diff --git "a/trainer_state.json" "b/trainer_state.json" new file mode 100644--- /dev/null +++ "b/trainer_state.json" @@ -0,0 +1,4860 @@ +{ + "best_metric": null, + "best_model_checkpoint": null, + "epoch": 1.0, + "eval_steps": 200, + "global_step": 3348, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.0002986857825567503, + "grad_norm": 0.3188975789150384, + "learning_rate": 5.970149253731343e-07, + "loss": 0.2045, + "step": 1 + }, + { + "epoch": 0.0014934289127837516, + "grad_norm": 0.5466095149619057, + "learning_rate": 2.9850746268656716e-06, + "loss": 0.5187, + "step": 5 + }, + { + "epoch": 0.002986857825567503, + "grad_norm": 0.46728261805758203, + "learning_rate": 5.970149253731343e-06, + "loss": 0.4648, + "step": 10 + }, + { + "epoch": 0.004480286738351254, + "grad_norm": 0.6171065397357978, + "learning_rate": 8.955223880597016e-06, + "loss": 0.475, + "step": 15 + }, + { + "epoch": 0.005973715651135006, + "grad_norm": 0.49334289479515664, + "learning_rate": 1.1940298507462686e-05, + "loss": 0.4602, + "step": 20 + }, + { + "epoch": 0.007467144563918757, + "grad_norm": 0.5174483814171045, + "learning_rate": 1.4925373134328357e-05, + "loss": 0.4816, + "step": 25 + }, + { + "epoch": 0.008960573476702509, + "grad_norm": 0.4471748736139888, + "learning_rate": 1.791044776119403e-05, + "loss": 0.3534, + "step": 30 + }, + { + "epoch": 0.01045400238948626, + "grad_norm": 0.5227128696793767, + "learning_rate": 2.0895522388059702e-05, + "loss": 0.455, + "step": 35 + }, + { + "epoch": 0.011947431302270013, + "grad_norm": 0.57335506943901, + "learning_rate": 2.3880597014925373e-05, + "loss": 0.4381, + "step": 40 + }, + { + "epoch": 0.013440860215053764, + "grad_norm": 0.5672666580878948, + "learning_rate": 2.6865671641791047e-05, + "loss": 0.4183, + "step": 45 + }, + { + "epoch": 0.014934289127837515, + "grad_norm": 0.615033725558188, + "learning_rate": 2.9850746268656714e-05, + "loss": 0.4493, + "step": 50 + }, + { + "epoch": 0.016427718040621268, + "grad_norm": 0.5694295537449918, + "learning_rate": 3.283582089552239e-05, + "loss": 0.3214, + "step": 55 + }, + { + "epoch": 0.017921146953405017, + "grad_norm": 0.4317314011547438, + "learning_rate": 3.582089552238806e-05, + "loss": 0.3697, + "step": 60 + }, + { + "epoch": 0.01941457586618877, + "grad_norm": 0.5686093679349059, + "learning_rate": 3.8805970149253736e-05, + "loss": 0.3266, + "step": 65 + }, + { + "epoch": 0.02090800477897252, + "grad_norm": 0.5271452706616764, + "learning_rate": 4.1791044776119404e-05, + "loss": 0.365, + "step": 70 + }, + { + "epoch": 0.022401433691756272, + "grad_norm": 0.7400114199406753, + "learning_rate": 4.477611940298508e-05, + "loss": 0.4381, + "step": 75 + }, + { + "epoch": 0.023894862604540025, + "grad_norm": 0.5987079894007333, + "learning_rate": 4.7761194029850745e-05, + "loss": 0.3233, + "step": 80 + }, + { + "epoch": 0.025388291517323774, + "grad_norm": 0.5231052349056124, + "learning_rate": 5.074626865671642e-05, + "loss": 0.3543, + "step": 85 + }, + { + "epoch": 0.026881720430107527, + "grad_norm": 0.6540900409905596, + "learning_rate": 5.373134328358209e-05, + "loss": 0.3387, + "step": 90 + }, + { + "epoch": 0.028375149342891277, + "grad_norm": 0.5922650571689663, + "learning_rate": 5.671641791044776e-05, + "loss": 0.3648, + "step": 95 + }, + { + "epoch": 0.02986857825567503, + "grad_norm": 0.671441503007203, + "learning_rate": 5.970149253731343e-05, + "loss": 