289, -1.5273723602294922, 2.404378890991211, 3.6677894592285156, 84.53724670410156, 10.058820724487305], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000272.npy"}
|
|
{"epoch": 0.41118669690098264, "step": 273, "batch_size": 64, "mean": 22.425758361816406, "std": 39.576908111572266, "min": -70.25868225097656, "p10": -33.26486587524413, "median": 26.51580810546875, "p90": 63.799059295654295, "max": 125.43414306640625, "pos_frac": 0.71875, "sample": [-1.5955467224121094, 34.30571746826172, 60.03410339355469, -36.57013702392578, 57.568817138671875, -7.688972473144531, -45.73016357421875, 49.9996452331543, -51.56110763549805, 32.92455291748047, 27.363296508789062, -25.552566528320312, 33.246849060058594, -1.139730453491211, 11.366207122802734, 29.14542007446289, 71.7940444946289, 11.864788055419922, 63.85426330566406, 52.26982116699219, 63.670249938964844, 15.381839752197266, -3.8105812072753906, -11.06344223022461, -70.25868225097656, 49.68036651611328, 37.245513916015625, 5.866355895996094, -16.964441299438477, 4.319408416748047, 35.141326904296875, 125.43414306640625, 60.063934326171875, -59.25655746459961, -0.391448974609375, 34.71989440917969, 44.00216293334961, 51.21220016479492, 11.366950988769531, 55.448333740234375, -41.36567687988281, 25.668319702148438, 53.47117614746094, -5.27362060546875, 28.353992462158203, 113.72038269042969, 85.76728820800781, 40.83506774902344, 11.47998046875, 43.05180740356445, 1.0112285614013672, 18.085845947265625, 1.5970077514648438, -11.922515869140625, 11.138526916503906, 78.40840911865234, 2.3068313598632812, 35.68827819824219, 16.696483612060547, -12.571327209472656, 61.71543884277344, 85.79938507080078, -40.1055908203125, 33.98500061035156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000273.npy"}
|
|
{"epoch": 0.4126984126984127, "step": 274, "batch_size": 64, "mean": 23.824462890625, "std": 37.32594680786133, "min": -65.39856719970703, "p10": -16.53703899383545, "median": 19.06736946105957, "p90": 74.98372039794921, "max": 89.19393157958984, "pos_frac": 0.765625, "sample": [31.39228057861328, -16.635047912597656, -7.0031585693359375, 1.3195343017578125, -7.44207763671875, -55.04682922363281, 65.19644165039062, 65.01904296875, 13.662086486816406, 17.695493698120117, 37.885902404785156, 56.31470489501953, 1.9793853759765625, -7.379608154296875, 29.9844913482666, 84.32133483886719, 83.42486572265625, -43.80194854736328, 10.826618194580078, 19.197956085205078, 69.71881103515625, 67.38386535644531, 40.513118743896484, 75.02001190185547, 43.209007263183594, 2.4282970428466797, 73.64789581298828, 86.50144958496094, 20.40913200378418, 0.225341796875, 4.370552062988281, 7.52838134765625, 60.94261932373047, -5.996513366699219, -5.120573043823242, 38.50238037109375, -45.750244140625, -22.35638427734375, 51.12682342529297, 3.6676025390625, -16.308351516723633, 37.68867492675781, 9.924129486083984, 14.65655517578125, -5.656435012817383, 23.781021118164062, 19.971160888671875, -65.39856719970703, 23.18732452392578, 12.393817901611328, 79.99304962158203, 70.50835418701172, 5.907649993896484, 18.936782836914062, 6.7992095947265625, -11.347570419311523, -29.957820892333984, 9.375797271728516, 81.30059814453125, 74.89904022216797, 35.10798263549805, 23.623573303222656, 89.19393157958984, 69.30257415771484], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000274.npy"}
|
|
{"epoch": 0.41421012849584277, "step": 275, "batch_size": 64, "mean": 25.821151733398438, "std": 34.42043685913086, "min": -74.8513412475586, "p10": -10.529314422607417, "median": 20.325023651123047, "p90": 75.59595642089845, "max": 97.00695037841797, "pos_frac": 0.8125, "sample": [28.602264404296875, 14.467231750488281, -13.078460693359375, 49.29130554199219, 73.53570556640625, 16.183937072753906, 27.826560974121094, -74.8513412475586, 86.08892059326172, 65.31082153320312, 11.36395263671875, 18.364273071289062, 0.01331329345703125, 85.0808334350586, -1.495431900024414, 8.846920013427734, 62.46014404296875, 36.539154052734375, 74.15455627441406, -20.553604125976562, -58.97468566894531, 90.32808685302734, 6.312896728515625, 6.5054931640625, 51.845802307128906, 1.1752700805664062, -15.072792053222656, 1.6087112426757812, 6.168720245361328, 9.323486328125, 25.07415771484375, 97.00695037841797, 21.948280334472656, 23.40215301513672, 76.21369934082031, 41.67549133300781, 41.18791961669922, 24.42572593688965, -0.280670166015625, -2.3558082580566406, 34.94929504394531, 3.7827987670898438, 18.320838928222656, 0.4889106750488281, 15.048036575317383, 81.6407699584961, 40.84149169921875, 50.977867126464844, -14.85067367553711, 27.723255157470703, 3.1996536254882812, 5.0924835205078125, -4.088754653930664, 13.411178588867188, 81.79887390136719, 64.71395111083984, -4.581306457519531, 45.35332489013672, 43.474159240722656, 61.38047790527344, 40.50275421142578, 18.701766967773438, -13.319581985473633, 42.32209777832031], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000275.npy"}
|
|
{"epoch": 0.41572184429327286, "step": 276, "batch_size": 64, "mean": 24.507638931274414, "std": 31.960922241210938, "min": -50.62937927246094, "p10": -6.469737243652343, "median": 19.428871154785156, "p90": 69.03575744628907, "max": 94.76016235351562, "pos_frac": 0.84375, "sample": [68.87925720214844, 7.425472259521484, 4.31749153137207, 39.254600524902344, 8.180978775024414, 20.678810119628906, 94.76016235351562, -36.581336975097656, 31.409263610839844, 8.305879592895508, 19.653362274169922, 9.673086166381836, 29.70269012451172, 40.62879943847656, 2.2139892578125, 77.15638732910156, -2.610828399658203, 3.2943267822265625, 14.844802856445312, 7.596469879150391, 18.453205108642578, 63.76641082763672, 58.505104064941406, 82.45463562011719, 10.043296813964844, -34.87334442138672, 53.24736022949219, -10.700164794921875, 32.887672424316406, 63.436805725097656, 6.442287445068359, 13.070335388183594, -15.629646301269531, -50.62937927246094, 45.47796630859375, -6.15997314453125, 69.10282897949219, 0.41574859619140625, 26.619827270507812, 46.57652282714844, -46.908592224121094, 21.90355682373047, 82.52970123291016, 2.3509979248046875, -3.9635772705078125, 17.107177734375, 58.98371887207031, 26.913330078125, 63.74360656738281, 45.21942901611328, -6.6024932861328125, 0.2463226318359375, 10.972496032714844, 19.20438003540039, 17.001220703125, 85.949951171875, 71.71837615966797, 27.934673309326172, 28.438217163085938, 35.18114471435547, 7.799386978149414, 37.10667419433594, 42.481380462646484, 1.8866653442382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000276.npy"}
|
|
{"epoch": 0.41723356009070295, "step": 277, "batch_size": 64, "mean": 14.846869468688965, "std": 32.294734954833984, "min": -52.570926666259766, "p10": -25.253006362915034, "median": 9.580495834350586, "p90": 60.67423400878907, "max": 95.78424072265625, "pos_frac": 0.6875, "sample": [45.460609436035156, -42.195098876953125, 52.71106719970703, -0.1726531982421875, 40.280601501464844, -33.14448165893555, 9.738239288330078, 10.574539184570312, 63.517845153808594, 22.214218139648438, -19.646034240722656, -5.976387023925781, 14.09393310546875, 5.31317138671875, -16.51056671142578, 32.18235778808594, 22.075376510620117, -3.7028427124023438, 4.060508728027344, 11.949542999267578, 24.98879051208496, 41.64002227783203, 3.943145751953125, 7.56787109375, -2.208820343017578, 12.852294921875, 2.5082778930664062, -18.88671875, -12.079147338867188, 58.59260559082031, 0.0015926361083984375, 12.181358337402344, -30.48439598083496, 65.05073547363281, 19.213138580322266, 54.77281951904297, -19.08765411376953, -39.586463928222656, 28.076812744140625, 7.893856048583984, 5.5060882568359375, -27.655994415283203, 77.31077575683594, 61.56636047363281, 23.2723388671875, 9.422752380371094, -2.4117813110351562, 31.226417541503906, -36.139747619628906, 95.78424072265625, 36.11259460449219, -13.764589309692383, 71.65274047851562, 29.152084350585938, -6.293006896972656, 1.492238998413086, 54.80255126953125, -52.570926666259766, 46.5679931640625, 81.39077758789062, -8.341018676757812, 5.592315673828125, 0.4313831329345703, 36.31897735595703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000277.npy"}
|
|
{"epoch": 0.41874527588813304, "step": 278, "batch_size": 64, "mean": 24.3956298828125, "std": 36.3751335144043, "min": -62.75355529785156, "p10": -14.068092536926269, "median": 21.382603645324707, "p90": 78.95611419677735, "max": 97.01594543457031, "pos_frac": 0.796875, "sample": [-14.583982467651367, -44.044677734375, 64.26130676269531, 31.050994873046875, 7.617454528808594, -2.6778621673583984, -57.63787841796875, 19.318008422851562, 81.68535614013672, 97.01594543457031, 17.926345825195312, 6.399749755859375, 69.13191986083984, 81.18949890136719, 36.23657989501953, -6.46412467956543, 15.857101440429688, 37.8150520324707, 40.82373046875, 55.33881759643555, 37.0113525390625, 34.313323974609375, 12.498367309570312, 7.496498107910156, -62.75355529785156, 90.6005859375, -10.666122436523438, 49.892974853515625, 24.175384521484375, 8.790603637695312, 10.410209655761719, 27.34638214111328, -14.696136474609375, -12.864349365234375, 60.873809814453125, 5.352546691894531, 58.80033874511719, 28.176528930664062, 40.0194206237793, 52.162132263183594, 88.20548248291016, 88.16291809082031, -8.395526885986328, 35.30859375, 23.44719886779785, 37.33258819580078, 10.60965347290039, 2.5359020233154297, -42.513572692871094, 5.024749755859375, 23.47815704345703, 55.626312255859375, -45.16987609863281, -3.0509166717529297, 1.8801803588867188, 80.17915344238281, 17.809722900390625, 76.10235595703125, 29.200801849365234, 12.738189697265625, 8.812005996704102, 0.5993385314941406, 64.12220001220703, 16.07513427734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000278.npy"}
|
|
{"epoch": 0.42025699168556313, "step": 279, "batch_size": 64, "mean": 31.532020568847656, "std": 32.745277404785156, "min": -52.83957290649414, "p10": -6.0512863159179675, "median": 25.684070587158203, "p90": 79.9479476928711, "max": 95.00604248046875, "pos_frac": 0.84375, "sample": [-1.3205547332763672, 6.541801452636719, 95.00604248046875, 14.187294006347656, -52.83957290649414, 59.583740234375, 8.899463653564453, -4.8286895751953125, -6.57525634765625, 35.542930603027344, 22.150352478027344, 72.52059936523438, 30.96526336669922, 77.0101318359375, 32.69645309448242, 93.52592468261719, 31.95635986328125, 57.232852935791016, -17.174943923950195, 2.506969451904297, -17.427322387695312, 65.93082427978516, 25.656517028808594, 25.711624145507812, 12.469993591308594, 30.91931915283203, 25.076807022094727, -1.2843818664550781, 17.175331115722656, 68.14210510253906, 15.766265869140625, 49.07460021972656, 83.3870620727539, 16.5911865234375, -14.558067321777344, 62.91674041748047, 33.172733306884766, 46.096282958984375, -9.717094421386719, 87.42466735839844, 81.89665985107422, 11.646099090576172, 80.49800109863281, 7.122978210449219, 7.086181640625, 19.953834533691406, 12.11440658569336, 6.990520477294922, 53.29168701171875, 78.66448974609375, 50.63728713989258, -12.0318603515625, 31.96834945678711, 77.52035522460938, 2.310708999633789, 11.715744018554688, 22.391281127929688, 7.828216552734375, 51.74324417114258, 53.41175842285156, 46.35783386230469, 39.03346252441406, 88.08148193359375, 7.70416259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000279.npy"}
|
|
{"epoch": 0.4217687074829932, "step": 280, "batch_size": 64, "mean": 33.42193603515625, "std": 32.332977294921875, "min": -21.880355834960938, "p10": -0.8644289016723631, "median": 23.830416679382324, "p90": 83.36447677612306, "max": 108.25390625, "pos_frac": 0.875, "sample": [54.07024002075195, 91.78907775878906, 6.675008773803711, 46.930076599121094, 0.33219146728515625, 89.6217041015625, 2.0539817810058594, -1.2770767211914062, 8.913995742797852, 60.0250244140625, 18.657791137695312, 15.080848693847656, 56.374176025390625, 41.48289489746094, 76.10794067382812, 78.01748657226562, 30.592201232910156, 32.302066802978516, 79.91114807128906, 17.666717529296875, 2.09136962890625, 6.730012893676758, -0.9660072326660156, 1.3836307525634766, 86.29421997070312, 52.02693176269531, 17.28238296508789, 44.688804626464844, 76.48843383789062, 108.25390625, 30.629375457763672, 7.724884033203125, 11.961715698242188, 34.693603515625, 5.392955780029297, 38.51422119140625, 97.402587890625, -7.729240417480469, 89.96007537841797, 68.2028579711914, -21.880355834960938, -13.43826675415039, -1.3034820556640625, -0.6274127960205078, 33.31087875366211, 42.53709030151367, 24.776174545288086, -3.3342971801757812, 22.884658813476562, 10.350557327270508, 84.84447479248047, 15.650754928588867, 13.34720230102539, 64.45635223388672, 25.60370445251465, 11.826797485351562, 18.128650665283203, 9.32415771484375, 13.360345840454102, 13.725975036621094, 5.779335021972656, 63.94981384277344, 73.27124786376953, 56.10546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000280.npy"}
|
|
{"epoch": 0.42328042328042326, "step": 281, "batch_size": 64, "mean": 30.297103881835938, "std": 36.555335998535156, "min": -60.51171112060547, "p10": -11.480802345275878, "median": 24.89885425567627, "p90": 81.11811904907226, "max": 95.28807067871094, "pos_frac": 0.765625, "sample": [-20.156688690185547, 53.35075759887695, 10.633766174316406, -1.365814208984375, 73.60527038574219, 53.26953887939453, 23.75565528869629, 61.593109130859375, 38.47479248046875, 91.75178527832031, 17.679336547851562, 80.93753814697266, 44.66976547241211, 41.357994079589844, -2.3079681396484375, 8.649789810180664, 12.793035507202148, 81.19551086425781, -3.483823776245117, 67.99420928955078, 73.2255859375, 4.547052383422852, 8.317428588867188, 37.62274932861328, 51.552459716796875, 13.268779754638672, 14.259414672851562, -4.828327178955078, 25.362701416015625, -60.51171112060547, 10.468132019042969, 3.6670913696289062, -0.6130828857421875, 24.435007095336914, 82.7373046875, -33.82898712158203, 56.31871032714844, 95.28807067871094, 7.662506103515625, -17.159908294677734, 85.05530548095703, 45.11309814453125, 90.4389877319336, 53.329559326171875, 81.30680084228516, 32.12531280517578, 60.018123626708984, 10.579486846923828, 74.65947723388672, 29.683185577392578, -1.7537994384765625, 5.090995788574219, 80.12782287597656, -35.22089385986328, 10.378215789794922, 66.47988891601562, 56.546932220458984, -11.765609741210938, 52.06718063354492, 60.78754425048828, -6.072154998779297, -10.816251754760742, 7.581483840942383, -22.914487838745117], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000281.npy"}
|
|
{"epoch": 0.42479213907785335, "step": 282, "batch_size": 64, "mean": 27.98813247680664, "std": 40.309295654296875, "min": -61.36808395385742, "p10": -23.562158584594723, "median": 22.143760681152344, "p90": 86.62294692993164, "max": 114.077392578125, "pos_frac": 0.71875, "sample": [29.259361267089844, -24.365699768066406, -31.07062530517578, 9.361373901367188, 6.063407897949219, 55.87263488769531, 45.17788314819336, 12.564386367797852, -5.61810302734375, 92.99362182617188, 62.58074188232422, 73.62603759765625, 21.425378799438477, 44.196014404296875, 80.26571655273438, 26.75067138671875, 95.48553466796875, 34.8503303527832, -6.7412109375, 49.88264846801758, 22.377182006835938, 21.389266967773438, 3.3393611907958984, 79.35079956054688, 14.862588882446289, -3.217620849609375, -29.293319702148438, 67.51517486572266, 89.54776000976562, 18.731822967529297, -21.68722915649414, -9.703756332397461, -61.36808395385742, 77.07858276367188, 6.474996566772461, -29.0682373046875, 87.29019927978516, 10.686630249023438, -19.337814331054688, 34.253875732421875, 55.85381317138672, 67.34154510498047, 7.6359405517578125, 39.57345962524414, 85.06602478027344, 114.077392578125, 36.55361557006836, -0.36974334716796875, -4.6298065185546875, 21.91033935546875, -13.085372924804688, -21.05902862548828, -8.990585327148438, 14.491504669189453, -29.965309143066406, 102.71361541748047, 94.18156433105469, 51.28246307373047, -33.063804626464844, 23.10040283203125, 42.3074951171875, 40.52556610107422, 60.127933502197266, 13.879112243652344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000282.npy"}
|
|
{"epoch": 0.42630385487528344, "step": 283, "batch_size": 64, "mean": 24.515975952148438, "std": 40.21672821044922, "min": -65.52972412109375, "p10": -25.49234619140625, "median": 12.536491394042969, "p90": 78.66199111938477, "max": 100.69617462158203, "pos_frac": 0.765625, "sample": [75.92948913574219, -44.64582824707031, -30.27838897705078, -31.325477600097656, 39.90216827392578, 34.320369720458984, 4.453895568847656, 0.26482391357421875, 74.15904235839844, 0.4303627014160156, -8.893089294433594, 59.00932312011719, 13.716705322265625, 9.41061019897461, 80.6342544555664, 12.370708465576172, 11.241851806640625, 88.11254119873047, -26.572174072265625, 23.686004638671875, 77.22161102294922, 44.05195617675781, 72.78330993652344, 1.6256332397460938, 38.207275390625, 85.9233627319336, 65.01671600341797, 11.268424987792969, 100.69617462158203, 12.049240112304688, 10.978157043457031, 69.24772644042969, -50.39125061035156, 11.721099853515625, -1.51971435546875, 46.395782470703125, -50.61041259765625, 12.9981689453125, 32.04591369628906, 71.0850601196289, 85.61322021484375, -19.32533073425293, 71.23783874511719, 75.74360656738281, 8.518619537353516, 12.702274322509766, 13.604911804199219, -65.52972412109375, 5.194303512573242, -19.27423095703125, 56.899925231933594, 5.936260223388672, 79.279296875, -0.09302902221679688, 56.9581298828125, 48.17338562011719, -3.3787841796875, -9.435413360595703, 7.931535720825195, 6.195953369140625, -22.972747802734375, 92.78782653808594, 3.2896728515625, 32.24359893798828], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000283.npy"}
|
|
{"epoch": 0.42781557067271353, "step": 284, "batch_size": 64, "mean": 26.66477394104004, "std": 36.58161163330078, "min": -75.04081726074219, "p10": -2.4871522903442376, "median": 17.66067886352539, "p90": 85.38203430175786, "max": 103.61018371582031, "pos_frac": 0.84375, "sample": [-1.300628662109375, 6.1596832275390625, 30.401823043823242, 16.244873046875, 100.66800689697266, 47.69158935546875, 2.6369991302490234, -12.521804809570312, 96.40632629394531, 5.384979248046875, 8.125476837158203, 22.623149871826172, 2.5131988525390625, 6.0543212890625, 62.49664306640625, -10.10770034790039, 8.005439758300781, 58.42543029785156, 3.0660858154296875, 98.00210571289062, 4.8511199951171875, 32.26588439941406, 3.89569091796875, -15.975421905517578, -14.929374694824219, 20.04492950439453, 7.083335876464844, 12.80926513671875, 29.692943572998047, -1.7306556701660156, 74.57582092285156, 36.79125213623047, 47.233612060546875, 39.354515075683594, 14.761306762695312, 19.07648468017578, 54.106842041015625, 94.36656188964844, 89.36361694335938, 6.041130065917969, 59.546241760253906, 103.61018371582031, 21.854476928710938, 0.15497207641601562, -2.8113651275634766, 0.19444656372070312, 66.46914672851562, 14.35038948059082, 37.37348937988281, 76.0916748046875, 3.2637557983398438, 33.707942962646484, 65.28314208984375, 13.148460388183594, 10.579399108886719, 34.44898986816406, 101.37277221679688, 23.289306640625, 31.165462493896484, -75.04081726074219, 42.679115295410156, -58.42535400390625, 0.9533615112304688, -1.368520736694336], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000284.npy"}
|
|
{"epoch": 0.4293272864701436, "step": 285, "batch_size": 64, "mean": 24.112892150878906, "std": 37.73579788208008, "min": -56.41380310058594, "p10": -20.121009826660156, "median": 16.787006378173828, "p90": 80.61451721191406, "max": 95.7332763671875, "pos_frac": 0.75, "sample": [30.79724884033203, 8.218421936035156, -1.616241455078125, 80.0833969116211, 6.88385009765625, -29.122806549072266, -56.41380310058594, 7.8076324462890625, 37.08756637573242, -17.170969009399414, 35.15204620361328, 85.15031433105469, 70.93384552001953, 26.677940368652344, 80.02133178710938, -38.33637237548828, 37.56243896484375, 91.16407012939453, 3.399066925048828, 5.733081817626953, 21.562698364257812, 0.83917236328125, -24.345333099365234, -26.92562484741211, 0.6563339233398438, -38.22088623046875, -5.06475830078125, 16.473892211914062, 36.361572265625, -19.15447235107422, -13.003387451171875, 87.18516540527344, -15.979759216308594, 80.8421401977539, 66.10150146484375, 35.332149505615234, 24.6046142578125, 11.617507934570312, 61.924171447753906, 70.63037109375, 43.927215576171875, 1.3810272216796875, 3.5774765014648438, 93.1933822631836, 64.42533111572266, 17.100120544433594, 1.5508403778076172, 95.7332763671875, -5.8639678955078125, -2.677417755126953, 36.30104064941406, 52.539276123046875, 8.248725891113281, 57.818023681640625, 2.3456573486328125, 18.97698402404785, 88.0682144165039, 75.25833129882812, -20.535240173339844, 11.507774353027344, 3.1209869384765625, 23.382415771484375, -1.5394134521484375, 39.935882568359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000285.npy"}
|
|
{"epoch": 0.4308390022675737, "step": 286, "batch_size": 64, "mean": 37.48759078979492, "std": 39.43290710449219, "min": -46.738372802734375, "p10": -3.390267372131342, "median": 29.75531768798828, "p90": 94.16571655273438, "max": 116.6873779296875, "pos_frac": 0.890625, "sample": [39.50091552734375, 7.463113784790039, 5.087030410766602, 33.255615234375, 12.142471313476562, 66.44993591308594, 8.689254760742188, 2.394624710083008, 19.584375381469727, -5.8695068359375, 47.4720458984375, 21.001571655273438, 29.128631591796875, 78.23097229003906, 77.45745086669922, 105.63485717773438, 76.56590270996094, 39.278778076171875, -25.17046356201172, 14.998085021972656, 53.988197326660156, -9.942726135253906, 21.832530975341797, 93.31547546386719, 3.552440643310547, -37.51936340332031, 50.09022521972656, 34.31576156616211, -20.42333984375, 57.09596252441406, 81.0924301147461, 59.96522903442383, 86.40245056152344, 93.57235717773438, 52.86076354980469, 112.36111450195312, 18.166454315185547, 75.8010025024414, -29.782398223876953, 7.0938873291015625, 99.37843322753906, 116.6873779296875, 20.537567138671875, 8.01470947265625, 45.50518035888672, 22.838165283203125, 6.435281753540039, 20.227813720703125, 101.01839447021484, 94.42001342773438, 89.46435546875, 25.615432739257812, 18.210002899169922, 5.089071273803711, 5.995784759521484, 30.382003784179688, 65.16460418701172, 19.826374053955078, -46.738372802734375, 102.830322265625, 40.232627868652344, 5.603534698486328, 39.42619323730469, 5.906761169433594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000286.npy"}
|
|
{"epoch": 0.4323507180650038, "step": 287, "batch_size": 64, "mean": 37.57877731323242, "std": 41.94770431518555, "min": -72.52266693115234, "p10": -4.664424705505371, "median": 29.69623565673828, "p90": 98.24327392578125, "max": 116.17694091796875, "pos_frac": 0.828125, "sample": [46.143821716308594, -17.41241455078125, 12.289398193359375, 73.03958129882812, 116.17694091796875, 85.75875854492188, 7.6664581298828125, 22.712675094604492, 39.38762664794922, -3.18743896484375, 37.68592071533203, 24.622249603271484, 29.171630859375, 30.220840454101562, 101.85910034179688, 105.61736297607422, 108.48513793945312, 40.03334045410156, 83.70976257324219, 49.45048904418945, 39.508262634277344, 34.612762451171875, 76.0810546875, 4.421525955200195, 8.036014556884766, 69.95013427734375, -1.0015754699707031, -29.245704650878906, 69.60382080078125, 3.0748634338378906, 64.78691864013672, 6.40582275390625, -72.52266693115234, 4.795600891113281, -4.88554573059082, 69.72496032714844, 14.106071472167969, 11.16696548461914, 21.44171905517578, 75.07034301757812, 79.53887939453125, -13.769744873046875, 98.30130004882812, 84.1562728881836, -22.738967895507812, 7.625617980957031, 105.5710678100586, 5.861263275146484, 84.02409362792969, 75.11699676513672, 0.4223003387451172, 6.958648681640625, 50.66039276123047, 8.428243637084961, -0.3408775329589844, 16.21387481689453, 95.16847229003906, 15.599906921386719, 56.98694610595703, -7.344882965087891, -4.148475646972656, 104.72708129882812, 98.10787963867188, 1.3529891967773438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000287.npy"}
|
|
{"epoch": 0.43386243386243384, "step": 288, "batch_size": 64, "mean": 25.4874210357666, "std": 40.9996337890625, "min": -71.09213256835938, "p10": -25.51014022827148, "median": 24.765670776367188, "p90": 86.13301391601564, "max": 106.65221405029297, "pos_frac": 0.734375, "sample": [37.822269439697266, 3.1671142578125, -29.218353271484375, -71.09213256835938, 28.439064025878906, 10.778606414794922, 24.245275497436523, -21.37741470336914, 58.70603942871094, 36.92346954345703, 15.627910614013672, 49.96802520751953, 16.273887634277344, 82.35630798339844, -27.505645751953125, 56.33740234375, -26.948715209960938, 36.54117202758789, -9.06719970703125, 32.09640121459961, -47.01848602294922, 89.69197082519531, 106.65221405029297, 87.75160217285156, -10.745803833007812, 101.65899658203125, 25.28606605529785, 43.99272918701172, 7.218040466308594, 17.699647903442383, 104.15355682373047, -43.557464599609375, 46.96064758300781, 50.337032318115234, 72.41484069824219, 38.42024230957031, 31.169662475585938, -18.8902587890625, 20.834091186523438, -22.153465270996094, -18.596710205078125, -50.43182373046875, 89.61109161376953, 1.2121734619140625, 3.3559799194335938, -20.48187255859375, 66.69693756103516, -10.79779052734375, 63.015113830566406, 58.90362548828125, 68.33860778808594, 0.9223480224609375, 53.44732666015625, 14.35833740234375, -5.003562927246094, -2.176319122314453, 52.26576232910156, 22.622413635253906, 0.11911201477050781, 88.60647583007812, 21.3267765045166, 65.217041015625, 25.519332885742188, 37.19516372680664], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000288.npy"}
|
|
{"epoch": 0.43537414965986393, "step": 289, "batch_size": 64, "mean": 25.91498565673828, "std": 44.931705474853516, "min": -74.31658935546875, "p10": -23.743520164489745, "median": 15.239624977111816, "p90": 90.96512145996094, "max": 127.03372192382812, "pos_frac": 0.71875, "sample": [22.104816436767578, 14.868522644042969, 86.42111206054688, -18.71229362487793, -29.345802307128906, -17.349105834960938, -7.956035614013672, 30.236366271972656, 43.71717071533203, 27.531700134277344, 6.645011901855469, 32.59312438964844, 70.11117553710938, 77.42012023925781, 23.490768432617188, 15.610727310180664, 84.55093383789062, -37.711273193359375, -1.8936119079589844, 88.830322265625, 11.317726135253906, 95.70048522949219, 0.779632568359375, -14.233619689941406, 73.75963592529297, 12.679794311523438, 91.88003540039062, 4.387500762939453, -8.942535400390625, 99.81102752685547, 3.7667999267578125, 19.11128044128418, -19.654117584228516, -74.31658935546875, 114.87567138671875, -10.715604782104492, 1.8282489776611328, 127.03372192382812, 7.948951721191406, 19.787513732910156, 2.691740036010742, 25.265602111816406, -20.93573760986328, 14.669126510620117, 41.61931228637695, -24.946855545043945, 28.39177703857422, -2.2250328063964844, -30.863067626953125, -57.79111862182617, 63.03134536743164, 27.345916748046875, 7.4620513916015625, 20.139862060546875, 95.2688980102539, 85.0606689453125, 74.33943176269531, 88.27902221679688, 92.44013977050781, 9.990760803222656, 6.512531280517578, -31.004831314086914, 80.87139892578125, -5.023250579833984], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000289.npy"}
|
|
{"epoch": 0.436885865457294, "step": 290, "batch_size": 64, "mean": 31.22284698486328, "std": 44.732173919677734, "min": -72.33040618896484, "p10": -10.478400421142576, "median": 29.746904373168945, "p90": 97.93758850097657, "max": 106.41625213623047, "pos_frac": 0.78125, "sample": [48.86179733276367, 34.74324035644531, 73.87335968017578, 105.49996948242188, 31.06390380859375, 7.847509384155273, -22.1790828704834, 0.2752971649169922, 4.192901611328125, 101.00960540771484, 39.432861328125, 19.764507293701172, 19.896560668945312, -56.37793731689453, 37.953643798828125, 14.210578918457031, -7.636192321777344, -8.489799499511719, -1.0411491394042969, -5.819145202636719, 37.93775939941406, 19.19329071044922, 100.09593200683594, 94.34526062011719, 11.195411682128906, 19.53723907470703, 7.216426849365234, -11.330657958984375, 16.727521896362305, -57.68933868408203, 90.9816665649414, 102.77496337890625, 76.52810668945312, 29.20706558227539, 36.808815002441406, 106.41625213623047, 95.8099136352539, 38.55683898925781, 33.285987854003906, -44.922271728515625, 83.10783386230469, 76.45465087890625, 9.141704559326172, -72.33040618896484, 3.1393814086914062, -1.0182647705078125, -4.279176712036133, 105.4925537109375, 89.4722671508789, 80.79806518554688, 14.942920684814453, 34.4539794921875, 8.862960815429688, 17.009017944335938, -65.68350219726562, 91.34628295898438, 20.69086456298828, 98.84944915771484, 31.69024658203125, 34.14727020263672, 31.189525604248047, 30.2867431640625, 41.281829833984375, -0.5427360534667969], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000290.npy"}
|
|
{"epoch": 0.4383975812547241, "step": 291, "batch_size": 64, "mean": 32.851356506347656, "std": 45.36640548706055, "min": -83.95588684082031, "p10": -16.031135940551756, "median": 17.37517547607422, "p90": 99.98351974487305, "max": 140.93756103515625, "pos_frac": 0.765625, "sample": [112.69784545898438, 100.2232437133789, 9.659027099609375, -21.5068359375, 18.339508056640625, 18.755477905273438, -6.634559631347656, 2.7445945739746094, 99.42416381835938, 2.5177993774414062, 66.05492401123047, 60.863624572753906, -18.56283187866211, -0.6024112701416016, 67.61036682128906, 82.15447998046875, -13.703697204589844, -9.427848815917969, 16.410842895507812, 54.36908721923828, -0.445892333984375, 3.040283203125, 36.53776550292969, 35.108665466308594, 52.36948776245117, 82.54767608642578, 86.46405029296875, 0.7024726867675781, 106.19154357910156, 5.4349517822265625, 42.96275329589844, 107.44357299804688, 0.006488800048828125, 5.803993225097656, -15.409164428710938, 2.440980911254883, 65.09088134765625, -4.046661376953125, 103.14761352539062, 14.101844787597656, -8.847114562988281, 98.2220230102539, -83.95588684082031, 27.646705627441406, 101.89703369140625, 37.99546813964844, 50.38581848144531, 85.98553466796875, 38.366188049316406, 140.93756103515625, -20.65738296508789, 13.351043701171875, -16.29769515991211, 14.835731506347656, -35.294654846191406, 8.486873626708984, 11.262893676757812, 96.94955444335938, 11.763786315917969, 53.30200958251953, 78.99580383300781, 41.532318115234375, 4.968994140625, -20.225990295410156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000291.npy"}
|
|
{"epoch": 0.4399092970521542, "step": 292, "batch_size": 64, "mean": 26.647615432739258, "std": 45.476104736328125, "min": -80.83549499511719, "p10": -33.51409530639648, "median": 21.932933807373047, "p90": 84.22667617797852, "max": 117.95455932617188, "pos_frac": 0.75, "sample": [-80.83549499511719, -8.892021179199219, 68.39400482177734, 0.27509117126464844, 102.50209045410156, 60.054473876953125, 26.859058380126953, 117.95455932617188, 65.35906982421875, -20.685714721679688, -36.29058074951172, 4.103717803955078, 23.265771865844727, 16.399024963378906, -49.46910095214844, -4.39013671875, 50.272613525390625, 56.07856750488281, 73.04031372070312, 12.414169311523438, 35.86321258544922, 66.65803527832031, 60.637001037597656, -38.767616271972656, 5.856109619140625, 5.140693664550781, 5.514209747314453, 85.1558609008789, -27.035629272460938, 62.57838439941406, 20.97924041748047, 60.856719970703125, 7.29258918762207, -5.88031005859375, -0.7950057983398438, -67.7117691040039, 82.05857849121094, -75.65325927734375, 0.7503929138183594, 14.558509826660156, -6.7844696044921875, 73.72301483154297, 58.92205810546875, -6.549646377563477, 104.20826721191406, 53.66161346435547, 17.629364013671875, 60.214393615722656, 21.413822174072266, 8.731735229492188, 87.92807006835938, 66.4687728881836, 2.127237319946289, 69.85224914550781, 43.10890579223633, 7.875114440917969, 39.52883529663086, 22.452045440673828, 102.51152038574219, 27.02752685546875, 24.069427490234375, -13.9781494140625, -38.80529022216797, 105.64543914794922], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000292.npy"}
|
|
{"epoch": 0.4414210128495843, "step": 293, "batch_size": 64, "mean": 37.90253448486328, "std": 48.89636993408203, "min": -100.40058898925781, "p10": -10.759159088134766, "median": 33.142677307128906, "p90": 98.63114852905274, "max": 125.8896484375, "pos_frac": 0.765625, "sample": [4.060825347900391, 22.873519897460938, 13.193805694580078, 97.4338607788086, 58.24192810058594, 22.062938690185547, -8.657794952392578, 21.42452621459961, 52.40215301513672, 9.51312255859375, 99.14427185058594, -7.370183944702148, 99.51713562011719, 74.1600112915039, 74.20893859863281, 3.8239364624023438, 76.40428924560547, 57.51243591308594, 68.76219940185547, 82.88172149658203, 62.707950592041016, -23.716163635253906, -10.765693664550781, -55.00236511230469, 80.06889343261719, 3.32952880859375, 60.665531158447266, -26.751789093017578, 25.855552673339844, 125.8896484375, 3.7162113189697266, 113.98483276367188, 95.57955932617188, -55.27318572998047, 71.10549926757812, -0.7194328308105469, 97.27592468261719, -100.40058898925781, -6.222827911376953, 3.21533203125, 25.979888916015625, 103.16262817382812, -41.576541900634766, -9.108314514160156, -4.899200439453125, 3.0052108764648438, 42.328956604003906, 79.04447937011719, 114.42266845703125, 29.708602905273438, 92.33192443847656, 11.9110107421875, 51.912925720214844, 3.6185569763183594, 80.1552734375, -10.743911743164062, 87.33592224121094, 20.030975341796875, 52.06730651855469, 106.07963562011719, 36.576751708984375, -5.022891998291016, 83.18812561035156, 88.11624145507812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000293.npy"}
|
|
{"epoch": 0.4429327286470144, "step": 294, "batch_size": 64, "mean": 27.64769744873047, "std": 40.75501251220703, "min": -67.6138687133789, "p10": -11.364003562927243, "median": 14.298001289367676, "p90": 98.18683090209962, "max": 126.46749114990234, "pos_frac": 0.78125, "sample": [100.4442138671875, 21.1552734375, 37.31932067871094, 18.8157958984375, 86.02070617675781, 22.149574279785156, 107.45619201660156, -19.166229248046875, -26.59233856201172, 7.334373474121094, 43.295379638671875, -67.6138687133789, -12.561422348022461, 21.9376220703125, 14.779085159301758, 80.40880584716797, 98.95662689208984, 9.777137756347656, 96.39064025878906, 23.293235778808594, 4.1920013427734375, 11.168418884277344, 4.135560989379883, 45.54167175292969, -6.158393859863281, 111.28743743896484, -8.570026397705078, 37.899967193603516, -4.800811767578125, 126.46749114990234, 89.10162353515625, -12.816280364990234, 3.652536392211914, 100.55947875976562, 6.555440902709961, 5.1753692626953125, 3.0353469848632812, -4.538536071777344, -4.38037109375, 26.64654541015625, 13.816917419433594, 1.897705078125, 89.88085174560547, 0.7312850952148438, 8.870733261108398, 50.838768005371094, 25.821311950683594, 37.59675979614258, 8.360418319702148, 106.74546813964844, -15.577136993408203, 40.52412033081055, 64.84246826171875, 4.784854888916016, 15.38105583190918, 39.84199142456055, 1.769989013671875, -2.854276657104492, 13.38543701171875, 3.846527099609375, 57.324867248535156, -23.347301483154297, 33.83504867553711, -6.619842529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000294.npy"}
|
|
{"epoch": 0.4444444444444444, "step": 295, "batch_size": 64, "mean": 25.159759521484375, "std": 44.5301399230957, "min": -65.4049301147461, "p10": -27.01750183105468, "median": 11.360718727111816, "p90": 85.3521209716797, "max": 127.5115966796875, "pos_frac": 0.71875, "sample": [98.85063171386719, 7.1965484619140625, -4.289207458496094, -4.6739654541015625, 55.39158630371094, 51.75238037109375, 89.26123809814453, 18.332942962646484, 18.32040023803711, 25.0877685546875, 111.01549530029297, -8.225582122802734, 127.5115966796875, 16.056381225585938, 10.44296646118164, 9.234840393066406, 2.915252685546875, 6.894462585449219, -8.636672973632812, 75.25614929199219, 86.24127197265625, -53.79014205932617, -2.541553497314453, 6.00794792175293, 71.93435668945312, 5.9058837890625, 5.877702713012695, -18.22125244140625, 83.27743530273438, 4.702354431152344, -33.721405029296875, -6.120771408081055, 111.81724548339844, 73.26759338378906, 14.366310119628906, -53.168052673339844, 75.55635070800781, 20.50925064086914, 35.74129867553711, 100.323974609375, 58.689300537109375, -65.4049301147461, 0.7411270141601562, -46.8587646484375, 63.55567932128906, 46.24860382080078, -30.787322998046875, -7.041923522949219, -38.40515899658203, 52.51844787597656, -4.636503219604492, 5.991605758666992, 17.607810974121094, 9.73388671875, 12.278470993041992, 10.156229019165039, 75.29551696777344, -8.465682983398438, 13.686744689941406, 0.2550201416015625, 66.30663299560547, 77.76731872558594, 82.01153564453125, -6.680107116699219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000295.npy"}
|
|
{"epoch": 0.4459561602418745, "step": 296, "batch_size": 64, "mean": 30.493009567260742, "std": 46.33211135864258, "min": -76.85610961914062, "p10": -31.54567260742187, "median": 24.954450607299805, "p90": 96.82270812988286, "max": 126.19860076904297, "pos_frac": 0.765625, "sample": [45.68492126464844, 77.9429931640625, 71.240478515625, 13.84980583190918, -0.10640716552734375, 15.864349365234375, 3.127531051635742, 119.01828002929688, -0.4063873291015625, 57.14015579223633, 24.262767791748047, 10.001506805419922, -40.59410858154297, 101.562255859375, 6.118221282958984, 28.968582153320312, 85.86619567871094, 58.70710754394531, 10.514602661132812, 33.55738830566406, 3.5207061767578125, 18.41427993774414, 8.50079345703125, -64.3702392578125, -32.611053466796875, -18.78643798828125, 3.7939910888671875, 35.976112365722656, 15.928596496582031, 33.64096450805664, 19.774459838867188, -33.866024017333984, 39.984588623046875, 6.37298583984375, 115.3963623046875, 25.646133422851562, -1.2475814819335938, -23.57217788696289, 74.73300170898438, -38.91453552246094, 33.140663146972656, 57.093994140625, 110.63396453857422, -76.85610961914062, 83.40647888183594, 30.5245361328125, 73.77903747558594, 71.04826354980469, 63.59312438964844, 69.4175033569336, 76.68238830566406, 7.661376953125, 101.51835632324219, 14.785581588745117, 126.19860076904297, -4.092193603515625, 7.9173431396484375, 105.16460418701172, -19.884963989257812, -29.059783935546875, -38.784759521484375, 69.7305908203125, 59.35821533203125, 47.94062805175781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000296.npy"}
|
|
{"epoch": 0.4474678760393046, "step": 297, "batch_size": 64, "mean": 32.65121078491211, "std": 40.3453369140625, "min": -55.66358947753906, "p10": -6.143429946899412, "median": 23.148399353027344, "p90": 94.36439437866211, "max": 109.60345458984375, "pos_frac": 0.796875, "sample": [64.6717529296875, -27.455894470214844, 7.714759826660156, 68.44344329833984, 30.319122314453125, 25.943801879882812, 14.017189025878906, 94.42478942871094, -2.35003662109375, -55.66358947753906, 68.24754333496094, -1.2453765869140625, 17.256000518798828, 22.467281341552734, 5.321361541748047, 102.42913055419922, 18.14923095703125, 42.16450119018555, 8.95053482055664, -2.3182373046875, 29.891311645507812, 4.299896240234375, -34.45521545410156, -43.20464324951172, -17.786808013916016, 102.48014068603516, 29.78943634033203, 6.6063079833984375, 82.31816864013672, 8.593147277832031, 90.65821075439453, 108.77522277832031, -3.8886566162109375, -0.2149066925048828, 11.34415054321289, 11.918691635131836, 14.753768920898438, -6.997283935546875, 80.87249755859375, 67.55652618408203, 22.510284423828125, 10.540802001953125, 70.97235107421875, 36.41914367675781, 23.786514282226562, 56.32328796386719, 39.454811096191406, 1.1952123641967773, -14.742645263671875, 94.22347259521484, 15.11911392211914, 69.55908203125, 82.38583374023438, 60.13981628417969, -4.151103973388672, 39.48729705810547, 97.455810546875, 39.20465087890625, 2.246856689453125, 109.60345458984375, 1.5145797729492188, 32.2718391418457, 104.72525787353516, 54.63446044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000297.npy"}
|
|
{"epoch": 0.4489795918367347, "step": 298, "batch_size": 64, "mean": 25.53311538696289, "std": 48.5272102355957, "min": -88.00865173339844, "p10": -33.13512954711914, "median": 14.499300003051758, "p90": 98.2300880432129, "max": 122.41839599609375, "pos_frac": 0.75, "sample": [25.000816345214844, -65.26329803466797, 122.41839599609375, 13.877716064453125, 105.83257293701172, 9.784622192382812, 70.77670288085938, 13.976409912109375, -51.7247314453125, 57.31066131591797, 108.88102722167969, 89.4665298461914, -36.340240478515625, 11.742828369140625, 4.895957946777344, 39.55784225463867, 57.92572021484375, -17.499107360839844, 77.36184692382812, 70.14021301269531, 6.308094024658203, -8.762165069580078, 81.79077911376953, 32.56660461425781, 38.26216125488281, 18.69708251953125, -64.72728729248047, 11.557668685913086, 53.54121398925781, 103.1136245727539, 4.166315078735352, 106.46609497070312, -4.83721923828125, 18.926902770996094, 21.136402130126953, 73.8056640625, 6.811864852905273, 101.02926635742188, 4.725564956665039, 5.236488342285156, 0.53564453125, -12.973104476928711, -19.566875457763672, 15.162887573242188, 21.03656768798828, 99.68008422851562, 72.64686584472656, -5.437042236328125, -10.754138946533203, 7.792205810546875, 32.95357894897461, 9.331283569335938, -32.456016540527344, -33.426177978515625, 4.4897003173828125, 69.69519805908203, -59.29652786254883, 56.79798889160156, -15.811614990234375, -88.00865173339844, 94.84676361083984, 15.02219009399414, 3.5332107543945312, 90.38761901855469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000298.npy"}
|
|
{"epoch": 0.4504913076341648, "step": 299, "batch_size": 64, "mean": 23.363990783691406, "std": 48.161651611328125, "min": -80.7428970336914, "p10": -42.72441368103027, "median": 15.835539817810059, "p90": 85.11325836181642, "max": 122.07351684570312, "pos_frac": 0.703125, "sample": [71.86752319335938, 69.06825256347656, -1.8818950653076172, -13.851829528808594, 13.56089973449707, 82.79261779785156, 23.596603393554688, 46.63430404663086, 122.07351684570312, 0.31525421142578125, -42.740882873535156, 46.15650939941406, -80.7428970336914, 5.770042419433594, -67.29243469238281, -13.147529602050781, 10.327224731445312, 94.78687286376953, 41.38673400878906, -42.86334228515625, 81.63031005859375, -19.026531219482422, 20.602115631103516, 49.7137336730957, -41.94392395019531, -62.29924774169922, 3.2229537963867188, 10.038532257080078, 113.89979553222656, -0.8285293579101562, 86.10781860351562, 5.143913269042969, 39.05591583251953, -66.62843322753906, 7.8057403564453125, 20.30224609375, 34.9459228515625, 100.87071228027344, 17.12195587158203, 11.358299255371094, 34.714019775390625, -23.12023162841797, 71.9062271118164, 75.47579193115234, 107.69136047363281, -34.25511932373047, 14.549123764038086, 79.2221450805664, 54.881622314453125, 56.98497009277344, 0.6257038116455078, 20.143943786621094, 54.95045471191406, -42.96925354003906, 11.130279541015625, 66.36824035644531, -2.83154296875, 54.43597412109375, -0.3567676544189453, 11.146804809570312, -0.7758560180664062, 43.01972961425781, -42.68598556518555, 108.13499450683594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000299.npy"}
|
|
{"epoch": 0.4520030234315949, "step": 300, "batch_size": 64, "mean": 28.89687728881836, "std": 46.893089294433594, "min": -69.5416259765625, "p10": -32.03212127685546, "median": 25.872092247009277, "p90": 97.58719863891602, "max": 113.65403747558594, "pos_frac": 0.78125, "sample": [25.1817684173584, 93.48860931396484, 6.954345703125, 34.092315673828125, 1.2051506042480469, 9.70928955078125, 68.1990737915039, 13.14794921875, 4.813650131225586, -66.62448120117188, 107.90373229980469, 26.562416076660156, 65.65425872802734, 100.31211853027344, 108.44883728027344, 62.55009460449219, 98.166748046875, -23.728134155273438, -13.469612121582031, 41.415374755859375, 2.2430286407470703, 57.058494567871094, 29.191513061523438, -22.08991241455078, -38.78764343261719, 54.874359130859375, 71.73867797851562, -20.094314575195312, 20.68846893310547, 49.3198356628418, 7.050022125244141, -69.5416259765625, 61.123687744140625, 1.0439796447753906, -63.22235107421875, 42.85932540893555, 21.238134384155273, 108.88240051269531, 5.2982025146484375, 96.56165313720703, 59.85418701171875, 30.64352798461914, 3.4278564453125, 113.65403747558594, 47.509605407714844, -35.590972900390625, 1.4778175354003906, 73.55001831054688, -6.24578857421875, 63.612491607666016, 60.64013671875, 13.202526092529297, 38.678314208984375, 38.366004943847656, -11.326042175292969, -39.05725860595703, 23.739242553710938, 84.70623779296875, 74.12362670898438, -64.0965805053711, 5.232475280761719, 6.371650695800781, 98.02671813964844, -10.519088745117188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000300.npy"}
|
|
{"epoch": 0.45351473922902497, "step": 301, "batch_size": 64, "mean": 24.680017471313477, "std": 46.31254196166992, "min": -93.8125228881836, "p10": -25.36355323791504, "median": 20.165165901184082, "p90": 93.95572509765626, "max": 147.25759887695312, "pos_frac": 0.75, "sample": [74.42594909667969, 15.077186584472656, 34.177940368652344, 21.33724594116211, 15.248687744140625, -22.519317626953125, 53.42613983154297, 101.50552368164062, 72.2374267578125, 9.144874572753906, 22.72149085998535, 15.969812393188477, 20.48126220703125, -26.165050506591797, -76.74996185302734, 23.760833740234375, 1.9259490966796875, 55.668487548828125, 73.5185317993164, 3.4556198120117188, -39.00355529785156, 30.185932159423828, 99.85623168945312, -10.706031799316406, 42.73060607910156, 3.2077713012695312, -2.0106563568115234, 92.39390563964844, 59.601593017578125, -22.916465759277344, -22.664180755615234, 19.44446563720703, 39.44270706176758, 74.66622924804688, -26.951431274414062, 3.0657806396484375, -23.493392944335938, 147.25759887695312, -43.11700439453125, 20.414348602294922, -7.846372604370117, 19.915983200073242, 34.40946960449219, -51.39365005493164, 4.243896484375, 69.58702087402344, 80.14826965332031, 38.04658508300781, 11.34576416015625, 96.587646484375, 35.410369873046875, 50.1063232421875, 48.14613342285156, -19.332721710205078, -21.477035522460938, 11.725818634033203, 94.62507629394531, 97.13980865478516, 14.909408569335938, 0.8178024291992188, 20.470001220703125, -93.8125228881836, 105.36864471435547, 10.326347351074219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000301.npy"}
|
|
{"epoch": 0.455026455026455, "step": 302, "batch_size": 64, "mean": 18.546327590942383, "std": 44.266754150390625, "min": -88.72087097167969, "p10": -30.71582927703857, "median": 16.181096076965332, "p90": 74.97491149902343, "max": 131.33251953125, "pos_frac": 0.71875, "sample": [42.223106384277344, 112.7635498046875, -1.5590972900390625, 131.33251953125, 29.202362060546875, 4.296052932739258, 93.48741149902344, -13.893470764160156, 105.9754867553711, 11.637542724609375, 31.185470581054688, 20.994400024414062, 10.946908950805664, 3.2657814025878906, 63.7773551940918, 98.7315673828125, 3.4336395263671875, 15.096488952636719, 2.2293319702148438, 8.727142333984375, 15.629375457763672, -71.19384002685547, -62.81959533691406, 28.421249389648438, 25.078781127929688, -18.154767990112305, 23.75804901123047, 33.45769500732422, 0.3502464294433594, -6.1781005859375, -19.845550537109375, 51.55711364746094, 42.84101867675781, 16.732816696166992, -25.51673698425293, 75.16586303710938, 95.15953063964844, -14.556976318359375, 4.5020751953125, 22.232624053955078, 51.28456115722656, 74.52935791015625, 1.7150611877441406, 26.794189453125, 35.782958984375, 28.082550048828125, -78.00485229492188, -17.266990661621094, -6.4168853759765625, 23.591835021972656, 59.56982421875, 28.313064575195312, -7.608924865722656, 10.649040222167969, 21.34872817993164, -53.13494873046875, -37.65374755859375, 67.34175109863281, -9.45086669921875, -32.94401168823242, -88.72087097167969, 14.899566650390625, 54.941627502441406, 28.848474502563477], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000302.npy"}
|
|
{"epoch": 0.4565381708238851, "step": 303, "batch_size": 64, "mean": 35.51524353027344, "std": 50.13938522338867, "min": -75.89732360839844, "p10": -34.48801231384277, "median": 42.64141845703125, "p90": 104.96328735351562, "max": 122.87164306640625, "pos_frac": 0.703125, "sample": [-11.3809814453125, 47.71562194824219, 70.6212158203125, 26.517446517944336, 111.43476867675781, 83.62550354003906, 63.41230773925781, -42.34858703613281, 83.71640014648438, 45.0955810546875, 92.88209533691406, -34.982826232910156, 32.996639251708984, -5.743562698364258, -63.42716979980469, -49.113433837890625, -41.96516418457031, 81.74127197265625, -75.89732360839844, -17.890472412109375, -0.09181022644042969, 11.96522331237793, 122.87164306640625, 55.61762619018555, -35.64729309082031, 54.50037384033203, 77.43940734863281, -29.65625762939453, -3.7832374572753906, 22.7777099609375, 47.04071807861328, 2.52490234375, 10.074668884277344, 20.347991943359375, 15.908447265625, -3.5823326110839844, 104.58218383789062, 7.868095397949219, 79.49397277832031, 63.15165710449219, 14.107872009277344, 77.21728515625, 55.242164611816406, 105.12661743164062, -0.42778778076171875, 76.2872314453125, 113.21012878417969, 29.082748413085938, 58.35603332519531, 84.2374038696289, -12.612319946289062, -4.164728164672852, -33.33344650268555, 40.187255859375, 108.33065795898438, 62.70599365234375, 23.60471534729004, 46.45948791503906, 74.31298065185547, 93.81082153320312, 105.83541870117188, -31.144384384155273, 58.53485107421875, 107.59577941894531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000303.npy"}
|
|
{"epoch": 0.4580498866213152, "step": 304, "batch_size": 64, "mean": 36.22197341918945, "std": 40.27058410644531, "min": -82.36907958984375, "p10": -9.031035995483398, "median": 33.81931114196777, "p90": 92.0260726928711, "max": 112.40951538085938, "pos_frac": 0.828125, "sample": [97.12971496582031, -8.311614990234375, 58.94889831542969, 89.50642395019531, 6.5428924560546875, 41.377559661865234, 36.13282775878906, 47.666107177734375, -15.820051193237305, 31.706058502197266, 7.2333831787109375, 76.82008361816406, 65.81607055664062, 30.451644897460938, 90.66325378417969, 40.519447326660156, -1.5140495300292969, 34.933834075927734, 92.85284423828125, -82.36907958984375, 25.959335327148438, 76.23643493652344, 26.887115478515625, 24.244720458984375, -2.6188735961914062, 8.706039428710938, 7.288263320922852, 32.73823165893555, 39.46888732910156, 97.53768920898438, 55.523704528808594, 2.629598617553711, 92.61013793945312, 50.34850311279297, 2.4686660766601562, -54.055389404296875, 51.23491668701172, 68.697021484375, 22.188392639160156, 34.900390625, 1.1311359405517578, -17.34709358215332, 112.40951538085938, 30.328216552734375, 3.0441017150878906, 8.4686279296875, 44.064964294433594, 58.648414611816406, 87.6630859375, 109.63316345214844, -14.379150390625, 0.7109146118164062, 79.51907348632812, -0.9282646179199219, 79.2311782836914, -9.339359283447266, 42.688316345214844, 104.58660125732422, 55.950531005859375, 27.9498291015625, 23.430038452148438, -18.729320526123047, 29.76995849609375, 76.42179870605469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000304.npy"}
|
|
{"epoch": 0.4595616024187453, "step": 305, "batch_size": 64, "mean": 27.846446990966797, "std": 48.11347198486328, "min": -105.60385131835938, "p10": -35.386479187011716, "median": 22.30913543701172, "p90": 88.3303825378418, "max": 116.74798583984375, "pos_frac": 0.71875, "sample": [81.5218276977539, 13.633750915527344, 8.408279418945312, -41.70830535888672, -6.1447296142578125, 1.8785896301269531, 116.74798583984375, -44.846778869628906, 36.72869873046875, 104.38444519042969, 12.940364837646484, 106.51138305664062, -32.546417236328125, 57.4214973449707, -58.12309265136719, -19.00996208190918, 25.881973266601562, 4.313358306884766, -40.725738525390625, 45.196380615234375, 14.485088348388672, 32.11860656738281, 17.82453155517578, 53.701324462890625, 45.06621551513672, 50.755645751953125, 47.815589904785156, -0.0186614990234375, 53.175628662109375, -105.60385131835938, -0.4822540283203125, 42.13440704345703, 87.69454956054688, 87.4445571899414, -18.114532470703125, 18.846357345581055, 110.05372619628906, -6.12530517578125, 20.971406936645508, 26.482513427734375, 9.385040283203125, 21.066322326660156, 59.31144714355469, 108.64469146728516, 83.58472442626953, -10.362606048583984, 17.857093811035156, -34.38993835449219, 95.41136169433594, 69.36814880371094, 66.3016586303711, 23.55194854736328, 20.839008331298828, 88.6028823852539, 51.473182678222656, 78.95156860351562, -19.8077392578125, 40.1689453125, -63.25210952758789, 84.84160614013672, -16.00349998474121, 19.64751625061035, 72.10581970214844, -35.813568115234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000305.npy"}
|
|
{"epoch": 0.46107331821617537, "step": 306, "batch_size": 64, "mean": 26.485097885131836, "std": 37.41227722167969, "min": -78.21794891357422, "p10": -14.16507053375244, "median": 21.31429386138916, "p90": 71.3425064086914, "max": 132.44277954101562, "pos_frac": 0.828125, "sample": [102.50529479980469, 47.49529266357422, 6.89155387878418, 97.47738647460938, 14.310897827148438, 55.54113006591797, 10.098831176757812, 35.03374481201172, 22.33664321899414, 51.48902130126953, 77.3516616821289, 66.30181884765625, -27.39794158935547, 13.592464447021484, -16.677234649658203, -78.21794891357422, 71.31852722167969, 32.38420104980469, 39.09946060180664, 4.672809600830078, 1.9746017456054688, 54.253517150878906, 64.04133605957031, 11.692304611206055, 13.137472152709961, 8.369514465332031, -15.137998580932617, 52.17570495605469, 8.399185180664062, 33.707427978515625, -22.323165893554688, 71.352783203125, -3.209453582763672, 1.7244281768798828, 47.4951171875, 27.114463806152344, 89.59083557128906, 108.88450622558594, 3.909759521484375, 22.800262451171875, 20.970354080200195, 7.76020622253418, 11.496540069580078, 45.24732208251953, -37.223899841308594, 13.852775573730469, -5.888433456420898, 36.894622802734375, 2.9451446533203125, 25.1107177734375, 61.792701721191406, 54.19537353515625, 3.2757720947265625, 5.3846282958984375, 22.479324340820312, 4.032566070556641, 13.520256042480469, -11.894905090332031, 40.849708557128906, 132.44277954101562, -34.71745300292969, 21.658233642578125, 59.674102783203125, -4.37251091003418], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000306.npy"}
|
|
{"epoch": 0.46258503401360546, "step": 307, "batch_size": 64, "mean": 17.150941848754883, "std": 47.93115997314453, "min": -70.77275085449219, "p10": -39.70662727355957, "median": 6.978604316711426, "p90": 88.7851028442383, "max": 117.62867736816406, "pos_frac": 0.5625, "sample": [73.38396453857422, -19.52123260498047, 9.678142547607422, -21.983253479003906, -58.36250305175781, 54.34849548339844, 117.62867736816406, -22.32268524169922, 95.84967041015625, 53.110107421875, -26.844573974609375, 27.63416290283203, 16.506725311279297, 100.98172760009766, -70.77275085449219, 49.39363098144531, 23.100479125976562, -0.282379150390625, 61.94783020019531, -35.00439453125, 48.2137451171875, -54.83776092529297, -10.279281616210938, -41.8533935546875, 27.51605224609375, -52.7257080078125, -30.716896057128906, 39.45039367675781, 68.25381469726562, -3.3549060821533203, -12.716484069824219, 24.286733627319336, -3.733631134033203, -36.07424545288086, -36.995262145996094, 38.426971435546875, -27.09454345703125, 0.5585861206054688, 60.19450378417969, -13.947555541992188, 4.27906608581543, -23.019241333007812, 2.8752174377441406, 100.66862487792969, 90.85792541503906, -20.670135498046875, 35.190025329589844, 84.3557357788086, 90.68340301513672, 57.93585968017578, -59.585845947265625, 37.264862060546875, 76.60637664794922, 0.11573410034179688, 61.00535583496094, 45.49424743652344, -1.6029930114746094, 97.14369201660156, -34.88623809814453, -16.006309509277344, -40.8686408996582, 37.00475311279297, 67.54224395751953, -5.764457702636719], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000307.npy"}
|
|
{"epoch": 0.46409674981103555, "step": 308, "batch_size": 64, "mean": 25.8104248046875, "std": 44.972599029541016, "min": -98.37969970703125, "p10": -34.141022491455075, "median": 23.73297882080078, "p90": 85.24247512817384, "max": 108.06830596923828, "pos_frac": 0.71875, "sample": [100.88329315185547, 54.38835144042969, 58.74797058105469, -31.762313842773438, 83.08482360839844, 61.26979064941406, -13.939414978027344, 31.539459228515625, 27.213836669921875, -6.141242980957031, 79.84811401367188, 67.23271179199219, 85.87389373779297, 19.379135131835938, 2.9698638916015625, -38.55720138549805, 28.40851593017578, 83.7691650390625, 60.11455535888672, 32.721744537353516, 53.20654296875, -22.862751007080078, 98.2118148803711, 54.03995895385742, 22.22338104248047, 49.292510986328125, -2.4357986450195312, 25.242576599121094, 45.635498046875, -62.35223388671875, 108.06830596923828, -6.111732482910156, 16.955963134765625, -35.16046905517578, 54.55018615722656, 1.384857177734375, 65.10941314697266, -1.9386711120605469, 6.6439056396484375, -98.37969970703125, 9.164306640625, 6.229938507080078, 50.34571838378906, 95.95294189453125, 14.46807861328125, 27.89288330078125, 42.23577117919922, -26.218185424804688, 55.139808654785156, -0.6153221130371094, 79.00081634521484, 2.9847412109375, -11.60235595703125, 64.39794921875, 90.21307373046875, 11.387245178222656, 96.94605255126953, -48.726806640625, -35.50103759765625, 11.39265251159668, 2.651113510131836, -55.30375671386719, 13.08685302734375, -2.0237789154052734], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000308.npy"}
|
|
{"epoch": 0.4656084656084656, "step": 309, "batch_size": 64, "mean": 26.08074188232422, "std": 41.253135681152344, "min": -67.92491149902344, "p10": -17.281287384033202, "median": 25.41902446746826, "p90": 77.78068542480469, "max": 117.19108581542969, "pos_frac": 0.703125, "sample": [65.6685562133789, 117.19108581542969, 10.805213928222656, 77.64717102050781, 36.82914733886719, 0.5384902954101562, 60.86100769042969, -13.978073120117188, 33.10279846191406, 8.600395202636719, 37.506378173828125, -1.1427440643310547, -40.56593704223633, -14.676923751831055, 12.953933715820312, -67.92491149902344, 16.706052780151367, 76.9832992553711, -29.338645935058594, -1.0030536651611328, 3.9585838317871094, 58.8415641784668, 84.99907684326172, -0.2991809844970703, 52.594024658203125, 58.920654296875, -10.771064758300781, 21.06148338317871, 0.11782073974609375, -2.659515380859375, 12.455999374389648, -1.077606201171875, -16.214309692382812, 77.83790588378906, 46.4439697265625, 32.084407806396484, 36.60801696777344, 8.161304473876953, -6.169610977172852, 45.48523712158203, -9.81707763671875, 96.502197265625, 75.26356506347656, 34.489646911621094, 116.62248229980469, -60.90887451171875, 74.4234390258789, 34.89286422729492, 18.439878463745117, 42.194252014160156, -17.738563537597656, 6.41058349609375, 29.776565551757812, 32.47834777832031, 102.2370834350586, 34.21201705932617, 40.726585388183594, 62.44928741455078, -23.748489379882812, -42.5732421875, 8.365234375, -12.293670654296875, 36.938385009765625, 100.68292236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000309.npy"}
|
|
{"epoch": 0.4671201814058957, "step": 310, "batch_size": 64, "mean": 16.290987014770508, "std": 42.496063232421875, "min": -56.74293518066406, "p10": -37.733465576171874, "median": 11.226820945739746, "p90": 70.58680648803713, "max": 121.19308471679688, "pos_frac": 0.59375, "sample": [67.47013854980469, -1.4689788818359375, 24.185726165771484, 31.229421615600586, 23.882553100585938, -47.222572326660156, -17.622230529785156, 71.63125610351562, 72.42635345458984, 1.5276737213134766, -24.271390914916992, -36.20037841796875, 39.01496887207031, 102.73831176757812, 15.692413330078125, -1.0494728088378906, 1.518850326538086, 42.087921142578125, 113.36958312988281, -22.771224975585938, 9.346736907958984, -20.054431915283203, -50.40869903564453, 121.19308471679688, -20.83686065673828, -29.761619567871094, 45.79450225830078, -15.071754455566406, 20.243637084960938, 64.47884368896484, -46.54566955566406, 40.76983642578125, -3.31353759765625, 63.506561279296875, -18.487380981445312, 39.58197021484375, 7.7330322265625, 50.198936462402344, 15.907264709472656, 47.34803771972656, 4.437339782714844, 17.017807006835938, -42.87873840332031, -38.3905029296875, 45.535728454589844, 20.446578979492188, -2.4798202514648438, 41.48814392089844, -2.108123779296875, -12.2342529296875, 9.781318664550781, 86.83604431152344, 68.1497573852539, -56.74293518066406, -0.2771587371826172, -26.98881721496582, -10.3311767578125, 99.83354949951172, 33.78174591064453, -18.55707550048828, -47.55210876464844, 65.39240264892578, 12.672323226928711, 17.999755859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000310.npy"}
|
|
{"epoch": 0.46863189720332576, "step": 311, "batch_size": 64, "mean": 26.693599700927734, "std": 43.612186431884766, "min": -104.39488220214844, "p10": -18.3430233001709, "median": 23.36012840270996, "p90": 75.20859756469727, "max": 182.28445434570312, "pos_frac": 0.75, "sample": [72.19808959960938, -0.4443702697753906, 15.355964660644531, -16.21630096435547, -11.389068603515625, -18.757064819335938, 36.69864273071289, 83.47000885009766, 2.776918411254883, 65.83050537109375, 11.113006591796875, -104.39488220214844, 6.664772033691406, 61.800086975097656, 48.023651123046875, -1.963430404663086, 62.222320556640625, 22.09807586669922, 80.67495727539062, 27.350601196289062, 4.6160888671875, 34.346221923828125, 2.8640708923339844, 27.59148406982422, 75.10601043701172, 5.264728546142578, -26.142276763916016, -17.37692642211914, 3.1870574951171875, 27.442474365234375, -25.668624877929688, 105.76617431640625, 129.62794494628906, 24.470701217651367, 15.041618347167969, 46.36894989013672, 16.63904571533203, 48.127994537353516, 24.426834106445312, 43.74772644042969, 182.28445434570312, 18.136398315429688, -8.875556945800781, 75.2525634765625, -30.170808792114258, 13.674850463867188, 14.735183715820312, -7.927696228027344, 44.23297119140625, 22.29342269897461, 109.18058776855469, 45.396087646484375, 35.31419372558594, -32.084983825683594, 27.481369018554688, 59.555274963378906, 3.3846702575683594, -32.85856628417969, -4.547859191894531, 25.127105712890625, -5.286659240722656, 27.97643280029297, 72.99403381347656, 44.563262939453125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000311.npy"}
|
|
{"epoch": 0.47014361300075586, "step": 312, "batch_size": 64, "mean": 34.544288635253906, "std": 38.5275764465332, "min": -53.840789794921875, "p10": -10.779659652709958, "median": 28.283781051635742, "p90": 90.26222381591802, "max": 112.67182159423828, "pos_frac": 0.84375, "sample": [25.45438003540039, 4.927284240722656, -7.64080810546875, 10.809104919433594, 30.106689453125, 6.893608093261719, 37.00593185424805, -6.946647644042969, 19.877166748046875, 16.270835876464844, 110.99317169189453, 59.77745056152344, 68.5378189086914, 64.54364776611328, 1.8456916809082031, 54.111473083496094, 12.67608642578125, 1.9920291900634766, 12.83803939819336, 63.381011962890625, 98.32306671142578, 20.45380401611328, -0.7610282897949219, 23.60063934326172, 28.577178955078125, 95.60736083984375, 10.37479019165039, -14.909332275390625, 25.134674072265625, 76.84796905517578, 105.4141845703125, -53.840789794921875, 52.89990234375, 65.71688079833984, -45.28491973876953, 27.99038314819336, 72.57354736328125, 24.466407775878906, 69.35623168945312, 106.05320739746094, 112.67182159423828, 2.198009490966797, 32.24525451660156, 41.82542419433594, 71.17256164550781, 17.142959594726562, 27.87635040283203, 77.79023742675781, 42.02958679199219, -12.124881744384766, -14.81304931640625, 56.202392578125, 100.67593383789062, 51.23797607421875, 12.65460205078125, 49.17974853515625, 43.51982116699219, 43.46099853515625, -47.42293930053711, 70.78976440429688, 44.28557586669922, -18.038963317871094, 17.412750244140625, 12.814422607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000312.npy"}
|
|
{"epoch": 0.47165532879818595, "step": 313, "batch_size": 64, "mean": 19.196598052978516, "std": 42.38386154174805, "min": -71.96452331542969, "p10": -29.09229774475097, "median": 12.661306381225586, "p90": 78.80340042114258, "max": 138.51539611816406, "pos_frac": 0.6875, "sample": [0.36362266540527344, 26.531578063964844, 40.949012756347656, 19.100078582763672, 13.175773620605469, 32.818023681640625, 10.698070526123047, 15.628637313842773, 66.92528533935547, -41.669281005859375, 1.8222274780273438, 25.369279861450195, 18.494216918945312, 22.076766967773438, 7.218757629394531, -0.618438720703125, -2.1886215209960938, -21.776805877685547, -0.798583984375, -8.074470520019531, 35.845252990722656, 54.184234619140625, 3.1186466217041016, 112.64102172851562, -71.96452331542969, -32.227508544921875, -1.6012496948242188, 20.12322998046875, 73.57977294921875, 32.019081115722656, 106.68240356445312, 79.212158203125, -49.6728515625, -7.074213027954102, 93.44740295410156, 33.85264587402344, 2.3202381134033203, 6.864780426025391, -4.4653167724609375, 103.37030792236328, 17.238170623779297, 12.146839141845703, 48.25337219238281, -8.626182556152344, 138.51539611816406, 1.6811752319335938, -10.070724487304688, -4.992805480957031, 20.484237670898438, 6.4194793701171875, 77.8496322631836, 33.23692321777344, 30.361923217773438, 0.3044776916503906, -4.752010345458984, -62.71501159667969, 50.004051208496094, -57.28013229370117, 82.26441192626953, -20.68975830078125, -40.77363204956055, 21.246978759765625, 9.596809387207031, 72.57796478271484], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000313.npy"}
|
|
{"epoch": 0.47316704459561604, "step": 314, "batch_size": 64, "mean": 34.41596984863281, "std": 40.995914459228516, "min": -38.6536865234375, "p10": -6.696667861938476, "median": 30.349600791931152, "p90": 95.84505081176758, "max": 123.537353515625, "pos_frac": 0.78125, "sample": [30.512054443359375, 13.211002349853516, 14.496940612792969, 100.32633209228516, 41.9725227355957, 48.57843780517578, 4.739845275878906, 63.84191131591797, -37.07927703857422, 48.99052429199219, -3.615234375, -7.160152435302734, 118.41476440429688, 33.325164794921875, -4.890472412109375, 123.537353515625, 40.22403335571289, 30.18714714050293, 83.3153305053711, -1.7404403686523438, 6.6478271484375, 2.86761474609375, 81.2274169921875, 94.87923431396484, 93.58074951171875, -38.6536865234375, 19.466033935546875, 95.34466552734375, 12.417404174804688, 4.8947906494140625, 64.69522857666016, 34.718170166015625, 25.691364288330078, -3.8267059326171875, 2.008136749267578, 25.018978118896484, 59.317352294921875, -19.690841674804688, 11.404680252075195, 44.05086898803711, 0.40093421936035156, -5.615203857421875, 96.05950164794922, 32.157283782958984, 71.11598205566406, 18.901016235351562, 61.435577392578125, 98.63780212402344, -0.9828414916992188, 96.7408676147461, 32.14312744140625, 92.17826843261719, 49.968597412109375, 12.582908630371094, 63.619354248046875, 24.97503662109375, -19.453872680664062, 34.942970275878906, 13.487831115722656, -0.45758819580078125, 37.514373779296875, -32.603858947753906, -36.349796295166016, 103.97661590576172], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000314.npy"}
|
|
{"epoch": 0.47467876039304613, "step": 315, "batch_size": 64, "mean": 27.152679443359375, "std": 38.2616081237793, "min": -70.78770446777344, "p10": -26.03059539794921, "median": 30.041645050048828, "p90": 78.55562057495118, "max": 114.159912109375, "pos_frac": 0.703125, "sample": [-13.707717895507812, 49.10612487792969, 82.99373626708984, -0.9880847930908203, -0.9104385375976562, 24.042245864868164, -70.78770446777344, -1.4639396667480469, -35.28099822998047, 29.31482696533203, -37.41089630126953, 38.56748962402344, -9.472881317138672, 82.71739959716797, -1.97698974609375, 46.85517120361328, 22.853660583496094, -7.270576477050781, 15.156742095947266, -0.1763744354248047, 34.43195343017578, 39.41398620605469, 1.6443634033203125, 30.75403594970703, 21.763320922851562, 35.840816497802734, 55.27178955078125, -18.24176025390625, 22.202205657958984, 42.37672424316406, 33.998199462890625, 42.69660568237305, 76.9154281616211, 18.199234008789062, 20.858596801757812, 64.92048645019531, 62.84446716308594, -29.368667602539062, -34.93168640136719, 40.04118347167969, 74.61971282958984, -1.25244140625, 68.11558532714844, 81.51976776123047, -17.919540405273438, 47.41560363769531, 52.031585693359375, 114.159912109375, -35.7380485534668, 29.329254150390625, 40.90998840332031, 4.9286956787109375, 64.78173828125, 0.1339092254638672, 49.606258392333984, 11.343578338623047, 79.25856018066406, 37.74751281738281, 92.84408569335938, -29.862585067749023, 72.07768249511719, -1.0834484100341797, 93.78575134277344, 35.226348876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000315.npy"}
|
|
{"epoch": 0.47619047619047616, "step": 316, "batch_size": 64, "mean": 34.32353973388672, "std": 39.463172912597656, "min": -37.47089385986328, "p10": -15.856701660156249, "median": 31.500656127929688, "p90": 91.83494873046875, "max": 113.72108459472656, "pos_frac": 0.765625, "sample": [-12.528694152832031, 13.4691162109375, 53.05690002441406, -15.602020263671875, -32.261444091796875, -7.905632019042969, 36.90850830078125, 25.593345642089844, 11.776710510253906, 76.68206024169922, 101.26921844482422, 80.1287841796875, 56.82758712768555, 85.64010620117188, 40.145599365234375, -20.396530151367188, -37.47089385986328, 13.936180114746094, -16.483335494995117, 58.13139343261719, 18.92332649230957, 92.33206176757812, 12.959728240966797, 20.07476043701172, 27.204078674316406, 27.439281463623047, 39.41869354248047, 4.588531494140625, -13.945426940917969, 32.797706604003906, 77.49735260009766, 81.39283752441406, 65.87266540527344, 113.72108459472656, -9.081132888793945, 56.46776580810547, -15.965850830078125, 58.130714416503906, 52.481651306152344, 55.44957733154297, 90.67501831054688, 43.077842712402344, 26.485721588134766, 3.151256561279297, 95.16349792480469, 2.5252552032470703, -0.0293121337890625, 99.427734375, -13.12322998046875, 52.980926513671875, -20.97486114501953, -28.747026443481445, 4.2714691162109375, -10.525703430175781, 30.20360565185547, 30.008773803710938, 1.9248771667480469, 95.55206298828125, 56.99485778808594, 70.54962921142578, 55.87335968017578, 43.41813659667969, 97.34890747070312, 61.79737091064453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000316.npy"}
|
|
{"epoch": 0.47770219198790626, "step": 317, "batch_size": 64, "mean": 36.151817321777344, "std": 39.50017547607422, "min": -52.899993896484375, "p10": -10.278323364257812, "median": 31.16257095336914, "p90": 92.74092102050783, "max": 124.31745147705078, "pos_frac": 0.8125, "sample": [58.60790252685547, 59.19514083862305, 50.629154205322266, 77.05450439453125, 29.382858276367188, 124.31745147705078, -11.953758239746094, 35.123817443847656, -16.754430770874023, 26.798851013183594, 6.014308929443359, 26.796817779541016, -8.654792785644531, 3.1732254028320312, 37.134613037109375, 10.970897674560547, 76.69651794433594, 21.2158203125, 68.25933074951172, 99.25057983398438, -25.559371948242188, -3.2501754760742188, 30.98822784423828, -38.95576477050781, 31.323265075683594, 28.860870361328125, 70.84767150878906, 47.070167541503906, 63.31394958496094, -52.899993896484375, 106.4786376953125, 12.337936401367188, 52.631370544433594, 110.72737121582031, 93.96820831298828, -36.54478454589844, 31.001876831054688, -1.0947685241699219, 42.64080047607422, 97.71428680419922, 19.355899810791016, 33.877716064453125, 108.42298126220703, -10.463813781738281, 16.761093139648438, 13.374711990356445, 23.285003662109375, 77.14797973632812, 13.657821655273438, 2.9217376708984375, 54.865142822265625, 25.85605239868164, 40.40876007080078, 89.87725067138672, -0.43881988525390625, 41.80223083496094, -9.845512390136719, 41.09820556640625, 81.0313491821289, 77.32441711425781, 3.107044219970703, 80.97283935546875, 19.436355590820312, 35.021087646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000317.npy"}
|
|
{"epoch": 0.47921390778533635, "step": 318, "batch_size": 64, "mean": 25.087976455688477, "std": 50.919368743896484, "min": -72.68806457519531, "p10": -44.41297607421875, "median": 21.98741912841797, "p90": 101.64470520019532, "max": 131.67955017089844, "pos_frac": 0.703125, "sample": [30.707942962646484, 73.52093505859375, 17.879417419433594, -54.675933837890625, 82.42728424072266, 81.2144775390625, 24.481689453125, -50.0052490234375, 40.943359375, 112.56652069091797, 13.057361602783203, -42.68029022216797, 49.89969253540039, 47.98328399658203, 88.26388549804688, 0.8055610656738281, 16.3787841796875, -31.27357292175293, 29.220170974731445, 5.560178756713867, -4.118049621582031, 56.7447509765625, 107.31401824951172, -23.812225341796875, 15.109428405761719, 51.24003219604492, 104.14854431152344, -72.68806457519531, 66.56393432617188, 61.29139709472656, -9.388923645019531, 32.588623046875, -33.43482971191406, 14.877735137939453, 2.3427581787109375, 131.36956787109375, 131.67955017089844, -25.02056121826172, 99.6180419921875, 4.107795715332031, -27.532958984375, 19.493148803710938, 50.13603973388672, -66.98540496826172, -4.007453918457031, 49.89170837402344, 2.4484481811523438, 107.12451171875, 0.2915325164794922, 72.13803100585938, 2.348712921142578, 102.51327514648438, 29.843549728393555, 46.61701965332031, -56.390750885009766, -18.181373596191406, 44.17742919921875, -60.60825729370117, 30.627090454101562, -2.391204833984375, 59.944610595703125, -45.155555725097656, -20.34637451171875, 42.825721740722656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000318.npy"}
|
|
{"epoch": 0.48072562358276644, "step": 319, "batch_size": 64, "mean": 34.94247055053711, "std": 45.98122787475586, "min": -51.22559356689453, "p10": -22.639825439453126, "median": 33.01838493347168, "p90": 99.289697265625, "max": 142.59271240234375, "pos_frac": 0.75, "sample": [36.51591491699219, 107.0352783203125, 18.617645263671875, 84.87518310546875, 16.11683464050293, 83.09161376953125, 5.462100982666016, 31.581893920898438, 27.019119262695312, 38.03650665283203, 85.8022232055664, 83.14842987060547, -3.6108760833740234, -6.816154479980469, 16.319290161132812, 43.22187805175781, 10.354110717773438, 129.9557342529297, 5.808311462402344, 53.308319091796875, 6.150012969970703, -23.74073028564453, 29.7080078125, 42.9269905090332, 129.74827575683594, -3.8706283569335938, 104.47075653076172, 52.083709716796875, 94.92716979980469, 99.33038330078125, -22.791473388671875, -30.41301727294922, 34.78941345214844, 142.59271240234375, -9.445274353027344, 48.33110046386719, -19.926025390625, 69.19235229492188, 64.16181182861328, -38.770572662353516, 99.19476318359375, 33.40382766723633, -8.665756225585938, 1.3984756469726562, 2.4263229370117188, 7.777738571166992, -41.69078063964844, -7.459205627441406, -29.703798294067383, 73.10081481933594, -6.817316055297852, 37.966400146484375, 21.411117553710938, -22.285980224609375, 32.63294219970703, 82.23887634277344, 44.49485778808594, 53.45406723022461, 47.60375213623047, 16.120363235473633, 52.49974060058594, -51.22559356689453, 113.72502899169922, 49.419044494628906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000319.npy"}
|
|
{"epoch": 0.48223733938019653, "step": 320, "batch_size": 64, "mean": 25.913326263427734, "std": 46.68253707885742, "min": -97.91407775878906, "p10": -23.37678623199463, "median": 13.834674835205078, "p90": 92.18739929199221, "max": 161.94091796875, "pos_frac": 0.703125, "sample": [65.8635482788086, -32.572566986083984, 0.9451808929443359, 37.5635986328125, 11.541740417480469, 18.635671615600586, 97.13613891601562, 0.8738822937011719, 56.236900329589844, -2.1946868896484375, -28.472824096679688, 87.09150695800781, 0.5961380004882812, -0.905426025390625, -45.220436096191406, 73.3968505859375, 50.44457244873047, 5.468742370605469, -8.760955810546875, 67.77099609375, -33.06861114501953, -0.4598731994628906, -97.91407775878906, 2.0530033111572266, 7.6058807373046875, 161.94091796875, 124.07967376708984, 74.26343536376953, 60.701080322265625, -22.0584716796875, 23.479156494140625, -20.154417037963867, 14.001289367675781, 13.668060302734375, -0.732696533203125, 94.37135314941406, 7.7233734130859375, -11.634117126464844, -8.273204803466797, -11.38260269165039, 29.552303314208984, 107.14237976074219, 10.32861328125, 31.38677978515625, 15.312530517578125, 28.921369552612305, 104.37530517578125, 36.33900833129883, 6.658439636230469, -23.9417781829834, 2.5264549255371094, -48.98192596435547, 103.97228240966797, -11.971321105957031, 9.041675567626953, 51.020835876464844, 19.244857788085938, 66.82685852050781, 75.74192810058594, 28.573348999023438, 44.86302185058594, -3.429107666015625, 69.62081146240234, 71.68045043945312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000320.npy"}
|
|
{"epoch": 0.4837490551776266, "step": 321, "batch_size": 64, "mean": 28.785263061523438, "std": 45.28849792480469, "min": -72.74876403808594, "p10": -32.62412338256835, "median": 27.326221466064453, "p90": 90.95676574707032, "max": 115.09402465820312, "pos_frac": 0.78125, "sample": [65.27729797363281, 59.535858154296875, 1.5873851776123047, 64.35892486572266, 55.31068420410156, -9.503105163574219, 110.62295532226562, 51.35137176513672, 92.41802978515625, 29.42603874206543, 111.37359619140625, 14.449954986572266, 31.218250274658203, 58.324058532714844, -71.81570434570312, 5.517620086669922, 16.00615692138672, 42.613677978515625, 28.032928466796875, 73.90035247802734, 29.17041015625, 15.344093322753906, -61.424034118652344, -9.775711059570312, 26.61951446533203, -50.72316360473633, -56.16997528076172, -52.12113952636719, 46.0858154296875, 33.46180725097656, 19.472869873046875, 40.115013122558594, 72.58685302734375, 48.5925407409668, 6.900733947753906, 51.92655944824219, 40.1856689453125, -3.6366958618164062, 15.603050231933594, 78.86682891845703, 2.2788009643554688, -6.108722686767578, 14.711711883544922, 75.54683685302734, 13.520950317382812, -72.74876403808594, 87.05440521240234, 104.9380111694336, 96.87832641601562, 45.80562973022461, 9.657257080078125, 13.686395645141602, 16.9010009765625, 96.90699768066406, 87.54714965820312, -10.148521423339844, 18.200050354003906, 26.318344116210938, 115.09402465820312, -36.79442596435547, 17.006324768066406, -2.6663055419921875, 30.477340698242188, -22.893417358398438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000321.npy"}