0.3842, + "step": 100 + }, + { + "epoch": 0.03136200716845878, + "grad_norm": 0.5589374162458473, + "learning_rate": 6.268656716417911e-05, + "loss": 0.3074, + "step": 105 + }, + { + "epoch": 0.032855436081242535, + "grad_norm": 0.5380602393505818, + "learning_rate": 6.567164179104478e-05, + "loss": 0.3507, + "step": 110 + }, + { + "epoch": 0.034348864994026285, + "grad_norm": 0.6063877341870142, + "learning_rate": 6.865671641791044e-05, + "loss": 0.3401, + "step": 115 + }, + { + "epoch": 0.035842293906810034, + "grad_norm": 0.5260512769152438, + "learning_rate": 7.164179104477612e-05, + "loss": 0.3724, + "step": 120 + }, + { + "epoch": 0.03733572281959379, + "grad_norm": 0.5959345681808844, + "learning_rate": 7.46268656716418e-05, + "loss": 0.3713, + "step": 125 + }, + { + "epoch": 0.03882915173237754, + "grad_norm": 0.5141504485060038, + "learning_rate": 7.761194029850747e-05, + "loss": 0.306, + "step": 130 + }, + { + "epoch": 0.04032258064516129, + "grad_norm": 0.45162013987431543, + "learning_rate": 8.059701492537314e-05, + "loss": 0.3279, + "step": 135 + }, + { + "epoch": 0.04181600955794504, + "grad_norm": 0.5664477964459901, + "learning_rate": 8.358208955223881e-05, + "loss": 0.2853, + "step": 140 + }, + { + "epoch": 0.043309438470728795, + "grad_norm": 0.4935236502406577, + "learning_rate": 8.656716417910447e-05, + "loss": 0.3515, + "step": 145 + }, + { + "epoch": 0.044802867383512544, + "grad_norm": 0.49883746692891684, + "learning_rate": 8.955223880597016e-05, + "loss": 0.3112, + "step": 150 + }, + { + "epoch": 0.046296296296296294, + "grad_norm": 0.5370459530188539, + "learning_rate": 9.253731343283582e-05, + "loss": 0.2918, + "step": 155 + }, + { + "epoch": 0.04778972520908005, + "grad_norm": 0.5147364002910404, + "learning_rate": 9.552238805970149e-05, + "loss": 0.3427, + "step": 160 + }, + { + "epoch": 0.0492831541218638, + "grad_norm": 0.48526830628523326, + "learning_rate": 9.850746268656717e-05, + "loss": 0.3272, + "step": 165 + }, + { + "epoch": 0.05077658303464755, + "grad_norm": 0.44536442368732077, + "learning_rate": 0.00010149253731343284, + "loss": 0.3216, + "step": 170 + }, + { + "epoch": 0.052270011947431305, + "grad_norm": 0.6529775264003637, + "learning_rate": 0.0001044776119402985, + "loss": 0.3343, + "step": 175 + }, + { + "epoch": 0.053763440860215055, + "grad_norm": 0.500538842360273, + "learning_rate": 0.00010746268656716419, + "loss": 0.2882, + "step": 180 + }, + { + "epoch": 0.055256869772998804, + "grad_norm": 0.45720922405183984, + "learning_rate": 0.00011044776119402987, + "loss": 0.3565, + "step": 185 + }, + { + "epoch": 0.05675029868578255, + "grad_norm": 0.395572707884918, + "learning_rate": 0.00011343283582089552, + "loss": 0.2879, + "step": 190 + }, + { + "epoch": 0.05824372759856631, + "grad_norm": 0.38317766271575404, + "learning_rate": 0.0001164179104477612, + "loss": 0.326, + "step": 195 + }, + { + "epoch": 0.05973715651135006, + "grad_norm": 0.5636861131014499, + "learning_rate": 0.00011940298507462686, + "loss": 0.3403, + "step": 200 + }, + { + "epoch": 0.05973715651135006, + "eval_loss": 0.3074450194835663, + "eval_runtime": 675.2136, + "eval_samples_per_second": 5.924, + "eval_steps_per_second": 0.741, + "step": 200 + }, + { + "epoch": 0.06123058542413381, + "grad_norm": 0.5285903241151306, + "learning_rate": 0.00012238805970149255, + "loss": 0.3054, + "step": 205 + }, + { + "epoch": 0.06272401433691756, + "grad_norm": 0.462285447656946, + "learning_rate": 0.00012537313432835822, + "loss": 0.3289, + "step": 210 + }, + { + "epoch": 0.06421744324970131, + "grad_norm": 0.41949653076343973, + "learning_rate": 0.00012835820895522389, + "loss": 0.3235, + "step": 215 + }, + { + "epoch": 0.06571087216248507, + "grad_norm": 0.37413394493320784, + "learning_rate": 0.00013134328358208955, + "loss": 0.3284, + "step": 220 + }, + { + "epoch": 0.06720430107526881, + "grad_norm": 0.43599604159082916, + "learning_rate": 0.00013432835820895525, + "loss": 0.3509, + "step": 225 + }, + { + "epoch": 0.06869772998805257, + "grad_norm": 0.45911550206829044, + "learning_rate": 0.0001373134328358209, + "loss": 0.2856, + "step": 230 + }, + { + "epoch": 0.07019115890083633, + "grad_norm": 0.6197373846263158, + "learning_rate": 0.00014029850746268658, + "loss": 0.3381, + "step": 235 + }, + { + "epoch": 0.07168458781362007, + "grad_norm": 0.5162001410157283, + "learning_rate": 0.00014328358208955225, + "loss": 0.2987, + "step": 240 + }, + { + "epoch": 0.07317801672640382, + "grad_norm": 0.3756162304697717, + "learning_rate": 0.00014626865671641792, + "loss": 0.3016, + "step": 245 + }, + { + "epoch": 0.07467144563918758, + "grad_norm": 0.4427925525549972, + "learning_rate": 0.0001492537313432836, + "loss": 0.3301, + "step": 250 + }, + { + "epoch": 0.07616487455197132, + "grad_norm": 0.4587549108919026, + "learning_rate": 0.00015223880597014925, + "loss": 0.2864, + "step": 255 + }, + { + "epoch": 0.07765830346475508, + "grad_norm": 0.4539343579685493, + "learning_rate": 0.00015522388059701495, + "loss": 0.3263, + "step": 260 + }, + { + "epoch": 0.07915173237753884, + "grad_norm": 0.4601786532582476, + "learning_rate": 0.00015820895522388059, + "loss": 0.3257, + "step": 265 + }, + { + "epoch": 0.08064516129032258, + "grad_norm": 0.4952572370631647, + "learning_rate": 0.00016119402985074628, + "loss": 0.3227, + "step": 270 + }, + { + "epoch": 0.08213859020310633, + "grad_norm": 0.5594600091045848, + "learning_rate": 0.00016417910447761195, + "loss": 0.3662, + "step": 275 + }, + { + "epoch": 0.08363201911589008, + "grad_norm": 0.4150317719098373, + "learning_rate": 0.00016716417910447761, + "loss": 0.251, + "step": 280 + }, + { + "epoch": 0.08512544802867383, + "grad_norm": 0.45530314509050923, + "learning_rate": 0.00017014925373134328, + "loss": 0.3277, + "step": 285 + }, + { + "epoch": 0.08661887694145759, + "grad_norm": 0.3707658637725497, + "learning_rate": 0.00017313432835820895, + "loss": 0.2958, + "step": 290 + }, + { + "epoch": 0.08811230585424133, + "grad_norm": 0.3532362466068546, + "learning_rate": 0.00017611940298507464, + "loss": 0.3275, + "step": 295 + }, + { + "epoch": 0.08960573476702509, + "grad_norm": 0.41975328143934704, + "learning_rate": 0.0001791044776119403, + "loss": 0.361, + "step": 300 + }, + { + "epoch": 0.09109916367980884, + "grad_norm": 0.3889061650864088, + "learning_rate": 0.00018208955223880598, + "loss": 0.2696, + "step": 305 + }, + { + "epoch": 0.09259259259259259, + "grad_norm": 0.46543359027880427, + "learning_rate": 0.00018507462686567165, + "loss": 0.3443, + "step": 310 + }, + { + "epoch": 0.09408602150537634, + "grad_norm": 0.5714057611200172, + "learning_rate": 0.00018805970149253734, + "loss": 0.3374, + "step": 315 + }, + { + "epoch": 0.0955794504181601, + "grad_norm": 0.3948901958880527, + "learning_rate": 