|
|
{"epoch": 0.4852607709750567, "step": 322, "batch_size": 64, "mean": 38.12841796875, "std": 41.72116470336914, "min": -66.92654418945312, "p10": -11.708669281005857, "median": 33.08979797363281, "p90": 99.66664962768554, "max": 113.06498718261719, "pos_frac": 0.875, "sample": [-12.797979354858398, 49.74961853027344, 106.97864532470703, 13.270416259765625, 82.11695861816406, 15.487403869628906, 99.78597259521484, 61.009681701660156, 82.90849304199219, 33.14259338378906, 30.65302276611328, 39.58600616455078, 99.38822937011719, 113.06498718261719, 54.249420166015625, 6.129413604736328, 33.186431884765625, 33.03700256347656, 5.11798095703125, 13.084136962890625, 52.28028869628906, 3.6474075317382812, -43.119354248046875, 100.67635345458984, 87.02284240722656, 18.603591918945312, 14.497642517089844, 29.54136085510254, 0.248748779296875, 10.339202880859375, -30.519241333007812, 59.23170471191406, 28.037918090820312, 84.32432556152344, 72.71774291992188, 34.23218536376953, 12.757514953613281, 105.87679290771484, 8.890304565429688, -12.4620361328125, 49.346923828125, 39.04492950439453, 2.081787109375, 98.73096466064453, 27.100807189941406, 101.65066528320312, 96.85875701904297, 55.2043342590332, -33.83563232421875, -66.92654418945312, 25.26327133178711, 22.021041870117188, 14.995243072509766, 37.906532287597656, 86.99110412597656, 16.798553466796875, 44.78987121582031, 13.070159912109375, 42.928192138671875, -27.5689697265625, 112.81394958496094, -9.950813293457031, 71.73735046386719, 23.192392349243164], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000322.npy"}
|
|
{"epoch": 0.48677248677248675, "step": 323, "batch_size": 64, "mean": 27.029254913330078, "std": 52.199363708496094, "min": -91.72499084472656, "p10": -25.87557601928711, "median": 9.774009704589844, "p90": 105.08823165893556, "max": 147.74981689453125, "pos_frac": 0.734375, "sample": [11.536003112792969, -51.28138732910156, 49.99107360839844, 20.274169921875, -8.232673645019531, 88.87171936035156, 22.503662109375, 15.419940948486328, -4.091667175292969, 106.38606262207031, 26.0927734375, -20.125638961791992, 102.0599594116211, 25.46595001220703, 85.18370819091797, -23.40085220336914, -45.59761047363281, 62.14569091796875, 70.5157699584961, 7.598876953125, 29.943862915039062, 77.92939758300781, 7.171669006347656, 9.279541015625, 126.13971710205078, 2.2683944702148438, -26.298171997070312, 56.058837890625, 0.990997314453125, -13.965667724609375, 147.74981689453125, -24.88951873779297, 6.882659912109375, -39.04008102416992, 96.60740661621094, 9.056022644042969, -39.533447265625, 85.30314636230469, 59.976165771484375, 2.299417495727539, 10.133781433105469, 52.711326599121094, -10.479660034179688, 1.8001632690429688, 119.43875122070312, 128.6991424560547, 115.5527114868164, -89.54611206054688, 115.18560791015625, 6.021080017089844, -91.72499084472656, -2.8245697021484375, 60.67420196533203, 9.414237976074219, 1.6072998046875, 63.312660217285156, 57.95845031738281, 52.12611389160156, 2.6441802978515625, -10.72781753540039, 19.97897720336914, 5.709613800048828, 9.241867065429688, -12.280445098876953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000323.npy"}
|
|
{"epoch": 0.48828420256991684, "step": 324, "batch_size": 64, "mean": 32.98882293701172, "std": 41.49715805053711, "min": -78.39552307128906, "p10": -10.000151252746578, "median": 25.17487907409668, "p90": 91.94470596313477, "max": 130.9846649169922, "pos_frac": 0.828125, "sample": [56.91228485107422, 30.020050048828125, 21.709579467773438, 7.7113800048828125, -43.31632614135742, 105.27041625976562, 36.34343719482422, 37.08921813964844, -16.769790649414062, -6.458763122558594, 18.194969177246094, 90.50619506835938, 48.503692626953125, 36.714359283447266, 62.88446044921875, 14.88467025756836, 86.84197235107422, -28.0994873046875, 43.207305908203125, 9.634521484375, 42.68809509277344, -11.39849853515625, 9.139801025390625, 48.66761779785156, 27.19814682006836, 16.394906997680664, -22.144065856933594, 12.076278686523438, 130.9846649169922, 17.802066802978516, 16.36084747314453, -78.39552307128906, 43.128562927246094, 95.56461334228516, 23.151611328125, 32.818294525146484, 16.510169982910156, -6.737340927124023, 51.9782600402832, 59.376739501953125, 33.84612274169922, -28.1964111328125, 117.81570434570312, 9.509841918945312, 115.8161392211914, 12.964252471923828, 2.5721511840820312, 79.61325073242188, 35.97783279418945, 92.56121063232422, 116.92831420898438, 1.8227958679199219, 8.947368621826172, 73.04149627685547, 74.62370300292969, 4.467315673828125, 22.82172393798828, 6.6702728271484375, -2.8744735717773438, 30.466228485107422, 76.47370910644531, 73.34561157226562, 17.777969360351562, -2.6567249298095703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000324.npy"}
|
|
{"epoch": 0.4897959183673469, "step": 325, "batch_size": 64, "mean": 32.93022155761719, "std": 50.13009262084961, "min": -133.30494689941406, "p10": -25.870690155029294, "median": 32.976497650146484, "p90": 97.59084930419922, "max": 128.21286010742188, "pos_frac": 0.765625, "sample": [32.27061080932617, 95.92184448242188, 1.8697624206542969, 69.88772583007812, -15.07598876953125, 45.51557922363281, 71.41952514648438, 43.82953643798828, 43.62654113769531, 46.740989685058594, 63.10554504394531, 73.90631103515625, 0.03809928894042969, -23.904434204101562, 43.93721008300781, -8.606014251708984, 93.35090637207031, 71.35552978515625, 55.51404571533203, -54.165985107421875, 128.21286010742188, -26.71337127685547, 85.29324340820312, 1.593719482421875, -1.0602836608886719, 123.93548583984375, 9.094837188720703, 70.3813705444336, -38.57724380493164, 39.85132598876953, -50.11396789550781, 30.09414291381836, 65.89899444580078, 21.4443359375, -50.22571563720703, 56.22370910644531, -28.724529266357422, 78.59679412841797, 39.49989318847656, 124.60275268554688, 34.69640350341797, 11.987852096557617, 16.19256591796875, 126.27456665039062, -133.30494689941406, 18.827247619628906, 21.65771484375, 7.887847900390625, 98.30613708496094, 44.085601806640625, 119.00218200683594, -5.542510986328125, 32.78215026855469, -13.994186401367188, 5.738555908203125, -4.543956756591797, 33.17084503173828, 45.671504974365234, 120.22975158691406, 29.233673095703125, 17.957275390625, 36.70343780517578, 20.576904296875, -5.9082183837890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000325.npy"}
|
|
{"epoch": 0.491307634164777, "step": 326, "batch_size": 64, "mean": 16.623451232910156, "std": 46.69992446899414, "min": -83.60859680175781, "p10": -52.85559349060058, "median": 17.797712326049805, "p90": 74.0636085510254, "max": 114.19454956054688, "pos_frac": 0.59375, "sample": [-21.890174865722656, -7.014598846435547, 21.129596710205078, 14.110713958740234, 38.91899108886719, -83.60859680175781, 110.07394409179688, 72.50833129882812, -3.109130859375, 52.15618133544922, -25.769264221191406, 21.236919403076172, 18.816944122314453, -4.39056396484375, 40.06309127807617, 66.25732421875, 4.471746444702148, 41.73780822753906, -6.605823516845703, 45.460693359375, 74.73015594482422, 0.353302001953125, 22.417465209960938, 56.85215759277344, -41.094337463378906, -10.6708984375, -76.30616760253906, -11.102127075195312, 15.317718505859375, -5.149314880371094, 114.19454956054688, -70.22164154052734, 16.778480529785156, -6.416934967041016, 49.16846466064453, 27.770584106445312, 7.3032989501953125, -12.512481689453125, -28.50750732421875, -31.058258056640625, 76.58116149902344, 99.14498138427734, -55.563140869140625, -4.7048797607421875, 77.34834289550781, 19.915740966796875, 32.61807632446289, -55.96080780029297, 69.77854919433594, 103.01240539550781, -1.0348014831542969, -6.198892593383789, -58.99101257324219, 71.31316375732422, -1.22991943359375, 22.44969940185547, -56.35009765625, 57.34284973144531, 47.281612396240234, 19.807861328125, 71.46575164794922, -46.53798294067383, 34.205413818359375, 61.80624008178711], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000326.npy"}
|
|
{"epoch": 0.4928193499622071, "step": 327, "batch_size": 64, "mean": 27.292985916137695, "std": 50.72620391845703, "min": -100.73716735839844, "p10": -38.26316223144531, "median": 23.61175537109375, "p90": 98.68785247802737, "max": 144.70169067382812, "pos_frac": 0.734375, "sample": [42.48790740966797, -24.09215545654297, -24.662757873535156, 108.02571105957031, 27.581222534179688, 41.819427490234375, -65.29724884033203, 51.07704162597656, -53.958229064941406, 53.498111724853516, 104.04595184326172, 12.972850799560547, -47.92829895019531, 93.49368286132812, 41.754493713378906, 93.14002227783203, 68.01788330078125, 7.320579528808594, 19.388198852539062, -30.891773223876953, 1.4433116912841797, 11.466312408447266, 56.99707794189453, 112.90576171875, 100.91392517089844, 4.7655487060546875, 35.62689208984375, 45.70851135253906, 74.08881378173828, -1.9843292236328125, 25.630935668945312, 9.093097686767578, -42.87726593017578, 81.24778747558594, 72.58915710449219, -3.5905990600585938, 27.77203369140625, 19.788162231445312, -6.75262451171875, 1.7337646484375, 114.44309997558594, 12.264225006103516, -100.73716735839844, 0.884063720703125, 71.95045471191406, 144.70169067382812, -23.635704040527344, 62.40110778808594, 103.74388885498047, 30.13562774658203, 88.42847442626953, -10.69660758972168, 21.592575073242188, 7.4481353759765625, -41.42232894897461, -20.290346145629883, 79.6540298461914, 60.95905303955078, 32.13688659667969, 0.4440574645996094, 0.9367752075195312, -62.304481506347656, 42.86360168457031, -13.508888244628906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000327.npy"}
|
|
{"epoch": 0.4943310657596372, "step": 328, "batch_size": 64, "mean": 32.809654235839844, "std": 42.69374084472656, "min": -57.14281463623047, "p10": -12.091287612915037, "median": 21.621732711791992, "p90": 94.33800659179688, "max": 127.20291137695312, "pos_frac": 0.75, "sample": [94.57859802246094, 5.920625686645508, 10.682485580444336, 24.80042266845703, 74.21174621582031, 8.116537094116211, -11.057392120361328, 65.64950561523438, 41.28173828125, 17.110389709472656, 6.048011779785156, 89.10650634765625, 104.03529357910156, 18.980587005615234, 28.678009033203125, -28.943450927734375, 106.74622344970703, 24.669126510620117, 31.195724487304688, -10.700359344482422, 14.491744995117188, 1.058990478515625, 7.220392227172852, 43.15236282348633, -10.634552001953125, -12.534385681152344, 121.9208984375, 16.466636657714844, -45.484161376953125, 44.181114196777344, -57.14281463623047, 73.6799087524414, 68.74404907226562, 10.742919921875, 88.20852661132812, -14.694122314453125, 15.076263427734375, -1.59490966796875, 127.20291137695312, 10.4427490234375, 88.79544067382812, 85.27253723144531, -5.191917419433594, -16.474252700805664, 14.037307739257812, 68.18022918701172, -0.5512561798095703, 6.8827667236328125, 2.754302978515625, 31.275028228759766, 56.75389862060547, -15.683685302734375, 39.374595642089844, 24.26287841796875, -2.9707565307617188, 49.81020736694336, 63.39558410644531, -2.517232894897461, 94.735107421875, 41.555023193359375, 71.08252716064453, -1.3444976806640625, 110.99254608154297, 93.77662658691406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000328.npy"}
|
|
{"epoch": 0.4958427815570673, "step": 329, "batch_size": 64, "mean": 34.222068786621094, "std": 47.9725456237793, "min": -53.034934997558594, "p10": -22.51208953857422, "median": 22.209625244140625, "p90": 109.44518890380861, "max": 127.8558349609375, "pos_frac": 0.734375, "sample": [15.447698593139648, 11.336627960205078, 35.410125732421875, 39.284027099609375, -22.346755981445312, 35.89837646484375, 124.76620483398438, -12.633895874023438, 86.0303955078125, 18.66020965576172, 72.61051940917969, 42.705230712890625, -12.630651473999023, 66.45618438720703, -0.1389751434326172, 11.058799743652344, 22.724258422851562, 92.25927734375, -22.58294677734375, 28.279869079589844, -3.676422119140625, 21.45432472229004, 119.57792663574219, 55.843536376953125, -53.034934997558594, 115.98246002197266, 127.8558349609375, 70.65833282470703, 23.783050537109375, 6.906707763671875, 96.93639373779297, 83.89537811279297, -32.16393280029297, 84.50833129882812, 10.245199203491211, 6.0860595703125, 110.93121337890625, -3.3036975860595703, 86.20429229736328, 69.23338317871094, 112.83328247070312, 82.65853881835938, 6.892940521240234, -29.79522705078125, 10.639934539794922, 105.97779846191406, 0.8284797668457031, 17.867416381835938, 29.947845458984375, 26.448406219482422, -7.237575531005859, 49.96437072753906, 1.7566680908203125, -18.053150177001953, 121.55681610107422, 78.22015380859375, -25.528383255004883, 61.88927459716797, -8.994064331054688, -47.703857421875, 21.694992065429688, 6.521575927734375, -1.3859119415283203, -37.30604553222656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000329.npy"}
|
|
{"epoch": 0.4973544973544973, "step": 330, "batch_size": 64, "mean": 35.02935028076172, "std": 55.2324333190918, "min": -115.56983184814453, "p10": -29.15935611724853, "median": 27.96980571746826, "p90": 100.74257431030274, "max": 136.89779663085938, "pos_frac": 0.765625, "sample": [1.7196426391601562, -19.208595275878906, -11.603551864624023, -93.1755142211914, 24.969512939453125, -115.56983184814453, 82.03694152832031, 18.75103759765625, -13.969154357910156, 99.4894027709961, 90.257080078125, 34.98863983154297, -68.38251495361328, 4.066307067871094, 30.36685562133789, 104.9344482421875, 117.34069061279297, -44.744537353515625, 136.89779663085938, 101.27964782714844, 62.71332931518555, -2.2503128051757812, 68.185302734375, 3.931203842163086, 95.28643798828125, 95.41056823730469, 7.7029266357421875, 9.2513427734375, -25.399383544921875, 13.340492248535156, 27.855260848999023, 72.42742156982422, 6.719482421875, 47.18331527709961, -55.67631149291992, 89.7367935180664, 77.93010711669922, 92.6033935546875, 73.42497253417969, 26.297588348388672, 0.09488677978515625, 47.42057800292969, 31.317031860351562, 72.12095642089844, 89.84864044189453, 91.59587860107422, 2.892650604248047, 31.48775863647461, -3.7208099365234375, 23.074783325195312, -3.7870025634765625, 28.0843505859375, 115.35529327392578, 76.56889343261719, -12.195236206054688, 17.321792602539062, 17.826034545898438, -30.77077293395996, 126.08897399902344, -49.3953857421875, 76.28893280029297, 87.44242095947266, 19.751556396484375, 120.04801940917969], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000330.npy"}
|
|
{"epoch": 0.4988662131519274, "step": 331, "batch_size": 64, "mean": 46.47313690185547, "std": 55.25177764892578, "min": -118.90702819824219, "p10": -8.583943939208979, "median": 48.206214904785156, "p90": 117.48829803466799, "max": 174.8258819580078, "pos_frac": 0.796875, "sample": [123.21208190917969, 16.15351104736328, 20.284442901611328, 72.89869689941406, 82.92889404296875, 104.37557983398438, 70.75296020507812, 70.20321655273438, 53.73649597167969, -1.4252166748046875, -2.081592559814453, 134.0803985595703, 112.20045471191406, 119.7545166015625, 90.12376403808594, 5.611778259277344, 130.614990234375, 40.11395263671875, 32.15623474121094, 122.19910430908203, 57.66890335083008, 7.853361129760742, 10.702140808105469, 14.376510620117188, -10.901763916015625, 104.9100112915039, 70.14176940917969, 174.8258819580078, 9.939102172851562, 57.903350830078125, 58.74043273925781, 34.343223571777344, 110.99038696289062, 80.84609985351562, -3.1756973266601562, -2.3330307006835938, 38.33207702636719, -59.04901123046875, -50.76399230957031, 76.03255462646484, -33.08404541015625, 128.47349548339844, 45.67289733886719, 0.68963623046875, 26.034013748168945, -118.90702819824219, 50.739532470703125, -3.1470794677734375, -25.634742736816406, 16.68183708190918, 13.583000183105469, 82.49765014648438, 87.54617309570312, 27.239974975585938, 38.479156494140625, 6.487037658691406, -1.3176374435424805, 92.9510498046875, 70.55400085449219, -58.4393424987793, 107.36168670654297, 61.10847854614258, 79.3299560546875, 100.10448455810547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000331.npy"}
|
|
{"epoch": 0.5003779289493575, "step": 332, "batch_size": 64, "mean": 37.30747985839844, "std": 47.78764724731445, "min": -50.68118667602539, "p10": -5.900407409667968, "median": 20.51661777496338, "p90": 113.51650619506837, "max": 142.1309814453125, "pos_frac": 0.828125, "sample": [-30.948505401611328, 5.7638702392578125, 7.696403503417969, 3.1237106323242188, 20.98325538635254, 71.72322845458984, 18.79998016357422, 23.95677375793457, 12.0322265625, 30.826213836669922, 5.035209655761719, 142.1309814453125, 92.45561218261719, 58.92840576171875, 114.31204223632812, 9.427703857421875, 21.0125732421875, 0.7204418182373047, 63.459564208984375, 23.410690307617188, 68.61319732666016, 3.06280517578125, -50.68118667602539, 118.48011779785156, 73.16716003417969, 78.1353530883789, -33.96623229980469, 20.04998016357422, 6.3204498291015625, 6.517326354980469, 16.354835510253906, 111.6602554321289, 19.583984375, -6.0451202392578125, -5.562744140625, 41.073638916015625, -1.687601089477539, 135.08184814453125, -45.623382568359375, 1.9388980865478516, 66.12158203125, 100.5784912109375, 59.280792236328125, 1.851511001586914, 22.307327270507812, -1.1707382202148438, 51.88188171386719, 86.19267272949219, 86.03024291992188, 38.068084716796875, 122.74543762207031, 16.956275939941406, 47.809757232666016, -2.7733154296875, 12.180274963378906, -17.206253051757812, 116.67048645019531, 4.934619903564453, 19.45477294921875, 104.96440124511719, 135.82179260253906, 80.69416809082031, -28.666370391845703, 11.626708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000332.npy"}
|
|
{"epoch": 0.5018896447467877, "step": 333, "batch_size": 64, "mean": 41.38634490966797, "std": 46.83847427368164, "min": -64.6052474975586, "p10": -6.172984313964844, "median": 34.341514587402344, "p90": 113.73513641357424, "max": 141.94375610351562, "pos_frac": 0.796875, "sample": [17.349754333496094, -7.290334701538086, 14.987808227539062, 69.454833984375, 23.117843627929688, -0.21129608154296875, 39.1130485534668, 54.772705078125, 6.414608001708984, -6.200958251953125, 57.32349395751953, 59.1352424621582, 55.186607360839844, 78.97064208984375, 96.45521545410156, 19.757492065429688, 42.8878173828125, 107.3143310546875, 124.95938110351562, 128.09886169433594, -2.9159698486328125, 31.040725708007812, 33.58393096923828, 21.6446533203125, 42.484588623046875, -35.36321258544922, -20.559955596923828, 46.885047912597656, 11.350906372070312, 40.40149688720703, 101.79884338378906, 116.21997833251953, -6.1077117919921875, 40.69287872314453, 65.89873504638672, 32.98293685913086, 141.94375610351562, 1.4967727661132812, 6.760097503662109, 15.95477294921875, -1.6157093048095703, 30.504165649414062, -64.6052474975586, 9.969057083129883, 35.099098205566406, -36.23472595214844, 136.22265625, 19.30864715576172, -3.4753952026367188, 35.711082458496094, 10.588851928710938, -0.51275634765625, 67.1700210571289, 121.44296264648438, 13.373895645141602, -13.186180114746094, 63.18999481201172, 7.117317199707031, 107.93717193603516, 44.25469207763672, 125.11456298828125, 66.0807113647461, 103.07345581054688, 104.4073715209961], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000333.npy"}
|
|
{"epoch": 0.5034013605442177, "step": 334, "batch_size": 64, "mean": 28.861309051513672, "std": 55.15010070800781, "min": -102.08997344970703, "p10": -31.461503601074217, "median": 20.690322875976562, "p90": 107.6853813171387, "max": 151.6025390625, "pos_frac": 0.6875, "sample": [-47.76850128173828, -102.08997344970703, 50.89574432373047, 63.83985137939453, 50.61248779296875, -12.710830688476562, -71.58029174804688, 82.10399627685547, 16.38085174560547, -86.62743377685547, 2.7214813232421875, 151.6025390625, -12.315624237060547, 40.86542510986328, 5.0078125, 63.286705017089844, -7.7508544921875, -17.294565200805664, 26.529022216796875, 72.13285827636719, 60.46598815917969, 130.37742614746094, 25.81116485595703, 48.58747863769531, 67.7698974609375, -46.05326461791992, 21.6468505859375, 2.448091506958008, 94.98927307128906, 8.82156753540039, -4.120325088500977, -1.8179779052734375, 118.89739227294922, -67.67642211914062, 66.3163070678711, 114.64434814453125, 51.62286376953125, 4.5002593994140625, 31.246002197265625, -2.1796951293945312, 9.834800720214844, -20.040504455566406, -4.0706024169921875, 1.8066864013671875, 99.72611236572266, -2.9121932983398438, 0.753265380859375, 15.75274658203125, -28.73126220703125, 94.255859375, 82.27474975585938, 47.995643615722656, -32.63160705566406, 51.100990295410156, -27.428037643432617, 19.733795166015625, 35.79817199707031, 118.69839477539062, -3.849517822265625, 10.930093765258789, 78.94561767578125, 52.21892547607422, 141.72714233398438, 111.09649658203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000334.npy"}
|
|
{"epoch": 0.5049130763416477, "step": 335, "batch_size": 64, "mean": 33.206573486328125, "std": 54.390350341796875, "min": -77.3192138671875, "p10": -32.60515213012695, "median": 21.280014038085938, "p90": 119.96056900024415, "max": 139.3416748046875, "pos_frac": 0.734375, "sample": [65.74021911621094, 54.85112762451172, -28.604013442993164, 139.3416748046875, 18.129669189453125, 77.26284790039062, 36.40791320800781, 1.258188247680664, 117.14773559570312, 58.573822021484375, 128.56454467773438, 112.79412078857422, 20.527664184570312, 44.76707458496094, 129.07937622070312, 116.28236389160156, 8.103431701660156, 41.76408386230469, 10.823638916015625, 123.47293090820312, -39.3755989074707, -1.678924560546875, 35.91351318359375, 122.78300476074219, -34.919158935546875, 17.66217041015625, 66.9629898071289, -30.796485900878906, -33.38029479980469, 55.375038146972656, 91.32708740234375, 4.466194152832031, 6.422477722167969, 42.47454071044922, 83.89218139648438, -14.935649871826172, 121.16606903076172, -20.753437042236328, 18.838172912597656, 122.9746322631836, 6.046112060546875, -0.7851734161376953, 20.167343139648438, 27.790794372558594, 38.011322021484375, 88.37378692626953, -5.967796325683594, -60.453125, 27.108829498291016, 2.513181686401367, 22.032363891601562, -22.211902618408203, -7.546195983886719, -70.8658447265625, -77.3192138671875, -51.73112487792969, 100.45463562011719, 59.60210037231445, 4.1381988525390625, 67.8514404296875, -12.173660278320312, 6.588966369628906, 18.0113525390625, 54.877349853515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000335.npy"}
|
|
{"epoch": 0.5064247921390779, "step": 336, "batch_size": 64, "mean": 33.61456298828125, "std": 54.78846740722656, "min": -80.03787994384766, "p10": -30.65946807861328, "median": 23.786460876464844, "p90": 116.67548446655275, "max": 184.3339385986328, "pos_frac": 0.75, "sample": [-32.141357421875, 6.35736083984375, 106.7729721069336, 25.675277709960938, 149.38873291015625, 101.13424682617188, 46.40724563598633, -3.4995651245117188, 124.26465606689453, 2.206754684448242, 1.6901569366455078, -8.518655776977539, 118.27361297607422, -8.235137939453125, 112.94651794433594, 25.242321014404297, 3.0829620361328125, 67.08006286621094, 137.25543212890625, -52.905242919921875, -28.3973388671875, 39.29474639892578, 126.18047332763672, 32.60912322998047, 1.18621826171875, 82.11781311035156, 22.33060073852539, -80.03787994384766, -22.704376220703125, -53.19440460205078, 120.96485900878906, -38.198097229003906, 31.80536460876465, 26.591094970703125, 3.1094398498535156, 6.70277214050293, 2.2461395263671875, -12.893226623535156, 39.03722381591797, 92.72895812988281, 17.750389099121094, -12.985984802246094, 39.291961669921875, 18.12420654296875, 97.21424865722656, 27.05023193359375, 43.65765380859375, 3.872589111328125, 22.240262985229492, 86.19134521484375, 8.859054565429688, -5.72308349609375, 49.93292236328125, 97.35134887695312, 184.3339385986328, 52.61399841308594, -36.4818115234375, 12.8369140625, 56.004417419433594, -11.531494140625, 27.877410888671875, 68.98104858398438, 21.541412353515625, -31.628952026367188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000336.npy"}
|
|
{"epoch": 0.5079365079365079, "step": 337, "batch_size": 64, "mean": 26.96866798400879, "std": 49.676448822021484, "min": -93.74345397949219, "p10": -27.99220085144043, "median": 21.73798370361328, "p90": 87.77854461669922, "max": 150.527587890625, "pos_frac": 0.71875, "sample": [4.648551940917969, -57.037933349609375, -93.74345397949219, -17.803237915039062, 32.60986328125, 85.13153839111328, -14.35151481628418, 91.9134521484375, 15.938873291015625, 1.6730117797851562, 100.0765380859375, 13.942710876464844, 43.64537811279297, -82.30938720703125, -29.90673828125, -12.702827453613281, 28.133926391601562, 78.50918579101562, 16.716285705566406, -16.462661743164062, 14.264991760253906, -0.5822620391845703, -28.069774627685547, -49.7786979675293, 70.04638671875, 59.950599670410156, -14.934234619140625, 45.70487594604492, 44.411651611328125, 130.6684112548828, 30.60334014892578, 4.364387512207031, 103.89945220947266, -8.370929718017578, 87.49313354492188, 40.330657958984375, 83.93943786621094, -1.0817108154296875, 8.37115478515625, 66.15440368652344, 107.92821502685547, 8.895011901855469, 80.59230041503906, 23.97857666015625, 40.544219970703125, -15.907852172851562, 50.102088928222656, 14.209884643554688, 37.95280456542969, 19.497390747070312, 64.11222839355469, 150.527587890625, 1.6005630493164062, -11.08258056640625, 27.10797119140625, 76.43031311035156, 7.805917739868164, -27.811195373535156, 74.47618103027344, 55.95929718017578, -66.67239379882812, 9.76092529296875, 87.90086364746094, 32.0796012878418], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000337.npy"}
|
|
{"epoch": 0.509448223733938, "step": 338, "batch_size": 64, "mean": 28.09157943725586, "std": 52.39634323120117, "min": -106.30271911621094, "p10": -26.762415313720698, "median": 16.3034725189209, "p90": 101.97246398925782, "max": 130.14154052734375, "pos_frac": 0.75, "sample": [-32.5682373046875, 17.33908462524414, 98.20320129394531, 86.98599243164062, 101.44265747070312, 72.84774017333984, -68.47731018066406, 89.33113861083984, 17.058395385742188, 103.44535827636719, 130.14154052734375, 11.529731750488281, -6.647300720214844, 32.53501892089844, 5.334403991699219, -62.738868713378906, 12.124290466308594, -78.75381469726562, 1.2341766357421875, 2.6806716918945312, -14.774284362792969, 102.19952392578125, 3.2686729431152344, -0.3233509063720703, 6.188072204589844, 59.70404052734375, -106.30271911621094, 72.93541717529297, 60.45231628417969, 57.008880615234375, 5.303123474121094, 57.36260986328125, 12.666202545166016, 19.09719467163086, 43.38829803466797, 57.260955810546875, 15.54854965209961, 10.780548095703125, 122.0807113647461, 5.8602447509765625, 9.863243103027344, 95.52888488769531, -28.828826904296875, 83.85810852050781, 1.5013084411621094, 22.308700561523438, -19.93785858154297, -21.94078826904297, 18.401451110839844, -18.548206329345703, 21.57281494140625, 122.20428466796875, 9.061786651611328, 2.0740089416503906, 41.930301666259766, 59.53622817993164, -6.1497955322265625, 106.30874633789062, 96.57176971435547, -43.65689468383789, -13.709558486938477, 22.652236938476562, 120.56562042236328, -6.059383392333984], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000338.npy"}
|
|
{"epoch": 0.5109599395313681, "step": 339, "batch_size": 64, "mean": 32.250999450683594, "std": 48.23143005371094, "min": -79.91911315917969, "p10": -20.494920921325683, "median": 29.684809684753418, "p90": 105.64643173217777, "max": 133.73583984375, "pos_frac": 0.75, "sample": [43.93963623046875, 31.633575439453125, 7.5096893310546875, -15.278892517089844, -9.439582824707031, 109.55326843261719, 15.102119445800781, 29.865732192993164, 11.02109146118164, -43.38020324707031, 46.36772155761719, -19.516220092773438, 8.92608642578125, 29.503887176513672, 33.15793228149414, -2.1408462524414062, -57.190223693847656, 85.89715576171875, 85.09497833251953, 4.665933609008789, 96.53047943115234, 20.457496643066406, 40.64115905761719, 68.40213775634766, 10.012962341308594, 110.99287414550781, 3.103748321533203, 18.208240509033203, -28.31464385986328, 56.64554214477539, 50.580379486083984, 68.35838317871094, 118.51045227050781, 128.23785400390625, 62.737648010253906, 6.820774078369141, 66.77069854736328, -19.597793579101562, 65.16824340820312, 48.77723693847656, 116.09475708007812, 31.615402221679688, -20.879404067993164, -10.012657165527344, 25.8082275390625, -3.2385101318359375, 18.2032470703125, 50.72314453125, -64.20594024658203, -32.651611328125, 11.99305534362793, 8.498859405517578, 70.15393829345703, 93.91780090332031, -79.91911315917969, 133.73583984375, -4.923797607421875, 34.94452667236328, 4.081024169921875, 74.06266021728516, 59.630828857421875, 36.53470993041992, -0.3201904296875, 121.88047790527344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000339.npy"}
|
|
{"epoch": 0.5124716553287982, "step": 340, "batch_size": 64, "mean": 34.10358428955078, "std": 59.364532470703125, "min": -113.11793518066406, "p10": -38.567757415771474, "median": 24.162578582763672, "p90": 122.67723083496094, "max": 137.28488159179688, "pos_frac": 0.640625, "sample": [46.07859802246094, 114.30635070800781, 20.780689239501953, 51.21113204956055, -8.769088745117188, 131.03982543945312, -1.475961685180664, -5.773721694946289, -25.120826721191406, 126.99014282226562, 9.623291015625, 68.64266967773438, 36.776092529296875, 82.97908020019531, 126.17040252685547, 85.27418518066406, 13.327859878540039, -50.08433532714844, 47.322296142578125, 62.3177490234375, -14.81492805480957, -9.220392227172852, -52.60660171508789, 59.89258575439453, -10.17633056640625, 123.42979431152344, -9.187698364257812, 108.96456909179688, -13.095787048339844, 120.92124938964844, -63.95878601074219, 100.63761901855469, 10.011871337890625, 32.54533386230469, 109.01014709472656, 54.58608627319336, 110.80963897705078, 1.576517105102539, -3.6199951171875, 119.36827087402344, 71.35884094238281, -16.635910034179688, -6.453834533691406, 124.389404296875, 45.99729537963867, -12.450603485107422, -43.413787841796875, -46.550601959228516, 3.8531570434570312, 100.9464111328125, -9.135971069335938, 41.30544662475586, 67.5648422241211, -13.522119522094727, -113.11793518066406, -48.697296142578125, 124.68543243408203, 11.900955200195312, 42.71826171875, 137.28488159179688, 27.54446792602539, 1.7619247436523438, -27.260353088378906, 11.867088317871094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000340.npy"}
|
|
{"epoch": 0.5139833711262283, "step": 341, "batch_size": 64, "mean": 37.2657470703125, "std": 55.70365905761719, "min": -92.72731018066406, "p10": -28.739953613281248, "median": 32.8682804107666, "p90": 116.17916870117189, "max": 129.10775756835938, "pos_frac": 0.75, "sample": [-64.7467041015625, 65.97978973388672, 102.5783462524414, 50.36271667480469, 92.90760803222656, 58.23063659667969, 97.87830352783203, 128.1436767578125, 109.36793518066406, -25.881378173828125, 111.75989532470703, 16.834312438964844, -68.60321807861328, -3.8098182678222656, 12.322481155395508, 3.3845901489257812, 92.93537139892578, 4.460044860839844, 91.78407287597656, 129.10775756835938, 86.84904479980469, 118.0731430053711, 119.19168090820312, 57.392601013183594, -61.760009765625, 120.15123748779297, 76.56098937988281, 7.510387420654297, -11.321189880371094, -8.340286254882812, 35.6146125793457, 13.566146850585938, -14.206268310546875, -1.7771224975585938, 19.837631225585938, 105.59685516357422, 39.793514251708984, 13.972715377807617, 40.948204040527344, -37.898841857910156, -25.519569396972656, -36.6824951171875, 38.11884307861328, 103.89654541015625, 7.56976318359375, 123.52003479003906, 30.1219482421875, 59.60615539550781, 68.01564025878906, 124.58905792236328, 47.49188232421875, 14.977508544921875, 76.50848388671875, -29.965057373046875, 7.0891571044921875, 81.03727722167969, 49.54920196533203, 22.453750610351562, 0.3555431365966797, -15.447463989257812, -16.71550750732422, 13.449295043945312, 8.96388053894043, -92.72731018066406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000341.npy"}
|
|
{"epoch": 0.5154950869236583, "step": 342, "batch_size": 64, "mean": 40.62333297729492, "std": 54.21014404296875, "min": -69.745361328125, "p10": -18.393797111511226, "median": 27.4831600189209, "p90": 124.73884658813478, "max": 164.0762176513672, "pos_frac": 0.75, "sample": [20.83551788330078, 24.26995849609375, -4.8137054443359375, -5.448284149169922, 2.341768264770508, 105.2005615234375, 19.464435577392578, 92.57403564453125, -30.081924438476562, 25.73858642578125, 1.5629596710205078, 30.54730224609375, -2.776479721069336, 123.35311126708984, 14.346931457519531, -34.98628234863281, 0.8975963592529297, -69.745361328125, -1.7994403839111328, -14.465904235839844, 31.974384307861328, 113.14079284667969, 84.51165771484375, 118.1771011352539, -21.569080352783203, 46.85797882080078, 28.788589477539062, 51.96076202392578, 86.39767456054688, 42.39889144897461, 34.813053131103516, 35.61056137084961, 19.33843994140625, 103.32281494140625, 32.2119140625, -15.838434219360352, 94.91132354736328, 3.418365478515625, -19.48895263671875, 117.04080200195312, 10.342926025390625, 129.69540405273438, 60.734954833984375, 59.621482849121094, -31.730722427368164, 132.98837280273438, 3.7752761840820312, -5.210731506347656, 164.0762176513672, 135.3364715576172, 73.10762786865234, -37.95642852783203, 19.708885192871094, 127.46466064453125, -11.69039535522461, 125.33273315429688, 57.14475631713867, -2.9830245971679688, 6.097221374511719, 57.873714447021484, 63.729164123535156, 26.177730560302734, 5.6024627685546875, 145.66061401367188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000342.npy"}
|
|
{"epoch": 0.5170068027210885, "step": 343, "batch_size": 64, "mean": 24.843975067138672, "std": 49.083457946777344, "min": -96.84977722167969, "p10": -16.2294334411621, "median": 17.846792221069336, "p90": 82.04468231201173, "max": 168.21701049804688, "pos_frac": 0.734375, "sample": [46.53645324707031, 6.217262268066406, -19.02442169189453, 4.6221160888671875, -6.290313720703125, 60.193519592285156, -4.763916015625, -1.1824359893798828, -6.6639251708984375, -5.0128631591796875, -90.76236724853516, 17.4765625, 30.696212768554688, 53.38697814941406, 168.21701049804688, -5.051723480224609, 104.27162170410156, 126.32025146484375, 78.94096374511719, 68.14271545410156, 26.65557861328125, 5.43275260925293, 57.192970275878906, 75.81824493408203, 6.052684783935547, 43.56779861450195, 12.49462890625, -42.246070861816406, 28.759803771972656, -4.980756759643555, -64.6948471069336, 34.21937561035156, 87.94593811035156, 18.217021942138672, 11.98486328125, 6.647758483886719, 66.313720703125, 83.37484741210938, -96.84977722167969, 117.71516418457031, -9.707794189453125, 32.634002685546875, 37.450279235839844, 106.680419921875, 29.918914794921875, 12.135581970214844, 45.25984191894531, 21.952190399169922, 8.012306213378906, 76.2771987915039, 75.27501678466797, 8.194389343261719, 4.807090759277344, 57.193519592285156, 38.12042999267578, 27.13201904296875, -3.8390350341796875, 3.6181488037109375, 40.60173034667969, 6.472496032714844, -94.0205078125, -9.47015380859375, -31.469707489013672, 6.894552230834961], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000343.npy"}
|
|
{"epoch": 0.5185185185185185, "step": 344, "batch_size": 64, "mean": 38.47479248046875, "std": 51.59884262084961, "min": -81.3990249633789, "p10": -19.473090171813965, "median": 30.959251403808594, "p90": 107.15454330444337, "max": 174.8153076171875, "pos_frac": 0.765625, "sample": [-49.13310241699219, 100.25294494628906, 12.727333068847656, 30.583110809326172, 111.05501556396484, -1.277425765991211, 8.55855941772461, 31.335391998291016, 16.964515686035156, 43.480316162109375, 3.1843719482421875, 14.868301391601562, 17.203153610229492, 86.32513427734375, 60.12406921386719, 50.88521957397461, 66.90995788574219, 174.8153076171875, 8.932716369628906, 64.84268188476562, -18.375261306762695, -19.943588256835938, -30.499435424804688, 134.16580200195312, 6.044940948486328, -8.535255432128906, -36.39942932128906, 2.7820968627929688, 131.0651397705078, 57.18001174926758, -26.896560668945312, 114.39752960205078, -2.4467697143554688, 71.86610412597656, 44.38224792480469, 110.11738586425781, 108.38130187988281, 15.285896301269531, 63.696685791015625, -59.664649963378906, -0.2857780456542969, 7.083248138427734, -9.047599792480469, 2.0704479217529297, -7.059837341308594, -8.911920547485352, 104.29210662841797, 98.91600036621094, 44.209625244140625, 53.16361618041992, 69.5999526977539, 26.801345825195312, 76.89032745361328, 17.899145126342773, 98.1707992553711, 12.287569046020508, 34.63214111328125, 88.86802673339844, 9.685623168945312, 90.95572662353516, -81.3990249633789, 71.99783325195312, 83.40958404541016, 68.91596984863281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000344.npy"}
|
|
{"epoch": 0.5200302343159486, "step": 345, "batch_size": 64, "mean": 35.34967803955078, "std": 52.99925994873047, "min": -89.11937713623047, "p10": -27.85866050720215, "median": 25.948780059814453, "p90": 101.18712615966798, "max": 158.74374389648438, "pos_frac": 0.796875, "sample": [38.35370635986328, 12.057441711425781, 134.53616333007812, -70.18732452392578, 16.384052276611328, 4.3813323974609375, 13.697967529296875, -36.13798522949219, 18.95624542236328, 3.1334915161132812, 96.01119995117188, 96.05317687988281, 72.80309295654297, 74.86824035644531, -37.94467544555664, 72.74752044677734, -51.030479431152344, 31.68708038330078, 25.74640655517578, 74.80435180664062, 158.74374389648438, 8.462011337280273, 73.10292053222656, 52.593292236328125, 87.84556579589844, 4.4597015380859375, 0.26329803466796875, 33.08354949951172, 15.015029907226562, 26.981529235839844, 67.5069351196289, 128.66592407226562, 92.51092529296875, 53.65935516357422, 32.393890380859375, -75.1207504272461, 129.71043395996094, 130.743896484375, 64.41752624511719, -10.40958023071289, 103.38739013671875, 17.224937438964844, -28.126373291015625, 95.02180480957031, 34.93994140625, -21.196331024169922, 25.591468811035156, -11.351509094238281, 26.151153564453125, 19.480491638183594, 25.022354125976562, 13.495697021484375, -9.223526000976562, 24.869873046875, -1.3208999633789062, 70.97607421875, 0.24439048767089844, -89.11937713623047, -27.233997344970703, 17.51239776611328, 56.967437744140625, 53.83955383300781, 84.43609619140625, 115.24002075195312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000345.npy"}
|
|
{"epoch": 0.5215419501133787, "step": 346, "batch_size": 64, "mean": 41.12611770629883, "std": 56.59311294555664, "min": -110.92037200927734, "p10": -25.34870567321777, "median": 29.82209014892578, "p90": 120.58455886840821, "max": 141.70620727539062, "pos_frac": 0.765625, "sample": [88.73790740966797, 101.6402587890625, 65.87115478515625, 97.92021942138672, 79.02998352050781, 41.34410095214844, 23.865005493164062, 64.13386535644531, 99.51251983642578, 126.4512939453125, 121.89046478271484, 4.640071868896484, 56.876678466796875, 104.45438385009766, -27.305248260498047, 128.7204132080078, -71.77314758300781, 27.051265716552734, -20.78343963623047, 35.59981155395508, 58.39308547973633, 53.0362548828125, 29.694107055664062, 30.139976501464844, 108.06326293945312, 141.70620727539062, -110.92037200927734, 29.9500732421875, 7.988018035888672, -61.68194580078125, 5.579742431640625, -40.89446258544922, 105.98092651367188, -4.0413055419921875, 21.59121322631836, 26.10852813720703, 6.472816467285156, -11.017677307128906, 97.61611938476562, 107.76759338378906, 124.65231323242188, 84.0718994140625, -2.2666015625, 125.43995666503906, 25.05280303955078, 20.837013244628906, 79.23768615722656, 38.37663269042969, -37.75572967529297, 59.517242431640625, 12.586174011230469, 14.710212707519531, 25.656850814819336, -5.367046356201172, -40.303550720214844, 11.513381958007812, 15.849693298339844, 89.86357116699219, -17.72978973388672, 136.26290893554688, -9.063968658447266, -4.07353401184082, 18.056350708007812, 117.53744506835938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000346.npy"}
|
|
{"epoch": 0.5230536659108088, "step": 347, "batch_size": 64, "mean": 31.370941162109375, "std": 51.241493225097656, "min": -67.8994369506836, "p10": -30.021241569519038, "median": 18.277095794677734, "p90": 109.27608337402344, "max": 146.46261596679688, "pos_frac": 0.75, "sample": [37.09619903564453, 2.3242759704589844, 38.407630920410156, 51.81837463378906, -67.8994369506836, 16.91907501220703, 63.833648681640625, 33.764427185058594, 75.19317626953125, 107.74085998535156, 4.4495086669921875, 101.12973022460938, 18.311111450195312, -48.326316833496094, -36.139060974121094, 141.66888427734375, 21.222091674804688, 39.6960334777832, -13.738128662109375, 5.368537902832031, 15.826644897460938, 4.229646682739258, 56.31590270996094, 4.90625, -36.79951477050781, 18.140018463134766, 71.86243438720703, 93.82875061035156, 113.20991516113281, -25.522552490234375, 18.243080139160156, -11.537839889526367, -33.64618682861328, 84.61995697021484, -12.676103591918945, 43.16184616088867, -11.67110824584961, 73.0766830444336, 3.12310791015625, 27.946556091308594, 44.335750579833984, 4.4025115966796875, -18.85980224609375, 64.43840026855469, 6.722723007202148, 122.76309204101562, -31.949251174926758, -16.834434509277344, 146.46261596679688, 132.46104431152344, 109.93403625488281, 88.88960266113281, 17.183792114257812, 21.70440673828125, -66.57785034179688, 48.86004638671875, 55.05479431152344, 7.698657989501953, 1.0179023742675781, -0.368194580078125, -13.924304962158203, 3.5180816650390625, 116.38914489746094, 74.93936157226562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000347.npy"}
|
|
{"epoch": 0.5245653817082389, "step": 348, "batch_size": 64, "mean": 39.01272201538086, "std": 59.43128204345703, "min": -77.0037841796875, "p10": -22.819043731689455, "median": 24.82831573486328, "p90": 128.5936462402344, "max": 177.65505981445312, "pos_frac": 0.6875, "sample": [104.8582763671875, 122.28948211669922, 2.6497344970703125, 2.4833335876464844, 124.36875915527344, 51.00709533691406, -16.561279296875, -18.570709228515625, -77.0037841796875, 8.539009094238281, 143.5877685546875, -36.98590087890625, 25.930740356445312, 50.93280792236328, 89.26132202148438, 106.58000183105469, 23.72589111328125, 143.61448669433594, 29.456134796142578, 141.90087890625, 7.165996551513672, 97.15589904785156, 106.6985092163086, 1.8736495971679688, 21.558143615722656, 112.33346557617188, -3.220489501953125, 112.04698181152344, -22.939775466918945, -31.898277282714844, 65.39823150634766, -1.7978744506835938, 4.606203079223633, 177.65505981445312, 38.158042907714844, -52.37127685546875, 45.548587799072266, -40.59360885620117, -10.4117431640625, 51.64459228515625, 131.5011444091797, 130.40431213378906, 30.625228881835938, -0.5807075500488281, 45.54948425292969, -11.834869384765625, -22.537336349487305, -26.09954071044922, -0.48249053955078125, 56.889225006103516, -15.158733367919922, 139.84146118164062, -19.957448959350586, 60.93939971923828, 68.46638488769531, 2.3223495483398438, 54.5648193359375, 18.569656372070312, -21.037017822265625, 5.455602645874023, -13.200607299804688, 56.45539855957031, 117.10221099853516, 8.341869354248047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000348.npy"}
|
|
{"epoch": 0.5260770975056689, "step": 349, "batch_size": 64, "mean": 37.89960479736328, "std": 54.11586380004883, "min": -90.52372741699219, "p10": -14.933248901367186, "median": 26.06684112548828, "p90": 114.26652450561524, "max": 147.44271850585938, "pos_frac": 0.71875, "sample": [-40.59163284301758, 27.77996826171875, 113.31169128417969, 27.634498596191406, 34.65113067626953, 4.3708953857421875, 49.69683074951172, 113.3714599609375, 40.13838195800781, 20.645095825195312, -3.7991695404052734, 17.43409538269043, 102.22848510742188, 9.137985229492188, 24.499183654785156, 77.51646423339844, 65.10693359375, -39.41522216796875, -90.52372741699219, 114.6501235961914, 28.567840576171875, 97.51181030273438, -18.040847778320312, 6.027614593505859, 96.69090270996094, 10.387008666992188, 14.384117126464844, 50.9760856628418, -1.9539260864257812, 69.77951049804688, 94.95722961425781, 96.3572998046875, 2.2440643310546875, 21.625028610229492, -10.781524658203125, 121.21480560302734, 14.459478378295898, 141.9420928955078, -55.89759063720703, 45.59977340698242, 135.48593139648438, 1.2615432739257812, 5.48675537109375, 29.317413330078125, 45.0866813659668, 128.34228515625, 104.7813720703125, -0.7168235778808594, -12.619796752929688, -2.9239578247070312, 72.36677551269531, -11.598129272460938, -1.4617080688476562, -4.794151306152344, 30.890344619750977, 146.266845703125, -15.924728393554688, -0.5432586669921875, -27.193504333496094, 98.38339233398438, -1.2103137969970703, 1.2691879272460938, 64.28559112548828, 147.44271850585938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000349.npy"}
|
|
{"epoch": 0.527588813303099, "step": 350, "batch_size": 64, "mean": 47.50630187988281, "std": 53.91734313964844, "min": -67.98899841308594, "p10": -11.580286598205564, "median": 49.52082061767578, "p90": 123.94836578369143, "max": 154.84092712402344, "pos_frac": 0.78125, "sample": [117.95313262939453, -7.969451904296875, 64.49203491210938, -7.3473968505859375, 68.96215057373047, 13.201982498168945, 119.0518798828125, 59.93333435058594, 58.276432037353516, 109.10509490966797, 13.101119995117188, -32.919677734375, 37.8106689453125, 81.16728210449219, 39.37688446044922, 59.8824348449707, 103.80411529541016, 11.511669158935547, -12.419292449951172, -9.62260627746582, 49.56768798828125, 65.58340454101562, -62.24047088623047, 72.69140625, 49.59606170654297, -1.28704833984375, -67.98899841308594, 143.84144592285156, -8.171243667602539, 22.595882415771484, 103.04711151123047, -43.49189758300781, 98.42263793945312, 6.4563140869140625, 95.59774780273438, 127.6513671875, 45.32574462890625, 103.06497192382812, 126.93556213378906, 154.84092712402344, 36.912315368652344, 142.39349365234375, 126.04685974121094, -9.561637878417969, 42.86884689331055, 6.0529327392578125, 7.608238220214844, 8.559640884399414, -34.70486831665039, 53.68274688720703, 62.689178466796875, 52.902034759521484, 1.1782398223876953, 44.44886016845703, 25.171592712402344, -39.27107238769531, 56.96160888671875, 49.47395324707031, -2.2745285034179688, 43.096771240234375, 133.6014404296875, 76.00955200195312, 74.24088287353516, 112.9259033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000350.npy"}
|
|
{"epoch": 0.5291005291005291, "step": 351, "batch_size": 64, "mean": 38.6037712097168, "std": 60.222503662109375, "min": -127.76897430419922, "p10": -29.474383544921874, "median": 35.06870651245117, "p90": 119.18153533935548, "max": 130.04653930664062, "pos_frac": 0.765625, "sample": [-1.9961490631103516, 79.23821258544922, 123.29781341552734, 94.28368377685547, -1.3915596008300781, 92.78976440429688, -15.684833526611328, 33.443328857421875, 120.38166046142578, 18.940841674804688, 27.292327880859375, 116.70176696777344, 22.917081832885742, -5.616729736328125, 23.51207733154297, -19.27375030517578, 12.64141845703125, 9.344348907470703, 87.4766845703125, 67.94146728515625, 81.8393783569336, -63.66419219970703, 120.24429321289062, 13.898818969726562, 104.19833374023438, 123.12167358398438, 64.50897216796875, 87.90914154052734, -121.16222381591797, 37.89434814453125, 101.22634887695312, -52.87578582763672, 84.10910034179688, 1.36065673828125, 83.28194427490234, -45.60663604736328, -30.46269989013672, 128.6781005859375, 36.275779724121094, 23.22787857055664, 74.15193939208984, 130.04653930664062, 27.332855224609375, -27.168312072753906, 68.12362670898438, 23.9923095703125, 56.59210968017578, 59.719085693359375, 5.9608612060546875, 76.42598724365234, -24.53211212158203, -3.373554229736328, 81.98936462402344, 68.99385833740234, 121.35125732421875, 33.86163330078125, 22.449325561523438, 79.07148742675781, -127.76897430419922, 14.983861923217773, -109.13845825195312, 75.60660552978516, 73.1077880859375, 4.6197357177734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000351.npy"}
|
|
{"epoch": 0.5306122448979592, "step": 352, "batch_size": 64, "mean": 38.38576126098633, "std": 53.5926399230957, "min": -77.1496810913086, "p10": -14.670447158813475, "median": 27.29497241973877, "p90": 116.1514663696289, "max": 178.54721069335938, "pos_frac": 0.765625, "sample": [107.99598693847656, 151.720947265625, 100.4983901977539, 4.711273193359375, 110.06027221679688, -74.44412231445312, -14.973064422607422, 0.4838600158691406, 31.470439910888672, 45.520938873291016, 54.10894775390625, 87.51210021972656, -48.1810302734375, -1.8650074005126953, 57.9488525390625, 13.429397583007812, -6.08331298828125, -6.009307861328125, -39.39977264404297, 19.27448272705078, 76.15341186523438, 44.29755401611328, 22.837966918945312, 125.61698913574219, 117.24928283691406, 46.9305419921875, 30.760894775390625, -77.1496810913086, 32.48115921020508, -2.9411087036132812, 19.52486801147461, 1.4008846282958984, 5.5876007080078125, 61.56927490234375, 44.64506530761719, 71.20135498046875, 20.417394638061523, 116.400146484375, 125.6910400390625, 27.917869567871094, 115.57121276855469, -13.191841125488281, 178.54721069335938, 11.20829963684082, 65.43155670166016, 104.89491271972656, -18.847373962402344, 13.863204956054688, -13.119455337524414, 46.098270416259766, -8.584075927734375, 43.098358154296875, 20.3193359375, 96.8268051147461, 88.7070541381836, 5.35479736328125, -17.645729064941406, 119.7589111328125, -13.964340209960938, 7.4076080322265625, 0.5400276184082031, 67.73397064208984, 25.635229110717773, 26.672075271606445], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000352.npy"}
|
|
{"epoch": 0.5321239606953893, "step": 353, "batch_size": 64, "mean": 40.618675231933594, "std": 56.27607345581055, "min": -129.7491455078125, "p10": -11.179858779907224, "median": 28.7528715133667, "p90": 118.73152618408203, "max": 141.91949462890625, "pos_frac": 0.765625, "sample": [70.49102020263672, 127.80612182617188, -49.4617805480957, 35.43893051147461, -64.71784973144531, 23.665325164794922, 64.66382598876953, -12.047969818115234, 118.28581237792969, 23.749649047851562, 1.8605613708496094, 42.516761779785156, 72.865966796875, -12.850433349609375, 123.78897094726562, 11.289323806762695, 37.469573974609375, 92.15386199951172, 101.29629516601562, 118.92254638671875, 88.00798034667969, 136.76669311523438, 2.0566463470458984, 66.33586883544922, -6.58990478515625, -3.455188751220703, 127.00060272216797, 26.33177947998047, 117.44709014892578, 54.7451171875, 61.188568115234375, 2.4549789428710938, 58.7640495300293, -0.8261451721191406, -129.7491455078125, 107.27983856201172, -46.66486358642578, -35.625850677490234, 107.86138153076172, 110.7686996459961, 63.80221176147461, 124.04688262939453, 141.91949462890625, 10.252265930175781, 4.1663970947265625, 23.508541107177734, 82.12548828125, 15.494836807250977, -5.448947906494141, 9.732933044433594, 31.17396354675293, 14.379096984863281, -9.154266357421875, 35.09503173828125, -0.3563518524169922, 16.567916870117188, 9.517616271972656, 7.03936767578125, -4.432807922363281, 5.5882568359375, 93.97906494140625, -5.3176727294921875, 117.16587829589844, 47.46527099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000353.npy"}
|
|
{"epoch": 0.5336356764928194, "step": 354, "batch_size": 64, "mean": 43.53630828857422, "std": 58.10167694091797, "min": -78.23066711425781, "p10": -41.511751556396476, "median": 40.228097915649414, "p90": 121.16160583496094, "max": 150.3041229248047, "pos_frac": 0.796875, "sample": [141.326171875, 40.39303970336914, -30.988525390625, 87.82275390625, -46.021705627441406, 36.07795715332031, 20.06983184814453, 78.62415313720703, 103.87109375, 6.485340118408203, 89.46500396728516, 126.68234252929688, 70.18165588378906, 36.287994384765625, 10.438003540039062, 1.6143341064453125, 100.13121795654297, -26.808361053466797, 106.12606811523438, 150.3041229248047, 64.6956787109375, 110.81463623046875, 40.06315612792969, 47.1878662109375, -63.32366180419922, 3.389944076538086, -78.23066711425781, 101.50361633300781, 31.241966247558594, 44.720298767089844, 40.52996826171875, 90.71003723144531, 119.73492431640625, 44.90252685546875, 37.30018615722656, 27.91460418701172, 62.08042526245117, -46.163665771484375, 128.94020080566406, 8.35595703125, 71.5466537475586, 114.94764709472656, -57.05951690673828, -57.13481140136719, 123.96504211425781, -56.62562561035156, 126.83755493164062, 100.17652893066406, 11.035917282104492, -8.211738586425781, 23.219505310058594, 36.93572998046875, 93.40948486328125, 121.77304077148438, 45.63264083862305, 24.205120086669922, 4.108116149902344, 2.169384002685547, -11.775924682617188, 75.11042785644531, 23.91408348083496, 105.10701751708984, -30.796401977539062, -14.616497039794922], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000354.npy"}
|
|
{"epoch": 0.5351473922902494, "step": 355, "batch_size": 64, "mean": 43.829795837402344, "std": 60.69679260253906, "min": -98.85359954833984, "p10": -35.76871128082274, "median": 50.05369758605957, "p90": 121.76437683105469, "max": 146.71707153320312, "pos_frac": 0.75, "sample": [62.76652908325195, 133.54116821289062, -24.634292602539062, 11.4730224609375, 107.62091064453125, 51.43592834472656, 32.186241149902344, -26.27643394470215, -47.286766052246094, 49.93480682373047, 67.1319351196289, 1.761444091796875, 70.52531433105469, -47.510223388671875, 95.84859466552734, 92.44865417480469, -4.404300689697266, 52.84822463989258, -60.67253494262695, 74.43775177001953, -46.38081359863281, 54.21645736694336, 28.45098876953125, 113.51719665527344, 115.63593292236328, 32.709197998046875, 65.26993560791016, 23.22248077392578, 104.19222259521484, 146.71707153320312, -39.836830139160156, 30.08087921142578, -3.5834312438964844, 65.367431640625, 9.944610595703125, -98.85359954833984, 69.87307739257812, -6.8254852294921875, 50.17258834838867, 118.12767028808594, 4.352409362792969, 17.61199951171875, 108.60271453857422, -7.6219482421875, 3.669870376586914, 128.27276611328125, 102.81716918945312, 7.723850250244141, 129.0894775390625, -95.83135223388672, 122.05174255371094, 107.29383850097656, 121.09385681152344, -4.879039764404297, 129.958740234375, -19.30999183654785, -5.466545104980469, 10.481571197509766, 127.50203704833984, 0.6747283935546875, 95.75048065185547, 18.346200942993164, 74.97347259521484, 102.75537872314453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000355.npy"}
|
|
{"epoch": 0.5366591080876795, "step": 356, "batch_size": 64, "mean": 35.68011474609375, "std": 67.85032653808594, "min": -130.9059295654297, "p10": -46.883685302734364, "median": 27.44455337524414, "p90": 124.69528732299806, "max": 136.73028564453125, "pos_frac": 0.734375, "sample": [128.9431610107422, 125.57388305664062, 26.619537353515625, -77.61219787597656, 100.46324920654297, 8.076988220214844, 3.4642677307128906, 130.1818389892578, 63.82215881347656, 1.7312755584716797, -92.690185546875, -121.66260528564453, 55.602752685546875, -71.64234161376953, 118.23844146728516, 136.73028564453125, 8.274642944335938, 26.467403411865234, 110.98469543457031, 0.11431312561035156, 118.03125, -130.9059295654297, 119.27841186523438, 20.53741455078125, 12.951248168945312, 122.20285034179688, -37.57087326049805, -30.175537109375, 36.56028366088867, 28.78221893310547, 50.144691467285156, 6.97247314453125, -13.01542854309082, 2.4622631072998047, -50.704139709472656, 95.05813598632812, 97.87261199951172, -33.23204040527344, 60.4166259765625, 98.85926055908203, 12.192554473876953, 3.36053466796875, 122.79032897949219, -10.258987426757812, 43.97686767578125, 61.54022979736328, 85.65705871582031, 8.771263122558594, -37.96929168701172, 54.94313049316406, -1.4944190979003906, 71.9080810546875, 125.03472137451172, 133.00021362304688, -5.401786804199219, 82.40633392333984, 25.08041000366211, -18.611106872558594, 123.90327453613281, -12.951034545898438, -77.66004943847656, 129.59539794921875, 28.269569396972656, 79.2366943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000356.npy"}
|
|
{"epoch": 0.5381708238851096, "step": 357, "batch_size": 64, "mean": 31.173824310302734, "std": 56.25328063964844, "min": -112.81633758544922, "p10": -38.169156265258785, "median": 25.753963470458984, "p90": 107.93027648925782, "max": 132.39468383789062, "pos_frac": 0.75, "sample": [106.05447387695312, -1.1917858123779297, 102.3613510131836, -23.940284729003906, -106.8015365600586, 128.76614379882812, 27.133087158203125, 54.42117691040039, 20.06920623779297, 18.5362548828125, -12.787399291992188, 48.30510711669922, -36.426902770996094, 18.815555572509766, 132.39468383789062, 62.08056640625, 110.76077270507812, 53.63874816894531, 63.12318420410156, 6.162334442138672, 60.53606414794922, 43.066314697265625, 37.50742721557617, 7.924654006958008, 20.090126037597656, 44.287681579589844, -2.2322826385498047, 5.061225891113281, -42.521453857421875, 44.28889846801758, -54.2271728515625, 4.034215927124023, -24.483959197998047, 68.44184112548828, 67.6181869506836, 123.18717193603516, 103.60151672363281, 38.62016296386719, 78.9234619140625, 123.9455795288086, -14.542625427246094, 49.36625671386719, 10.057037353515625, 0.9646835327148438, 108.73419189453125, 36.695281982421875, 2.436370849609375, -12.908287048339844, -38.915836334228516, 24.374839782714844, 29.478656768798828, 122.59294128417969, 70.9351577758789, 19.042354583740234, 92.08234405517578, 12.766510009765625, 6.213228225708008, 4.530975341796875, 101.27906036376953, -76.81660461425781, -58.43244934082031, -112.81633758544922, 99.02804565429688, -0.16532516479492188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000357.npy"}
|
|
{"epoch": 0.5396825396825397, "step": 358, "batch_size": 64, "mean": 43.42327117919922, "std": 58.716373443603516, "min": -80.65330505371094, "p10": -26.65189056396484, "median": 29.250125885009766, "p90": 129.2271255493164, "max": 172.0169677734375, "pos_frac": 0.78125, "sample": [83.72406768798828, 59.75425720214844, 74.48837280273438, 132.2281494140625, -21.28517723083496, 125.16639709472656, 125.03101348876953, -22.61009979248047, -4.486133575439453, 16.6749267578125, -9.671566009521484, 75.36373901367188, 29.46398162841797, 76.9979248046875, 119.19214630126953, 23.578201293945312, -44.413414001464844, 22.000228881835938, 86.88650512695312, 24.720796585083008, 31.403106689453125, 20.682170867919922, -28.38408660888672, 99.58319854736328, 102.13384246826172, 3.3478469848632812, 8.596084594726562, 8.523025512695312, -16.79705810546875, 8.55527114868164, -56.39008331298828, 23.39398956298828, -42.40374755859375, 7.079618453979492, -17.05640411376953, 47.25114440917969, 64.61598205566406, 45.240936279296875, 14.445564270019531, -0.7425479888916016, 134.53036499023438, 135.79937744140625, 172.0169677734375, 4.920928955078125, 148.9112548828125, 130.96743774414062, 84.89730072021484, -36.176822662353516, 29.036270141601562, 50.75341796875, 1.8249664306640625, -49.65690612792969, -80.65330505371094, 116.47539520263672, 0.9419097900390625, 108.21383666992188, 27.363475799560547, 59.53301239013672, 13.5323486328125, 72.09437561035156, 67.1460189819336, 97.01923370361328, 44.119232177734375, 149.59727478027344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000358.npy"}
|
|
{"epoch": 0.5411942554799698, "step": 359, "batch_size": 64, "mean": 22.790054321289062, "std": 52.63726043701172, "min": -79.03219604492188, "p10": -41.44717025756836, "median": 18.910823822021484, "p90": 100.4476585388184, "max": 150.01991271972656, "pos_frac": 0.671875, "sample": [-31.053115844726562, 55.20368957519531, -18.947174072265625, 10.25860595703125, 43.72032928466797, 43.469032287597656, -29.948421478271484, 17.771045684814453, -43.5322265625, -25.984405517578125, 150.01991271972656, 48.4208984375, -1.0667724609375, 34.98236083984375, -5.141078948974609, 51.60002136230469, 111.74288940429688, -29.37104034423828, 4.7081756591796875, -75.36561584472656, 8.749217987060547, 3.5320167541503906, 18.559814453125, -41.54658508300781, 7.8741302490234375, 114.75167846679688, 49.58100891113281, 26.743549346923828, 28.42975616455078, -68.87210083007812, 123.13494873046875, 129.85414123535156, -50.382568359375, 90.16873168945312, 24.135894775390625, 19.26183319091797, 5.628776550292969, 55.300933837890625, 23.735198974609375, 28.40581512451172, -41.21520233154297, 73.41896057128906, 104.85291290283203, 65.52723693847656, 9.057884216308594, 39.682159423828125, -32.188392639160156, -66.81964111328125, -34.518760681152344, 51.303436279296875, 24.467853546142578, -6.624061584472656, -79.03219604492188, -5.242471694946289, 75.2694320678711, 1.3093719482421875, 74.28028869628906, 15.95303726196289, 40.853729248046875, 84.93980407714844, 31.935726165771484, 129.68858337402344, -1.5491752624511719, -5.3203277587890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000359.npy"}
|
|
{"epoch": 0.5427059712773998, "step": 360, "batch_size": 64, "mean": 36.89454650878906, "std": 54.36711120605469, "min": -83.85198211669922, "p10": -21.135464477539056, "median": 22.837358474731445, "p90": 114.60849075317383, "max": 148.82809448242188, "pos_frac": 0.703125, "sample": [-2.867340087890625, 133.5172119140625, 4.442596435546875, 13.972091674804688, 31.76532745361328, 72.66271209716797, -6.454689025878906, 35.314762115478516, 21.815967559814453, 99.093994140625, -1.9797687530517578, -53.83879089355469, -0.23016357421875, -51.586978912353516, 99.83424377441406, -38.54985046386719, 132.94830322265625, -30.840198516845703, 2.5058746337890625, 113.10101318359375, 26.018070220947266, -1.351776123046875, -33.528411865234375, 23.858749389648438, 4.834251403808594, -0.9773178100585938, 100.02549743652344, 16.740135192871094, 77.76696014404297, 3.7827606201171875, 35.014808654785156, 17.047706604003906, 67.18299865722656, -24.40606689453125, 21.253753662109375, 96.99311065673828, -13.504058837890625, 52.235740661621094, 113.89812469482422, 2.6257858276367188, 148.82809448242188, -4.1333160400390625, 91.31974792480469, -12.616674423217773, 146.23263549804688, 35.094993591308594, 119.78295135498047, 11.93353271484375, 27.266490936279297, 114.91293334960938, 104.00113677978516, 50.26116943359375, 129.6574249267578, -0.7649631500244141, -5.235555648803711, 41.48857116699219, -2.4042186737060547, -83.85198211669922, 17.784547805786133, 82.52423095703125, 31.589691162109375, 4.9635772705078125, 62.64402770996094, 89.83476257324219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000360.npy"}
|
|
{"epoch": 0.54421768707483, "step": 361, "batch_size": 64, "mean": 44.295894622802734, "std": 53.943641662597656, "min": -76.3538818359375, "p10": -15.503910064697262, "median": 43.092952728271484, "p90": 118.98207321166993, "max": 149.26553344726562, "pos_frac": 0.78125, "sample": [52.210472106933594, 66.42707061767578, 6.972133636474609, -11.546810150146484, 126.75924682617188, 48.673866271972656, -33.447147369384766, -76.3538818359375, 33.61433029174805, 42.39934158325195, 66.37480163574219, -47.60195541381836, 0.991729736328125, 122.45662689208984, 93.45425415039062, 74.96024322509766, -17.199810028076172, 7.869197845458984, 49.592315673828125, 9.242584228515625, 55.27777099609375, 61.44890594482422, 16.335594177246094, 127.64459991455078, 99.95014190673828, 84.32225036621094, 83.15116882324219, -74.29379272460938, 119.76839447021484, 78.9180908203125, 43.786563873291016, -1.824432373046875, 94.81710815429688, 149.26553344726562, 102.16241455078125, 52.0145263671875, 115.552490234375, 67.1678466796875, -36.66972351074219, -6.523735046386719, -1.033334732055664, 9.304489135742188, 117.14732360839844, 107.1048583984375, 14.859355926513672, 129.9491424560547, 90.87939453125, 9.434181213378906, -44.124916076660156, -10.352645874023438, 75.71894836425781, 27.001663208007812, 75.4210205078125, 21.409481048583984, 137.78909301757812, 17.532155990600586, 94.63796997070312, 13.973564147949219, -4.0887908935546875, 40.00122833251953, 41.717262268066406, 30.059560775756836, 0.4294242858886719, -7.9537200927734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000361.npy"}
|
|
{"epoch": 0.54572940287226, "step": 362, "batch_size": 64, "mean": 45.63495635986328, "std": 64.44790649414062, "min": -109.30011749267578, "p10": -22.175243377685543, "median": 37.12725067138672, "p90": 122.54052505493163, "max": 255.31134033203125, "pos_frac": 0.78125, "sample": [-20.043685913085938, 5.200544357299805, 15.776290893554688, 1.4115371704101562, 3.0849380493164062, -109.30011749267578, 34.820960998535156, 62.478912353515625, 255.31134033203125, 92.82933044433594, 11.961860656738281, -12.613208770751953, 35.162620544433594, -3.2187423706054688, -2.939556121826172, -7.029270172119141, 61.753326416015625, 73.91763305664062, 66.90243530273438, 128.97421264648438, -8.484930038452148, 19.972686767578125, 20.822874069213867, 53.070465087890625, 119.59263610839844, 12.670501708984375, 102.46841430664062, 12.96771240234375, 1.9187335968017578, 79.5693359375, 166.44683837890625, 4.71617317199707, -19.75830841064453, 114.33503723144531, 22.878578186035156, 128.24786376953125, 93.23656463623047, 4.317352294921875, 49.04065704345703, -23.088768005371094, 2.3232154846191406, 111.61255645751953, 122.57913970947266, 122.45042419433594, 23.635391235351562, 42.02349853515625, 85.845458984375, 140.25877380371094, 93.36448669433594, 94.04248046875, -26.425111770629883, 21.27191162109375, 110.78206634521484, -61.02992248535156, 39.091880798339844, 56.550872802734375, 120.19155883789062, -56.6605224609375, -46.037208557128906, 69.07295227050781, 150.4512939453125, -67.19341278076172, 59.11735534667969, 63.93641662597656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000362.npy"}
|
|
{"epoch": 0.54724111866969, "step": 363, "batch_size": 64, "mean": 44.824920654296875, "std": 57.4576301574707, "min": -110.85472106933594, "p10": -12.244223403930665, "median": 38.89578819274902, "p90": 129.91102600097656, "max": 192.49710083007812, "pos_frac": 0.84375, "sample": [77.46773529052734, -72.46709442138672, 130.8831787109375, 55.169273376464844, 44.071319580078125, 39.01959991455078, 8.293128967285156, 132.62730407714844, 12.248626708984375, -29.085994720458984, -5.9331817626953125, 43.895904541015625, 192.49710083007812, 141.64007568359375, 7.743217468261719, 146.8463134765625, 38.62833023071289, 47.601402282714844, 21.369338989257812, 105.77040100097656, 13.859344482421875, 60.455936431884766, 13.312349319458008, 131.53224182128906, 127.14845275878906, 26.205965042114258, 2.930938720703125, 25.544692993164062, -12.237945556640625, 17.050254821777344, -18.5395565032959, -110.85472106933594, 47.739906311035156, -61.14572525024414, 76.13204956054688, 56.30659484863281, 30.09992790222168, 81.41021728515625, 43.254371643066406, -1.214426040649414, 8.77969741821289, 9.188957214355469, 41.05638122558594, 118.73094177246094, 14.791593551635742, 139.55816650390625, 88.60223388671875, 28.246383666992188, 118.46084594726562, 3.1083221435546875, 112.8698501586914, 4.1072235107421875, 91.34894561767578, 62.481231689453125, -33.25749206542969, 25.60736083984375, 127.64266967773438, -12.24691390991211, 69.14927673339844, 38.771976470947266, 0.0021190643310546875, 47.632110595703125, 59.398406982421875, 17.48781394958496], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000363.npy"}
|
|
{"epoch": 0.5487528344671202, "step": 364, "batch_size": 64, "mean": 47.91945266723633, "std": 68.43669128417969, "min": -123.2496337890625, "p10": -22.547470092773438, "median": 48.87984848022461, "p90": 139.77841796875, "max": 189.2482147216797, "pos_frac": 0.765625, "sample": [146.5216064453125, -2.8839111328125, 17.74169158935547, 141.80227661132812, 140.89642333984375, 12.030662536621094, 58.464942932128906, 89.8154067993164, -22.94835662841797, 144.05514526367188, 163.4048309326172, 74.643310546875, -11.816936492919922, -50.51111602783203, 84.31382751464844, 130.36378479003906, 43.416595458984375, -16.376651763916016, 189.2482147216797, -3.4470672607421875, 137.16973876953125, 33.67058563232422, 76.99906921386719, 119.82167053222656, -86.40434265136719, -45.799095153808594, 134.9652099609375, 0.49391746520996094, 1.2426166534423828, 82.52392578125, 118.2747802734375, 54.343101501464844, -4.424598693847656, 55.959617614746094, 22.233016967773438, 7.1038665771484375, 96.82131958007812, 127.13446044921875, -18.401512145996094, 14.764139175415039, 80.62537384033203, 67.6622085571289, -21.61206817626953, -52.88676452636719, 151.23077392578125, 9.027896881103516, 5.9536895751953125, 103.94638061523438, 6.1227874755859375, 99.35211181640625, 3.6973724365234375, 90.40172576904297, -79.31243896484375, 60.30225372314453, -123.2496337890625, 11.320175170898438, 73.84131622314453, 91.7359619140625, 0.4912548065185547, 12.633392333984375, 2.638538360595703, -18.14069366455078, 128.46397399902344, 105.37333679199219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000364.npy"}
|
|
{"epoch": 0.5502645502645502, "step": 365, "batch_size": 64, "mean": 44.768638610839844, "std": 51.508304595947266, "min": -68.86378479003906, "p10": -8.679346084594721, "median": 39.373199462890625, "p90": 110.49596405029298, "max": 184.03912353515625, "pos_frac": 0.84375, "sample": [23.341550827026367, 12.783990859985352, 75.65969848632812, 1.9999656677246094, 34.54412841796875, 2.337606430053711, 106.8251724243164, 97.09976196289062, 101.60990905761719, 71.81205749511719, 88.17097473144531, 43.67701721191406, -68.86378479003906, 4.274677276611328, 18.178756713867188, 95.74922180175781, -2.7820892333984375, 6.3442535400390625, -14.628227233886719, 0.6162509918212891, 51.12791442871094, 75.77863311767578, 87.61479949951172, 4.912683486938477, -0.7689990997314453, -61.03367614746094, 45.57439422607422, 138.60580444335938, 26.327247619628906, 73.37393188476562, 12.946914672851562, 15.68073844909668, 104.36392211914062, 25.857501983642578, 102.33903503417969, 54.17941665649414, 121.04882049560547, -4.085475921630859, 37.90715408325195, 61.168434143066406, 127.714111328125, 67.41389465332031, 48.55458068847656, -12.8228759765625, 62.65974426269531, 120.57437896728516, 4.76112174987793, -19.2373104095459, 184.03912353515625, 38.076141357421875, 75.8207778930664, 0.27860450744628906, 99.86853790283203, 9.66644287109375, 0.914794921875, 40.670257568359375, 1.1498794555664062, -36.97709274291992, 57.63494873046875, 4.0893707275390625, 136.2458038330078, 112.06916046142578, -10.648147583007812, 81.00642395019531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000365.npy"}
|
|
{"epoch": 0.5517762660619804, "step": 366, "batch_size": 64, "mean": 39.542518615722656, "std": 63.52601623535156, "min": -97.29533386230469, "p10": -37.81134643554687, "median": 39.61838150024414, "p90": 122.24505844116212, "max": 158.30198669433594, "pos_frac": 0.6875, "sample": [-1.22589111328125, 117.71456909179688, 93.69310760498047, 27.06772232055664, -6.586933135986328, 113.08132934570312, -40.852325439453125, -21.085927963256836, -18.86517333984375, -16.35400390625, -30.715728759765625, 82.76927947998047, 19.727977752685547, 63.66579055786133, 61.57643127441406, 33.95672607421875, -49.89459228515625, 53.18658447265625, 3.123411178588867, 79.0475845336914, 57.74029541015625, 50.566932678222656, 42.38688659667969, -9.75406265258789, 20.946195602416992, 87.96708679199219, 44.50748062133789, 122.7626953125, -97.29533386230469, -12.015504837036133, 150.00144958496094, -0.16602325439453125, -4.453500747680664, 158.30198669433594, -53.90345001220703, 2.049001693725586, -24.888893127441406, -73.38121032714844, 21.633121490478516, 37.95417022705078, 6.66851806640625, -13.752567291259766, 8.011655807495117, 43.740447998046875, -49.75688552856445, 4.699974060058594, 10.153226852416992, 152.1278533935547, 141.95602416992188, 41.2825927734375, 49.85906982421875, 115.23822021484375, 94.19508361816406, -92.18388366699219, 116.37409973144531, -20.07704734802246, 126.54415130615234, 97.26466369628906, 140.96900939941406, 96.56001281738281, 117.74401092529297, 72.04954528808594, 121.03723907470703, 66.02680969238281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000366.npy"}
|
|
{"epoch": 0.5532879818594104, "step": 367, "batch_size": 64, "mean": 35.85786437988281, "std": 55.85295486450195, "min": -124.57630920410156, "p10": -28.121626472473146, "median": 31.20323944091797, "p90": 112.08380050659181, "max": 133.50390625, "pos_frac": 0.8125, "sample": [3.760976791381836, 133.50390625, 47.632843017578125, -5.592567443847656, 50.73577117919922, 0.325225830078125, 109.26089477539062, 118.8552017211914, 70.18338012695312, 3.4580039978027344, 18.310134887695312, 12.440109252929688, 7.364356994628906, 16.442184448242188, 121.78606414794922, 13.58578872680664, 63.184326171875, -2.8519363403320312, -27.923986434936523, 93.3602523803711, 20.265045166015625, 46.74983215332031, -7.479154586791992, -32.676597595214844, 56.124267578125, -124.57630920410156, -36.865577697753906, 86.06087493896484, 15.528709411621094, 37.725830078125, 89.37881469726562, 83.50656127929688, 52.87135314941406, 52.3258056640625, 52.19807815551758, 5.7829742431640625, 36.02154541015625, -52.102294921875, 9.866090774536133, 126.76039123535156, 16.521438598632812, -122.9476089477539, 13.988325119018555, 128.1918487548828, 76.00981903076172, -28.206329345703125, 65.02264404296875, 13.416316986083984, 77.24087524414062, 17.679271697998047, 26.384933471679688, 17.681854248046875, -80.8266830444336, 104.46057891845703, 67.35433959960938, 106.89022064208984, 25.21416473388672, -15.816076278686523, 72.54667663574219, 123.95896911621094, 20.2689208984375, 49.44872283935547, 41.83937072753906, 113.29361724853516], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000367.npy"}
|
|
{"epoch": 0.5547996976568406, "step": 368, "batch_size": 64, "mean": 34.7327995300293, "std": 57.268455505371094, "min": -123.31732177734375, "p10": -33.96074066162109, "median": 35.42912292480469, "p90": 105.07051162719729, "max": 140.77182006835938, "pos_frac": 0.71875, "sample": [-123.31732177734375, 74.56303405761719, 0.8785629272460938, 8.563810348510742, 66.806640625, 86.37787628173828, 97.81167602539062, 37.73262023925781, -12.573341369628906, -16.056427001953125, 122.26847839355469, 9.161190032958984, 33.12562561035156, 29.49829864501953, 128.44598388671875, 98.13783264160156, -22.21015167236328, -35.3763427734375, -51.918601989746094, 98.91564178466797, 29.959991455078125, 25.115564346313477, 55.75385284423828, -27.524459838867188, 122.88965606689453, 19.15488052368164, -23.88970947265625, 107.70831298828125, 140.77182006835938, 49.146175384521484, -13.13984489440918, 114.80484008789062, 66.73015594482422, 1.2001457214355469, 14.164600372314453, 85.86135864257812, 20.04137420654297, 80.44590759277344, 8.896354675292969, 74.21965789794922, 42.91166687011719, -2.379270553588867, 93.64222717285156, 27.118927001953125, -40.20613098144531, -28.09210968017578, 29.40890121459961, -13.161575317382812, 111.7033462524414, -50.93055725097656, 58.94709777832031, 98.35079956054688, 81.2574462890625, 98.5727767944336, -1.136587142944336, 52.37379455566406, 41.134620666503906, -30.657669067382812, -74.28546905517578, 67.59440612792969, 83.87313842773438, 38.58018493652344, -86.24882507324219, 41.382171630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000368.npy"}
|
|
{"epoch": 0.5563114134542706, "step": 369, "batch_size": 64, "mean": 32.24602508544922, "std": 56.65467834472656, "min": -101.71308898925781, "p10": -26.51274585723877, "median": 23.857738494873047, "p90": 112.2237632751465, "max": 173.2786865234375, "pos_frac": 0.734375, "sample": [21.056800842285156, -60.060142517089844, 39.40594482421875, 57.0548095703125, -60.25421142578125, 71.15045166015625, 43.69981002807617, 3.2283706665039062, 102.36032104492188, 21.024822235107422, 13.454437255859375, 91.44346618652344, 99.92733001708984, 87.0243148803711, 119.38407135009766, 5.193607330322266, 51.28962707519531, 34.61317443847656, 25.63933563232422, 32.15183639526367, 19.2171630859375, -101.71308898925781, 39.27738952636719, -67.27197265625, 28.457717895507812, 69.91394805908203, -17.02255630493164, 62.64887237548828, 39.70199203491211, -26.98138999938965, 105.13198852539062, -9.828781127929688, 32.504425048828125, 119.64789581298828, 65.51827239990234, 114.66845703125, 140.239013671875, -4.0269775390625, 2.3308258056640625, -58.62653350830078, -4.49407958984375, 22.076141357421875, -11.855060577392578, 64.60872650146484, -4.379558563232422, 9.370109558105469, 173.2786865234375, 6.438331604003906, 102.75439453125, -11.277931213378906, -7.7351837158203125, 19.870864868164062, 0.2975654602050781, 113.98611450195312, 3.047504425048828, -14.054162979125977, 2.3873653411865234, 140.00283813476562, 17.107521057128906, 34.91015625, -25.41924285888672, 42.700164794921875, -70.56224060058594, 108.11161041259766], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000369.npy"}
|
|