0.00019104477611940298, + "loss": 0.321, + "step": 320 + }, + { + "epoch": 0.09707287933094384, + "grad_norm": 0.5540915005920514, + "learning_rate": 0.00019402985074626867, + "loss": 0.3277, + "step": 325 + }, + { + "epoch": 0.0985663082437276, + "grad_norm": 0.48042615725477694, + "learning_rate": 0.00019701492537313434, + "loss": 0.2833, + "step": 330 + }, + { + "epoch": 0.10005973715651136, + "grad_norm": 0.35451829573422367, + "learning_rate": 0.0002, + "loss": 0.2882, + "step": 335 + }, + { + "epoch": 0.1015531660692951, + "grad_norm": 0.37754764974968846, + "learning_rate": 0.00019999864102799163, + "loss": 0.2981, + "step": 340 + }, + { + "epoch": 0.10304659498207885, + "grad_norm": 0.41754484552953547, + "learning_rate": 0.0001999945641489025, + "loss": 0.3385, + "step": 345 + }, + { + "epoch": 0.10454002389486261, + "grad_norm": 0.48242379028499566, + "learning_rate": 0.00019998776947353995, + "loss": 0.3462, + "step": 350 + }, + { + "epoch": 0.10603345280764635, + "grad_norm": 0.43615681396287914, + "learning_rate": 0.00019997825718657945, + "loss": 0.2829, + "step": 355 + }, + { + "epoch": 0.10752688172043011, + "grad_norm": 0.5382667372932559, + "learning_rate": 0.0001999660275465596, + "loss": 0.3086, + "step": 360 + }, + { + "epoch": 0.10902031063321387, + "grad_norm": 0.4221095850996317, + "learning_rate": 0.00019995108088587528, + "loss": 0.2853, + "step": 365 + }, + { + "epoch": 0.11051373954599761, + "grad_norm": 0.4049511576507522, + "learning_rate": 0.00019993341761076824, + "loss": 0.3144, + "step": 370 + }, + { + "epoch": 0.11200716845878136, + "grad_norm": 0.41672237725910044, + "learning_rate": 0.00019991303820131645, + "loss": 0.3262, + "step": 375 + }, + { + "epoch": 0.1135005973715651, + "grad_norm": 0.37665369388894876, + "learning_rate": 0.00019988994321142088, + "loss": 0.2746, + "step": 380 + }, + { + "epoch": 0.11499402628434886, + "grad_norm": 0.38890047129761735, + "learning_rate": 0.00019986413326879035, + "loss": 0.3194, + "step": 385 + }, + { + "epoch": 0.11648745519713262, + "grad_norm": 0.37534439629256694, + "learning_rate": 0.00019983560907492476, + "loss": 0.3134, + "step": 390 + }, + { + "epoch": 0.11798088410991636, + "grad_norm": 0.3401949072629029, + "learning_rate": 0.00019980437140509563, + "loss": 0.3198, + "step": 395 + }, + { + "epoch": 0.11947431302270012, + "grad_norm": 0.4404312617632526, + "learning_rate": 0.00019977042110832537, + "loss": 0.3224, + "step": 400 + }, + { + "epoch": 0.11947431302270012, + "eval_loss": 0.29538044333457947, + "eval_runtime": 671.0671, + "eval_samples_per_second": 5.961, + "eval_steps_per_second": 0.745, + "step": 400 + }, + { + "epoch": 0.12096774193548387, + "grad_norm": 0.3877778527340937, + "learning_rate": 0.00019973375910736408, + "loss": 0.2605, + "step": 405 + }, + { + "epoch": 0.12246117084826762, + "grad_norm": 0.42449178944770155, + "learning_rate": 0.00019969438639866436, + "loss": 0.3292, + "step": 410 + }, + { + "epoch": 0.12395459976105137, + "grad_norm": 0.4421146156020614, + "learning_rate": 0.00019965230405235443, + "loss": 0.2847, + "step": 415 + }, + { + "epoch": 0.12544802867383512, + "grad_norm": 0.33313138965230243, + "learning_rate": 0.00019960751321220887, + "loss": 0.3346, + "step": 420 + }, + { + "epoch": 0.12694145758661887, + "grad_norm": 0.35768831528441764, + "learning_rate": 0.0001995600150956177, + "loss": 0.2896, + "step": 425 + }, + { + "epoch": 