{"epoch": 0.5578231292517006, "step": 370, "batch_size": 64, "mean": 44.62942886352539, "std": 66.01347351074219, "min": -138.7526397705078, "p10": -20.32953643798828, "median": 37.24666976928711, "p90": 135.10694427490236, "max": 154.42221069335938, "pos_frac": 0.765625, "sample": [48.179962158203125, 134.21826171875, 84.40131378173828, 31.199424743652344, 154.42221069335938, 4.306098937988281, 147.95445251464844, 3.1093597412109375, 139.25982666015625, 12.241039276123047, 8.853485107421875, 28.375350952148438, 0.2973365783691406, 104.2342529296875, 154.17173767089844, 72.37234497070312, 43.16654968261719, -54.88860321044922, -18.599777221679688, 61.3758544921875, 51.783416748046875, 64.33744812011719, 51.308963775634766, 62.52464294433594, 125.47544860839844, 79.66168212890625, 75.83944702148438, 11.379650115966797, 2.4007110595703125, -6.7682647705078125, -0.0958099365234375, -1.36590576171875, -7.582252502441406, -74.34620666503906, -6.6123046875, 73.07654571533203, 141.30905151367188, 36.98809814453125, 132.17433166503906, 103.37699890136719, 145.05567932128906, -138.7526397705078, 37.50524139404297, 74.41815185546875, 14.396638870239258, 135.48780822753906, 73.49739074707031, 93.09288787841797, 115.4102554321289, 124.943603515625, -133.42367553710938, 14.853073120117188, 117.43756103515625, -4.12761116027832, -34.450927734375, 21.00274658203125, 19.003524780273438, -3.571685791015625, -49.884544372558594, 1.4234256744384766, 35.56476593017578, 115.47966003417969, -21.07086181640625, 29.476816177368164], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000370.npy"}
|
|
{"epoch": 0.5593348450491308, "step": 371, "batch_size": 64, "mean": 42.01428985595703, "std": 58.205352783203125, "min": -59.500579833984375, "p10": -36.27098159790039, "median": 39.26255798339844, "p90": 124.49847183227541, "max": 162.39529418945312, "pos_frac": 0.734375, "sample": [111.85377502441406, -59.500579833984375, 145.490966796875, 62.28600311279297, 89.84898376464844, 0.670989990234375, 87.89668273925781, 13.612390518188477, -6.826873779296875, 11.298685073852539, -7.1608123779296875, 81.46212768554688, -41.93635940551758, 143.5506134033203, 3.9736404418945312, 151.52798461914062, -2.0113391876220703, 89.04645538330078, 16.367389678955078, 133.80563354492188, 31.384445190429688, -49.15156555175781, -10.371980667114258, 12.004009246826172, 45.00787353515625, 101.80931091308594, 86.12997436523438, 162.39529418945312, 97.20899963378906, 62.08837127685547, 90.4371337890625, 126.67195892333984, -52.94629669189453, 4.3283233642578125, 57.07244873046875, 31.185836791992188, -7.610267639160156, -40.560089111328125, -15.109542846679688, 140.73789978027344, -40.12776184082031, 21.571842193603516, 0.9110431671142578, 48.11005783081055, 56.88055419921875, -23.339462280273438, 67.82301330566406, 65.11521911621094, -34.062744140625, 33.517242431640625, 74.27442932128906, -16.947399139404297, 4.562740325927734, 72.12568664550781, 98.61349487304688, 0.43918800354003906, 65.60401153564453, 23.231163024902344, 80.73269653320312, 56.079933166503906, 119.427001953125, 84.30998992919922, -37.217369079589844, -30.688777923583984], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000371.npy"}
|
|
{"epoch": 0.5608465608465608, "step": 372, "batch_size": 64, "mean": 25.308969497680664, "std": 68.0642318725586, "min": -122.12499237060547, "p10": -53.25455665588379, "median": 17.92519187927246, "p90": 127.72814025878911, "max": 153.380859375, "pos_frac": 0.71875, "sample": [-35.42622375488281, 18.012943267822266, 107.7941665649414, 80.10435485839844, 61.15472412109375, 38.641258239746094, 140.8086395263672, -30.521278381347656, 106.88184356689453, 102.0586929321289, 146.37640380859375, 0.26739501953125, 81.28982543945312, 153.380859375, 40.061767578125, 65.81370544433594, 88.25027465820312, -109.23808288574219, 9.706642150878906, 22.833168029785156, 9.76247787475586, 0.3596057891845703, 22.441421508789062, 115.21365356445312, 7.030963897705078, -6.815132141113281, -86.12156677246094, 10.198112487792969, 140.20689392089844, -47.32789611816406, 62.953086853027344, 32.26690673828125, -9.367015838623047, -8.18404769897461, -54.44498825073242, 20.2041072845459, -122.12499237060547, 12.432548522949219, 97.07482147216797, 29.355398178100586, 136.79428100585938, 47.48231887817383, 83.11166381835938, 48.419097900390625, -50.47688293457031, 37.254425048828125, -107.65066528320312, 1.4611492156982422, 8.084527969360352, 17.837440490722656, 3.6840286254882812, -119.59449768066406, 69.37091064453125, 3.6690406799316406, -26.255525588989258, -35.68878173828125, -86.66485595703125, 44.36217498779297, 134.50662231445312, 133.09149169921875, -40.05014419555664, -21.860595703125, 10.517585754394531, 15.003826141357422], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000372.npy"}
|
|
{"epoch": 0.562358276643991, "step": 373, "batch_size": 64, "mean": 32.62363052368164, "std": 68.84725189208984, "min": -114.26799011230469, "p10": -71.09233779907225, "median": 28.45964527130127, "p90": 120.37649536132812, "max": 162.58070373535156, "pos_frac": 0.703125, "sample": [-90.58080291748047, 102.54178619384766, 120.61448669433594, 82.5765380859375, 112.12335205078125, -74.4814224243164, -20.725364685058594, 107.06149291992188, 71.1038589477539, 23.31327247619629, 82.24848937988281, -89.23365020751953, 35.3896484375, -40.30497741699219, -63.79808807373047, -32.34135818481445, 30.53835105895996, 65.78787994384766, -114.26799011230469, -26.376251220703125, 58.54520034790039, -14.363853454589844, 66.9142837524414, -56.40142059326172, 113.57797241210938, 119.82118225097656, -7.5893707275390625, 6.479156494140625, 77.86347198486328, 83.97753143310547, 136.37208557128906, 0.3610992431640625, 26.380939483642578, 117.70748138427734, -58.39240264892578, 0.2512626647949219, -4.224090576171875, 70.10061645507812, 0.38404083251953125, 115.17340087890625, 87.63345336914062, -78.80548095703125, -74.21844482421875, -15.003223419189453, 15.016159057617188, 78.41023254394531, -75.75784301757812, 126.7232666015625, 15.077095031738281, 39.01789855957031, -16.304656982421875, 133.87010192871094, 3.9717864990234375, 162.58070373535156, 25.24534034729004, 37.578365325927734, 49.70376205444336, 59.856842041015625, 25.549957275390625, 162.2548065185547, 38.43316650390625, 7.704967498779297, 17.058975219726562, 128.18739318847656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000373.npy"}
|
|
{"epoch": 0.563869992441421, "step": 374, "batch_size": 64, "mean": 54.3033447265625, "std": 58.349945068359375, "min": -94.87937927246094, "p10": -2.9443962097167966, "median": 49.54737663269043, "p90": 131.81131362915042, "max": 185.80520629882812, "pos_frac": 0.859375, "sample": [185.80520629882812, 106.18563842773438, 88.63919067382812, -2.09259033203125, 39.7789192199707, 104.14286804199219, 137.50033569335938, 122.40216827392578, 137.7947998046875, 50.516273498535156, 74.12694549560547, 14.39453125, 89.81318664550781, 150.61044311523438, 67.5669937133789, 11.866714477539062, 122.4818344116211, 5.6938629150390625, 10.699569702148438, 50.356361389160156, 23.24102020263672, 95.49443054199219, 27.54257583618164, -67.3795166015625, 82.22024536132812, 68.19329071044922, 38.69095230102539, 111.34607696533203, -2.7771987915039062, 74.69520568847656, 49.136322021484375, 64.33676147460938, 2.225330352783203, 2.5683975219726562, 53.072296142578125, 49.958431243896484, 33.10572814941406, -94.87937927246094, 135.80966186523438, 70.66535949707031, 34.71834945678711, -68.89460754394531, 15.980649948120117, 158.53089904785156, 153.12557983398438, 26.918594360351562, -47.5159912109375, 16.879837036132812, 45.72889709472656, 40.303688049316406, 119.60295104980469, 103.37887573242188, 113.83653259277344, 36.26373291015625, 30.88022232055664, 2.564657211303711, 39.3656005859375, 72.46670532226562, 93.6312484741211, 119.69161987304688, -42.9884033203125, -3.01605224609375, 34.88616943359375, -6.4750823974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000374.npy"}
|
|
{"epoch": 0.5653817082388511, "step": 375, "batch_size": 64, "mean": 36.52012252807617, "std": 61.960628509521484, "min": -127.55668640136719, "p10": -50.44816665649414, "median": 33.571041107177734, "p90": 119.71471786499023, "max": 162.43850708007812, "pos_frac": 0.703125, "sample": [-52.172279357910156, -53.033668518066406, 37.197776794433594, 61.67305374145508, 64.82369232177734, 16.41817283630371, 141.82254028320312, -1.2760047912597656, 7.5792999267578125, 66.881591796875, -14.216682434082031, 96.23118591308594, -5.788818359375, 129.14810180664062, 6.914646148681641, -5.228931427001953, 35.27758026123047, -56.248382568359375, 73.51760864257812, -5.246726989746094, 35.136314392089844, 120.31822204589844, 129.86203002929688, -51.472625732421875, 80.06763458251953, 118.3065414428711, 1.5837783813476562, -36.847198486328125, 34.64928436279297, 101.36042785644531, 4.443378448486328, 0.8024272918701172, 9.779104232788086, -48.057762145996094, 66.32763671875, 52.84588623046875, 66.89032745361328, 22.770095825195312, 141.58641052246094, 47.03730010986328, 100.31112670898438, 32.4927978515625, 108.35865783691406, -127.55668640136719, -3.2496185302734375, 14.535247802734375, 130.44595336914062, 9.350830078125, 162.43850708007812, 62.36113357543945, 111.4088134765625, 63.71958923339844, -14.370269775390625, 103.94671630859375, -15.365612030029297, 19.30925750732422, 66.72762298583984, -65.51469421386719, -21.26494598388672, 117.46027374267578, -7.429710388183594, 91.88715362548828, 13.127937316894531, -57.50513458251953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000375.npy"}
|
|
{"epoch": 0.5668934240362812, "step": 376, "batch_size": 64, "mean": 47.989322662353516, "std": 68.40380859375, "min": -100.425537109375, "p10": -39.17854881286621, "median": 38.98215866088867, "p90": 137.55483551025392, "max": 212.70919799804688, "pos_frac": 0.734375, "sample": [37.953857421875, 3.6211624145507812, 1.4043254852294922, 30.1719970703125, 2.9544906616210938, -21.791072845458984, -0.6208114624023438, 113.63578796386719, -0.498443603515625, 38.347816467285156, 78.8865966796875, 81.65571594238281, -100.425537109375, -3.1666030883789062, -62.768798828125, 6.18438720703125, 149.06658935546875, 52.31477355957031, 11.118122100830078, 212.70919799804688, 39.61650085449219, 138.69602966308594, 7.252716064453125, 115.87062072753906, 164.4068603515625, -10.025772094726562, 117.54322052001953, 124.88513946533203, 111.66255950927734, 92.75615692138672, 96.97036743164062, 36.5789794921875, 131.0404052734375, 43.75627136230469, 47.22578430175781, 135.8120880126953, 116.83279418945312, -38.4190788269043, 139.6378631591797, -66.60517883300781, 43.382164001464844, -46.43907165527344, 110.45331573486328, 99.355712890625, 78.52108001708984, 138.30172729492188, -39.50403594970703, 85.43643188476562, -29.002975463867188, 114.21033477783203, -61.966758728027344, 28.299766540527344, 13.526710510253906, 20.380245208740234, 139.46568298339844, -4.247446060180664, 69.01515197753906, -5.957561492919922, 132.33978271484375, 68.82514953613281, -19.754547119140625, 22.08491325378418, 2.1846485137939453, -63.84161376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000376.npy"}
|
|
{"epoch": 0.5684051398337112, "step": 377, "batch_size": 64, "mean": 34.98808288574219, "std": 58.97484588623047, "min": -106.27184295654297, "p10": -42.25645446777343, "median": 33.696590423583984, "p90": 113.20317306518555, "max": 153.860595703125, "pos_frac": 0.6875, "sample": [80.97903442382812, -35.224143981933594, 89.79047393798828, 85.42811584472656, 57.68821716308594, -4.785591125488281, -68.52050018310547, -61.952178955078125, 50.60272979736328, 33.63463592529297, 135.21792602539062, 4.216094970703125, 97.3022232055664, -0.4468994140625, 102.95523071289062, 107.13846588134766, -62.470680236816406, 77.47733306884766, -50.43012237548828, -0.6015548706054688, 118.93940734863281, -2.852466583251953, 74.80313110351562, 23.12232208251953, 66.83726501464844, -97.09071350097656, 23.510730743408203, 153.860595703125, 16.694049835205078, -0.7646007537841797, 98.52422332763672, -10.267908096313477, 76.7789306640625, 1.6863040924072266, 7.9376373291015625, -45.270301818847656, 33.758544921875, 44.553977966308594, 117.666259765625, 31.037521362304688, -4.811332702636719, 85.15084075927734, 9.851556777954102, 56.77629089355469, 41.3673095703125, 137.328857421875, 33.799495697021484, -9.607135772705078, -15.266082763671875, 40.02021789550781, 111.04200744628906, 14.113525390625, -6.043985366821289, 135.66390991210938, 0.4044647216796875, -20.56279945373535, 114.12938690185547, 66.9937515258789, -106.27184295654297, 10.532218933105469, 55.799842834472656, -7.197242736816406, 62.20808410644531, 62.352169036865234], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000377.npy"}
|
|
{"epoch": 0.5699168556311414, "step": 378, "batch_size": 64, "mean": 32.11248016357422, "std": 58.19516372680664, "min": -117.74447631835938, "p10": -16.318192481994625, "median": 19.096742630004883, "p90": 114.81733551025391, "max": 159.27154541015625, "pos_frac": 0.734375, "sample": [84.42134857177734, 55.097862243652344, 3.606729507446289, 14.800052642822266, 73.28314971923828, 101.74126434326172, -33.20768737792969, 2.8706207275390625, 39.45338439941406, 114.86341857910156, -117.74447631835938, 6.970819473266602, 159.27154541015625, -0.9616813659667969, 108.08210754394531, 72.77727508544922, 31.367231369018555, 6.316133499145508, 132.065185546875, -3.603199005126953, 75.30513000488281, 23.431167602539062, 1.1810874938964844, 35.443992614746094, 123.07748413085938, 10.232566833496094, 35.90107727050781, 34.857086181640625, -64.08833312988281, 11.63485336303711, 68.05732727050781, 23.793716430664062, -54.26423645019531, -7.921369552612305, 74.41302490234375, 16.49551773071289, -17.486190795898438, 8.129257202148438, 96.95722198486328, -10.974397659301758, 148.40916442871094, 60.11708068847656, 4.933990478515625, 114.70980834960938, 21.697967529296875, 84.06959533691406, -6.629062652587891, 131.2736053466797, 46.668785095214844, 7.461713790893555, 5.332271575927734, -4.26708984375, 11.32358169555664, -88.29170227050781, -10.298114776611328, 88.45639038085938, -13.592863082885742, 47.297760009765625, 3.1270503997802734, 38.38849639892578, -6.130832672119141, -94.52650451660156, -4.066057205200195, 134.08648681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000378.npy"}
|
|
{"epoch": 0.5714285714285714, "step": 379, "batch_size": 64, "mean": 43.51008605957031, "std": 59.817230224609375, "min": -82.52468872070312, "p10": -19.22555999755859, "median": 33.326995849609375, "p90": 126.92794189453127, "max": 181.75018310546875, "pos_frac": 0.828125, "sample": [51.51081466674805, -1.1199798583984375, 116.24945831298828, 0.98565673828125, 21.760498046875, 49.92259216308594, 44.55205535888672, -82.52468872070312, 43.11317825317383, 11.613517761230469, 8.714820861816406, 18.6181640625, -51.5928955078125, 90.5301513671875, 8.671470642089844, 181.75018310546875, 31.942153930664062, -19.865360260009766, 96.94091033935547, 117.93122863769531, 176.29824829101562, 83.95932006835938, 129.8998260498047, 77.9485855102539, 33.382080078125, -57.44384765625, 56.88690185546875, -51.19743347167969, 43.486412048339844, -39.528472900390625, 174.416748046875, 41.17356872558594, 5.1978912353515625, 24.52695083618164, -53.782928466796875, 102.00286865234375, 119.99354553222656, 10.130096435546875, 16.56403350830078, -17.73269271850586, -17.028770446777344, 15.494827270507812, 34.91930389404297, 138.91502380371094, 116.30062866210938, 14.30484390258789, 136.1859588623047, 150.0679931640625, 1.5750160217285156, 4.0580596923828125, 102.97538757324219, 9.984100341796875, 13.18890380859375, 39.83146667480469, 92.16098022460938, 1.5424346923828125, 33.27191162109375, 62.90126037597656, 22.250110626220703, 3.2182750701904297, 78.17807006835938, 50.64396667480469, 64.18220520019531, -0.361846923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000379.npy"}
|
|
{"epoch": 0.5729402872260015, "step": 380, "batch_size": 64, "mean": 51.607852935791016, "std": 63.45000457763672, "min": -113.05413818359375, "p10": -20.08927974700927, "median": 47.4591064453125, "p90": 126.96358489990234, "max": 161.29351806640625, "pos_frac": 0.828125, "sample": [27.682537078857422, 124.53030395507812, 45.19671630859375, 27.0789794921875, -37.84700012207031, -69.5417709350586, 126.92219543457031, 143.91098022460938, -12.30105209350586, 95.48367309570312, 77.44123840332031, 10.920345306396484, -4.611576080322266, 62.1533317565918, 1.8239936828613281, -46.64131164550781, 3.909778594970703, 38.856170654296875, 111.35877990722656, 129.25119018554688, 85.41558837890625, 43.65674591064453, -12.250320434570312, -113.05413818359375, 114.8835678100586, 2.9426345825195312, 94.88957977294922, 64.46210479736328, -16.133623123168945, 33.558990478515625, 87.93548583984375, 120.34619140625, 49.72149658203125, 26.69992446899414, 161.29351806640625, 8.878593444824219, 122.25149536132812, 89.09207916259766, 10.692916870117188, 11.462745666503906, 118.1613540649414, -100.96402740478516, 123.84820556640625, 137.82171630859375, 42.283241271972656, 97.43518829345703, 0.9249801635742188, 87.59063720703125, -21.784561157226562, 94.02169799804688, 10.966331481933594, 6.604219436645508, -37.37477111816406, 153.21731567382812, 124.70405578613281, 111.5740737915039, 9.334817886352539, 135.41754150390625, 85.18516540527344, 126.9813232421875, 50.21662139892578, 67.76260375976562, 18.873319625854492, 17.778594970703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000380.npy"}
|
|
{"epoch": 0.5744520030234316, "step": 381, "batch_size": 64, "mean": 20.611921310424805, "std": 62.59636306762695, "min": -154.03016662597656, "p10": -48.736593627929686, "median": 10.209853172302246, "p90": 117.44828109741213, "max": 181.150634765625, "pos_frac": 0.65625, "sample": [10.678680419921875, 0.6451263427734375, 1.6586685180664062, -55.39549255371094, -154.03016662597656, 7.339611053466797, 3.1064376831054688, 18.14694595336914, 144.38128662109375, 52.92277145385742, 9.373771667480469, -8.710208892822266, 1.4927749633789062, 28.222557067871094, -6.08244514465332, 181.150634765625, 67.50741577148438, 0.3535575866699219, 38.01545715332031, -1.1214561462402344, 15.733915328979492, -60.42790985107422, 14.889959335327148, 129.67564392089844, -74.97319793701172, 73.62772369384766, -39.8360595703125, -43.445640563964844, 75.90623474121094, -2.7569408416748047, 45.70355987548828, 102.06402587890625, 16.575706481933594, 16.392135620117188, -9.088565826416016, 122.51023864746094, 27.322246551513672, -16.256702423095703, -83.06135559082031, 11.173782348632812, 120.04524993896484, -11.5213623046875, 20.98436737060547, 9.741025924682617, 31.697830200195312, 141.39219665527344, 66.87889862060547, 52.026283264160156, -3.186891555786133, 8.560249328613281, 6.080879211425781, 92.83473205566406, 111.38868713378906, -16.337814331054688, 20.764934539794922, -3.12652587890625, 90.4119873046875, -117.58404541015625, -50.07311248779297, 26.13148307800293, -1.347381591796875, -45.61804962158203, -22.40479278564453, 130.03952026367188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000381.npy"}
|
|
{"epoch": 0.5759637188208617, "step": 382, "batch_size": 64, "mean": 34.857486724853516, "std": 52.370582580566406, "min": -103.63990783691406, "p10": -18.433259201049804, "median": 27.769153594970703, "p90": 113.3430877685547, "max": 162.43994140625, "pos_frac": 0.734375, "sample": [-12.11627197265625, -4.451139450073242, 147.38040161132812, 38.17139434814453, 126.42347717285156, -8.405372619628906, 32.77958679199219, 26.126869201660156, 97.96343994140625, 55.276123046875, 46.1485595703125, -3.4663543701171875, -16.449691772460938, 27.726593017578125, 109.95536804199219, 16.9832763671875, -31.92840576171875, 1.718475341796875, -103.63990783691406, 83.855224609375, -1.732198715209961, 30.396316528320312, 29.862777709960938, 0.804107666015625, -24.804229736328125, 9.700286865234375, 55.832183837890625, 27.728851318359375, -52.13538360595703, 142.08340454101562, 39.53668212890625, 20.054584503173828, -34.94611740112305, -31.691816329956055, 127.46453857421875, -2.2856216430664062, -13.827205657958984, 0.1764373779296875, 27.80945587158203, 114.79496765136719, 39.066246032714844, 162.43994140625, 7.0749969482421875, 33.46299743652344, 15.361640930175781, 72.53472137451172, -13.348033905029297, -5.3627471923828125, 13.381256103515625, 73.06678771972656, 57.234161376953125, 89.9298095703125, -19.28335952758789, 32.475215911865234, 79.26287841796875, 1.9864387512207031, 62.748382568359375, 23.14914321899414, 104.37952423095703, 132.73179626464844, 65.37738037109375, 55.15992736816406, 34.57344436645508, 18.60291290283203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000382.npy"}
|
|
{"epoch": 0.5774754346182918, "step": 383, "batch_size": 64, "mean": 40.4869384765625, "std": 65.93172454833984, "min": -100.86428833007812, "p10": -43.562364196777345, "median": 31.696495056152344, "p90": 131.30811462402346, "max": 167.23858642578125, "pos_frac": 0.703125, "sample": [-0.2043323516845703, 101.9736328125, 90.60617065429688, 11.723922729492188, 34.72814178466797, -34.61518478393555, 73.24478149414062, 33.982322692871094, 5.036216735839844, -47.39997100830078, 9.866840362548828, -46.46537780761719, 51.98664474487305, 53.999786376953125, 80.88483428955078, 34.109344482421875, 101.91358184814453, -50.530128479003906, 140.7238311767578, 95.94783020019531, 34.141754150390625, -9.88629150390625, 128.1540985107422, -22.5660400390625, 126.76113891601562, 29.09527587890625, 31.772125244140625, -21.431686401367188, 28.028078079223633, 155.19161987304688, 115.55428314208984, -27.872085571289062, -8.393608093261719, 166.92410278320312, -5.988475799560547, 63.10292053222656, 108.81502532958984, -100.86428833007812, 13.393707275390625, 55.012298583984375, -66.33216094970703, 15.907562255859375, 32.69889831542969, 31.184724807739258, 141.74105834960938, -43.94374084472656, 17.780174255371094, 31.620864868164062, -42.6724853515625, 167.23858642578125, 22.359060287475586, 150.606689453125, 97.06986236572266, 9.563003540039062, -4.291328430175781, 126.98303985595703, 108.52763366699219, 14.25762939453125, -9.197189331054688, -81.23493957519531, 117.50608825683594, -34.08384704589844, 132.6598358154297, 54.7584228515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000383.npy"}
|
|
{"epoch": 0.5789871504157218, "step": 384, "batch_size": 64, "mean": 39.862464904785156, "std": 54.007843017578125, "min": -124.39144897460938, "p10": -21.6344814300537, "median": 40.06710433959961, "p90": 121.82428741455081, "max": 147.51589965820312, "pos_frac": 0.8125, "sample": [-36.02496337890625, 116.23809814453125, 37.5159912109375, 72.28656005859375, 9.63168716430664, 55.990379333496094, 81.23693084716797, 62.56946563720703, 124.21836853027344, -11.730770111083984, 38.18806457519531, 67.29606628417969, 28.00860595703125, 4.243339538574219, -6.910194396972656, 9.5291748046875, 6.194854736328125, 142.99530029296875, 7.3609619140625, 126.74198913574219, -8.431259155273438, 43.033935546875, 23.54521942138672, -25.878929138183594, -48.30724334716797, 57.03266906738281, 145.98480224609375, 43.27278137207031, -70.14249420166016, 72.98655700683594, 136.6725616455078, 82.11631774902344, 3.866544723510742, 40.34095001220703, 48.34867858886719, 5.3389739990234375, 147.51589965820312, 23.681697845458984, -124.39144897460938, 101.0671157836914, 52.200042724609375, 57.69923400878906, 83.38021087646484, -29.12481689453125, 135.26678466796875, 68.12905883789062, 46.303409576416016, 7.885950088500977, 13.243541717529297, 73.3304672241211, 65.19357299804688, 9.21331787109375, 39.79325866699219, 29.642486572265625, 13.051536560058594, -0.13573074340820312, -49.833412170410156, 48.83387756347656, -2.00128173828125, 5.628719329833984, 87.30484008789062, 78.51587677001953, 61.073577880859375, 23.370155334472656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000384.npy"}
|
|
{"epoch": 0.5804988662131519, "step": 385, "batch_size": 64, "mean": 48.64518737792969, "std": 55.96137619018555, "min": -46.85029602050781, "p10": -20.33966484069824, "median": 50.24795913696289, "p90": 131.7388916015625, "max": 152.35244750976562, "pos_frac": 0.75, "sample": [-45.13953399658203, -0.3806343078613281, 125.21733856201172, 81.53070068359375, 47.05381774902344, 20.350082397460938, 42.042236328125, 65.24545288085938, 13.411872863769531, 35.9185791015625, -30.292308807373047, 61.55438232421875, 50.689369201660156, 116.6585693359375, 56.57573699951172, 97.06141662597656, 129.714599609375, 93.28668212890625, -2.4270401000976562, 58.41160202026367, -17.132049560546875, -37.61943054199219, 8.664352416992188, 132.6064453125, 128.0890655517578, 135.12705993652344, 35.2187385559082, 49.806549072265625, 26.389862060546875, 73.81524658203125, 71.85856628417969, -42.03196716308594, 53.91188049316406, -1.0546112060546875, 61.615421295166016, 136.6402587890625, 106.7225341796875, 77.65408325195312, 51.064231872558594, 58.86412811279297, -20.66970443725586, -19.56957244873047, 68.79705810546875, 40.98345947265625, 4.959075927734375, 141.6190643310547, 143.5312042236328, -46.85029602050781, 3.9290504455566406, -10.615447998046875, -18.406890869140625, 36.60230255126953, -16.477325439453125, -0.8091278076171875, 83.86673736572266, 81.6596908569336, 6.662071228027344, 119.28463745117188, -25.334800720214844, 152.35244750976562, 7.560619354248047, 5.451469421386719, 137.67941284179688, 110.39329528808594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000385.npy"}
|
|
{"epoch": 0.582010582010582, "step": 386, "batch_size": 64, "mean": 32.316810607910156, "std": 55.39605712890625, "min": -121.38362121582031, "p10": -32.01499938964844, "median": 27.368663787841797, "p90": 99.68653717041016, "max": 151.11105346679688, "pos_frac": 0.703125, "sample": [94.55526733398438, -54.83518981933594, -1.6118583679199219, 73.61530303955078, 75.8582763671875, -5.785675048828125, -0.9933242797851562, 56.73484802246094, 48.679508209228516, 57.74254608154297, 12.53253173828125, 36.9747314453125, -85.06353759765625, 20.395553588867188, 11.831405639648438, 110.13785552978516, 81.80484008789062, 100.17678833007812, 93.39886474609375, 64.94905090332031, 19.853322982788086, 82.9331283569336, 48.872520446777344, -33.12993240356445, -28.666709899902344, 86.65211486816406, 105.239501953125, -33.30546569824219, 58.65766143798828, 55.690284729003906, 21.65900421142578, 64.99478149414062, -10.841911315917969, 11.1951904296875, -22.469005584716797, -30.154769897460938, 22.306861877441406, 26.001693725585938, 40.348934173583984, 124.00835418701172, 17.241729736328125, -25.079429626464844, 58.771949768066406, 68.7735595703125, 107.16315460205078, -21.604312896728516, -32.81224060058594, 28.735633850097656, -0.20108604431152344, 151.11105346679688, 2.4988327026367188, -23.864852905273438, -84.88116455078125, 98.54261779785156, 142.31622314453125, 43.187782287597656, 23.897705078125, 43.85493469238281, 79.06062316894531, 89.13238525390625, 22.130706787109375, -13.314327239990234, 14.054824829101562, -121.38362121582031], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000386.npy"}
|
|
{"epoch": 0.5835222978080121, "step": 387, "batch_size": 64, "mean": 45.44010925292969, "std": 54.053218841552734, "min": -67.60176086425781, "p10": -12.01925106048584, "median": 35.41669845581055, "p90": 128.93484954833986, "max": 191.28041076660156, "pos_frac": 0.84375, "sample": [41.58856201171875, 5.592437744140625, 100.87100982666016, -11.143899917602539, 3.1533966064453125, 135.4601593017578, 35.84608459472656, -13.706485748291016, 46.32445526123047, 191.28041076660156, 30.741060256958008, -2.331544876098633, 17.956939697265625, 34.1331787109375, 131.2407989501953, 4.586418151855469, 31.08453369140625, 97.30381774902344, 109.8150634765625, 21.857635498046875, 79.84979248046875, 138.36892700195312, 76.10212707519531, -20.86281967163086, 2.0382919311523438, 81.39166259765625, 0.9124526977539062, 8.278852462768555, 119.89619445800781, 46.75489807128906, 0.6755561828613281, -33.35240173339844, 34.98731231689453, 131.04275512695312, -31.494293212890625, 100.99547576904297, 62.398746490478516, 15.356462478637695, 10.32919692993164, 2.63262939453125, 2.279409408569336, 40.02545166015625, 78.19969177246094, -67.60176086425781, -22.846481323242188, 39.925716400146484, 79.0480728149414, 150.72061157226562, 28.755460739135742, 114.9497299194336, -10.554821014404297, 31.814682006835938, 19.048446655273438, 43.2908935546875, 38.24205017089844, 30.27658462524414, -12.394401550292969, 12.797872543334961, 51.948486328125, 124.01640319824219, 169.14645385742188, 38.95020294189453, 36.04833984375, 54.123931884765625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000387.npy"}
|
|
{"epoch": 0.5850340136054422, "step": 388, "batch_size": 64, "mean": 30.568408966064453, "std": 63.11652755737305, "min": -104.66266632080078, "p10": -37.10470123291015, "median": 19.99579620361328, "p90": 114.50547103881837, "max": 188.8636474609375, "pos_frac": 0.6875, "sample": [103.05316162109375, 61.08538818359375, 82.82572937011719, 153.57113647460938, 23.943824768066406, 98.04051208496094, 32.76240539550781, 20.76690673828125, 8.465003967285156, -17.378250122070312, 25.47132110595703, 107.8889389038086, -9.023002624511719, 64.47947692871094, -14.464645385742188, 188.8636474609375, 0.09722518920898438, 165.7609405517578, 16.762664794921875, 71.61488342285156, 91.09063720703125, 2.2487030029296875, 115.0611801147461, 10.390151977539062, 115.98868560791016, -104.66266632080078, 49.49322509765625, 44.01588439941406, 6.880403518676758, 5.187532424926758, -4.779397964477539, -13.629926681518555, -7.0225982666015625, 113.20881652832031, -17.795686721801758, 27.721649169921875, 25.469432830810547, -31.195281982421875, 44.32960510253906, -70.11820983886719, 57.182823181152344, 109.9185791015625, -87.43807983398438, 112.13346862792969, 105.16629028320312, 20.218490600585938, -3.1380844116210938, 6.918403625488281, 19.773101806640625, 116.47389221191406, -0.7180023193359375, 129.77903747558594, -35.95433044433594, 0.6995086669921875, -82.48876190185547, -37.59771728515625, -14.86236572265625, 42.390045166015625, -11.096920013427734, 4.270271301269531, 38.489990234375, -63.49403381347656, -65.35040283203125, 8.63365364074707], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000388.npy"}
|
|
{"epoch": 0.5865457294028723, "step": 389, "batch_size": 64, "mean": 44.879356384277344, "std": 60.86525344848633, "min": -118.03050994873047, "p10": -20.165683364868165, "median": 44.091819763183594, "p90": 130.43324203491213, "max": 165.88296508789062, "pos_frac": 0.71875, "sample": [-33.21772766113281, -13.793722152709961, 62.24488830566406, 55.96842956542969, 66.68211364746094, 39.94624328613281, 86.42228698730469, 109.71260833740234, 160.55657958984375, 73.90850830078125, -31.618194580078125, -22.522811889648438, 72.77732849121094, 77.08868408203125, 28.445999145507812, 66.79878997802734, 60.331024169921875, 120.26265716552734, 56.475677490234375, 27.950664520263672, 112.23260498046875, 97.82504272460938, 0.7885532379150391, 22.427154541015625, 93.11447143554688, -9.488670349121094, 106.40910339355469, -16.524314880371094, 20.572784423828125, 59.985321044921875, -12.076744079589844, 98.36669921875, -14.026924133300781, 125.61408233642578, 137.68331909179688, 23.677398681640625, 144.58441162109375, 76.41239166259766, 10.080150604248047, -118.03050994873047, 136.00906372070312, -4.219547271728516, 11.926229476928711, -9.870277404785156, 59.831687927246094, 1.40216064453125, 106.99736022949219, -31.14040756225586, -75.33958435058594, 16.547271728515625, -16.149620056152344, 7.539886474609375, 14.81280517578125, -5.867269515991211, -19.826377868652344, 143.55856323242188, 132.49859619140625, 90.69517517089844, 48.237396240234375, 165.88296508789062, -6.01756477355957, 96.46607971191406, 4.569089889526367, -20.311100006103516], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000389.npy"}
|
|
{"epoch": 0.5880574452003023, "step": 390, "batch_size": 64, "mean": 36.71540069580078, "std": 54.24372100830078, "min": -102.9295654296875, "p10": -27.648554229736327, "median": 33.57826042175293, "p90": 107.82233200073243, "max": 155.65261840820312, "pos_frac": 0.765625, "sample": [-39.88525390625, 48.01612091064453, 46.285804748535156, 13.806777954101562, 102.88703918457031, 31.683937072753906, 83.20506286621094, 92.55397033691406, -37.53346252441406, 51.052040100097656, 1.9864063262939453, -26.055465698242188, 82.58026123046875, 11.011701583862305, -2.3192214965820312, 155.65261840820312, 101.72020721435547, -29.032730102539062, 67.56116485595703, -28.33130645751953, 83.99005126953125, 6.089851379394531, 13.2940673828125, 127.49286651611328, 108.34393310546875, 106.60526275634766, 11.688507080078125, -54.85856628417969, 4.606861114501953, 91.10232543945312, -15.213325500488281, 9.2859525680542, -17.417936325073242, 39.447425842285156, -12.056575775146484, -20.83252716064453, -102.9295654296875, 90.7066650390625, 121.49197387695312, 0.8677635192871094, 121.61022186279297, 12.597253799438477, 11.12164306640625, 35.59352111816406, -0.9473228454589844, 53.14495849609375, 14.964714050292969, 29.028526306152344, 128.98532104492188, 39.32984161376953, 100.33216094970703, 61.44712448120117, 7.219306945800781, 38.116363525390625, 6.1079254150390625, 71.98367309570312, 4.700794219970703, -51.22392272949219, 35.47258377075195, -11.15363883972168, 55.0662841796875, 53.88232421875, 70.84180450439453, 143.01345825195312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000390.npy"}
|
|
{"epoch": 0.5895691609977324, "step": 391, "batch_size": 64, "mean": 45.503570556640625, "std": 66.39805603027344, "min": -119.27469635009766, "p10": -32.962836456298824, "median": 48.398298263549805, "p90": 127.32094116210938, "max": 194.62635803222656, "pos_frac": 0.75, "sample": [2.368896484375, -119.27469635009766, 135.41058349609375, 55.31358337402344, 49.052284240722656, 123.11299896240234, -1.9393959045410156, 125.09913635253906, 126.44416809082031, 184.826171875, 39.495452880859375, -95.68199157714844, 92.15834045410156, -13.661766052246094, 68.77003479003906, 151.60804748535156, -80.5467758178711, 26.663002014160156, -30.102066040039062, 20.145347595214844, 0.4395332336425781, 90.85711669921875, 73.397705078125, 54.88953399658203, 17.330467224121094, -28.191587448120117, -76.83628845214844, -18.486812591552734, 73.06333923339844, -9.441108703613281, 194.62635803222656, 85.05789184570312, 64.8896255493164, -14.196800231933594, 47.100921630859375, -0.4753684997558594, 24.160860061645508, 70.84100341796875, 68.02040100097656, 96.39746856689453, 71.58158874511719, 68.96427917480469, -57.62065124511719, -36.16062927246094, 64.05242156982422, 31.323741912841797, 136.5318603515625, 18.626426696777344, 118.3957748413086, 110.22307586669922, 6.60614013671875, -14.004013061523438, 127.69670104980469, 107.55238342285156, 140.74276733398438, 78.72697448730469, 86.83946228027344, 44.58790588378906, -34.188880920410156, 12.331745147705078, 21.210189819335938, 47.74431228637695, 83.08575439453125, 4.6736297607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000391.npy"}
|
|
{"epoch": 0.5910808767951625, "step": 392, "batch_size": 64, "mean": 48.990028381347656, "std": 68.50502014160156, "min": -120.02066040039062, "p10": -24.405120086669918, "median": 36.63347244262695, "p90": 144.76822814941409, "max": 232.15399169921875, "pos_frac": 0.828125, "sample": [-87.4544677734375, 0.2451629638671875, 37.06505584716797, 31.929183959960938, 17.776695251464844, -9.888496398925781, 130.89488220214844, -18.253700256347656, 85.28591918945312, 117.35884857177734, 36.20188903808594, -120.02066040039062, 81.93385314941406, 122.84822082519531, 157.48614501953125, 43.15370178222656, 7.438140869140625, 108.43373107910156, 136.7207489013672, 19.039321899414062, 38.96172332763672, 4.4931793212890625, 60.25507354736328, -60.30558776855469, 96.21017456054688, 98.50594329833984, 138.2844696044922, 42.489540100097656, 76.03121948242188, -63.65931701660156, 30.78582763671875, 149.89894104003906, 87.510498046875, 9.777725219726562, -49.14088439941406, 47.34093475341797, 19.72710609436035, 72.13224792480469, 73.79743957519531, 109.26356506347656, 159.84188842773438, 155.86083984375, -50.73926544189453, 131.88067626953125, 0.8711452484130859, 3.1366729736328125, 6.101490020751953, 13.4739990234375, 147.54698181152344, 12.295425415039062, -4.4288787841796875, 232.15399169921875, 5.416431427001953, 33.82196044921875, -27.04144287109375, 7.738386154174805, 157.87139892578125, -14.581085205078125, 19.294097900390625, 92.29818725585938, 12.46147346496582, 80.53068542480469, 17.48206329345703, 61.52088928222656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000392.npy"}
|
|
{"epoch": 0.5925925925925926, "step": 393, "batch_size": 64, "mean": 49.60386657714844, "std": 57.80107116699219, "min": -70.97676086425781, "p10": -13.306502342224118, "median": 40.866220474243164, "p90": 131.1794006347656, "max": 173.95220947265625, "pos_frac": 0.765625, "sample": [49.466468811035156, -4.11821174621582, 101.119384765625, 131.30101013183594, 132.16705322265625, 74.83768463134766, 23.121501922607422, 54.552345275878906, 53.600929260253906, 36.10774612426758, -6.812719345092773, 1.4170608520507812, 146.61932373046875, 99.99845886230469, 47.982582092285156, -18.824867248535156, 8.140022277832031, 131.59622192382812, -1.6725196838378906, 2.848804473876953, -10.143539428710938, 6.072046279907227, 32.463104248046875, 148.69818115234375, 83.36750793457031, 121.70332336425781, -14.662057876586914, 107.81217956542969, 91.05813598632812, 110.09518432617188, 49.96418762207031, -43.65109634399414, 130.89564514160156, 128.49586486816406, 1.7644844055175781, 45.62469482421875, -70.97676086425781, 137.53260803222656, -0.681549072265625, -4.239225387573242, 47.65387725830078, 18.06451416015625, 92.28843688964844, -17.510723114013672, 33.928016662597656, 83.44107055664062, 114.98220825195312, 25.556442260742188, 23.938480377197266, 101.60641479492188, 121.61299133300781, 173.95220947265625, 60.080970764160156, -7.920387268066406, 20.74810028076172, 79.22472381591797, 14.23419189453125, 31.354019165039062, -64.21150207519531, -23.922164916992188, 16.403728485107422, 106.88639831542969, -7.5121917724609375, 15.12652587890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000393.npy"}
|
|
{"epoch": 0.5941043083900227, "step": 394, "batch_size": 64, "mean": 39.414154052734375, "std": 50.62601852416992, "min": -102.59820556640625, "p10": -19.476305580139158, "median": 40.602752685546875, "p90": 104.04091644287111, "max": 161.9539794921875, "pos_frac": 0.828125, "sample": [26.056716918945312, 131.9415283203125, 99.70623779296875, 11.727378845214844, 7.230690002441406, 26.946487426757812, 2.3842620849609375, 31.104042053222656, 130.10321044921875, 44.84489440917969, 119.98092651367188, 61.85063934326172, 22.808650970458984, 33.2654914855957, -38.33638381958008, 54.263282775878906, -1.26287841796875, -20.421438217163086, -17.27099609375, 21.957439422607422, 74.99293518066406, -26.167675018310547, 52.685543060302734, 69.30085754394531, 59.11997985839844, 147.3551788330078, 85.21350860595703, 82.4329833984375, 61.67219543457031, 0.7332229614257812, -68.2786865234375, 14.024703979492188, 121.97675323486328, 56.2373046875, 27.59846305847168, 48.214569091796875, 0.7261867523193359, 38.31663513183594, 43.73046875, 13.57094955444336, 78.83232116699219, -33.34617233276367, -13.734100341796875, 39.290260314941406, 61.85850524902344, 161.9539794921875, -102.59820556640625, 88.35836791992188, -1.6305160522460938, 43.73936462402344, -69.77752685546875, 76.7473373413086, 33.580528259277344, 47.00138854980469, 42.08912658691406, 41.915245056152344, 44.73304748535156, 75.61705780029297, 11.590568542480469, 23.658905029296875, 9.933944702148438, 105.89863586425781, 37.067161560058594, 67.3902359008789], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000394.npy"}
|
|
{"epoch": 0.5956160241874527, "step": 395, "batch_size": 64, "mean": 34.710330963134766, "std": 55.7655029296875, "min": -88.86983489990234, "p10": -19.230447769165032, "median": 21.833864212036133, "p90": 113.41269912719727, "max": 174.15890502929688, "pos_frac": 0.75, "sample": [76.57013702392578, 101.98287963867188, 80.75247192382812, 4.9688568115234375, 95.0641098022461, 36.95555114746094, -2.6221141815185547, 59.41552734375, 138.64285278320312, -8.045000076293945, 13.839447021484375, 52.866668701171875, 112.60157775878906, 90.86746978759766, 12.805618286132812, 137.47653198242188, -5.591850280761719, 64.35401916503906, -6.7777099609375, 10.458641052246094, -1.4300384521484375, 11.154747009277344, 14.944244384765625, 36.23159408569336, 84.38449096679688, -88.86983489990234, 37.21580505371094, 113.76032257080078, 31.70418930053711, 1.5649948120117188, 43.481529235839844, 174.15890502929688, 30.504104614257812, 30.18193817138672, -51.32560729980469, 25.083465576171875, -21.786617279052734, 7.083026885986328, 99.44718170166016, -13.26605224609375, -3.988800048828125, 1.6500186920166016, 123.40528869628906, 12.885631561279297, 19.906970977783203, 50.46044921875, 1.4903640747070312, 134.34466552734375, -54.03019714355469, -8.177240371704102, 6.4156646728515625, 169.6149444580078, -33.897674560546875, 3.104816436767578, 23.760757446289062, 103.16732025146484, -43.27095413208008, 24.069473266601562, 48.02318572998047, -59.92833709716797, 11.398807525634766, -7.0402984619140625, 6.35875129699707, 60.929542541503906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000395.npy"}
|
|
{"epoch": 0.5971277399848829, "step": 396, "batch_size": 64, "mean": 41.708580017089844, "std": 56.201751708984375, "min": -97.21480560302734, "p10": -22.143891525268554, "median": 30.747821807861328, "p90": 121.7097869873047, "max": 190.6778106689453, "pos_frac": 0.796875, "sample": [137.07034301757812, 190.6778106689453, 115.75190734863281, 124.17713165283203, 100.18089294433594, 75.42616271972656, 19.164974212646484, 113.3621826171875, 32.86414337158203, 21.458206176757812, 60.02452850341797, 12.280471801757812, 86.29747772216797, -26.925861358642578, 31.70690155029297, -1.0296554565429688, 34.5352783203125, 73.38612365722656, 0.2836570739746094, 45.54479217529297, 40.474090576171875, -27.93701171875, 145.49404907226562, 36.753150939941406, 11.274375915527344, 31.29993438720703, 77.47329711914062, -97.21480560302734, 94.59039306640625, 72.28152465820312, 15.042682647705078, -34.96419906616211, 2.126678466796875, 115.75148010253906, 30.195709228515625, 23.299827575683594, 42.44472122192383, 14.314781188964844, 144.80819702148438, 141.40411376953125, -10.775074005126953, 20.509389877319336, -0.2618827819824219, 62.547882080078125, 10.417182922363281, 48.89308166503906, 24.10247802734375, 8.040504455566406, 61.2869873046875, -22.559425354003906, 17.794265747070312, -21.174312591552734, 16.50423240661621, 13.858551025390625, -4.738056182861328, 112.65518188476562, 41.88959503173828, 118.94114685058594, 5.3388824462890625, 19.010862350463867, -33.09419250488281, -64.84835815429688, -3.0367507934570312, 122.89634704589844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000396.npy"}
|
|
{"epoch": 0.5986394557823129, "step": 397, "batch_size": 64, "mean": 21.93632698059082, "std": 66.23133087158203, "min": -139.1895294189453, "p10": -59.45835189819335, "median": 22.328442573547363, "p90": 99.42006607055664, "max": 192.65432739257812, "pos_frac": 0.734375, "sample": [0.8521862030029297, 17.662086486816406, -65.02174377441406, 45.61933898925781, -27.73229217529297, 46.43498229980469, 70.0794677734375, 28.976341247558594, 97.44999694824219, 3.6383209228515625, 3.686939239501953, 50.74666976928711, 15.651176452636719, 73.55821228027344, 61.249000549316406, 64.71393585205078, -137.47817993164062, -80.95079040527344, 25.624256134033203, -8.922828674316406, 74.34691619873047, -19.907730102539062, 49.28290557861328, 87.52410888671875, 18.847389221191406, -139.1895294189453, -47.494110107421875, 9.704742431640625, 123.89647674560547, 65.13986206054688, 30.158287048339844, 19.032629013061523, 106.84972381591797, 33.092803955078125, 116.50630950927734, -10.229705810546875, 35.56940841674805, 81.31568908691406, 74.69461822509766, 4.861085891723633, 1.0767059326171875, 139.33367919921875, 100.2643814086914, 38.58641052246094, -139.08279418945312, 40.71148681640625, 192.65432739257812, -46.575927734375, -36.15705871582031, 0.2543792724609375, 2.1976661682128906, 33.92119598388672, 2.4782028198242188, 6.527313232421875, 0.7877674102783203, -20.433242797851562, 33.60859680175781, 34.32714080810547, -22.9373779296875, 89.93379211425781, -124.55096435546875, -64.58588409423828, -2.9662303924560547, 144.71234130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000397.npy"}
|
|
{"epoch": 0.600151171579743, "step": 398, "batch_size": 64, "mean": 49.83014678955078, "std": 58.95198059082031, "min": -51.498741149902344, "p10": -10.302153015136719, "median": 37.14751434326172, "p90": 135.1312683105469, "max": 240.30032348632812, "pos_frac": 0.765625, "sample": [101.34376525878906, 52.25594711303711, 88.19015502929688, 44.26255798339844, 135.4420166015625, 9.686580657958984, 140.5951385498047, 131.71881103515625, -12.633491516113281, 140.1289520263672, 77.86795043945312, 18.585983276367188, -9.719192504882812, -5.6453704833984375, 89.56292724609375, -1.3346405029296875, 83.82474517822266, -11.720321655273438, 45.24699401855469, -37.87189865112305, 70.8204345703125, 2.631612777709961, 142.95382690429688, 165.04478454589844, -2.0778274536132812, 123.46431732177734, 114.22351837158203, 72.92247772216797, 36.62012481689453, 29.5152587890625, -21.398540496826172, 36.702423095703125, 18.91448974609375, 80.6081314086914, 3.3237991333007812, -10.455284118652344, 27.1849365234375, 109.29475402832031, 10.040939331054688, 44.805755615234375, -26.157562255859375, -9.944847106933594, 106.37211608886719, 37.59260559082031, 76.41459655761719, 99.69820404052734, 12.198934555053711, 6.655967712402344, -6.995334625244141, 51.10075759887695, 5.173622131347656, 240.30032348632812, 26.082435607910156, -4.398582458496094, 57.14042663574219, -9.719047546386719, 134.40618896484375, -51.498741149902344, 41.24559783935547, 5.478067398071289, 13.909671783447266, 100.81436157226562, 138.0680389404297, 10.264200210571289], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000398.npy"}
|
|
{"epoch": 0.6016628873771731, "step": 399, "batch_size": 64, "mean": 55.661277770996094, "std": 65.16482543945312, "min": -65.47323608398438, "p10": -20.724655151367188, "median": 47.624637603759766, "p90": 137.15745849609377, "max": 272.8056640625, "pos_frac": 0.765625, "sample": [72.07096862792969, 35.171470642089844, 90.01329803466797, 135.5572509765625, -27.548858642578125, 79.50018310546875, 105.79414367675781, 23.786056518554688, -3.587329864501953, -7.579856872558594, -44.944610595703125, 78.60589599609375, 60.340606689453125, -55.038761138916016, 38.97431945800781, 47.96604919433594, 54.60478210449219, 60.008758544921875, 49.48777770996094, 165.27102661132812, -29.071746826171875, -4.8455810546875, -65.47323608398438, 89.88674926757812, 190.99801635742188, 26.884201049804688, 23.636821746826172, 272.8056640625, 19.48133087158203, 12.10687255859375, 15.040290832519531, -2.632078170776367, 40.263275146484375, 25.172985076904297, 27.964492797851562, -19.659263610839844, 47.751304626464844, 130.41217041015625, 48.98082733154297, 143.62294006347656, 44.419700622558594, 0.30823707580566406, 128.30007934570312, -8.8189697265625, 132.4456329345703, 79.04891204833984, 47.49797058105469, 39.788875579833984, -26.470840454101562, -21.181251525878906, 131.20480346679688, 137.84326171875, 133.18704223632812, 79.11894989013672, 98.32827758789062, 105.56716918945312, 93.9423828125, 160.80706787109375, 33.68898010253906, -16.494510650634766, 149.22833251953125, -6.086273193359375, 53.45594024658203, 41.412742614746094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000399.npy"}
|
|
{"epoch": 0.6031746031746031, "step": 400, "batch_size": 64, "mean": 32.258888244628906, "std": 58.32211685180664, "min": -104.13074493408203, "p10": -22.70907974243164, "median": 15.512781143188477, "p90": 112.93882141113282, "max": 166.17417907714844, "pos_frac": 0.6875, "sample": [-5.201328277587891, 62.69414520263672, 14.405303955078125, -21.391708374023438, 70.09039306640625, 93.80894470214844, 13.978263854980469, 84.76458740234375, 73.41854858398438, 113.87649536132812, 26.692691802978516, -41.267723083496094, 29.344539642333984, 2.836576461791992, -1.6221809387207031, 9.265899658203125, 0.172088623046875, 121.32366943359375, 91.11937713623047, 92.66917419433594, -21.626571655273438, 15.60611343383789, 61.006103515625, 8.012933731079102, -56.9501953125, 63.11817169189453, 12.474660873413086, 17.2957763671875, 12.82382583618164, -4.865606307983398, -67.41781616210938, -4.350368499755859, 77.30854797363281, 9.363021850585938, 142.1905059814453, -2.192584991455078, -61.34474182128906, 10.58620834350586, 110.75091552734375, 82.12355041503906, 130.77215576171875, 94.50714874267578, -104.13074493408203, 76.31121826171875, -0.5788650512695312, 15.419448852539062, 6.0866546630859375, -9.399795532226562, -7.49871826171875, 166.17417907714844, 30.14730453491211, 91.2037124633789, -23.173011779785156, 32.041160583496094, -16.770475387573242, 143.79588317871094, 149.0736541748047, 48.009613037109375, 18.190887451171875, -1.15850830078125, -85.3074951171875, -6.386974334716797, 17.91071319580078, 64.43954467773438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000400.npy"}
|
|
{"epoch": 0.6046863189720333, "step": 401, "batch_size": 64, "mean": 49.36147689819336, "std": 57.893821716308594, "min": -42.9671516418457, "p10": -19.97995319366455, "median": 37.04583740234375, "p90": 139.2564697265625, "max": 171.81790161132812, "pos_frac": 0.796875, "sample": [-0.89971923828125, 95.89551544189453, 96.85376739501953, 56.56083679199219, -21.05276870727539, 1.355010986328125, 17.57623291015625, 31.292728424072266, -29.997135162353516, 1.8501205444335938, 133.0301513671875, -1.5261688232421875, -42.9671516418457, -2.101226806640625, 35.93114471435547, -26.742530822753906, 85.26727294921875, 159.1707763671875, 102.0020751953125, 128.01034545898438, 52.213584899902344, 49.11781692504883, 20.55493927001953, 30.05026626586914, -30.799118041992188, -31.99425506591797, 33.47332763671875, 61.093589782714844, 38.16053009033203, 39.152503967285156, 90.52542114257812, 29.14264678955078, 15.928693771362305, 116.58614349365234, -4.13177490234375, 143.8697509765625, 4.124631881713867, 73.51934051513672, 8.623857498168945, 126.03424072265625, 2.913330078125, 140.257568359375, 74.15992736816406, -17.476716995239258, 137.49151611328125, 6.761249542236328, 27.724578857421875, 171.39834594726562, 127.23245239257812, 82.72460174560547, 50.94487762451172, 4.267751693725586, 18.815052032470703, 13.282310485839844, -32.252933502197266, 55.15174865722656, 51.26267623901367, 19.926538467407227, 140.01287841796875, 150.72848510742188, 49.55339050292969, -11.357067108154297, 39.038639068603516, 171.81790161132812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000401.npy"}
|
|
{"epoch": 0.6061980347694633, "step": 402, "batch_size": 64, "mean": 38.89942932128906, "std": 61.18667221069336, "min": -78.58012390136719, "p10": -46.740101623535146, "median": 30.98989963531494, "p90": 120.39372024536135, "max": 160.0277862548828, "pos_frac": 0.75, "sample": [69.5065689086914, 130.56802368164062, 88.10928344726562, 96.22371673583984, 82.75993347167969, 61.00233459472656, 160.0277862548828, 18.779220581054688, 30.417922973632812, 65.89490509033203, -8.835418701171875, 157.17385864257812, 33.173255920410156, -50.26470947265625, 37.896270751953125, 32.387996673583984, 2.467498779296875, 19.947023391723633, 18.726783752441406, 30.148086547851562, 108.58396911621094, 2.758373260498047, 2.1478805541992188, -78.58012390136719, 138.62619018554688, 3.113574981689453, 59.480018615722656, 104.12197875976562, 42.8175163269043, 22.332530975341797, 142.79490661621094, 104.63214111328125, -11.7000732421875, 59.79638671875, -4.3808441162109375, 31.56187629699707, 123.26740264892578, 113.68846130371094, 21.175926208496094, 61.00093078613281, -49.98243713378906, -59.124263763427734, -78.01321411132812, 73.26622009277344, 21.796539306640625, 10.119222640991211, 33.24684143066406, -27.179840087890625, 4.339927673339844, -0.20918655395507812, 95.21847534179688, 111.52078247070312, 96.41453552246094, 146.11817932128906, -17.747848510742188, -5.987569808959961, -8.529548645019531, -77.87789916992188, -72.49757385253906, 59.224761962890625, 23.502601623535156, 104.97649383544922, -39.174652099609375, 22.79369354248047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000402.npy"}
|
|
{"epoch": 0.6077097505668935, "step": 403, "batch_size": 64, "mean": 51.62788009643555, "std": 65.73644256591797, "min": -104.88151550292969, "p10": -23.91880760192871, "median": 34.107643127441406, "p90": 138.2383041381836, "max": 225.71572875976562, "pos_frac": 0.8125, "sample": [-4.3441925048828125, 1.93096923828125, 171.67666625976562, 26.509735107421875, 132.6636962890625, 26.509994506835938, 166.8719024658203, 86.08799743652344, 13.982627868652344, 11.671356201171875, 59.3001708984375, -26.4818115234375, 112.497314453125, -17.16940689086914, 125.90565490722656, 74.40963745117188, 38.022438049316406, 12.124837875366211, 23.846511840820312, 7.925355911254883, -11.36854362487793, -41.309261322021484, 225.71572875976562, 138.94406127929688, 0.584747314453125, 86.58122253417969, 131.78424072265625, 5.092229843139648, 134.03233337402344, 24.38697052001953, -0.10522842407226562, 45.78868103027344, 32.36109924316406, 23.819541931152344, 56.00060272216797, -25.460941314697266, 14.562582015991211, 139.7951202392578, 26.723054885864258, 61.419769287109375, 35.85418701171875, 0.8549823760986328, 136.59153747558594, 36.14521408081055, 106.04004669189453, -20.32049560546875, 92.953857421875, 99.72867584228516, 60.84530258178711, 30.344329833984375, 104.08609008789062, 142.80450439453125, -104.88151550292969, -26.75333023071289, 142.5880126953125, -83.9676513671875, 115.02046966552734, 55.625511169433594, 28.28622055053711, 132.25877380371094, -38.622535705566406, 6.701545715332031, 32.009925842285156, 106.70111083984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000403.npy"}
|
|
{"epoch": 0.6092214663643235, "step": 404, "batch_size": 64, "mean": 49.39562225341797, "std": 66.52606201171875, "min": -79.06941223144531, "p10": -31.75748672485351, "median": 43.069313049316406, "p90": 138.0685791015625, "max": 190.29632568359375, "pos_frac": 0.734375, "sample": [3.99658203125, 35.80853271484375, 80.69969177246094, 7.6111602783203125, 37.11177062988281, -7.919853210449219, 84.7237777709961, 13.817802429199219, -24.80185890197754, -27.674942016601562, 138.19635009765625, -12.04583740234375, -74.15109252929688, -79.06941223144531, -0.5973663330078125, 147.3673095703125, -16.929336547851562, -5.683887481689453, 137.64297485351562, 133.52871704101562, 76.55952453613281, 95.62946319580078, 91.08828735351562, 160.72781372070312, -37.2398681640625, 112.50971221923828, -33.50714874267578, 139.91348266601562, 90.16896057128906, 104.01155090332031, 12.375190734863281, 89.45500183105469, -11.37660026550293, 99.93956756591797, -14.031394958496094, 8.967933654785156, 20.378944396972656, 97.03278350830078, 69.02562713623047, 6.476696014404297, -4.117057800292969, 7.7271270751953125, 76.007080078125, 43.72291564941406, 124.34294891357422, 27.410770416259766, 81.71917724609375, 137.77044677734375, 150.6381072998047, 42.41571044921875, 1.9875144958496094, 190.29632568359375, 181.83572387695312, 45.886444091796875, 97.80923461914062, 97.4608383178711, 132.73118591308594, -49.59979248046875, -44.35585021972656, -57.42027282714844, 45.76445007324219, 55.75801467895508, 7.289508819580078, 18.502853393554688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000404.npy"}
|
|
{"epoch": 0.6107331821617535, "step": 405, "batch_size": 64, "mean": 39.45708465576172, "std": 58.6667594909668, "min": -76.11527252197266, "p10": -23.592994499206544, "median": 21.232280731201172, "p90": 129.52738037109376, "max": 161.15032958984375, "pos_frac": 0.65625, "sample": [126.13348388671875, -14.88558578491211, -36.58571243286133, 129.89218139648438, 35.76445770263672, 5.197090148925781, 126.16867065429688, -13.268402099609375, -1.3034400939941406, -42.38983154296875, -8.442794799804688, 6.152860641479492, 3.745523452758789, -10.462686538696289, 55.761085510253906, 14.573272705078125, 161.15032958984375, -2.810750961303711, -4.575462341308594, 80.6065673828125, 128.67617797851562, 110.41204833984375, 22.411537170410156, 73.57615661621094, -2.4902706146240234, 27.852359771728516, 95.83175659179688, 138.18092346191406, -24.395706176757812, 78.95133972167969, 17.520320892333984, -4.279193878173828, 96.81723022460938, 106.22283172607422, -21.850982666015625, 64.34845733642578, 18.26352310180664, -76.11527252197266, 42.603172302246094, 48.327117919921875, 20.053024291992188, -23.996110916137695, 37.41449737548828, 23.16862678527832, -2.7050552368164062, 12.348089218139648, 14.317787170410156, 36.42393493652344, 151.8919219970703, 106.17345428466797, -28.61971664428711, -5.435028076171875, 67.9315414428711, 19.976226806640625, 136.19622802734375, 143.88577270507812, -22.652389526367188, 45.81229782104492, 80.96277618408203, -40.19732666015625, -8.356971740722656, 148.41360473632812, 80.92985534667969, -19.967994689941406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000405.npy"}
|
|
{"epoch": 0.6122448979591837, "step": 406, "batch_size": 64, "mean": 50.081642150878906, "std": 64.3166732788086, "min": -125.40477752685547, "p10": -28.40374088287353, "median": 46.976585388183594, "p90": 134.18961486816406, "max": 162.7838592529297, "pos_frac": 0.75, "sample": [122.06629180908203, 20.88299560546875, -29.48177146911621, -12.314590454101562, 73.66264343261719, 85.07563018798828, 114.50959777832031, 125.06509399414062, -30.09089469909668, 159.2648162841797, 83.89949035644531, 92.9271011352539, 3.9631385803222656, -12.99169921875, 124.41014862060547, 149.0790252685547, 134.56167602539062, 22.9282169342041, 41.626220703125, 137.1056365966797, 71.3984375, 26.388587951660156, 77.53559875488281, 85.33084869384766, 25.361244201660156, 81.15968322753906, -13.68838882446289, 81.47314453125, -3.332040786743164, 26.513134002685547, -1.2454795837402344, -2.8921966552734375, -80.64089965820312, -10.628437042236328, 46.49717712402344, 127.70369720458984, 29.638978958129883, 143.218994140625, -47.70309829711914, 141.29525756835938, 73.08760070800781, 18.840774536132812, 32.813072204589844, 162.7838592529297, 9.80523681640625, 133.32147216796875, -53.91086196899414, 12.37858772277832, 71.47419738769531, 131.75335693359375, 20.58851432800293, -25.888336181640625, 103.07038879394531, -51.875091552734375, 16.509872436523438, -125.40477752685547, 50.20660400390625, 91.36807250976562, -8.306884765625, 18.152694702148438, 115.52610778808594, 61.340484619140625, 47.45599365234375, 90.60125732421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000406.npy"}
|
|
{"epoch": 0.6137566137566137, "step": 407, "batch_size": 64, "mean": 40.23088836669922, "std": 64.3543701171875, "min": -117.94451904296875, "p10": -28.80393085479736, "median": 31.716745376586914, "p90": 130.38849487304688, "max": 179.2663116455078, "pos_frac": 0.703125, "sample": [4.2816162109375, 7.850648880004883, -9.373672485351562, 79.33223724365234, -19.479896545410156, 116.16435241699219, 134.6887969970703, -29.769210815429688, -80.2674560546875, 84.55249786376953, 43.08085632324219, 28.605640411376953, 1.3645477294921875, -0.0207061767578125, -59.955596923828125, 10.42645263671875, 53.21012878417969, 92.16504669189453, 148.24484252929688, 131.00167846679688, 34.827850341796875, 5.414918899536133, 54.337398529052734, 44.554595947265625, 53.10186767578125, 54.946685791015625, 2.5828018188476562, -9.038330078125, 93.425537109375, 163.35317993164062, 76.61286926269531, 2.2528305053710938, 131.01406860351562, 124.02108764648438, 104.95782470703125, 1.8154296875, -51.813873291015625, 115.08309936523438, -39.35835266113281, 24.876243591308594, -4.1976776123046875, 98.99774169921875, 110.15869140625, 52.83654022216797, 136.68453979492188, 179.2663116455078, 85.95262908935547, 12.446083068847656, 128.95773315429688, -1.5339622497558594, 18.876174926757812, -117.94451904296875, -63.49346923828125, -16.431549072265625, -26.551610946655273, 23.465496063232422, -19.481950759887695, 110.90301513671875, -26.342891693115234, -5.968559265136719, -4.479316711425781, 65.18984985351562, 58.07050323486328, 56.326873779296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000407.npy"}
|
|
{"epoch": 0.6152683295540439, "step": 408, "batch_size": 64, "mean": 57.371498107910156, "std": 60.55350112915039, "min": -80.39784240722656, "p10": -15.459552192687989, "median": 45.42769813537598, "p90": 142.37139587402345, "max": 166.35565185546875, "pos_frac": 0.8125, "sample": [22.425691604614258, 134.81936645507812, 120.3907470703125, 63.79670333862305, -1.9141731262207031, -48.98188781738281, -80.39784240722656, 7.844017028808594, 103.42465209960938, 49.853973388671875, 33.40458679199219, 146.69940185546875, -1.2177104949951172, 75.31317138671875, 109.56675720214844, 141.312744140625, 22.518306732177734, -26.08965301513672, 73.2746810913086, 22.746139526367188, 71.74221801757812, 4.529502868652344, -5.620706558227539, 91.53943634033203, 166.35565185546875, -37.99504089355469, 35.42060089111328, 144.61721801757812, 93.28462982177734, 104.57232666015625, 142.82510375976562, -15.604183197021484, 84.62041473388672, 4.579671859741211, 18.77880859375, 133.31558227539062, 41.00142288208008, -18.54473304748535, 160.23190307617188, 40.04185485839844, 40.8897705078125, 123.3834228515625, 131.3097381591797, 151.67396545410156, 111.84419250488281, 18.8619441986084, 118.54351043701172, -15.122079849243164, 33.83543395996094, 11.251998901367188, 110.7852554321289, 68.78416442871094, 143.9534912109375, -61.653297424316406, 91.58988189697266, 80.37362670898438, 33.23149490356445, -2.4659500122070312, 18.442031860351562, 27.730892181396484, 87.84830474853516, 38.5591926574707, 5.71636962890625, 73.92695617675781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000408.npy"}
|
|
{"epoch": 0.6167800453514739, "step": 409, "batch_size": 64, "mean": 29.109418869018555, "std": 59.37889862060547, "min": -107.0844955444336, "p10": -31.787866210937498, "median": 19.533392906188965, "p90": 114.28034591674805, "max": 157.43392944335938, "pos_frac": 0.625, "sample": [3.0208740234375, 6.203630447387695, -23.796951293945312, -16.801895141601562, 3.7953758239746094, -12.064868927001953, 31.10032844543457, -18.499553680419922, 26.248126983642578, -22.64527130126953, -23.644763946533203, 141.378173828125, -3.0800342559814453, 150.08993530273438, 82.94090270996094, 8.08294677734375, 50.346435546875, -5.669269561767578, -22.523487091064453, 146.32528686523438, 101.51522064208984, -33.360721588134766, 92.30262756347656, 113.07804870605469, 15.109596252441406, -36.51918029785156, -6.4736328125, 49.47650146484375, 61.23448181152344, -4.07208251953125, -27.966644287109375, 41.96026611328125, 75.06266021728516, -28.117870330810547, 76.15401458740234, -4.245328903198242, 114.79561614990234, 98.41535186767578, 43.43778991699219, 23.957189559936523, -38.4083137512207, -24.19994354248047, 51.16719055175781, -54.14046859741211, 14.912040710449219, -11.067535400390625, 6.7312469482421875, 33.000003814697266, -91.90383911132812, 145.16844177246094, 25.279422760009766, 86.359130859375, -0.298553466796875, 149.6292724609375, -107.0844955444336, 9.117889404296875, 32.985137939453125, 32.865623474121094, 64.83185577392578, 68.58358764648438, 157.43392944335938, -34.94830322265625, 49.960906982421875, 30.478771209716797], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000409.npy"}
|
|
{"epoch": 0.618291761148904, "step": 410, "batch_size": 64, "mean": 46.21394729614258, "std": 51.27942657470703, "min": -69.76958465576172, "p10": -16.57084808349609, "median": 39.63755798339844, "p90": 109.16769104003907, "max": 166.6189727783203, "pos_frac": 0.828125, "sample": [36.70777130126953, 47.043983459472656, 18.625408172607422, -0.9203262329101562, 139.85220336914062, 14.8819580078125, 73.60801696777344, 38.42396926879883, -22.437240600585938, -69.76958465576172, 64.34123229980469, 107.16401672363281, 3.8671188354492188, -40.42107391357422, 101.8968505859375, 69.0452880859375, 95.52757263183594, -17.17620849609375, 54.224517822265625, 137.9491424560547, 166.6189727783203, 40.00740051269531, 18.838775634765625, 66.74964141845703, 9.88568115234375, 71.67328643798828, 14.470626831054688, 75.27655029296875, 39.22765350341797, -22.705116271972656, 78.7995376586914, -15.158340454101562, 25.000957489013672, 109.67042541503906, 30.231216430664062, 160.34732055664062, 89.85159301757812, 16.89194107055664, 25.951723098754883, 62.20416259765625, -11.389289855957031, -31.260604858398438, 46.86936950683594, 39.26771545410156, 87.25455474853516, 77.837158203125, 107.99464416503906, -2.267993927001953, 16.081134796142578, 54.906097412109375, 22.645965576171875, 55.853973388671875, 164.87103271484375, 71.44793701171875, 22.18927764892578, 58.591026306152344, 6.898357391357422, 20.06300163269043, 5.034332275390625, 80.21630859375, 122.99285888671875, 42.71520233154297, -39.82489013671875, 22.43675994873047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000410.npy"}
|
|
{"epoch": 0.6198034769463341, "step": 411, "batch_size": 64, "mean": 46.03551483154297, "std": 65.43930053710938, "min": -131.93154907226562, "p10": -25.566012573242187, "median": 41.847076416015625, "p90": 134.64074096679687, "max": 217.87435913085938, "pos_frac": 0.703125, "sample": [-12.970191955566406, 128.72706604003906, 103.11164855957031, 122.91180419921875, 74.54798126220703, -9.513679504394531, -9.685256958007812, -131.93154907226562, 66.75531768798828, 0.3163909912109375, 52.03990936279297, 145.76690673828125, 80.7592544555664, 60.752559661865234, 125.57278442382812, 57.664695739746094, 68.69296264648438, -32.420654296875, 54.22248077392578, 43.23895263671875, 21.387123107910156, 141.6875762939453, 135.42153930664062, 17.76536750793457, -18.22900390625, 86.13526916503906, 3.3280296325683594, 29.470172882080078, -24.81802749633789, -52.45414733886719, 12.26312255859375, -14.863321304321289, -29.66529083251953, 40.4552001953125, 55.85221481323242, 32.19405746459961, -7.780908584594727, 132.92889404296875, -3.913177490234375, 217.87435913085938, 96.67922973632812, 34.95548629760742, -25.271682739257812, 135.3743896484375, 115.72599029541016, 142.66177368164062, -53.58100128173828, 30.747390747070312, -6.591678619384766, -31.2838134765625, 30.606767654418945, -25.692153930664062, 132.72767639160156, 105.48654174804688, 50.08910369873047, 92.4875717163086, 61.59747314453125, 22.321748733520508, 177.88677978515625, 43.4599609375, -8.455116271972656, 2.50836181640625, -10.957473754882812, 69.19100189208984], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000411.npy"}
|
|
{"epoch": 0.6213151927437641, "step": 412, "batch_size": 64, "mean": 42.40864181518555, "std": 56.323463439941406, "min": -54.51789093017578, "p10": -23.961349868774413, "median": 36.478261947631836, "p90": 120.66222229003907, "max": 166.1996307373047, "pos_frac": 0.78125, "sample": [-54.51789093017578, 120.82139587402344, 44.670494079589844, 149.06427001953125, 41.97681427001953, 76.79407501220703, 38.18605041503906, 114.62692260742188, -35.418521881103516, 0.43433570861816406, 58.152854919433594, 4.092826843261719, 53.8641357421875, 141.6045684814453, -2.401214599609375, -24.071937561035156, 104.32124328613281, -46.702938079833984, 25.342979431152344, 1.148468017578125, 127.28407287597656, 71.90248107910156, 16.304428100585938, 38.82832336425781, 44.54425048828125, -51.13318634033203, 25.830123901367188, 21.24813461303711, 42.75113296508789, 24.37939453125, 7.64207649230957, 150.8264923095703, 4.689455032348633, 166.1996307373047, -17.094100952148438, 10.188491821289062, 32.6419677734375, -13.108638763427734, 98.679931640625, -2.4452743530273438, -19.18900489807129, 0.0992279052734375, 76.82377624511719, -39.09959030151367, 15.66973876953125, 120.29081726074219, 100.19375610351562, 91.3849105834961, 148.5754852294922, -23.703311920166016, 43.72716522216797, 69.09703063964844, 64.79014587402344, 6.331428527832031, 34.77047348022461, 110.35888671875, -2.2572879791259766, 52.635772705078125, -27.53366470336914, 47.45008850097656, 119.82455444335938, 1.1014080047607422, 105.12003326416016, 5.54315185546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000412.npy"}
|
|
{"epoch": 0.6228269085411943, "step": 413, "batch_size": 64, "mean": 39.000335693359375, "std": 71.28355407714844, "min": -139.8255157470703, "p10": -45.87035636901855, "median": 40.87965774536133, "p90": 128.70740051269533, "max": 238.02024841308594, "pos_frac": 0.734375, "sample": [4.923595428466797, -54.16902160644531, 79.33443450927734, 40.32820129394531, -8.377601623535156, -15.836380004882812, 2.522613525390625, 150.25576782226562, -45.128875732421875, 143.56069946289062, -113.85930633544922, 113.35233306884766, 75.29682159423828, -39.34584045410156, -4.0293121337890625, 16.549148559570312, -45.78439712524414, 125.963134765625, -0.432586669921875, 33.437294006347656, 160.013671875, 57.83360290527344, 42.90476989746094, 35.55302810668945, 84.6800537109375, 38.830116271972656, 238.02024841308594, 70.1656494140625, 3.9354171752929688, 34.44731903076172, 50.36891174316406, 87.01026153564453, 65.39457702636719, 43.922019958496094, 150.8241424560547, -64.26950073242188, 63.332244873046875, 12.088760375976562, 50.81774139404297, 14.975204467773438, -108.90648651123047, 114.84412384033203, 118.76396942138672, -53.08539581298828, 62.655517578125, -45.907196044921875, -139.8255157470703, 41.431114196777344, 48.45209503173828, 14.420063018798828, -13.712066650390625, -38.0943603515625, 129.88351440429688, 0.17162704467773438, 34.170074462890625, 90.01371765136719, 45.13084030151367, 124.43109130859375, 26.96872329711914, 67.31795501708984, -43.67364501953125, 92.21247863769531, 137.35484313964844, 91.59556579589844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000413.npy"}
|
|
{"epoch": 0.6243386243386243, "step": 414, "batch_size": 64, "mean": 46.04102325439453, "std": 59.27863311767578, "min": -105.95695495605469, "p10": -20.42852706909179, "median": 42.535823822021484, "p90": 132.73482742309574, "max": 158.59022521972656, "pos_frac": 0.75, "sample": [106.32166290283203, -23.28199005126953, 90.96372985839844, 70.86709594726562, 9.75439453125, 39.94011688232422, -74.27459716796875, 99.88941955566406, 51.40270233154297, 15.296829223632812, 48.146148681640625, 106.29096984863281, 97.13216400146484, -36.449459075927734, 125.92511749267578, 59.10648727416992, -68.40666198730469, -37.12346649169922, 17.772071838378906, 50.527809143066406, 32.930694580078125, 45.13153076171875, 156.62744140625, -5.032600402832031, 147.3521728515625, 22.199188232421875, 56.53419876098633, 28.05506134033203, 67.2684326171875, 86.09236145019531, 135.6532745361328, 18.766464233398438, 79.6755142211914, -5.412406921386719, 146.58372497558594, 49.57646942138672, 112.13089752197266, 38.439483642578125, 61.40093994140625, 105.26473999023438, 35.73857116699219, -13.77044677734375, 12.959217071533203, -0.8449840545654297, 80.17460632324219, -4.1216278076171875, -1.0346298217773438, 82.57656860351562, -105.95695495605469, 31.923606872558594, 105.363037109375, -3.6011199951171875, -3.3594436645507812, 20.829875946044922, 141.56982421875, 155.28741455078125, 22.625892639160156, 8.97802734375, 61.18746566772461, 69.55301666259766, -40.65171813964844, -4.145429611206055, 7.716289520263672, 158.59022521972656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000414.npy"}
|
|
{"epoch": 0.6258503401360545, "step": 415, "batch_size": 64, "mean": 42.49724578857422, "std": 64.37648010253906, "min": -98.69310760498047, "p10": -34.191297912597655, "median": 33.71407699584961, "p90": 135.00726013183598, "max": 193.82293701171875, "pos_frac": 0.734375, "sample": [-5.773378372192383, 93.36314392089844, 32.65386199951172, 72.13955688476562, 22.022823333740234, 13.457328796386719, 112.14179992675781, 148.76547241210938, -32.747344970703125, 98.6092758178711, 57.85907745361328, -43.114990234375, 106.36119842529297, 126.07713317871094, 4.608917236328125, 138.83445739746094, 159.08853149414062, 54.92295837402344, 82.46592712402344, -2.3154735565185547, -13.252304077148438, 72.82167053222656, 50.49176025390625, -0.26197052001953125, 34.7742919921875, 141.4539794921875, -92.28511047363281, 19.23741912841797, 64.76597595214844, 95.12991333007812, -22.666458129882812, 85.89198303222656, 49.6142578125, -8.676658630371094, -97.0025863647461, 193.82293701171875, 80.0856704711914, 86.31820678710938, 118.63018798828125, -34.81013488769531, 65.29203033447266, -85.19874572753906, -35.01374816894531, -98.69310760498047, 22.751625061035156, 63.567543029785156, -0.8346633911132812, 25.338333129882812, 7.158058166503906, 4.782304763793945, 2.174276351928711, 51.744789123535156, 31.82513427734375, 13.143718719482422, 14.65057373046875, 54.554325103759766, -1.9944591522216797, 94.8064193725586, 84.79067993164062, -3.1564102172851562, 25.78960418701172, 27.705223083496094, 149.5203399658203, 141.61676025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000415.npy"}
|
|
{"epoch": 0.6273620559334845, "step": 416, "batch_size": 64, "mean": 34.15831756591797, "std": 67.86822509765625, "min": -113.92300415039062, "p10": -35.20976867675781, "median": 29.384897232055664, "p90": 133.25684051513673, "max": 235.34011840820312, "pos_frac": 0.703125, "sample": [7.697502136230469, -113.92300415039062, -10.592060089111328, 55.99229431152344, 24.55694580078125, 138.8946990966797, -31.013687133789062, -33.05000305175781, 134.19754028320312, 148.53756713867188, 31.688003540039062, 142.99693298339844, 235.34011840820312, 11.632080078125, 0.3975944519042969, 45.92174530029297, 2.0838184356689453, 62.992637634277344, -102.48678588867188, -35.311309814453125, 13.617979049682617, 76.49958801269531, 35.028961181640625, 116.94253540039062, -1.310049057006836, 78.399658203125, 63.413787841796875, 152.25784301757812, 163.47213745117188, 69.3221206665039, 35.72496032714844, 66.30227661132812, -11.74676513671875, -53.981239318847656, 7.320680618286133, -0.9690780639648438, -48.37547302246094, -89.8138427734375, 5.596372604370117, 7.9114837646484375, -33.87724685668945, 36.01750946044922, -14.495231628417969, 3.9384918212890625, 45.128204345703125, -21.163127899169922, 97.04425811767578, -77.7791748046875, 0.40343284606933594, 78.9302978515625, 38.494171142578125, 6.803247451782227, -34.97283935546875, 86.12984466552734, -10.638908386230469, 27.081790924072266, -1.3676071166992188, 43.67536926269531, 72.46955871582031, 44.855979919433594, 120.04080200195312, 104.23661041259766, 131.06187438964844, 41.94842529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000416.npy"}
|
|
{"epoch": 0.6288737717309146, "step": 417, "batch_size": 64, "mean": 47.312744140625, "std": 68.42310333251953, "min": -180.66580200195312, "p10": -30.44071578979492, "median": 48.59893989562988, "p90": 126.33426971435549, "max": 223.95887756347656, "pos_frac": 0.78125, "sample": [24.03728485107422, 1.7703514099121094, -11.266563415527344, -133.3863525390625, 133.99703979492188, 111.80696868896484, -180.66580200195312, 134.16226196289062, 48.465213775634766, 91.51243591308594, 223.95887756347656, 37.30851745605469, 32.64801788330078, 56.8775634765625, 62.576805114746094, -66.70103454589844, 127.47337341308594, 65.82840728759766, 4.317474365234375, 58.09796905517578, 83.21934509277344, 48.732666015625, 65.22454833984375, 38.151763916015625, 60.44305419921875, -21.399917602539062, 112.22697448730469, 38.47029113769531, 17.684276580810547, 96.9540786743164, -32.558937072753906, 117.38531494140625, 141.04649353027344, -60.35270690917969, 105.60916137695312, -19.90569305419922, 34.70198059082031, -5.9504241943359375, 113.02130889892578, 86.12413787841797, 132.9194793701172, 10.211936950683594, 123.67636108398438, 63.654380798339844, -40.74688720703125, 29.060256958007812, 13.628999710083008, 49.92937469482422, -0.54632568359375, 60.819244384765625, 25.829360961914062, 104.79409790039062, 9.457901000976562, -25.498199462890625, 78.61508178710938, -36.191749572753906, 69.3791275024414, 43.216190338134766, 43.301849365234375, -2.2512683868408203, 101.19525146484375, 41.97574996948242, 116.331298828125, 173.60772705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000417.npy"}
|
|
{"epoch": 0.6303854875283447, "step": 418, "batch_size": 64, "mean": 35.868492126464844, "std": 67.08555603027344, "min": -126.64920043945312, "p10": -41.55298919677734, "median": 29.402145385742188, "p90": 116.11672821044925, "max": 254.52735900878906, "pos_frac": 0.703125, "sample": [66.81524658203125, 90.4311294555664, 98.95623016357422, 39.857330322265625, 42.402530670166016, 108.5381088256836, 27.97229766845703, 87.25669860839844, 149.83119201660156, -40.60582733154297, 119.12946319580078, 145.6342315673828, -21.13677978515625, 86.24867248535156, 48.81438446044922, -60.363250732421875, 95.59910583496094, 11.653892517089844, 105.97581481933594, -17.83983612060547, 4.593223571777344, 254.52735900878906, 57.71283721923828, 1.843994140625, 136.89004516601562, 32.318702697753906, -23.20318603515625, 8.095939636230469, 18.513744354248047, -19.575958251953125, -126.64920043945312, -66.2313232421875, 143.6166534423828, 32.23974609375, 149.90965270996094, -15.070816040039062, 9.344745635986328, -15.851943969726562, 29.1021728515625, 5.312774658203125, -45.281944274902344, -41.95891571044922, -75.65974426269531, -9.055635452270508, 70.61652374267578, -27.007308959960938, -5.834648132324219, 33.77573013305664, 6.135992050170898, -4.315788269042969, 24.84449005126953, 61.15374755859375, 109.0870132446289, 104.84930419921875, 25.94792938232422, 19.007904052734375, -76.68275451660156, 61.93522644042969, 66.57472229003906, 87.41122436523438, 63.54706573486328, -33.02104187011719, 47.202423095703125, 29.702117919921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000418.npy"}
|
|
{"epoch": 0.6318972033257747, "step": 419, "batch_size": 64, "mean": 52.30290222167969, "std": 67.78700256347656, "min": -99.27902221679688, "p10": -21.453517150878906, "median": 34.16430473327637, "p90": 150.99341583251953, "max": 202.1077117919922, "pos_frac": 0.84375, "sample": [-14.30599594116211, 192.53285217285156, 16.791770935058594, 12.161048889160156, -49.68931198120117, 29.9710693359375, 150.53257751464844, 27.648509979248047, 3.3257904052734375, 21.290695190429688, -99.27902221679688, 126.54472351074219, 125.33848571777344, 78.34254455566406, -28.16065216064453, 17.367538452148438, 65.86495971679688, 30.068389892578125, 151.19091796875, 14.936477661132812, 42.942054748535156, 73.0374984741211, -22.347488403320312, 62.71183776855469, 11.69818115234375, 3.7436676025390625, 161.54409790039062, -19.367584228515625, 165.94456481933594, 86.15029907226562, 7.055744171142578, -61.85557556152344, -24.065677642822266, 138.48788452148438, 202.1077117919922, 59.48910140991211, 102.201416015625, 17.641769409179688, 6.9391632080078125, 47.90837097167969, 7.189199447631836, 151.3962860107422, 12.257881164550781, 9.712173461914062, 138.42942810058594, 11.7340087890625, 73.82788848876953, 38.26021957397461, 14.378253936767578, 145.93572998046875, 129.3273162841797, 0.3728599548339844, 41.020565032958984, 21.058509826660156, 102.4270248413086, -6.197935104370117, 82.43927764892578, 87.31137084960938, 168.78053283691406, 4.407501220703125, 68.07810974121094, 63.9107666015625, -75.78533172607422, 122.67381286621094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000419.npy"}
|
|
{"epoch": 0.6334089191232048, "step": 420, "batch_size": 64, "mean": 27.279382705688477, "std": 71.1070556640625, "min": -156.86508178710938, "p10": -44.213609313964845, "median": 16.497394561767578, "p90": 129.20306396484378, "max": 174.72943115234375, "pos_frac": 0.65625, "sample": [-11.260330200195312, -41.710636138916016, 111.42986297607422, 8.671928405761719, 92.49085235595703, -102.58207702636719, 19.761398315429688, 11.967571258544922, 16.13788604736328, -138.41207885742188, -13.02471923828125, 54.40673828125, 7.28033447265625, 105.68900299072266, 0.74761962890625, 109.69638061523438, 163.52841186523438, 73.86233520507812, 153.4091339111328, 115.12207794189453, 16.075048446655273, -31.89101791381836, 3.0297412872314453, 41.49549102783203, 58.83296203613281, 132.69168090820312, -64.64289855957031, -69.65325164794922, 44.062835693359375, 8.953144073486328, -67.08209228515625, 16.856903076171875, -21.164344787597656, 22.428699493408203, 0.5612297058105469, -44.80702209472656, -12.611457824707031, -34.32610321044922, 118.37307739257812, 38.70704650878906, -12.040435791015625, -156.86508178710938, -42.8289794921875, -4.3875274658203125, 121.06295776367188, 45.484886169433594, 29.429325103759766, 142.83485412597656, 3.6999359130859375, 45.66400146484375, 146.2784423828125, 51.449989318847656, 91.66566467285156, -0.9277439117431641, 27.504451751708984, 84.173095703125, 147.1312255859375, -24.58324432373047, 18.809154510498047, 26.178817749023438, -0.8624420166015625, 174.72943115234375, -37.06459045410156, -23.757015228271484], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000420.npy"}
|
|
{"epoch": 0.6349206349206349, "step": 421, "batch_size": 64, "mean": 39.28024673461914, "std": 77.73573303222656, "min": -142.1602325439453, "p10": -73.21754989624021, "median": 31.891183853149414, "p90": 138.97351684570313, "max": 226.55838012695312, "pos_frac": 0.765625, "sample": [-108.49526977539062, 128.93710327148438, 5.373218536376953, 117.60218811035156, 106.08175659179688, 26.81060791015625, 58.04939270019531, 3.774322509765625, 2.1377182006835938, 226.55838012695312, 183.6084747314453, -56.71025085449219, 138.87789916992188, 49.68169403076172, -23.45248794555664, 30.766178131103516, 22.691627502441406, -29.972198486328125, 139.38116455078125, 76.98883056640625, 93.53385925292969, 77.72772216796875, 1.7634105682373047, 2.490192413330078, 54.9736328125, -3.1582679748535156, 81.61992645263672, -142.1602325439453, -106.5037612915039, -38.43618392944336, 21.961061477661133, 133.55599975585938, 109.13436126708984, 9.418478012084961, 37.065711975097656, 145.56597900390625, 139.01449584960938, 113.54683685302734, 16.496429443359375, 3.6448326110839844, 133.3566131591797, -4.840312957763672, 41.576812744140625, -113.07345581054688, -80.29210662841797, 78.220947265625, 0.544403076171875, 44.51384735107422, -88.1455078125, 17.064620971679688, 155.8060302734375, 141.65774536132812, 8.185264587402344, 33.01618957519531, 20.383682250976562, -83.03265380859375, 35.40227508544922, 59.91217041015625, 70.15715026855469, -38.577850341796875, 128.01121520996094, 106.23079681396484, 2.3651046752929688, -4.452213287353516], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000421.npy"}
|
|
{"epoch": 0.636432350718065, "step": 422, "batch_size": 64, "mean": 51.23059844970703, "std": 62.362464904785156, "min": -63.57942199707031, "p10": -26.80270309448242, "median": 44.258052825927734, "p90": 127.30783233642578, "max": 182.12240600585938, "pos_frac": 0.78125, "sample": [120.6663589477539, 179.85787963867188, -61.08965301513672, -2.831584930419922, 126.53536987304688, -12.507049560546875, 124.47897338867188, -63.57942199707031, 64.18733215332031, 171.9642333984375, 8.845535278320312, 43.97685241699219, 5.1321868896484375, 50.607696533203125, 127.38845825195312, 29.187908172607422, 115.54276275634766, 137.75765991210938, -27.15625, 105.02462005615234, 78.18421936035156, -44.91815185546875, -11.057605743408203, 14.301937103271484, 73.54106140136719, 127.11970520019531, 99.74635314941406, 98.15953826904297, 170.78851318359375, 64.25584411621094, 67.99916076660156, 77.7185287475586, 40.36972427368164, -11.823183059692383, 20.39484405517578, 43.34816360473633, 79.78693389892578, 27.371049880981445, 128.590087890625, 18.379852294921875, -13.864463806152344, -53.308414459228516, 108.30960083007812, 84.4716796875, 20.651962280273438, -33.494956970214844, -6.7096710205078125, 88.53650665283203, 10.612419128417969, 44.53925323486328, 30.357913970947266, 182.12240600585938, 28.064538955688477, -25.977760314941406, 16.124401092529297, 48.2987060546875, 11.910551071166992, -60.027015686035156, 26.26893424987793, 84.22527313232422, 120.77084350585938, 73.0771713256836, 7.9468231201171875, 79.60507202148438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000422.npy"}
|
|
{"epoch": 0.6379440665154951, "step": 423, "batch_size": 64, "mean": 52.90401840209961, "std": 65.08995819091797, "min": -88.26663208007812, "p10": -17.41249847412109, "median": 34.66228485107422, "p90": 150.5980163574219, "max": 182.31809997558594, "pos_frac": 0.796875, "sample": [35.4169921875, -6.4384613037109375, 116.68232727050781, -34.936683654785156, 161.88417053222656, 5.82221794128418, 46.53207015991211, -43.77387237548828, 143.18060302734375, 33.425628662109375, -12.4049072265625, 128.5455780029297, 5.67938232421875, -19.558609008789062, 6.965522766113281, 9.47979736328125, 8.845474243164062, -5.5624237060546875, 147.51364135742188, -54.48487854003906, 64.23149108886719, 74.31461334228516, -38.51116943359375, -1.6230335235595703, 57.14435958862305, 155.0240020751953, 112.08014678955078, 4.844913482666016, -88.26663208007812, 71.06031799316406, 153.63958740234375, 117.3193130493164, 74.33522033691406, 53.867218017578125, 89.28589630126953, 34.27032470703125, 25.941635131835938, 142.3899383544922, 170.59608459472656, 151.91989135742188, 16.33684539794922, 12.444198608398438, 15.877670288085938, 85.2293701171875, 102.14928436279297, 128.656982421875, 124.4809799194336, 27.89530372619629, -0.48346710205078125, 50.19573974609375, 182.31809997558594, 84.66546630859375, 2.4880313873291016, 35.05424499511719, 102.04901885986328, 156.09503173828125, 11.601837158203125, 4.2704315185546875, -19.87409210205078, 27.319900512695312, 127.50137329101562, 2.808349609375, 20.085758209228516, -9.987089157104492], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000423.npy"}
|
|
{"epoch": 0.6394557823129252, "step": 424, "batch_size": 64, "mean": 54.999149322509766, "std": 66.54378509521484, "min": -84.8169937133789, "p10": -24.453794670104976, "median": 38.46010208129883, "p90": 152.67937164306642, "max": 207.11233520507812, "pos_frac": 0.796875, "sample": [-11.534820556640625, 164.84518432617188, 39.17649841308594, 117.11893463134766, -38.95477294921875, 118.37786102294922, 207.11233520507812, 113.25225067138672, -65.5166015625, -47.29766845703125, 9.892932891845703, 16.502779006958008, 31.99016761779785, -84.8169937133789, 103.96852111816406, 13.059799194335938, 119.78337097167969, -10.612174987792969, 57.90362548828125, 7.744873046875, -20.2887020111084, 79.04972076416016, 118.0685806274414, 37.74370574951172, -26.238834381103516, 123.96487426757812, 166.6431884765625, 16.939498901367188, 163.486572265625, 31.087785720825195, 82.2591323852539, 67.76457977294922, 34.877647399902344, 30.884239196777344, 27.4102783203125, 51.99714660644531, 109.88081359863281, 164.4123992919922, 98.25635528564453, 114.7064437866211, 59.974143981933594, 22.729469299316406, 12.293292999267578, 154.75448608398438, 19.347938537597656, 154.40541076660156, 100.84785461425781, -0.14577293395996094, -50.260780334472656, 6.348562240600586, 138.8688201904297, 57.50010681152344, 114.035888671875, 148.65194702148438, -1.1372451782226562, 21.54247283935547, 43.91520690917969, -56.818115234375, 61.91752624511719, 110.81982421875, 9.938697814941406, 28.17443084716797, -3.859466552734375, 31.199295043945312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000424.npy"}
|
|
{"epoch": 0.6409674981103552, "step": 425, "batch_size": 64, "mean": 41.52496337890625, "std": 65.11768341064453, "min": -97.63389587402344, "p10": -39.0960033416748, "median": 31.581607818603516, "p90": 137.14102935791016, "max": 165.38693237304688, "pos_frac": 0.765625, "sample": [107.71735382080078, 100.06888580322266, 25.98285675048828, 4.9961090087890625, 137.40362548828125, -24.64535903930664, 165.38693237304688, -6.445178985595703, 57.31348419189453, 50.17209243774414, 95.2264404296875, -54.5404052734375, 50.74768829345703, -16.065948486328125, -49.08529281616211, 81.72821044921875, 30.38292694091797, 148.50949096679688, 13.686065673828125, 42.98695755004883, -72.47647094726562, 34.19038391113281, 2.4849205017089844, 46.135093688964844, 83.82254028320312, 0.9253158569335938, 3.4804248809814453, 158.9044952392578, 2.1153030395507812, 150.72268676757812, 136.52830505371094, -29.028461456298828, 79.27845764160156, 9.183181762695312, 43.263458251953125, 32.418434143066406, 24.016525268554688, 147.98138427734375, 24.687000274658203, 33.11983871459961, 133.36529541015625, -81.24871063232422, 43.324981689453125, 76.87716674804688, 30.744781494140625, -97.63389587402344, 34.015228271484375, 50.268856048583984, 21.756423950195312, -10.370651245117188, 136.15174865722656, -10.828609466552734, -40.423667907714844, 124.1229248046875, 13.885673522949219, 5.786663055419922, 112.82781982421875, 28.390411376953125, -10.443492889404297, 156.99163818359375, -35.99811935424805, 22.05718994140625, 127.13777160644531, -46.43949890136719], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000425.npy"}
|
|
{"epoch": 0.6424792139077853, "step": 426, "batch_size": 64, "mean": 25.169984817504883, "std": 74.15911102294922, "min": -179.09312438964844, "p10": -61.527179718017564, "median": 18.775737762451172, "p90": 118.85806808471683, "max": 191.20574951171875, "pos_frac": 0.6875, "sample": [12.216968536376953, -3.340747833251953, 95.15320587158203, 124.48356628417969, 30.464942932128906, 105.27662658691406, 9.154617309570312, 148.9100799560547, 17.68675994873047, 34.74122619628906, -18.93378448486328, -25.753692626953125, 34.09800720214844, -102.4388656616211, 64.40132141113281, 19.864715576171875, 68.5749282836914, 1.2349853515625, 44.368797302246094, -24.73389434814453, -134.85543823242188, -16.053436279296875, 15.461395263671875, 43.096893310546875, -17.38751983642578, 7.8569793701171875, 64.75425720214844, -179.09312438964844, 80.4349365234375, 21.354148864746094, 122.14556884765625, 81.86149597167969, 99.81009674072266, -27.934247970581055, -155.17552185058594, 50.47689437866211, -8.074325561523438, 4.027528762817383, 0.0645904541015625, 191.20574951171875, 134.3397216796875, 166.6042938232422, -94.14741516113281, -87.05850219726562, -0.07956314086914062, 165.08331298828125, -10.024063110351562, 1.572540283203125, 0.5083503723144531, -67.74371337890625, 49.12470245361328, 111.1872329711914, 8.675369262695312, 87.30230712890625, 44.55743408203125, -47.021934509277344, 13.64316177368164, 78.97836303710938, 25.705413818359375, -43.9866943359375, 49.45013427734375, 56.473236083984375, 108.85182189941406, -20.523067474365234], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000426.npy"}
|
|
{"epoch": 0.6439909297052154, "step": 427, "batch_size": 64, "mean": 41.73516845703125, "std": 73.08354949951172, "min": -120.01620483398438, "p10": -52.302783203124996, "median": 29.843780517578125, "p90": 143.4964538574219, "max": 241.35894775390625, "pos_frac": 0.75, "sample": [62.50522232055664, 23.869718551635742, -15.90787124633789, -72.220947265625, -43.22328186035156, 99.58042907714844, 148.53927612304688, 17.778934478759766, 0.1607208251953125, -51.22205352783203, 129.11190795898438, -17.190326690673828, -58.334808349609375, -58.241851806640625, 48.270240783691406, 122.2967529296875, 69.88321685791016, 57.59721374511719, -2.9976577758789062, -19.304031372070312, 0.09522056579589844, 54.89229965209961, 104.86630249023438, 73.24415588378906, 27.68927001953125, 121.62197875976562, 86.07536315917969, 111.0174331665039, 24.103271484375, 28.466758728027344, 154.3358154296875, 8.73577880859375, 75.21405029296875, 31.220802307128906, 21.573097229003906, 136.0924072265625, -4.774711608886719, -120.01620483398438, 131.713623046875, 146.66961669921875, 6.392601013183594, 10.790885925292969, 40.383941650390625, 11.32672119140625, -94.86859130859375, -52.765953063964844, 112.79638671875, 63.296417236328125, 22.217422485351562, -94.36331176757812, 147.75201416015625, 175.85760498046875, 0.1406707763671875, 58.763648986816406, 65.61038970947266, 14.326934814453125, 60.745880126953125, 58.83298873901367, 4.88726806640625, -12.478668212890625, 50.435157775878906, -40.764312744140625, 241.35894775390625, 166.58868408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000427.npy"}
|
|
{"epoch": 0.6455026455026455, "step": 428, "batch_size": 64, "mean": 49.80167007446289, "std": 72.620361328125, "min": -103.56037902832031, "p10": -25.706016159057615, "median": 39.462764739990234, "p90": 147.7404541015625, "max": 262.7655944824219, "pos_frac": 0.75, "sample": [-89.72714233398438, 64.47962188720703, 156.8826446533203, -2.1561813354492188, 2.2391433715820312, 18.714927673339844, 21.84897232055664, -71.178955078125, -56.81584167480469, 39.652427673339844, 154.55340576171875, -20.60974884033203, 96.15091705322266, -26.65113067626953, 126.5823745727539, 49.153533935546875, 134.68177795410156, 20.728179931640625, -12.5618896484375, 34.93646240234375, 129.47369384765625, 3.4816551208496094, 1.2489395141601562, 11.652961730957031, 149.4489288330078, 65.84335327148438, 39.273101806640625, -103.56037902832031, 68.25712585449219, -81.14276123046875, 79.9306640625, -8.104393005371094, 16.885417938232422, 37.882076263427734, 53.29505157470703, 161.27041625976562, 122.86737823486328, 120.53265380859375, 59.057395935058594, -20.059118270874023, 143.75401306152344, 262.7655944824219, 69.951171875, 126.22843170166016, -26.232406616210938, -4.15985107421875, 82.02665710449219, 52.215003967285156, 3.6531543731689453, -17.02764892578125, 11.133621215820312, 16.706642150878906, -0.9575958251953125, 2.5772056579589844, 142.5254364013672, 149.92625427246094, 157.094970703125, 80.7892074584961, 108.98231506347656, 117.56726837158203, 86.9971923828125, 95.73161315917969, 1.0987167358398438, -24.477771759033203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000428.npy"}
|
|
{"epoch": 0.6470143613000756, "step": 429, "batch_size": 64, "mean": 38.78816223144531, "std": 67.32217407226562, "min": -121.80945587158203, "p10": -50.22602195739746, "median": 34.29494857788086, "p90": 132.0724609375, "max": 159.6457977294922, "pos_frac": 0.75, "sample": [70.06138610839844, -69.51983642578125, 1.7949638366699219, 6.949653625488281, 122.81385040283203, 77.75530242919922, 43.14860153198242, 26.77745819091797, -13.451290130615234, 21.916824340820312, 27.594512939453125, -77.6793212890625, 39.59523010253906, 33.19293212890625, 3.3601455688476562, 92.18527221679688, -121.80945587158203, 20.67327117919922, -0.7331695556640625, 89.68534088134766, 101.58370971679688, -10.831233978271484, 86.7133560180664, 132.7842559814453, 142.35617065429688, 146.18692016601562, 147.39810180664062, -8.31695556640625, 35.39696502685547, -47.51808547973633, -51.386566162109375, -42.713314056396484, 7.605445861816406, -12.642601013183594, 101.71112823486328, 7.3544769287109375, -58.280548095703125, 72.71112823486328, -0.16930389404296875, 2.1836624145507812, 130.41160583496094, 118.10669708251953, 36.299522399902344, 129.66192626953125, 116.94871520996094, -119.46842193603516, -66.33085632324219, 24.275283813476562, 6.017219543457031, 91.70276641845703, 16.122535705566406, 48.41997528076172, 16.59471893310547, -21.520519256591797, 40.70009994506836, 56.77781677246094, 159.6457977294922, 53.10411834716797, 157.6077423095703, 133.82769775390625, 6.427093505859375, 88.89479064941406, 37.61415100097656, 74.16350555419922], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000429.npy"}
|
|
{"epoch": 0.6485260770975056, "step": 430, "batch_size": 64, "mean": 57.286834716796875, "std": 67.5430908203125, "min": -163.18588256835938, "p10": -7.147391510009762, "median": 50.49277114868164, "p90": 147.54747009277344, "max": 173.17449951171875, "pos_frac": 0.8125, "sample": [36.96253967285156, 25.210264205932617, -8.594375610351562, 24.90076446533203, -3.7710952758789062, 139.78639221191406, -0.1806182861328125, 144.28244018554688, 23.9232234954834, 65.55524444580078, -67.28449249267578, 91.93277740478516, 155.30340576171875, 29.816373825073242, 148.02944946289062, 56.52162170410156, 91.47290802001953, 20.708038330078125, 163.00364685058594, -12.310049057006836, 44.5704345703125, 173.17449951171875, 97.05011749267578, 71.65354919433594, -0.29531097412109375, 149.76356506347656, 25.433914184570312, 146.4228515625, 79.73854064941406, 126.80319213867188, 71.65536499023438, 6.60595703125, 76.75233459472656, 113.37673950195312, -38.7866096496582, 130.67799377441406, 159.2529296875, 95.32825469970703, -70.09861755371094, 119.59021759033203, 11.966827392578125, 56.41510772705078, 18.93313980102539, 43.10169219970703, 33.09864807128906, 103.19050598144531, 14.278343200683594, 1.5036067962646484, -163.18588256835938, 4.317558288574219, -1.98724365234375, 117.88636016845703, -1.7555007934570312, 119.44078063964844, 27.327301025390625, 156.4393310546875, 140.2870635986328, 89.77536010742188, 110.73014068603516, 19.900314331054688, 66.74107360839844, -49.684207916259766, 21.561004638671875, 22.137725830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000430.npy"}
|
|
{"epoch": 0.6500377928949358, "step": 431, "batch_size": 64, "mean": 42.55097961425781, "std": 64.47652435302734, "min": -127.39696502685547, "p10": -22.24824066162109, "median": 30.63453197479248, "p90": 130.2634246826172, "max": 181.29330444335938, "pos_frac": 0.796875, "sample": [-127.39696502685547, 91.18465423583984, 54.82661819458008, 30.25676727294922, 6.890911102294922, 123.3162612915039, 92.611572265625, 106.40947723388672, 107.78007507324219, -5.471811294555664, 3.8802452087402344, 152.51316833496094, 106.404052734375, -12.372245788574219, 38.28166198730469, 20.030996322631836, 181.29330444335938, -7.592071533203125, 126.64279174804688, 23.626190185546875, 41.472801208496094, -108.96609497070312, 10.530403137207031, 86.2769775390625, -38.68150329589844, 145.3164520263672, 49.8394775390625, -49.568328857421875, 64.22975158691406, -13.39204216003418, 31.012296676635742, 8.726470947265625, 110.39606475830078, 130.49246215820312, -65.91287231445312, 17.49732208251953, 140.79806518554688, 71.34422302246094, -24.80096435546875, -29.816787719726562, -16.291885375976562, 1.0267333984375, 0.9491348266601562, 106.62690734863281, 5.926170349121094, 15.257537841796875, 107.24017333984375, 44.978424072265625, 27.129283905029297, 176.7293701171875, 15.7674560546875, 43.1822624206543, 2.785736083984375, 24.056415557861328, 74.71012115478516, 9.98361587524414, 36.41943359375, 0.19603729248046875, 15.336933135986328, 129.72900390625, 133.7237548828125, 54.756980895996094, 32.62980651855469, -9.496551513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000431.npy"}
|
|
{"epoch": 0.6515495086923658, "step": 432, "batch_size": 64, "mean": 44.27925109863281, "std": 67.7125015258789, "min": -140.63983154296875, "p10": -27.69722557067871, "median": 32.784257888793945, "p90": 142.57356719970704, "max": 161.67111206054688, "pos_frac": 0.75, "sample": [-7.301605224609375, -28.287569046020508, -31.349273681640625, -140.63983154296875, 2.6683483123779297, 1.1426410675048828, 7.8870391845703125, 62.700599670410156, -21.29669952392578, 143.13516235351562, 59.54169845581055, -87.85808563232422, 16.1397705078125, 51.54486083984375, 141.62049865722656, -94.16156005859375, 68.05196380615234, -3.482208251953125, 161.0967559814453, 87.86495971679688, 5.492378234863281, 92.77344512939453, 118.83287811279297, -8.430084228515625, -6.322353363037109, 142.98202514648438, 2.9886550903320312, 136.07791137695312, 103.77925109863281, -19.43678855895996, -39.74314880371094, 28.151042938232422, 8.823614120483398, -31.448257446289062, 29.6654052734375, 43.75554656982422, 146.85971069335938, 85.01847076416016, 42.39115905761719, 116.46949005126953, -8.628910064697266, 35.90311050415039, 98.64044189453125, 14.499984741210938, 116.97869873046875, 0.8444366455078125, 19.42888069152832, 148.28701782226562, 17.368465423583984, 157.03065490722656, 84.0107650756836, 72.99143981933594, 85.42649841308594, 0.199798583984375, 63.66558837890625, 161.67111206054688, -17.55791473388672, 115.03370666503906, 68.29574584960938, 5.100776672363281, -26.31975746154785, 125.62338256835938, 4.4915771484375, 103.18878173828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000432.npy"}
|
|
{"epoch": 0.6530612244897959, "step": 433, "batch_size": 64, "mean": 54.992530822753906, "std": 67.26995849609375, "min": -135.9022216796875, "p10": -25.69538097381592, "median": 44.68544387817383, "p90": 145.9962600708008, "max": 167.73912048339844, "pos_frac": 0.796875, "sample": [41.1502571105957, 93.0201644897461, 12.76949691772461, 44.27456283569336, 13.181343078613281, 36.88104248046875, -1.0372333526611328, 99.85296630859375, -21.64004898071289, 16.160736083984375, 13.179473876953125, 43.43462371826172, 82.10009765625, 19.584877014160156, 151.55465698242188, -72.8382339477539, 52.89280700683594, 114.44245910644531, 98.5622329711914, -38.906009674072266, 55.892059326171875, 37.582305908203125, 5.566154479980469, 128.23617553710938, -135.9022216796875, 45.0963249206543, 117.41426086425781, 12.55316162109375, 107.74835205078125, 107.56877136230469, 23.51036834716797, 100.23506927490234, -25.62175178527832, 3.1618118286132812, 47.681209564208984, -10.729629516601562, -29.69510269165039, 68.59483337402344, 66.968505859375, 153.4113311767578, 167.73912048339844, 148.97879028320312, 161.76983642578125, 38.660484313964844, 127.9365234375, -81.93077850341797, 32.04939270019531, 80.666259765625, -19.00061798095703, 141.07101440429688, -18.551559448242188, 71.28531646728516, 107.50300598144531, -30.2347412109375, 32.64331817626953, 142.1132049560547, 109.97988891601562, 40.19818115234375, 147.27880859375, 142.999755859375, 143.00364685058594, 33.30079650878906, 147.89678955078125, -25.72693634033203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000433.npy"}
|
|
{"epoch": 0.654572940287226, "step": 434, "batch_size": 64, "mean": 54.47591781616211, "std": 75.58665466308594, "min": -153.32553100585938, "p10": -31.818343734741205, "median": 65.51350402832031, "p90": 146.55767059326172, "max": 177.62452697753906, "pos_frac": 0.8125, "sample": [-130.75497436523438, 61.7666015625, 96.14427185058594, 2.325927734375, 108.73640441894531, 137.9044952392578, -33.42905044555664, 143.95455932617188, -54.4298095703125, 33.179412841796875, -0.40773582458496094, -47.858428955078125, -153.32553100585938, 61.2342529296875, 102.1631851196289, -28.060028076171875, 93.52360534667969, 71.8465576171875, 168.98443603515625, 156.20262145996094, 69.23739624023438, 2.9196834564208984, 177.62452697753906, 6.101737976074219, -89.43301391601562, 37.52485656738281, 147.47972106933594, 82.82856750488281, 3.0181312561035156, 87.20978546142578, 16.218746185302734, 111.81863403320312, 79.48085021972656, 149.49490356445312, 162.78982543945312, 136.52392578125, -25.156593322753906, 103.4852066040039, 144.40621948242188, 2.082286834716797, 131.09304809570312, 5.192840576171875, 30.875988006591797, 88.82649230957031, 0.7855606079101562, 22.21038055419922, 21.03228187561035, 92.99227142333984, 61.78961181640625, 99.61857604980469, 5.911308288574219, 151.585693359375, -117.77175903320312, -14.588912963867188, 2.7665634155273438, 74.74398040771484, 119.42189025878906, 136.82562255859375, 93.06475830078125, -11.82177734375, 47.84421920776367, 73.14234161376953, 128.6455078125, 46.9161376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000434.npy"}
|
|
{"epoch": 0.656084656084656, "step": 435, "batch_size": 64, "mean": 46.55925750732422, "std": 68.19964599609375, "min": -97.91596221923828, "p10": -40.7843141555786, "median": 47.923160552978516, "p90": 138.863232421875, "max": 164.794921875, "pos_frac": 0.71875, "sample": [87.37283325195312, 56.14052963256836, -4.564720153808594, 47.96049499511719, -66.70442962646484, 7.983802795410156, 12.657852172851562, 0.7187767028808594, 89.85352325439453, -6.3860321044921875, 111.40328979492188, 139.95928955078125, -22.301055908203125, 93.74907684326172, 139.00881958007812, 137.51651000976562, 17.81902313232422, -46.87336730957031, -12.624031066894531, 138.27703857421875, 65.37261199951172, 4.436515808105469, 125.9518814086914, 162.67347717285156, 16.479705810546875, 40.80145263671875, -12.872833251953125, 48.78718566894531, 145.14588928222656, 137.3533935546875, -7.129859924316406, 70.64700317382812, -97.91596221923828, 89.50926208496094, 51.701725006103516, 68.09687805175781, -4.571434020996094, -24.61053466796875, 85.76874542236328, 138.52352905273438, 134.99468994140625, 164.794921875, 151.45266723632812, 57.619667053222656, 0.8289337158203125, 51.495948791503906, 28.476932525634766, -29.2144832611084, -22.29928207397461, 152.33584594726562, -45.74281311035156, 21.748382568359375, -46.17369842529297, 62.971824645996094, 47.885826110839844, 72.33660888671875, 127.39762878417969, -62.38922119140625, -92.03424835205078, 25.431808471679688, 0.66571044921875, -0.488677978515625, 122.74773406982422, 29.833984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000435.npy"}
|
|
{"epoch": 0.6575963718820862, "step": 436, "batch_size": 64, "mean": 41.385353088378906, "std": 70.56678771972656, "min": -116.562744140625, "p10": -49.9293472290039, "median": 32.11033630371094, "p90": 140.19811096191407, "max": 161.15249633789062, "pos_frac": 0.6875, "sample": [97.09651184082031, -6.421230316162109, 134.18222045898438, 2.1516361236572266, 89.04839324951172, 5.974153518676758, 50.54009246826172, 123.24989318847656, -50.71485137939453, 21.99303436279297, 35.56217956542969, 141.63279724121094, 9.481285095214844, 75.49819946289062, -13.252140045166016, 151.9178466796875, 125.760009765625, 95.25361633300781, -98.75177001953125, 68.41246795654297, 21.857250213623047, 61.820465087890625, -21.59703826904297, -2.4531936645507812, 33.675987243652344, -2.8608779907226562, 124.38330078125, -6.989885330200195, 136.8505096435547, 20.649208068847656, 60.37934494018555, -16.763916015625, -116.562744140625, 69.74992370605469, -15.516252517700195, 30.54468536376953, 124.43731689453125, -48.09650421142578, -4.85614013671875, -59.326690673828125, 112.13911437988281, 103.82376861572266, 33.73842239379883, 8.554489135742188, -85.61990356445312, 83.40141296386719, -35.45226287841797, 156.62899780273438, 72.04611206054688, 161.15249633789062, 98.9404067993164, 1.4185943603515625, 152.45538330078125, 87.94718933105469, -17.210411071777344, 16.551403045654297, 143.3331298828125, -29.41461944580078, 16.488746643066406, 0.1917400360107422, -51.444610595703125, 159.1083526611328, -64.88550567626953, 76.83110809326172], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000436.npy"}
|
|
{"epoch": 0.6591080876795162, "step": 437, "batch_size": 64, "mean": 42.4582633972168, "std": 78.9501724243164, "min": -189.7606201171875, "p10": -53.883251953125, "median": 40.06365203857422, "p90": 150.7042251586914, "max": 204.16061401367188, "pos_frac": 0.671875, "sample": [204.16061401367188, 155.44935607910156, 134.0106201171875, 70.84149169921875, 34.08338928222656, -42.72246170043945, 114.53184509277344, 82.037353515625, 178.79794311523438, 31.910785675048828, 142.64747619628906, -6.736541748046875, -189.7606201171875, 74.98933410644531, -70.48974609375, -52.882545471191406, 24.309219360351562, 67.44871520996094, 99.21755981445312, -83.61767578125, 164.14930725097656, 152.60360717773438, 136.6591033935547, -16.532176971435547, -34.55705642700195, 68.82044982910156, 149.9451904296875, -17.105571746826172, -93.38131713867188, -54.31212615966797, -14.404983520507812, -36.262298583984375, 36.51567840576172, -10.736412048339844, -0.10547637939453125, 151.02952575683594, 54.898681640625, -22.899612426757812, 62.23149871826172, 65.46646118164062, 22.727279663085938, -4.103404998779297, -29.106632232666016, 122.49516296386719, 13.396224975585938, 129.0902099609375, 43.61162567138672, 72.95880126953125, 79.63511657714844, -81.4482192993164, 44.00934600830078, 70.74664306640625, 119.1784439086914, 116.68914794921875, 18.640724182128906, 17.194007873535156, 26.361099243164062, 75.16676330566406, -71.205810546875, -1.3963050842285156, 0.31487083435058594, 53.91487121582031, 1.83111572265625, 166.37911987304688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000437.npy"}
|
|
{"epoch": 0.6606198034769464, "step": 438, "batch_size": 64, "mean": 59.02305603027344, "std": 73.29878234863281, "min": -88.9160385131836, "p10": -13.599872970581053, "median": 40.61395835876465, "p90": 142.6837875366211, "max": 293.03558349609375, "pos_frac": 0.765625, "sample": [172.46644592285156, -12.891571044921875, 293.03558349609375, -53.140403747558594, -0.3556632995605469, -39.202613830566406, -13.903430938720703, 120.07247924804688, 4.754350662231445, 155.7249298095703, -9.655956268310547, 169.781494140625, -19.435020446777344, 7.035392761230469, 136.2020263671875, 0.4247322082519531, 103.09647369384766, -88.9160385131836, 118.83576965332031, 77.98966979980469, 121.78302764892578, 121.34165954589844, -9.771034240722656, 37.53338623046875, 45.65082931518555, 128.92901611328125, 45.61810302734375, 25.034313201904297, 145.27784729003906, 112.58770751953125, 81.90087890625, -18.10552978515625, 124.7490463256836, 9.645261764526367, 0.83624267578125, 129.62213134765625, 125.04087829589844, -9.317380905151367, 82.35877990722656, 42.45188903808594, 127.21176147460938, 1.4625701904296875, 38.77602767944336, 203.64071655273438, 124.90667724609375, 23.52496337890625, 104.4013442993164, 26.232955932617188, 13.910453796386719, 2.953929901123047, -12.146800994873047, -6.781023025512695, 19.226346969604492, 117.38607788085938, 117.79650115966797, -62.57495880126953, 136.6309814453125, 20.840545654296875, -6.583473205566406, 149.8380584716797, 12.953632354736328, 24.44310188293457, 82.50820922851562, 51.83128356933594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000438.npy"}
|
|
{"epoch": 0.6621315192743764, "step": 439, "batch_size": 64, "mean": 50.01641082763672, "std": 66.44976806640625, "min": -159.8109130859375, "p10": -16.589583587646484, "median": 31.73589324951172, "p90": 154.36370391845702, "max": 176.9484405517578, "pos_frac": 0.796875, "sample": [93.63298797607422, -0.5094261169433594, 15.711446762084961, -32.235313415527344, 38.28546905517578, 121.30693054199219, 64.65099334716797, 30.870513916015625, -18.987831115722656, 160.9390106201172, 176.9484405517578, 1.9133281707763672, 146.68943786621094, 19.822280883789062, 31.303359985351562, 159.38946533203125, 124.38076782226562, 7.276237487792969, 127.9332275390625, -17.146469116210938, 6.462139129638672, 83.91243743896484, 44.36870574951172, -4.8904571533203125, 135.286865234375, 28.02068328857422, 154.43704223632812, -42.80131912231445, 4.003007888793945, 160.48660278320312, -17.71314811706543, 154.1925811767578, -22.20081329345703, 3.89764404296875, 64.4731674194336, 14.265396118164062, 6.534027099609375, 12.437309265136719, -159.8109130859375, 16.448421478271484, 1.54449462890625, 31.561203002929688, 126.33195495605469, 45.3056640625, 53.021358489990234, 136.6068878173828, 70.21633911132812, 171.49156188964844, 31.91058349609375, 39.85406494140625, 16.143451690673828, 59.127227783203125, 6.606254577636719, 93.28131866455078, 64.83937072753906, 173.6561737060547, 83.3734130859375, -5.56939697265625, -15.290184020996094, -5.5998382568359375, 18.064495086669922, 66.17176818847656, 49.265586853027344, -4.847724914550781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000439.npy"}
|
|
{"epoch": 0.6636432350718064, "step": 440, "batch_size": 64, "mean": 51.436134338378906, "std": 71.94380187988281, "min": -106.67120361328125, "p10": -40.68874549865723, "median": 36.188602447509766, "p90": 153.43307495117188, "max": 185.5030517578125, "pos_frac": 0.765625, "sample": [-19.992374420166016, 33.073333740234375, 60.591033935546875, 180.8140869140625, 153.82180786132812, 185.5030517578125, 158.4864501953125, -8.674934387207031, 143.64404296875, 22.305999755859375, 25.386314392089844, 144.46502685546875, 39.303871154785156, 117.36408996582031, 82.59295654296875, 29.19708251953125, 133.27066040039062, -41.367408752441406, -73.2146224975586, 41.54301452636719, 65.5420913696289, 1.0567150115966797, 8.128089904785156, 101.26691436767578, 42.49861145019531, 101.91875457763672, 70.64451599121094, 84.654541015625, 9.505300521850586, 18.006858825683594, 11.121381759643555, 66.05345916748047, 100.04313659667969, -0.6077766418457031, 3.2788772583007812, 32.70660400390625, 28.53070640563965, -86.8906478881836, 28.93340492248535, 156.2037353515625, 111.62982177734375, 149.9769287109375, -6.55181884765625, 146.25588989257812, -59.06438064575195, 17.89850616455078, 157.0204315185547, 27.121360778808594, 119.59515380859375, -18.648033142089844, -39.10519790649414, -29.40559959411621, 59.691184997558594, -51.705474853515625, 59.402687072753906, 27.694644927978516, -20.722469329833984, -50.369529724121094, 126.05763244628906, -106.67120361328125, 32.1451416015625, 157.27740478515625, 152.52603149414062, 79.1545639038086], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000440.npy"}
|
|
{"epoch": 0.6651549508692366, "step": 441, "batch_size": 64, "mean": 56.87614440917969, "std": 58.87123489379883, "min": -54.526695251464844, "p10": -12.790094184875487, "median": 51.645751953125, "p90": 143.35904388427736, "max": 158.3564453125, "pos_frac": 0.828125, "sample": [100.05355834960938, 50.543060302734375, 146.78822326660156, 143.92373657226562, -3.114461898803711, 77.12371826171875, 131.570068359375, -5.779731750488281, 91.77987670898438, -10.044189453125, -24.32223129272461, 129.45217895507812, -13.377311706542969, 138.81631469726562, 17.634540557861328, 9.697021484375, 58.69280242919922, 8.647186279296875, -32.429935455322266, 148.49142456054688, 42.25938415527344, 54.897727966308594, 101.86553955078125, 138.26144409179688, 19.059913635253906, 151.01702880859375, 8.19723129272461, 12.171745300292969, 157.890625, 57.16082763671875, 39.370765686035156, 52.54347229003906, -11.419919967651367, 99.13359069824219, 104.87858581542969, 4.335258483886719, -44.70021057128906, 142.0414276123047, 53.792457580566406, 46.232147216796875, 36.254058837890625, 147.33255004882812, 12.735343933105469, 21.406841278076172, 93.77671813964844, 48.547271728515625, 60.8914909362793, 10.668708801269531, 82.81039428710938, -14.358444213867188, 131.2344970703125, 16.061426162719727, 139.7764892578125, 158.3564453125, -38.51202392578125, 51.519805908203125, 51.771697998046875, 10.637214660644531, 4.388648986816406, 80.27094268798828, -54.526695251464844, 53.335792541503906, 109.07855224609375, 33.48069763183594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000441.npy"}
|
|
{"epoch": 0.6666666666666666, "step": 442, "batch_size": 64, "mean": 54.09173583984375, "std": 67.61331939697266, "min": -112.70775604248047, "p10": -29.43664855957031, "median": 49.799842834472656, "p90": 141.32360229492187, "max": 193.28433227539062, "pos_frac": 0.78125, "sample": [26.908401489257812, 165.47836303710938, 62.55171203613281, -26.53216552734375, 47.223236083984375, 193.28433227539062, 140.392333984375, 151.98959350585938, 74.33551025390625, 129.61737060546875, 58.3934211730957, 110.0052261352539, 124.07986450195312, -33.45989990234375, 98.41345977783203, 81.87789154052734, -0.0414581298828125, 10.4810791015625, 27.979400634765625, -26.206356048583984, -57.76954650878906, 123.08514404296875, 76.95639038085938, -2.5174331665039062, 121.37245178222656, 5.031867980957031, -51.25865936279297, 87.85662841796875, 16.180646896362305, 137.8599090576172, 3.0479736328125, 44.684364318847656, 103.30525970458984, 134.56295776367188, 157.82089233398438, 64.83602905273438, -13.080207824707031, 3.5218238830566406, 149.34864807128906, 74.35806274414062, 68.94420623779297, 4.431739807128906, 11.93720817565918, 27.821895599365234, 0.8430881500244141, -112.70775604248047, -47.5244140625, 151.63677978515625, -30.681427001953125, -16.921157836914062, 137.825927734375, 108.8764419555664, -51.239044189453125, 41.73210144042969, 76.72647094726562, -20.169118881225586, 78.57878875732422, 22.07965087890625, 51.975982666015625, 128.82977294921875, 47.62370300292969, 141.72271728515625, 16.920921325683594, 26.631933212280273], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000442.npy"}
|
|
{"epoch": 0.6681783824640968, "step": 443, "batch_size": 64, "mean": 66.78971862792969, "std": 75.92330169677734, "min": -139.53460693359375, "p10": -11.76922607421875, "median": 69.03491973876953, "p90": 149.39045104980468, "max": 270.4560241699219, "pos_frac": 0.796875, "sample": [7.725971221923828, 127.33259582519531, 69.14451599121094, 135.46337890625, 26.64440155029297, 56.4737548828125, 93.32493591308594, -73.40451049804688, 40.40381622314453, 88.99928283691406, -11.013160705566406, -14.690729141235352, 31.289012908935547, 133.87435913085938, 51.35616683959961, 165.86215209960938, 13.095804214477539, 98.52751922607422, 72.99988555908203, 128.90005493164062, 270.4560241699219, -139.53460693359375, -21.489707946777344, 129.40988159179688, -1.6746234893798828, 183.497802734375, 108.2496566772461, 59.07147979736328, 147.10693359375, 132.82687377929688, 112.90827941894531, 97.98733520507812, 100.12186431884766, 124.74629974365234, 69.19186401367188, 147.40997314453125, 137.5718231201172, 151.45358276367188, 109.55339050292969, 145.0301971435547, 7.171596527099609, 125.62355041503906, -9.422494888305664, 63.400596618652344, 150.23922729492188, 139.19105529785156, -12.093254089355469, 18.30938720703125, 14.170417785644531, 7.439516067504883, 3.3256797790527344, -41.40534210205078, 68.92532348632812, 1.3355255126953125, 27.72384262084961, 52.00416946411133, -111.84361267089844, -9.542648315429688, 162.9584197998047, 28.092164993286133, -0.7624893188476562, -2.6867599487304688, 91.0006103515625, 195.1840057373047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000443.npy"}
|
|
{"epoch": 0.6696900982615268, "step": 444, "batch_size": 64, "mean": 25.20024299621582, "std": 71.15889739990234, "min": -144.65078735351562, "p10": -65.85600090026855, "median": 15.66521167755127, "p90": 127.39149856567383, "max": 179.08914184570312, "pos_frac": 0.640625, "sample": [96.47917938232422, -1.4654350280761719, -27.284278869628906, 3.8531665802001953, 21.662750244140625, 36.95853805541992, 66.4423828125, -26.436477661132812, -1.0959205627441406, 101.4629135131836, 2.6787338256835938, 76.0014877319336, -80.11471557617188, -94.13223266601562, 34.26002502441406, 18.091094970703125, 64.33488464355469, 38.52201843261719, 35.587833404541016, -89.43453979492188, 141.79315185546875, 119.66570281982422, 0.9874763488769531, -20.845733642578125, -13.214019775390625, 28.51563262939453, 4.880271911621094, -7.536405563354492, -43.560791015625, -54.9234619140625, 113.10232543945312, 11.356868743896484, 127.44316864013672, 45.4971923828125, 13.357757568359375, 110.64855194091797, 20.98340606689453, 128.90798950195312, -10.821678161621094, -144.65078735351562, 19.756683349609375, 179.08914184570312, -97.97991943359375, -45.112953186035156, -5.802274703979492, 138.92311096191406, 107.8883056640625, 127.27093505859375, 147.10958862304688, 56.01947021484375, 20.883737564086914, 63.434722900390625, 7.336498260498047, 9.761756896972656, 121.8694839477539, -20.122194290161133, -63.75198745727539, 17.972665786743164, 7.63916015625, -66.75772094726562, -103.74515533447266, 147.51307678222656, -2.8184871673583984, -1.5201282501220703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000444.npy"}
|
|
{"epoch": 0.671201814058957, "step": 445, "batch_size": 64, "mean": 55.75970458984375, "std": 70.0136947631836, "min": -101.9571762084961, "p10": -27.423861694335933, "median": 52.95687675476074, "p90": 148.2859619140625, "max": 177.54978942871094, "pos_frac": 0.828125, "sample": [39.83376693725586, 177.54978942871094, 10.326103210449219, 0.8249664306640625, -23.78045654296875, 20.837142944335938, 142.78240966796875, 106.90440368652344, 42.76237487792969, 49.87868118286133, 3.985107421875, -61.527565002441406, -28.985321044921875, 16.388565063476562, -86.68975830078125, 134.03530883789062, -50.33103942871094, 5.875831604003906, 22.587554931640625, 37.91448211669922, 2.2491378784179688, 138.32080078125, -101.9571762084961, 26.2364501953125, -60.93505096435547, 141.0612030029297, 18.11902618408203, 143.231689453125, 32.08574676513672, 5.4042816162109375, -46.827266693115234, -12.392129898071289, 56.38011169433594, 159.3323516845703, 148.8992919921875, 56.28517150878906, 134.16383361816406, 149.13589477539062, 47.933555603027344, -4.956169128417969, 146.8548583984375, 65.2097396850586, 67.49208068847656, 166.47084045410156, 133.60877990722656, 141.05113220214844, 159.8033447265625, 10.760149002075195, 91.98533630371094, -23.672414779663086, 73.29975891113281, 1.5652847290039062, 5.38531494140625, 56.47270965576172, 67.31188201904297, 69.72944641113281, 141.61587524414062, 122.38272857666016, 56.035072326660156, 63.475257873535156, 105.6310806274414, 85.87188720703125, 153.2169952392578, 14.120903015136719], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000445.npy"}
|
|
{"epoch": 0.672713529856387, "step": 446, "batch_size": 64, "mean": 62.465049743652344, "std": 64.22469329833984, "min": -58.461326599121094, "p10": -10.027786254882812, "median": 49.86690902709961, "p90": 154.11269683837892, "max": 211.71160888671875, "pos_frac": 0.828125, "sample": [4.781465530395508, 6.627952575683594, -56.83158493041992, 113.79571533203125, 39.951622009277344, 76.62294006347656, 34.89542007446289, -11.941261291503906, 211.71160888671875, 58.071746826171875, 22.45218276977539, 112.68788146972656, 93.86439514160156, 117.9228515625, 88.6037368774414, 152.33470153808594, 33.428688049316406, 72.8434066772461, 77.27581024169922, 37.063499450683594, 4.122913360595703, 98.71263122558594, -10.301300048828125, 126.15035247802734, 157.98223876953125, 50.405052185058594, 6.824237823486328, 123.86780548095703, 152.51467895507812, -17.93810272216797, 119.36324310302734, 0.8745880126953125, -58.461326599121094, -3.9186935424804688, -5.6164093017578125, 1.040170669555664, 4.882057189941406, -21.18459701538086, 59.895652770996094, 108.27086639404297, 45.348419189453125, 167.9639434814453, 155.89573669433594, 127.4358901977539, 190.446044921875, 0.15856361389160156, 156.44503784179688, 136.0978240966797, -15.134553909301758, 42.605224609375, 60.49767303466797, 32.27037048339844, 49.328765869140625, 110.73902893066406, 143.51467895507812, 154.7975616455078, 14.401058197021484, 30.97765350341797, 7.6731109619140625, 76.88308715820312, -3.762664794921875, 47.34440231323242, -9.38958740234375, 91.5769271850586], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000446.npy"}
|
|
{"epoch": 0.674225245653817, "step": 447, "batch_size": 64, "mean": 67.43780517578125, "std": 76.2525863647461, "min": -143.14175415039062, "p10": -34.207490539550776, "median": 78.99031066894531, "p90": 157.50799865722655, "max": 174.41506958007812, "pos_frac": 0.796875, "sample": [174.41506958007812, 153.85194396972656, 117.35954284667969, 17.54401397705078, 20.728836059570312, -12.171348571777344, 168.99349975585938, -6.5532379150390625, -102.0932846069336, 31.938796997070312, 65.75090789794922, 111.7261734008789, 8.063026428222656, 72.28390502929688, 146.98947143554688, 158.53025817871094, 54.80406951904297, 80.85440063476562, 39.40806198120117, 49.842838287353516, 26.1953125, -26.83324432373047, 42.847564697265625, -37.367881774902344, 24.112037658691406, 164.95046997070312, 157.5972900390625, 133.4581756591797, 110.17291259765625, -40.005584716796875, 82.68505859375, 97.67359924316406, 137.24964904785156, 135.1641845703125, 80.22970581054688, 134.73098754882812, 169.9993896484375, 99.26296997070312, 157.29965209960938, 6.873401641845703, 36.95631408691406, 59.329345703125, 145.81033325195312, -65.7661361694336, -14.848491668701172, -105.037109375, 26.45519256591797, 143.9730224609375, 113.54631042480469, 89.54011535644531, -11.212295532226562, 140.26019287109375, -55.81877517700195, 148.7857666015625, 94.76162719726562, 46.853973388671875, 85.17201232910156, -143.14175415039062, 77.75091552734375, 161.8124542236328, 154.78524780273438, -4.80560302734375, 72.562744140625, 109.73153686523438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000447.npy"}
|
|
{"epoch": 0.6757369614512472, "step": 448, "batch_size": 64, "mean": 53.45562744140625, "std": 69.70098876953125, "min": -146.12010192871094, "p10": -24.74644680023192, "median": 48.27450942993164, "p90": 143.56926727294922, "max": 196.76382446289062, "pos_frac": 0.8125, "sample": [117.13375091552734, -97.49839782714844, 143.8762664794922, 52.430152893066406, 87.47830963134766, 84.34766387939453, -3.9180221557617188, 105.40879821777344, 64.21470642089844, -2.2162094116210938, 105.33070373535156, 96.11083221435547, 30.597557067871094, 0.68548583984375, 105.5770034790039, 152.79873657226562, -7.686714172363281, 163.4066162109375, 31.973106384277344, 11.89309310913086, 5.0769500732421875, 142.85293579101562, 45.903480529785156, 66.10273742675781, 97.9909439086914, 2.108379364013672, 142.34963989257812, 44.12651062011719, 17.55207061767578, -30.879714965820312, -50.724365234375, 17.597976684570312, 58.207481384277344, 1.5406646728515625, 11.864494323730469, 78.99678039550781, -31.828460693359375, 31.066102981567383, 174.60247802734375, -10.435487747192383, 10.080343246459961, -146.12010192871094, 81.04985046386719, 30.159133911132812, 69.17520141601562, 114.08927917480469, 79.1551513671875, 128.98153686523438, 55.368614196777344, 25.05106544494629, -32.99797821044922, 41.07615661621094, 50.645538330078125, 122.13603210449219, 0.6716537475585938, 196.76382446289062, -3.953460693359375, -85.41531372070312, 141.70770263671875, 145.76783752441406, 138.79217529296875, 27.844146728515625, 10.448936462402344, 164.66781616210938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000448.npy"}
|
|
{"epoch": 0.6772486772486772, "step": 449, "batch_size": 64, "mean": 27.46868133544922, "std": 62.488956451416016, "min": -96.39117431640625, "p10": -53.051588439941405, "median": 17.65884780883789, "p90": 119.99193038940432, "max": 169.227783203125, "pos_frac": 0.65625, "sample": [1.9110965728759766, 35.45709228515625, -90.74639892578125, 50.2435302734375, 8.817527770996094, -6.5108795166015625, -10.995498657226562, 53.09601593017578, 28.100318908691406, 155.08856201171875, 88.61919403076172, 51.653053283691406, -1.0233993530273438, -60.98396301269531, -7.8482513427734375, 4.03326416015625, 169.227783203125, 5.249843597412109, 4.6541900634765625, 52.98381423950195, -25.453746795654297, 53.964454650878906, 153.10775756835938, 130.64089965820312, -96.39117431640625, 9.330631256103516, 88.10556030273438, -51.629638671875, 95.8520736694336, 27.959304809570312, 41.81892395019531, -18.62841033935547, 57.96026611328125, -56.5131721496582, 121.86383819580078, 13.491607666015625, 134.5274200439453, -2.4486045837402344, 27.661128997802734, 35.36083984375, 7.412832260131836, -5.6755828857421875, -47.83956527709961, 28.972518920898438, -53.66099548339844, 70.40261840820312, -7.5280914306640625, 129.44244384765625, 10.04745864868164, 57.86237335205078, 112.37077331542969, -3.3019866943359375, 21.826087951660156, 26.28363037109375, -50.39042663574219, 100.80557250976562, 87.35559844970703, -66.87747955322266, 115.6241455078125, -30.54694366455078, 62.97001647949219, -80.35769653320312, 9.862186431884766, -8.670814514160156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000449.npy"}
|
|
{"epoch": 0.6787603930461074, "step": 450, "batch_size": 64, "mean": 40.254066467285156, "std": 73.12519836425781, "min": -91.73638916015625, "p10": -41.93066062927246, "median": 22.545151710510254, "p90": 149.64212646484376, "max": 197.22970581054688, "pos_frac": 0.671875, "sample": [1.0099315643310547, -29.939773559570312, -38.568199157714844, -31.057777404785156, 194.24099731445312, 163.82492065429688, 53.63121032714844, 41.823333740234375, 126.78218078613281, 158.13412475585938, 58.72148895263672, 14.01371955871582, 55.951385498046875, 30.664398193359375, -2.1885452270507812, 135.92135620117188, 2.349029541015625, 30.857086181640625, 124.61177062988281, 197.22970581054688, -2.032520294189453, 47.610477447509766, -4.22247314453125, -12.887962341308594, 62.09967041015625, -49.942134857177734, 20.08208465576172, 131.61685180664062, 37.421653747558594, -6.412664413452148, -2.3608627319335938, 123.77645874023438, 16.647987365722656, 76.47682189941406, -91.73638916015625, 146.46339416503906, 113.3385238647461, -0.4273414611816406, 157.36834716796875, 32.880367279052734, 14.047637939453125, -2.2347183227539062, 69.40644073486328, -86.51529693603516, 31.18616485595703, 8.967432022094727, 88.34727478027344, 3.7432861328125, 93.49334716796875, -40.99001693725586, 23.087244033813477, -80.7750015258789, 20.29997444152832, 81.30712890625, 194.74200439453125, 9.773788452148438, -58.75508117675781, 117.48092651367188, -69.65999603271484, 151.0044403076172, -42.33379364013672, 22.00305938720703, -35.955963134765625, -19.18256378173828], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000450.npy"}
|
|
{"epoch": 0.6802721088435374, "step": 451, "batch_size": 64, "mean": 40.047096252441406, "std": 68.0593032836914, "min": -117.99314880371094, "p10": -29.292755126953125, "median": 26.923751831054688, "p90": 130.30914306640625, "max": 189.91177368164062, "pos_frac": 0.6875, "sample": [86.49775695800781, -1.7822265625, 49.08518981933594, 29.714630126953125, 111.04573059082031, 38.74993896484375, -117.99314880371094, 47.527679443359375, 6.182703018188477, 146.70155334472656, 49.200347900390625, 52.848533630371094, 130.886474609375, 23.998077392578125, 128.9620361328125, 31.842926025390625, 44.57239532470703, 21.23719024658203, 20.358301162719727, 70.00275421142578, -0.9754924774169922, -86.65840911865234, -63.87546920776367, 189.91177368164062, 127.72795867919922, -29.159988403320312, 72.59112548828125, -82.00213623046875, -29.349655151367188, -0.055606842041015625, 57.939842224121094, 174.01254272460938, -21.637645721435547, 2.746877670288086, 110.43736267089844, 63.33928680419922, 149.52911376953125, 100.94439697265625, -17.225189208984375, 14.082878112792969, 6.7158355712890625, 98.13641357421875, 1.9343719482421875, -3.0746307373046875, 76.01028442382812, 14.176177978515625, 4.752208709716797, 125.14997863769531, 161.03411865234375, 63.809295654296875, -11.06915283203125, 118.5633316040039, 24.13287353515625, -32.709388732910156, -2.7068710327148438, -9.478713989257812, 123.92425537109375, 0.7326164245605469, -4.09507942199707, 52.9782600402832, 156.178466796875, -11.603946685791016, -85.69276428222656, -6.746116638183594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000451.npy"}
|
|
{"epoch": 0.6817838246409675, "step": 452, "batch_size": 64, "mean": 45.00667953491211, "std": 74.19367980957031, "min": -133.60679626464844, "p10": -45.75568695068359, "median": 38.690391540527344, "p90": 135.8440887451172, "max": 190.45191955566406, "pos_frac": 0.78125, "sample": [28.440406799316406, -16.887367248535156, 20.37965202331543, 78.47857666015625, 13.611038208007812, 57.459068298339844, 21.56914520263672, 136.5977783203125, 130.9005889892578, 142.2801513671875, 14.390804290771484, 184.20082092285156, -133.10733032226562, 98.16270446777344, 133.92276000976562, 90.96269989013672, 190.45191955566406, 74.28240966796875, 92.1826171875, 84.16973876953125, 16.636192321777344, -19.926162719726562, 172.16848754882812, 100.654052734375, 14.86001205444336, -17.67179298400879, 119.46734619140625, 34.01617431640625, 5.583900451660156, 18.076576232910156, 3.4878273010253906, 174.0042266845703, -64.60263061523438, 134.08547973632812, -46.37467956542969, 77.40882873535156, 32.243690490722656, 69.79646301269531, -87.87702178955078, 17.918102264404297, 73.51407623291016, -44.311370849609375, 69.4273910522461, 48.577056884765625, 60.90373229980469, -95.2845687866211, 156.93856811523438, 96.85033416748047, 43.36460876464844, 2.1602630615234375, 84.09671783447266, -13.222362518310547, 13.728113174438477, 100.17579650878906, 2.496112823486328, 1.3797149658203125, -33.726524353027344, 132.29051208496094, 114.9421615600586, -59.131103515625, -10.164779663085938, 61.36084747314453, -133.60679626464844, 11.265670776367188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000452.npy"}
|
|
{"epoch": 0.6832955404383976, "step": 453, "batch_size": 64, "mean": 50.110694885253906, "std": 74.74535369873047, "min": -97.69188690185547, "p10": -41.31236648559569, "median": 40.24397659301758, "p90": 156.82335510253907, "max": 199.2140655517578, "pos_frac": 0.6875, "sample": [95.8440170288086, 39.87443542480469, -30.129974365234375, -4.3846893310546875, 157.13299560546875, 16.331619262695312, -5.2277069091796875, 130.3697052001953, 102.75990295410156, -11.175315856933594, -88.07404327392578, 2.1498641967773438, 46.447113037109375, -5.115165710449219, 119.65433502197266, 141.67257690429688, 199.2140655517578, 49.11865234375, 156.66546630859375, 58.62559127807617, 2.498922348022461, -19.180397033691406, 51.83592224121094, 21.2681884765625, 2.7874679565429688, -97.69188690185547, -81.67273712158203, 156.89102172851562, -18.08789825439453, 142.12515258789062, 165.86648559570312, 27.132667541503906, 94.58708190917969, -49.9117431640625, -8.052181243896484, 40.61351776123047, 133.92791748046875, 116.71945190429688, -53.39710235595703, 143.6337890625, 42.416664123535156, 177.8940887451172, -46.55015182495117, -2.340564727783203, -15.080978393554688, 37.93824768066406, -46.104820251464844, 26.101032257080078, 113.34929656982422, 70.92831420898438, 4.802829742431641, -20.463214874267578, 54.75243377685547, 72.0790023803711, 77.84312438964844, -24.681568145751953, -0.24925613403320312, 87.1678466796875, 153.96499633789062, 35.75384521484375, 177.79727172851562, 111.09776306152344, 168.2601318359375, 6.7610321044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000453.npy"}
|
|
{"epoch": 0.6848072562358276, "step": 454, "batch_size": 64, "mean": 52.0062255859375, "std": 74.9896240234375, "min": -179.76715087890625, "p10": -21.08816184997558, "median": 38.84870910644531, "p90": 144.01159973144533, "max": 227.525390625, "pos_frac": 0.796875, "sample": [135.46307373046875, -40.838836669921875, -14.738174438476562, 142.7174072265625, -179.76715087890625, -12.761484146118164, 121.68072509765625, 16.93982696533203, 10.665393829345703, 38.202362060546875, 66.96633911132812, 142.4246368408203, 28.491363525390625, -145.37330627441406, 153.802001953125, 70.56523132324219, 150.244873046875, 65.35955810546875, 139.08285522460938, 31.768783569335938, -8.390617370605469, -85.59628295898438, 1.1823978424072266, 142.30178833007812, 158.24203491210938, 227.525390625, -6.196504592895508, 25.961013793945312, -17.58374786376953, 23.63756561279297, -23.949264526367188, 67.70624542236328, 112.17800903320312, 22.680259704589844, 57.759979248046875, 35.93326187133789, 171.7216033935547, -45.21432876586914, 156.35572814941406, 41.317649841308594, 67.49876403808594, 93.77055358886719, -2.4935684204101562, 0.5021543502807617, 5.2678375244140625, 24.37586212158203, 144.56625366210938, 109.07220458984375, 9.078607559204102, 39.49505615234375, 98.11753845214844, 120.86695861816406, 9.208324432373047, 124.39865112304688, 111.857177734375, 33.31852722167969, 87.20158386230469, 75.38360595703125, 8.304267883300781, 105.08743286132812, 81.8470230102539, 25.530681610107422, 0.26503944396972656, -22.59005355834961], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000454.npy"}
|
|
{"epoch": 0.6863189720332578, "step": 455, "batch_size": 64, "mean": 52.12287902832031, "std": 69.4990234375, "min": -88.37081146240234, "p10": -29.477046966552734, "median": 36.236040115356445, "p90": 151.59687194824218, "max": 210.91537475585938, "pos_frac": 0.765625, "sample": [96.333984375, 22.806671142578125, 54.49054718017578, 168.6572265625, 56.665184020996094, 0.6598377227783203, 175.7830352783203, 136.90184020996094, -27.70154571533203, 19.23236083984375, 68.81089782714844, -88.37081146240234, 77.55453491210938, 62.45466232299805, 61.024993896484375, 210.91537475585938, -12.448661804199219, 7.9514923095703125, 21.94134521484375, -45.80335235595703, 151.85769653320312, 1.2673206329345703, 121.92150115966797, -2.781085968017578, -14.2183837890625, 135.131103515625, 41.33403015136719, 102.83059692382812, 148.57449340820312, 32.71813201904297, 14.35434341430664, 16.169809341430664, 150.98828125, 17.21216583251953, 44.07585144042969, -1.0272598266601562, -66.57778930664062, 80.67960357666016, 116.05384063720703, 149.69725036621094, -42.24618911743164, 12.692062377929688, 67.989990234375, -30.23797607421875, 132.86354064941406, 175.49815368652344, 50.842750549316406, 51.066261291503906, 12.252899169921875, 7.3674468994140625, 160.39169311523438, 52.38811492919922, -1.9096260070800781, 119.77891540527344, -4.6686553955078125, 30.35369873046875, 39.75394821166992, -44.40538787841797, -34.691436767578125, 25.866249084472656, 29.74171257019043, 30.15811538696289, -7.548152923583984, 194.44509887695312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000455.npy"}
|
|
{"epoch": 0.6878306878306878, "step": 456, "batch_size": 64, "mean": 45.853267669677734, "std": 69.36479949951172, "min": -94.76585388183594, "p10": -18.69857654571533, "median": 30.4954833984375, "p90": 145.76733551025393, "max": 230.63735961914062, "pos_frac": 0.765625, "sample": [89.70732116699219, -3.9491958618164062, 0.2438678741455078, 72.97872161865234, 230.63735961914062, 0.3076629638671875, -12.189250946044922, 68.05879211425781, -3.677165985107422, 115.7091064453125, 56.875885009765625, 43.22928237915039, 156.84640502929688, 138.28515625, 26.505111694335938, 8.486526489257812, 72.6710205078125, 54.250404357910156, 8.749326705932617, 138.11041259765625, -58.827911376953125, 10.637855529785156, 58.13703155517578, -11.859413146972656, 107.77311706542969, -1.1839847564697266, -94.76585388183594, -93.48036193847656, 102.30287170410156, -36.946929931640625, 12.66683578491211, 47.45759582519531, 18.238962173461914, 84.18907165527344, 15.842979431152344, 86.09014129638672, 32.55415344238281, -0.4530487060546875, 83.70623779296875, 5.698009490966797, -19.519506454467773, 16.82761001586914, 38.751121520996094, 54.42433166503906, -76.61834716796875, -93.17232513427734, 113.37417602539062, 6.9247589111328125, -16.78307342529297, -4.050975799560547, 103.70426940917969, 156.1027069091797, 11.220062255859375, 176.5458221435547, 28.436813354492188, 125.16079711914062, 148.97398376464844, 14.953193664550781, 92.938720703125, 152.43124389648438, 26.921113967895508, 16.136505126953125, 48.20091247558594, 183.11102294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000456.npy"}
|
|
{"epoch": 0.6893424036281179, "step": 457, "batch_size": 64, "mean": 47.18499755859375, "std": 61.23504638671875, "min": -92.73633575439453, "p10": -14.237775230407712, "median": 36.36514091491699, "p90": 133.28848419189453, "max": 170.54820251464844, "pos_frac": 0.796875, "sample": [13.950859069824219, -0.3011474609375, 4.381710052490234, 86.90191650390625, 54.81361389160156, -69.42840576171875, 34.6403694152832, 48.49052429199219, 16.00835418701172, 5.076416015625, 25.558401107788086, 142.6666259765625, 115.68038940429688, 80.69763946533203, -8.15896987915039, -60.03807067871094, 132.64662170410156, 18.968116760253906, 0.03978919982910156, 114.13571166992188, 133.56356811523438, 78.10885620117188, 170.54820251464844, 23.110763549804688, -2.8762969970703125, 67.83998107910156, -15.583108901977539, 129.8055419921875, -11.098663330078125, 10.977039337158203, -52.983154296875, -92.73633575439453, 7.491996765136719, 92.3251953125, 23.11075210571289, -22.360504150390625, -73.39393615722656, 41.946929931640625, 57.851112365722656, 19.541839599609375, 81.14960479736328, 5.481559753417969, 144.32254028320312, 119.00618743896484, -1.046478271484375, 15.598365783691406, 95.88786315917969, 10.683359146118164, 17.390830993652344, 149.19003295898438, 97.44071197509766, 63.04944610595703, 118.99948120117188, 87.90444946289062, 95.78254699707031, 154.32032775878906, 147.99609375, 38.08991241455078, 21.779279708862305, 20.515331268310547, 90.30760192871094, -1.815399169921875, 55.339691162109375, 50.54634094238281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000457.npy"}
|
|
{"epoch": 0.690854119425548, "step": 458, "batch_size": 64, "mean": 51.03356170654297, "std": 72.397705078125, "min": -157.39508056640625, "p10": -23.927894210815428, "median": 38.191802978515625, "p90": 151.23799591064454, "max": 202.97669982910156, "pos_frac": 0.78125, "sample": [98.61418151855469, 119.62508392333984, 5.311433792114258, 110.55401611328125, 48.29932403564453, 34.87955093383789, 74.38087463378906, -157.39508056640625, -0.08490753173828125, 12.729476928710938, -12.342117309570312, -60.4649658203125, -24.231491088867188, 139.7031707763672, 136.13584899902344, 24.41303062438965, 42.792030334472656, 136.72116088867188, 34.17963409423828, 56.16841125488281, 14.467367172241211, 29.007232666015625, 38.259254455566406, -12.048698425292969, 125.20811462402344, 16.144882202148438, 147.6229705810547, 74.66992950439453, 25.84626007080078, 116.64896392822266, 167.1925048828125, 46.89637756347656, 35.19337844848633, -52.26636505126953, 36.16950988769531, -0.19403648376464844, 17.98785400390625, 28.13427734375, 54.937774658203125, 140.54515075683594, 96.86892700195312, 42.41103744506836, -123.49970245361328, -2.023792266845703, 160.476806640625, 8.265178680419922, 93.41902160644531, -2.8063602447509766, 152.78729248046875, 167.50872802734375, -23.219501495361328, 202.97669982910156, 93.22975158691406, 38.124351501464844, -60.66260528564453, 159.4899444580078, 46.665122985839844, 125.53589630126953, -46.290924072265625, 7.318382263183594, 8.791133880615234, 26.767288208007812, 162.10366821289062, 61.5], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000458.npy"}
|
|
{"epoch": 0.6923658352229781, "step": 459, "batch_size": 64, "mean": 63.13481903076172, "std": 73.86588287353516, "min": -126.38214111328125, "p10": -19.770085716247557, "median": 59.936241149902344, "p90": 161.96355895996095, "max": 199.15724182128906, "pos_frac": 0.828125, "sample": [55.38361358642578, 96.84586334228516, -19.32636260986328, 99.43089294433594, 7.335624694824219, -126.38214111328125, 5.84056282043457, 144.64346313476562, 80.172607421875, 123.017333984375, 165.97189331054688, -28.432315826416016, 73.33876037597656, -31.031700134277344, 157.85885620117188, 173.0159149169922, 62.95564270019531, 72.58370971679688, 15.876277923583984, 115.91378021240234, 112.95233917236328, 128.5046844482422, 52.269046783447266, 24.552268981933594, 12.273767471313477, 135.45611572265625, 25.639556884765625, 121.39681243896484, 50.34624099731445, 114.64718627929688, 175.7681884765625, 11.916877746582031, -19.96025276184082, 15.01869010925293, -0.32537078857421875, 56.916839599609375, 70.31909942626953, -99.34434509277344, 140.1710205078125, -11.768901824951172, 0.07332420349121094, 89.4298095703125, 134.2563018798828, 4.10693359375, 199.15724182128906, -55.45091247558594, 76.3653793334961, 14.404899597167969, -78.40948486328125, 123.08379364013672, 2.7381057739257812, 13.185379028320312, -13.335485458374023, 156.877685546875, 87.71662902832031, 25.363967895507812, 133.0755157470703, 55.917999267578125, 163.72271728515625, 152.69961547851562, 168.28253173828125, 8.026496887207031, 165.6492462158203, 51.928993225097656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000459.npy"}
|
|
{"epoch": 0.6938775510204082, "step": 460, "batch_size": 64, "mean": 62.04758071899414, "std": 71.3957290649414, "min": -78.95889282226562, "p10": -20.06517639160156, "median": 40.30424118041992, "p90": 154.14468994140626, "max": 221.88230895996094, "pos_frac": 0.796875, "sample": [168.9994354248047, 80.52729797363281, 188.06900024414062, -19.227859497070312, 34.524635314941406, 149.7355499267578, 167.8943328857422, 38.45758056640625, 221.88230895996094, 4.509071350097656, -27.479171752929688, -18.55841064453125, 19.055992126464844, 15.133342742919922, 7.86639404296875, 2.7612380981445312, 18.79334831237793, 137.99485778808594, -21.84130859375, 149.13717651367188, 41.01085662841797, 153.95742797851562, 34.26453399658203, 154.22494506835938, 136.77377319335938, 96.84223937988281, 18.513500213623047, 47.671173095703125, 141.9527587890625, 57.84716033935547, 86.75384521484375, 146.5499267578125, 120.12445068359375, 81.98408508300781, 15.832094192504883, 67.79214477539062, 9.421283721923828, 34.564666748046875, 19.341644287109375, 0.3941173553466797, 89.22225952148438, 138.4290313720703, -16.018775939941406, -0.2538585662841797, 96.97988891601562, 157.66351318359375, -29.73651123046875, -20.424026489257812, 113.94155883789062, -78.95889282226562, -1.1287517547607422, 39.597625732421875, 141.109619140625, 24.620817184448242, 18.409194946289062, 148.285400390625, 41.10706329345703, 173.25042724609375, -51.54157638549805, -39.97175598144531, 89.52511596679688, 3.763702392578125, -4.643623352050781, 153.76638793945312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000460.npy"}
|
|
{"epoch": 0.6953892668178382, "step": 461, "batch_size": 64, "mean": 61.22492980957031, "std": 69.67225646972656, "min": -62.27079391479492, "p10": -26.90047512054443, "median": 49.709699630737305, "p90": 154.77180938720704, "max": 185.4267578125, "pos_frac": 0.765625, "sample": [160.071533203125, 72.1326675415039, 36.872222900390625, 42.05531311035156, 133.75808715820312, 155.94056701660156, 39.750389099121094, 125.43025970458984, 128.68814086914062, -40.12273025512695, -40.87211608886719, 39.23603820800781, 53.98133850097656, 5.043600082397461, 96.00421142578125, -13.401412963867188, -31.283065795898438, 158.10955810546875, 2.978485107421875, 6.006780624389648, 95.81041717529297, 120.18389892578125, 114.48114776611328, 12.663299560546875, 19.56214714050293, 89.1414794921875, -28.52465057373047, -6.410774230957031, -22.79163360595703, 77.31134033203125, 135.02066040039062, -35.20983123779297, -27.0881290435791, 59.574668884277344, 96.27275085449219, 152.04470825195312, 185.4267578125, 120.3001708984375, 35.215362548828125, 182.92837524414062, 133.89576721191406, 147.91268920898438, 111.67501831054688, 57.16187286376953, 4.216020584106445, -16.388870239257812, 45.43806076049805, -1.6199932098388672, 132.91050720214844, 23.81122589111328, 142.37120056152344, 75.94580841064453, 20.911956787109375, 17.968551635742188, 149.96826171875, -26.462615966796875, 166.89352416992188, -62.27079391479492, 17.935476303100586, -19.595577239990234, 144.69595336914062, 4.131309509277344, 158.89410400390625, -18.296188354492188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000461.npy"}
|
|
{"epoch": 0.6969009826152683, "step": 462, "batch_size": 64, "mean": 66.06465148925781, "std": 71.17271423339844, "min": -127.30052947998047, "p10": -10.910999870300287, "median": 64.71345520019531, "p90": 148.4502960205078, "max": 201.82708740234375, "pos_frac": 0.8125, "sample": [143.89930725097656, 166.9272918701172, 161.16622924804688, 138.69712829589844, 4.63970947265625, 33.35279083251953, 110.70972442626953, -2.7394676208496094, -1.2460098266601562, 3.8360977172851562, -23.079612731933594, 149.0599365234375, 141.154052734375, 30.48766326904297, 137.5191650390625, 154.13946533203125, 3.9763336181640625, 134.24180603027344, 28.36528778076172, 93.84400177001953, 114.88090515136719, 48.62239074707031, 42.5855712890625, 109.0261459350586, 56.173095703125, 8.431297302246094, 123.30821228027344, 147.02780151367188, -5.723140716552734, 115.81280517578125, -12.96087646484375, -21.584518432617188, 7.393547058105469, 139.75466918945312, 140.64617919921875, -41.869842529296875, -27.4593563079834, 85.09097290039062, 71.29988098144531, 70.81108093261719, 141.14096069335938, 146.9892578125, 77.8508529663086, 28.26378631591797, 30.965377807617188, 154.7511749267578, 5.7711944580078125, 22.23398208618164, 129.86395263671875, 122.7552261352539, 58.61582946777344, -110.19853210449219, 124.54027557373047, 149.74063110351562, 114.09663391113281, 6.2119293212890625, 23.164766311645508, 201.82708740234375, 85.7833480834961, 44.094825744628906, -127.30052947998047, -2.7037277221679688, -6.127954483032227, 25.58981704711914], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000462.npy"}
|
|
{"epoch": 0.6984126984126984, "step": 463, "batch_size": 64, "mean": 39.97034454345703, "std": 64.672119140625, "min": -140.18328857421875, "p10": -32.81437168121337, "median": 37.277137756347656, "p90": 121.63692626953126, "max": 164.1253204345703, "pos_frac": 0.75, "sample": [-5.259944915771484, 109.03654479980469, 89.06575012207031, -140.18328857421875, 82.39303588867188, 95.19387817382812, -18.91228675842285, 151.21609497070312, 114.2549057006836, 53.688270568847656, -43.23764419555664, -27.344955444335938, 13.678718566894531, 47.21989440917969, 112.47271728515625, -60.3590087890625, 27.368690490722656, 44.61686706542969, 2.68011474609375, -54.56914520263672, 57.61372375488281, 17.235260009765625, -17.562728881835938, 139.70721435546875, 57.153167724609375, -0.799072265625, 16.603012084960938, 18.57038116455078, -9.480968475341797, 2.4613609313964844, 17.223800659179688, 18.65180206298828, 150.6603546142578, 127.34007263183594, 39.34231185913086, 115.2640609741211, 101.8173599243164, 29.882911682128906, 4.237861633300781, 162.6333465576172, 53.109066009521484, 9.025932312011719, 37.92803955078125, -60.451416015625, -17.196941375732422, -34.03070068359375, 164.1253204345703, 91.37443542480469, 120.89578247070312, 90.24900817871094, 46.252628326416016, -102.0057601928711, 2.3622970581054688, -29.97627067565918, 36.62623596191406, 121.95455932617188, 53.74563980102539, 17.51372528076172, 77.4169692993164, 98.06224822998047, -11.212997436523438, 38.45091247558594, 6.442380905151367, 105.86628723144531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000463.npy"}
|
|
{"epoch": 0.6999244142101285, "step": 464, "batch_size": 64, "mean": 58.41905212402344, "std": 78.41796112060547, "min": -128.47994995117188, "p10": -39.69790802001951, "median": 56.263450622558594, "p90": 165.80153961181642, "max": 175.41627502441406, "pos_frac": 0.734375, "sample": [-23.138031005859375, 21.002334594726562, 144.99331665039062, -46.79499816894531, 140.4087677001953, -1.6217041015625, 0.19738006591796875, 141.2611083984375, 83.2984619140625, 91.672119140625, 135.28350830078125, -20.457382202148438, 113.00577545166016, 84.91169738769531, 68.58352661132812, -59.741790771484375, 164.9403076171875, 89.44638061523438, 18.076919555664062, 129.90121459960938, 22.479507446289062, 175.41627502441406, -20.103235244750977, 84.01969909667969, 21.501907348632812, 115.86758422851562, 34.10248565673828, 174.5036163330078, 90.42610168457031, 109.65620422363281, 57.584442138671875, 9.790191650390625, 151.71177673339844, 170.18577575683594, -107.2828369140625, -9.252395629882812, 157.58438110351562, 54.94245910644531, -74.55517578125, -60.09109115600586, 142.80079650878906, 20.62390899658203, 36.8256950378418, 166.70892333984375, -53.509117126464844, -6.149360656738281, 14.312980651855469, 45.603790283203125, -2.6125335693359375, 171.08580017089844, 96.47897338867188, 166.17063903808594, -128.47994995117188, -14.83187484741211, 31.041725158691406, -21.55904769897461, 22.855857849121094, 150.701171875, 96.2302474975586, 175.0196075439453, -20.22185516357422, 94.5628890991211, 113.58231353759766, 7.86114501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000464.npy"}
|
|
{"epoch": 0.7014361300075586, "step": 465, "batch_size": 64, "mean": 47.65568542480469, "std": 74.96724700927734, "min": -92.61013793945312, "p10": -56.387245178222656, "median": 42.28005027770996, "p90": 148.19021453857422, "max": 169.87466430664062, "pos_frac": 0.6875, "sample": [-50.77680969238281, 166.26553344726562, 125.42650604248047, -78.50914001464844, 114.23422241210938, 129.16299438476562, -19.289413452148438, 155.98573303222656, 26.673629760742188, 53.200706481933594, 40.8515625, 146.4514617919922, 108.32475280761719, 169.87466430664062, -57.71521759033203, 132.51358032226562, -8.442085266113281, 148.93539428710938, 85.89202880859375, 4.778587341308594, -32.459808349609375, -81.05101013183594, 141.1148681640625, 166.91107177734375, 89.77873229980469, 153.62892150878906, -53.28864288330078, -62.99263000488281, -20.262428283691406, 43.401283264160156, 5.405113220214844, -21.87689971923828, 48.64385986328125, 60.924903869628906, 85.02220153808594, 8.038482666015625, -59.639892578125, -7.223960876464844, 1.725229263305664, 125.69155883789062, -82.12428283691406, 22.29224395751953, 42.78385543823242, 35.15747833251953, -12.095756530761719, -92.61013793945312, -22.721324920654297, 41.7762451171875, 32.07745361328125, 144.03982543945312, -1.940399169921875, 135.64483642578125, 20.62207794189453, 80.23786926269531, 23.97071075439453, -20.068655014038086, 57.852088928222656, -5.516790390014648, 92.25257110595703, 81.4760971069336, 155.2064971923828, 81.68743896484375, 116.93746948242188, 137.69677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000465.npy"}
|
|
{"epoch": 0.7029478458049887, "step": 466, "batch_size": 64, "mean": 31.08350372314453, "std": 67.55465698242188, "min": -160.79583740234375, "p10": -48.3255443572998, "median": 22.27791976928711, "p90": 131.22273330688478, "max": 184.41921997070312, "pos_frac": 0.6875, "sample": [-9.877708435058594, 119.76026153564453, -37.40374755859375, -89.57159423828125, 35.406837463378906, -4.838998794555664, 11.200149536132812, -11.328781127929688, 29.301170349121094, -160.79583740234375, 50.082855224609375, 45.966094970703125, 138.14813232421875, -59.849586486816406, 60.948272705078125, 12.262680053710938, 87.72874450683594, 184.41921997070312, 31.080429077148438, 168.2550048828125, 54.065853118896484, 16.900062561035156, 133.85888671875, -0.15795135498046875, -1.38348388671875, 9.486686706542969, 79.90553283691406, -42.3499870300293, 173.28482055664062, 22.72784423828125, -2.2739944458007812, 67.19236755371094, -11.282691955566406, 41.41114807128906, 119.31062316894531, 29.451229095458984, 101.73579406738281, 29.30286407470703, -23.288700103759766, 31.62480926513672, -11.750953674316406, -88.3397216796875, -50.886497497558594, -52.67997741699219, 76.88528442382812, 26.149673461914062, 0.7622833251953125, 81.26921844482422, 177.0081787109375, -1.1305351257324219, -60.0911865234375, 44.28260803222656, -14.176719665527344, 39.28816223144531, 18.973068237304688, 2.402872085571289, 11.756895065307617, 13.593854904174805, 23.317045211791992, 125.07170867919922, 150.63470458984375, 13.495159149169922, 11.265785217285156, 21.82799530029297], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000466.npy"}
|
|
{"epoch": 0.7044595616024187, "step": 467, "batch_size": 64, "mean": 44.47019958496094, "std": 66.21611785888672, "min": -110.07079315185547, "p10": -18.39465751647949, "median": 25.84877109527588, "p90": 127.44457855224613, "max": 282.1828308105469, "pos_frac": 0.765625, "sample": [6.6363677978515625, 103.59922790527344, 25.232391357421875, 47.873992919921875, 14.042709350585938, 54.22742462158203, -0.97344970703125, -19.435787200927734, -6.227962493896484, 53.67566680908203, 86.61868286132812, -48.36931228637695, 145.24713134765625, -12.764127731323242, 61.31303405761719, 115.874755859375, 63.63811492919922, 25.993240356445312, -68.57749938964844, 9.072830200195312, 104.20077514648438, -82.65765380859375, -15.965354919433594, 105.4240493774414, 159.66184997558594, 18.29877471923828, 16.362892150878906, 74.13954162597656, -110.07079315185547, 95.15475463867188, -5.397636413574219, 49.80540466308594, 11.907552719116211, 28.036773681640625, -1.973114013671875, 14.105789184570312, 86.11102294921875, 156.6319580078125, 53.92604064941406, 72.68576049804688, 0.557861328125, -0.24094581604003906, -4.288215637207031, 25.704301834106445, 5.742254257202148, -23.329971313476562, 110.51773834228516, 2.876953125, 110.52131652832031, 94.59652709960938, 282.1828308105469, 9.954742431640625, 18.535850524902344, 10.280912399291992, 67.0494384765625, 61.629119873046875, 12.354911804199219, 158.99557495117188, 30.301227569580078, 146.7458953857422, -28.85626220703125, 120.62800598144531, 130.365966796875, 16.180801391601562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000467.npy"}