0.12843488649940263, + "grad_norm": 0.4073503627739374, + "learning_rate": 0.0001995098109935531, + "loss": 0.2963, + "step": 430 + }, + { + "epoch": 0.12992831541218638, + "grad_norm": 0.4634204409239979, + "learning_rate": 0.00019945690227053445, + "loss": 0.309, + "step": 435 + }, + { + "epoch": 0.13142174432497014, + "grad_norm": 0.48789887803899984, + "learning_rate": 0.00019940129036459121, + "loss": 0.2778, + "step": 440 + }, + { + "epoch": 0.13291517323775387, + "grad_norm": 0.41135714129715417, + "learning_rate": 0.0001993429767872239, + "loss": 0.312, + "step": 445 + }, + { + "epoch": 0.13440860215053763, + "grad_norm": 0.48911130889740795, + "learning_rate": 0.00019928196312336285, + "loss": 0.354, + "step": 450 + }, + { + "epoch": 0.13590203106332138, + "grad_norm": 0.31087210214054994, + "learning_rate": 0.00019921825103132531, + "loss": 0.2581, + "step": 455 + }, + { + "epoch": 0.13739545997610514, + "grad_norm": 0.34319695770593983, + "learning_rate": 0.00019915184224277032, + "loss": 0.2861, + "step": 460 + }, + { + "epoch": 0.1388888888888889, + "grad_norm": 0.39072017479958165, + "learning_rate": 0.00019908273856265152, + "loss": 0.28, + "step": 465 + }, + { + "epoch": 0.14038231780167265, + "grad_norm": 0.38108092212692735, + "learning_rate": 0.00019901094186916825, + "loss": 0.2938, + "step": 470 + }, + { + "epoch": 0.14187574671445638, + "grad_norm": 0.40284399729512566, + "learning_rate": 0.00019893645411371447, + "loss": 0.3232, + "step": 475 + }, + { + "epoch": 0.14336917562724014, + "grad_norm": 0.41081349055067273, + "learning_rate": 0.00019885927732082563, + "loss": 0.2682, + "step": 480 + }, + { + "epoch": 0.1448626045400239, + "grad_norm": 0.35453420174268324, + "learning_rate": 0.00019877941358812382, + "loss": 0.3099, + "step": 485 + }, + { + "epoch": 0.14635603345280765, + "grad_norm": 0.4219522172174877, + "learning_rate": 0.00019869686508626054, + "loss": 0.2929, + "step": 490 + }, + { + "epoch": 0.1478494623655914, + "grad_norm": 0.433146389991006, + "learning_rate": 0.00019861163405885787, + "loss": 0.3255, + "step": 495 + }, + { + "epoch": 0.14934289127837516, + "grad_norm": 0.43636524610968963, + "learning_rate": 0.0001985237228224474, + "loss": 0.3251, + "step": 500 + }, + { + "epoch": 0.1508363201911589, + "grad_norm": 0.39903987473575914, + "learning_rate": 0.00019843313376640732, + "loss": 0.2786, + "step": 505 + }, + { + "epoch": 0.15232974910394265, + "grad_norm": 0.42757498238366765, + "learning_rate": 0.0001983398693528975, + "loss": 0.3365, + "step": 510 + }, + { + "epoch": 0.1538231780167264, + "grad_norm": 0.4030570437823937, + "learning_rate": 0.00019824393211679246, + "loss": 0.2501, + "step": 515 + }, + { + "epoch": 0.15531660692951016, + "grad_norm": 0.3624497193987114, + "learning_rate": 0.00019814532466561259, + "loss": 0.2987, + "step": 520 + }, + { + "epoch": 0.15681003584229392, + "grad_norm": 0.4694503509924643, + "learning_rate": 0.00019804404967945315, + "loss": 0.3244, + "step": 525 + }, + { + "epoch": 0.15830346475507767, + "grad_norm": 0.43327424018713545, + "learning_rate": 0.00019794010991091164, + "loss": 0.2656, + "step": 530 + }, + { + "epoch": 0.1597968936678614, + "grad_norm": 0.36124038110578643, + "learning_rate": 0.00019783350818501272, + "loss": 0.2949, + "step": 535 + }, + { + "epoch": 0.16129032258064516, + "grad_norm": 0.4813284517384988, + "learning_rate": 