|
|
{"epoch": 0.7059712773998488, "step": 468, "batch_size": 64, "mean": 66.48197174072266, "std": 73.9620132446289, "min": -106.83613586425781, "p10": -22.330417823791503, "median": 68.11100769042969, "p90": 162.93050079345707, "max": 189.32217407226562, "pos_frac": 0.828125, "sample": [5.95549201965332, 37.83360290527344, 78.06800842285156, 188.24916076660156, 32.97870635986328, -20.818296432495117, 5.20775032043457, 129.96035766601562, 113.11383056640625, 40.98004150390625, 114.32496643066406, 170.40628051757812, 150.0288543701172, -106.83613586425781, 18.631057739257812, 77.62033081054688, 56.937599182128906, -33.3465576171875, -84.21884155273438, 82.75416564941406, 153.172607421875, 80.25662231445312, 131.13412475585938, 166.21066284179688, 21.5071964263916, 110.818115234375, 156.2823486328125, 7.110429763793945, -6.609153747558594, -77.89942169189453, 165.7797088623047, -69.55853271484375, 42.771728515625, -3.9892539978027344, 22.216941833496094, -41.52106475830078, 94.54395294189453, 90.36148834228516, -22.978469848632812, 139.85667419433594, 43.33784484863281, 69.1168212890625, 142.60626220703125, 171.09552001953125, 144.34230041503906, 5.359561920166016, 32.07636260986328, 35.49504089355469, 17.78099822998047, 15.890630722045898, 150.85385131835938, 55.91181182861328, 86.29232788085938, 39.592437744140625, 189.32217407226562, 74.98471069335938, -10.5775146484375, 181.95114135742188, 141.04196166992188, 146.82577514648438, 144.29025268554688, 91.61133575439453, 67.10519409179688, 1.2422561645507812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000468.npy"}
|
|
{"epoch": 0.7074829931972789, "step": 469, "batch_size": 64, "mean": 56.68113708496094, "std": 64.6186294555664, "min": -84.99180603027344, "p10": -7.982396697998043, "median": 54.27886199951172, "p90": 147.0340789794922, "max": 202.703125, "pos_frac": 0.8125, "sample": [20.239578247070312, 132.89215087890625, 97.25440979003906, 87.58167266845703, 10.183975219726562, 6.481292724609375, 32.28004455566406, 85.43824768066406, 72.9677734375, -84.99180603027344, 141.24436950683594, 58.62240982055664, 84.55335998535156, 0.10436248779296875, -9.771156311035156, 119.58990478515625, 95.25272369384766, 44.749107360839844, 69.01520538330078, 154.443115234375, 154.00848388671875, 8.757011413574219, 96.59662628173828, 7.12799072265625, 117.19842529296875, -77.75872039794922, 54.394920349121094, 10.9398193359375, 2.8845043182373047, 40.97943115234375, -2.4510498046875, 1.1543254852294922, 94.31427001953125, 168.8151397705078, 141.71475219726562, -15.300865173339844, 0.9305496215820312, -1.7286148071289062, 148.27874755859375, 113.9685287475586, 144.12985229492188, 71.96202850341797, 17.767635345458984, 89.23287200927734, 1.6908645629882812, -3.808624267578125, 69.87881469726562, -0.6018142700195312, 150.49867248535156, 33.824188232421875, -28.2525634765625, -1.7415504455566406, 187.19766235351562, 1.9971237182617188, 202.703125, 7.904716491699219, 69.91984558105469, 54.162803649902344, 98.30934143066406, -28.286123275756836, 91.86559295654297, 10.397077560424805, -16.672943115234375, 120.55908203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000469.npy"}
|
|
{"epoch": 0.708994708994709, "step": 470, "batch_size": 64, "mean": 54.28352355957031, "std": 72.57247924804688, "min": -164.2198486328125, "p10": -26.744008445739745, "median": 53.362815856933594, "p90": 146.7494064331055, "max": 180.02645874023438, "pos_frac": 0.765625, "sample": [120.1668701171875, 68.01526641845703, 180.02645874023438, -164.2198486328125, 49.78660583496094, 134.2979736328125, 46.95098876953125, 30.827964782714844, 138.87786865234375, 106.943603515625, 135.45059204101562, -68.41268157958984, 159.76614379882812, 17.29973602294922, 130.1546630859375, 17.14440155029297, 80.81625366210938, -39.43354797363281, -6.3898162841796875, 127.85497283935547, -42.12504959106445, 158.61090087890625, 14.890390396118164, -54.77947998046875, 150.5240020751953, 45.205894470214844, 130.27105712890625, 122.60450744628906, 15.651802062988281, -27.199769973754883, 66.252197265625, 1.3443450927734375, 103.8011245727539, 6.726249694824219, -50.69525146484375, 9.165794372558594, 60.89751052856445, 4.696104049682617, 56.93902587890625, -24.029312133789062, 142.78932189941406, 147.78977966308594, 144.32186889648438, 2.5409927368164062, -25.680564880371094, -22.985031127929688, 19.97764015197754, 121.61618041992188, 80.36541748046875, 125.46443939208984, 92.716552734375, 156.69137573242188, -20.91254425048828, 82.20278930664062, 20.176963806152344, 95.168212890625, 64.83504486083984, 77.85543823242188, -7.7014312744140625, -10.781732559204102, 164.47708129882812, -2.2287158966064453, 5.646739959716797, 5.123268127441406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000470.npy"}
|
|
{"epoch": 0.7105064247921391, "step": 471, "batch_size": 64, "mean": 63.787384033203125, "std": 74.7765884399414, "min": -140.07618713378906, "p10": -19.114245605468746, "median": 57.51313781738281, "p90": 155.6711181640625, "max": 203.41644287109375, "pos_frac": 0.796875, "sample": [10.104179382324219, 180.94570922851562, 133.79859924316406, 119.63233184814453, 51.15679931640625, -20.214645385742188, 140.18597412109375, 71.99567413330078, 143.14056396484375, 169.85610961914062, 82.23412322998047, 37.84950256347656, 161.68191528320312, 10.354660034179688, -98.48587036132812, -2.988058090209961, 55.08080291748047, 164.37625122070312, 145.17684936523438, -64.45541381835938, 23.968936920166016, 60.4628791809082, 120.8617935180664, 55.651763916015625, 128.77716064453125, -140.07618713378906, 88.23119354248047, 92.01065063476562, 155.06056213378906, 7.470348358154297, 126.85057067871094, 68.28529357910156, 27.204540252685547, 21.11419677734375, 59.37451171875, 68.87439727783203, 152.2716064453125, -23.974868774414062, 13.213722229003906, 143.7658233642578, 16.195655822753906, -7.58416748046875, 155.9327850341797, 150.24838256835938, -8.567548751831055, 22.516891479492188, 44.75169372558594, 96.90959167480469, -41.11774444580078, 113.55766296386719, -16.546646118164062, 153.773681640625, 43.368736267089844, 9.806205749511719, 109.4903793334961, -4.350971221923828, 17.772857666015625, 94.23271179199219, -63.74775695800781, 203.41644287109375, 15.9783935546875, 54.7490234375, -2.7743148803710938, 183.5557098388672], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000471.npy"}
|
|
{"epoch": 0.7120181405895691, "step": 472, "batch_size": 64, "mean": 50.75205612182617, "std": 83.91259002685547, "min": -155.57675170898438, "p10": -57.52799072265623, "median": 39.564971923828125, "p90": 157.77535095214844, "max": 261.38079833984375, "pos_frac": 0.765625, "sample": [7.881736755371094, 37.12921905517578, 55.607765197753906, -37.28141784667969, 154.33067321777344, 139.69210815429688, -17.962631225585938, 125.34712219238281, 163.96197509765625, 113.86651611328125, 48.62997055053711, 188.27349853515625, 23.456579208374023, -33.354827880859375, -5.564352035522461, 156.1048583984375, 9.609174728393555, 69.16021728515625, 161.8297119140625, 21.096328735351562, -66.20509338378906, 43.902557373046875, -23.413715362548828, 140.36801147460938, 7.073577880859375, 120.93287658691406, 21.518081665039062, 157.33462524414062, 157.9642333984375, -26.074565887451172, 14.691436767578125, 62.10796356201172, 43.292266845703125, -80.79059600830078, -30.253219604492188, 35.946563720703125, 166.4308624267578, 261.38079833984375, -4.5189971923828125, 10.15970230102539, 135.50990295410156, 158.0745391845703, -155.57675170898438, 154.2457275390625, -88.44731140136719, -74.76213836669922, 138.3344268798828, 12.358491897583008, -111.37141418457031, 90.32146453857422, 6.737525939941406, 44.76469421386719, 11.04150390625, 15.950172424316406, 24.017425537109375, 42.00072479248047, 59.88726806640625, 20.574920654296875, 153.81219482421875, 30.419296264648438, 77.82145690917969, 50.249122619628906, -79.83128356933594, 138.33807373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000472.npy"}
|
|
{"epoch": 0.7135298563869993, "step": 473, "batch_size": 64, "mean": 53.142547607421875, "std": 86.78927612304688, "min": -125.54631042480469, "p10": -51.782799530029294, "median": 36.13990020751953, "p90": 176.33277282714843, "max": 239.43524169921875, "pos_frac": 0.75, "sample": [186.91604614257812, 168.748291015625, 23.471601486206055, -105.61209106445312, -72.91534423828125, 118.11683654785156, 32.023353576660156, 29.549957275390625, 54.03772735595703, 161.99961853027344, 145.22689819335938, 30.069669723510742, -3.1273059844970703, 58.78529357910156, 11.163658142089844, -38.163330078125, 199.60777282714844, 176.24777221679688, 23.596153259277344, 158.11782836914062, -97.70884704589844, 25.72185516357422, 104.74090576171875, -39.183677673339844, -53.65950012207031, 118.09284973144531, 91.51207733154297, 135.04611206054688, 13.71837043762207, 0.9439163208007812, -10.754058837890625, 3.0371055603027344, -125.54631042480469, -17.764076232910156, -27.655107498168945, 77.44125366210938, 239.43524169921875, 191.21185302734375, 18.816158294677734, 7.595026016235352, 192.64785766601562, 126.35978698730469, 178.7786865234375, 101.027587890625, 63.58134078979492, 57.86360549926758, 40.256446838378906, 9.7139892578125, 26.03650665283203, 138.4850311279297, 85.96220397949219, -1.5347442626953125, 72.20718383789062, 68.88092803955078, -17.20980453491211, 61.071434020996094, 176.36920166015625, -98.94081115722656, 8.4361572265625, -47.403831481933594, 12.346000671386719, 116.47728729248047, 125.86182403564453, -109.05224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000473.npy"}
|
|
{"epoch": 0.7150415721844293, "step": 474, "batch_size": 64, "mean": 44.872154235839844, "std": 80.47221374511719, "min": -137.02989196777344, "p10": -38.81858825683594, "median": 25.501426696777344, "p90": 152.03923339843752, "max": 217.8150177001953, "pos_frac": 0.6875, "sample": [162.77487182617188, 139.969970703125, 92.32481384277344, -48.327823638916016, -11.306842803955078, 2.731822967529297, 96.87020874023438, 171.15054321289062, 81.21873474121094, 24.497711181640625, 61.353607177734375, 75.89830017089844, -137.02989196777344, 88.90863037109375, 46.79847717285156, 145.99790954589844, -81.4319076538086, 0.005252838134765625, -27.326873779296875, 25.231048583984375, -38.35032653808594, -17.62567138671875, 123.94398498535156, 25.235383987426758, 101.7913818359375, 17.000764846801758, -0.9922809600830078, -39.01927185058594, -5.325836181640625, -17.37067413330078, 119.27265167236328, 33.54634475708008, -124.61924743652344, 5.7147979736328125, -15.109683990478516, 130.6895751953125, 13.370803833007812, 162.65338134765625, 153.38302612304688, 0.48571014404296875, 25.76746940612793, 28.030120849609375, 177.44607543945312, 105.31663513183594, -24.804813385009766, 217.8150177001953, 125.48553466796875, 55.61317443847656, 130.66653442382812, -9.118326187133789, 148.90371704101562, 176.7767333984375, -11.991195678710938, 26.00939178466797, 19.72509002685547, -18.440509796142578, 144.7585906982422, -98.89093017578125, 21.325374603271484, 46.59504699707031, 134.86111450195312, -96.28619384765625, -2.03741455078125, 9.308361053466797], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000474.npy"}
|
|
{"epoch": 0.7165532879818595, "step": 475, "batch_size": 64, "mean": 71.81007385253906, "std": 73.77912902832031, "min": -61.107749938964844, "p10": -8.97530117034912, "median": 64.07228660583496, "p90": 168.94536895751955, "max": 217.09747314453125, "pos_frac": 0.859375, "sample": [-0.13300514221191406, 8.477985382080078, 16.956016540527344, 4.6327667236328125, 134.3573455810547, 87.1714096069336, 151.47061157226562, 165.9102020263672, 4.072309494018555, -25.272850036621094, 121.5343246459961, -52.79620361328125, 132.60794067382812, 196.98629760742188, 9.607004165649414, 23.301925659179688, 116.64933013916016, 131.49314880371094, 15.990283966064453, 111.4815444946289, 44.20171356201172, 13.838088989257812, -8.422510147094727, 25.58629608154297, 109.21752166748047, 7.076114654541016, 27.81993865966797, 150.2905731201172, 98.70531463623047, -9.212211608886719, 171.36776733398438, 158.892333984375, 148.5155792236328, 173.80709838867188, 40.00743865966797, 2.029102325439453, 105.55428314208984, 4.810089111328125, 110.70466613769531, 26.473751068115234, 189.814697265625, 125.1964111328125, 11.092708587646484, 22.04595947265625, 118.34788513183594, -46.310848236083984, 15.405391693115234, 81.4422607421875, 217.09747314453125, 23.02478790283203, 152.3682098388672, 192.97842407226562, 170.24615478515625, 110.50150299072266, 138.37142944335938, -29.780738830566406, -61.107749938964844, 149.88088989257812, 74.75808715820312, 147.41702270507812, 13.45730972290039, 4.883857727050781, -34.43431091308594, 53.3864860534668], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000475.npy"}
|
|
{"epoch": 0.7180650037792895, "step": 476, "batch_size": 64, "mean": 47.22654342651367, "std": 76.94304656982422, "min": -110.65455627441406, "p10": -43.278640747070305, "median": 38.72284698486328, "p90": 153.44300231933596, "max": 225.75735473632812, "pos_frac": 0.71875, "sample": [167.78790283203125, 51.30898666381836, -23.562353134155273, 57.16893005371094, -37.76310729980469, 37.454193115234375, -23.564308166503906, 154.62649536132812, 225.75735473632812, 27.526687622070312, 188.69912719726562, -1.7210845947265625, 150.6815185546875, 118.38983154296875, 112.8555908203125, 91.82389068603516, 86.08990478515625, 8.183662414550781, 64.0360336303711, 29.848114013671875, -86.75650787353516, 3.213825225830078, 94.11245727539062, 127.14364624023438, 39.57814025878906, -110.65455627441406, 7.113899230957031, -7.051658630371094, -21.701324462890625, 20.139728546142578, 169.2705078125, -27.874893188476562, 100.32798767089844, 17.492610931396484, 58.14824676513672, -45.64244079589844, -79.11701965332031, -89.65498352050781, 56.464019775390625, 15.535789489746094, -10.711959838867188, 90.05799865722656, 8.957672119140625, 2.644134521484375, 72.98403930664062, 48.92523193359375, 100.05509948730469, 2.8813400268554688, -18.01849365234375, 127.82683563232422, 88.90119171142578, -77.90240478515625, 156.07530212402344, -30.125732421875, 81.40754699707031, 189.86285400390625, 0.9289417266845703, 37.8675537109375, 139.7957763671875, 112.8350830078125, -26.069292068481445, 125.23863220214844, 120.971435546875, -48.60491180419922], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000476.npy"}
|
|
{"epoch": 0.7195767195767195, "step": 477, "batch_size": 64, "mean": 48.207820892333984, "std": 89.41880798339844, "min": -154.10910034179688, "p10": -79.11202163696288, "median": 37.080379486083984, "p90": 159.14734497070313, "max": 219.88446044921875, "pos_frac": 0.765625, "sample": [-122.97254943847656, -82.84919738769531, 123.1586685180664, 41.270347595214844, 8.23614501953125, 145.3342742919922, 1.1014022827148438, 16.12871742248535, 38.3360595703125, 1.8275604248046875, 149.70440673828125, 29.181182861328125, -102.37815856933594, -4.796577453613281, 33.534759521484375, 4.774513244628906, -17.187393188476562, 202.91397094726562, 89.17494201660156, -52.1162109375, 147.50257873535156, 126.09580993652344, 10.974334716796875, 83.52489471435547, 87.47068786621094, -18.915863037109375, -70.3919448852539, 3.4931564331054688, 150.62786865234375, 11.715225219726562, -110.10957336425781, 35.82469940185547, 29.76533317565918, 22.44692611694336, -154.10910034179688, -36.062679290771484, -24.672653198242188, 42.559837341308594, 88.1077880859375, 42.05341339111328, 162.01698303222656, -2.6209945678710938, 161.86607360839844, 160.2041015625, 84.63658142089844, 167.2381591796875, -124.9502182006836, 108.47064208984375, 113.1033706665039, 14.542205810546875, 219.88446044921875, 101.17863464355469, 10.319414138793945, 138.33424377441406, 169.4890899658203, 50.16017150878906, -145.65554809570312, 130.19778442382812, 129.5624237060547, 156.68157958984375, 27.677858352661133, 145.61505126953125, 15.161117553710938, 121.90966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000477.npy"}
|
|
{"epoch": 0.7210884353741497, "step": 478, "batch_size": 64, "mean": 41.46323776245117, "std": 80.45807647705078, "min": -135.39202880859375, "p10": -60.16899261474608, "median": 37.77024459838867, "p90": 151.4689651489258, "max": 177.88406372070312, "pos_frac": 0.671875, "sample": [-7.311826705932617, 153.95957946777344, 55.89031982421875, 129.81039428710938, 83.84365844726562, -0.8315887451171875, 164.2932586669922, 101.70513916015625, -14.215738296508789, 85.7562255859375, 6.8064117431640625, 11.588577270507812, 93.638671875, 177.88406372070312, -30.92498016357422, 16.199249267578125, -89.18336486816406, 38.13584899902344, -13.554710388183594, 44.115318298339844, -27.83766746520996, -29.609895706176758, 136.8487548828125, 100.70648193359375, 136.35052490234375, -78.28836822509766, 113.2549057006836, 90.30740356445312, 68.3443603515625, -135.39202880859375, 56.810791015625, 6.677356719970703, -119.40889739990234, -129.41094970703125, 19.068805694580078, 124.29127502441406, 145.65753173828125, 15.519317626953125, 133.8431396484375, 37.404640197753906, 161.61544799804688, -2.5656909942626953, -66.71804809570312, 38.891815185546875, 160.2677001953125, 59.68986511230469, -1.9753036499023438, 165.6479949951172, 23.607017517089844, -5.150371551513672, 116.03584289550781, 91.52552032470703, -40.26959228515625, 94.91230773925781, -1.4457855224609375, 166.9910888671875, 118.79458618164062, -43.34601593017578, 2.9683685302734375, 25.178131103515625, -44.88786315917969, -113.40261840820312, 62.099769592285156, 12.441102981567383], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000478.npy"}
|
|
{"epoch": 0.7226001511715797, "step": 479, "batch_size": 64, "mean": 36.16193389892578, "std": 72.8821792602539, "min": -183.3060302734375, "p10": -42.61075439453125, "median": 29.2693452835083, "p90": 126.36541290283205, "max": 231.2264404296875, "pos_frac": 0.703125, "sample": [16.757152557373047, 33.89615249633789, 65.0791015625, -75.43910217285156, -37.326507568359375, 47.96930694580078, 172.4268035888672, 85.25460815429688, 81.27642822265625, -42.677024841308594, 33.20498275756836, -20.547588348388672, 11.05853271484375, -16.647523880004883, -8.906791687011719, 33.239959716796875, 171.9175567626953, 113.90055084228516, -25.032331466674805, 1.1013145446777344, -43.952823638916016, 18.11764144897461, 231.2264404296875, 154.5196990966797, 30.482812881469727, -64.28533935546875, 28.055877685546875, 20.1488037109375, -61.057945251464844, -42.45612335205078, 1.1714019775390625, 87.66492462158203, 0.12798309326171875, 123.09312438964844, 9.98028564453125, 42.03132247924805, -114.26130676269531, -15.54443359375, 17.54376983642578, 127.767822265625, 95.97406005859375, -183.3060302734375, -7.3958740234375, 62.80975341796875, 135.18170166015625, 150.84315490722656, 85.13199615478516, -30.932876586914062, 4.1163177490234375, 54.418235778808594, 57.445770263671875, 101.53646087646484, -32.93446350097656, 105.64175415039062, -6.01902961730957, 96.18121337890625, 87.12678527832031, 22.83922004699707, 58.806785583496094, 110.65176391601562, -1.8920536041259766, 1.1390533447265625, 59.174171447753906, 96.94639587402344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000479.npy"}
|
|
{"epoch": 0.7241118669690099, "step": 480, "batch_size": 64, "mean": 57.10227966308594, "std": 79.7073745727539, "min": -167.89389038085938, "p10": -26.02982711791992, "median": 63.947715759277344, "p90": 145.19690551757816, "max": 252.39694213867188, "pos_frac": 0.78125, "sample": [89.93657684326172, 4.7540283203125, 137.5850067138672, -117.61155700683594, 119.4797134399414, 12.832098007202148, 123.57184600830078, -167.89389038085938, 13.595718383789062, 252.39694213867188, -10.228111267089844, 53.37699508666992, 131.7693328857422, -21.244834899902344, 62.893402099609375, -11.688186645507812, -34.13883972167969, 117.55708312988281, -23.376068115234375, 8.803180694580078, -82.40524291992188, 118.58777618408203, 44.68036651611328, -86.97229766845703, 148.19015502929688, 122.72858428955078, 219.18618774414062, 73.3282470703125, 125.48320007324219, 62.165802001953125, 84.97171783447266, 1.5394668579101562, 22.80982208251953, 95.85952758789062, 48.32691192626953, 87.48661804199219, 149.5403289794922, 103.05577087402344, -109.11432647705078, 84.80755615234375, 139.18307495117188, 103.90770721435547, 65.00202941894531, 118.58335876464844, 29.604415893554688, 107.74540710449219, 11.799064636230469, -8.619903564453125, 58.57865524291992, 78.24336242675781, -1.6625041961669922, 6.83087158203125, 147.77426147460938, 83.37678527832031, 107.10482788085938, 92.57283782958984, 161.0550537109375, 137.67066955566406, 26.107982635498047, -27.167152404785156, 8.933731079101562, 162.53729248046875, -16.37237548828125, 5.1298370361328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000480.npy"}
|
|
{"epoch": 0.7256235827664399, "step": 481, "batch_size": 64, "mean": 53.69420623779297, "std": 77.43951416015625, "min": -138.88169860839844, "p10": -32.798208618164054, "median": 41.97603225708008, "p90": 145.7267623901367, "max": 237.12542724609375, "pos_frac": 0.765625, "sample": [69.1461181640625, 12.457817077636719, 151.85162353515625, 42.18682098388672, -34.82135009765625, 102.7494125366211, 120.76361846923828, 62.485233306884766, -55.91419982910156, -46.275081634521484, 129.21157836914062, 47.412109375, 140.82167053222656, 87.14215850830078, -3.8144454956054688, 124.47897338867188, 133.03955078125, 29.950820922851562, 109.6302490234375, 132.09051513671875, -134.5517120361328, -3.5624008178710938, 143.43392944335938, 41.76524353027344, 0.2954521179199219, -23.88338279724121, 17.376266479492188, 137.38694763183594, 17.194026947021484, 79.40780639648438, 0.4156360626220703, 146.02418518066406, 25.626663208007812, -16.891082763671875, 45.92076110839844, 3.9913482666015625, 90.59037017822266, 16.164592742919922, 3.4175148010253906, 9.821693420410156, 22.10992431640625, 45.147789001464844, 144.24314880371094, 165.80264282226562, -10.05224609375, 170.2552947998047, -28.077545166015625, -11.450050354003906, 153.39077758789062, 145.03277587890625, 120.08950805664062, 33.260292053222656, 31.76183319091797, 8.223228454589844, -37.080230712890625, 237.12542724609375, -138.88169860839844, 71.4009017944336, 80.51022338867188, 227.6575927734375, 79.85244750976562, 39.00303649902344, -54.405120849609375, -13.027759552001953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000481.npy"}
|
|
{"epoch": 0.72713529856387, "step": 482, "batch_size": 64, "mean": 48.69911193847656, "std": 91.59036254882812, "min": -193.8697967529297, "p10": -59.004071044921865, "median": 28.733318328857422, "p90": 166.9149124145508, "max": 266.7454528808594, "pos_frac": 0.71875, "sample": [-2.8644256591796875, -193.8697967529297, -18.87664031982422, 83.99724578857422, 45.243263244628906, 104.42584991455078, 26.565155029296875, -14.900657653808594, 5.4766998291015625, 163.3811492919922, 65.66877746582031, 149.600830078125, 120.71546936035156, 35.23633575439453, 123.99836730957031, 91.3764419555664, 180.53689575195312, 15.679481506347656, -144.97691345214844, 82.54615020751953, 170.82696533203125, -9.889617919921875, 257.0522766113281, -46.52732849121094, -0.431488037109375, 102.79435729980469, 130.69512939453125, 61.429935455322266, -8.124313354492188, 6.872653961181641, -5.678565979003906, 126.70838165283203, -83.17204284667969, 80.70714569091797, 168.42938232421875, 140.12149047851562, 73.10832977294922, 19.893733978271484, 6.492839813232422, -30.949142456054688, 266.7454528808594, 78.31045532226562, 25.809852600097656, 13.377685546875, 11.9093017578125, 145.2689666748047, 30.90148162841797, 9.176248550415039, 231.3623809814453, -90.39851379394531, 86.53797149658203, 22.960830688476562, 68.60836791992188, 20.77813720703125, 109.94544982910156, 0.33298683166503906, -50.500389099121094, 198.11595153808594, -62.64850616455078, -71.64964294433594, 101.05696105957031, -8.294242858886719, -118.3699951171875, 18.086145401000977], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000482.npy"}
|
|
{"epoch": 0.7286470143613001, "step": 483, "batch_size": 64, "mean": 59.417572021484375, "std": 73.6280746459961, "min": -157.7246856689453, "p10": -13.272258377075195, "median": 50.4278564453125, "p90": 154.80380249023438, "max": 189.9386444091797, "pos_frac": 0.765625, "sample": [11.925666809082031, 157.1953582763672, -11.358379364013672, 66.58882141113281, 117.53273010253906, 150.13087463378906, 127.80799865722656, -7.7546539306640625, -13.658172607421875, -12.37179183959961, 57.24085998535156, 122.61048889160156, 106.08596801757812, 134.05276489257812, -33.79543685913086, 123.98149108886719, 63.43122482299805, -47.83936309814453, 6.73712158203125, 139.27999877929688, 42.53547668457031, 154.7779541015625, 148.06475830078125, 33.06687927246094, 158.305908203125, 55.258888244628906, 160.63804626464844, -6.299247741699219, -11.962905883789062, 54.032806396484375, 77.24117279052734, -111.78717041015625, 153.27114868164062, 2.700925827026367, 1.95880126953125, 175.57015991210938, 28.857627868652344, -157.7246856689453, 36.4034423828125, -27.99897003173828, 47.20215606689453, 29.37676239013672, 138.47535705566406, 59.96527862548828, -0.5908126831054688, 59.59714126586914, 189.9386444091797, 154.81488037109375, 53.65355682373047, 112.36056518554688, 29.307174682617188, 44.30525207519531, 35.463829040527344, 167.57464599609375, -4.84263801574707, 20.17140007019043, 38.672000885009766, 7.365976333618164, 153.4229736328125, 110.31908416748047, 12.183517456054688, -15.184288024902344, -3.3150062561035156, 137.752685546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000483.npy"}
|
|
{"epoch": 0.7301587301587301, "step": 484, "batch_size": 64, "mean": 44.27375030517578, "std": 59.38957214355469, "min": -96.9074935913086, "p10": -27.40899543762206, "median": 38.52699661254883, "p90": 124.18641586303714, "max": 175.08282470703125, "pos_frac": 0.765625, "sample": [32.69749450683594, 61.19769287109375, -3.3517913818359375, 13.082992553710938, 166.6673583984375, 8.281152725219727, 52.26310729980469, -33.44648742675781, 128.12855529785156, -13.275039672851562, 28.13141632080078, -96.9074935913086, 31.978824615478516, 22.019817352294922, 86.49055480957031, 81.259033203125, -30.770938873291016, -10.11935806274414, 80.00381469726562, 60.69829559326172, -1.7288074493408203, 112.53965759277344, 111.27235412597656, 21.71186065673828, 89.30154418945312, 84.761962890625, 154.24215698242188, 2.72674560546875, 77.60638427734375, 2.9471302032470703, 23.24605941772461, 37.55561065673828, -7.2129364013671875, -36.85240936279297, 175.08282470703125, -5.493602752685547, 33.69435119628906, 45.301971435546875, 59.70577621459961, 12.3248291015625, 114.98809051513672, -19.56446075439453, 168.49415588378906, -37.32183837890625, 107.51212310791016, 71.7419204711914, 148.26251220703125, -13.103923797607422, 79.54898071289062, 42.027435302734375, 154.32176208496094, 27.862415313720703, 54.44990539550781, 8.407730102539062, -48.74742889404297, -72.99858093261719, 49.89374923706055, 31.651580810546875, 78.40129089355469, 85.8951187133789, 39.498382568359375, 59.60809326171875, 41.44174575805664, 3.4868316650390625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000484.npy"}
|
|
{"epoch": 0.7316704459561603, "step": 485, "batch_size": 64, "mean": 47.15207290649414, "std": 69.46649169921875, "min": -140.70538330078125, "p10": -15.279150390624999, "median": 26.112791061401367, "p90": 156.02053070068362, "max": 212.331298828125, "pos_frac": 0.765625, "sample": [26.820030212402344, 74.02538299560547, 84.39039611816406, 77.23197937011719, 61.90740966796875, 2.4011611938476562, 7.234594345092773, 0.5387630462646484, 47.24028015136719, -43.083961486816406, -99.12606811523438, 23.78216552734375, 158.22738647460938, 140.97927856445312, -9.21783447265625, 149.75111389160156, 7.945659637451172, 148.6792449951172, 192.7292938232422, 20.849485397338867, 13.140996932983398, 79.22488403320312, -7.7527008056640625, 162.7540283203125, -1.94989013671875, 0.31662750244140625, -140.70538330078125, 81.53636169433594, 45.362220764160156, 72.5260009765625, -18.80583953857422, 0.9386787414550781, 62.80522918701172, -0.11664390563964844, 90.26426696777344, -52.79246520996094, 212.331298828125, 165.177001953125, -14.750091552734375, 115.11940002441406, 25.40555191040039, -20.54477882385254, 50.180633544921875, 5.151161193847656, 50.854347229003906, -1.3742694854736328, -4.243661880493164, 150.8844451904297, 95.92179870605469, 4.9810333251953125, 16.264232635498047, 42.16004180908203, 100.86673736572266, 159.55470275878906, 35.91320037841797, 98.11715698242188, 72.53209686279297, 158.22171020507812, -15.505889892578125, -5.212274551391602, 15.0430908203125, 3.2654075622558594, 18.307510375976562, 23.05889892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000485.npy"}
|
|
{"epoch": 0.7331821617535903, "step": 486, "batch_size": 64, "mean": 41.430946350097656, "std": 82.5999526977539, "min": -144.3964080810547, "p10": -49.923261642456055, "median": 33.89793395996094, "p90": 161.13499145507814, "max": 287.1193542480469, "pos_frac": 0.671875, "sample": [20.426719665527344, 2.509674072265625, 176.92388916015625, 13.215951919555664, -8.739715576171875, -2.8792152404785156, 31.703079223632812, 287.1193542480469, 18.737937927246094, -6.5518646240234375, -12.802736282348633, 99.6905517578125, 85.19763946533203, 50.38287353515625, 65.9107437133789, 13.732406616210938, -11.583908081054688, 163.41714477539062, -1.2270240783691406, 125.34439086914062, 84.69737243652344, 33.800079345703125, -50.12381362915039, -38.38824462890625, 177.68154907226562, -102.5362548828125, 39.804351806640625, -54.067440032958984, 63.44945526123047, -111.99717712402344, 162.49227905273438, 61.88221740722656, 33.99578857421875, -110.6465835571289, 11.755067825317383, 82.76231384277344, 45.15998077392578, 107.83396911621094, -73.7555160522461, 18.116939544677734, -0.3493003845214844, 151.4289093017578, 50.4127197265625, 170.73672485351562, 157.96798706054688, -40.311317443847656, 62.74229431152344, 106.7075424194336, 46.7540283203125, 154.2524871826172, 73.876220703125, -33.47282409667969, 25.413406372070312, 137.4384765625, -10.303825378417969, -49.45530700683594, 0.16678237915039062, 71.800048828125, 191.7534942626953, -12.816797256469727, 39.064483642578125, -144.3964080810547, 40.26499557495117, -30.538307189941406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000486.npy"}
|
|
{"epoch": 0.7346938775510204, "step": 487, "batch_size": 64, "mean": 50.10889434814453, "std": 74.32987976074219, "min": -122.79695892333984, "p10": -26.819139480590813, "median": 37.1124153137207, "p90": 148.8478713989258, "max": 281.3051452636719, "pos_frac": 0.765625, "sample": [105.62007141113281, -38.640716552734375, 79.59992218017578, 93.42092895507812, 36.34968566894531, 25.881582260131836, -19.67388916015625, 37.581932067871094, 91.53751373291016, 117.921142578125, 9.143035888671875, -19.203628540039062, 25.651901245117188, -30.255477905273438, -18.265525817871094, 161.11460876464844, 103.95013427734375, 44.104007720947266, 250.07403564453125, 52.168548583984375, -5.321083068847656, 107.12506103515625, 36.64289855957031, 154.18348693847656, 86.87430572509766, -29.881389617919922, 33.78418731689453, -113.998046875, 7.314445495605469, 36.380523681640625, 144.29046630859375, 44.36847686767578, 52.936729431152344, -3.2467784881591797, 150.36322021484375, 5.706489562988281, 92.39103698730469, -40.10845947265625, 34.81004333496094, 85.06582641601562, 153.80340576171875, 16.456470489501953, 36.005950927734375, 123.01541137695312, -106.25885009765625, 69.20790100097656, 4.1386260986328125, -14.402206420898438, 53.3077392578125, -122.79695892333984, -9.857431411743164, 154.5673828125, 28.825050354003906, 30.6309814453125, 11.6053466796875, 97.89012908935547, 33.474430084228516, 40.0811767578125, -3.506561279296875, 145.3120574951172, 80.36429595947266, 41.65282440185547, 281.3051452636719, 74.38561248779297], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000487.npy"}
|
|
{"epoch": 0.7362055933484505, "step": 488, "batch_size": 64, "mean": 56.83726119995117, "std": 72.02116394042969, "min": -117.85260772705078, "p10": -13.368602752685543, "median": 38.308176040649414, "p90": 162.49911804199218, "max": 196.816650390625, "pos_frac": 0.765625, "sample": [13.602325439453125, -2.1791458129882812, -3.516489028930664, -5.63770866394043, 9.914482116699219, -32.36759948730469, -21.350601196289062, 144.15928649902344, 157.1777801513672, 34.661556243896484, 23.666105270385742, 11.390892028808594, 162.61953735351562, 11.660202026367188, 4.459098815917969, 194.22329711914062, 8.314155578613281, 144.57180786132812, 2.5621795654296875, 54.41884994506836, 88.12030792236328, 34.30419158935547, 2.7914505004882812, -7.0865325927734375, -9.609825134277344, 41.954795837402344, -53.712432861328125, 165.87008666992188, 10.683258056640625, 162.2181396484375, 55.38482666015625, -48.96296691894531, -9.124486923217773, 151.71031188964844, 143.8901824951172, 58.698814392089844, 152.29185485839844, -19.97185516357422, 175.095458984375, 156.11961364746094, 154.6605987548828, 91.13948822021484, 98.58761596679688, 16.084457397460938, 77.96247100830078, 89.06433868408203, -9.265304565429688, 72.25534057617188, -14.979507446289062, 60.02415466308594, 42.11619567871094, 66.85083770751953, -117.85260772705078, 99.41180419921875, 16.86492919921875, 93.60272979736328, 196.816650390625, 56.38947677612305, 164.8948516845703, 30.39181137084961, 9.868698120117188, 21.643508911132812, -6.816215515136719, 164.8531494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000488.npy"}
|
|
{"epoch": 0.7377173091458806, "step": 489, "batch_size": 64, "mean": 33.52085876464844, "std": 76.8851547241211, "min": -137.12197875976562, "p10": -60.520115280151366, "median": 23.414491653442383, "p90": 136.40122833251954, "max": 198.91937255859375, "pos_frac": 0.625, "sample": [-122.45233154296875, 106.5186996459961, 28.555740356445312, 106.77843475341797, 103.75028991699219, 145.3624267578125, -26.92748260498047, 108.3805160522461, 45.25979232788086, 183.56005859375, 87.8897933959961, 14.741783142089844, 51.776336669921875, -81.88424682617188, -3.822174072265625, -38.40718078613281, -24.163612365722656, -18.721424102783203, 127.52947998046875, 48.885162353515625, 160.55230712890625, 92.10560607910156, 87.95307159423828, 10.976512908935547, 94.93489074707031, 126.18849182128906, -15.081184387207031, -32.01555633544922, -7.625823974609375, 24.864154815673828, -137.12197875976562, -60.67007064819336, 135.61106872558594, 21.28191375732422, -2.0709991455078125, -13.764152526855469, -19.471641540527344, -0.7323017120361328, 136.7398681640625, 83.24561309814453, 30.303739547729492, 58.883575439453125, 14.461174011230469, 198.91937255859375, 38.62195587158203, -83.46311950683594, 154.7493896484375, 9.005027770996094, 74.68287658691406, 69.15847778320312, -110.1360855102539, 8.207366943359375, -8.354259490966797, 3.639585494995117, -7.3777618408203125, -107.8338623046875, 116.89068603515625, 36.162376403808594, -60.17021942138672, 45.994171142578125, -32.16314697265625, 152.9437255859375, -8.264646530151367, 21.964828491210938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000489.npy"}
|
|
{"epoch": 0.7392290249433107, "step": 490, "batch_size": 64, "mean": 42.458702087402344, "std": 90.40245819091797, "min": -175.03292846679688, "p10": -71.49959869384764, "median": 42.784332275390625, "p90": 163.858447265625, "max": 199.73654174804688, "pos_frac": 0.640625, "sample": [-18.71613311767578, -21.002792358398438, -175.03292846679688, 22.71491241455078, -17.54872703552246, 67.81622314453125, 23.094680786132812, 137.8477020263672, 172.21817016601562, 199.73654174804688, -134.85562133789062, 163.64337158203125, -157.47103881835938, 9.124238967895508, 52.42124938964844, 80.04598236083984, 118.6043930053711, 66.00402069091797, 169.164794921875, 19.880043029785156, 8.306907653808594, 157.76693725585938, 131.416015625, 152.8338623046875, -74.61088562011719, -24.921661376953125, 155.26014709472656, -0.9868240356445312, 33.99055862426758, 30.09447479248047, -109.39132690429688, 70.86917114257812, 44.8165283203125, 87.7984848022461, -2.8999500274658203, -4.596832275390625, -18.444358825683594, 166.46612548828125, -79.28050231933594, -10.13565444946289, 75.0835189819336, -64.23992919921875, 64.09916687011719, 112.75023651123047, -2.525247573852539, 103.36227416992188, 17.037506103515625, -1.8483333587646484, -45.077369689941406, -13.128753662109375, 188.432861328125, 119.5657730102539, -7.969047546386719, 70.78141784667969, 185.75148010253906, 89.81256103515625, 163.95062255859375, 49.96465301513672, 51.28124237060547, 154.02130126953125, -150.61827087402344, -56.30735397338867, 40.75213623046875, 80.38404846191406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000490.npy"}
|
|
{"epoch": 0.7407407407407407, "step": 491, "batch_size": 64, "mean": 47.570377349853516, "std": 70.74996185302734, "min": -130.6225128173828, "p10": -28.95525512695312, "median": 28.616958618164062, "p90": 147.40860290527348, "max": 237.6094512939453, "pos_frac": 0.75, "sample": [101.93840026855469, -18.01844024658203, 98.91015625, 32.964508056640625, 136.11199951171875, 79.06314086914062, -1.135660171508789, 83.590576171875, 121.22515106201172, 23.311290740966797, 21.96905517578125, -86.6977310180664, 158.22406005859375, 25.1474609375, 139.57711791992188, -54.831642150878906, 9.36648178100586, 110.27273559570312, 8.595409393310547, 68.27951049804688, -42.733154296875, 84.26895141601562, 153.8122100830078, -0.4193611145019531, 70.3243637084961, 124.992919921875, 10.513862609863281, -11.503524780273438, 14.379081726074219, 13.258655548095703, 158.84573364257812, 83.06890869140625, -24.420059204101562, 19.22809600830078, 119.2416763305664, 44.99075698852539, 16.269737243652344, 237.6094512939453, 32.086456298828125, -30.898910522460938, 150.76495361328125, 7.607017517089844, 16.06366729736328, 0.5922527313232422, -0.3338508605957031, 60.45882034301758, -13.028717041015625, -53.79663848876953, 88.00421905517578, 156.43504333496094, 17.834205627441406, 112.75894165039062, 168.51580810546875, 134.5513916015625, 14.940393447875977, -8.010971069335938, 46.368309020996094, 47.79657745361328, 68.1180648803711, -130.6225128173828, -61.24504089355469, 5.230710983276367, -13.844013214111328, 98.56607055664062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000491.npy"}
|
|
{"epoch": 0.7422524565381708, "step": 492, "batch_size": 64, "mean": 37.207618713378906, "std": 80.87089538574219, "min": -157.011474609375, "p10": -45.20123443603516, "median": 20.964638710021973, "p90": 149.15524291992188, "max": 209.3871307373047, "pos_frac": 0.671875, "sample": [209.3871307373047, 134.60223388671875, -68.60243225097656, 7.499576568603516, 165.38064575195312, -3.4806461334228516, 97.77293395996094, 43.68584060668945, 72.28404235839844, -7.038980484008789, 99.83735656738281, 149.19268798828125, 37.874542236328125, -19.03618621826172, 17.43365478515625, 94.3370361328125, -157.011474609375, 46.97755432128906, 5.805908203125, 93.9034423828125, 122.38534545898438, -44.44171905517578, 0.990325927734375, -24.411842346191406, 8.159263610839844, 81.28982543945312, 72.84345245361328, 177.2581787109375, -41.14741516113281, -25.48073959350586, 155.9981689453125, 24.495622634887695, -32.083709716796875, 69.28602600097656, -7.195600509643555, 10.716598510742188, -138.58668518066406, 179.58013916015625, -70.6845703125, -13.150459289550781, 120.94929504394531, 6.611717224121094, 70.8822250366211, 4.370586395263672, -32.75439453125, 125.1030502319336, 44.122859954833984, 41.706695556640625, -45.52674102783203, 119.84749603271484, -138.76071166992188, -29.95568084716797, -6.3232421875, 116.45549011230469, 149.06787109375, 12.978740692138672, 100.96978759765625, 39.922393798828125, 86.93763732910156, 4.5345001220703125, 3.320394515991211, -75.73312377929688, -30.758567810058594, 166.69418334960938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000492.npy"}
|
|
{"epoch": 0.7437641723356009, "step": 493, "batch_size": 64, "mean": 57.51554489135742, "std": 75.61233520507812, "min": -145.76739501953125, "p10": -38.49077033996582, "median": 43.8641357421875, "p90": 160.98193664550783, "max": 185.51083374023438, "pos_frac": 0.796875, "sample": [8.60009765625, 10.546001434326172, -12.587982177734375, 129.46157836914062, -6.487373352050781, 41.2330322265625, -131.61334228515625, 26.714908599853516, 13.974580764770508, 1.834320068359375, 53.43414306640625, 152.32073974609375, 103.56369018554688, 170.5513916015625, 39.33047103881836, -37.50472640991211, 125.22817993164062, -0.36041259765625, 107.90029907226562, 141.11843872070312, 23.138994216918945, 23.235435485839844, -2.862119674682617, 51.15526580810547, 53.171531677246094, -43.424835205078125, 132.6815643310547, 15.457618713378906, 173.27601623535156, 104.6865005493164, 77.802734375, 145.06065368652344, -40.12837219238281, 6.662609100341797, -38.913360595703125, 98.7685317993164, 157.23492431640625, -39.52326965332031, 124.37091064453125, 5.02537727355957, 185.45248413085938, 100.62034606933594, 104.40553283691406, 185.51083374023438, 44.39362335205078, 173.49046325683594, -44.38008117675781, 144.9573516845703, 116.71837615966797, 0.45771026611328125, -6.416786193847656, 125.42711639404297, 43.33464813232422, 5.9501190185546875, 169.76742553710938, 94.89817810058594, 41.37074279785156, 30.188446044921875, 84.64102172851562, 162.58779907226562, 12.416725158691406, 40.23113250732422, 46.604164123535156, -145.76739501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000493.npy"}
|
|
{"epoch": 0.745275888133031, "step": 494, "batch_size": 64, "mean": 44.52660369873047, "std": 86.65257263183594, "min": -188.81820678710938, "p10": -63.66748199462889, "median": 44.403913497924805, "p90": 154.11524505615236, "max": 261.05230712890625, "pos_frac": 0.703125, "sample": [109.13824462890625, 22.353614807128906, 45.402671813964844, 25.684478759765625, 142.5380859375, 7.3559417724609375, -34.39372253417969, 121.24691772460938, 87.93476104736328, -68.5980224609375, 18.531326293945312, -52.53950500488281, -5.036163330078125, 113.30597686767578, 55.11764144897461, 58.58758544921875, 145.33355712890625, -0.30388641357421875, 44.7426643371582, -68.43661499023438, 9.187244415283203, 261.05230712890625, 90.56111145019531, 3.7484664916992188, 44.065162658691406, 155.79501342773438, 140.33815002441406, -2.8462142944335938, 99.28115844726562, 150.19578552246094, 104.62847137451172, -148.22897338867188, -162.95843505859375, 121.31266784667969, -10.736001968383789, -81.11711120605469, -70.89739227294922, 119.1729507446289, 17.74445343017578, 98.83064270019531, 40.42103576660156, 0.11181449890136719, -188.81820678710938, -3.010662078857422, 41.45608139038086, 54.715728759765625, 178.70594787597656, 8.721696853637695, -25.19794464111328, 24.2965087890625, 169.38198852539062, 47.94830322265625, 51.25080871582031, 201.93484497070312, 46.652931213378906, 79.00263214111328, -18.455413818359375, -16.096790313720703, 96.6217041015625, 91.52841186523438, 157.18704223632812, -23.84417724609375, -32.24637985229492, 160.33963012695312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000494.npy"}
|
|
{"epoch": 0.7467876039304611, "step": 495, "batch_size": 64, "mean": 39.45659255981445, "std": 77.54976654052734, "min": -115.13369750976562, "p10": -37.45547943115234, "median": 24.261802673339844, "p90": 152.22631988525396, "max": 269.70098876953125, "pos_frac": 0.65625, "sample": [-69.07881927490234, -33.563377380371094, -115.13369750976562, -113.03823852539062, 142.86746215820312, -8.794273376464844, -78.88339233398438, 24.725204467773438, 60.876834869384766, -23.5828857421875, -12.067977905273438, -89.34400177001953, -7.575408935546875, 58.6869010925293, 43.590309143066406, 160.14181518554688, -13.707695007324219, 39.0078125, 80.6443862915039, -1.632833480834961, 51.64071273803711, 80.32707977294922, 110.61726379394531, 1.859811782836914, 124.39693450927734, 156.2372589111328, -17.455890655517578, 78.86237335205078, 1.4639663696289062, 29.168262481689453, 269.70098876953125, 14.447595596313477, 12.558158874511719, -42.325984954833984, 25.761327743530273, 21.775672912597656, -1.633941650390625, 104.59759521484375, 141.45016479492188, 121.44464874267578, -36.136741638183594, 2.3399124145507812, 59.36990737915039, 80.58078002929688, 179.95298767089844, 0.5468902587890625, 55.062782287597656, -4.030731201171875, -4.0695343017578125, 161.53839111328125, -3.1472949981689453, 121.46157836914062, 45.964088439941406, -22.084510803222656, -38.020652770996094, 212.08810424804688, 29.13983154296875, -0.311187744140625, 23.79840087890625, 43.072357177734375, 76.88410949707031, 9.42459487915039, 192.9750518798828, 9.79056167602539], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000495.npy"}
|
|
{"epoch": 0.7482993197278912, "step": 496, "batch_size": 64, "mean": 49.96495056152344, "std": 88.66837310791016, "min": -109.01669311523438, "p10": -50.97002067565917, "median": 38.777732849121094, "p90": 159.96271667480468, "max": 413.0125732421875, "pos_frac": 0.6875, "sample": [94.54725646972656, 98.30105590820312, 176.2657470703125, 110.67823791503906, -69.5891342163086, 39.98387145996094, 58.20716857910156, 187.6943359375, 49.229331970214844, 97.78216552734375, -78.90391540527344, 18.301063537597656, 51.79509735107422, 45.76362609863281, 6.186811447143555, -7.61187744140625, -4.015033721923828, 33.111907958984375, 115.00230407714844, 32.34253692626953, -42.84960174560547, 413.0125732421875, -2.262798309326172, 158.81195068359375, 66.11772155761719, -109.01669311523438, 225.88601684570312, 136.40240478515625, -2.4247779846191406, 168.39703369140625, -54.132301330566406, 14.819721221923828, 160.45590209960938, 147.36465454101562, 0.1616668701171875, 6.287309646606445, -71.97602844238281, -14.309465408325195, 40.751220703125, 87.9347152709961, 39.407989501953125, 44.6002197265625, 155.19573974609375, 70.66876983642578, 94.51736450195312, 31.102022171020508, -35.25148010253906, 68.69573974609375, -11.543800354003906, 14.693721771240234, 180.19398498535156, -100.6637191772461, -31.221755981445312, 4.473958969116211, 120.40451049804688, 109.8809814453125, -5.576454162597656, 116.69799041748047, -43.591365814208984, 20.518814086914062, -57.95262145996094, 38.14747619628906, -6.6821441650390625, -3.4630966186523438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000496.npy"}
|
|
{"epoch": 0.7498110355253212, "step": 497, "batch_size": 64, "mean": 56.30817794799805, "std": 74.4541015625, "min": -130.28033447265625, "p10": -21.1615737915039, "median": 59.7045955657959, "p90": 145.16714782714845, "max": 204.01666259765625, "pos_frac": 0.75, "sample": [159.36212158203125, 39.009849548339844, 204.01666259765625, 139.2606201171875, 94.91067504882812, 0.47867584228515625, -16.409561157226562, -3.891204833984375, 16.671972274780273, 48.18286895751953, 120.03291320800781, 81.26675415039062, 101.49149322509766, 12.804336547851562, 124.79708862304688, 8.080070495605469, 104.29277038574219, -130.28033447265625, 142.6344757080078, 152.8953857421875, -2.6864776611328125, 1.8251533508300781, -2.4049453735351562, -5.599822998046875, 33.99665832519531, -26.247207641601562, -28.398590087890625, 132.9303741455078, 123.88101196289062, 149.65390014648438, -6.40873908996582, 67.56443786621094, 103.80426025390625, 130.6793212890625, -122.47295379638672, 127.74705505371094, 134.84274291992188, 81.79620361328125, 74.32785034179688, -9.409698486328125, 143.6222686767578, 72.86876678466797, -6.639591217041016, 67.71792602539062, 62.99464416503906, -23.198150634765625, 181.17071533203125, 124.41571807861328, 4.259765625, 23.844886779785156, 59.24375534057617, 43.67596435546875, 46.003204345703125, 91.67747497558594, 31.60138702392578, 185.67218017578125, -116.97628784179688, 4.849281311035156, -6.373981475830078, 145.82923889160156, 60.165435791015625, 46.75830841064453, -72.08799743652344, 73.60049438476562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000497.npy"}
|
|
{"epoch": 0.7513227513227513, "step": 498, "batch_size": 64, "mean": 55.8685188293457, "std": 68.13394165039062, "min": -167.03750610351562, "p10": -14.03450469970703, "median": 42.796579360961914, "p90": 148.81077880859374, "max": 198.68788146972656, "pos_frac": 0.796875, "sample": [-16.456771850585938, 76.59626770019531, 144.0979766845703, 56.988765716552734, 25.8095703125, -12.522758483886719, 124.97301483154297, 134.85479736328125, -10.271366119384766, -17.3631591796875, 122.69438171386719, 42.90287399291992, 57.986846923828125, -8.046287536621094, 111.00233459472656, 51.42811584472656, 8.384376525878906, -24.086782455444336, 29.626121520996094, 109.33048248291016, 198.68788146972656, 42.690284729003906, 148.1918487548828, 17.604190826416016, 40.16204833984375, 72.86604309082031, -27.239707946777344, 143.4916534423828, 126.1556396484375, 29.61041259765625, 149.07603454589844, 141.1016387939453, 1.3215293884277344, 37.22509765625, 52.45574188232422, -36.0035400390625, -4.449161529541016, 54.32887268066406, 2.2023448944091797, -2.993124008178711, 72.69963073730469, 160.4580841064453, -167.03750610351562, 91.23689270019531, 65.89503479003906, 38.833648681640625, -0.08111190795898438, 40.47148895263672, 24.100231170654297, 51.6803092956543, 169.16259765625, 166.49366760253906, 78.80950164794922, 17.076866149902344, 136.66197204589844, 7.865549087524414, 164.50267028808594, 24.896881103515625, 1.9570941925048828, 3.906951904296875, 43.26838684082031, -14.682395935058594, 18.50194549560547, 184.4922637939453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000498.npy"}
|
|
{"epoch": 0.7528344671201814, "step": 499, "batch_size": 64, "mean": 54.894493103027344, "std": 87.70376586914062, "min": -141.85400390625, "p10": -65.42530822753906, "median": 62.011056900024414, "p90": 158.49153747558594, "max": 255.91790771484375, "pos_frac": 0.75, "sample": [87.65193939208984, 151.44314575195312, -141.85400390625, 186.20806884765625, 15.842071533203125, 60.6742057800293, 43.84228515625, -113.68218231201172, 16.339946746826172, 70.23530578613281, 93.79203033447266, 81.2825927734375, 147.5308380126953, -92.84024047851562, 156.11257934570312, 3.0264511108398438, 255.91790771484375, -90.3846206665039, 165.39254760742188, 155.46694946289062, 56.48011779785156, -29.69635009765625, -25.78129005432129, -65.92935180664062, -42.419349670410156, 77.50906372070312, 52.65106201171875, 148.457763671875, -31.137113571166992, 82.68604278564453, -12.232791900634766, 14.683795928955078, 108.3994140625, 110.53117370605469, 59.81852340698242, -18.47559356689453, 20.197418212890625, 110.30713653564453, -2.873537063598633, 68.77604675292969, 176.90692138671875, 12.511955261230469, 158.36526489257812, 50.10481643676758, -127.27928161621094, 96.89522552490234, 63.34790802001953, 103.73824310302734, 123.99307250976562, 197.83372497558594, 158.545654296875, 5.988838195800781, 38.27305603027344, -64.24920654296875, -107.02900695800781, 130.54071044921875, 100.9129638671875, 87.01032257080078, -29.0460205078125, 161.7223358154297, 5.684173583984375, 148.67819213867188, 70.8938217163086, 14.9537353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000499.npy"}
|
|
{"epoch": 0.7543461829176115, "step": 500, "batch_size": 64, "mean": 49.16947937011719, "std": 79.02659606933594, "min": -165.552734375, "p10": -47.209840774536126, "median": 48.07783317565918, "p90": 150.79942474365234, "max": 177.01150512695312, "pos_frac": 0.765625, "sample": [-41.898223876953125, 79.37630462646484, 18.614395141601562, 7.457334518432617, 105.74591064453125, 48.81959533691406, 65.36408996582031, 117.42686462402344, 52.5980224609375, 163.8012237548828, 46.7808837890625, 160.0807647705078, 151.33255004882812, -34.560707092285156, 149.5554656982422, 43.10261535644531, 160.2162628173828, 51.549072265625, 36.47321319580078, -15.125, 167.93507385253906, 124.20654296875, 34.727378845214844, 108.19117736816406, -8.825546264648438, 153.23989868164062, 75.65301513671875, 12.50369644165039, -16.66417694091797, -96.8564682006836, -79.90184783935547, 47.3360710144043, 13.27597427368164, 29.704147338867188, -77.03065490722656, -5.210748672485352, 50.033111572265625, 146.5444793701172, 143.85208129882812, 3.9004898071289062, 92.64028930664062, 9.914693832397461, 118.50529479980469, 41.35991287231445, -49.48624801635742, 92.09138488769531, 38.478233337402344, -165.552734375, 144.97149658203125, -139.50669860839844, 177.01150512695312, -41.441253662109375, 2.521350860595703, 75.83565521240234, 99.94953918457031, 26.121601104736328, 10.025772094726562, 59.06071472167969, 103.57343292236328, -9.935701370239258, -94.68180084228516, 94.3910903930664, 148.3544464111328, 119.3202133178711], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000500.npy"}
|
|
{"epoch": 0.7558578987150416, "step": 501, "batch_size": 64, "mean": 45.30480194091797, "std": 57.435970306396484, "min": -134.5987548828125, "p10": -9.13991012573242, "median": 30.764796257019043, "p90": 136.1842239379883, "max": 155.1857452392578, "pos_frac": 0.765625, "sample": [23.82103729248047, 135.30303955078125, 8.627853393554688, 56.376529693603516, 67.01203918457031, 89.89360809326172, 15.846023559570312, 23.846961975097656, 10.675533294677734, 116.70047760009766, 136.56187438964844, 20.776792526245117, 81.17025756835938, 14.549118041992188, -0.6131343841552734, -1.475149154663086, 51.41594696044922, 149.10440063476562, -9.990455627441406, -6.698225021362305, 33.672706604003906, -2.3407135009765625, -2.8963050842285156, 153.49252319335938, 152.16363525390625, 121.60759735107422, 96.36994934082031, 153.23709106445312, 35.27810287475586, -12.36602783203125, 29.58495330810547, -15.130462646484375, -3.0470504760742188, 3.671527862548828, 49.89048767089844, 41.069580078125, 95.03070831298828, 4.942115783691406, 5.933895111083984, 47.42085266113281, 31.944639205932617, 132.12307739257812, 11.361682891845703, 50.947914123535156, -1.4666824340820312, -134.5987548828125, 25.59461212158203, 105.21226501464844, -42.9196662902832, 51.18602752685547, 52.70849609375, 66.83313751220703, 18.347084045410156, -11.475234985351562, 56.944091796875, -12.934625625610352, 19.845458984375, 22.533205032348633, 155.1857452392578, 146.9420928955078, 82.06301879882812, 17.099994659423828, 92.6954345703125, -7.155303955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000501.npy"}
|
|
{"epoch": 0.7573696145124716, "step": 502, "batch_size": 64, "mean": 47.52082061767578, "std": 64.30711364746094, "min": -80.91444396972656, "p10": -30.210643768310536, "median": 40.977928161621094, "p90": 134.10370483398438, "max": 189.75184631347656, "pos_frac": 0.765625, "sample": [-18.97425079345703, 18.67361831665039, 15.034332275390625, -34.38621520996094, 114.1756591796875, 94.54862976074219, 13.131355285644531, 14.718475341796875, -77.43488311767578, 80.82196044921875, 19.469547271728516, 159.67242431640625, -3.8918075561523438, -35.638282775878906, 90.72634887695312, 189.75184631347656, 22.7305908203125, -80.91444396972656, -7.568611145019531, 43.33628845214844, 24.361282348632812, -16.057750701904297, 58.37689971923828, 106.21475219726562, 108.02420043945312, 52.1439208984375, 123.21018981933594, -20.243091583251953, -0.06702232360839844, 38.61956787109375, 13.222915649414062, 9.701665878295898, 72.96875, 6.09309196472168, 61.49026107788086, 62.890655517578125, 161.28048706054688, 184.6704559326172, -41.35675048828125, 1.6002693176269531, -20.46764373779297, 29.156036376953125, 70.13834381103516, 3.800567626953125, 77.85234069824219, 3.4287796020507812, 27.893707275390625, 123.75016021728516, 97.42610931396484, -61.79359436035156, 75.42726135253906, 72.82449340820312, 121.0451889038086, 60.08184814453125, 135.85617065429688, -52.89239501953125, 52.047393798828125, 130.01461791992188, -6.38397216796875, 154.3125, 12.112491607666016, 89.54410552978516, 136.365234375, 84.66541290283203], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000502.npy"}
|
|
{"epoch": 0.7588813303099018, "step": 503, "batch_size": 64, "mean": 48.986656188964844, "std": 77.63191223144531, "min": -150.72610473632812, "p10": -33.12084903717041, "median": 33.07769775390625, "p90": 154.4410125732422, "max": 168.20416259765625, "pos_frac": 0.78125, "sample": [119.7143783569336, -2.4933509826660156, -24.904510498046875, 25.96598243713379, 90.0302963256836, 18.77654266357422, 165.617919921875, -150.72610473632812, 5.462074279785156, 89.20968627929688, 82.08267211914062, 9.422592163085938, 26.41162109375, 165.3754119873047, -33.73384094238281, 92.94084167480469, -31.690534591674805, 34.336395263671875, 164.5317840576172, -47.174781799316406, 13.213951110839844, 90.25692749023438, 146.3363494873047, 57.21136474609375, 26.11969757080078, 29.395370483398438, 111.23872375488281, 168.20416259765625, 13.786090850830078, 109.34420776367188, -14.155437469482422, 38.395050048828125, 159.15638732910156, 157.31985473632812, 156.07754516601562, -8.688995361328125, 71.25392150878906, -16.667091369628906, -92.90179443359375, 28.19568634033203, 2.4705467224121094, 31.819000244140625, 2.956605911254883, 71.27621459960938, -0.2502593994140625, 10.568893432617188, 46.81622314453125, 142.70921325683594, 107.49128723144531, 10.300304412841797, 145.76742553710938, 148.44432067871094, -127.94963073730469, 5.011730194091797, 150.6224365234375, 110.50094604492188, 5.13298225402832, -76.9057846069336, 87.24232482910156, 132.80296325683594, 16.348133087158203, 94.8729476928711, 103.2017593383789, -128.35166931152344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000503.npy"}
|
|
{"epoch": 0.7603930461073318, "step": 504, "batch_size": 64, "mean": 65.2579345703125, "std": 67.06529998779297, "min": -87.58032989501953, "p10": -4.353078460693359, "median": 59.495174407958984, "p90": 156.63138275146486, "max": 221.6778564453125, "pos_frac": 0.859375, "sample": [2.228363037109375, 151.8031005859375, -9.346633911132812, 95.73123168945312, 165.62013244628906, -10.913742065429688, 97.05392456054688, 41.485801696777344, 168.86041259765625, 4.27287483215332, 34.98388671875, 62.3436279296875, 20.945602416992188, 164.0339813232422, 41.93315505981445, 128.00167846679688, 131.7713165283203, -77.81417083740234, 85.2432861328125, 127.88895416259766, 1.6782798767089844, 70.0379638671875, 2.1427841186523438, 138.92845153808594, 158.56263732910156, -23.3668212890625, 59.57202911376953, 152.1251220703125, 59.41831970214844, 221.6778564453125, 151.48590087890625, 160.80734252929688, -4.168048858642578, 58.94450759887695, 60.20269775390625, 3.1571807861328125, 4.344635009765625, -80.82250213623047, 138.59335327148438, 54.07875061035156, 131.08013916015625, -3.7853546142578125, -4.432376861572266, 29.476226806640625, 21.244049072265625, 146.16073608398438, 75.76699829101562, 57.86438751220703, 41.909149169921875, 71.55616760253906, 107.30621337890625, 43.99335861206055, -87.58032989501953, 83.26507568359375, 101.46702575683594, 23.375267028808594, 20.165693283081055, 69.69790649414062, 98.76631164550781, 22.312471389770508, 17.877206802368164, 177.2161865234375, 72.65393829345703, 45.62450408935547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000504.npy"}
|
|
{"epoch": 0.7619047619047619, "step": 505, "batch_size": 64, "mean": 41.974151611328125, "std": 84.10599517822266, "min": -145.0826416015625, "p10": -56.11108474731445, "median": 25.673725128173828, "p90": 158.3820510864258, "max": 216.61370849609375, "pos_frac": 0.65625, "sample": [88.04910278320312, 2.2693824768066406, 60.48057556152344, 29.085399627685547, -0.15822982788085938, 216.61370849609375, -107.79104614257812, 201.0027313232422, 123.50929260253906, 153.41690063476562, 159.60598754882812, -14.888580322265625, 26.540382385253906, 5.13916015625, 19.8477783203125, -15.0775146484375, -1.5279998779296875, -5.992420196533203, 92.86397552490234, 11.265316009521484, -70.33599853515625, 131.37655639648438, 116.47319793701172, 29.82797622680664, 77.34229278564453, 193.92694091796875, 49.82994842529297, -42.00062561035156, 33.00584411621094, -0.5872688293457031, -25.441848754882812, -15.890357971191406, 150.5819091796875, -94.61095428466797, 176.2227325439453, 123.86437225341797, 91.91985321044922, 120.89896392822266, -1.0558300018310547, -49.76539611816406, 24.80706787109375, 14.426307678222656, -2.6670455932617188, -126.93912506103516, 33.355133056640625, 38.34407043457031, 62.17987823486328, -22.991960525512695, 172.9038848876953, -96.3743896484375, -19.451454162597656, 129.25146484375, 155.5261993408203, 38.049285888671875, 148.18572998046875, 19.366592407226562, 99.5506591796875, -8.605926513671875, 8.349105834960938, 162.6463623046875, 20.392454147338867, 0.11867523193359375, -58.830665588378906, -145.0826416015625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000505.npy"}
|
|
{"epoch": 0.763416477702192, "step": 506, "batch_size": 64, "mean": 46.02772521972656, "std": 80.12324523925781, "min": -164.31494140625, "p10": -45.785075378417964, "median": 44.7231330871582, "p90": 146.82628936767583, "max": 204.7708740234375, "pos_frac": 0.75, "sample": [57.340126037597656, 138.57302856445312, 3.1235694885253906, -20.93756866455078, 102.28440856933594, 117.21678161621094, 65.05390930175781, -29.200328826904297, 116.22125244140625, -53.580108642578125, 2.098905563354492, 73.57075500488281, 31.12877655029297, 34.408973693847656, 52.189945220947266, 86.70036315917969, -144.47305297851562, 184.7942657470703, 204.7708740234375, 9.659004211425781, 137.23678588867188, -72.68022918701172, -6.4163665771484375, -164.31494140625, -75.98869323730469, -147.40713500976562, 17.481521606445312, 165.72265625, 175.44924926757812, 136.8265380859375, 164.1080322265625, -40.36363220214844, 195.21798706054688, -14.106475830078125, 150.10877990722656, 22.99578094482422, 139.16714477539062, 39.81159210205078, 85.40192413330078, -34.17573547363281, -48.108551025390625, 4.992698669433594, 8.865623474121094, 21.557861328125, 91.60836791992188, 57.5541877746582, 3.1707191467285156, 36.926963806152344, 82.334716796875, 70.7402114868164, 70.93363952636719, 89.65133666992188, -0.8955154418945312, 103.63253021240234, 73.57195281982422, 4.031059265136719, 9.906761169433594, 91.61492156982422, -2.96246337890625, -8.268451690673828, 47.83525848388672, 96.81793212890625, 93.63313293457031, 41.61100769042969], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000506.npy"}
|
|
{"epoch": 0.764928193499622, "step": 507, "batch_size": 64, "mean": 49.553489685058594, "std": 83.49346160888672, "min": -117.30718994140625, "p10": -48.16331558227539, "median": 41.397499084472656, "p90": 166.61209411621095, "max": 263.39935302734375, "pos_frac": 0.6875, "sample": [119.92481231689453, -12.683319091796875, -18.420127868652344, 64.20834350585938, 263.39935302734375, 1.1324996948242188, 190.37664794921875, -117.30718994140625, -58.36079406738281, 78.70185852050781, 42.84297180175781, 40.24891662597656, -46.243896484375, 124.66474914550781, -31.6190185546875, 10.178955078125, 44.55696105957031, -25.012474060058594, 148.97518920898438, 19.079265594482422, 42.54608154296875, 44.99421691894531, 134.32388305664062, -33.19197082519531, 163.78985595703125, -3.197378158569336, 45.502769470214844, 167.82162475585938, 60.85477828979492, 4.62396240234375, 34.142578125, 19.97796630859375, 214.16091918945312, -15.063560485839844, 136.43963623046875, 79.75102996826172, -29.384185791015625, 133.43516540527344, -54.62684631347656, -9.25823974609375, -23.692398071289062, 42.895111083984375, -1.0106773376464844, 103.77095031738281, 97.68429565429688, 16.262863159179688, 49.773048400878906, 10.755569458007812, 235.16668701171875, 188.98239135742188, -61.798675537109375, 64.0645751953125, -48.985923767089844, 6.561866760253906, 89.0975341796875, -8.013298034667969, 137.34353637695312, -60.25767517089844, 39.370201110839844, -99.94508361816406, 18.009002685546875, 178.63772583007812, 64.20399475097656, 156.26162719726562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000507.npy"}
|
|
{"epoch": 0.7664399092970522, "step": 508, "batch_size": 64, "mean": 58.004920959472656, "std": 76.2405014038086, "min": -123.99542236328125, "p10": -24.979859161376954, "median": 49.72257995605469, "p90": 160.2904235839844, "max": 200.4735565185547, "pos_frac": 0.78125, "sample": [115.06745910644531, 10.04779052734375, 110.08282470703125, 32.37434387207031, 23.747604370117188, -36.742431640625, 195.8756103515625, -8.271148681640625, 103.74373626708984, 151.3878936767578, 200.4735565185547, 141.9834442138672, 23.296600341796875, 17.496047973632812, 31.520843505859375, 105.72844696044922, -123.99542236328125, 57.0628662109375, 61.764434814453125, -31.412616729736328, 39.23199462890625, 28.983253479003906, 115.10167694091797, -25.270957946777344, -29.73613739013672, -109.35581970214844, 60.00424575805664, 2.236612319946289, 189.80877685546875, 91.1041259765625, -13.131607055664062, 186.41273498535156, 58.41227722167969, 60.55689239501953, 104.70623016357422, 72.75991821289062, 62.269283294677734, 58.60565185546875, 158.77737426757812, 5.406715393066406, 0.6950454711914062, -23.42707061767578, 148.3922119140625, 143.25550842285156, -16.96148681640625, -22.273056030273438, 101.37863159179688, 151.2027587890625, 160.93887329101562, 12.146455764770508, 146.93185424804688, 1.0339279174804688, 31.012500762939453, -9.488279342651367, 27.723129272460938, -81.4616928100586, -24.300628662109375, 162.11502075195312, 131.6678466796875, 119.00320434570312, 42.382293701171875, 34.01727294921875, 12.191741943359375, 166.02389526367188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000508.npy"}
|
|
{"epoch": 0.7679516250944822, "step": 509, "batch_size": 64, "mean": 44.94239044189453, "std": 71.61153411865234, "min": -121.38874816894531, "p10": -57.33029136657714, "median": 48.22373390197754, "p90": 135.69960174560546, "max": 173.61410522460938, "pos_frac": 0.734375, "sample": [81.10821533203125, -18.62560272216797, 49.97066116333008, 133.63201904296875, 33.05891418457031, -0.33017539978027344, 14.848960876464844, -14.615716934204102, -82.69532012939453, 55.005271911621094, 18.515235900878906, 70.45004272460938, -69.41891479492188, 49.070194244384766, 121.31403350830078, -121.38874816894531, 101.50555419921875, 94.03002166748047, 0.3705596923828125, 18.383529663085938, 105.74285888671875, -59.66718292236328, 64.78985595703125, -25.49774169921875, 166.27133178710938, 52.91993713378906, -16.216957092285156, 12.059673309326172, 26.981109619140625, -3.28436279296875, 126.02548217773438, 46.09708023071289, 113.77825927734375, 97.05073547363281, 37.209381103515625, -1.9099864959716797, 138.425048828125, 90.30264282226562, 75.1427001953125, 151.62881469726562, 23.46465301513672, 1.2130279541015625, 151.79745483398438, -51.87754440307617, 104.19718933105469, 133.7310333251953, 42.56535339355469, 47.37727355957031, 173.61410522460938, 143.66738891601562, 136.54327392578125, -16.80376434326172, -28.82269287109375, -114.50839233398438, 9.883132934570312, 122.8336181640625, 81.5362548828125, -92.17373657226562, -68.98979949951172, 53.037818908691406, 22.386951446533203, 51.051124572753906, 129.40135192871094, 89.15058898925781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000509.npy"}
|
|
{"epoch": 0.7694633408919124, "step": 510, "batch_size": 64, "mean": 55.06531524658203, "std": 83.48746490478516, "min": -117.1601791381836, "p10": -47.16330413818359, "median": 40.52160453796387, "p90": 181.90606079101562, "max": 211.20181274414062, "pos_frac": 0.78125, "sample": [38.37928771972656, 135.21209716796875, 127.46697235107422, 80.40107727050781, -20.049102783203125, 0.975799560546875, 16.29043197631836, 186.47549438476562, 3.7472152709960938, 8.25189208984375, -18.47014617919922, 17.70421600341797, 21.980770111083984, 37.47395324707031, -109.25822448730469, -9.810234069824219, 8.440826416015625, 39.33073425292969, -10.111495971679688, 27.999696731567383, 57.058349609375, 139.5974578857422, 165.31455993652344, -67.70404815673828, 6.250831604003906, 140.15133666992188, 52.675682067871094, -12.44158935546875, 0.6910133361816406, 119.63436889648438, 66.97671508789062, 86.3956298828125, 207.35452270507812, 73.8666763305664, 20.622455596923828, 108.84912109375, 71.4901351928711, 197.98629760742188, 34.217437744140625, -49.017333984375, 16.05134391784668, -102.66194915771484, 137.93246459960938, 73.00126647949219, 181.90802001953125, 20.39002227783203, 91.46576690673828, 193.95346069335938, 211.20181274414062, -42.83723449707031, 52.231101989746094, 41.71247482299805, 149.96942138671875, 104.43482971191406, -86.30397033691406, 11.803627014160156, 153.525634765625, 181.9014892578125, -6.116874694824219, 72.67311096191406, 203.85537719726562, 56.20805358886719, -77.35980987548828, -117.1601791381836], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000510.npy"}
|
|
{"epoch": 0.7709750566893424, "step": 511, "batch_size": 64, "mean": 62.12137985229492, "std": 80.21784210205078, "min": -110.12503051757812, "p10": -36.70190277099608, "median": 45.69560241699219, "p90": 164.8777267456055, "max": 188.7430419921875, "pos_frac": 0.796875, "sample": [175.04364013671875, 153.38401794433594, 62.3182373046875, 162.96343994140625, 12.698856353759766, -1.6021041870117188, 25.806705474853516, 4.677513122558594, 134.9859161376953, 159.80789184570312, 45.935394287109375, -67.0915298461914, 17.642410278320312, 40.28546905517578, 153.74374389648438, 165.69813537597656, 121.80924987792969, -20.215843200683594, 133.18902587890625, 82.62649536132812, 150.39988708496094, -42.815460205078125, 33.43164825439453, -100.18614959716797, -103.95591735839844, -17.620880126953125, 174.60205078125, 104.7176513671875, 74.81720733642578, 37.91534423828125, 18.895662307739258, 94.0941162109375, 44.7515869140625, 45.455810546875, 131.42678833007812, 94.42155456542969, 140.60035705566406, 23.257083892822266, -22.436935424804688, 15.90119743347168, 33.13096618652344, 67.52310180664062, 124.93795776367188, 183.57363891601562, 184.65322875976562, 117.67928314208984, -86.52508544921875, 27.638015747070312, 3.2067394256591797, 159.22434997558594, 36.11621856689453, 1.8052215576171875, 104.42767333984375, 99.3763198852539, 11.312553405761719, 38.125221252441406, -66.08619689941406, 188.7430419921875, 147.36427307128906, 105.19996643066406, 169.9610595703125, -8.260334014892578, -110.12503051757812, -18.61312484741211], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000511.npy"}
|
|
{"epoch": 0.7724867724867724, "step": 512, "batch_size": 64, "mean": 44.54682922363281, "std": 78.15986633300781, "min": -154.74639892578125, "p10": -50.927678680419916, "median": 22.971195220947266, "p90": 144.99405975341796, "max": 177.42327880859375, "pos_frac": 0.703125, "sample": [120.43416595458984, 51.325408935546875, 42.86552429199219, -97.49041748046875, -52.27853012084961, 158.9426727294922, 13.741680145263672, 19.356054306030273, 21.134925842285156, 105.85818481445312, 119.91567993164062, 160.884521484375, -154.74639892578125, -47.775691986083984, 39.00437927246094, -9.656097412109375, -6.603233337402344, -6.649875640869141, 135.6182861328125, -1.0791378021240234, 164.46395874023438, 134.29795837402344, 8.920875549316406, 114.70360565185547, -5.6951141357421875, 129.010498046875, -83.47897338867188, 126.9412612915039, 2.5396041870117188, 143.6896209716797, 17.070049285888672, 107.49356079101562, 121.91279602050781, 4.648260116577148, 109.62261962890625, 15.75946044921875, 4.4338531494140625, 85.83918762207031, -7.816656112670898, 145.0126953125, -16.005722045898438, -104.03531646728516, -5.07078742980957, -21.770355224609375, 61.73603820800781, 154.1455535888672, 20.523475646972656, 20.71390151977539, 24.807464599609375, 3.503978729248047, 144.95057678222656, -1.1946487426757812, 17.1871337890625, -105.53274536132812, 47.24919128417969, 97.59689331054688, 177.42327880859375, 87.54794311523438, 159.30789184570312, 122.8576431274414, 46.28809356689453, -35.002540588378906, 66.6458740234375, -65.04706573486328], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000512.npy"}
|
|
{"epoch": 0.7739984882842026, "step": 513, "batch_size": 64, "mean": 51.648468017578125, "std": 68.45352935791016, "min": -111.23851013183594, "p10": -13.512488746643063, "median": 36.70175361633301, "p90": 166.34791564941406, "max": 226.888916015625, "pos_frac": 0.8125, "sample": [206.58050537109375, 92.09003448486328, 156.7277374267578, 103.51725769042969, 76.73839569091797, 56.87814712524414, 78.11113739013672, 11.80782699584961, 1.8001289367675781, -17.905662536621094, 50.99571990966797, -79.33125305175781, 129.8011016845703, 62.172332763671875, 48.443912506103516, 80.67753601074219, -6.712184906005859, 16.186180114746094, 167.5478057861328, 28.48143768310547, 44.02757263183594, 172.4619903564453, 91.46715545654297, 45.34975814819336, 168.37164306640625, 10.067626953125, -8.845321655273438, 81.38656616210938, 25.61870765686035, 70.6789779663086, 30.071800231933594, 32.04549026489258, 93.7750473022461, -111.23851013183594, 41.636009216308594, 18.66875648498535, -31.749839782714844, 88.56631469726562, 226.888916015625, 13.055530548095703, -1.8612022399902344, -14.97303581237793, 187.49386596679688, 0.9562530517578125, 108.19503784179688, 163.5481719970703, 177.66632080078125, -10.104545593261719, 92.65069580078125, 69.32772827148438, 30.528831481933594, 6.4845428466796875, 41.35801696777344, 26.624069213867188, -1.5633544921875, -37.75745391845703, 22.289024353027344, 10.946582794189453, 12.104068756103516, 15.250726699829102, 69.43765258789062, -50.121910095214844, 8.579635620117188, 11.530029296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000513.npy"}
|
|
{"epoch": 0.7755102040816326, "step": 514, "batch_size": 64, "mean": 59.078712463378906, "std": 75.47417449951172, "min": -151.232666015625, "p10": -29.278196525573726, "median": 68.00643539428711, "p90": 153.01130523681644, "max": 257.14471435546875, "pos_frac": 0.75, "sample": [7.07063102722168, 61.25749588012695, 81.85282897949219, 92.36933898925781, 95.8067626953125, 65.76895904541016, 124.2061767578125, -12.998397827148438, 34.397918701171875, -10.561145782470703, 53.658267974853516, 155.45904541015625, 78.71945190429688, 68.78567504882812, -3.7528629302978516, 57.48969268798828, -82.64971923828125, 88.21183776855469, 7.306488037109375, 67.2271957397461, -151.232666015625, 1.6124191284179688, 147.29991149902344, 120.8026351928711, 169.4693603515625, -12.352981567382812, 29.159149169921875, 124.6953125, 106.84933471679688, 170.54269409179688, -37.266937255859375, -31.41444969177246, 133.82382202148438, 121.16014862060547, 17.973060607910156, 73.20056915283203, 71.54006958007812, 132.31170654296875, -4.68450927734375, 80.0262451171875, 43.095054626464844, -41.684326171875, -24.29360580444336, 10.518096923828125, 96.98213195800781, 73.11620330810547, 101.09324645996094, -11.868576049804688, 19.280899047851562, 75.47528839111328, -20.72272491455078, 34.77241134643555, 10.931495666503906, 130.91372680664062, -88.6827392578125, 110.20299530029297, 162.3974151611328, 112.71450805664062, 188.61328125, -2.608224868774414, 257.14471435546875, 158.87503051757812, -51.13429260253906, 142.76483154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000514.npy"}
|
|
{"epoch": 0.7770219198790628, "step": 515, "batch_size": 64, "mean": 50.13121795654297, "std": 73.22516632080078, "min": -121.35375213623047, "p10": -40.620829010009764, "median": 43.04071617126465, "p90": 153.35699768066408, "max": 179.13282775878906, "pos_frac": 0.765625, "sample": [3.273181915283203, 83.18589782714844, 44.704872131347656, 46.893348693847656, 179.13282775878906, 129.30113220214844, 18.855464935302734, 148.87928771972656, 8.170875549316406, 143.79556274414062, 42.10541534423828, 51.77672576904297, 80.26317596435547, -19.659914016723633, 79.6100082397461, 21.8546142578125, -59.43510437011719, 159.62535095214844, -72.14788055419922, -37.5137939453125, 69.5899429321289, -2.1886978149414062, 0.955657958984375, 10.892669677734375, -121.35375213623047, 114.51077270507812, 162.14088439941406, 71.71873474121094, 8.338417053222656, -16.513084411621094, 75.29694366455078, 108.12767028808594, 4.935047149658203, 12.116250991821289, 36.60688018798828, 137.75550842285156, -78.58282470703125, 113.54429626464844, -6.526769638061523, 99.64122009277344, -42.5022087097168, 43.976016998291016, 176.16725158691406, -89.14757537841797, 40.08540344238281, 153.95335388183594, 16.317522048950195, -25.21071434020996, 81.76900482177734, 148.68814086914062, 40.06079864501953, 90.08536529541016, 75.740478515625, 171.03440856933594, 173.46641540527344, -24.84756088256836, 101.2782211303711, 6.118656158447266, 151.9654998779297, -41.952415466308594, 28.73265838623047, 80.53408813476562, -2.5038318634033203, 0.9124374389648438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000515.npy"}
|
|
{"epoch": 0.7785336356764928, "step": 516, "batch_size": 64, "mean": 42.742279052734375, "std": 76.42302703857422, "min": -165.46981811523438, "p10": -45.546193695068354, "median": 41.33094787597656, "p90": 141.0275405883789, "max": 178.49913024902344, "pos_frac": 0.734375, "sample": [144.9146270751953, 172.4046630859375, 97.03878784179688, -55.897918701171875, -46.75421142578125, 25.428241729736328, -28.560646057128906, 88.95285034179688, 123.5245361328125, 97.07699584960938, 37.58264923095703, 41.19134521484375, 93.50592041015625, 178.49913024902344, 18.088375091552734, 2.168344497680664, 74.08282470703125, 117.15731811523438, 160.76600646972656, 86.34815979003906, -42.72748565673828, -154.61483764648438, 47.843597412109375, 86.3102798461914, 148.9910888671875, 59.43431091308594, 100.89537048339844, 9.925224304199219, 125.93841552734375, 95.48003387451172, -165.46981811523438, -4.15362548828125, 1.0098552703857422, 1.2058486938476562, -5.067604064941406, 142.84246826171875, 107.68378448486328, -6.051263809204102, 3.230510711669922, -11.87874984741211, 84.6038818359375, -123.59601593017578, 81.87554931640625, 148.22161865234375, 41.470550537109375, 28.473602294921875, 115.98887634277344, 23.945194244384766, 40.073516845703125, 61.954620361328125, 5.4576263427734375, 81.20818328857422, 11.338539123535156, -29.10375213623047, -117.1224365234375, 92.155029296875, -30.784404754638672, 79.29840087890625, -6.698398590087891, 16.91378402709961, 79.69410705566406, -0.0128021240234375, -54.98741912841797, 136.79270935058594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000516.npy"}
|
|
{"epoch": 0.780045351473923, "step": 517, "batch_size": 64, "mean": 36.760196685791016, "std": 70.7247085571289, "min": -86.89962768554688, "p10": -46.70819931030273, "median": 22.597171783447266, "p90": 146.52850341796878, "max": 203.88055419921875, "pos_frac": 0.65625, "sample": [21.04824447631836, -9.215606689453125, -12.900333404541016, 76.28843688964844, -38.975059509277344, 29.073932647705078, -37.92778015136719, 49.63017272949219, 173.003173828125, 27.640430450439453, 20.651016235351562, -18.114925384521484, -45.518890380859375, 15.2891845703125, 19.48766326904297, -32.330101013183594, -0.6020374298095703, 5.837928771972656, 56.83174133300781, -0.6390151977539062, 93.26850891113281, -83.4658203125, 164.54739379882812, 118.62068939208984, -59.154876708984375, 41.54890441894531, -5.749298095703125, 68.9688491821289, 29.356903076171875, 115.87615966796875, 51.01856231689453, 142.7269744873047, 16.990083694458008, 37.83283233642578, -28.4300537109375, -68.1494369506836, 11.006156921386719, 119.77018737792969, -13.799066543579102, -47.21790313720703, 58.61273956298828, 148.15773010253906, 203.88055419921875, 24.146099090576172, 70.6629638671875, 125.2413101196289, 20.24020767211914, 60.20090103149414, -72.44706726074219, -15.679389953613281, -51.73243713378906, 75.71553039550781, 8.43585205078125, 160.8551025390625, 190.32447814941406, 148.93463134765625, 40.21369934082031, -13.219375610351562, 25.388809204101562, 17.287113189697266, 140.99295043945312, -86.89962768554688, 80.28695678710938, -11.07098388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000517.npy"}
|
|
{"epoch": 0.781557067271353, "step": 518, "batch_size": 64, "mean": 44.792510986328125, "std": 71.79142761230469, "min": -108.86726379394531, "p10": -48.19257049560546, "median": 30.45583724975586, "p90": 141.56781921386724, "max": 216.56752014160156, "pos_frac": 0.703125, "sample": [73.72683715820312, 0.57080078125, -1.4153366088867188, 158.35537719726562, -11.869598388671875, 3.9589691162109375, 13.779396057128906, 95.44273376464844, -62.64436721801758, 33.995948791503906, 41.86332321166992, -52.77423095703125, 42.89440155029297, 84.87089538574219, 105.99776458740234, -51.79978942871094, 54.48597717285156, 110.98595428466797, -63.05316925048828, 127.15203857421875, 173.6429443359375, -108.86726379394531, 3.8736114501953125, 63.68693161010742, 128.1463165283203, 85.51188659667969, -58.52984619140625, 95.61955261230469, -39.775726318359375, 117.22762298583984, 1.4225044250488281, -32.951637268066406, -0.6222324371337891, 79.406494140625, 41.62602233886719, -3.80621337890625, 11.241474151611328, -12.842376708984375, 43.70658874511719, 216.56752014160156, 26.915725708007812, 26.293495178222656, 2.7282962799072266, 170.90850830078125, 112.88392639160156, 173.093017578125, -31.936431884765625, 129.94845581054688, 146.54754638671875, -53.14025115966797, -10.124940872192383, -5.447122573852539, 99.59765625, 0.7221794128417969, 24.812583923339844, 121.32196807861328, 89.11441040039062, 45.8480224609375, -1.1524829864501953, -7.6299896240234375, 17.45130157470703, 4.648618698120117, 188.74002075195312, 85.76824951171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000518.npy"}
|
|
{"epoch": 0.783068783068783, "step": 519, "batch_size": 64, "mean": 28.961193084716797, "std": 90.16047668457031, "min": -225.97714233398438, "p10": -85.6501365661621, "median": 21.147628784179688, "p90": 149.34280090332032, "max": 189.94061279296875, "pos_frac": 0.625, "sample": [-89.44860076904297, 148.24612426757812, 40.587371826171875, -160.9628448486328, -40.33601760864258, 133.4735107421875, -4.938846588134766, 189.94061279296875, 19.429733276367188, 116.93966674804688, -8.252593994140625, 13.990537643432617, 111.01892852783203, -11.369132995605469, 48.010826110839844, 59.346595764160156, 146.98321533203125, 19.617280960083008, -126.19061279296875, -31.17413330078125, 136.26109313964844, 111.01322937011719, -123.41316223144531, 122.73500061035156, 104.12409973144531, 163.67263793945312, -50.53459548950195, 20.306488037109375, 70.00431823730469, 93.2850341796875, -32.648582458496094, 134.24139404296875, 19.707260131835938, -54.72013854980469, 62.94291687011719, -58.24127197265625, -25.104293823242188, 4.925380706787109, 172.14634704589844, -59.7227783203125, 50.974853515625, 7.0753021240234375, 101.06398010253906, 50.32982635498047, 33.435142517089844, -225.97714233398438, 40.68244934082031, -19.663619995117188, 42.055755615234375, -86.48224639892578, 149.81280517578125, -28.538536071777344, -84.96664428710938, 174.87802124023438, 26.661331176757812, 21.98876953125, 153.07154846191406, 89.4245376586914, 1.1791572570800781, -85.94306182861328, -36.70873260498047, -6.679834365844727, 159.47430419921875, -59.52364730834961], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000519.npy"}
|
|
{"epoch": 0.7845804988662132, "step": 520, "batch_size": 64, "mean": 41.260108947753906, "std": 70.55509948730469, "min": -96.3083267211914, "p10": -43.07108154296874, "median": 32.978370666503906, "p90": 143.08482971191407, "max": 225.21072387695312, "pos_frac": 0.78125, "sample": [104.81050109863281, -13.716623306274414, -2.7499656677246094, -2.124225616455078, 144.12237548828125, 0.508056640625, 68.7552490234375, 5.871818542480469, 3.2818946838378906, 225.21072387695312, 122.01661682128906, 27.60178565979004, 5.2064666748046875, -36.80335998535156, 56.0091552734375, 43.224815368652344, 3.3282623291015625, -45.75724792480469, 6.819316864013672, 144.088623046875, 140.74264526367188, 58.7150764465332, 42.436927795410156, 159.38426208496094, 60.74420166015625, 58.643470764160156, 59.51393127441406, -90.68672180175781, -96.3083267211914, 111.89897155761719, 72.20991516113281, 4.716453552246094, 14.017005920410156, 24.030471801757812, 2.0867462158203125, 0.1112518310546875, 159.0886688232422, 77.6822738647461, -5.069915771484375, -79.84803009033203, -93.82186889648438, 29.291030883789062, 36.66571044921875, -87.86811828613281, 51.62083435058594, 37.93035888671875, 25.930801391601562, 76.02020263671875, 72.31810760498047, 25.712661743164062, 44.90556335449219, 163.33258056640625, 126.08403015136719, 26.096412658691406, 195.7630157470703, 37.140052795410156, 60.501617431640625, 7.854728698730469, 0.8845787048339844, -1.1747112274169922, -3.4467544555664062, 121.50860595703125, 127.1745834350586, -73.59077453613281], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000520.npy"}
|
|
{"epoch": 0.7860922146636432, "step": 521, "batch_size": 64, "mean": 58.189048767089844, "std": 69.40008544921875, "min": -135.50033569335938, "p10": -11.73952178955078, "median": 55.53245162963867, "p90": 144.88397369384765, "max": 248.47952270507812, "pos_frac": 0.765625, "sample": [117.30506134033203, 34.316192626953125, 127.56602478027344, 43.699554443359375, 143.74842834472656, 66.90258026123047, 145.37063598632812, -2.4697837829589844, -4.5728759765625, 61.72418212890625, 77.53463745117188, -9.077686309814453, 109.22607421875, 54.203880310058594, 106.24468231201172, 94.57328033447266, 38.86570739746094, 25.14337158203125, -1.0796356201171875, 154.05560302734375, 0.5696640014648438, 120.20697784423828, 21.623748779296875, 10.0296630859375, -7.239809036254883, 100.79187774658203, -67.06660461425781, 129.87167358398438, -3.8759918212890625, 105.974609375, 189.02525329589844, 248.47952270507812, -75.09329986572266, 18.044361114501953, -1.8840408325195312, -14.597923278808594, 9.852592468261719, 76.83323669433594, 3.644866943359375, 167.3463134765625, 162.89109802246094, 9.34140396118164, -49.864768981933594, 70.58135223388672, -15.047849655151367, 28.703054428100586, 121.20433044433594, 73.3289794921875, 27.967388153076172, 105.73915100097656, 94.83097839355469, 56.86102294921875, -135.50033569335938, 179.80093383789062, 78.33255004882812, 53.918174743652344, 50.06079864501953, -12.008087158203125, 79.5723876953125, -11.112869262695312, 80.53074645996094, 47.88597869873047, 108.1646728515625, 102.10174560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000521.npy"}
|
|
{"epoch": 0.7876039304610734, "step": 522, "batch_size": 64, "mean": 41.04133605957031, "std": 89.3820571899414, "min": -186.3199462890625, "p10": -79.31229476928709, "median": 33.19508743286133, "p90": 170.97488708496095, "max": 214.4505157470703, "pos_frac": 0.703125, "sample": [-44.44769287109375, 72.84989166259766, 5.39442253112793, 52.958335876464844, 77.43854522705078, 143.54248046875, 27.66754913330078, 177.85675048828125, 132.108642578125, 42.98484802246094, -128.91590881347656, -89.97945404052734, 43.17759704589844, 172.38040161132812, 31.693206787109375, 131.73928833007812, -56.42748260498047, 55.89128494262695, 107.60487365722656, 27.756025314331055, 169.39134216308594, 11.39688491821289, 45.93944549560547, -186.3199462890625, 193.79501342773438, 4.245536804199219, 4.931125640869141, -52.00029754638672, -129.46502685546875, 9.413240432739258, -14.251945495605469, 158.9073486328125, -16.945327758789062, 171.65354919433594, 103.57333374023438, 60.326934814453125, -6.162933349609375, 44.999359130859375, -20.440135955810547, 34.69696807861328, 65.86368560791016, 1.9768142700195312, 7.250724792480469, -11.760963439941406, 81.8369140625, -1.515899658203125, 124.80203247070312, -139.21701049804688, 175.31390380859375, -33.11402893066406, 5.431587219238281, -93.16767883300781, 116.3966293334961, 214.4505157470703, -0.511138916015625, 9.913658142089844, -89.12007141113281, 187.0763702392578, 110.7110595703125, 112.48273468017578, 15.747079849243164, -28.060840606689453, 86.69757843017578, 136.20391845703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000522.npy"}
|
|
{"epoch": 0.7891156462585034, "step": 523, "batch_size": 64, "mean": 46.14569854736328, "std": 77.18418884277344, "min": -174.97198486328125, "p10": -60.152575683593746, "median": 51.6284122467041, "p90": 131.2216781616211, "max": 177.3218536376953, "pos_frac": 0.796875, "sample": [23.914417266845703, 164.2611083984375, 13.698917388916016, 9.681350708007812, 97.71033477783203, 83.0864486694336, 40.61668395996094, 33.236656188964844, 102.99201965332031, 38.43772888183594, 1.3925037384033203, 80.25718688964844, 140.00502014160156, -117.80561828613281, 74.98139953613281, 40.23625183105469, 84.262939453125, -4.428571701049805, 40.82403564453125, -120.30477142333984, 15.827285766601562, 134.94412231445312, -30.867080688476562, 65.46985626220703, -94.33866882324219, 118.609619140625, -32.444725036621094, 5.250469207763672, 10.716590881347656, 111.0926513671875, 27.055994033813477, 84.30186462402344, 113.843505859375, 88.9466781616211, 127.20832824707031, 110.59133911132812, 132.5643768310547, 28.089202880859375, 29.053306579589844, -174.97198486328125, 49.828216552734375, -74.83138275146484, 177.3218536376953, 88.75927734375, 128.08871459960938, -7.661933898925781, 27.851947784423828, 67.1226806640625, 124.88027954101562, -62.937843322753906, 171.51011657714844, 53.42860794067383, 107.18157958984375, -53.65361785888672, 134.04156494140625, 98.26898956298828, 61.06574249267578, 120.518310546875, 17.117111206054688, -136.74867248535156, -40.24877166748047, 71.13693237304688, 16.2353515625, 117.05075073242188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000523.npy"}
|
|
{"epoch": 0.7906273620559335, "step": 524, "batch_size": 64, "mean": 45.3804817199707, "std": 65.57249450683594, "min": -150.43295288085938, "p10": -31.907807159423825, "median": 43.10912895202637, "p90": 136.37902069091797, "max": 183.37042236328125, "pos_frac": 0.765625, "sample": [101.2723388671875, 40.01616668701172, 181.55776977539062, 44.99978256225586, -7.072296142578125, -21.83135223388672, -32.38933563232422, -37.540618896484375, 82.48280334472656, 35.696964263916016, 137.4801025390625, -0.80194091796875, -45.74406433105469, 40.19146728515625, 36.459259033203125, 153.11785888671875, -112.03646850585938, -30.78424072265625, 26.44369888305664, 113.94723510742188, 119.67704010009766, -150.43295288085938, 117.87516784667969, -0.028377532958984375, 7.00323486328125, 26.276893615722656, 156.00387573242188, 67.31029510498047, 78.88957977294922, 1.54608154296875, -12.188392639160156, 76.26042938232422, 23.009689331054688, 57.097068786621094, 87.4879150390625, 55.933109283447266, 46.22631072998047, 77.72975158691406, 23.467575073242188, -39.0560302734375, 47.950775146484375, -14.613433837890625, 6.285423278808594, 183.37042236328125, 23.230417251586914, 92.32477569580078, 133.80982971191406, 1.0860633850097656, 45.41682434082031, 41.218475341796875, 56.01141357421875, 19.97484588623047, 54.00098419189453, 159.0437469482422, 47.2314338684082, 87.82884979248047, 140.1302490234375, 60.77934265136719, -38.56361389160156, 22.21080780029297, 125.89695739746094, -6.172096252441406, 84.90504455566406, 5.439907073974609], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000524.npy"}
|
|
{"epoch": 0.7921390778533636, "step": 525, "batch_size": 64, "mean": 37.24139404296875, "std": 71.94461059570312, "min": -128.4527587890625, "p10": -38.580946350097655, "median": 25.270998001098633, "p90": 141.71195220947266, "max": 223.51441955566406, "pos_frac": 0.671875, "sample": [31.73065185546875, -117.5594482421875, -53.246856689453125, -22.026729583740234, -73.2685775756836, 24.947181701660156, 27.2396240234375, 33.198883056640625, 174.18731689453125, -25.717994689941406, 72.44926452636719, 10.423080444335938, 79.70753479003906, 22.620147705078125, 50.052581787109375, 142.8616943359375, 60.34187316894531, -39.10051727294922, 103.65556335449219, -21.73150634765625, 34.79975128173828, -17.194217681884766, 162.72378540039062, 138.7307891845703, -1.2290420532226562, 87.96177673339844, -17.919937133789062, 88.2088623046875, 42.920188903808594, 79.36207580566406, 10.533920288085938, 121.1558837890625, 23.97344207763672, 75.71098327636719, 72.89625549316406, -25.964126586914062, 25.59481430053711, 18.1767578125, 223.51441955566406, -20.412256240844727, -26.763076782226562, 38.142250061035156, 11.32735824584961, -128.4527587890625, -5.94403076171875, 139.0292205810547, 1.6624336242675781, 10.747215270996094, 21.207061767578125, 131.93832397460938, -37.368614196777344, 155.75155639648438, 21.787155151367188, -11.703536987304688, 95.44660186767578, 150.94094848632812, -50.58467102050781, 60.8922119140625, 76.77145385742188, 74.51390075683594, -9.1444091796875, 148.56512451171875, -1.927988052368164, -87.69239807128906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000525.npy"}
|
|
{"epoch": 0.7936507936507936, "step": 526, "batch_size": 64, "mean": 35.366519927978516, "std": 68.94192504882812, "min": -135.91848754882812, "p10": -49.48865013122557, "median": 37.3592529296875, "p90": 121.42113113403322, "max": 199.39303588867188, "pos_frac": 0.734375, "sample": [173.37416076660156, 13.340648651123047, -20.367822647094727, 10.035085678100586, -65.00740814208984, 9.562023162841797, 5.465740203857422, 157.26881408691406, -119.37913513183594, -79.68360900878906, 36.662200927734375, 54.6666145324707, -86.41117095947266, 13.45037841796875, 90.24795532226562, -19.47254180908203, 52.99653244018555, 116.61009979248047, 23.672080993652344, 94.38822937011719, 41.32977294921875, 135.46661376953125, 22.511276245117188, 112.17269897460938, 123.48300170898438, 13.116863250732422, 37.781593322753906, 66.0285415649414, -18.340377807617188, 92.53750610351562, -2.9170665740966797, 55.781280517578125, 66.0404281616211, -19.935014724731445, 125.8427505493164, -10.53227424621582, 76.17559051513672, 18.88684844970703, 53.58447265625, 11.457221984863281, 84.93877410888672, 55.951297760009766, 3.1159820556640625, 81.69788360595703, 57.69856643676758, -135.91848754882812, -3.3991737365722656, 67.60234832763672, 94.90292358398438, -18.573974609375, 63.73944091796875, -6.693294525146484, 199.39303588867188, 49.160526275634766, 155.075927734375, 12.80645751953125, -58.553104400634766, 79.97190856933594, 36.936912536621094, 30.644546508789062, 55.27117919921875, -28.3382568359375, 59.691986083984375, -135.55679321289062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000526.npy"}
|
|
{"epoch": 0.7951625094482238, "step": 527, "batch_size": 64, "mean": 29.875896453857422, "std": 72.94585418701172, "min": -144.77442932128906, "p10": -51.42031707763672, "median": 15.610345840454102, "p90": 134.11583862304687, "max": 180.6656494140625, "pos_frac": 0.65625, "sample": [-7.37541389465332, 25.143707275390625, 23.87885284423828, 88.42076110839844, 14.449111938476562, 2.5770492553710938, 168.40463256835938, 49.24874496459961, 125.67631530761719, 160.26162719726562, 134.19149780273438, -30.48419189453125, 113.6170654296875, 4.803428649902344, 41.22716522216797, -51.82434844970703, 154.861083984375, 1.933511734008789, 64.89234924316406, -39.184852600097656, -72.60530090332031, -50.477577209472656, -44.52146911621094, -57.17295837402344, 91.157470703125, 18.73553466796875, 180.6656494140625, 27.012435913085938, 24.980449676513672, 10.293060302734375, 69.99908447265625, 12.7279052734375, -22.853607177734375, 55.88290023803711, 16.77157974243164, 164.23074340820312, 56.915794372558594, -7.810211181640625, -26.852140426635742, 117.58155822753906, 72.51605224609375, 2.487804412841797, 90.25604248046875, -32.94838333129883, -25.808242797851562, -5.8583526611328125, -59.57579803466797, 150.89767456054688, -10.456619262695312, -8.830183029174805, -11.14080810546875, 5.198530197143555, 73.20501708984375, 86.98304748535156, -127.01744079589844, -18.72486114501953, 62.28570556640625, 8.947517395019531, 92.89656066894531, 133.93930053710938, -144.77442932128906, -113.58061218261719, 68.26982116699219, 13.510978698730469], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000527.npy"}
|
|
{"epoch": 0.7966742252456538, "step": 528, "batch_size": 64, "mean": 31.643692016601562, "std": 79.32357788085938, "min": -164.59149169921875, "p10": -38.497798919677734, "median": 24.619765281677246, "p90": 131.2139701843262, "max": 179.06585693359375, "pos_frac": 0.6875, "sample": [118.05674743652344, -164.59149169921875, 156.51641845703125, -4.477357864379883, 52.98169708251953, 89.29759979248047, 64.07170104980469, 10.382553100585938, 11.830337524414062, 3.9560012817382812, -6.535392761230469, -38.01605224609375, 118.93927001953125, -150.24945068359375, -5.925954818725586, -6.054534912109375, 132.92123413085938, 15.002233505249023, 19.414243698120117, -13.942733764648438, -27.271888732910156, 179.06585693359375, 0.04093170166015625, -145.96127319335938, -1.6139144897460938, 45.063629150390625, -38.704261779785156, -24.29431915283203, 18.53474235534668, 115.3595962524414, 71.90213012695312, 3.296142578125, 38.85024642944336, 173.64950561523438, 80.91816711425781, 1.7482357025146484, 178.68836975097656, 45.66228485107422, 56.07090759277344, 42.958251953125, 127.23035430908203, 136.9812469482422, 15.744977951049805, 36.206146240234375, -132.0692901611328, 46.80830383300781, 13.386005401611328, -1.739837646484375, 97.40695190429688, 4.751441955566406, 29.825286865234375, -130.25094604492188, -122.94444274902344, 163.25119018554688, 118.24575805664062, 40.818546295166016, -3.1783103942871094, -6.887413024902344, 122.25215911865234, -5.799772262573242, 83.40691375732422, 30.650131225585938, 90.73392486572266, 52.82652282714844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000528.npy"}
|
|
{"epoch": 0.7981859410430839, "step": 529, "batch_size": 64, "mean": 71.42031860351562, "std": 73.63651275634766, "min": -138.36544799804688, "p10": -17.123058319091797, "median": 81.72113037109375, "p90": 165.80318298339844, "max": 196.31124877929688, "pos_frac": 0.84375, "sample": [109.25556945800781, -35.042476654052734, 84.07693481445312, 10.290451049804688, 70.33380126953125, 6.1816253662109375, 196.31124877929688, 168.26365661621094, -20.982601165771484, 11.108123779296875, 3.5308895111083984, 0.8650722503662109, 12.826690673828125, 80.55595397949219, 109.62504577636719, 85.3853988647461, -64.62749481201172, 127.36100006103516, -16.35601806640625, 126.37902069091797, 151.55123901367188, -12.446456909179688, -26.417266845703125, 113.88134002685547, 80.8661117553711, 155.6693115234375, -17.45178985595703, 7.238151550292969, 116.53709411621094, 32.06019592285156, 168.30819702148438, 126.2532958984375, 101.89073181152344, 161.55804443359375, 80.64543151855469, 95.04390716552734, 139.62353515625, 158.6533660888672, 3.0400657653808594, 149.7557373046875, 108.37680053710938, 143.123046875, 191.01882934570312, 26.031265258789062, 89.75260162353516, 146.20335388183594, 53.21833801269531, -2.3468856811523438, 82.5761489868164, 50.74139404296875, 26.23339080810547, 89.49913787841797, 33.32422637939453, 168.5382843017578, -79.6739501953125, 167.82369995117188, 54.503639221191406, 156.26002502441406, 91.05377197265625, 71.85252380371094, -138.36544799804688, 18.33061981201172, 3.600921630859375, 167.62252807617188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000529.npy"}
|
|
{"epoch": 0.799697656840514, "step": 530, "batch_size": 64, "mean": 43.41593933105469, "std": 67.25553131103516, "min": -97.73687744140625, "p10": -19.721403503417967, "median": 28.622288703918457, "p90": 141.06814880371095, "max": 207.92861938476562, "pos_frac": 0.734375, "sample": [207.92861938476562, 56.41806411743164, 27.574251174926758, 6.320178985595703, 18.549781799316406, -10.450210571289062, 155.96014404296875, -14.277290344238281, 121.87867736816406, 112.24620056152344, 18.470386505126953, 145.1883544921875, -14.883485794067383, 25.85578155517578, 72.15596008300781, 56.985069274902344, 29.670326232910156, 52.13751220703125, 141.57496643066406, -5.969902038574219, 0.7312660217285156, -19.999961853027344, -82.29127502441406, 125.17562103271484, 108.33280944824219, 0.8035449981689453, 79.12742614746094, 51.58462905883789, 22.849510192871094, 154.994140625, 45.44123077392578, 20.326820373535156, 162.99395751953125, 176.82522583007812, 15.726028442382812, 16.20979881286621, 35.20344161987305, 93.17025756835938, -69.19929504394531, 7.779438018798828, 8.304733276367188, 72.46138000488281, -11.167686462402344, 53.60478210449219, -53.09290313720703, -19.071434020996094, 138.7559356689453, -15.561103820800781, -97.73687744140625, -30.628631591796875, -7.407510757446289, -47.37689208984375, 136.9205322265625, -5.881862640380859, 139.8855743408203, 38.21427917480469, 96.33760070800781, -17.591400146484375, 37.063018798828125, 5.9381866455078125, 120.2176513671875, 39.980987548828125, 5.908977508544922, 41.42509460449219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000530.npy"}
|
|
{"epoch": 0.8012093726379441, "step": 531, "batch_size": 64, "mean": 44.665199279785156, "std": 73.0637435913086, "min": -147.0755157470703, "p10": -27.89951934814453, "median": 36.670265197753906, "p90": 152.50930175781252, "max": 171.9420166015625, "pos_frac": 0.703125, "sample": [-65.46867370605469, 101.95618438720703, 158.65692138671875, 76.36843872070312, 40.67707824707031, 70.70133972167969, 3.666534423828125, 160.76617431640625, 38.46504211425781, -2.3920021057128906, 31.519309997558594, 0.6008586883544922, 171.9420166015625, -6.3981781005859375, 119.1821060180664, 143.71893310546875, -107.1671371459961, -28.881797790527344, 15.769676208496094, 146.2204132080078, 41.56892395019531, 164.35191345214844, -71.34331512451172, -14.636955261230469, -23.096923828125, 34.87548828125, 60.53699493408203, 47.0203857421875, 31.357833862304688, 4.089408874511719, 51.651283264160156, -8.918355941772461, -1.4284801483154297, 7.9647369384765625, -147.0755157470703, -4.364604949951172, 50.066253662109375, 116.22408294677734, 27.159439086914062, -49.1356201171875, 55.12123107910156, 64.59722137451172, 147.73602294921875, 143.93112182617188, -70.29798889160156, -25.60753631591797, 154.55499267578125, 146.68263244628906, 63.576385498046875, 19.5450439453125, -7.284904479980469, 20.072124481201172, 59.03619384765625, 125.40794372558594, 42.66453552246094, -13.86532974243164, 83.35977172851562, -5.552669525146484, 122.58642578125, 167.61419677734375, -0.2495574951171875, 0.07813072204589844, 163.62246704101562, 14.474090576171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000531.npy"}
|
|
{"epoch": 0.8027210884353742, "step": 532, "batch_size": 64, "mean": 51.352699279785156, "std": 75.74872589111328, "min": -112.88005828857422, "p10": -50.625601196289054, "median": 59.61493682861328, "p90": 158.99124450683595, "max": 194.19361877441406, "pos_frac": 0.765625, "sample": [84.97384643554688, 67.96949768066406, 142.17971801757812, 157.55221557617188, -45.651153564453125, -3.7048568725585938, 83.30587005615234, 167.98046875, 2.719707489013672, 97.51173400878906, -61.74915313720703, 105.64927673339844, 138.0995635986328, 28.614486694335938, 72.81170654296875, 32.666961669921875, 65.61630249023438, 101.50694274902344, 17.615171432495117, -95.57005310058594, -31.071489334106445, 10.191370010375977, 137.50015258789062, 33.32623291015625, 16.35638427734375, 70.93031311035156, 80.0235595703125, 194.19361877441406, 73.99690246582031, 169.876953125, 135.15834045410156, 29.063716888427734, 23.179443359375, 172.58096313476562, 11.690460205078125, 53.61357116699219, 77.52420043945312, -55.86823272705078, 4.319122314453125, 11.586153030395508, -52.75750732421875, -17.315048217773438, 15.34682846069336, -16.360599517822266, 166.16207885742188, 42.92644500732422, 159.60797119140625, -29.055931091308594, 45.31636047363281, 99.33999633789062, 94.38929748535156, 82.2741470336914, -15.483497619628906, 164.98939514160156, -112.88005828857422, 96.49052429199219, -105.99615478515625, 72.15707397460938, 137.92056274414062, -96.44451141357422, 1.5092792510986328, -18.285242080688477, 100.37580871582031, 94.07539367675781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000532.npy"}
|
|
{"epoch": 0.8042328042328042, "step": 533, "batch_size": 64, "mean": 47.43366241455078, "std": 70.49414825439453, "min": -153.833251953125, "p10": -17.3102783203125, "median": 31.866703033447266, "p90": 145.1284301757813, "max": 200.25765991210938, "pos_frac": 0.765625, "sample": [46.178199768066406, 122.06074523925781, 6.070045471191406, 37.523681640625, 71.22958374023438, 73.97930145263672, 5.5132904052734375, -9.930706024169922, 136.48883056640625, 90.97921752929688, -17.722183227539062, 5.731288909912109, 3.1511573791503906, -1.9333324432373047, 90.25555419921875, 148.83111572265625, 11.401117324829102, 121.14966583251953, -41.89521789550781, -7.007896423339844, 10.349372863769531, 6.811250686645508, 66.90702819824219, 16.745361328125, -65.3119888305664, 109.58163452148438, 98.9227294921875, 24.32196044921875, 22.758995056152344, -123.9516372680664, -6.4681243896484375, -10.713397979736328, 154.61431884765625, 128.051513671875, -8.360170364379883, 153.92079162597656, 6.301788330078125, 200.25765991210938, 34.50257110595703, -24.963809967041016, 29.18145751953125, 106.99401092529297, 88.25277709960938, 95.64286041259766, 127.95335388183594, 162.53738403320312, 40.14958190917969, -56.6060905456543, 8.165779113769531, -7.359413146972656, 3.138975143432617, 110.72364807128906, 92.24091339111328, -16.349166870117188, 3.1369457244873047, 24.76781463623047, 108.2771224975586, 60.09693908691406, 29.2308349609375, 158.5032501220703, 116.314208984375, 154.5977325439453, -153.833251953125, 63.665687561035156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000533.npy"}
|
|
{"epoch": 0.8057445200302343, "step": 534, "batch_size": 64, "mean": 49.772850036621094, "std": 80.86235046386719, "min": -149.936767578125, "p10": -41.642341232299806, "median": 46.823997497558594, "p90": 156.4381088256836, "max": 229.5037841796875, "pos_frac": 0.65625, "sample": [24.35633087158203, -12.958093643188477, 116.35104370117188, 33.73005676269531, -5.62800407409668, 50.118858337402344, -0.1551666259765625, -149.936767578125, 64.64151000976562, -64.30477905273438, -4.583436965942383, 17.77569580078125, 141.98736572265625, 125.2526626586914, -40.03894805908203, 229.5037841796875, 88.39325714111328, 68.76332092285156, 61.61686706542969, -19.920257568359375, 145.31744384765625, 182.4095001220703, 80.97763061523438, 61.52069091796875, -11.243034362792969, -67.43197631835938, -141.5028076171875, -7.621862411499023, 174.99453735351562, 31.266754150390625, 54.033294677734375, -42.32950973510742, 174.02435302734375, 155.8026123046875, 156.71046447753906, 109.82839965820312, -54.51265335083008, 43.41040802001953, 198.9080352783203, 146.640380859375, 84.06971740722656, 145.95013427734375, -36.691871643066406, 15.58544921875, 124.76305389404297, 90.161865234375, -9.69333267211914, 98.39015197753906, 10.251213073730469, -16.9952392578125, 19.43505859375, 161.93626403808594, 55.11476135253906, -19.266040802001953, -16.51294708251953, 9.641448974609375, 95.4104995727539, -42.38777160644531, 66.86970520019531, -2.6559677124023438, 76.75672149658203, 43.529136657714844, 141.80072021484375, -26.168079376220703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000534.npy"}
|
|
{"epoch": 0.8072562358276644, "step": 535, "batch_size": 64, "mean": 40.693458557128906, "std": 75.89592742919922, "min": -164.99392700195312, "p10": -33.70376815795898, "median": 28.322263717651367, "p90": 127.96705322265628, "max": 204.49464416503906, "pos_frac": 0.671875, "sample": [22.061508178710938, -31.650531768798828, 95.15243530273438, 44.86307907104492, 10.676567077636719, 34.60432434082031, -135.08985900878906, -26.24765396118164, -78.08002471923828, 42.35736083984375, 102.73187255859375, -48.16871643066406, 36.62952423095703, 98.82782745361328, -18.24457550048828, -0.22336578369140625, -34.13177490234375, 27.732833862304688, -0.774566650390625, -26.53753662109375, 27.88758087158203, 86.13760375976562, 112.34613037109375, -28.812225341796875, 121.67555236816406, 4.920433044433594, 168.982421875, -88.21379089355469, -32.70508575439453, 14.074274063110352, 96.65327453613281, 144.71673583984375, 78.79582214355469, -7.826133728027344, 67.12042236328125, 118.6993637084961, 25.90418243408203, 204.49464416503906, -38.23689270019531, 97.88780975341797, 45.21765899658203, 4.467662811279297, 199.57028198242188, 92.93501281738281, -6.6866912841796875, -164.99392700195312, -21.09381103515625, 103.5815658569336, -1.3367805480957031, 68.1095199584961, 94.40999603271484, -23.92679214477539, 28.756946563720703, 174.4468994140625, 20.718902587890625, 96.95167541503906, 13.67510986328125, 112.77761840820312, 122.46856689453125, 1.8942337036132812, 44.1136589050293, -5.111289978027344, 182.1212158203125, 130.32354736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000535.npy"}
|
|
{"epoch": 0.8087679516250945, "step": 536, "batch_size": 64, "mean": 62.283599853515625, "std": 64.63536834716797, "min": -50.77445983886719, "p10": -5.362652587890624, "median": 51.308834075927734, "p90": 155.84038696289062, "max": 209.6444854736328, "pos_frac": 0.84375, "sample": [61.03432083129883, 11.010337829589844, -38.46015167236328, 152.2384796142578, 110.00262451171875, 23.370956420898438, -2.572540283203125, -13.031730651855469, 56.22955322265625, 10.084085464477539, -50.77445983886719, 53.49578094482422, 83.08580780029297, 164.79132080078125, 133.82479858398438, 110.0286636352539, 49.12188720703125, 151.51553344726562, 209.6444854736328, 54.874916076660156, 25.039819717407227, 8.489473342895508, 145.50799560546875, 92.78619384765625, 18.529998779296875, 166.27349853515625, 86.30830383300781, 3.5553150177001953, 92.16264343261719, 66.9940185546875, 170.3026123046875, -6.764698028564453, 4.007114410400391, 62.62607192993164, -13.22418212890625, 12.409210205078125, 144.3848419189453, 16.02166748046875, 76.8897705078125, 1.5375308990478516, 154.38595581054688, 150.08651733398438, 69.14554595947266, -4.611177444458008, 11.428836822509766, 34.100982666015625, 15.637992858886719, 57.949073791503906, 48.13900375366211, 156.46371459960938, -1.6427288055419922, 138.9136505126953, 38.14219665527344, 156.61489868164062, 119.02604675292969, 37.05767822265625, 25.912925720214844, 183.9286651611328, -5.684713363647461, -33.58955383300781, 22.65822410583496, 100.34326171875, 3.1737060546875, 5.2176055908203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000536.npy"}
|
|
{"epoch": 0.8102796674225246, "step": 537, "batch_size": 64, "mean": 48.76762390136719, "std": 76.72247314453125, "min": -94.20277404785156, "p10": -57.95054016113281, "median": 45.59852981567383, "p90": 151.146745300293, "max": 220.2465362548828, "pos_frac": 0.703125, "sample": [90.19938659667969, 72.53411102294922, 7.2039642333984375, 89.6025390625, 186.34173583984375, 132.8428497314453, 103.59394836425781, -0.33173179626464844, 95.8790283203125, 63.06499099731445, 46.76439666748047, 5.865800857543945, 157.71595764160156, 156.7464599609375, 116.00740051269531, 147.24920654296875, 143.57830810546875, 10.113775253295898, -11.530326843261719, -20.6297664642334, -2.3464584350585938, 41.790496826171875, 27.673599243164062, 6.105705261230469, -4.797548294067383, 159.09967041015625, 63.28802490234375, 9.654769897460938, -65.52542877197266, -7.804563522338867, 47.34117889404297, 26.456668853759766, 72.94159698486328, -22.6522216796875, -60.36592102050781, 6.483066558837891, 194.2012939453125, 81.63609313964844, 9.410919189453125, -8.392589569091797, 26.708023071289062, 152.51332092285156, -71.15333557128906, 44.43266296386719, -25.22779655456543, -30.209218978881836, 80.95294189453125, -32.022552490234375, 54.31581115722656, 49.29383850097656, 4.251220703125, 119.03970336914062, -52.31465148925781, 142.85568237304688, 147.95806884765625, 220.2465362548828, -71.25430297851562, 83.65238952636719, -64.0215072631836, 58.16314697265625, 141.39468383789062, 147.78518676757812, -94.20277404785156, -79.03942108154297], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000537.npy"}
|
|
{"epoch": 0.8117913832199547, "step": 538, "batch_size": 64, "mean": 54.715885162353516, "std": 70.05072784423828, "min": -136.10455322265625, "p10": -26.487525177001952, "median": 46.01313591003418, "p90": 152.69129180908206, "max": 240.94435119628906, "pos_frac": 0.8125, "sample": [-40.974021911621094, 107.197998046875, 32.413734436035156, 21.058208465576172, 48.78245162963867, -136.10455322265625, 52.77445602416992, -15.005447387695312, 66.19195556640625, 124.57321166992188, 61.102806091308594, 148.4808349609375, 4.85986328125, 129.50201416015625, 41.943878173828125, 36.82389831542969, 106.99272918701172, 154.4957733154297, 15.163047790527344, 55.51747512817383, 38.661155700683594, 1.834716796875, 161.2581787109375, 77.40151977539062, 29.47948455810547, -40.27116012573242, 92.8065185546875, 11.844093322753906, 187.6773681640625, 33.686058044433594, -24.22222900390625, 102.25365447998047, -24.191574096679688, -6.35211181640625, 0.31943511962890625, 30.491928100585938, 43.2364501953125, 80.53153228759766, 0.3770599365234375, 40.19158935546875, 191.41583251953125, -42.172760009765625, 156.68209838867188, 73.8203125, -60.31749725341797, 87.63594055175781, -27.45836639404297, 133.62538146972656, 94.25140380859375, 240.94435119628906, 125.83218383789062, 43.24382019042969, 67.38615417480469, -52.754852294921875, 101.19259643554688, 61.61228942871094, 170.04393005371094, 1.4108314514160156, 104.06791687011719, 12.476593017578125, 8.622978210449219, 72.50796508789062, 97.29334259033203, -12.349925994873047], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000538.npy"}
|
|
{"epoch": 0.8133030990173847, "step": 539, "batch_size": 64, "mean": 40.80616760253906, "std": 65.64129638671875, "min": -94.57168579101562, "p10": -23.150367736816406, "median": 28.26344871520996, "p90": 141.99639587402345, "max": 198.28076171875, "pos_frac": 0.75, "sample": [61.30500793457031, 102.25405883789062, 1.495513916015625, 163.20135498046875, 46.670982360839844, -4.802604675292969, 144.21881103515625, 115.29421997070312, 78.3636474609375, 8.229438781738281, 65.48968505859375, 40.915283203125, 30.33111572265625, 7.980926513671875, -23.337615966796875, -9.117034912109375, -8.754669189453125, -3.7182693481445312, 167.64231872558594, 57.536399841308594, 15.528718948364258, -35.341732025146484, 50.901397705078125, -94.57168579101562, 6.405063629150391, -84.75599670410156, -6.174383163452148, 74.044189453125, 198.28076171875, 43.74787902832031, 16.458824157714844, 7.406829833984375, -11.546417236328125, 11.61187744140625, 163.52334594726562, 43.930335998535156, 8.782554626464844, 19.064422607421875, 136.81076049804688, 11.979288101196289, -22.713455200195312, 39.012664794921875, 41.2729606628418, 1.6376113891601562, 87.81585693359375, 45.44233703613281, -81.13365936279297, 127.72444915771484, 16.767333984375, -28.698486328125, 113.95584106445312, 62.00878143310547, 14.505115509033203, -2.7566757202148438, 97.49336242675781, 58.204681396484375, -0.4064483642578125, 26.195781707763672, -83.20498657226562, 100.48817443847656, 31.129838943481445, 157.8138427734375, 182.20474243164062, 9.55047607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000539.npy"}
|
|
{"epoch": 0.8148148148148148, "step": 540, "batch_size": 64, "mean": 59.29026412963867, "std": 69.34873962402344, "min": -104.39628601074219, "p10": -23.18605976104736, "median": 48.45354461669922, "p90": 153.81943359375, "max": 186.89202880859375, "pos_frac": 0.828125, "sample": [-29.608169555664062, 5.1852569580078125, 186.89202880859375, 151.2417449951172, 5.451482772827148, -104.39628601074219, 92.7552490234375, 61.39143371582031, 50.88678741455078, 20.875125885009766, 40.79719161987305, 98.633544921875, 154.31497192382812, 51.16765594482422, -57.71453094482422, 29.761140823364258, 147.6568603515625, 84.13509368896484, 7.250530242919922, 68.21543884277344, 47.89637756347656, -48.66920471191406, 67.1619873046875, 38.62117004394531, 181.67886352539062, 133.47964477539062, -20.9863338470459, 62.839508056640625, -39.486602783203125, 29.552749633789062, 45.224090576171875, 161.21385192871094, 137.57302856445312, 59.67747497558594, 148.36322021484375, -1.7510871887207031, -6.66278076171875, 127.01631927490234, -24.128799438476562, 2.057901382446289, 0.5472068786621094, 126.21571350097656, 141.2109375, 170.43157958984375, 166.28639221191406, 18.021774291992188, 145.1158905029297, 38.61305618286133, 40.735137939453125, -54.56913757324219, 44.46528625488281, 86.20806884765625, 35.72770690917969, 3.8466110229492188, 5.6414031982421875, 49.010711669921875, -15.582817077636719, 179.6166534423828, 27.266372680664062, 70.69970703125, 125.1677474975586, 8.476499557495117, 63.19749450683594, 152.66317749023438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000540.npy"}
|
|
{"epoch": 0.8163265306122449, "step": 541, "batch_size": 64, "mean": 57.04608917236328, "std": 66.90148162841797, "min": -97.60089874267578, "p10": -14.99986019134521, "median": 49.749752044677734, "p90": 149.47283630371095, "max": 165.1106719970703, "pos_frac": 0.703125, "sample": [135.41885375976562, 45.9268798828125, 87.95062255859375, 156.65240478515625, -3.086700439453125, 147.12005615234375, -2.7791671752929688, 118.38084411621094, 145.98312377929688, -0.2723541259765625, -6.2710113525390625, -3.697357177734375, 27.313400268554688, 79.83566284179688, -4.457725524902344, -97.60089874267578, 114.784912109375, 155.10000610351562, 27.944168090820312, 150.48117065429688, -41.492652893066406, 42.66859817504883, 165.1106719970703, 135.38018798828125, 135.01876831054688, 86.2914810180664, 119.79347229003906, -10.927642822265625, -45.290035247802734, -4.893688201904297, 96.62703704833984, 21.290315628051758, 163.2491455078125, 163.25384521484375, 6.014183044433594, 107.57174682617188, 72.54169464111328, -35.720855712890625, 91.92608642578125, -4.554636001586914, 6.057710647583008, -47.15127944946289, 93.66555786132812, 128.0556640625, 38.56330108642578, 69.06813049316406, 121.3780517578125, 104.67023468017578, -48.87840270996094, -3.333892822265625, 2.435606002807617, 123.07546997070312, 62.27216339111328, 49.31901550292969, -6.3153533935546875, 18.74810791015625, 41.171356201171875, 60.93196487426758, -16.74509620666504, 26.538658142089844, 86.56562042236328, -10.480928421020508, 162.5726776123047, 50.18048858642578], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000541.npy"}
|
|
{"epoch": 0.817838246409675, "step": 542, "batch_size": 64, "mean": 62.12199020385742, "std": 78.8272705078125, "min": -132.14390563964844, "p10": -43.74644203186035, "median": 53.94924545288086, "p90": 161.53887176513672, "max": 201.42926025390625, "pos_frac": 0.8125, "sample": [121.78834533691406, 100.29644775390625, 13.100494384765625, 188.26492309570312, 42.78324890136719, 12.117889404296875, 47.420318603515625, 61.474281311035156, 82.37413024902344, 62.560951232910156, 45.37022018432617, -1.1328506469726562, -42.794593811035156, 130.0057373046875, 32.82007598876953, 122.59558868408203, 53.476829528808594, 201.42926025390625, -80.06358337402344, 156.302734375, 148.60446166992188, 171.55810546875, 22.19713592529297, 4.327911376953125, -109.2847900390625, -132.14390563964844, 108.80026245117188, 54.421661376953125, 90.19098663330078, -52.53717041015625, 151.87889099121094, 161.490966796875, 1.603445053100586, 145.87808227539062, -53.303932189941406, -10.62838363647461, 0.1585693359375, 78.67228698730469, -2.3684310913085938, 102.39918518066406, 31.26312255859375, 20.438735961914062, 161.5594024658203, -44.15437698364258, 72.24617004394531, 3.3147659301757812, 154.84503173828125, 21.879356384277344, 38.24207305908203, 173.77029418945312, 131.29522705078125, 175.7825469970703, -69.97236633300781, 33.803489685058594, -3.1169605255126953, 5.226490020751953, 19.4193115234375, 159.4548797607422, 136.27438354492188, 49.98103332519531, 69.00568389892578, 173.01727294921875, 70.13663482666016, 159.9892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000542.npy"}
|
|
{"epoch": 0.8193499622071051, "step": 543, "batch_size": 64, "mean": 55.2010612487793, "std": 70.38705444335938, "min": -166.9195556640625, "p10": -11.54035415649414, "median": 45.35458755493164, "p90": 149.56314544677736, "max": 178.19717407226562, "pos_frac": 0.84375, "sample": [37.38856506347656, 29.372447967529297, 64.59635925292969, 0.03433418273925781, 74.11122131347656, 24.113780975341797, 94.54470825195312, 140.41363525390625, -127.16685485839844, 144.08035278320312, 53.61805725097656, 124.50765991210938, 107.22781372070312, 136.74256896972656, 178.19717407226562, 68.98088073730469, 156.56048583984375, -72.14627075195312, 48.335975646972656, 149.78675842285156, 25.102941513061523, 6.657402038574219, 2.9009971618652344, 32.798797607421875, 10.958568572998047, 138.9266815185547, -166.9195556640625, 1.2070999145507812, 26.372440338134766, -41.717437744140625, 34.58424377441406, 153.2545623779297, 149.0413818359375, 33.56077194213867, 127.31134796142578, 53.22276306152344, 161.50411987304688, -3.977365493774414, -21.311683654785156, -7.875762939453125, 75.01240539550781, 83.06084442138672, 90.48239135742188, 28.816650390625, 146.67898559570312, 38.847434997558594, 87.5809555053711, 168.75973510742188, 5.640697479248047, 42.373199462890625, -12.306224822998047, 49.86711883544922, 31.560531616210938, 4.902015686035156, 67.77574157714844, 27.004669189453125, 74.725341796875, 21.97639274597168, 7.6513214111328125, -24.47354507446289, 119.10212707519531, 169.84945678710938, -9.75332260131836, 88.83094787597656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000543.npy"}
|
|
{"epoch": 0.8208616780045351, "step": 544, "batch_size": 64, "mean": 63.5430908203125, "std": 78.17076873779297, "min": -104.01992797851562, "p10": -39.73048782348633, "median": 58.00074768066406, "p90": 162.3791198730469, "max": 219.4183349609375, "pos_frac": 0.8125, "sample": [21.108489990234375, -72.8485107421875, 58.137359619140625, 24.729951858520508, 121.79212188720703, 46.015899658203125, -69.94107055664062, 44.98487091064453, 18.64984130859375, 40.046043395996094, 29.981353759765625, 61.52727127075195, 16.390350341796875, 136.73623657226562, 84.83667755126953, 26.187353134155273, -104.01992797851562, 85.04810333251953, 88.27091217041016, 64.76962280273438, 69.0665283203125, 87.36075592041016, 114.61865234375, 153.19439697265625, 156.66061401367188, 46.62532043457031, 21.044021606445312, -7.642112731933594, 1.098562240600586, -92.32688903808594, 219.4183349609375, -94.83799743652344, 130.4401092529297, -1.9661178588867188, 164.49966430664062, 97.42507934570312, 190.6602783203125, -41.70844268798828, 24.810516357421875, 39.28699493408203, 120.88496398925781, 32.025611877441406, 187.51055908203125, 20.304533004760742, 22.14641571044922, 67.45207214355469, 123.52980041503906, 83.99532318115234, 190.57113647460938, -36.15895080566406, 2.9848499298095703, 154.28814697265625, 185.03427124023438, 210.56117248535156, -41.261146545410156, -2.7717437744140625, 147.93032836914062, 22.160430908203125, 157.43118286132812, 148.5977783203125, -2.4933395385742188, 57.8641357421875, 86.25642395019531, 127.782958984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000544.npy"}
|
|
{"epoch": 0.8223733938019653, "step": 545, "batch_size": 64, "mean": 40.036678314208984, "std": 73.54252624511719, "min": -161.61105346679688, "p10": -41.52947998046875, "median": 26.34117889404297, "p90": 150.85923461914064, "max": 195.6271514892578, "pos_frac": 0.6875, "sample": [76.06477355957031, 85.69708251953125, -42.29335021972656, 168.98585510253906, 21.524913787841797, 5.44622802734375, -27.573875427246094, -34.42938995361328, -19.389780044555664, 39.48952865600586, 57.978240966796875, -8.61859130859375, 150.24417114257812, 18.241228103637695, 28.979650497436523, 39.85877227783203, 36.08148193359375, 134.2019805908203, -41.53886413574219, 83.13456726074219, -1.4070510864257812, 102.7657470703125, -8.004005432128906, 100.32693481445312, 84.17098999023438, 10.017433166503906, 28.697593688964844, 19.70371437072754, -0.368743896484375, 10.902219772338867, -0.3594627380371094, -41.50758361816406, -63.814453125, 89.14314270019531, -66.93701171875, -30.29353141784668, 159.21778869628906, 162.81712341308594, 122.4041748046875, -4.290122985839844, -5.608066558837891, 167.43106079101562, 125.7685546875, 23.984764099121094, 11.729938507080078, 87.01558685302734, -15.777000427246094, 105.51184844970703, 46.464080810546875, -161.61105346679688, 13.41400146484375, 89.05157470703125, 37.99017333984375, 195.6271514892578, 92.61746215820312, -82.24995422363281, 2.0343284606933594, 94.4625473022461, -105.43742370605469, 0.9730625152587891, 67.52503204345703, 151.12283325195312, 162.35421752929688, 12.683162689208984], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000545.npy"}
|
|
{"epoch": 0.8238851095993953, "step": 546, "batch_size": 64, "mean": 47.9470100402832, "std": 71.43206024169922, "min": -125.78521728515625, "p10": -55.793245315551744, "median": 47.13375473022461, "p90": 143.8185836791992, "max": 167.96224975585938, "pos_frac": 0.796875, "sample": [146.512451171875, -25.370941162109375, 143.17396545410156, 43.424644470214844, 93.30085754394531, 136.45587158203125, 7.481475830078125, -26.249862670898438, 12.177230834960938, -29.809064865112305, -125.78521728515625, 94.40303039550781, 21.388683319091797, 23.47795867919922, -110.977783203125, 128.36590576171875, 20.173337936401367, 10.929847717285156, -40.575714111328125, 40.07494354248047, -80.06422424316406, 60.822044372558594, 55.26593017578125, 88.00059509277344, 44.12494659423828, 73.6259765625, 111.08427429199219, 22.010467529296875, 109.76097106933594, 167.96224975585938, 91.21437072753906, 11.200843811035156, 128.1874237060547, 36.62736511230469, 60.986942291259766, -66.53681945800781, 27.063514709472656, 158.6181640625, 60.60466003417969, 68.41676330566406, 17.904590606689453, 32.92999267578125, -62.31504440307617, 61.58007049560547, 38.746726989746094, 98.28572845458984, -26.161346435546875, 23.878047943115234, 86.27698516845703, 113.29617309570312, 112.925048828125, 29.25690460205078, 164.70761108398438, -73.8974609375, -9.418708801269531, 50.14256286621094, 106.23434448242188, 75.9478988647461, 163.0701446533203, 59.876976013183594, 158.11209106445312, -90.72428894042969, 144.0948486328125, 2.310789108276367], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000546.npy"}
|
|
{"epoch": 0.8253968253968254, "step": 547, "batch_size": 64, "mean": 58.534942626953125, "std": 74.24851989746094, "min": -161.91616821289062, "p10": -14.005144500732413, "median": 56.42059898376465, "p90": 149.60095672607423, "max": 215.34193420410156, "pos_frac": 0.828125, "sample": [4.083320617675781, 6.7960052490234375, 25.86458396911621, 12.354110717773438, 142.49752807617188, 44.193504333496094, -68.87886810302734, 193.97573852539062, 69.62887573242188, 124.67214965820312, 93.96710205078125, 96.90357208251953, 41.904273986816406, 65.01841735839844, 58.09684371948242, -1.9377517700195312, 80.77235412597656, 149.90138244628906, 17.404613494873047, 32.654075622558594, 123.15509033203125, 143.3810272216797, 159.31951904296875, -1.416351318359375, 153.57666015625, 101.98866271972656, 5.0201263427734375, 73.41802215576172, 65.82052612304688, 21.62488555908203, 70.68211364746094, -1.2244129180908203, -161.91616821289062, 113.24656677246094, -49.50407409667969, 63.35665512084961, 147.44525146484375, 25.33700180053711, 144.77957153320312, 53.17584228515625, 91.06710052490234, 163.07701110839844, 93.38540649414062, 112.5906982421875, -75.02960205078125, 54.744354248046875, 7.834953308105469, -129.24176025390625, 4.559112548828125, 35.09623718261719, 168.2935791015625, 133.8238067626953, 215.34193420410156, -18.148521423339844, 69.62232208251953, -32.763580322265625, 36.63783645629883, 131.5203857421875, 0.9961071014404297, 30.90935516357422, 49.03776550292969, 17.18071746826172, 148.89996337890625, -4.3372650146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000547.npy"}
|
|
{"epoch": 0.8269085411942555, "step": 548, "batch_size": 64, "mean": 38.3332633972168, "std": 72.73468780517578, "min": -91.34162139892578, "p10": -54.31920852661133, "median": 25.82855796813965, "p90": 139.7552673339844, "max": 192.8960418701172, "pos_frac": 0.6875, "sample": [83.45045471191406, 119.2909164428711, -17.247467041015625, -39.17283630371094, -91.34162139892578, 57.38481140136719, -21.225936889648438, 192.8960418701172, 122.24151611328125, 57.50782012939453, -20.664215087890625, 163.30738830566406, 0.6279792785644531, 13.340003967285156, 20.93865966796875, 1.310140609741211, -77.87628173828125, 34.30809783935547, -0.8703231811523438, 72.86112976074219, 140.20376586914062, 62.95228576660156, -77.67512512207031, 14.612831115722656, 111.6390380859375, 28.312702178955078, 178.1551971435547, 138.70877075195312, 19.36016845703125, 100.63533020019531, -24.27493667602539, 69.70135498046875, -13.052093505859375, 93.94856262207031, 19.538482666015625, -10.492843627929688, 177.28004455566406, 41.847084045410156, -28.84912872314453, 61.04216766357422, 57.571388244628906, 0.41628074645996094, 62.589378356933594, -64.59564208984375, 184.45257568359375, 1.006561279296875, -18.955726623535156, 50.28324890136719, -55.19165802001953, 71.57337188720703, 19.753448486328125, -91.2943115234375, 42.523521423339844, 107.41378784179688, 61.13764190673828, -15.596595764160156, 110.23491668701172, -23.71978759765625, 23.34441375732422, -52.28349304199219, 167.68885803222656, 112.18409729003906, -72.55833435058594, 0.6908187866210938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000548.npy"}
|
|
{"epoch": 0.8284202569916855, "step": 549, "batch_size": 64, "mean": 43.183135986328125, "std": 76.1004867553711, "min": -162.09254455566406, "p10": -54.33896102905273, "median": 40.715721130371094, "p90": 148.70928344726565, "max": 172.18753051757812, "pos_frac": 0.6875, "sample": [17.725128173828125, 28.81104278564453, -53.051353454589844, 67.30424499511719, -19.26622772216797, -54.89079284667969, 68.28083038330078, 4.138406753540039, 61.197227478027344, 172.18753051757812, 155.40968322753906, 11.5755615234375, 68.95751953125, 105.75430297851562, -10.2452392578125, -15.459663391113281, -95.48976135253906, 72.75634765625, 86.20185852050781, 108.70475769042969, 111.12242889404297, 117.04653930664062, 123.14703369140625, -67.78193664550781, 50.0086669921875, 4.419158935546875, 17.278514862060547, 159.57003784179688, 93.80087280273438, -56.226097106933594, 151.7388916015625, 118.11641693115234, 164.73040771484375, -21.005578994750977, 106.60734558105469, 168.99371337890625, -5.211336135864258, -72.23275756835938, -26.74932861328125, 119.2919692993164, -125.94717407226562, 26.098617553710938, -11.689979553222656, 23.649837493896484, 63.23847579956055, 105.5926513671875, -39.07421112060547, 99.75077819824219, 141.64019775390625, 31.422775268554688, -5.027976989746094, 92.32478332519531, 105.8353271484375, 72.69075012207031, 66.80235290527344, 19.544586181640625, 18.480737686157227, 3.7784996032714844, -7.917667388916016, 92.53008270263672, 154.4989471435547, -162.09254455566406, -33.82632827758789, -5.8490753173828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000549.npy"}
|
|
{"epoch": 0.8299319727891157, "step": 550, "batch_size": 64, "mean": 31.665681838989258, "std": 69.73125457763672, "min": -130.7543182373047, "p10": -62.21151924133301, "median": 27.0690860748291, "p90": 123.17795715332032, "max": 181.10952758789062, "pos_frac": 0.71875, "sample": [45.55967712402344, 171.22479248046875, 162.0415496826172, 32.93357849121094, -93.60230255126953, 5.960762023925781, 0.7151050567626953, -16.505416870117188, -10.692512512207031, 43.63346862792969, 103.53926086425781, 31.203086853027344, 46.3941650390625, 2.1468772888183594, 20.187349319458008, 147.810302734375, 172.0911865234375, 81.1938705444336, 16.779449462890625, 38.150386810302734, 109.6568374633789, -5.260768890380859, 88.62438201904297, 83.20416259765625, 110.21427917480469, -60.57817077636719, 181.10952758789062, -4.428077697753906, 18.020584106445312, -69.13691711425781, -67.1950912475586, 15.004989624023438, -51.288055419921875, 54.636962890625, 5.56500244140625, 54.542755126953125, 41.960533142089844, -9.92626953125, -33.30968475341797, 5.7717742919921875, -62.91152572631836, 74.6221923828125, 16.864795684814453, 27.220630645751953, 8.993240356445312, 73.71650695800781, 26.91754150390625, 31.72691535949707, 34.05439758300781, 96.38597106933594, -80.43215942382812, 98.90931701660156, 122.29934692382812, 49.52100372314453, 123.55450439453125, 22.893678665161133, -130.7543182373047, 4.478391647338867, 134.43821716308594, -122.7348403930664, 71.93759155273438, -20.34192657470703, -14.090282440185547, -28.618915557861328], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000550.npy"}
|
|
{"epoch": 0.8314436885865457, "step": 551, "batch_size": 64, "mean": 42.0129280090332, "std": 66.58758544921875, "min": -129.43487548828125, "p10": -17.393956756591795, "median": 17.678848266601562, "p90": 130.0168167114258, "max": 300.6473388671875, "pos_frac": 0.765625, "sample": [17.982437133789062, 64.7044677734375, 77.82402038574219, 17.375259399414062, 101.72041320800781, 53.903724670410156, 300.6473388671875, 0.21137619018554688, 17.066434860229492, 34.896629333496094, -35.053775787353516, 17.10533905029297, 117.11771392822266, 4.713693618774414, 71.67068481445312, 34.37301254272461, 91.78341674804688, 46.579795837402344, -129.43487548828125, -19.480979919433594, -1.3374805450439453, 16.036231994628906, 85.21752166748047, 14.5098876953125, 51.12385559082031, 164.95675659179688, -0.896148681640625, 67.17774200439453, 49.47723388671875, 126.12126159667969, 38.27751922607422, -19.948089599609375, 0.6114349365234375, 12.318389892578125, 35.64540100097656, 177.61712646484375, -0.58673095703125, 123.48019409179688, -8.853857040405273, 20.467361450195312, 1.2442398071289062, 43.460594177246094, -16.71991729736328, 15.385456085205078, 118.44587707519531, 159.05850219726562, 131.68634033203125, 13.583011627197266, 7.253059387207031, -0.10321807861328125, 43.79553985595703, 59.939231872558594, 3.274810791015625, -34.83696365356445, 3.0411911010742188, -6.600372314453125, -9.492656707763672, 151.63693237304688, -17.682830810546875, 8.264095306396484, -54.4954833984375, 4.664241790771484, 145.913818359375, 80.99026489257812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000551.npy"}
|
|
{"epoch": 0.8329554043839759, "step": 552, "batch_size": 64, "mean": 49.126773834228516, "std": 65.05878448486328, "min": -150.38436889648438, "p10": -9.054973793029783, "median": 53.10337448120117, "p90": 119.04282836914064, "max": 182.1974639892578, "pos_frac": 0.84375, "sample": [-6.71607780456543, 31.75018310546875, 44.67774200439453, -4.7154388427734375, 102.34903717041016, 57.02895736694336, 1.4570293426513672, 11.917068481445312, 30.14056396484375, 108.90701293945312, 129.72702026367188, 182.1974639892578, 112.68536376953125, 43.644004821777344, -6.198272705078125, -54.62694549560547, 0.318145751953125, 101.7474365234375, 44.79759979248047, -10.057357788085938, 40.797950744628906, 97.41105651855469, 86.43376159667969, 93.1149673461914, 103.03929138183594, 16.724151611328125, 70.10692596435547, 114.53031921386719, 111.58029174804688, -150.38436889648438, 13.319501876831055, 39.96253204345703, 98.56634521484375, -121.42916870117188, 73.39105987548828, 0.026042938232421875, 122.4541015625, 52.62205505371094, 1.5609817504882812, 3.7191009521484375, 136.98577880859375, 21.686492919921875, -43.793495178222656, 90.00263977050781, 87.26187133789062, 12.793212890625, 78.06871032714844, 43.75579071044922, -130.81617736816406, 121.25880432128906, 162.28369140625, 76.8102035522461, 13.808584213256836, -60.160423278808594, 95.89865112304688, 65.65030670166016, 53.401954650878906, 36.73998260498047, 105.3011474609375, 52.80479431152344, 74.67955780029297, 82.62640380859375, 120.97676086425781, 57.51091003417969], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000552.npy"}
|
|
{"epoch": 0.8344671201814059, "step": 553, "batch_size": 64, "mean": 44.01886749267578, "std": 81.3061752319336, "min": -159.28054809570312, "p10": -37.30851402282715, "median": 34.2902946472168, "p90": 165.2100494384766, "max": 234.19329833984375, "pos_frac": 0.671875, "sample": [196.22982788085938, -27.71396255493164, -2.0611419677734375, 71.42430114746094, 4.929649353027344, 0.6661434173583984, 136.48760986328125, 125.31353759765625, -20.79914093017578, -9.326942443847656, 34.42646026611328, 63.292205810546875, 169.57406616210938, 182.3311309814453, -54.797096252441406, 83.80923461914062, 6.658220291137695, 63.74811553955078, 39.38920593261719, -159.28054809570312, -8.441614151000977, 108.94486999511719, -12.903450012207031, 94.68609619140625, 129.86965942382812, 40.755699157714844, 0.7524490356445312, 25.94140625, -5.723114013671875, -45.06398010253906, 203.91424560546875, 134.39540100097656, 81.6941909790039, 234.19329833984375, 30.9818115234375, 83.43900299072266, 155.02734375, 91.65379333496094, 59.802581787109375, 74.08379364013672, 11.693687438964844, 47.74441146850586, 11.332883834838867, -11.007762908935547, -36.76112365722656, 25.368995666503906, -8.550331115722656, 182.84115600585938, -141.68975830078125, -78.5858383178711, 64.90664672851562, 104.85073852539062, 101.89450073242188, -32.906944274902344, -24.609580993652344, 34.15412902832031, -20.67398452758789, -80.03111267089844, 79.84515380859375, 74.43274688720703, -20.14452362060547, -37.54310989379883, 182.23257446289062, 6.1096954345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000553.npy"}
|
|
{"epoch": 0.8359788359788359, "step": 554, "batch_size": 64, "mean": 53.0959358215332, "std": 73.8534927368164, "min": -158.51565551757812, "p10": -18.680526733398438, "median": 40.03156280517578, "p90": 155.94975891113282, "max": 203.46441650390625, "pos_frac": 0.78125, "sample": [44.05305480957031, 174.89857482910156, 163.3236846923828, 166.56625366210938, 1.9415206909179688, 46.8957633972168, 25.630525588989258, 26.647212982177734, 52.33900451660156, 109.73236846923828, -3.9481964111328125, -14.689308166503906, -3.5142059326171875, 105.85240936279297, -158.51565551757812, 20.66260528564453, 98.96034240722656, -19.256118774414062, 85.24793243408203, 11.540288925170898, -8.652473449707031, 106.8931884765625, 36.01007080078125, -60.92071533203125, 124.15879821777344, 122.1295394897461, 157.37725830078125, 16.826595306396484, -10.783039093017578, 30.27867889404297, 161.16976928710938, 5.3109588623046875, 103.46739196777344, 34.310794830322266, 3.5686721801757812, -34.775848388671875, 121.07797241210938, 12.482561111450195, 94.47502899169922, -11.00604248046875, -117.93919372558594, 116.31483459472656, 47.8839111328125, 51.08333206176758, -17.337478637695312, 79.39675903320312, 12.284292221069336, 83.27557373046875, 32.87474060058594, 3.75360107421875, 115.71163940429688, 13.658134460449219, -19.869083404541016, 139.66690063476562, 179.07334899902344, 23.79082489013672, 92.38236999511719, 106.67967224121094, 73.29761505126953, 149.2235870361328, 203.46441650390625, -87.05171203613281, 152.61892700195312, 26.1356201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000554.npy"}
|
|
{"epoch": 0.8374905517762661, "step": 555, "batch_size": 64, "mean": 45.452369689941406, "std": 79.89541625976562, "min": -176.26904296875, "p10": -58.26800117492675, "median": 44.68128967285156, "p90": 143.1341506958008, "max": 169.68971252441406, "pos_frac": 0.765625, "sample": [5.270025253295898, -167.54391479492188, 38.60669708251953, 39.88690185546875, 119.5920181274414, 14.089408874511719, -117.05781555175781, -5.74163818359375, -60.62411880493164, 73.09416198730469, 161.743408203125, -44.22889709472656, 69.87451171875, 15.63498306274414, 123.74322509765625, 165.34820556640625, 33.05200958251953, 8.823143005371094, 17.552627563476562, 143.61810302734375, 69.84101104736328, -6.1888885498046875, 169.68971252441406, 49.475677490234375, 145.54019165039062, 38.65672302246094, 74.42996215820312, 136.68519592285156, 72.82569885253906, 154.33352661132812, 114.705078125, 105.35984802246094, 88.91172790527344, 122.41690826416016, -65.3115234375, 142.0049285888672, -176.26904296875, 83.11647033691406, -52.77039337158203, 39.78254318237305, 84.64999389648438, 58.05474853515625, -13.363731384277344, 15.203468322753906, -37.60289001464844, 30.128856658935547, 66.60078430175781, 70.08454132080078, 136.23033142089844, -20.485336303710938, 82.85406494140625, 8.380516052246094, -135.28919982910156, -5.729362487792969, 129.3394012451172, 25.677520751953125, 84.84982299804688, 26.7174072265625, 73.17611694335938, -60.79827880859375, 164.05844116210938, 38.163116455078125, 136.0810546875, 10.001775741577148], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000555.npy"}
|
|
{"epoch": 0.8390022675736961, "step": 556, "batch_size": 64, "mean": 40.992000579833984, "std": 80.38300323486328, "min": -143.12423706054688, "p10": -57.183563995361325, "median": 24.92243003845215, "p90": 146.92704315185546, "max": 231.80462646484375, "pos_frac": 0.6875, "sample": [96.5258560180664, 24.938873291015625, 29.481918334960938, -12.16348648071289, -4.2159881591796875, 130.66552734375, 171.95152282714844, 51.26704025268555, -12.69862174987793, -40.620025634765625, 13.26896858215332, 7.3893280029296875, 88.46265411376953, 131.91778564453125, 8.72793960571289, -28.55059814453125, 34.23661804199219, -4.724922180175781, 92.38424682617188, 146.12525939941406, -13.810646057128906, 110.07667541503906, 130.23696899414062, 24.29668426513672, -51.91020965576172, 18.74903106689453, -125.30921936035156, 34.580772399902344, 105.97554016113281, -71.94296264648438, -8.121313095092773, -143.12423706054688, 81.64085388183594, 6.041709899902344, 15.505294799804688, 5.817289352416992, 108.11616516113281, 134.1717987060547, 6.790428161621094, 13.630613327026367, 159.4751434326172, 231.80462646484375, 27.830921173095703, 41.09013366699219, 165.77232360839844, 137.94879150390625, -42.05064392089844, 24.69927978515625, -89.42547607421875, -6.866798400878906, 169.85073852539062, -59.443572998046875, -23.79071807861328, 130.98695373535156, 24.905986785888672, -62.2789306640625, 34.240386962890625, -64.86737823486328, 58.576446533203125, 146.8275909423828, 166.01736450195312, 146.96966552734375, 50.846092224121094, -51.41202926635742], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000556.npy"}
|
|
{"epoch": 0.8405139833711263, "step": 557, "batch_size": 64, "mean": 58.25011444091797, "std": 67.8838119506836, "min": -46.299224853515625, "p10": -23.926494216918943, "median": 54.290931701660156, "p90": 157.6219421386719, "max": 174.8380889892578, "pos_frac": 0.75, "sample": [168.557373046875, 93.1292495727539, 68.61285400390625, 17.691787719726562, 5.919696807861328, -31.96344757080078, 165.82908630371094, 20.68505859375, 143.52056884765625, 78.8092041015625, 9.702713012695312, -8.85650634765625, -2.3360595703125, 97.39783477783203, -3.7870140075683594, 47.83415603637695, 75.33006286621094, 8.743148803710938, 128.7717742919922, 84.48236846923828, -2.474506378173828, -16.86053466796875, 151.32928466796875, -34.60675048828125, -45.307952880859375, -46.299224853515625, 158.95114135742188, 174.8380889892578, 24.243038177490234, -32.99311065673828, 154.52047729492188, 70.85541534423828, 116.54007720947266, -20.81374740600586, 0.2747344970703125, 164.26605224609375, 69.80213928222656, 126.01184844970703, 52.61346435546875, 171.91433715820312, 6.3048858642578125, 23.603944778442383, -16.843162536621094, -20.309722900390625, 0.2862834930419922, -25.260528564453125, 132.7495880126953, 134.50051879882812, 129.01548767089844, 19.51305389404297, 6.7525787353515625, 61.425941467285156, 33.12217712402344, 110.01947021484375, 122.64762115478516, 61.627777099609375, 133.63255310058594, 61.07753372192383, 35.245174407958984, 55.96839904785156, -27.37115478515625, -15.861831665039062, 170.67626953125, 130.6062469482422], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000557.npy"}
|
|
{"epoch": 0.8420256991685563, "step": 558, "batch_size": 64, "mean": 64.52188110351562, "std": 72.5560302734375, "min": -97.5450668334961, "p10": -16.89672164916992, "median": 57.61283493041992, "p90": 156.9566619873047, "max": 221.28512573242188, "pos_frac": 0.78125, "sample": [110.7489242553711, 155.8387451171875, 160.79727172851562, 7.157432556152344, 146.54153442382812, 6.422000885009766, 155.92193603515625, -32.46804428100586, 3.074596405029297, 94.12318420410156, -12.661140441894531, 130.57046508789062, -13.192794799804688, 25.785789489746094, -1.00128173828125, 128.7311553955078, 136.8605499267578, 10.600677490234375, 61.90793228149414, 65.42710876464844, 221.28512573242188, 121.32421875, 29.564212799072266, 149.22021484375, 107.02436828613281, -30.930023193359375, 145.05511474609375, 21.219017028808594, 119.18534088134766, 68.99427032470703, 149.83038330078125, 19.195083618164062, 57.238006591796875, 183.84178161621094, 6.1764678955078125, 1.959035873413086, 74.59538269042969, 7.100341796875, 36.827423095703125, 197.43887329101562, 114.21820068359375, -22.304168701171875, 198.90914916992188, -17.980255126953125, 159.59292602539062, 157.40011596679688, -12.482711791992188, -20.096031188964844, 138.5975799560547, 26.1162109375, 63.551300048828125, -97.5450668334961, 57.98766326904297, -23.841339111328125, 22.586517333984375, -4.01763916015625, 103.73199462890625, 17.301673889160156, -14.368476867675781, -3.27740478515625, 132.69918823242188, 45.37714385986328, 69.2286605834961, 10.684844970703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000558.npy"}
|
|
{"epoch": 0.8435374149659864, "step": 559, "batch_size": 64, "mean": 54.650325775146484, "std": 82.44042205810547, "min": -112.20875549316406, "p10": -25.700985527038572, "median": 35.02008056640625, "p90": 165.03953094482424, "max": 310.87677001953125, "pos_frac": 0.734375, "sample": [1.6257476806640625, -22.34789276123047, 9.03934097290039, -11.02151870727539, 91.0055923461914, 310.87677001953125, 178.18820190429688, 196.001708984375, 23.163734436035156, -2.121265411376953, -111.09718322753906, 117.31649780273438, 50.12133026123047, 80.01094055175781, 13.602109909057617, 160.67958068847656, 100.6894760131836, 39.967918395996094, 128.62831115722656, 94.48435974121094, 144.97695922851562, 27.35418701171875, -16.643096923828125, 104.20010375976562, 1.5634536743164062, 19.15430450439453, -72.08172607421875, 148.975830078125, 38.000457763671875, 166.9080810546875, 32.039703369140625, 53.05485534667969, 9.678153991699219, 152.99520874023438, 167.28604125976562, -27.138025283813477, 85.73153686523438, 0.35893821716308594, -11.804893493652344, -14.371025085449219, -4.2900848388671875, -18.435508728027344, -18.933040618896484, 41.535491943359375, 84.92727661132812, 154.7711639404297, 95.28778839111328, -34.301918029785156, 133.44187927246094, -7.0068817138671875, 15.752622604370117, 172.94712829589844, -88.9837646484375, 5.7728729248046875, 117.6329345703125, 179.50144958496094, 13.294475555419922, 92.43573760986328, 7.799072265625, 10.697099685668945, 106.787353515625, 131.61863708496094, -41.475120544433594, -112.20875549316406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000559.npy"}
|
|
{"epoch": 0.8450491307634165, "step": 560, "batch_size": 64, "mean": 50.07002258300781, "std": 70.01318359375, "min": -63.54411315917969, "p10": -41.985111999511716, "median": 35.19263458251953, "p90": 154.56063842773438, "max": 171.85743713378906, "pos_frac": 0.71875, "sample": [68.43460845947266, 48.1514892578125, 149.3164825439453, 35.438804626464844, -55.37751770019531, -44.66002655029297, 164.24072265625, 34.61460876464844, 161.32302856445312, 2.5881099700927734, 68.04673767089844, 7.840400695800781, 45.31776428222656, 43.393463134765625, 7.814605712890625, -39.218597412109375, -0.9925689697265625, -8.111028671264648, 18.553592681884766, 103.05551147460938, -48.84318542480469, 117.72830963134766, 39.12092590332031, 4.080574035644531, 155.03904724121094, 111.63821411132812, -16.70117950439453, 141.34791564941406, 153.44435119628906, 22.63045310974121, 37.560272216796875, -10.372230529785156, -28.17334747314453, 171.85743713378906, 122.41120147705078, -59.50201416015625, 23.293113708496094, -43.17076110839844, 160.33543395996094, -38.723533630371094, 163.44772338867188, 113.86676025390625, 34.94646453857422, 85.16995239257812, 6.102546691894531, 97.5879898071289, -2.1932601928710938, 157.9296875, 28.365215301513672, -7.07781982421875, -14.685150146484375, 129.66787719726562, -63.54411315917969, -2.7129058837890625, 104.7891845703125, 109.9874038696289, -61.9276123046875, 31.073810577392578, 138.7130126953125, 38.1005744934082, 147.03604125976562, 90.70314025878906, 25.458377838134766, 28.905384063720703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000560.npy"}
|
|
{"epoch": 0.8465608465608465, "step": 561, "batch_size": 64, "mean": 52.991817474365234, "std": 74.41658020019531, "min": -75.91769409179688, "p10": -37.650828933715815, "median": 35.47640609741211, "p90": 160.30832214355473, "max": 214.2837371826172, "pos_frac": 0.734375, "sample": [3.65765380859375, 10.440177917480469, 8.360870361328125, -0.7867183685302734, 35.52238464355469, 99.1777114868164, 54.88962173461914, -20.07025909423828, 146.00686645507812, 27.879501342773438, 39.758209228515625, 108.45042419433594, 7.219898223876953, 105.48060607910156, 3.845672607421875, -75.91769409179688, 22.567703247070312, 141.51416015625, 172.34011840820312, -3.42474365234375, -27.638916015625, 121.52362060546875, 101.48876953125, -33.796382904052734, 79.09002685546875, -4.4882659912109375, 104.51618194580078, -41.72610855102539, -63.723480224609375, 197.40386962890625, -20.51882553100586, 39.26966857910156, 149.3284912109375, 164.3916473388672, 38.96134948730469, 27.399932861328125, 46.39582824707031, 34.59864044189453, 13.986722946166992, 201.55767822265625, 214.2837371826172, 170.2129364013672, 134.96060180664062, 170.5578155517578, 107.12608337402344, -66.44554901123047, -39.856529235839844, 35.43042755126953, 71.43980407714844, 16.484968185424805, -16.117767333984375, 39.135704040527344, -0.33113861083984375, 4.563692092895508, 102.88070678710938, 150.7805633544922, 12.17011833190918, 101.37362670898438, 100.32522583007812, -60.099822998046875, -4.178571701049805, 20.857492446899414, 150.29237365722656, -39.302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000561.npy"}
|
|
{"epoch": 0.8480725623582767, "step": 562, "batch_size": 64, "mean": 59.33275604248047, "std": 75.72676849365234, "min": -84.59882354736328, "p10": -34.71458358764648, "median": 57.708518981933594, "p90": 150.1046157836914, "max": 231.75250244140625, "pos_frac": 0.703125, "sample": [15.338138580322266, 140.143798828125, -70.65914916992188, 134.35189819335938, 140.51539611816406, 132.90695190429688, 6.548759460449219, 226.4453125, 148.626953125, -31.400184631347656, 62.88514709472656, 51.63836669921875, -65.98175048828125, 11.467037200927734, 68.30072021484375, -1.3511924743652344, 11.1871337890625, 70.22607421875, 91.26193237304688, -40.53065490722656, 99.60648345947266, 150.0271453857422, -40.65447235107422, -20.169349670410156, -8.55731201171875, 157.3660888671875, -1.3257198333740234, -4.622200012207031, 47.5518798828125, 19.881053924560547, 132.29196166992188, -36.135040283203125, 141.74966430664062, 148.22149658203125, 55.89942932128906, 231.75250244140625, -20.10550308227539, -2.5944061279296875, -0.14792251586914062, 177.4384765625, 8.158843994140625, 2.7203750610351562, 71.0514144897461, 92.26788330078125, 95.30160522460938, 159.26611328125, 146.66448974609375, 100.00732421875, 117.45650482177734, -5.785919189453125, 150.1378173828125, 59.517608642578125, -9.95195198059082, 35.403533935546875, 106.53598022460938, 62.46581268310547, 133.9036865234375, 162.3068389892578, 61.48676300048828, -84.59882354736328, -1.676361083984375, -44.8704833984375, 7.927574157714844, 42.20481872558594], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000562.npy"}
|
|
{"epoch": 0.8495842781557067, "step": 563, "batch_size": 64, "mean": 51.89995193481445, "std": 78.37530517578125, "min": -155.49880981445312, "p10": -40.33888549804687, "median": 52.72649002075195, "p90": 155.9464874267578, "max": 185.1681365966797, "pos_frac": 0.734375, "sample": [53.60210418701172, 108.58100128173828, 38.75045394897461, 0.45130157470703125, 148.8694305419922, 106.52853393554688, -155.49880981445312, 154.853271484375, 85.08705139160156, -18.07892608642578, 15.063365936279297, 97.35151672363281, 6.778739929199219, -39.40272521972656, 143.6328125, 46.35362243652344, -93.88398742675781, 68.93859100341797, -23.12908935546875, 185.1681365966797, 38.6966552734375, 175.579833984375, -38.73689270019531, 130.59228515625, 126.01093292236328, 63.35162353515625, 100.99311828613281, 13.066749572753906, 71.33914184570312, -83.53560638427734, -5.4294891357421875, -23.099552154541016, -1.0418548583984375, 23.308250427246094, -15.338565826416016, 111.88136291503906, 37.752506256103516, -86.8551025390625, 81.65978240966797, 22.605541229248047, 164.44329833984375, 162.6358642578125, 144.1011199951172, -40.74009704589844, -76.00274658203125, 51.85087585449219, 49.18360900878906, -21.813987731933594, 156.41500854492188, 83.9927978515625, -75.36160278320312, 71.80018615722656, 57.45742416381836, 115.53235626220703, 93.46879577636719, 47.98480224609375, 175.79843139648438, 176.45834350585938, 28.229629516601562, 131.4247283935547, 75.93232727050781, -33.11851501464844, 100.29753112792969, 8.809707641601562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000563.npy"}
|
|
{"epoch": 0.8510959939531368, "step": 564, "batch_size": 64, "mean": 67.04069519042969, "std": 80.26325988769531, "min": -136.0011749267578, "p10": -38.18096160888671, "median": 64.53370666503906, "p90": 174.4405700683594, "max": 198.47682189941406, "pos_frac": 0.8125, "sample": [159.98843383789062, 3.6095046997070312, 175.01303100585938, 130.379638671875, 94.55191802978516, 188.48825073242188, 117.45127868652344, -126.91972351074219, -9.927650451660156, 5.1142425537109375, 173.10482788085938, 138.06329345703125, 42.93339157104492, -39.93971252441406, 16.081405639648438, 8.798721313476562, -16.327186584472656, 101.26469421386719, 97.94683074951172, 51.54778289794922, 51.4042854309082, 0.7056369781494141, 91.9892578125, 85.0980224609375, 138.55149841308594, 121.98168182373047, 181.56854248046875, -60.583160400390625, 2.935365676879883, 84.55116271972656, 184.86761474609375, 198.47682189941406, 178.728759765625, 74.07925415039062, 49.16056823730469, -76.05034637451172, 119.89414978027344, 42.956783294677734, 50.689422607421875, 9.852439880371094, 151.56044006347656, 65.95695495605469, 57.82453918457031, -9.635042190551758, -55.044979095458984, 46.328155517578125, 56.97663116455078, 130.3946533203125, 59.26854705810547, -60.19480895996094, 84.09536743164062, 87.85995483398438, 150.362060546875, -8.285116195678711, 143.85870361328125, 19.391502380371094, 152.6377410888672, 181.32589721679688, -136.0011749267578, 161.47055053710938, -34.07720947265625, 20.29507827758789, 63.11045837402344, 119.04499816894531], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000564.npy"}