0.00019772424739913168, + "loss": 0.2527, + "step": 540 + }, + { + "epoch": 0.1627837514934289, + "grad_norm": 0.517676779630334, + "learning_rate": 0.00019761233052291544, + "loss": 0.3322, + "step": 545 + }, + { + "epoch": 0.16427718040621267, + "grad_norm": 0.41586609149970155, + "learning_rate": 0.0001974977605982021, + "loss": 0.3058, + "step": 550 + }, + { + "epoch": 0.16577060931899643, + "grad_norm": 0.38142158790681063, + "learning_rate": 0.00019738054073893807, + "loss": 0.2617, + "step": 555 + }, + { + "epoch": 0.16726403823178015, + "grad_norm": 0.4508850152729104, + "learning_rate": 0.00019726067413109347, + "loss": 0.3124, + "step": 560 + }, + { + "epoch": 0.1687574671445639, + "grad_norm": 0.38711879954880984, + "learning_rate": 0.0001971381640325756, + "loss": 0.2886, + "step": 565 + }, + { + "epoch": 0.17025089605734767, + "grad_norm": 0.3713159311530026, + "learning_rate": 0.00019701301377314038, + "loss": 0.2884, + "step": 570 + }, + { + "epoch": 0.17174432497013142, + "grad_norm": 0.4253789489790291, + "learning_rate": 0.0001968852267543018, + "loss": 0.3403, + "step": 575 + }, + { + "epoch": 0.17323775388291518, + "grad_norm": 0.3716154947554444, + "learning_rate": 0.00019675480644923944, + "loss": 0.2563, + "step": 580 + }, + { + "epoch": 0.17473118279569894, + "grad_norm": 0.4067400864065092, + "learning_rate": 0.00019662175640270424, + "loss": 0.3139, + "step": 585 + }, + { + "epoch": 0.17622461170848266, + "grad_norm": 0.3587456607358574, + "learning_rate": 0.00019648608023092195, + "loss": 0.2851, + "step": 590 + }, + { + "epoch": 0.17771804062126642, + "grad_norm": 0.35869894756290216, + "learning_rate": 0.00019634778162149497, + "loss": 0.299, + "step": 595 + }, + { + "epoch": 0.17921146953405018, + "grad_norm": 0.4291965063428789, + "learning_rate": 0.00019620686433330207, + "loss": 0.3055, + "step": 600 + }, + { + "epoch": 0.17921146953405018, + "eval_loss": 0.28864020109176636, + "eval_runtime": 670.6057, + "eval_samples_per_second": 5.965, + "eval_steps_per_second": 0.746, + "step": 600 + }, + { + "epoch": 0.18070489844683393, + "grad_norm": 0.4107636120154602, + "learning_rate": 0.00019606333219639624, + "loss": 0.2674, + "step": 605 + }, + { + "epoch": 0.1821983273596177, + "grad_norm": 0.4094821475481708, + "learning_rate": 0.00019591718911190066, + "loss": 0.2748, + "step": 610 + }, + { + "epoch": 0.18369175627240145, + "grad_norm": 0.42712744392327107, + "learning_rate": 0.00019576843905190253, + "loss": 0.2913, + "step": 615 + }, + { + "epoch": 0.18518518518518517, + "grad_norm": 0.3838363588569511, + "learning_rate": 0.00019561708605934515, + "loss": 0.3019, + "step": 620 + }, + { + "epoch": 0.18667861409796893, + "grad_norm": 0.4060361946228662, + "learning_rate": 0.0001954631342479182, + "loss": 0.3028, + "step": 625 + }, + { + "epoch": 0.1881720430107527, + "grad_norm": 0.36895107070060834, + "learning_rate": 0.0001953065878019457, + "loss": 0.2912, + "step": 630 + }, + { + "epoch": 0.18966547192353644, + "grad_norm": 0.5389015474723242, + "learning_rate": 0.0001951474509762724, + "loss": 0.3277, + "step": 635 + }, + { + "epoch": 0.1911589008363202, + "grad_norm": 0.3713681816986798, + "learning_rate": 0.0001949857280961481, + "loss": 0.2564, + "step": 640 + }, + { + "epoch": 0.19265232974910393, + "grad_norm": 0.39787955963512267, + "learning_rate": 0.00019482142355711023, + "loss": 0.3167, + "step": 645 + }, + { + "epoch": 