|
|
{"epoch": 0.8526077097505669, "step": 565, "batch_size": 64, "mean": 45.354454040527344, "std": 71.05998229980469, "min": -108.5728759765625, "p10": -34.26854705810546, "median": 39.75825119018555, "p90": 141.1159439086914, "max": 225.9752197265625, "pos_frac": 0.734375, "sample": [22.137245178222656, -19.48769760131836, -17.29145050048828, 29.902267456054688, 13.384925842285156, 32.350555419921875, 64.95858764648438, 37.58245086669922, -3.5018310546875, 17.380958557128906, 29.565330505371094, 87.19287109375, 173.89071655273438, -108.5728759765625, -9.0045166015625, 113.06832885742188, 57.856937408447266, 91.54165649414062, 225.9752197265625, 173.60302734375, 136.01416015625, -63.58006286621094, 2.774871826171875, 7.784339904785156, 55.090049743652344, 106.00968933105469, 42.71736145019531, 41.934051513671875, 56.88141632080078, 79.54578399658203, 145.58493041992188, 8.072744369506836, -30.352142333984375, -62.72012710571289, 88.75706481933594, -48.80250930786133, 21.81492805480957, 104.46086120605469, -90.75089263916016, 55.6461181640625, 83.17147827148438, -22.57950210571289, -29.58465576171875, 84.5697021484375, -35.94700622558594, 90.77473449707031, 13.940032958984375, 139.931884765625, 182.35084533691406, 14.638389587402344, 115.90985107421875, -13.704193115234375, -92.06854248046875, -0.4612464904785156, 23.173736572265625, 19.59830093383789, -25.77374267578125, 105.71688842773438, 45.281272888183594, 72.73284912109375, 165.01170349121094, 92.28186798095703, 62.681724548339844, 141.62339782714844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000565.npy"}
|
|
{"epoch": 0.854119425547997, "step": 566, "batch_size": 64, "mean": 46.76872634887695, "std": 86.159912109375, "min": -187.97914123535156, "p10": -53.55961303710937, "median": 49.725318908691406, "p90": 157.98909454345704, "max": 215.9680633544922, "pos_frac": 0.6875, "sample": [66.58818054199219, 30.102012634277344, -39.19947814941406, 100.69059753417969, 4.1847381591796875, 50.00172805786133, 160.141357421875, 129.5345458984375, 4.473257064819336, 108.49006652832031, 56.85654830932617, 60.19660949707031, 47.427154541015625, 49.448909759521484, 20.864585876464844, 200.47921752929688, -187.97914123535156, 97.59888458251953, 148.0714111328125, 112.33415985107422, 161.7906036376953, -10.456771850585938, -38.41560745239258, 103.36727142333984, 2.389291763305664, -40.44880676269531, 90.62120056152344, 146.228515625, 144.922119140625, -0.4943389892578125, 141.888916015625, 99.53384399414062, 117.78291320800781, 55.3167724609375, 15.004364013671875, -153.38702392578125, -55.015846252441406, 33.72614669799805, 190.8850555419922, -50.16173553466797, 30.2479248046875, -40.56752014160156, 215.9680633544922, -42.2786865234375, 152.96714782714844, 10.800132751464844, -55.572540283203125, -92.70333099365234, 54.300575256347656, -1.7956390380859375, -30.382366180419922, -0.33344268798828125, -23.114608764648438, 144.93612670898438, -21.497798919677734, 43.18156814575195, -64.36702728271484, 163.40060424804688, 50.56194305419922, -90.72943115234375, 105.60530090332031, 71.15442657470703, 68.38650512695312, 169.64828491210938], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000566.npy"}
|
|
{"epoch": 0.8556311413454271, "step": 567, "batch_size": 64, "mean": 53.4029541015625, "std": 69.26298522949219, "min": -85.39898681640625, "p10": -42.76461563110351, "median": 53.620628356933594, "p90": 153.3176696777344, "max": 182.70867919921875, "pos_frac": 0.78125, "sample": [98.28184509277344, 8.774456024169922, -5.800079345703125, 154.4124755859375, -21.675750732421875, 85.03294372558594, 169.31234741210938, 126.28996276855469, 157.6560821533203, 8.413230895996094, 125.14646911621094, 51.825897216796875, 51.941978454589844, -44.21037292480469, 61.42925262451172, -5.512123107910156, -5.019336700439453, 32.760955810546875, 11.211921691894531, 50.62137222290039, 80.4019775390625, 6.1857147216796875, 1.3836040496826172, 130.14373779296875, -39.39118194580078, 0.8836536407470703, -85.39898681640625, -51.452484130859375, 97.90482330322266, -60.113304138183594, 136.2962188720703, 57.589813232421875, 150.76312255859375, 182.70867919921875, 164.18515014648438, 2.478618621826172, 71.21162414550781, 23.909683227539062, 74.87814331054688, 167.95887756347656, -64.45044708251953, 84.48040771484375, 30.52996826171875, 48.608978271484375, 118.4542236328125, 108.05484008789062, -3.2724609375, 6.372802734375, 136.48599243164062, 96.25865173339844, 71.47908782958984, 13.55978012084961, 128.4891357421875, 3.663818359375, 11.245376586914062, -71.92749786376953, -52.36616134643555, 82.14396667480469, 168.48463439941406, -13.31640625, 88.85026550292969, 55.299278259277344, 60.53862762451172, 86.7011947631836], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000567.npy"}
|
|
{"epoch": 0.8571428571428571, "step": 568, "batch_size": 64, "mean": 63.64884567260742, "std": 73.98210906982422, "min": -59.36931610107422, "p10": -28.405567932128903, "median": 57.655025482177734, "p90": 158.1019027709961, "max": 280.25518798828125, "pos_frac": 0.796875, "sample": [87.15775299072266, 28.796998977661133, 69.45465087890625, 18.384613037109375, 75.501708984375, 103.78440856933594, 95.3460693359375, -59.36931610107422, 23.819522857666016, -31.28466033935547, -53.44208526611328, 143.8450469970703, 29.16655158996582, 177.13375854492188, -16.043296813964844, -54.26875305175781, 38.923973083496094, 4.510650634765625, 62.95867156982422, 85.79523468017578, -58.22550964355469, 7.1085205078125, 16.97188949584961, 146.72430419921875, 156.7454071044922, 139.17926025390625, 110.62517547607422, 195.73306274414062, -54.69953536987305, 26.996017456054688, 94.75569152832031, 52.35137939453125, 63.513572692871094, -14.332710266113281, 112.96202087402344, 22.594865798950195, -0.0201263427734375, 80.9954605102539, -4.36473274230957, 124.50286865234375, 31.17877960205078, 135.2618408203125, 71.62130737304688, 73.72128295898438, 127.40576171875, 29.278053283691406, 33.05815124511719, -0.55303955078125, 21.742019653320312, 29.543182373046875, 67.81524658203125, 118.29403686523438, -25.9185791015625, 110.74027252197266, 142.79066467285156, 41.47174072265625, 1.918670654296875, 226.23089599609375, 158.68325805664062, 177.95167541503906, -29.471420288085938, 35.87871551513672, 164.33985900878906, 280.25518798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000568.npy"}
|
|
{"epoch": 0.8586545729402872, "step": 569, "batch_size": 64, "mean": 39.753517150878906, "std": 73.62667083740234, "min": -128.61257934570312, "p10": -43.639263916015615, "median": 32.3258171081543, "p90": 151.15475311279297, "max": 188.56883239746094, "pos_frac": 0.71875, "sample": [145.91783142089844, -13.385391235351562, 14.51350212097168, 147.51528930664062, 34.324310302734375, 157.45068359375, 2.9914703369140625, 156.8992919921875, 60.111358642578125, 161.85537719726562, 82.69917297363281, -21.29131317138672, 10.484722137451172, 40.22282409667969, -128.61257934570312, 48.521240234375, 15.61355209350586, 188.56883239746094, -7.179067611694336, 65.24616241455078, 138.9334716796875, 89.50677490234375, 31.522747039794922, 171.662109375, -24.036026000976562, 141.5349884033203, 66.37322998046875, 19.671398162841797, 24.141937255859375, -93.01615142822266, -101.46367645263672, 35.05803298950195, -0.010223388671875, 18.337764739990234, -13.202033996582031, -9.117324829101562, 172.61050415039062, 9.71837043762207, 39.668548583984375, 13.589853286743164, 43.50196075439453, -28.601593017578125, -73.85557556152344, 2.7198867797851562, 66.65338897705078, -86.20272064208984, 27.340133666992188, -28.630050659179688, 33.12888717651367, 83.51122283935547, 43.36832809448242, 57.66230773925781, 135.03399658203125, -33.13500213623047, -48.141090393066406, 30.753259658813477, 85.52130126953125, 38.10606384277344, 147.94680786132812, -18.018768310546875, -65.54715728759766, 27.841690063476562, 152.5295867919922, 56.786521911621094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000569.npy"}
|
|
{"epoch": 0.8601662887377173, "step": 570, "batch_size": 64, "mean": 68.99894714355469, "std": 74.68815612792969, "min": -105.87970733642578, "p10": -16.90215167999267, "median": 66.03765106201172, "p90": 169.29181365966798, "max": 283.5580749511719, "pos_frac": 0.765625, "sample": [148.522705078125, -68.46881866455078, 111.9251480102539, 46.11147689819336, 89.21053314208984, 46.48948669433594, -23.748458862304688, -5.5947113037109375, 48.09437561035156, -12.511184692382812, 47.93408203125, 50.58924102783203, 30.458112716674805, 114.49984741210938, -3.6505050659179688, 119.05183410644531, -2.790742874145508, -0.8951759338378906, 54.13294982910156, 91.87220764160156, 89.40475463867188, -67.349365234375, 18.97826385498047, 140.27040100097656, 172.14016723632812, 2.650876998901367, 51.386287689208984, 88.78215026855469, 169.4447021484375, -10.721305847167969, 31.35796356201172, 101.51187896728516, 128.737548828125, -54.08689880371094, 30.530784606933594, 74.84081268310547, 139.91668701171875, 78.02302551269531, 158.8634033203125, 57.23448944091797, 178.36956787109375, 126.12931823730469, 181.19369506835938, -18.783994674682617, 27.330726623535156, 86.83988952636719, 139.32952880859375, 168.93507385253906, 85.93551635742188, -105.87970733642578, 134.7917938232422, -30.20117950439453, 38.5634765625, 135.39271545410156, 171.317626953125, 19.880096435546875, 76.4671630859375, -2.493009567260742, 95.2962646484375, 139.31771850585938, 283.5580749511719, 34.49794387817383, 171.07235717773438, -4.076995849609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000570.npy"}
|
|
{"epoch": 0.8616780045351474, "step": 571, "batch_size": 64, "mean": 38.720985412597656, "std": 70.84283447265625, "min": -124.20176696777344, "p10": -64.23846054077147, "median": 53.21860122680664, "p90": 117.92451095581056, "max": 175.83883666992188, "pos_frac": 0.75, "sample": [84.60357666015625, 52.83501434326172, 36.94526672363281, 16.87995719909668, 0.8234405517578125, 97.92628479003906, 47.648406982421875, 74.78680419921875, 100.48286437988281, 175.83883666992188, 61.080543518066406, 65.33634948730469, -69.99981689453125, 9.730724334716797, -3.3475494384765625, 97.10160827636719, 18.568267822265625, 113.78473663330078, 5.116172790527344, 95.4714126586914, 120.4158935546875, 5.830223083496094, 95.04034423828125, 55.49440002441406, 125.85094451904297, 95.07476806640625, 81.70700073242188, 75.52896118164062, -115.97218322753906, 38.575050354003906, 1.4028491973876953, -77.32933044433594, 0.32437896728515625, -83.2553939819336, 69.06333923339844, 0.5583286285400391, -50.79529571533203, -0.6937065124511719, -48.92247009277344, -14.379592895507812, 79.35192108154297, 71.31869506835938, 113.60629272460938, 99.15441131591797, 48.81692123413086, 153.60903930664062, 134.0638427734375, -124.20176696777344, 29.80150604248047, -11.800460815429688, -45.54167175292969, -108.17556762695312, 71.6171646118164, 53.60218811035156, 83.28732299804688, -22.58031463623047, 80.55490112304688, 119.69869995117188, 69.99488067626953, 102.11735534667969, -29.37533950805664, 4.748439788818359, -114.09559631347656, 163.43899536132812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000571.npy"}
|
|
{"epoch": 0.8631897203325775, "step": 572, "batch_size": 64, "mean": 47.77949905395508, "std": 65.07282257080078, "min": -79.59829711914062, "p10": -38.048292541503905, "median": 42.16059494018555, "p90": 140.64554290771486, "max": 174.26718139648438, "pos_frac": 0.75, "sample": [5.474555969238281, -55.646453857421875, 2.2941207885742188, 78.00257873535156, 112.3607177734375, 46.76817321777344, 144.42098999023438, 43.40522003173828, -13.042621612548828, 170.24429321289062, -0.0282440185546875, 96.5350570678711, 104.66705322265625, 142.3280029296875, -14.933879852294922, 134.51116943359375, -39.394195556640625, 15.638483047485352, 16.9034481048584, 100.58195495605469, 77.5479736328125, 46.514984130859375, 67.29313659667969, 136.7198028564453, 131.27438354492188, 49.52899932861328, -13.612518310546875, 68.94689178466797, -15.823997497558594, 8.852371215820312, 174.26718139648438, 146.27626037597656, 153.92559814453125, -79.59829711914062, 95.51293182373047, 40.08839797973633, 40.91596984863281, -53.86085510253906, 5.511936187744141, 63.82610321044922, 82.43608856201172, 18.118318557739258, 105.22747039794922, 87.14601135253906, 29.515342712402344, 101.36324310302734, 16.689285278320312, -68.53471374511719, 169.0152587890625, 39.55088806152344, 64.28819274902344, 110.16127014160156, -10.916763305664062, 30.3896484375, -60.19036865234375, -34.90785217285156, 18.89557647705078, 71.55961608886719, -12.841949462890625, -59.64501953125, 92.5362548828125, -2.381765365600586, 16.98712730407715, 18.229257583618164], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000572.npy"}
|
|
{"epoch": 0.8647014361300076, "step": 573, "batch_size": 64, "mean": 47.72744369506836, "std": 85.88606262207031, "min": -145.57391357421875, "p10": -39.31252059936523, "median": 41.50811958312988, "p90": 171.06248779296877, "max": 248.9026641845703, "pos_frac": 0.734375, "sample": [35.387054443359375, 79.30087280273438, -138.10377502441406, 137.96585083007812, -32.32257843017578, 36.4755859375, -5.832395553588867, 29.686134338378906, 44.089935302734375, -42.17073059082031, -145.57391357421875, -105.26361846923828, -17.755704879760742, 0.9306182861328125, -13.888938903808594, 102.91888427734375, -8.404434204101562, 31.582733154296875, 166.54415893554688, -140.34390258789062, 45.586341857910156, 43.18791961669922, 169.72332763671875, 53.198020935058594, -19.910125732421875, 10.95401382446289, 248.9026641845703, 9.762264251708984, 131.56137084960938, 42.308265686035156, 31.160293579101562, 21.373291015625, 63.943634033203125, 44.31935119628906, 125.53680419921875, 24.08376693725586, 71.53294372558594, 40.70797348022461, 32.83765411376953, 205.49935913085938, 134.30691528320312, 122.38035583496094, 110.65847778320312, -2.3763561248779297, -27.266708374023438, 55.441932678222656, 193.77455139160156, 28.934219360351562, 30.396766662597656, 52.390594482421875, 179.0499267578125, 192.0274658203125, 59.92376708984375, 97.1710205078125, -7.0847625732421875, -94.41265869140625, 217.0909423828125, -32.64336395263672, 49.59305191040039, 81.03650665283203, 87.12203979492188, 18.245399475097656, -74.33108520507812, 171.63641357421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000573.npy"}
|
|
{"epoch": 0.8662131519274376, "step": 574, "batch_size": 64, "mean": 61.10847473144531, "std": 64.33439636230469, "min": -96.01526641845703, "p10": -4.5766059875488265, "median": 57.28807067871094, "p90": 160.5542175292969, "max": 238.03872680664062, "pos_frac": 0.8125, "sample": [48.96075439453125, -23.95069122314453, -8.951072692871094, -96.01526641845703, 69.17634582519531, 88.06607818603516, 27.380172729492188, 17.262557983398438, 18.020326614379883, -15.740158081054688, -8.14996337890625, 73.29411315917969, 15.257492065429688, 15.984407424926758, -5.3714141845703125, 0.7159194946289062, -1.6787452697753906, 54.94038391113281, 47.70433807373047, 86.37926483154297, 147.54478454589844, 41.21729278564453, 80.89666748046875, 100.83819580078125, -2.7220535278320312, 173.1675567626953, 35.30030059814453, 129.9163360595703, 1.2567157745361328, 158.38385009765625, 88.12298583984375, 0.222412109375, 66.18365478515625, 28.136978149414062, 76.09054565429688, 90.20823669433594, -49.35942840576172, 107.99769592285156, 162.7662353515625, 5.11534309387207, 77.86245727539062, 24.40557861328125, 161.484375, 102.61427307128906, 121.4705810546875, 164.74368286132812, 81.55787658691406, 81.27593994140625, 59.63575744628906, 22.12012481689453, -2.426177978515625, 164.5233612060547, 20.936195373535156, 72.46549987792969, 88.86963653564453, 3.8364791870117188, 50.17876052856445, 156.55224609375, 124.04962921142578, 168.5238494873047, -1.1752395629882812, 85.13166809082031, 238.03872680664062, -0.3020172119140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000574.npy"}
|
|
{"epoch": 0.8677248677248677, "step": 575, "batch_size": 64, "mean": 40.411102294921875, "std": 82.38896179199219, "min": -141.65553283691406, "p10": -70.8141544342041, "median": 29.261615753173828, "p90": 161.46279602050785, "max": 214.43185424804688, "pos_frac": 0.6875, "sample": [49.119834899902344, -97.40263366699219, -121.98654174804688, -8.292383193969727, 210.42633056640625, 51.36396789550781, -9.094123840332031, 76.59700775146484, 6.543426513671875, -23.043018341064453, 101.36555480957031, -41.153282165527344, 176.05316162109375, -36.563968658447266, -88.38365173339844, 77.53837585449219, -75.5422134399414, 68.04920959472656, 9.506355285644531, 13.043609619140625, 42.096778869628906, 149.708251953125, 99.91316986083984, 21.5596923828125, -1.9633102416992188, 17.139297485351562, -91.742431640625, 93.93810272216797, 125.12179565429688, 3.7058334350585938, 17.923070907592773, 43.728599548339844, 166.50045776367188, 211.3056182861328, 133.80166625976562, -73.895751953125, 185.97073364257812, -2.723358154296875, 187.36553955078125, 121.66893005371094, -2.0181732177734375, 28.809097290039062, 60.21405792236328, 31.804107666015625, 5.6319427490234375, -3.2540969848632812, 18.65449333190918, -141.65553283691406, 62.13508605957031, 214.43185424804688, 131.0976104736328, 67.90711975097656, 22.780136108398438, 29.89952850341797, -63.62376022338867, 29.714134216308594, 40.880043029785156, -14.596080780029297, 27.77018165588379, 98.19474792480469, -24.2991943359375, 119.15567779541016, 87.75880432128906, -30.349170684814453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000575.npy"}
|
|
{"epoch": 0.8692365835222978, "step": 576, "batch_size": 64, "mean": 60.56700897216797, "std": 76.40544128417969, "min": -154.89524841308594, "p10": -25.805554008483885, "median": 52.23602294921875, "p90": 162.52512817382814, "max": 230.15391540527344, "pos_frac": 0.765625, "sample": [44.915130615234375, 142.36756896972656, -5.5647735595703125, 31.567716598510742, 89.32281494140625, 50.364013671875, 165.22998046875, 40.95689392089844, 57.83099365234375, 84.16117858886719, -19.3886775970459, -46.829837799072266, 127.94219970703125, -43.17071533203125, 75.62284088134766, -78.48701477050781, -37.68151092529297, 129.7311248779297, 230.15391540527344, 64.78616333007812, 141.9258575439453, 173.74169921875, 90.1041259765625, 1.0167369842529297, 108.04592895507812, 76.81795501708984, 43.78168487548828, -30.184526443481445, 195.50949096679688, -26.469528198242188, 25.985687255859375, 4.792716979980469, 38.57377624511719, 152.49044799804688, 54.1080322265625, 25.936996459960938, 193.3422393798828, -8.804977416992188, -2.715301513671875, 17.537086486816406, 143.29098510742188, -6.473278045654297, 147.70741271972656, 5.711156845092773, 156.21380615234375, 1.2799224853515625, 174.3270721435547, 114.29708862304688, -2.606172561645508, 165.28720092773438, 137.9584197998047, -154.89524841308594, 151.68727111816406, 73.43116760253906, 43.578285217285156, 116.09716796875, -24.25628089904785, -3.863330841064453, 57.332984924316406, 76.92056274414062, 0.6751155853271484, 28.627161026000977, 8.849266052246094, 85.74479675292969], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000576.npy"}
|
|
{"epoch": 0.8707482993197279, "step": 577, "batch_size": 64, "mean": 43.29014587402344, "std": 75.6006088256836, "min": -137.2376708984375, "p10": -29.07851734161376, "median": 41.63595962524414, "p90": 147.12607421875, "max": 193.00961303710938, "pos_frac": 0.75, "sample": [47.67871856689453, 8.606155395507812, 28.506568908691406, 42.39464569091797, -8.455673217773438, 73.55677795410156, 11.34958267211914, -21.119508743286133, 167.56759643554688, -11.408866882324219, -14.310997009277344, 142.86279296875, 96.97134399414062, -18.963546752929688, -60.29976272583008, 166.6241455078125, 130.64608764648438, 148.3014373779297, -137.2376708984375, 79.4830322265625, 156.65184020996094, 73.50775909423828, 20.681743621826172, 133.92404174804688, 0.2798309326171875, -68.33037567138672, -104.71409606933594, 102.61943054199219, 107.1113510131836, 50.271392822265625, 12.68341064453125, 118.00053405761719, -15.0167236328125, 42.8773193359375, 23.849884033203125, -16.556442260742188, 47.13259506225586, 30.20819091796875, 45.55363464355469, 0.13464736938476562, 29.167007446289062, 36.11065673828125, 144.38356018066406, 5.802022933959961, 165.7135772705078, 193.00961303710938, -132.32186889648438, 50.978271484375, 50.848785400390625, 62.89204406738281, -14.236312866210938, 76.23207092285156, 166.9127197265625, 140.49758911132812, 40.87727355957031, 27.216312408447266, -133.27252197265625, -14.860078811645508, -32.48952102661133, 95.5392074584961, 16.316001892089844, 73.68405151367188, 73.03788757324219, 14.908279418945312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000577.npy"}
|
|
{"epoch": 0.872260015117158, "step": 578, "batch_size": 64, "mean": 38.197776794433594, "std": 74.97052764892578, "min": -131.28268432617188, "p10": -39.70995674133301, "median": 24.58087158203125, "p90": 143.18692474365236, "max": 223.34490966796875, "pos_frac": 0.625, "sample": [59.66966247558594, -35.85474395751953, 49.240386962890625, 29.272674560546875, 48.84369659423828, -29.00358772277832, 49.79812240600586, 13.547309875488281, 99.806396484375, 4.258974075317383, -5.7321624755859375, 223.34490966796875, -12.738693237304688, 77.37760925292969, 135.7304229736328, -131.28268432617188, 130.11940002441406, 136.31491088867188, 90.92425537109375, 112.22908020019531, 140.44956970214844, 59.27214813232422, 11.153280258178711, 13.13437271118164, -14.198923110961914, 79.69951629638672, 135.4180908203125, -4.868110656738281, 62.12616729736328, -6.5312957763671875, 152.42080688476562, 107.11194610595703, 157.11703491210938, -51.846466064453125, -35.86878967285156, 23.466720581054688, 6.063089370727539, 89.82496643066406, 37.14374542236328, 54.20903015136719, 121.67868041992188, -89.92669677734375, 25.695022583007812, -37.725807189941406, -14.716690063476562, -9.878211975097656, -12.729026794433594, 151.55624389648438, -40.560306549072266, 165.73931884765625, 60.48456573486328, -66.52413177490234, 172.0762481689453, -5.593385696411133, -20.399246215820312, -71.07037353515625, 12.728134155273438, 36.131385803222656, -27.64270782470703, 3.6032238006591797, -18.122066497802734, -5.465629577636719, -90.20384979248047, 144.36007690429688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000578.npy"}
|
|
{"epoch": 0.873771730914588, "step": 579, "batch_size": 64, "mean": 61.96001052856445, "std": 67.34143829345703, "min": -112.38807678222656, "p10": -18.349396514892575, "median": 55.517887115478516, "p90": 155.49180450439454, "max": 187.21676635742188, "pos_frac": 0.8125, "sample": [81.11487579345703, 56.665435791015625, 55.31885528564453, 35.507362365722656, 34.29991912841797, -80.46798706054688, 149.110107421875, 40.25511169433594, 89.54501342773438, 173.33473205566406, 110.98506164550781, 155.9967803955078, 90.64736938476562, 171.37625122070312, 107.60203552246094, 25.62291717529297, 155.5228271484375, -19.251541137695312, -19.741615295410156, 94.23564147949219, 187.21676635742188, -52.34147262573242, 155.41941833496094, -2.8351802825927734, 50.38169860839844, 89.18281555175781, 167.0343017578125, 22.333206176757812, 153.5644989013672, 113.45527648925781, -32.63733673095703, 40.88265609741211, 128.25802612304688, 11.05925178527832, 88.94200897216797, 99.24542236328125, 68.45266723632812, 6.310930252075195, 2.9223785400390625, 150.6754150390625, -8.484214782714844, 54.452110290527344, -5.892295837402344, -14.371604919433594, 52.956783294677734, 8.066560745239258, -16.24439239501953, -112.38807678222656, 167.69232177734375, 95.76032257080078, 53.277313232421875, 68.48597717285156, 22.451507568359375, 134.1697235107422, 74.64811706542969, 9.207380294799805, 111.67216491699219, 122.00201416015625, 22.075429916381836, 55.7169189453125, -29.401153564453125, 15.934879302978516, 94.24713897705078, 34.20363235473633], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000579.npy"}
|
|
{"epoch": 0.8752834467120182, "step": 580, "batch_size": 64, "mean": 47.156192779541016, "std": 83.0543212890625, "min": -136.6153564453125, "p10": -53.776134109497065, "median": 38.15962219238281, "p90": 172.5503646850586, "max": 189.19496154785156, "pos_frac": 0.71875, "sample": [-94.32647705078125, -65.90994262695312, -4.0247039794921875, 19.551298141479492, -30.268142700195312, 58.22248840332031, 143.0686492919922, 69.94869995117188, 58.948944091796875, 10.157379150390625, 157.52378845214844, 72.94731903076172, 12.632537841796875, 3.768321990966797, -43.310218811035156, 37.40538024902344, 114.95809936523438, 81.10846710205078, 0.03410148620605469, 33.43641662597656, -3.4320030212402344, 79.5348892211914, 187.85256958007812, 52.55712127685547, 86.19335174560547, -17.677108764648438, 185.67681884765625, -4.679985046386719, 3.499826431274414, 183.2698211669922, 173.2242431640625, 163.48117065429688, 3.7023239135742188, 91.25179290771484, -25.79974365234375, 170.9779815673828, -49.80355453491211, -136.6153564453125, 14.075122833251953, 187.644775390625, -120.59617614746094, 22.07537841796875, -55.478668212890625, 84.39141845703125, 189.19496154785156, -28.28422737121582, 27.739980697631836, 36.70983123779297, 161.36428833007812, 73.52023315429688, 98.56588745117188, 117.52762603759766, -24.94481658935547, -116.04911804199219, 19.363948822021484, -18.463464736938477, 87.45668029785156, 111.17889404296875, 70.64308166503906, 176.02059936523438, 38.91386413574219, -73.72409057617188, 55.60395050048828, 104.45993041992188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000580.npy"}
|
|
{"epoch": 0.8767951625094482, "step": 581, "batch_size": 64, "mean": 45.74122619628906, "std": 75.48744201660156, "min": -100.9111328125, "p10": -52.91797142028808, "median": 39.38780975341797, "p90": 138.1748489379883, "max": 212.99667358398438, "pos_frac": 0.6875, "sample": [116.100830078125, 127.44141387939453, 11.481269836425781, 7.9303131103515625, 74.44029235839844, 6.5311737060546875, -24.988388061523438, -51.445037841796875, 70.93907165527344, 61.62561798095703, -72.68060302734375, 13.819477081298828, 17.25501823425293, 40.89771270751953, 98.79216766357422, -55.367584228515625, 114.006591796875, 34.820587158203125, 97.1232681274414, 7.485563278198242, 137.84483337402344, 14.777389526367188, 161.24386596679688, -8.6495361328125, 187.54661560058594, 170.87869262695312, 114.8179931640625, -12.9312744140625, 37.877906799316406, 72.13880920410156, -8.94378662109375, 89.0036392211914, 24.71926498413086, 107.98627471923828, 67.58820343017578, 104.98414611816406, 52.00627136230469, -100.9111328125, 205.51661682128906, 138.3162841796875, -16.0115966796875, -71.0479736328125, 58.99237060546875, -6.157562255859375, 66.0155029296875, -90.26722717285156, 212.99667358398438, -12.536422729492188, -10.37295150756836, -94.11236572265625, 118.7073745727539, -15.983634948730469, 86.60258483886719, 102.42684936523438, 17.002140045166016, 105.88334655761719, -1.5267868041992188, -53.54922866821289, -37.300537109375, 44.793033599853516, 7.25860595703125, 132.50634765625, 152.5159912109375, -21.4158935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000581.npy"}
|
|
{"epoch": 0.8783068783068783, "step": 582, "batch_size": 64, "mean": 49.674530029296875, "std": 73.79070281982422, "min": -110.74752807617188, "p10": -27.366620635986326, "median": 37.50057315826416, "p90": 142.8125427246094, "max": 237.02133178710938, "pos_frac": 0.75, "sample": [107.7962417602539, 162.27947998046875, 18.26409149169922, 98.84124755859375, 5.5506744384765625, 16.48342514038086, 154.59913635253906, 59.86549758911133, 107.49114990234375, 137.58206176757812, 51.785125732421875, 47.251792907714844, 131.8040313720703, 237.02133178710938, -27.772796630859375, 142.36517333984375, -16.9329833984375, -9.801742553710938, 59.416015625, 122.14446258544922, 1.4056015014648438, 104.86022186279297, 14.049518585205078, 170.57870483398438, 115.61688232421875, 143.0042724609375, 27.749353408813477, 12.129602432250977, -88.45626831054688, -107.30528259277344, -3.4969234466552734, 26.334747314453125, 92.99108123779297, -28.147415161132812, -16.512727737426758, 98.35127258300781, 115.6221923828125, 74.83882141113281, 159.74874877929688, 5.353580474853516, 117.65414428710938, 66.4948501586914, 53.812522888183594, 11.459260940551758, -5.033426284790039, -10.967849731445312, -44.288551330566406, 81.40930938720703, 7.030570983886719, -110.74752807617188, 4.203813552856445, 13.256118774414062, 0.06938552856445312, 1.0965118408203125, -69.45429992675781, 64.04603576660156, 124.66952514648438, -20.926490783691406, 151.35462951660156, -26.41887664794922, -10.359527587890625, 7.400352478027344, 106.40087127685547, 142.2590789794922], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000582.npy"}
|
|
{"epoch": 0.8798185941043084, "step": 583, "batch_size": 64, "mean": 55.59295654296875, "std": 82.79890441894531, "min": -165.06283569335938, "p10": -33.658647537231445, "median": 60.742387771606445, "p90": 161.36123046875, "max": 182.3106689453125, "pos_frac": 0.6875, "sample": [56.00432205200195, -9.687118530273438, -28.09149932861328, -17.056171417236328, 161.7229461669922, 71.22840881347656, -63.66667938232422, -39.96662139892578, -92.15013122558594, -7.6902618408203125, 139.34271240234375, -34.65409469604492, -18.396221160888672, 123.22955322265625, 52.7822151184082, -26.69196319580078, 71.61634826660156, -5.076238632202148, 176.80654907226562, 135.7658233642578, 65.48045349121094, -101.05582427978516, 106.84465026855469, 3.3672714233398438, 83.43470764160156, 105.06755065917969, 179.85305786132812, -122.22413635253906, 30.388050079345703, 139.83981323242188, 74.87495422363281, 39.829078674316406, -1.6878204345703125, 47.48539733886719, 113.34194946289062, 178.0115203857422, -31.3359375, -13.841842651367188, 134.36727905273438, 160.51722717285156, 43.37637710571289, 180.59368896484375, -165.06283569335938, 182.3106689453125, 7.163858413696289, 111.27386474609375, 107.33767700195312, 55.404808044433594, 97.36219024658203, -19.940635681152344, 158.923828125, 74.3042221069336, 142.5067596435547, 8.45414924621582, 68.5986328125, 100.94033813476562, -15.70930290222168, 86.51297760009766, -1.3812179565429688, 138.85003662109375, 176.16668701171875, 159.28219604492188, 16.493972778320312, 6.257011413574219], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000583.npy"}
|
|
{"epoch": 0.8813303099017384, "step": 584, "batch_size": 64, "mean": 48.627113342285156, "std": 85.13157653808594, "min": -136.19935607910156, "p10": -62.23723983764647, "median": 36.653961181640625, "p90": 158.75752410888674, "max": 220.40692138671875, "pos_frac": 0.671875, "sample": [174.89358520507812, 137.37591552734375, 71.13203430175781, -6.284034729003906, 44.08476257324219, -1.9885196685791016, 187.21392822265625, 156.22146606445312, 112.24707794189453, 106.60252380371094, 73.43070983886719, -16.422327041625977, -6.648435592651367, -47.176734924316406, -82.91752624511719, 81.12828826904297, 82.59551239013672, 159.8444061279297, 137.92015075683594, -14.704715728759766, 118.69615936279297, 176.2748565673828, -0.5865402221679688, 116.28459167480469, 151.89736938476562, 139.14358520507812, 54.07301330566406, 18.819015502929688, 18.50897216796875, 13.286834716796875, 155.89459228515625, -52.13722229003906, 103.9566650390625, 25.585357666015625, -70.22831726074219, 143.21566772460938, -28.901275634765625, -37.8768310546875, -8.256950378417969, -18.74555778503418, 12.485210418701172, 76.20621490478516, -66.81941223144531, 45.87620544433594, 85.2686996459961, 165.97381591796875, -0.24205589294433594, 54.504913330078125, 76.84381866455078, -66.5658187866211, -136.03555297851562, 26.692035675048828, -39.91600799560547, 153.8515625, 2.4319229125976562, 220.40692138671875, 193.79559326171875, 4.927606582641602, 13.251920700073242, 29.223159790039062, 28.402122497558594, 97.68154907226562, -97.36216735839844, -136.19935607910156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000584.npy"}
|
|
{"epoch": 0.8828420256991686, "step": 585, "batch_size": 64, "mean": 37.31257629394531, "std": 73.11202239990234, "min": -187.012451171875, "p10": -24.846855163574215, "median": 26.097142219543457, "p90": 146.1387329101563, "max": 196.53854370117188, "pos_frac": 0.78125, "sample": [40.03700637817383, 73.95317077636719, 25.80645751953125, -3.0641937255859375, 58.412574768066406, 57.955291748046875, 39.761497497558594, 133.89898681640625, -1.2332839965820312, 6.773773193359375, -147.40399169921875, 11.88576889038086, 98.04705810546875, 1.41644287109375, 0.9680595397949219, 196.53854370117188, -99.32755279541016, 0.21611785888671875, 42.92185974121094, 82.7462158203125, 103.4246826171875, 10.680219650268555, -5.051176071166992, -71.35629272460938, 26.714265823364258, 45.115257263183594, 0.14463043212890625, 3.7367019653320312, 121.62547302246094, 83.3589859008789, 0.3709907531738281, -5.530487060546875, 51.62422561645508, -2.9267234802246094, -107.79815673828125, 15.116813659667969, 10.573959350585938, 109.9092788696289, 151.38433837890625, 87.07688903808594, 23.138538360595703, 112.73240661621094, 48.77613830566406, -56.281089782714844, -187.012451171875, 168.17974853515625, 171.72669982910156, 162.9442901611328, -26.52977752685547, -20.92003631591797, 26.387826919555664, 25.678855895996094, -1.9784393310546875, 19.33622169494629, 20.944068908691406, 47.150596618652344, 49.583984375, 156.00131225585938, 162.44960021972656, 86.14663696289062, 50.05987548828125, 25.745010375976562, 54.25493621826172, 20.98644256591797], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000585.npy"}
|
|
{"epoch": 0.8843537414965986, "step": 586, "batch_size": 64, "mean": 52.18818664550781, "std": 61.5578498840332, "min": -77.7114486694336, "p10": -11.221081733703608, "median": 35.457271575927734, "p90": 148.2980514526367, "max": 184.1104736328125, "pos_frac": 0.796875, "sample": [117.75288391113281, 15.087104797363281, 100.7822265625, 50.66985321044922, 24.396560668945312, 137.60369873046875, 35.212493896484375, 5.108100891113281, 3.4524459838867188, -13.524946212768555, 83.01854705810547, -13.9737548828125, 15.676040649414062, 181.49624633789062, 153.0936737060547, -14.991743087768555, 76.64525604248047, 98.44760131835938, 39.866668701171875, 106.28226470947266, 18.240337371826172, 78.3355941772461, 24.91409683227539, 148.0561981201172, 13.0382080078125, -2.119039535522461, -5.644065856933594, 14.409881591796875, 117.31367492675781, 58.09066390991211, 124.37493896484375, 92.29689025878906, 35.702049255371094, 48.2859001159668, 154.0915985107422, 10.743255615234375, -70.46383666992188, 184.1104736328125, -3.5968589782714844, 7.499746322631836, 13.873010635375977, 29.78790283203125, -5.84539794921875, 94.91828918457031, 70.78544616699219, -14.629024505615234, -3.5172348022460938, 52.13585662841797, -2.5305023193359375, 81.90281677246094, 171.4392547607422, -32.84852600097656, 14.885848999023438, 31.408498764038086, 148.40170288085938, 28.39875030517578, -77.7114486694336, 149.310302734375, 0.9637908935546875, 16.945236206054688, 138.99517822265625, 40.47988510131836, 89.01805877685547, 53.695098876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000586.npy"}
|
|
{"epoch": 0.8858654572940288, "step": 587, "batch_size": 64, "mean": 51.51568603515625, "std": 91.45906829833984, "min": -302.5234069824219, "p10": -36.307583999633785, "median": 58.4794921875, "p90": 164.82706146240238, "max": 211.63485717773438, "pos_frac": 0.71875, "sample": [35.49380874633789, -118.67039489746094, -1.5919876098632812, 82.84683227539062, 118.32785034179688, -65.69148254394531, 57.49625015258789, -25.651275634765625, 142.34323120117188, 70.42739868164062, 12.186016082763672, 77.26531982421875, -37.269927978515625, -3.0626983642578125, -3.9599609375, 57.521583557128906, 104.63179016113281, -4.072540283203125, -5.4254302978515625, 183.0181427001953, 134.54344177246094, 89.43665313720703, -0.4188995361328125, 168.05799865722656, 88.98487854003906, 46.17103576660156, 176.23971557617188, 8.205734252929688, -302.5234069824219, -5.758197784423828, 76.1019058227539, 72.58921813964844, 88.14323425292969, 99.77133178710938, 24.620433807373047, 81.28074645996094, 81.90129089355469, -187.41600036621094, -61.37261962890625, 199.19924926757812, 143.85829162597656, -26.886959075927734, 134.85653686523438, 98.41033935546875, 2.387784957885742, 121.28166961669922, 10.12408447265625, 144.4521026611328, 127.36927032470703, 92.45681762695312, 177.1544647216797, -83.24766540527344, 113.85493469238281, 8.725196838378906, 54.112754821777344, 176.44956970214844, 15.281471252441406, -12.780677795410156, -34.06211471557617, 44.47614288330078, 59.437400817871094, 211.63485717773438, 157.2882080078125, 6.449287414550781], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000587.npy"}
|
|
{"epoch": 0.8873771730914588, "step": 588, "batch_size": 64, "mean": 53.87213134765625, "std": 78.00247955322266, "min": -106.60125732421875, "p10": -47.605690383911124, "median": 55.72893142700195, "p90": 155.58757324218752, "max": 226.1064453125, "pos_frac": 0.765625, "sample": [69.87472534179688, -57.64299011230469, -29.29645538330078, 99.79500579833984, -7.4381561279296875, 129.0395050048828, 144.89773559570312, 38.23521423339844, 137.0667724609375, 117.25338745117188, 198.31820678710938, 30.144655227661133, 29.36747932434082, 64.616455078125, 30.0965576171875, 89.4925537109375, 178.656005859375, 226.1064453125, 27.608945846557617, 222.45811462402344, 55.91773223876953, 99.53052520751953, -50.91660690307617, 162.50350952148438, 15.641382217407227, 86.00727844238281, 47.636383056640625, -31.494789123535156, 2.0896034240722656, -57.3084716796875, 5.517065048217773, 73.14002990722656, -99.55748748779297, 71.93235778808594, 99.7115249633789, 164.4119873046875, -9.966337203979492, 84.25337219238281, 55.540130615234375, -18.960105895996094, -36.048362731933594, -30.661590576171875, 65.27959442138672, 144.96420288085938, 95.73627471923828, 58.32112121582031, 130.11581420898438, 2.0044517517089844, 15.729812622070312, 114.19444274902344, 17.968948364257812, -81.03512573242188, 23.766845703125, 40.829872131347656, 67.64250183105469, 62.78411865234375, -39.880218505859375, 24.684532165527344, 32.53710174560547, -82.33724975585938, 124.13622283935547, 156.91488647460938, 152.49050903320312, -106.60125732421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000588.npy"}
|
|
{"epoch": 0.8888888888888888, "step": 589, "batch_size": 64, "mean": 56.97416687011719, "std": 64.6989517211914, "min": -84.18157958984375, "p10": -18.50485115051269, "median": 51.242828369140625, "p90": 144.88469848632815, "max": 226.86732482910156, "pos_frac": 0.8125, "sample": [28.042146682739258, 49.49951934814453, -68.31947326660156, 161.91632080078125, 58.58502960205078, 3.522266387939453, -12.557151794433594, -2.5380401611328125, 7.681221008300781, 1.05572509765625, -23.373458862304688, 20.84659767150879, 226.86732482910156, 6.623767852783203, 109.45045471191406, 53.15766906738281, 71.34345245361328, 120.39518737792969, 121.08354187011719, 105.88697052001953, 43.877593994140625, 85.65731811523438, 126.79353332519531, -42.342926025390625, 146.0096893310547, 82.88552856445312, -20.82074737548828, 109.80138397216797, 111.70927429199219, 91.69755554199219, 65.53302764892578, -11.789073944091797, 142.2597198486328, 31.04315185546875, -12.608261108398438, 68.3221664428711, 32.659934997558594, 25.95134735107422, 52.1666259765625, 164.3658447265625, 31.040908813476562, 75.275146484375, 140.35073852539062, 75.72993469238281, 65.31584167480469, 20.42706298828125, 85.45406341552734, -36.533512115478516, -13.101093292236328, 17.001991271972656, -84.18157958984375, 92.4188003540039, 148.05880737304688, -25.97985076904297, 46.19245147705078, 146.6604461669922, 117.02635955810547, 84.48445129394531, 14.433197021484375, 18.20349884033203, 50.31903076171875, 29.854907989501953, 204.53732299804688, 11.015878677368164], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000589.npy"}
|
|
{"epoch": 0.890400604686319, "step": 590, "batch_size": 64, "mean": 56.176185607910156, "std": 74.03482818603516, "min": -133.428955078125, "p10": -23.64844284057617, "median": 52.39865493774414, "p90": 158.8896697998047, "max": 190.63275146484375, "pos_frac": 0.75, "sample": [91.7885971069336, 23.04778289794922, 84.4365463256836, 6.296318054199219, 117.71200561523438, 141.28457641601562, 67.77888488769531, -125.71875, 26.293590545654297, -15.945701599121094, 34.19810485839844, 5.079807281494141, 5.2289886474609375, 32.896949768066406, -133.428955078125, 102.56330108642578, 133.74185180664062, 94.59988403320312, -31.188732147216797, -1.0343780517578125, -27.555213928222656, 93.73541259765625, -2.9103317260742188, 156.9678955078125, 169.207275390625, -9.875251770019531, -13.497344970703125, 40.94779968261719, 188.6195068359375, 41.66845703125, -21.360267639160156, 139.21127319335938, 17.32842445373535, 21.893798828125, 117.5367431640625, 79.91572570800781, 81.51457977294922, 12.7591552734375, 52.61663818359375, 14.34246826171875, 165.92022705078125, 81.60935974121094, 190.63275146484375, 131.4722900390625, 155.31100463867188, 92.8168716430664, 88.85870361328125, 106.73184204101562, -21.10601806640625, 60.607177734375, -33.432701110839844, 162.4386444091797, 99.31857299804688, -14.060501098632812, 5.1942291259765625, 148.07632446289062, 55.176971435546875, 52.18067169189453, 163.71385192871094, -4.716114044189453, 41.16353225708008, -24.62908935546875, 159.71328735351562, -80.41319274902344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000590.npy"}
|
|
{"epoch": 0.891912320483749, "step": 591, "batch_size": 64, "mean": 66.34883880615234, "std": 88.1004638671875, "min": -138.74578857421875, "p10": -61.88242797851562, "median": 69.02766036987305, "p90": 171.73624114990236, "max": 254.15740966796875, "pos_frac": 0.796875, "sample": [1.1711254119873047, 162.46807861328125, 99.76663970947266, 124.47835540771484, 121.33922576904297, 65.31173706054688, 162.57041931152344, 60.63591003417969, 81.94140625, 98.59613037109375, 38.06684875488281, -63.78363037109375, 129.5852508544922, -81.24750518798828, 89.28437042236328, -17.177566528320312, 56.96220397949219, 50.1015625, 95.40008544921875, 17.269710540771484, 117.89938354492188, 72.74358367919922, 210.58352661132812, -76.91398620605469, 148.5019989013672, -138.74578857421875, 96.62765502929688, 88.56427764892578, 14.999618530273438, 41.302459716796875, -105.68663024902344, -103.75262451171875, 190.81436157226562, 186.93910217285156, 9.203239440917969, 165.3854522705078, -15.263858795166016, 37.38508605957031, 152.08041381835938, 15.927837371826172, 2.2892837524414062, -14.738639831542969, 54.534576416015625, 146.2222137451172, 0.44322776794433594, 193.38897705078125, -18.2321720123291, -82.52311706542969, -11.092018127441406, 254.15740966796875, -57.4462890625, 1.4648971557617188, 187.20223999023438, 37.160438537597656, 39.58700180053711, 164.4652099609375, 144.3887939453125, 93.85182189941406, 136.4423370361328, 87.08744049072266, 174.4580078125, 138.8022003173828, 14.245141983032227, 158.83145141601562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000591.npy"}
|
|
{"epoch": 0.8934240362811792, "step": 592, "batch_size": 64, "mean": 51.988304138183594, "std": 74.30171966552734, "min": -70.41520690917969, "p10": -30.835396194458006, "median": 32.46206855773926, "p90": 165.80882110595704, "max": 247.80087280273438, "pos_frac": 0.796875, "sample": [65.81605529785156, 182.223388671875, 143.14610290527344, 36.347999572753906, 142.99522399902344, 162.62893676757812, 170.23770141601562, 33.37958908081055, 5.715679168701172, 31.54454803466797, -24.44580841064453, 75.99205017089844, 42.68244934082031, 10.16424560546875, -44.77119445800781, 76.85021209716797, 163.01210021972656, 100.66171264648438, 3.2703704833984375, 21.126365661621094, -53.96998596191406, 124.78046417236328, 66.05763244628906, 193.57833862304688, 21.127399444580078, 81.97285461425781, 82.12055206298828, 167.00741577148438, 132.614501953125, 19.148635864257812, -70.41520690917969, 97.40574645996094, -5.9404144287109375, 95.02629089355469, 25.56676483154297, 43.53792953491211, -69.96905517578125, 89.214599609375, -49.90741729736328, -24.36405372619629, -5.742671966552734, 12.495988845825195, 12.213470458984375, 10.353225708007812, -53.31166076660156, 15.746826171875, 12.800491333007812, 211.1960906982422, 188.3736572265625, 4.967071533203125, 247.80087280273438, -1.2766647338867188, 9.393856048583984, 0.7094268798828125, 36.67613220214844, 4.574609756469727, -30.931610107421875, 5.373191833496094, -30.610897064208984, 39.53890609741211, 71.3267822265625, 15.356010437011719, 77.47891235351562, 109.57852935791016], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000592.npy"}
|
|
{"epoch": 0.8949357520786092, "step": 593, "batch_size": 64, "mean": 50.3743896484375, "std": 80.56074523925781, "min": -128.150390625, "p10": -51.90472946166989, "median": 46.469661712646484, "p90": 156.3938400268555, "max": 188.77328491210938, "pos_frac": 0.75, "sample": [100.1278076171875, 18.668716430664062, 139.45632934570312, -12.710342407226562, 131.43212890625, 0.19478607177734375, 54.32355499267578, -128.150390625, 177.0824737548828, 43.90349578857422, 167.09786987304688, 19.30453872680664, 169.92002868652344, 137.8540496826172, 33.059417724609375, -7.800380706787109, 79.14981842041016, 80.21621704101562, 141.06082153320312, 98.37815856933594, 11.515859603881836, 66.23880004882812, 181.86572265625, 151.4302978515625, 0.385467529296875, 3.9304962158203125, -18.484304428100586, 132.73260498046875, 84.2442398071289, -99.60050964355469, 70.10416412353516, -76.03573608398438, -8.520248413085938, 63.600547790527344, 81.9308853149414, 130.06222534179688, 1.0345840454101562, -15.796890258789062, 117.4609375, -66.22776794433594, -100.44729614257812, -15.491741180419922, 173.90394592285156, -117.68128204345703, 151.2653350830078, 109.83973693847656, 134.09231567382812, 25.753076553344727, -12.854705810546875, 188.77328491210938, 41.764495849609375, -5.758928298950195, 97.54932403564453, 62.35293960571289, 49.03582763671875, -114.3530044555664, 21.739721298217773, 158.5210723876953, 1.0586700439453125, -4.076194763183594, 11.386138916015625, 33.208396911621094, 14.1513671875, 65.78815460205078], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000593.npy"}
|
|
{"epoch": 0.8964474678760394, "step": 594, "batch_size": 64, "mean": 46.536590576171875, "std": 68.00553894042969, "min": -150.57835388183594, "p10": -38.62346382141113, "median": 44.70326232910156, "p90": 131.18660125732424, "max": 180.96331787109375, "pos_frac": 0.703125, "sample": [24.085765838623047, 89.88136291503906, -37.05913162231445, 20.543434143066406, 71.62457275390625, 104.79058074951172, 34.18901824951172, 38.28877258300781, 49.63813781738281, -1.9822006225585938, -34.11891174316406, -1.0844764709472656, -20.467613220214844, 76.01530456542969, 26.882349014282227, -0.5257759094238281, -45.380157470703125, -1.7888069152832031, 82.7664794921875, 7.2393646240234375, 118.08854675292969, 45.690773010253906, 2.0634899139404297, 114.73780822753906, 169.02349853515625, 99.234375, 125.83854675292969, -39.29389190673828, 118.653076171875, -2.3610668182373047, 117.50360870361328, 105.51266479492188, 176.23440551757812, 132.42843627929688, 24.42505645751953, -21.84441375732422, -48.953834533691406, 72.3624496459961, 107.61997985839844, -61.02931213378906, 128.2889862060547, 1.0816516876220703, 94.80253601074219, 151.23739624023438, 4.297401428222656, 26.500219345092773, 58.95539855957031, 151.96966552734375, -3.442502975463867, 134.35430908203125, 56.696746826171875, -150.57835388183594, 96.56051635742188, 180.96331787109375, 64.48574829101562, 91.23634338378906, 12.123394012451172, 43.71575164794922, 51.55430603027344, -44.43457794189453, 72.23027038574219, -52.02174377441406, -17.272239685058594, -14.435012817382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000594.npy"}
|
|
{"epoch": 0.8979591836734694, "step": 595, "batch_size": 64, "mean": 62.99003601074219, "std": 83.77729034423828, "min": -158.0752716064453, "p10": -20.64822692871093, "median": 56.64980125427246, "p90": 166.6436477661133, "max": 236.25357055664062, "pos_frac": 0.828125, "sample": [-6.351173400878906, 68.54804229736328, 236.25357055664062, 75.78048706054688, 29.36846160888672, 16.154664993286133, 62.42565155029297, 26.257354736328125, 10.793136596679688, 156.84552001953125, 163.22528076171875, 25.35226058959961, 50.87395095825195, -74.77627563476562, 164.14251708984375, 17.846275329589844, 5.456075668334961, 1.1348876953125, 159.7641143798828, 25.738290786743164, 43.598167419433594, 75.58059692382812, 189.78968811035156, 29.12335205078125, 179.39471435546875, 32.986351013183594, 79.36937713623047, 74.93260192871094, 119.32951354980469, 125.4891357421875, -3.1409378051757812, -66.29780578613281, 47.328636169433594, -24.2918701171875, 143.3245086669922, 154.85472106933594, 179.9807891845703, -97.12167358398438, -158.0752716064453, 181.45480346679688, 173.3422088623047, 160.64707946777344, 69.02688598632812, 64.927001953125, -82.37540435791016, 163.04615783691406, 67.61091613769531, 100.954345703125, -12.146392822265625, 23.023120880126953, -0.7442226409912109, 156.60198974609375, 85.98373413085938, 17.534706115722656, 0.5018157958984375, 44.38103485107422, 167.71556091308594, 127.7738037109375, 26.771408081054688, 6.1746826171875, 4.51820182800293, -127.69292449951172, 143.39515686035156, 127.94892120361328], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000595.npy"}
|
|
{"epoch": 0.8994708994708994, "step": 596, "batch_size": 64, "mean": 31.32728385925293, "std": 72.07229614257812, "min": -141.44422912597656, "p10": -34.05284194946289, "median": 18.15302085876465, "p90": 155.02068176269532, "max": 189.6879425048828, "pos_frac": 0.625, "sample": [-61.11326599121094, 38.512184143066406, 172.6785430908203, 36.04210662841797, 59.318138122558594, -34.81829833984375, -6.494663238525391, 16.353790283203125, -17.80237579345703, 161.05462646484375, -68.92790222167969, 39.32437515258789, -83.17120361328125, -13.318954467773438, 189.1122589111328, 156.45816040039062, 122.42623901367188, 34.315670013427734, -87.06648254394531, 33.63459014892578, -25.7506103515625, 151.66656494140625, 34.60694885253906, 80.92672729492188, 167.1056365966797, 147.37081909179688, 56.45604705810547, -32.07554626464844, 23.319007873535156, -20.568359375, -24.18631362915039, 46.50067138671875, 22.4420166015625, 3.483041763305664, 50.83428192138672, -6.194286346435547, 32.66870880126953, 189.6879425048828, 6.642879486083984, -32.57311248779297, 31.905860900878906, 8.320762634277344, -1.8055915832519531, -141.44422912597656, -12.734878540039062, 12.364582061767578, -0.191650390625, -11.976654052734375, 167.943115234375, 67.84048461914062, 102.71759033203125, -6.3287200927734375, -16.928314208984375, 11.723793029785156, -30.40119171142578, 146.736083984375, 9.64056396484375, 74.7743148803711, -34.68701171875, 34.44416809082031, 19.952251434326172, 38.67460632324219, 6.876068115234375, -31.350488662719727], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000596.npy"}
|
|
{"epoch": 0.9009826152683296, "step": 597, "batch_size": 64, "mean": 46.612571716308594, "std": 71.81172180175781, "min": -90.07017517089844, "p10": -37.36870193481445, "median": 30.006983757019043, "p90": 161.97226257324223, "max": 188.40797424316406, "pos_frac": 0.71875, "sample": [-4.518360137939453, -59.350608825683594, -42.41543197631836, -10.459915161132812, 169.19888305664062, -6.144496917724609, 30.820545196533203, -68.38233184814453, -36.23246765136719, 168.23985290527344, 3.674724578857422, 92.440185546875, 6.045732498168945, 102.09323120117188, -27.831298828125, 10.788610458374023, 170.1208953857422, 147.75408935546875, 138.01022338867188, -37.85565948486328, 71.14945220947266, 14.184844970703125, -57.1695556640625, 18.68189239501953, 106.2640151977539, 105.82266998291016, 17.429807662963867, -28.843130111694336, 109.43439483642578, 5.640665054321289, 82.77951049804688, 30.198036193847656, 168.99716186523438, 41.095863342285156, 64.09558868408203, 30.628314971923828, -18.08624267578125, 80.00244140625, 36.7142333984375, 144.66580200195312, 1.9689750671386719, -63.54710388183594, 7.9593048095703125, -3.540220260620117, -16.30179786682129, -2.5263423919677734, 4.74772834777832, 19.164735794067383, 99.88573455810547, 49.27368927001953, 168.06576538085938, 7.543621063232422, 124.12354278564453, -6.6599884033203125, -90.07017517089844, 99.96771240234375, 5.624664306640625, 63.51988220214844, 188.40797424316406, 177.30319213867188, 29.81593132019043, 134.45835876464844, 104.55170440673828, 109.7855224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000597.npy"}
|
|
{"epoch": 0.9024943310657596, "step": 598, "batch_size": 64, "mean": 30.801076889038086, "std": 73.88812255859375, "min": -148.35067749023438, "p10": -61.753604125976565, "median": 21.043701171875, "p90": 138.29890747070314, "max": 204.89730834960938, "pos_frac": 0.640625, "sample": [-6.634424209594727, -66.72328186035156, 19.711387634277344, 23.99030303955078, 60.87090301513672, 1.7167015075683594, 11.694206237792969, 165.38619995117188, 112.34749603271484, -148.35067749023438, 35.481346130371094, 48.76123046875, 19.705326080322266, -71.30635070800781, 138.839599609375, -28.652687072753906, 183.7051544189453, -61.2664794921875, 115.42051696777344, 53.51042556762695, -26.910242080688477, -126.66383361816406, -44.54450988769531, 78.37145233154297, 20.986709594726562, 25.300048828125, 39.14242935180664, 53.672454833984375, -8.339607238769531, -22.335010528564453, -39.10108184814453, -29.409948348999023, 33.02970886230469, 82.85694885253906, 137.03729248046875, -2.217761993408203, 21.100692749023438, 2.182140350341797, 68.09840393066406, 21.94549560546875, 120.90293884277344, 20.6673583984375, -2.409181594848633, 163.9253692626953, -0.3697509765625, 89.98359680175781, 34.27064514160156, -2.4944229125976562, 98.56668853759766, 44.93548583984375, -28.57684326171875, -78.99279022216797, 5.603752136230469, 17.1484375, 47.35723876953125, -61.962371826171875, 146.54916381835938, 74.94966125488281, -7.9091033935546875, -12.125450134277344, 204.89730834960938, 145.55377197265625, -74.9848861694336, 133.37353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000598.npy"}
|
|
{"epoch": 0.9040060468631897, "step": 599, "batch_size": 64, "mean": 37.68316650390625, "std": 75.18990325927734, "min": -143.92620849609375, "p10": -53.53351974487304, "median": 34.926862716674805, "p90": 145.42339935302735, "max": 209.78501892089844, "pos_frac": 0.640625, "sample": [-39.431243896484375, -40.5091552734375, 42.92829132080078, 57.34606170654297, -27.362037658691406, 59.191612243652344, 13.527835845947266, 168.1475830078125, -68.01583862304688, -78.99984741210938, -13.554862976074219, -25.860666275024414, -4.126396179199219, 72.92045593261719, 91.38451385498047, -23.781105041503906, -102.25459289550781, -10.93552017211914, 2.607311248779297, -49.800750732421875, 146.0554656982422, 86.58912658691406, 41.83576965332031, -143.92620849609375, 6.205867767333984, 16.929168701171875, 7.1205291748046875, 158.94290161132812, -60.1260986328125, 29.636688232421875, 175.34495544433594, 135.55958557128906, -55.133277893066406, -27.77001190185547, 40.217037200927734, 19.333892822265625, 46.423622131347656, -8.504501342773438, 54.60362243652344, 111.63461303710938, 123.38632202148438, -15.313362121582031, 24.223588943481445, -39.704498291015625, 148.03903198242188, 71.02259826660156, 138.15078735351562, -1.905181884765625, -61.32967758178711, 25.532428741455078, 157.67520141601562, 143.94857788085938, 61.96014404296875, 92.50079345703125, 121.97962951660156, 47.223609924316406, 209.78501892089844, 90.79517364501953, 63.95118713378906, -1.3458614349365234, -22.541763305664062, 85.40027618408203, 93.67381286621094, 50.22034454345703], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000599.npy"}
|
|
{"epoch": 0.9055177626606198, "step": 600, "batch_size": 64, "mean": 44.304630279541016, "std": 75.25826263427734, "min": -126.516845703125, "p10": -50.307614898681635, "median": 32.0871696472168, "p90": 154.1640106201172, "max": 201.0743408203125, "pos_frac": 0.734375, "sample": [107.006591796875, 53.50263214111328, -4.155363082885742, 7.84027099609375, 169.1163330078125, 31.315048217773438, 31.217182159423828, -68.2188949584961, 145.7407989501953, 176.0098419189453, -14.694679260253906, 18.020404815673828, 92.0458755493164, 114.60502624511719, 4.435699462890625, 8.360246658325195, -35.264190673828125, -53.75615692138672, -42.261016845703125, 126.53370666503906, 13.432674407958984, 52.97662353515625, 134.46836853027344, 104.28787231445312, -30.293190002441406, 170.1844024658203, 66.47560119628906, 148.07342529296875, -9.54949951171875, 121.1990737915039, 156.77426147460938, 30.97266387939453, -126.516845703125, -16.663246154785156, 73.16517639160156, -91.72360229492188, -60.11407470703125, 88.07071685791016, 1.5457229614257812, 183.91098022460938, -1.4465713500976562, 72.46551513671875, 32.149749755859375, 139.0595703125, -12.701696395874023, 23.391117095947266, -59.07347106933594, -25.746110916137695, -101.13938903808594, 58.927730560302734, 1.4164009094238281, 75.71371459960938, 32.02458953857422, 25.176422119140625, 58.01274108886719, 8.76605224609375, 40.22947692871094, 36.094730377197266, 79.77574157714844, 169.53179931640625, 6.176475524902344, 41.31520080566406, 56.225738525390625, 201.0743408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000600.npy"}
|
|
{"epoch": 0.9070294784580499, "step": 601, "batch_size": 64, "mean": 58.88792037963867, "std": 81.41028594970703, "min": -127.521484375, "p10": -43.16235656738281, "median": 53.316253662109375, "p90": 160.98493041992188, "max": 246.01861572265625, "pos_frac": 0.765625, "sample": [121.31907653808594, 30.34941864013672, 129.7266845703125, -44.83199691772461, 153.5069580078125, 107.90496826171875, 246.01861572265625, -44.241607666015625, 49.2603759765625, -9.096847534179688, 181.66905212402344, -127.521484375, 157.5148162841797, 33.66914367675781, -13.270654678344727, 63.744346618652344, 126.7350082397461, 167.904052734375, -44.59184265136719, 101.5047836303711, 162.4721221923828, 3.215862274169922, 0.5977020263671875, 131.94271850585938, 170.13592529296875, 72.22871398925781, -2.158437728881836, 4.762912750244141, 111.9775619506836, 13.829299926757812, 22.541656494140625, 100.02894592285156, 94.30142211914062, -9.311714172363281, -2.4776859283447266, 122.60202026367188, 2.2439498901367188, 77.57403564453125, 43.925750732421875, 20.95062828063965, -72.86372375488281, 62.749298095703125, 10.00527572631836, 135.59275817871094, 57.37213134765625, 73.67036437988281, 41.284114837646484, -40.64410400390625, -26.105737686157227, 37.386497497558594, 10.323616027832031, 199.79244995117188, -7.273492813110352, 4.075595855712891, 116.22901916503906, 143.64675903320312, 199.00830078125, 103.93878173828125, 147.756103515625, 0.37494659423828125, -88.80400085449219, 138.11807250976562, 112.23506164550781, -117.69737243652344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000601.npy"}
|
|
{"epoch": 0.90854119425548, "step": 602, "batch_size": 64, "mean": 56.12777328491211, "std": 70.09522247314453, "min": -131.28204345703125, "p10": -7.5826587677001935, "median": 42.64584159851074, "p90": 151.53154754638672, "max": 225.17120361328125, "pos_frac": 0.84375, "sample": [-20.749725341796875, 112.03398895263672, 150.40826416015625, 225.17120361328125, 63.04539489746094, 35.840301513671875, 57.155517578125, 6.932889938354492, 64.46810913085938, 17.025630950927734, 181.10720825195312, -13.695537567138672, 152.01295471191406, 140.47299194335938, -70.33949279785156, 150.33743286132812, -85.91403198242188, 160.035888671875, 95.45509338378906, 1.1063270568847656, 3.76708984375, -1.7425003051757812, 7.657054901123047, -6.170623779296875, 16.526840209960938, 117.7402114868164, 44.83306884765625, 96.1120834350586, 30.314167022705078, 59.372711181640625, 5.821445465087891, 100.25054931640625, 84.84549713134766, 40.458614349365234, 40.211856842041016, 22.61331558227539, 14.464282989501953, 23.20372772216797, 152.07505798339844, 48.76209259033203, 5.923923492431641, 111.42916870117188, 132.42098999023438, 136.7515869140625, -1.7743072509765625, -131.28204345703125, 86.43712615966797, 36.19298553466797, 18.461669921875, 10.470748901367188, -8.187816619873047, 7.49578857421875, 120.71253967285156, 45.173912048339844, 11.935981750488281, 150.16619873046875, 44.88855743408203, 47.900264739990234, 176.59051513671875, 169.02479553222656, -49.76299285888672, 110.25706481933594, 22.33953094482422, 15.584381103515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000602.npy"}
|
|
{"epoch": 0.91005291005291, "step": 603, "batch_size": 64, "mean": 54.27854919433594, "std": 74.34972381591797, "min": -119.8145751953125, "p10": -40.85634078979492, "median": 49.65247344970703, "p90": 144.5316940307617, "max": 237.37567138671875, "pos_frac": 0.765625, "sample": [121.14761352539062, 117.21226501464844, 92.79678344726562, -27.907798767089844, -41.696197509765625, 44.46675109863281, 53.087989807128906, 58.847808837890625, -8.071990966796875, 54.12132263183594, 153.4373321533203, 43.29670715332031, 25.252906799316406, -53.77758026123047, 13.559532165527344, 138.67752075195312, -60.02499008178711, 47.660675048828125, 10.137947082519531, 237.37567138671875, 60.082130432128906, 13.067962646484375, 139.64100646972656, 35.98065185546875, -5.393817901611328, 56.4453125, 115.92622375488281, -55.282928466796875, 124.70494079589844, -108.65548706054688, 8.867744445800781, 22.23626708984375, 84.85065460205078, 84.42282104492188, -63.90998840332031, 126.75115966796875, 75.05203247070312, 25.782878875732422, -1.116689682006836, 161.484130859375, 83.98686218261719, 106.51847076416016, 51.64427185058594, 78.33077239990234, 110.6697006225586, 136.00775146484375, 30.713821411132812, -21.795488357543945, 40.04738235473633, 8.556474685668945, -3.69842529296875, 34.225746154785156, 112.31649780273438, 142.6714324951172, 95.7146224975586, -7.394050598144531, 28.33205795288086, -119.8145751953125, 150.72543334960938, 145.32894897460938, 159.05020141601562, 2.9919681549072266, 227.0567169189453, -38.89667510986328], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000603.npy"}
|
|
{"epoch": 0.9115646258503401, "step": 604, "batch_size": 64, "mean": 47.592529296875, "std": 81.14274597167969, "min": -179.41925048828125, "p10": -39.82253646850586, "median": 37.323158264160156, "p90": 160.42808380126954, "max": 254.9449920654297, "pos_frac": 0.734375, "sample": [63.818206787109375, 135.074951171875, -4.796764373779297, -75.23556518554688, 34.59475326538086, 91.26136779785156, -15.079883575439453, 10.440622329711914, 161.7734832763672, -25.807220458984375, 3.339874267578125, -16.58917999267578, 2.8270626068115234, 108.59407806396484, -36.60040283203125, -6.06353759765625, 139.51519775390625, 39.88832092285156, 60.04851531982422, 123.94376373291016, -115.7995834350586, -27.98724365234375, 10.2034912109375, -64.19049072265625, 56.16979217529297, 1.474863052368164, 8.90391731262207, 45.09855651855469, 51.85823059082031, -179.41925048828125, 56.66658401489258, -40.368934631347656, 47.69243621826172, -41.038848876953125, 157.288818359375, -5.297935485839844, 34.75799560546875, 12.907470703125, 174.22799682617188, 116.31053924560547, 86.3357925415039, 170.60816955566406, 254.9449920654297, 24.126480102539062, 122.82193756103516, 118.89431762695312, 22.772499084472656, 144.28260803222656, 175.6522216796875, -35.862060546875, 170.55941772460938, 11.800582885742188, 54.325538635253906, 19.769081115722656, 91.16209411621094, 48.6746826171875, 101.4263916015625, 16.466028213500977, 2.851865768432617, 188.09408569335938, 109.64987182617188, -38.547607421875, -52.39714813232422, 143.10403442382812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000604.npy"}
|
|
{"epoch": 0.9130763416477702, "step": 605, "batch_size": 64, "mean": 47.050899505615234, "std": 76.82124328613281, "min": -131.13433837890625, "p10": -55.58697814941406, "median": 44.00075149536133, "p90": 140.65245056152344, "max": 213.12994384765625, "pos_frac": 0.765625, "sample": [72.3413314819336, 46.673736572265625, -96.77838897705078, 213.12994384765625, 24.419898986816406, 149.9002685546875, 17.507354736328125, 112.52734375, 94.08878326416016, -54.04765319824219, 100.98473358154297, -25.085708618164062, 17.376014709472656, -15.934669494628906, 15.961261749267578, -89.00013732910156, 54.455413818359375, 129.68258666992188, 99.2196044921875, 71.46234130859375, -7.885841369628906, 6.33203125, 83.65282440185547, -131.13433837890625, 8.027702331542969, 87.73602294921875, 158.36871337890625, 121.03182983398438, 41.32776641845703, 24.698822021484375, 108.7563705444336, 23.346153259277344, -99.44184112548828, 139.18783569335938, 1.539031982421875, 140.87506103515625, 138.19410705566406, 88.61634826660156, 38.84431457519531, 104.10545349121094, -56.24668884277344, -0.057888031005859375, 106.99495697021484, 72.60978698730469, 169.2349090576172, 21.96765899658203, 140.13302612304688, 141.21299743652344, 113.23224639892578, -28.502685546875, 181.2896728515625, 56.11131286621094, 77.38749694824219, 3.2603988647460938, -69.79786682128906, 16.867761611938477, 24.669153213500977, -122.86185455322266, 25.595230102539062, -34.04387283325195, 6.840764999389648, 74.05769348144531, -11.090103149414062, 87.33103942871094], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000605.npy"}
|
|
{"epoch": 0.9145880574452003, "step": 606, "batch_size": 64, "mean": 55.24286651611328, "std": 75.54102325439453, "min": -87.5557632446289, "p10": -40.91037445068359, "median": 45.98482131958008, "p90": 154.72399139404297, "max": 233.44931030273438, "pos_frac": 0.703125, "sample": [-46.343788146972656, 165.93783569335938, 23.344467163085938, 173.38177490234375, -43.55182647705078, 12.11175537109375, 147.6225128173828, 8.623977661132812, -0.07211875915527344, 117.9910888671875, 150.01290893554688, -15.07293701171875, 90.26080322265625, 118.08758544921875, 112.85079193115234, -87.5557632446289, 95.90936279296875, 42.416259765625, -8.507761001586914, 69.99317932128906, 233.44931030273438, 38.79071044921875, 31.608028411865234, 108.3626937866211, -50.24517059326172, -34.746986389160156, 139.7723388671875, -18.13007926940918, 65.7501220703125, 123.35590362548828, 62.95579528808594, -6.155853271484375, -81.77188873291016, 154.3834228515625, 50.464447021484375, 6.546380996704102, 42.32048034667969, -10.171831130981445, 104.14276123046875, 16.63439178466797, 154.8699493408203, 189.94796752929688, 65.88341522216797, 109.24827575683594, 16.082542419433594, 200.7432098388672, -15.147552490234375, 39.31477355957031, -51.938995361328125, -22.509632110595703, 146.85885620117188, 11.503158569335938, -73.4883041381836, 81.31332397460938, -10.447334289550781, 95.8212661743164, 49.553382873535156, -12.49007797241211, 66.07284545898438, 169.28907775878906, 81.48343658447266, 103.00078582763672, 37.85157775878906, -2.0274829864501953], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000606.npy"}
|
|
{"epoch": 0.9160997732426304, "step": 607, "batch_size": 64, "mean": 53.149288177490234, "std": 75.91312408447266, "min": -124.47518920898438, "p10": -28.88330917358398, "median": 38.58739471435547, "p90": 154.80211029052737, "max": 217.98927307128906, "pos_frac": 0.703125, "sample": [8.844413757324219, 18.195526123046875, -61.370811462402344, 45.69651794433594, 79.26593017578125, 30.503150939941406, 101.5548324584961, 56.16033172607422, -124.47518920898438, -29.710060119628906, 0.3326683044433594, -10.022895812988281, 136.92730712890625, 157.03614807128906, 36.65727233886719, -12.666290283203125, -14.078437805175781, 6.888204574584961, 44.257469177246094, 165.62644958496094, 31.278961181640625, 164.8925018310547, 117.74044799804688, 110.52082824707031, 89.94194030761719, 22.30146026611328, -22.26325225830078, 119.38264465332031, -14.282173156738281, -11.25616455078125, 163.04660034179688, 70.21822357177734, 101.33894348144531, 100.8377914428711, -42.884002685546875, 174.92257690429688, 217.98927307128906, -3.0860862731933594, 122.08363342285156, -1.7130584716796875, -47.679779052734375, 11.363143920898438, 2.5108814239501953, 114.077392578125, 86.36824035644531, -26.9542236328125, -14.002557754516602, 121.28467559814453, 206.45095825195312, 136.40794372558594, 1.5847206115722656, -7.269622802734375, 80.92162322998047, 142.95947265625, 26.467731475830078, 149.0736846923828, -1.4193992614746094, 40.51751708984375, 84.86354064941406, 149.58935546875, 31.992416381835938, -91.0368423461914, -58.300804138183594, 115.1507568359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000607.npy"}
|
|
{"epoch": 0.9176114890400605, "step": 608, "batch_size": 64, "mean": 47.51267623901367, "std": 69.04450225830078, "min": -110.4946517944336, "p10": -20.34271659851074, "median": 26.61419677734375, "p90": 136.94870147705078, "max": 224.0789794921875, "pos_frac": 0.734375, "sample": [102.69949340820312, -34.67381286621094, 0.31435394287109375, 137.70384216308594, 0.06440162658691406, -17.80291748046875, 114.82791900634766, -0.746673583984375, 31.408000946044922, -54.30122375488281, 68.19306945800781, 53.347999572753906, 110.35319519042969, 118.45980834960938, 41.779571533203125, 85.60113525390625, 188.5683135986328, -110.4946517944336, -6.106462478637695, 8.383625030517578, 16.90435791015625, -21.431201934814453, -0.7680988311767578, 178.31329345703125, 135.18670654296875, 13.110164642333984, -9.461585998535156, 54.29996871948242, 12.97601318359375, 161.95065307617188, 123.289306640625, -42.68579864501953, -1.56842041015625, 75.82424926757812, 20.30157470703125, 86.55435180664062, 77.77413940429688, -17.043617248535156, -14.816146850585938, 8.465484619140625, -2.1524276733398438, 18.981369018554688, 28.573272705078125, 19.01226043701172, 30.964340209960938, 96.07728576660156, -13.4779052734375, 146.5467071533203, 224.0789794921875, 119.5357894897461, 60.811248779296875, 21.104827880859375, 31.65412139892578, 103.37594604492188, 2.9089317321777344, 117.22503662109375, -91.36947631835938, 162.25643920898438, 24.655120849609375, -27.651954650878906, 13.005561828613281, 120.75987243652344, 19.192031860351562, 119.989501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000608.npy"}
|
|
{"epoch": 0.9191232048374905, "step": 609, "batch_size": 64, "mean": 60.38801574707031, "std": 69.14053344726562, "min": -169.89767456054688, "p10": -27.11173858642578, "median": 72.64849090576172, "p90": 153.36116027832034, "max": 215.2572784423828, "pos_frac": 0.8125, "sample": [159.7102508544922, 79.31260681152344, 104.56999969482422, 10.399932861328125, 39.3656005859375, 17.464324951171875, 139.61204528808594, 75.89994812011719, -10.867889404296875, -2.8371963500976562, -27.894500732421875, 91.62814331054688, -10.826278686523438, 27.040355682373047, 215.2572784423828, 128.24545288085938, 168.31666564941406, 86.18569946289062, 66.36884307861328, 168.91372680664062, -42.564598083496094, 85.95986938476562, 83.2296142578125, 77.5723876953125, -42.15976333618164, 156.48104858398438, 110.68042755126953, 84.22875213623047, -48.869300842285156, 95.5255355834961, 52.42066192626953, -1.3231525421142578, 158.27915954589844, 85.58599090576172, 16.060943603515625, 91.51393127441406, 61.69741439819336, 0.019845962524414062, 133.60775756835938, 77.51136779785156, 41.6092529296875, -56.82270431518555, 23.537330627441406, 162.23760986328125, 146.0814208984375, -169.89767456054688, 21.811351776123047, 68.77951049804688, 87.28314208984375, 14.303075790405273, 25.18332290649414, 91.63167572021484, 74.96952819824219, 57.315773010253906, 10.080167770385742, 6.538692474365234, 119.73176574707031, 90.29737854003906, 128.49603271484375, -42.437599182128906, 129.32752990722656, 70.32745361328125, 28.411697387695312, -25.285293579101562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000609.npy"}
|
|
{"epoch": 0.9206349206349206, "step": 610, "batch_size": 64, "mean": 56.99535369873047, "std": 78.31913757324219, "min": -93.34817504882812, "p10": -36.043472290039055, "median": 43.845279693603516, "p90": 165.8217529296875, "max": 231.49972534179688, "pos_frac": 0.734375, "sample": [52.80940246582031, 204.09339904785156, 139.877197265625, 118.69108581542969, 41.77312469482422, 135.63365173339844, 23.60455322265625, 7.612756729125977, -56.4189453125, 14.739845275878906, 45.153404235839844, -78.26569366455078, 63.59233474731445, 42.53715515136719, -8.855880737304688, 54.974334716796875, 154.5398406982422, -18.75164031982422, -80.26443481445312, -13.336341857910156, 38.38172149658203, -47.777862548828125, 20.095924377441406, 91.286376953125, -37.72328186035156, -32.12391662597656, 9.672134399414062, 165.88204956054688, 62.72936248779297, 177.05709838867188, 40.086509704589844, 231.49972534179688, 165.68106079101562, 153.61526489257812, -0.61572265625, 128.91783142089844, 147.72389221191406, 0.9509830474853516, 119.38903045654297, 16.081634521484375, 21.001373291015625, 95.53851318359375, 54.242332458496094, -93.34817504882812, 153.46893310546875, -5.184307098388672, 12.416694641113281, -20.995349884033203, 154.2138214111328, 177.44662475585938, 57.32862854003906, 60.91925811767578, 29.358322143554688, -3.782106399536133, -1.255167007446289, -11.794181823730469, -60.47511291503906, 105.26000213623047, 187.06385803222656, 70.09025573730469, 89.92561340332031, 13.52362060546875, 187.20635986328125, 80.98402404785156], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000610.npy"}
|
|
{"epoch": 0.9221466364323507, "step": 611, "batch_size": 64, "mean": 43.20577621459961, "std": 64.4560775756836, "min": -123.41372680664062, "p10": -39.20527687072753, "median": 47.22472381591797, "p90": 132.52520599365235, "max": 189.38638305664062, "pos_frac": 0.703125, "sample": [86.80007934570312, 77.7200927734375, 17.133575439453125, 64.02122497558594, 127.21263122558594, -1.4954757690429688, -8.278961181640625, 55.73500061035156, -5.512256622314453, 49.48838806152344, -7.788234710693359, 140.22035217285156, 52.53839874267578, 83.77288818359375, -5.966911315917969, -9.15899658203125, -54.26381301879883, 134.1210479736328, -42.78664779663086, 189.38638305664062, 101.60462951660156, 102.2756118774414, 71.10256958007812, -42.0766487121582, -30.760730743408203, -5.949028015136719, 135.00387573242188, 57.30146026611328, 64.60535430908203, 123.84796142578125, 5.449378967285156, -52.101585388183594, -0.20098114013671875, 163.7572021484375, -14.2109375, 18.193374633789062, -123.41372680664062, 132.4426727294922, 36.14220428466797, 53.0489501953125, 92.62416076660156, 72.11295318603516, -85.96331024169922, 71.83685302734375, 22.87670135498047, 27.312232971191406, 60.30842590332031, 31.574737548828125, 81.73255920410156, 92.81692504882812, 83.7550048828125, 132.56057739257812, 16.415998458862305, 52.72852325439453, 6.854984283447266, 36.521942138671875, 146.17987060546875, -68.6210708618164, 120.20154571533203, 23.11920928955078, -32.505409240722656, 1.4248886108398438, -4.620141983032227, 44.9610595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000611.npy"}
|
|
{"epoch": 0.9236583522297808, "step": 612, "batch_size": 64, "mean": 59.692298889160156, "std": 76.77848815917969, "min": -167.5363006591797, "p10": -22.325021362304685, "median": 54.95181655883789, "p90": 162.69558105468752, "max": 218.1883544921875, "pos_frac": 0.78125, "sample": [-1.5178680419921875, -30.39404296875, 86.61767578125, -13.224464416503906, -16.260879516601562, 7.599601745605469, 125.02912902832031, 113.93496704101562, -6.4470672607421875, 16.373497009277344, 115.9928970336914, 56.35890197753906, 65.57994842529297, 60.26234436035156, 74.1584701538086, 38.677825927734375, 32.30927276611328, 14.277961730957031, 55.1607666015625, 89.78887939453125, 127.67514038085938, -68.48766326904297, 117.74791717529297, 33.42543029785156, 52.246002197265625, 20.588977813720703, 195.47779846191406, -19.446979522705078, -23.558467864990234, 14.991935729980469, 49.75513458251953, 147.1908416748047, 1.7946434020996094, 0.5899829864501953, -13.828437805175781, 165.58804321289062, 47.701332092285156, 101.81243133544922, 218.1883544921875, -122.61604309082031, 186.26805114746094, 43.67576217651367, 114.70686340332031, -55.41749572753906, 184.4085693359375, 106.86573791503906, 87.2122802734375, -25.716856002807617, 145.40907287597656, 190.87705993652344, 54.74286651611328, -3.9746627807617188, 203.03366088867188, 66.32771301269531, 32.91632080078125, 155.94650268554688, 77.37677764892578, 38.313926696777344, 70.97148132324219, -167.5363006591797, 54.55809783935547, 98.25843048095703, 103.81562042236328, 126.1534652709961], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000612.npy"}
|
|
{"epoch": 0.9251700680272109, "step": 613, "batch_size": 64, "mean": 60.82259750366211, "std": 74.56514739990234, "min": -136.32772827148438, "p10": -12.22029342651367, "median": 54.280982971191406, "p90": 164.59518127441407, "max": 194.20394897460938, "pos_frac": 0.78125, "sample": [-10.123481750488281, 51.321136474609375, -5.604705810546875, 30.644088745117188, 108.09058380126953, 109.8381118774414, 96.2005386352539, 63.11151123046875, -45.15972137451172, -2.217226028442383, 43.95525360107422, 53.144535064697266, -124.87681579589844, 86.56182861328125, 111.37273406982422, 50.72779846191406, 4.513425827026367, 23.17732810974121, 187.5167236328125, 107.89884185791016, 1.9343109130859375, -29.37860679626465, -136.32772827148438, 13.893205642700195, 156.47825622558594, 69.23507690429688, 129.63832092285156, 32.27642822265625, 20.539268493652344, 27.163429260253906, 194.20394897460938, 54.10544967651367, 72.5302734375, -3.722087860107422, -2.6456222534179688, 54.45651626586914, -50.371726989746094, 96.05429077148438, 87.04493713378906, 179.47201538085938, 150.84957885742188, 159.83258056640625, 57.232765197753906, 146.5699462890625, 171.2294921875, 1.355133056640625, 185.1865234375, 8.976587295532227, -13.118927001953125, 149.3568572998047, -7.8174591064453125, -6.4965362548828125, 126.0732421875, 184.99501037597656, 69.84674072265625, 83.79238891601562, 18.911701202392578, 164.915771484375, 84.63232421875, 163.84713745117188, 27.917076110839844, 56.00264358520508, 48.21337127685547, -46.330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000613.npy"}
|
|
{"epoch": 0.926681783824641, "step": 614, "batch_size": 64, "mean": 53.770999908447266, "std": 68.12210083007812, "min": -127.27287292480469, "p10": -28.6245979309082, "median": 60.33407974243164, "p90": 132.89063568115233, "max": 215.9451904296875, "pos_frac": 0.765625, "sample": [132.9124755859375, 100.72344207763672, 30.547626495361328, 104.7651138305664, 59.73383331298828, 120.33150482177734, 19.036327362060547, 0.3326568603515625, 61.05378341674805, 39.75685501098633, 136.8693389892578, 86.96837615966797, 132.8396759033203, 135.1829376220703, 97.91036987304688, 89.53985595703125, 101.97972106933594, 88.63697052001953, 105.17501068115234, 94.0252685546875, -54.457908630371094, 6.604034423828125, -25.815673828125, -118.84817504882812, 215.9451904296875, -18.318077087402344, -12.831991195678711, -47.014373779296875, 116.61050415039062, -1.87750244140625, -29.82842254638672, 8.136188507080078, -3.7715415954589844, 127.68596649169922, 30.10519790649414, 25.283411026000977, 118.27006530761719, 96.98092651367188, -127.27287292480469, 92.60965728759766, -11.598251342773438, 157.14340209960938, 94.7757339477539, 107.71390533447266, 63.93976593017578, 84.09420013427734, 26.181861877441406, 48.2794189453125, 116.4090576171875, 26.257064819335938, 27.687137603759766, 41.47471618652344, -35.91169738769531, -8.337989807128906, 0.12445449829101562, 149.67745971679688, 68.54960632324219, -23.20934295654297, 125.33197021484375, 60.934326171875, 46.538177490234375, 55.01893997192383, -56.749603271484375, 140.50393676757812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000614.npy"}
|
|
{"epoch": 0.9281934996220711, "step": 615, "batch_size": 64, "mean": 58.68397521972656, "std": 87.02568054199219, "min": -153.62167358398438, "p10": -28.236441802978515, "median": 49.82448196411133, "p90": 170.87498779296874, "max": 233.26849365234375, "pos_frac": 0.75, "sample": [37.059410095214844, 87.39778137207031, 137.5194854736328, 45.79985809326172, 56.59281921386719, 138.92001342773438, 54.57685089111328, -134.35110473632812, 163.14013671875, 54.878761291503906, -71.82247161865234, 64.32906341552734, -87.000244140625, 97.42359924316406, 46.42755889892578, -66.1263427734375, 96.14891815185547, 189.43710327148438, 94.1449966430664, -5.4564208984375, 148.67727661132812, -18.75518035888672, 176.92578125, -5.741096496582031, 123.10758209228516, 32.470947265625, -29.093528747558594, 54.4757080078125, 8.63629150390625, 170.85206604003906, 147.98104858398438, 40.22954559326172, -7.290473937988281, 8.230701446533203, 23.760475158691406, 162.02854919433594, 162.19534301757812, -26.236572265625, 2.9546356201171875, 166.34805297851562, 171.22369384765625, 181.00869750976562, 153.62144470214844, 24.966171264648438, 111.29312133789062, 23.707630157470703, 28.302322387695312, -22.873565673828125, 53.221405029296875, 170.8848114013672, -8.74029541015625, 187.09600830078125, -3.854766845703125, 81.910888671875, -4.4744415283203125, 233.26849365234375, 30.2843017578125, 77.215576171875, 161.12698364257812, 15.158782958984375, 31.619895935058594, 14.332313537597656, -141.70022583007812, -153.62167358398438], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000615.npy"}
|
|
{"epoch": 0.9297052154195011, "step": 616, "batch_size": 64, "mean": 47.39195251464844, "std": 72.93856811523438, "min": -116.40699768066406, "p10": -21.409159851074214, "median": 31.589420318603516, "p90": 153.60762634277347, "max": 199.98211669921875, "pos_frac": 0.703125, "sample": [175.59378051757812, 9.348922729492188, 15.2655029296875, -3.90496826171875, 129.1611328125, -16.34479331970215, 199.98211669921875, 67.40254211425781, -0.4297904968261719, 61.901832580566406, 43.233253479003906, 5.111042022705078, 53.66893768310547, -114.83302307128906, 15.938777923583984, -58.13425064086914, 27.576385498046875, -26.816577911376953, 1.6978988647460938, 176.34613037109375, 33.915992736816406, 167.16082763671875, -13.657880783081055, -22.451601028442383, 168.99221801757812, -2.5876636505126953, -5.227537155151367, -6.993125915527344, -15.747926712036133, 156.897216796875, 114.33271789550781, 6.369529724121094, 66.98515319824219, 118.55384826660156, 186.6659393310547, 109.12837219238281, -18.633098602294922, 68.83021545410156, 25.38226318359375, 145.93191528320312, 124.82522583007812, 77.41534423828125, 20.46894645690918, 66.51438903808594, 125.45487213134766, 134.59909057617188, -18.976797103881836, 11.322689056396484, -116.40699768066406, 10.749702453613281, -10.756633758544922, -69.3626708984375, 54.12122344970703, 29.262847900390625, 37.06806182861328, 144.33741760253906, 47.36463928222656, 8.92105484008789, 85.71476745605469, -11.697120666503906, 127.07439422607422, 104.85933685302734, 37.88165283203125, -33.2828369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000616.npy"}
|
|
{"epoch": 0.9312169312169312, "step": 617, "batch_size": 64, "mean": 44.974143981933594, "std": 71.14928436279297, "min": -115.38555908203125, "p10": -32.705802154541004, "median": 32.27157402038574, "p90": 136.06387786865236, "max": 230.10427856445312, "pos_frac": 0.78125, "sample": [31.34387969970703, 29.965572357177734, 1.4112701416015625, 45.273963928222656, 171.51414489746094, 84.25798034667969, 124.0098876953125, 132.45494079589844, 10.432910919189453, -53.506080627441406, 26.468521118164062, 29.670860290527344, -16.808334350585938, 13.725461959838867, 4.442478179931641, 116.282470703125, -4.213035583496094, 137.61056518554688, 28.691162109375, 8.649658203125, 230.10427856445312, -6.473636627197266, 7.302135467529297, 131.6260986328125, 63.69151306152344, -1.6425323486328125, -1.1997718811035156, 97.27658081054688, 37.39131164550781, 28.413524627685547, 43.74196243286133, -0.22907638549804688, 28.554443359375, 157.41091918945312, 74.28372192382812, 185.8494873046875, 37.50647735595703, -85.42467498779297, -47.124290466308594, 124.43087768554688, 112.10387420654297, 17.158981323242188, -37.221839904785156, 4.003963470458984, 153.96292114257812, 129.90769958496094, -112.59635162353516, -115.38555908203125, -108.42588806152344, 35.526397705078125, 68.74229431152344, 96.62553405761719, 75.46656799316406, 10.226339340209961, 152.44259643554688, 84.58726501464844, 37.92504119873047, 7.392601013183594, 62.078147888183594, 25.012176513671875, 33.19926834106445, 67.7997055053711, 72.81422424316406, -22.168380737304688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000617.npy"}
|
|
{"epoch": 0.9327286470143613, "step": 618, "batch_size": 64, "mean": 52.114044189453125, "std": 70.38123321533203, "min": -84.07301330566406, "p10": -23.510491561889634, "median": 36.222068786621094, "p90": 152.14276428222658, "max": 236.63742065429688, "pos_frac": 0.8125, "sample": [197.5670928955078, 150.37062072753906, 34.619564056396484, 115.60978698730469, 6.350868225097656, -4.69175910949707, -36.19071578979492, 2.9449996948242188, 148.29241943359375, 0.19741439819335938, 6.019432067871094, 152.6934814453125, 124.94361877441406, 49.94770812988281, 170.82809448242188, 4.1238555908203125, 37.223228454589844, 2.0944061279296875, 236.63742065429688, 88.77566528320312, 36.42285919189453, 188.71705627441406, 4.459497451782227, 35.423316955566406, -1.1419906616210938, 165.61715698242188, 51.448333740234375, 47.05170440673828, 90.38582611083984, 42.4635009765625, 7.5611419677734375, 29.69359588623047, 161.26312255859375, 10.879337310791016, 56.67759704589844, -0.85101318359375, 66.90829467773438, -29.733421325683594, 95.82077026367188, 33.62537384033203, -67.45835876464844, 0.5881767272949219, -8.99032211303711, -76.45155334472656, 66.58099365234375, -42.94432067871094, 33.827064514160156, -84.07301330566406, 94.4969711303711, 150.85775756835938, 38.643348693847656, 97.89291381835938, 13.760833740234375, 9.231353759765625, 32.372222900390625, 51.62982177734375, -8.268905639648438, 71.05205535888672, 85.53227233886719, 141.42916870117188, 7.8438568115234375, 148.26531982421875, -37.589141845703125, 36.021278381347656], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000618.npy"}
|
|
{"epoch": 0.9342403628117913, "step": 619, "batch_size": 64, "mean": 52.449317932128906, "std": 71.40641784667969, "min": -151.49517822265625, "p10": -25.03207340240478, "median": 52.93910217285156, "p90": 159.90054168701172, "max": 196.8798065185547, "pos_frac": 0.75, "sample": [76.89955139160156, -3.6334075927734375, 2.773347854614258, 14.619430541992188, 159.9253387451172, 159.84268188476562, 160.8297119140625, -40.47089385986328, -81.5279541015625, 196.8798065185547, 148.35531616210938, 57.00825119018555, 90.37583923339844, 5.997182846069336, 77.53302001953125, -19.241588592529297, 61.836204528808594, 73.0498046875, -6.300317764282227, -1.22967529296875, -68.5284652709961, 184.28158569335938, 14.155776977539062, -32.51399230957031, -20.538911819458008, 109.38809967041016, 95.58150482177734, 95.34294128417969, 2.0260448455810547, -15.884628295898438, 19.73119354248047, -151.49517822265625, -26.957714080810547, 127.30892944335938, 108.2297592163086, 111.0274658203125, 16.594268798828125, 162.92654418945312, 175.37350463867188, -40.95063781738281, 75.43429565429688, 149.36557006835938, 56.44068908691406, -1.862874984741211, 114.90974426269531, 92.62617492675781, 66.405517578125, 109.15401458740234, 86.16416931152344, 31.17385482788086, 22.103126525878906, 67.89736938476562, 49.43751525878906, 18.5670166015625, 172.2362060546875, 13.470176696777344, 31.096343994140625, 83.57860565185547, -3.248258590698242, 35.43891906738281, -4.377655029296875, 60.23949432373047, 16.382797241210938, 15.503681182861328], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000619.npy"}
|
|
{"epoch": 0.9357520786092215, "step": 620, "batch_size": 64, "mean": 45.923828125, "std": 72.46736907958984, "min": -86.5489501953125, "p10": -35.94379615783691, "median": 31.93645477294922, "p90": 144.83030548095704, "max": 198.00559997558594, "pos_frac": 0.734375, "sample": [24.90357208251953, -80.6976318359375, -25.40821075439453, 146.51239013671875, -80.83284759521484, 1.8766937255859375, 33.57283020019531, 145.24095153808594, 9.691810607910156, 78.58048248291016, 122.89501953125, 133.08560180664062, 77.95437622070312, 65.85940551757812, 7.949684143066406, 164.43951416015625, 94.99066162109375, 38.51606750488281, -25.742156982421875, 3.3388214111328125, -52.40191650390625, 15.889751434326172, -86.5489501953125, 190.08010864257812, -37.96734619140625, 141.2039337158203, -5.745002746582031, -13.044921875, 172.83554077148438, -20.62657928466797, 130.7671356201172, 130.53717041015625, -24.864307403564453, -17.46186637878418, -31.222179412841797, 18.387161254882812, 13.530170440673828, 37.28865051269531, 143.87213134765625, 6.620635986328125, 96.03041076660156, 27.756027221679688, 68.77238464355469, 53.77440643310547, 49.9472541809082, -65.37648010253906, 198.00559997558594, 169.1453857421875, -23.546377182006836, 101.39601135253906, 1.974405288696289, 143.65184020996094, 30.300079345703125, -9.791410446166992, -68.57737731933594, 20.465200424194336, 114.51205444335938, 26.698486328125, 43.69437789916992, 29.363494873046875, 94.75928497314453, 73.88529205322266, 75.95350646972656, 38.474761962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000620.npy"}
|
|
{"epoch": 0.9372637944066515, "step": 621, "batch_size": 64, "mean": 46.51704406738281, "std": 74.62489318847656, "min": -116.79678344726562, "p10": -34.975292587280265, "median": 35.206106185913086, "p90": 148.25009765625, "max": 254.8575439453125, "pos_frac": 0.734375, "sample": [-0.4783477783203125, 17.096839904785156, 20.012718200683594, -5.421230316162109, -56.745391845703125, -115.37680053710938, 75.82058715820312, 107.1380615234375, -26.411163330078125, 9.473794937133789, 58.35086441040039, 86.20791625976562, 156.44204711914062, 9.367589950561523, 161.4733428955078, 31.934783935546875, 44.74207305908203, 63.26704406738281, 70.53610229492188, 125.18011474609375, 59.98725891113281, 31.490478515625, 177.1171112060547, -116.79678344726562, 175.01657104492188, 117.89508056640625, 67.89622497558594, -101.87135314941406, 65.57684326171875, -38.645633697509766, 177.97836303710938, 71.85118865966797, 71.52706909179688, 132.87799072265625, 127.32228088378906, -0.502838134765625, 56.305423736572266, 6.867547988891602, 64.67982482910156, -24.40057373046875, -38.83265686035156, 12.678503036499023, 143.4110870361328, 30.6011962890625, 19.67867088317871, -9.965045928955078, 25.818038940429688, -6.2694549560546875, -11.170181274414062, 35.73508834838867, 2.1578197479248047, 106.72626495361328, 64.16519165039062, 16.670541763305664, 103.37306213378906, -15.522716522216797, -6.7012786865234375, 254.8575439453125, -102.3593978881836, 73.28004455566406, 34.6771240234375, 19.661361694335938, 150.32395935058594, 119.31292724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000621.npy"}
|
|
{"epoch": 0.9387755102040817, "step": 622, "batch_size": 64, "mean": 48.64423370361328, "std": 68.3425521850586, "min": -92.87818908691406, "p10": -27.52618961334228, "median": 45.36263656616211, "p90": 139.02353515625, "max": 188.46182250976562, "pos_frac": 0.734375, "sample": [108.78096771240234, 21.390933990478516, -3.6631088256835938, -92.87818908691406, 13.191261291503906, 33.550621032714844, -4.6763458251953125, 45.558074951171875, -53.276641845703125, -4.2754058837890625, 187.70108032226562, 94.23866271972656, -0.2580585479736328, -48.50114822387695, 137.65745544433594, 58.66326141357422, 140.68228149414062, 116.98080444335938, 76.47261047363281, 148.97540283203125, 95.4439468383789, 132.9285888671875, -8.979543685913086, -9.974287033081055, 188.46182250976562, 46.33478546142578, 48.18608093261719, 93.0853042602539, 1.7518806457519531, 68.41999816894531, 142.38818359375, -80.04782104492188, 47.37312316894531, 13.667118072509766, -13.830345153808594, 1.690408706665039, 138.339599609375, 87.7952651977539, 85.76580810546875, 17.05597686767578, 137.52426147460938, 19.741355895996094, 119.13308715820312, -20.876615524291992, 77.98316955566406, 125.67071533203125, 111.78045654296875, -65.53678131103516, -1.3695030212402344, 38.3380126953125, 139.316650390625, 12.035758972167969, 1.141143798828125, -30.376007080078125, 1.1098690032958984, -14.368942260742188, 159.2160186767578, 45.167198181152344, -64.22489929199219, 1.0623855590820312, 65.91944885253906, 10.775671005249023, 125.9966812133789, 45.9017333984375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000622.npy"}
|
|
{"epoch": 0.9402872260015117, "step": 623, "batch_size": 64, "mean": 52.31147003173828, "std": 70.39192199707031, "min": -98.8924560546875, "p10": -25.941195297241208, "median": 51.235870361328125, "p90": 143.94938659667972, "max": 198.81353759765625, "pos_frac": 0.734375, "sample": [121.7363052368164, 31.763633728027344, -13.899250030517578, -86.72608947753906, 54.41654968261719, 176.14505004882812, 84.29256439208984, 108.03012084960938, 105.971923828125, 77.45438385009766, -59.61070251464844, 40.1724853515625, 151.97879028320312, 42.09986114501953, -18.842727661132812, 190.18087768554688, -24.355121612548828, 118.83822631835938, 71.72223663330078, 72.3899154663086, 51.58018493652344, 27.377595901489258, 119.46173858642578, -17.557296752929688, -14.980430603027344, 81.50247192382812, 113.62843322753906, 145.62081909179688, -98.8924560546875, 33.00909423828125, 100.2252197265625, 98.01148986816406, 71.23477172851562, 92.90182495117188, -26.620941162109375, -83.14525604248047, 129.91822814941406, -9.279205322265625, 115.87577056884766, 140.04937744140625, 164.12069702148438, 171.53683471679688, 70.51243591308594, 5.266481399536133, -19.011680603027344, -10.284086227416992, -35.15697479248047, 22.72224235534668, 4.789531707763672, 97.42713928222656, 83.10220336914062, 198.81353759765625, 50.89155578613281, 74.02326965332031, 2.097381591796875, 23.39508056640625, -4.4541778564453125, 2.8791656494140625, 49.501373291015625, 9.304059982299805, 108.75322723388672, -9.245277404785156, 26.937896728515625, -53.668312072753906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000623.npy"}
|
|
{"epoch": 0.9417989417989417, "step": 624, "batch_size": 64, "mean": 56.121585845947266, "std": 79.31605529785156, "min": -141.0371551513672, "p10": -23.383354759216306, "median": 34.82566833496094, "p90": 164.8141845703125, "max": 213.42227172851562, "pos_frac": 0.703125, "sample": [-65.36105346679688, -22.041763305664062, 125.88189697265625, 85.90678405761719, 165.68893432617188, 149.3202667236328, -3.993865966796875, 14.116043090820312, -14.032135009765625, -17.489091873168945, 52.51195526123047, 28.35577392578125, 72.41716003417969, 86.35980224609375, 137.4983673095703, 99.67410278320312, 31.397653579711914, 144.257080078125, 177.95248413085938, 19.576271057128906, -5.930912017822266, -38.73494338989258, -23.958322525024414, 159.07208251953125, 3.421039581298828, 103.86736297607422, 27.698524475097656, -10.526283264160156, 31.0675048828125, 36.341217041015625, 192.85238647460938, 162.77310180664062, 94.10232543945312, 213.42227172851562, 19.517078399658203, -49.12704849243164, 168.9676513671875, 131.98863220214844, 204.1937713623047, 116.66461944580078, 49.19780731201172, 128.93637084960938, 81.67668914794922, 33.31011962890625, -36.81666564941406, -2.4195098876953125, 189.2919464111328, -4.40380859375, -14.69627571105957, 125.84123992919922, 67.79586791992188, 3.4199390411376953, -9.279426574707031, 5.09901237487793, -141.0371551513672, 94.8109359741211, 9.774124145507812, -13.092720031738281, 144.71987915039062, -16.669002532958984, -88.70719146728516, 1.4200706481933594, 42.178466796875, 135.76193237304688], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000624.npy"}
|
|
{"epoch": 0.9433106575963719, "step": 625, "batch_size": 64, "mean": 42.43467712402344, "std": 78.93307495117188, "min": -132.12155151367188, "p10": -50.96083946228027, "median": 30.265029907226562, "p90": 166.90492095947272, "max": 215.06222534179688, "pos_frac": 0.703125, "sample": [41.3673095703125, 91.1596450805664, 38.3204345703125, 36.1278076171875, -2.928476333618164, 104.5804443359375, 57.84849548339844, 55.51637268066406, 8.028379440307617, 215.06222534179688, 62.86796951293945, -31.906139373779297, 136.6302032470703, -12.093692779541016, 186.15521240234375, 29.152130126953125, 189.1250762939453, 22.056747436523438, -53.34444808959961, -3.2936019897460938, 82.28091430664062, 124.93814086914062, 13.890434265136719, -65.99884033203125, -10.091110229492188, 47.17131805419922, 30.2447509765625, 71.8474349975586, 13.802196502685547, 26.525665283203125, -3.853839874267578, 148.56883239746094, 187.54803466796875, -12.045372009277344, 92.52777099609375, -5.493473052978516, -7.159414291381836, 211.20626831054688, 0.16165924072265625, 14.544692993164062, 78.3399658203125, 134.94644165039062, 174.76324462890625, 85.13772583007812, 65.54192352294922, -16.405746459960938, -128.00540161132812, -90.38101196289062, -75.84210205078125, 101.50047302246094, 13.866279602050781, -93.69058227539062, -132.12155151367188, 176.68899536132812, 97.02715301513672, 3.3843536376953125, 91.75289916992188, 17.771865844726562, 53.822364807128906, -45.399085998535156, 30.285308837890625, 50.52029800415039, 10.284658432006836, -19.017295837402344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000625.npy"}
|
|
{"epoch": 0.9448223733938019, "step": 626, "batch_size": 64, "mean": 58.080322265625, "std": 88.23892211914062, "min": -105.78582763671875, "p10": -55.046099853515614, "median": 50.32401657104492, "p90": 161.07769470214845, "max": 286.7709655761719, "pos_frac": 0.765625, "sample": [154.45233154296875, 21.996475219726562, 141.66387939453125, -39.6387939453125, 254.50543212890625, 157.97579956054688, 168.38421630859375, -36.68315124511719, 80.89297485351562, 158.4823455810547, -27.3488826751709, 11.339714050292969, 42.372581481933594, 2.598529815673828, -47.53339385986328, 158.4351806640625, 92.81278228759766, -89.18528747558594, 186.39227294921875, 23.990312576293945, -101.14739990234375, 7.89306640625, 63.482826232910156, 160.02630615234375, 15.929229736328125, 26.59404945373535, 92.3951187133789, 58.27545166015625, -42.46607971191406, 286.7709655761719, 61.45800018310547, -40.231651306152344, 161.52828979492188, 128.39047241210938, 34.55632019042969, -105.78582763671875, -7.396453857421875, 89.5316162109375, 1.911325454711914, 93.23918914794922, 20.91840362548828, 27.65464210510254, 161.81423950195312, 25.411014556884766, -34.61674499511719, 11.73845100402832, 39.685577392578125, 151.71017456054688, 70.89610290527344, 114.53871154785156, 125.98710632324219, 6.286663055419922, -77.97601318359375, 116.99505615234375, -71.63116455078125, 92.752197265625, 19.661739349365234, 157.03001403808594, 135.6238250732422, 85.58465576171875, 194.2223358154297, -62.401458740234375, 62.660682678222656, -58.265830993652344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000626.npy"}
|
|
{"epoch": 0.9463340891912321, "step": 627, "batch_size": 64, "mean": 62.98246765136719, "std": 78.8995361328125, "min": -148.47088623046875, "p10": -17.352007675170896, "median": 62.586181640625, "p90": 167.30467987060547, "max": 210.5204315185547, "pos_frac": 0.78125, "sample": [57.614444732666016, 16.38051986694336, 73.69385528564453, 94.89254760742188, 70.83027648925781, 86.49644470214844, 150.70889282226562, 205.05406188964844, 205.404296875, 59.834136962890625, -15.488998413085938, 168.1021270751953, 164.20138549804688, 116.20208740234375, 39.70237731933594, 36.658447265625, -1.9579010009765625, 30.08643341064453, 177.8758544921875, 66.77456665039062, 141.58201599121094, 80.2368392944336, 26.405494689941406, 21.228515625, 14.867660522460938, -83.85719299316406, 167.43186950683594, -44.15669250488281, 115.93997192382812, -18.150440216064453, 3.0664539337158203, 167.00790405273438, 112.40341186523438, -3.7251663208007812, 64.74771118164062, 7.7786407470703125, -8.416038513183594, 62.26983642578125, -30.545244216918945, 162.28521728515625, 210.5204315185547, 64.9721450805664, -64.62176513671875, -11.804367065429688, 14.566505432128906, -2.9115142822265625, -6.840545654296875, 114.41641998291016, 32.97344970703125, 149.0790557861328, 60.87670135498047, -148.47088623046875, 26.811199188232422, 96.73822021484375, 157.38217163085938, 82.43142700195312, 9.704978942871094, -129.6502685546875, 62.90252685546875, 41.768165588378906, 102.3759765625, 132.73468017578125, 100.19042205810547, 173.26622009277344], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000627.npy"}
|
|
{"epoch": 0.9478458049886621, "step": 628, "batch_size": 64, "mean": 62.58487319946289, "std": 86.23270416259766, "min": -193.32920837402344, "p10": -57.38705139160154, "median": 68.88978576660156, "p90": 156.89990234375, "max": 207.20155334472656, "pos_frac": 0.796875, "sample": [175.24195861816406, 63.36317443847656, 61.80091857910156, 78.95555114746094, -27.613937377929688, -73.01739501953125, 141.607666015625, 11.565582275390625, 136.71783447265625, 107.38478088378906, 178.10372924804688, 181.70358276367188, 104.65396118164062, 38.82566452026367, 46.018959045410156, 88.99112701416016, -193.32920837402344, -125.95088958740234, 132.91294860839844, 30.35964584350586, 155.913818359375, 67.33538818359375, 30.66301155090332, -85.43089294433594, 144.5414276123047, 127.17156982421875, 70.44418334960938, 90.62506103515625, 54.73509216308594, 157.322509765625, 134.2512664794922, 154.82696533203125, 20.8446044921875, 169.0218505859375, 32.775020599365234, 158.98504638671875, -17.9307861328125, 2.5424728393554688, 137.19662475585938, -67.18495178222656, 108.59368896484375, 79.48029327392578, 45.284568786621094, 126.67313385009766, 63.00347137451172, 66.97758483886719, 149.59815979003906, 77.48056030273438, 142.025634765625, 147.5276336669922, 10.482732772827148, -34.52528381347656, 7.7634735107421875, -148.58998107910156, 51.835697174072266, 207.20155334472656, 7.966644287109375, 142.14695739746094, -32.59922790527344, 88.38143920898438, -28.866302490234375, 105.3930435180664, -3.868053436279297, -70.88023376464844], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000628.npy"}
|
|
{"epoch": 0.9493575207860923, "step": 629, "batch_size": 64, "mean": 49.82585906982422, "std": 65.65707397460938, "min": -85.69971466064453, "p10": -16.04401092529297, "median": 30.32353973388672, "p90": 160.61976928710942, "max": 193.28659057617188, "pos_frac": 0.796875, "sample": [-26.60314178466797, 61.704383850097656, -85.69971466064453, 77.55299377441406, 64.74406433105469, 31.275009155273438, 6.7564544677734375, 66.43927001953125, -70.21504211425781, 43.53968811035156, 9.184295654296875, 21.93097686767578, 7.705820083618164, 173.4322509765625, 15.099725723266602, -29.443256378173828, 16.193784713745117, 29.3720703125, 18.25830078125, 51.36895751953125, 3.44976806640625, -7.748039245605469, -8.531229019165039, 43.43715286254883, -26.979389190673828, 32.503089904785156, 181.83680725097656, 115.45431518554688, -4.483800888061523, 151.92086791992188, 118.4962158203125, -16.05071258544922, 62.34272766113281, 76.4596939086914, 12.11264419555664, 77.69601440429688, 95.65205383300781, 19.388153076171875, 7.879779815673828, -0.17992591857910156, 6.386199951171875, 38.054134368896484, 179.18899536132812, 3.5996742248535156, -9.437660217285156, 26.577476501464844, 26.634109497070312, 108.93260192871094, 94.21820068359375, 111.61900329589844, 14.099441528320312, 47.46835708618164, 180.12765502929688, 20.217317581176758, 69.96270751953125, 183.8767547607422, -16.02837371826172, 193.28659057617188, 150.10189819335938, 164.34786987304688, 21.61102294921875, 137.19488525390625, -25.40545654296875, 44.96862030029297], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000629.npy"}
|
|
{"epoch": 0.9508692365835223, "step": 630, "batch_size": 64, "mean": 58.50631332397461, "std": 74.19185638427734, "min": -101.32279205322266, "p10": -30.252120971679688, "median": 46.84414100646973, "p90": 159.89211273193362, "max": 233.02395629882812, "pos_frac": 0.765625, "sample": [-51.73398208618164, 89.5113525390625, 11.607067108154297, 5.521293640136719, 106.5811767578125, 169.6692657470703, 96.09856414794922, -4.056331634521484, 192.6220703125, 131.0296630859375, -2.9819869995117188, 127.348876953125, -11.9609375, 145.73480224609375, 12.499168395996094, 117.2781753540039, -60.3426513671875, 61.353736877441406, 120.8454360961914, -71.54733276367188, 24.685211181640625, -6.648704528808594, 57.40056610107422, 150.63046264648438, 156.36129760742188, 33.047935485839844, -41.74236297607422, 47.97288131713867, -101.32279205322266, 106.47460174560547, 2.8176422119140625, 34.17567443847656, 3.355457305908203, 186.42832946777344, 75.5799560546875, 25.897796630859375, 161.4053192138672, 14.616287231445312, 36.70039367675781, 84.04119110107422, 233.02395629882812, 30.937973022460938, 140.58087158203125, -57.111454010009766, 162.61257934570312, 96.84638214111328, 84.64422607421875, 40.809661865234375, 73.38523864746094, -30.639251708984375, 39.712318420410156, -29.34881591796875, -17.966625213623047, 166.97726440429688, 105.82127380371094, -1.0422840118408203, 58.19532012939453, 7.207477569580078, 128.0065155029297, 151.284423828125, -15.863662719726562, 84.36367797851562, 45.71540069580078, 9.297069549560547], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000630.npy"}
|
|
{"epoch": 0.9523809523809523, "step": 631, "batch_size": 64, "mean": 68.33146667480469, "std": 86.36024475097656, "min": -129.84500122070312, "p10": -34.11054801940918, "median": 56.698665618896484, "p90": 179.83003234863284, "max": 258.7642822265625, "pos_frac": 0.703125, "sample": [140.47348022460938, 51.10859680175781, -3.19769287109375, 166.72845458984375, 93.60970306396484, 157.979736328125, 156.0283203125, 115.67377471923828, -3.0516090393066406, -0.929473876953125, -13.977264404296875, -27.67764663696289, 103.26519012451172, -13.875442504882812, -6.809864044189453, 49.06812286376953, -32.29134750366211, 23.276485443115234, 66.13005828857422, 206.15142822265625, 198.60304260253906, -39.558441162109375, 175.14794921875, 169.8992156982422, 135.8072509765625, 68.41659545898438, 25.194732666015625, 208.21817016601562, -0.8691844940185547, 84.17582702636719, 36.689632415771484, 30.244659423828125, -99.14010620117188, 82.69789123535156, 60.86155319213867, 201.33641052246094, 148.1949005126953, 19.338348388671875, 258.7642822265625, 151.26036071777344, 30.921653747558594, -58.09385681152344, -50.67060852050781, 99.96861267089844, -14.297531127929688, -129.84500122070312, 8.649820327758789, 24.413667678833008, -49.56561279296875, 99.5762939453125, -34.89020538330078, 113.66584777832031, -3.679412841796875, 173.3269805908203, -15.775928497314453, 146.24533081054688, 35.28353500366211, 181.83663940429688, 52.5357780456543, 190.26190185546875, 89.45829772949219, 162.0277862548828, 37.68241500854492, 141.21144104003906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000631.npy"}
|
|
{"epoch": 0.9538926681783825, "step": 632, "batch_size": 64, "mean": 65.62427520751953, "std": 70.59999084472656, "min": -93.65469360351562, "p10": -19.161856460571286, "median": 58.82337188720703, "p90": 157.26509552001954, "max": 206.20111083984375, "pos_frac": 0.8125, "sample": [82.34590911865234, 116.29974365234375, 61.87969970703125, 6.424495697021484, 69.72123718261719, 157.60191345214844, -3.3060646057128906, 100.64936065673828, -14.491893768310547, 159.27737426757812, 59.412261962890625, -30.014312744140625, 45.20051574707031, -5.928524017333984, -52.23851776123047, 143.05604553222656, 158.77354431152344, 8.815666198730469, 152.9786376953125, -52.16590881347656, 48.80259704589844, 37.63435745239258, -78.30899047851562, 24.893386840820312, 38.25586700439453, 100.93987274169922, 42.726043701171875, 130.03933715820312, 52.57478713989258, 41.12314987182617, 137.5599365234375, 206.20111083984375, 54.57588195800781, -21.16326904296875, 119.04116821289062, 32.58735656738281, 149.47454833984375, 58.23448181152344, 91.6344223022461, 79.6627426147461, 143.332763671875, 140.42868041992188, 12.778366088867188, -4.666984558105469, 6.484737396240234, 167.64352416992188, 110.91085815429688, 61.364288330078125, 156.47918701171875, 134.06591796875, 167.6961669921875, -93.65469360351562, 63.545135498046875, 104.87887573242188, -7.2807464599609375, 16.402488708496094, 174.73208618164062, 146.98797607421875, 128.07615661621094, 28.76494026184082, 40.817665100097656, 31.601276397705078, -74.203125, 31.98836898803711], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000632.npy"}
|
|
{"epoch": 0.9554043839758125, "step": 633, "batch_size": 64, "mean": 49.17240905761719, "std": 74.78765106201172, "min": -103.31390380859375, "p10": -29.63682060241699, "median": 38.91164016723633, "p90": 156.89699707031252, "max": 209.75674438476562, "pos_frac": 0.734375, "sample": [-1.7587413787841797, 42.97334289550781, 28.942657470703125, 68.81645202636719, 110.69215393066406, 5.189552307128906, 64.89820861816406, 79.01087951660156, 29.960172653198242, 50.925819396972656, -0.1386871337890625, 9.248064041137695, 200.6760711669922, 44.532718658447266, -46.72800064086914, 113.90003204345703, 1.055419921875, 69.42417907714844, 46.97736358642578, -20.777923583984375, 7.171176910400391, -12.359046936035156, 84.09110260009766, 57.40790557861328, 109.93504333496094, 67.21099090576172, -25.263874053955078, -4.674674987792969, 97.91754150390625, 84.8062515258789, 90.69953918457031, -14.568763732910156, 159.06788635253906, 1.108001708984375, -9.218433380126953, 22.196361541748047, -31.510940551757812, -19.17621612548828, -61.696502685546875, -24.20403289794922, -75.61521911621094, 18.592609405517578, 2.283519744873047, 198.80227661132812, 50.628990173339844, 28.66588020324707, 34.849937438964844, -67.14606475830078, 48.35054016113281, 162.83958435058594, 151.8315887451172, 135.76161193847656, 138.587890625, 149.62884521484375, 11.8343505859375, -69.9341049194336, 2.618305206298828, 14.205623626708984, 138.80337524414062, 179.09677124023438, 209.75674438476562, -103.31390380859375, 197.8643035888672, 111.2818603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000633.npy"}
|
|
{"epoch": 0.9569160997732427, "step": 634, "batch_size": 64, "mean": 23.69026756286621, "std": 74.61470031738281, "min": -121.8653564453125, "p10": -66.60973815917968, "median": 7.231658935546875, "p90": 136.72037506103516, "max": 179.94309997558594, "pos_frac": 0.578125, "sample": [7.319606781005859, -68.21078491210938, -0.0327301025390625, 157.18038940429688, 35.50823974609375, -10.863866806030273, -11.50124740600586, -10.658857345581055, 19.231346130371094, 8.832992553710938, 23.43681526184082, 6.266685485839844, -2.959197998046875, -17.533653259277344, -17.288467407226562, 35.578407287597656, 78.50822448730469, 64.1226806640625, -46.872459411621094, -41.497413635253906, -34.91169738769531, -0.7846527099609375, 122.83183288574219, -118.21550750732422, 162.61569213867188, 0.6103286743164062, 138.0227813720703, 118.03899383544922, 179.94309997558594, -94.08180236816406, 2.395477294921875, 26.171838760375977, -1.3009757995605469, -26.55459213256836, 150.00595092773438, 2.6987667083740234, -53.657127380371094, -100.18760681152344, 166.2211456298828, 131.81808471679688, 92.72517395019531, -73.59994506835938, -76.06072998046875, 46.2805061340332, -62.87396240234375, -121.8653564453125, 7.143711090087891, 8.902236938476562, 156.311767578125, 49.80832290649414, -36.50984191894531, -28.265594482421875, 69.3696517944336, 21.58575439453125, 12.127595901489258, -1.4865608215332031, 129.59669494628906, 104.24627685546875, 59.70903778076172, 76.9816665649414, -26.334571838378906, 7.903387069702148, -13.446220397949219, 133.68142700195312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000634.npy"}
|
|
{"epoch": 0.9584278155706727, "step": 635, "batch_size": 64, "mean": 39.63690185546875, "std": 82.21954345703125, "min": -134.84344482421875, "p10": -71.11473541259765, "median": 18.957908630371094, "p90": 151.1818313598633, "max": 200.7286834716797, "pos_frac": 0.671875, "sample": [-46.553680419921875, 124.93888854980469, -91.45067596435547, 1.3102912902832031, 59.796356201171875, 7.231742858886719, -73.8768310546875, -101.81196594238281, -80.39089965820312, 157.36538696289062, 2.699848175048828, 123.73931121826172, -13.030746459960938, -13.867240905761719, 118.57914733886719, 167.40377807617188, 111.91641235351562, 13.171363830566406, 154.4981689453125, 128.03353881835938, -53.14784240722656, 21.596221923828125, -28.93378448486328, 183.87100219726562, 15.705169677734375, -64.66984558105469, 108.86970520019531, -1.7513713836669922, -23.682842254638672, 67.87245178222656, 25.13873291015625, -47.829734802246094, 143.44371032714844, 16.296363830566406, 27.636573791503906, 14.33377456665039, 138.0215301513672, 8.953861236572266, 123.28408813476562, -118.19784545898438, 86.35298156738281, 186.22793579101562, -17.619693756103516, 172.58766174316406, -7.0463104248046875, -31.915542602539062, 66.18325805664062, 78.87438201904297, 98.05897521972656, 16.319595336914062, 81.15694427490234, 200.7286834716797, -0.88311767578125, 44.518638610839844, -134.84344482421875, 78.36870574951172, 131.0242156982422, 13.431251525878906, 9.749969482421875, 97.8712158203125, -87.08523559570312, 88.94979858398438, 61.10710906982422, -1.8685989379882812], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000635.npy"}
|
|
{"epoch": 0.9599395313681028, "step": 636, "batch_size": 64, "mean": 56.66056823730469, "std": 72.48065948486328, "min": -129.53826904296875, "p10": -12.222594261169434, "median": 44.96332931518555, "p90": 155.17146301269537, "max": 230.104248046875, "pos_frac": 0.828125, "sample": [136.1088409423828, 5.265380859375, 27.510404586791992, 38.26872253417969, 94.2534408569336, 16.16136932373047, 67.65130615234375, 12.890708923339844, -13.586341857910156, 139.8506317138672, 87.03607177734375, -11.88926887512207, 105.89120483398438, 96.60946655273438, 43.079498291015625, 21.648635864257812, -21.235214233398438, 113.48722839355469, 38.28136444091797, 114.56949615478516, 96.49639129638672, 27.191612243652344, 1.921133041381836, 124.47598266601562, 10.197612762451172, 79.42253112792969, 202.13650512695312, 3.0229244232177734, -8.659162521362305, 71.34528350830078, 36.203433990478516, 37.82551574707031, 184.4622039794922, 58.42570495605469, -11.852180480957031, 126.4532241821289, 50.83726501464844, 85.59783172607422, -10.805274963378906, 161.73753356933594, 82.12896728515625, 166.86793518066406, -12.365447998046875, 230.104248046875, 3.7329063415527344, -129.53826904296875, 182.54774475097656, 26.39215087890625, 23.754486083984375, 138.49444580078125, 61.3868408203125, 44.669212341308594, 106.48408508300781, 177.02981567382812, 69.98298645019531, -104.02003479003906, -121.50186920166016, 15.136423110961914, 45.2574462890625, 104.90592193603516, 38.542381286621094, -47.881996154785156, 56.318634033203125, 29.558441162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000636.npy"}
|
|
{"epoch": 0.9614512471655329, "step": 637, "batch_size": 64, "mean": 42.193756103515625, "std": 83.66610717773438, "min": -197.74862670898438, "p10": -50.48247871398925, "median": 30.037370681762695, "p90": 166.48177795410157, "max": 251.36996459960938, "pos_frac": 0.71875, "sample": [17.157623291015625, 146.25054931640625, 33.098976135253906, 251.36996459960938, -16.78160858154297, 22.34490966796875, -38.506736755371094, 104.60372161865234, 179.4984130859375, 29.595447540283203, -25.94391632080078, 11.867591857910156, 80.41378784179688, -91.2265396118164, 89.6234359741211, 35.8045539855957, -8.975898742675781, 82.866455078125, 138.49197387695312, -8.60862922668457, 59.98847961425781, 53.48222351074219, 7.4398193359375, -32.453590393066406, 3.0143871307373047, 42.2781982421875, -61.331695556640625, 156.24378967285156, 184.03895568847656, 167.03179931640625, 167.2971649169922, 34.858314514160156, -87.12837219238281, 114.19180297851562, -33.34283447265625, 114.95068359375, -74.12265014648438, 17.092758178710938, 8.023887634277344, 48.77104949951172, 46.99220275878906, 170.3568115234375, -4.417058944702148, 26.154621124267578, 139.3240966796875, 19.251304626464844, -44.70027160644531, 47.551025390625, -52.960567474365234, 2.1585006713867188, 16.693603515625, 192.48374938964844, 50.62908935546875, 47.939674377441406, 30.479293823242188, -71.72541046142578, 162.12875366210938, 21.920557022094727, -18.2447509765625, -197.74862670898438, 65.74520111083984, 165.19839477539062, -44.23253631591797, 6.1546630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000637.npy"}
|
|
{"epoch": 0.9629629629629629, "step": 638, "batch_size": 64, "mean": 50.89472579956055, "std": 74.86235809326172, "min": -121.08181762695312, "p10": -28.69413795471191, "median": 30.350522994995117, "p90": 156.02502899169923, "max": 168.75144958496094, "pos_frac": 0.703125, "sample": [-90.26492309570312, 124.03855895996094, 108.61825561523438, -2.8016624450683594, -121.08181762695312, -4.51983642578125, 166.65113830566406, 56.029388427734375, 26.575603485107422, -14.670673370361328, 9.768768310546875, -48.13222122192383, -12.472885131835938, 1.74578857421875, -24.387008666992188, 158.84280395507812, -51.021026611328125, -41.281982421875, 147.78414916992188, 2.71917724609375, 62.42527770996094, 42.66522979736328, 137.70867919921875, 132.00860595703125, -59.447471618652344, 152.1285858154297, 94.49036407470703, 156.12142944335938, 2.9396514892578125, 55.757415771484375, 34.12544250488281, 163.22523498535156, 85.75590515136719, 147.63340759277344, 16.2674560546875, -13.05903434753418, 2.6933212280273438, 111.88764953613281, 6.4481353759765625, 147.05462646484375, 77.4203109741211, 154.14923095703125, 155.8000946044922, -10.183917999267578, 138.8622283935547, -3.526458740234375, -1.3503303527832031, 154.86386108398438, 25.673095703125, 44.43981170654297, -0.7361583709716797, 24.491451263427734, 168.27882385253906, 56.73480224609375, -30.540050506591797, 53.992008209228516, 1.6664161682128906, 168.75144958496094, 156.6005401611328, 5.369365692138672, -7.68212890625, -5.6479034423828125, 15.728607177734375, 43.137794494628906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000638.npy"}
|
|
{"epoch": 0.9644746787603931, "step": 639, "batch_size": 64, "mean": 56.4093017578125, "std": 76.26365661621094, "min": -80.41045379638672, "p10": -29.649546432495118, "median": 38.3980598449707, "p90": 173.66141662597659, "max": 225.20730590820312, "pos_frac": 0.71875, "sample": [106.8030014038086, 86.67842102050781, 56.540191650390625, -7.055885314941406, 75.97148132324219, 91.82835388183594, 11.495208740234375, -29.27231216430664, -41.610191345214844, 95.61622619628906, 27.996963500976562, 44.89455795288086, 217.53358459472656, 39.657691955566406, 118.40428161621094, -1.64306640625, 36.64033508300781, 0.5646743774414062, -80.41045379638672, 56.164024353027344, 82.51276397705078, 134.6863555908203, 2.6897335052490234, 171.06912231445312, -0.2854804992675781, 124.57958984375, 51.644412994384766, 127.54554748535156, -21.250656127929688, 175.16773986816406, 175.43637084960938, -22.101167678833008, 1.3428974151611328, 90.92810821533203, -5.783321380615234, 5.713905334472656, 109.6696548461914, -4.7637786865234375, 225.20730590820312, 133.05270385742188, 100.72460174560547, 33.01995849609375, 201.34713745117188, 57.084083557128906, -29.81121826171875, 13.889179229736328, -23.002525329589844, -15.691020965576172, 37.138427734375, 20.59194564819336, 167.51345825195312, 116.24552917480469, 17.79114532470703, 19.529144287109375, -12.75555419921875, -79.20045471191406, 20.107276916503906, -31.270828247070312, -52.28997039794922, 183.96360778808594, -34.10608673095703, 127.1409912109375, 133.605224609375, 174.77239990234375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000639.npy"}
|
|
{"epoch": 0.9659863945578231, "step": 640, "batch_size": 64, "mean": 50.65288543701172, "std": 75.13240051269531, "min": -163.93646240234375, "p10": -38.6843994140625, "median": 54.51410675048828, "p90": 148.51922302246095, "max": 190.4933319091797, "pos_frac": 0.796875, "sample": [130.81192016601562, 33.878421783447266, 190.4933319091797, -63.59169006347656, 9.875003814697266, 66.95126342773438, 110.09005737304688, 4.766746520996094, 89.81210327148438, 158.4087371826172, 66.31228637695312, 13.046897888183594, 86.04084777832031, 77.20777893066406, -104.12715148925781, 76.80899047851562, 51.23346710205078, 70.15115356445312, -23.343936920166016, 174.87123107910156, -157.93777465820312, 78.22197723388672, 176.5312042236328, 21.197799682617188, -39.57203674316406, -19.04877471923828, 99.61902618408203, 14.325645446777344, -39.51255798339844, -1.1754951477050781, 138.1060028076172, 0.8008270263671875, 26.905609130859375, 143.93673706054688, 71.83418273925781, -1.5984268188476562, 6.364067077636719, 27.945404052734375, -40.38963317871094, 67.18920135498047, 21.557769775390625, 145.4767608642578, 4.262302398681641, 55.6116943359375, 139.4086151123047, 97.7042007446289, 81.88263702392578, -163.93646240234375, 38.96717071533203, 131.06906127929688, 95.19569396972656, 164.828125, 55.12593078613281, 157.82403564453125, 1.3829498291015625, -16.650480270385742, 149.82313537597656, -36.75202941894531, 53.90228271484375, 83.58790588378906, 24.305347442626953, 39.28009033203125, 43.646934509277344, 80.84069061279297], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000640.npy"}
|
|
{"epoch": 0.9674981103552532, "step": 641, "batch_size": 64, "mean": 47.86433410644531, "std": 62.50070571899414, "min": -160.49549865722656, "p10": -8.017171478271484, "median": 40.10387420654297, "p90": 126.89223022460942, "max": 235.07546997070312, "pos_frac": 0.828125, "sample": [-32.49412155151367, -51.82726287841797, 89.61041259765625, 66.24751281738281, 68.27291870117188, 2.0356369018554688, 91.32719421386719, 12.891944885253906, -1.8600387573242188, 23.788894653320312, 61.05712890625, 20.885530471801758, 83.79656982421875, 153.0703582763672, 44.809234619140625, 10.49213981628418, -7.52935791015625, 150.7298583984375, 164.70449829101562, 66.85505676269531, 94.28641510009766, -1.2095565795898438, 31.279258728027344, 31.808792114257812, 68.67461395263672, 143.27012634277344, 16.389602661132812, 52.753509521484375, 25.963516235351562, 87.21408081054688, 17.070587158203125, 79.27371215820312, -8.226234436035156, 71.44630432128906, 8.7757568359375, -57.78881072998047, 88.02140808105469, 31.103378295898438, -36.315025329589844, 27.568634033203125, -160.49549865722656, 108.70181274414062, 58.34996032714844, 15.508346557617188, 61.95264434814453, 178.65614318847656, 5.5284881591796875, -1.8078975677490234, 109.7183837890625, 51.25782775878906, 235.07546997070312, 75.02101135253906, 131.75205993652344, 115.55262756347656, -23.35345458984375, 35.39851379394531, 65.27157592773438, 21.877975463867188, 24.299930572509766, 5.998565673828125, 5.263877868652344, 12.2640380859375, 45.962486267089844, 97.33820343017578], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000641.npy"}
|
|
{"epoch": 0.9690098261526833, "step": 642, "batch_size": 64, "mean": 38.03020477294922, "std": 76.95255279541016, "min": -144.68141174316406, "p10": -57.348080444335935, "median": 29.786803245544434, "p90": 134.50374908447267, "max": 223.587158203125, "pos_frac": 0.6875, "sample": [74.56946563720703, 41.049278259277344, 3.5318756103515625, -75.46583557128906, 24.469497680664062, -11.658025741577148, 70.65184020996094, 160.6956329345703, -27.90234375, -144.68141174316406, 223.587158203125, 26.74439239501953, 62.95741271972656, -55.10430145263672, 72.31430053710938, 14.102676391601562, 109.12995910644531, 184.0069580078125, -19.830121994018555, -21.494770050048828, 80.02088165283203, 82.96406555175781, 113.02367401123047, -27.44063949584961, 96.36885070800781, 103.1988525390625, 161.3994598388672, 39.449249267578125, 129.76547241210938, 62.32036590576172, 64.96271514892578, 83.15298461914062, 68.70407104492188, 24.082828521728516, -10.854530334472656, 132.78610229492188, -69.14421081542969, 36.212913513183594, 18.515220642089844, -48.42637634277344, -7.461132049560547, 23.364974975585938, 1.611551284790039, -26.322349548339844, 93.37115478515625, 29.394182205200195, -34.97758483886719, 105.66283416748047, 135.23988342285156, 166.74618530273438, 15.414291381835938, 11.159065246582031, 131.10125732421875, 87.41614532470703, 160.498291015625, 30.179424285888672, -58.30970001220703, 38.019630432128906, -100.27462005615234, -114.58785247802734, -8.740997314453125, -29.970596313476562, -83.86357116699219, 16.527175903320312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000642.npy"}
|
|
{"epoch": 0.9705215419501134, "step": 643, "batch_size": 64, "mean": 50.763084411621094, "std": 77.63394927978516, "min": -132.43939208984375, "p10": -48.51781883239746, "median": 45.79005813598633, "p90": 162.3182861328125, "max": 179.8270263671875, "pos_frac": 0.75, "sample": [38.27284240722656, 53.61557388305664, 43.022274017333984, 12.58135986328125, 3.1442785263061523, 179.80856323242188, 6.299798965454102, 129.48941040039062, 77.39110565185547, 19.680625915527344, 94.41485595703125, 24.7918758392334, 80.176025390625, 1.5225982666015625, 123.62186431884766, 107.85183715820312, -24.178665161132812, 9.512308120727539, 162.12969970703125, 175.0076904296875, 93.00003814697266, 30.218826293945312, 144.22409057617188, 44.91485595703125, -37.20763397216797, 161.7740020751953, 54.96427917480469, -39.810791015625, 166.49282836914062, 167.14114379882812, 50.58441925048828, 106.72080993652344, -21.988372802734375, 98.71269226074219, 23.353971481323242, -47.691036224365234, 26.539491653442383, -0.290740966796875, 176.5699462890625, 150.01779174804688, 125.96940612792969, -48.872154235839844, 105.69691467285156, -10.997909545898438, 103.92432403564453, -76.49140930175781, -91.88614654541016, -132.43939208984375, 4.316947937011719, 39.05422592163086, -75.47307586669922, 92.71392059326172, 117.61175537109375, 50.40008544921875, 21.54449462890625, -71.364990234375, -6.9895172119140625, 162.39910888671875, 58.57904815673828, 46.665260314941406, -30.25369644165039, 179.8270263671875, 84.51819610595703, -66.01168823242188], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000643.npy"}
|
|
{"epoch": 0.9720332577475435, "step": 644, "batch_size": 64, "mean": 41.363677978515625, "std": 80.56716918945312, "min": -169.40049743652344, "p10": -39.45143165588378, "median": 24.21911907196045, "p90": 153.52691192626955, "max": 175.86871337890625, "pos_frac": 0.75, "sample": [116.87421417236328, -46.54119873046875, -160.1504669189453, 137.493896484375, -32.35221481323242, -41.8872184753418, 12.17178726196289, 13.470701217651367, 134.31011962890625, 170.85845947265625, 72.27238464355469, 10.009037017822266, 146.56214904785156, 128.14694213867188, 51.70558166503906, 23.874176025390625, 56.82795333862305, 35.92473602294922, 0.5165634155273438, 159.34605407714844, 13.492511749267578, 133.9556427001953, -117.22248840332031, -12.503768920898438, -14.766525268554688, 93.48638916015625, 0.12711334228515625, 59.73927688598633, 0.2568511962890625, -33.76792907714844, 162.6721954345703, -15.91217041015625, 22.458450317382812, 95.26341247558594, 124.70915985107422, -4.991641998291016, -26.474884033203125, 175.86871337890625, 140.4967041015625, 17.441368103027344, 59.79914855957031, 161.04269409179688, 76.43482971191406, 24.564062118530273, 13.075302124023438, -169.40049743652344, -11.131072998046875, 118.62599182128906, 20.41527557373047, 140.31594848632812, -32.35722351074219, 8.946380615234375, 2.349802017211914, -66.65572357177734, 34.469581604003906, 25.23448944091797, -111.94610595703125, 75.38923645019531, 88.60426330566406, 3.4967041015625, 11.647186279296875, 156.51181030273438, 42.15692901611328, 171.92446899414062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000644.npy"}
|
|
{"epoch": 0.9735449735449735, "step": 645, "batch_size": 64, "mean": 46.735137939453125, "std": 68.38373565673828, "min": -150.04934692382812, "p10": -28.75754585266113, "median": 43.53399658203125, "p90": 130.6294891357422, "max": 214.08778381347656, "pos_frac": 0.75, "sample": [28.95126724243164, 157.95578002929688, 97.1710433959961, -62.415367126464844, 78.58439636230469, 196.68624877929688, 113.38814544677734, 10.534650802612305, -9.364250183105469, 7.583564758300781, -2.4355201721191406, -50.71482849121094, 34.80200958251953, -2.6511802673339844, 121.50164794921875, 23.01507568359375, 90.23896789550781, 2.391021728515625, -3.56231689453125, 0.6174144744873047, 160.82049560546875, -35.001869201660156, 40.96222686767578, 66.91075897216797, 72.1794204711914, -11.579345703125, 30.211753845214844, 53.44146728515625, 155.10794067382812, 88.6095962524414, 46.10576629638672, 2.2570667266845703, 56.49126434326172, 7.0938873291015625, 10.868606567382812, 83.6826171875, -77.45417785644531, -7.396308898925781, -51.821136474609375, 129.5928192138672, 91.27093505859375, 82.1969223022461, 214.08778381347656, 0.7405776977539062, 20.507266998291016, 65.87017059326172, 98.7164077758789, 52.309165954589844, 55.40032958984375, 92.94076538085938, 4.3338470458984375, 120.34725952148438, 56.89145278930664, 122.33776092529297, 79.8118896484375, -1.6977615356445312, 131.0737762451172, -26.895626068115234, -29.555511474609375, -150.04934692382812, 10.676567077636719, 150.1295623779297, 97.64859771728516, -1.4045486450195312], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000645.npy"}
|
|
{"epoch": 0.9750566893424036, "step": 646, "batch_size": 64, "mean": 44.6483154296875, "std": 78.50543975830078, "min": -154.49891662597656, "p10": -29.866193389892576, "median": 31.638078689575195, "p90": 150.2661911010742, "max": 230.11911010742188, "pos_frac": 0.734375, "sample": [-26.6695556640625, 2.1787853240966797, 3.5037918090820312, 109.727783203125, 156.59515380859375, 129.3239288330078, -87.64619445800781, 143.53854370117188, 77.12960815429688, 153.50460815429688, -7.9326324462890625, 21.15752410888672, 123.92789459228516, 147.54920959472656, 97.65339660644531, -4.3401031494140625, 68.77095794677734, 58.417694091796875, 159.99981689453125, 104.60627746582031, 163.60617065429688, 68.96841430664062, -132.54493713378906, -26.904090881347656, -71.15815734863281, 87.15990447998047, 3.9799842834472656, -12.249862670898438, 24.806350708007812, 20.109512329101562, 40.517364501953125, 6.927154541015625, 43.986610412597656, 150.3045196533203, 76.8919677734375, 6.562049865722656, 22.18082046508789, -14.902976989746094, 46.91925811767578, 90.03073120117188, 0.31206321716308594, -24.639572143554688, 66.86151123046875, -41.39192199707031, 117.14434814453125, 230.11911010742188, 150.1767578125, 105.51854705810547, 17.71263885498047, 199.98370361328125, 46.74122619628906, 8.135162353515625, 17.344879150390625, 3.4113311767578125, -154.49891662597656, 9.032012939453125, 70.02243041992188, -106.70573425292969, -31.135665893554688, 135.97140502929688, -9.16250991821289, -12.384506225585938, -5.732929229736328, 38.46980667114258], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000646.npy"}
|
|
{"epoch": 0.9765684051398337, "step": 647, "batch_size": 64, "mean": 67.81459045410156, "std": 74.73070526123047, "min": -108.99119567871094, "p10": -18.134431266784663, "median": 51.94602394104004, "p90": 170.82075653076177, "max": 244.6019287109375, "pos_frac": 0.796875, "sample": [-0.9881401062011719, 69.6016616821289, 148.024169921875, 15.081802368164062, 43.18562316894531, 105.70651245117188, 9.220947265625, 175.00120544433594, 74.9234619140625, 197.36907958984375, 42.36302947998047, 53.707122802734375, 93.13467407226562, -4.184749603271484, -25.916183471679688, 44.769775390625, 244.6019287109375, -20.7720890045166, -89.95220947265625, 104.09246826171875, 141.86427307128906, 46.36206817626953, 104.98346710205078, 58.108577728271484, 175.10931396484375, 123.01586151123047, 132.28652954101562, -5.840208053588867, 157.41844177246094, 25.20770263671875, 117.26628875732422, -4.96673583984375, 126.767333984375, 14.178573608398438, 120.798583984375, -31.320404052734375, 37.04648971557617, 161.06637573242188, 34.333778381347656, -45.39002227783203, 89.0165023803711, 42.95130157470703, 50.1849250793457, 115.90699005126953, -108.99119567871094, 24.78578758239746, 74.55475616455078, 37.714298248291016, 32.29847717285156, 35.886207580566406, -11.69598388671875, 191.13800048828125, 117.70091247558594, 13.63970947265625, -51.4208984375, 130.79669189453125, 22.59199333190918, 213.80023193359375, -11.979896545410156, 137.56271362304688, 180.27635192871094, 34.109474182128906, 96.17029571533203, 115.86953735351562], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000647.npy"}
|
|
{"epoch": 0.9780801209372638, "step": 648, "batch_size": 64, "mean": 50.06586456298828, "std": 80.55511474609375, "min": -159.7940216064453, "p10": -37.146985626220705, "median": 56.51596641540527, "p90": 143.99099731445315, "max": 237.4211883544922, "pos_frac": 0.71875, "sample": [121.0179672241211, 33.90093231201172, 44.74689483642578, -159.7940216064453, 130.7028045654297, -37.57719421386719, -156.5423583984375, 14.525054931640625, 147.2491455078125, 53.62175750732422, -46.49296951293945, 63.68122100830078, 35.25439453125, 3.1654739379882812, -30.718765258789062, -87.47657012939453, 59.41017532348633, -9.597160339355469, 237.4211883544922, 121.83258056640625, 164.4491424560547, 92.12718963623047, 101.43911743164062, 118.96576690673828, 119.896484375, -6.7727508544921875, 70.10467529296875, 166.82049560546875, 65.01030731201172, 92.32626342773438, 2.0815277099609375, -2.139881134033203, 196.0147705078125, 206.84152221679688, 2.8147926330566406, -53.00891876220703, 76.40797424316406, 63.92597198486328, 86.63603973388672, 25.926795959472656, 98.774169921875, 117.21214294433594, 146.93084716796875, -16.86899185180664, -36.143165588378906, 14.095207214355469, 46.68003845214844, 73.35872650146484, 76.80543518066406, 137.13134765625, 78.67617797851562, 124.50608825683594, 94.32330322265625, -118.96488952636719, -11.743438720703125, 47.79887390136719, 101.11935424804688, -29.96546173095703, 32.05414962768555, -35.16251754760742, -0.5147209167480469, -11.831125259399414, 51.9732666015625, 95.77288818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000648.npy"}
|
|
{"epoch": 0.9795918367346939, "step": 649, "batch_size": 64, "mean": 49.79816436767578, "std": 80.06608581542969, "min": -142.56272888183594, "p10": -58.18824081420898, "median": 51.35458564758301, "p90": 152.75591125488282, "max": 219.7524871826172, "pos_frac": 0.75, "sample": [132.38955688476562, 93.48835754394531, -33.74520492553711, 99.88703918457031, 147.19985961914062, 133.5634307861328, 38.19343566894531, 27.39185905456543, -4.634651184082031, 149.97280883789062, 2.575826644897461, 219.7524871826172, 96.55116271972656, -21.006641387939453, 85.52008056640625, 91.91654968261719, 89.65660095214844, 78.57017517089844, 79.67276763916016, 125.49772644042969, 162.20254516601562, -6.840600967407227, -106.00556182861328, -38.833152770996094, -142.56272888183594, 99.42344665527344, 11.372459411621094, 29.216594696044922, -58.315940856933594, -81.818359375, 54.00001907348633, 24.329078674316406, 133.39535522460938, 0.6135463714599609, 78.53590393066406, 101.16885375976562, 48.70915222167969, 98.965087890625, 112.4115982055664, 114.09906768798828, 3.2400436401367188, 164.76492309570312, 79.18831634521484, -11.99515151977539, 5.3775634765625, 160.186279296875, 1.8601493835449219, 9.795753479003906, 1.0665740966796875, 30.478073120117188, 2.0328292846679688, 191.74960327148438, -57.89027404785156, -76.8824462890625, 184.3195037841797, -30.476200103759766, -72.23103332519531, -67.12600708007812, 78.17671203613281, 153.94866943359375, 130.00827026367188, 57.65416717529297, 1.6699943542480469, -18.313232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000649.npy"}
|
|
{"epoch": 0.981103552532124, "step": 650, "batch_size": 64, "mean": 51.47490692138672, "std": 81.4701919555664, "min": -134.54054260253906, "p10": -41.230438995361325, "median": 36.8114013671875, "p90": 157.28876495361328, "max": 199.5574188232422, "pos_frac": 0.734375, "sample": [-37.28908157348633, 177.3927764892578, 93.39804077148438, 132.69309997558594, 9.516693115234375, 151.6397705078125, 112.82144165039062, 28.384552001953125, 6.669349670410156, 18.692089080810547, 146.5982666015625, 126.15068054199219, -65.00936889648438, 11.318496704101562, 3.6347904205322266, 104.61138153076172, 129.2441864013672, 53.74113464355469, 165.16526794433594, -0.7183818817138672, 110.40086364746094, 67.11026000976562, -44.34291076660156, 7.291393280029297, 160.17845153808594, 158.1588134765625, 167.5262451171875, 128.51754760742188, 80.70966339111328, -81.95259857177734, 4.070941925048828, -134.54054260253906, 12.064510345458984, 112.25315856933594, -129.92849731445312, 21.232955932617188, 155.25865173339844, 80.9085693359375, 11.046669006347656, -36.20478057861328, -11.60931396484375, 72.03221130371094, -93.53955078125, 145.56851196289062, -26.009140014648438, -11.289466857910156, 199.5574188232422, 151.43423461914062, 45.238250732421875, -42.252952575683594, 91.35855102539062, 5.061485290527344, 138.8184051513672, -1.0606269836425781, 26.545129776000977, 26.082429885864258, -18.962142944335938, 49.71357727050781, 137.48886108398438, 183.29888916015625, -38.844573974609375, -7.796390533447266, 49.92238998413086, 5.223197937011719], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000650.npy"}
|
|
{"epoch": 0.982615268329554, "step": 651, "batch_size": 64, "mean": 45.396671295166016, "std": 73.53605651855469, "min": -124.34098052978516, "p10": -35.719566345214844, "median": 25.486919403076172, "p90": 145.08242645263672, "max": 216.20272827148438, "pos_frac": 0.75, "sample": [-0.13925933837890625, 17.716995239257812, 64.52449035644531, -22.035400390625, 41.51616668701172, -62.087867736816406, -37.03582000732422, 63.144203186035156, 9.662178039550781, 142.8594970703125, 14.958778381347656, -4.28582763671875, -7.0104217529296875, 4.4878082275390625, 86.74478149414062, -124.34098052978516, 0.7130355834960938, 196.19827270507812, 56.80236053466797, 146.0351104736328, 140.72372436523438, -3.1617813110351562, -45.06544494628906, 72.49069213867188, 10.628486633300781, 9.236248016357422, 46.72369384765625, 100.60379028320312, -23.330123901367188, 140.96197509765625, 35.69038391113281, 166.02017211914062, 8.704601287841797, 1.9482955932617188, 42.60581970214844, 0.2248401641845703, 129.1859893798828, 38.93794250488281, -39.23418426513672, 41.15793228149414, 164.70281982421875, 27.837265014648438, 94.74885559082031, 8.138538360595703, -11.438434600830078, 169.0753173828125, 179.4044952392578, 127.57332611083984, -32.64830780029297, 142.36767578125, 18.219472885131836, 5.6082763671875, 4.1332244873046875, 107.48693084716797, 83.53202819824219, 86.8567886352539, 134.00559997558594, 10.30843734741211, 23.136573791503906, -86.56362915039062, 57.636741638183594, -67.28951263427734, 216.20272827148438, -21.129371643066406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000651.npy"}
|
|
{"epoch": 0.9841269841269841, "step": 652, "batch_size": 64, "mean": 43.015357971191406, "std": 79.63150787353516, "min": -126.02532196044922, "p10": -42.503238677978516, "median": 28.533966064453125, "p90": 152.64413452148438, "max": 235.32330322265625, "pos_frac": 0.671875, "sample": [4.240604400634766, 135.8908233642578, 153.9495849609375, 41.44940948486328, -41.29015350341797, 161.36029052734375, 22.6077938079834, 44.01095199584961, 156.279296875, 7.625396728515625, 138.48345947265625, -89.7254409790039, 0.2763099670410156, 36.51385498046875, 82.55659484863281, 127.93464660644531, 3.9766693115234375, 114.0134506225586, 55.62584686279297, -2.3272247314453125, -55.40888977050781, 123.23675537109375, 95.14302062988281, -1.2334213256835938, 12.236316680908203, -22.34624481201172, -122.07696533203125, -26.80584716796875, 178.5376739501953, 65.32561492919922, 25.029098510742188, -26.918670654296875, 64.19163513183594, 6.452415466308594, 121.96662139892578, -40.765052795410156, 125.0291519165039, 85.25508117675781, -11.816818237304688, -86.30641174316406, -26.536399841308594, 165.12014770507812, 1.4460906982421875, 97.4540786743164, -27.762069702148438, -15.910022735595703, 149.59808349609375, 111.55300903320312, 0.7825565338134766, 147.50335693359375, 235.32330322265625, 60.864356994628906, 85.73614501953125, -8.874061584472656, -13.99649429321289, 32.03883361816406, -62.12568664550781, 13.08149528503418, -43.02313232421875, 160.2851104736328, 45.53248596191406, -6.086578369140625, -126.02532196044922, 114.82635498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000652.npy"}
|
|
{"epoch": 0.9856386999244142, "step": 653, "batch_size": 64, "mean": 52.90898895263672, "std": 78.41134643554688, "min": -130.12203979492188, "p10": -31.60022048950195, "median": 50.50858116149902, "p90": 161.6865478515625, "max": 211.6064453125, "pos_frac": 0.78125, "sample": [20.90800666809082, 139.45384216308594, 7.2490692138671875, -2.636138916015625, 55.64121627807617, 23.588973999023438, 61.70585632324219, -32.301795959472656, 171.80723571777344, 6.6909332275390625, -16.214580535888672, 2.9020538330078125, 102.31228637695312, -29.963211059570312, 211.6064453125, 116.44593048095703, 74.13943481445312, 165.7310028076172, 13.278087615966797, 162.21804809570312, 97.54029083251953, 180.10536193847656, 75.12288665771484, -1.1338424682617188, 49.22564697265625, 14.800329208374023, 154.41830444335938, 22.315643310546875, -3.129283905029297, -11.261123657226562, 189.91937255859375, -59.75929260253906, -47.48200225830078, 23.212871551513672, 27.944246292114258, -64.1032485961914, 51.7915153503418, 65.20153045654297, 91.10086059570312, 85.56178283691406, 12.211174011230469, -128.0694122314453, 184.92071533203125, 65.7977294921875, 110.64484405517578, -130.12203979492188, 145.0896759033203, -6.0250091552734375, 123.63816833496094, 55.205718994140625, 5.661262512207031, 160.44638061523438, -121.31594848632812, 6.5280303955078125, 2.662443161010742, 25.663599014282227, 136.59127807617188, 139.04534912109375, 11.329448699951172, 105.78384399414062, 55.19603729248047, 98.1422119140625, 14.843666076660156, 116.35160064697266], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000653.npy"}
|
|
{"epoch": 0.9871504157218443, "step": 654, "batch_size": 64, "mean": 43.44438934326172, "std": 75.37660217285156, "min": -104.0870132446289, "p10": -41.93445053100585, "median": 30.804466247558594, "p90": 156.91722869873047, "max": 259.2958984375, "pos_frac": 0.734375, "sample": [9.488822937011719, 173.76722717285156, -104.0870132446289, 82.71368408203125, 41.47911071777344, -63.857177734375, 152.29115295410156, 55.26140594482422, 5.595829010009766, -33.64308166503906, -2.8721923828125, -44.996246337890625, 70.27031707763672, 0.34272003173828125, -11.658599853515625, 20.174726486206055, 143.6650390625, -66.64692687988281, 159.80227661132812, 9.082382202148438, 16.240699768066406, 12.106456756591797, 139.6676788330078, 179.34469604492188, -0.7179546356201172, -7.623390197753906, 6.099037170410156, -11.394111633300781, -34.790260314941406, 43.679569244384766, -2.1447372436523438, 102.11807250976562, 259.2958984375, 76.54466247558594, 57.87567901611328, 145.3946533203125, 119.32066345214844, -16.281097412109375, 21.837814331054688, 43.996002197265625, 102.12820434570312, 27.340347290039062, 65.72987365722656, -76.41813659667969, 16.283416748046875, 8.265853881835938, 172.14064025878906, 158.1908416748047, 34.75984191894531, 60.094383239746094, 55.38941955566406, 159.27471923828125, 4.2695770263671875, 45.080360412597656, 153.94546508789062, -64.93966674804688, 1.7584991455078125, 51.07406997680664, 30.664321899414062, 90.57920837402344, 36.41753005981445, 30.944610595703125, -27.676498413085938, -101.59927368164062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000654.npy"}
|
|
{"epoch": 0.9886621315192744, "step": 655, "batch_size": 64, "mean": 64.08234405517578, "std": 64.6014175415039, "min": -65.3758773803711, "p10": -9.668852996826171, "median": 58.96406555175781, "p90": 147.63128967285158, "max": 199.29312133789062, "pos_frac": 0.8125, "sample": [117.90361022949219, 79.742919921875, -4.5365753173828125, 19.844085693359375, 39.42066955566406, 94.08905029296875, 183.2929229736328, 98.14408874511719, 87.27303314208984, 12.219573974609375, 0.612579345703125, 106.17713165283203, 1.9601287841796875, -0.12943267822265625, -6.928016662597656, -15.114456176757812, 49.5286865234375, 165.6142578125, 136.34161376953125, 20.026172637939453, 163.79115295410156, -9.052993774414062, 110.58777618408203, 157.2876739501953, 42.20273208618164, -9.932792663574219, 66.6601791381836, -59.099483489990234, 173.17288208007812, -33.55326843261719, 139.1855926513672, 83.80532836914062, 140.94332885742188, 29.702682495117188, 50.82025146484375, 12.414813995361328, 49.35222625732422, 131.53518676757812, 81.781494140625, 199.29312133789062, 10.7911376953125, 5.1297607421875, -10.917022705078125, 136.03884887695312, 108.81109619140625, -59.644065856933594, 44.17973327636719, 111.24961853027344, 150.49755859375, 110.24934387207031, -4.951549530029297, 22.453079223632812, 123.15426635742188, 45.91143798828125, 30.641952514648438, 103.62542724609375, 96.49990844726562, 51.26795196533203, 118.40179443359375, 2.570831298828125, -65.3758773803711, 74.95719909667969, 97.40655517578125, 91.94107055664062], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000655.npy"}
|
|
{"epoch": 0.9901738473167044, "step": 656, "batch_size": 64, "mean": 53.9482536315918, "std": 72.55986785888672, "min": -116.80992126464844, "p10": -15.338326263427733, "median": 38.67782402038574, "p90": 165.63236236572266, "max": 183.16363525390625, "pos_frac": 0.796875, "sample": [141.06570434570312, 41.10710906982422, 142.3504638671875, 114.12859344482422, 12.225513458251953, 10.919265747070312, 11.41659164428711, -1.0861282348632812, 35.2899169921875, -11.405742645263672, 12.09658432006836, 183.16363525390625, -3.7638168334960938, 16.456573486328125, 2.728626251220703, 97.34669494628906, -116.80992126464844, -16.063316345214844, 85.16004180908203, 124.15481567382812, 177.91629028320312, 13.836769104003906, -109.97882080078125, 36.248538970947266, 166.4081573486328, -24.78066635131836, 52.328147888183594, 130.9369659423828, -13.646682739257812, 76.47718811035156, 137.67880249023438, 11.531593322753906, 23.6129207611084, 56.555572509765625, 63.66332244873047, 100.91416931152344, 46.334625244140625, 44.73347473144531, 171.12216186523438, 54.072509765625, 26.531814575195312, -72.09280395507812, 50.64513397216797, 179.2375030517578, 13.018646240234375, 150.09478759765625, 52.5418701171875, 81.61154174804688, 29.761287689208984, 106.15440368652344, 165.54013061523438, -19.067073822021484, -2.75958251953125, 181.3422088623047, 109.2708511352539, 11.353897094726562, 12.2239990234375, -81.57063293457031, -10.568321228027344, 10.186174392700195, 22.086074829101562, 33.87085723876953, 141.15733337402344, 165.67189025878906], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000656.npy"}
|
|
{"epoch": 0.9916855631141346, "step": 657, "batch_size": 64, "mean": 61.816036224365234, "std": 80.42914581298828, "min": -145.43251037597656, "p10": -8.766362571716307, "median": 74.98483276367188, "p90": 166.64861450195312, "max": 192.73641967773438, "pos_frac": 0.796875, "sample": [184.10130310058594, 77.29644775390625, -2.9169654846191406, 191.89508056640625, 82.01338195800781, 12.45301628112793, 192.73641967773438, 166.29049682617188, 49.63947296142578, -145.43251037597656, 9.728824615478516, -6.55430793762207, 107.81390380859375, 35.10939407348633, 179.76959228515625, 74.46220397949219, 90.8169937133789, 158.20266723632812, -85.8455810546875, 156.47068786621094, -9.714385986328125, 93.30288696289062, -106.35832977294922, 21.60251808166504, 81.62908935546875, 145.96670532226562, 40.8712158203125, 0.30023193359375, -2.068330764770508, 55.879310607910156, 83.84857177734375, 140.3059844970703, 8.562789916992188, 116.48738861083984, 80.68169403076172, 42.47090148925781, 104.35567474365234, 119.20024108886719, 164.7191162109375, 3.777231216430664, 180.80307006835938, 14.251235961914062, -4.1743316650390625, 163.07958984375, 1.1236495971679688, 57.711181640625, 182.331787109375, 3.5122509002685547, -51.025611877441406, 75.50746154785156, 11.329414367675781, -139.6873779296875, 131.09841918945312, -1.7451152801513672, 8.646726608276367, 75.92888641357422, -29.616249084472656, 86.77359771728516, 86.431396484375, 166.80209350585938, 2.8075904846191406, 100.21377563476562, -4.1779327392578125, 124.42969512939453], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000657.npy"}
|
|
{"epoch": 0.9931972789115646, "step": 658, "batch_size": 64, "mean": 47.44400405883789, "std": 89.08833312988281, "min": -154.52517700195312, "p10": -74.5908950805664, "median": 40.60212707519531, "p90": 159.13210906982425, "max": 273.80633544921875, "pos_frac": 0.71875, "sample": [60.951900482177734, 32.668399810791016, 147.71730041503906, -132.14547729492188, 127.74842834472656, 7.003122329711914, -134.8459014892578, 33.10935974121094, 144.2586669921875, 133.3533935546875, -4.7216949462890625, 71.31008911132812, -79.65484619140625, 77.76576232910156, 138.00633239746094, 50.63865280151367, -13.691253662109375, 115.33558654785156, 273.80633544921875, 7.259063720703125, -16.59653663635254, 22.73187255859375, 37.48438262939453, 98.0838394165039, 132.60125732421875, 162.1652374267578, -99.49658203125, 97.11846923828125, 87.56100463867188, 30.335433959960938, 5.34937858581543, -1.8328628540039062, 61.108421325683594, 102.34849548339844, -0.2846336364746094, 43.719871520996094, 181.94451904296875, -127.77041625976562, 48.535987854003906, 51.14715576171875, 161.71853637695312, 3.6639480590820312, 71.52311706542969, 146.73046875, -3.5897274017333984, 49.163047790527344, -62.77500915527344, 161.63075256347656, 28.55919647216797, 60.01565170288086, 15.892929077148438, -1.123819351196289, -29.96489715576172, -96.90318298339844, 30.183860778808594, 125.20792388916016, -14.945056915283203, -154.52517700195312, 209.3682861328125, -23.74474334716797, 205.78265380859375, 26.4686279296875, 153.30194091796875, 2.649564743041992], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000658.npy"}
|
|
{"epoch": 0.9947089947089947, "step": 659, "batch_size": 64, "mean": 42.28049850463867, "std": 76.93946075439453, "min": -136.19232177734375, "p10": -58.87153854370117, "median": 34.763973236083984, "p90": 146.1025131225586, "max": 187.29251098632812, "pos_frac": 0.734375, "sample": [38.24334716796875, 91.7823486328125, 153.281982421875, 114.2007064819336, -11.2327880859375, 185.88592529296875, 78.4532699584961, 87.82548522949219, 28.681520462036133, 162.33099365234375, 145.07211303710938, -6.8605804443359375, 25.78911590576172, 19.316387176513672, -0.37885284423828125, 12.028297424316406, 105.02304077148438, -100.9052505493164, 13.624212265014648, -67.39949035644531, 138.22531127929688, -42.204307556152344, 37.99882507324219, 40.56744384765625, 8.895864486694336, -3.8529815673828125, 21.606294631958008, 36.311920166015625, -12.726594924926758, 43.51158142089844, 27.109848022460938, 38.79637908935547, 119.76631164550781, 95.56425476074219, 59.760841369628906, 10.905006408691406, 164.72030639648438, 88.92506408691406, 162.82278442382812, 22.561504364013672, 83.56405639648438, -72.24774169921875, -77.43089294433594, 146.5441131591797, -33.38512420654297, 12.260543823242188, 4.35601806640625, 109.08507537841797, 115.9002456665039, -53.061546325683594, 25.885536193847656, 130.91140747070312, -136.19232177734375, -57.226966857910156, 143.57601928710938, 33.216026306152344, 12.186256408691406, -134.73452758789062, 187.29251098632812, 59.660064697265625, -16.398086547851562, 100.34947204589844, 47.3907470703125, -59.57635498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000659.npy"}
|
|
{"epoch": 0.9962207105064248, "step": 660, "batch_size": 64, "mean": 59.14536666870117, "std": 68.8411865234375, "min": -101.87828063964844, "p10": -36.14063720703124, "median": 50.36180305480957, "p90": 150.39470825195315, "max": 244.148193359375, "pos_frac": 0.84375, "sample": [164.6932373046875, 34.555702209472656, 32.09737014770508, -59.50800323486328, 45.83247375488281, -55.29905700683594, 21.087112426757812, 155.2972412109375, -101.87828063964844, -47.495361328125, 51.97910690307617, 169.45553588867188, 31.989124298095703, 107.58709716796875, 21.157638549804688, 217.19775390625, 112.18630981445312, 86.7474136352539, 28.03852653503418, -42.82783508300781, 69.16246032714844, 30.9945068359375, 65.56824493408203, 27.72600746154785, 52.78428649902344, 157.8932342529297, -39.108612060546875, 143.64242553710938, 51.77382278442383, -29.215362548828125, 64.36392211914062, 112.39178466796875, 244.148193359375, 55.90412902832031, -2.0288314819335938, -49.22087097167969, 14.954605102539062, 3.1604156494140625, 103.95972442626953, 48.94978332519531, 37.16852569580078, 16.788846969604492, 125.50611877441406, 126.02407836914062, 135.90737915039062, 135.3211669921875, -5.167449951171875, 40.805137634277344, 113.21698760986328, 108.44730377197266, 29.847148895263672, 153.28854370117188, 19.835975646972656, 4.783761978149414, 129.32058715820312, 56.54591751098633, 59.72585678100586, 103.56265258789062, 18.962844848632812, 2.523406982421875, 45.056671142578125, 61.158538818359375, 46.27775573730469, 119.69892120361328], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000660.npy"}
|
|
{"epoch": 0.9977324263038548, "step": 661, "batch_size": 64, "mean": 42.786922454833984, "std": 80.24382781982422, "min": -171.3659210205078, "p10": -47.72219696044922, "median": 42.67686653137207, "p90": 151.60226593017578, "max": 217.162353515625, "pos_frac": 0.734375, "sample": [16.498291015625, -17.1241455078125, -45.973724365234375, -90.07169342041016, 101.0980224609375, 85.91293334960938, 117.58819580078125, -2.385284423828125, -171.3659210205078, 108.99003601074219, -99.10382080078125, 2.7719268798828125, -14.317306518554688, 160.86756896972656, 152.49468994140625, 109.59513092041016, 167.8665313720703, 72.93415832519531, 217.162353515625, 48.39955139160156, -159.88706970214844, 59.44163513183594, 108.86251831054688, -48.47154235839844, -4.4630889892578125, 13.386199951171875, 21.511260986328125, -31.249027252197266, 3.7553367614746094, 16.83734893798828, 26.676136016845703, 63.79021453857422, -94.50712585449219, 43.875526428222656, 80.94000244140625, 10.806957244873047, 48.34172058105469, 149.5199432373047, 12.179279327392578, 142.90628051757812, 77.93954467773438, 28.08856201171875, -7.7356414794921875, 140.57241821289062, 147.64822387695312, 61.30913543701172, 168.383544921875, 42.43500518798828, 5.588430404663086, 66.80268096923828, -52.47638702392578, 1.0809555053710938, 3.6686553955078125, -22.13300895690918, 42.91872787475586, 123.0425033569336, 67.43970489501953, 78.33441162109375, 49.13865661621094, 162.18075561523438, 18.59417724609375, -4.2545166015625, -23.409019470214844, 177.11549377441406], "npy": "/workspace/dynamic-dpo-v4/outputs/llama-3-8b-base-new-dpo-hh-harmless-s_star0.6-4xh200-batch-64-20260421-213851/margin_logs/step_0000661.npy"}
|