0.19414575866188769, + "grad_norm": 0.4160144539607265, + "learning_rate": 0.0001946545418248641, + "loss": 0.3473, + "step": 650 + }, + { + "epoch": 0.19563918757467144, + "grad_norm": 0.4611729979684992, + "learning_rate": 0.00019448508743516186, + "loss": 0.2579, + "step": 655 + }, + { + "epoch": 0.1971326164874552, + "grad_norm": 0.37926676547852983, + "learning_rate": 0.00019431306499367886, + "loss": 0.2995, + "step": 660 + }, + { + "epoch": 0.19862604540023895, + "grad_norm": 0.41885156529130685, + "learning_rate": 0.00019413847917588878, + "loss": 0.3042, + "step": 665 + }, + { + "epoch": 0.2001194743130227, + "grad_norm": 0.427624169132239, + "learning_rate": 0.00019396133472693642, + "loss": 0.3173, + "step": 670 + }, + { + "epoch": 0.20161290322580644, + "grad_norm": 0.38475710691405496, + "learning_rate": 0.00019378163646150876, + "loss": 0.3021, + "step": 675 + }, + { + "epoch": 0.2031063321385902, + "grad_norm": 0.4138321670681785, + "learning_rate": 0.000193599389263704, + "loss": 0.274, + "step": 680 + }, + { + "epoch": 0.20459976105137395, + "grad_norm": 0.355884010075492, + "learning_rate": 0.00019341459808689898, + "loss": 0.3393, + "step": 685 + }, + { + "epoch": 0.2060931899641577, + "grad_norm": 0.3631472347986104, + "learning_rate": 0.00019322726795361443, + "loss": 0.2949, + "step": 690 + }, + { + "epoch": 0.20758661887694146, + "grad_norm": 0.4773277681201716, + "learning_rate": 0.0001930374039553785, + "loss": 0.301, + "step": 695 + }, + { + "epoch": 0.20908004778972522, + "grad_norm": 0.448239132237338, + "learning_rate": 0.00019284501125258835, + "loss": 0.3332, + "step": 700 + }, + { + "epoch": 0.21057347670250895, + "grad_norm": 0.38679234194418, + "learning_rate": 0.00019265009507436997, + "loss": 0.2821, + "step": 705 + }, + { + "epoch": 0.2120669056152927, + "grad_norm": 0.34050133768678587, + "learning_rate": 0.00019245266071843596, + "loss": 0.2971, + "step": 710 + }, + { + "epoch": 0.21356033452807646, + "grad_norm": 0.377904449126756, + "learning_rate": 0.00019225271355094155, + "loss": 0.2644, + "step": 715 + }, + { + "epoch": 0.21505376344086022, + "grad_norm": 0.43091731111413145, + "learning_rate": 0.00019205025900633884, + "loss": 0.3028, + "step": 720 + }, + { + "epoch": 0.21654719235364397, + "grad_norm": 0.35823568635784103, + "learning_rate": 0.00019184530258722899, + "loss": 0.2941, + "step": 725 + }, + { + "epoch": 0.21804062126642773, + "grad_norm": 0.3871906521592633, + "learning_rate": 0.00019163784986421276, + "loss": 0.2453, + "step": 730 + }, + { + "epoch": 0.21953405017921146, + "grad_norm": 0.3870097665982384, + "learning_rate": 0.00019142790647573902, + "loss": 0.3163, + "step": 735 + }, + { + "epoch": 0.22102747909199522, + "grad_norm": 0.4043861350049887, + "learning_rate": 0.00019121547812795152, + "loss": 0.2867, + "step": 740 + }, + { + "epoch": 0.22252090800477897, + "grad_norm": 0.3472296001968241, + "learning_rate": 0.00019100057059453381, + "loss": 0.308, + "step": 745 + }, + { + "epoch": 0.22401433691756273, + "grad_norm": 0.4555451484017787, + "learning_rate": 0.00019078318971655237, + "loss": 0.3094, + "step": 750 + }, + { + "epoch": 0.22550776583034648, + "grad_norm": 0.39417763113385473, + "learning_rate": 0.00019056334140229777, + "loss": 0.2673, + "step": 755 + }, + { + "epoch": 0.2270011947431302, + "grad_norm": 0.33903321452810364, + "learning_rate": 0.00019034103162712408, + "loss": 0.318, + "step": 760 + }, + { + "epoch": 0.22849462365591397, + "grad_norm": 0.37450897393444954, + "learning_rate": 0.0001901162664332866, + "loss": 0.2543, + "step": 765 + }, + { + "epoch": 0.22998805256869773, + "grad_norm": 0.3798157275859589, + "learning_rate": 0.0001898890519297774, + "loss": 0.2945, + "step": 770 + }, + { + "epoch": 0.23148148148148148, + "grad_norm": 0.3682423785435909, + "learning_rate": 0.00018965939429215948, + "loss": 0.3127, + "step": 775 + }, + { + "epoch": 0.23297491039426524, + "grad_norm": 0.3771090235525355, + "learning_rate": 0.0001894272997623989, + "loss": 0.2722, + "step": 780 + }, + { + "epoch": 0.234468339307049, + "grad_norm": 0.33941544966890785, + "learning_rate": 0.00018919277464869504, + "loss": 0.3163, + "step": 785 + }, + { + "epoch": 0.23596176821983272, + "grad_norm": 0.37376605453569783, + "learning_rate": 0.0001889558253253092, + "loss": 0.2598, + "step": 790 + }, + { + "epoch": 0.23745519713261648, + "grad_norm": 0.38959423980866825, + "learning_rate": 0.00018871645823239128, + "loss": 0.2983, + "step": 795 + }, + { + "epoch": 0.23894862604540024, + "grad_norm": 0.38139963613722794, + "learning_rate": 0.00018847467987580493, + "loss": 0.2899, + "step": 800 + }, + { + "epoch": 0.23894862604540024, + "eval_loss": 0.2804073989391327, + "eval_runtime": 670.1656, + "eval_samples_per_second": 5.969, + "eval_steps_per_second": 0.746, + "step": 800 + }, + { + "epoch": 0.240442054958184, + "grad_norm": 0.40419199411712203, + "learning_rate": 0.00018823049682695052, + "loss": 0.3008, + "step": 805 + }, + { + "epoch": 0.24193548387096775, + "grad_norm": 0.3612377971129905, + "learning_rate": 0.0001879839157225866, + "loss": 0.316, + "step": 810 + }, + { + "epoch": 0.2434289127837515, + "grad_norm": 0.4605084068043848, + "learning_rate": 0.0001877349432646495, + "loss": 0.2776, + "step": 815 + }, + { + "epoch": 0.24492234169653523, + "grad_norm": 0.39686115392847526, + "learning_rate": 0.0001874835862200713, + "loss": 0.2664, + "step": 820 + }, + { + "epoch": 0.246415770609319, + "grad_norm": 0.38181276219849275, + "learning_rate": 0.00018722985142059572, + "loss": 0.2896, + "step": 825 + }, + { + "epoch": 0.24790919952210275, + "grad_norm": 0.36529483511621264, + "learning_rate": 0.0001869737457625926, + "loss": 0.2939, + "step": 830 + }, + { + "epoch": 0.2494026284348865, + "grad_norm": 0.4035594565736224, + "learning_rate": 0.00018671527620687034, + "loss": 0.3102, + "step": 835 + }, + { + "epoch": 0.25089605734767023, + "grad_norm": 0.3854541007588229, + "learning_rate": 0.00018645444977848677, + "loss": 0.2769, + "step": 840 + }, + { + "epoch": 0.252389486260454, + "grad_norm": 0.42123392936346155, + "learning_rate": 0.00018619127356655813, + "loss": 0.308, + "step": 845 + }, + { + "epoch": 0.25388291517323774, + "grad_norm": 0.49950892415773196, + "learning_rate": 0.0001859257547240666, + "loss": 0.314, + "step": 850 + }, + { + "epoch": 0.2553763440860215, + "grad_norm": 0.40906372202780966, + "learning_rate": 0.00018565790046766564, + "loss": 0.2881, + "step": 855 + }, + { + "epoch": 0.25686977299880526, + "grad_norm": 0.38764669514338385, + "learning_rate": 0.000185387718077484, + "loss": 0.3027, + "step": 860 + }, + { + "epoch": 0.258363201911589, + "grad_norm": 0.42262538628601043, + "learning_rate": 0.00018511521489692775, + "loss": 0.2545, + "step": 865 + }, + { + "epoch": 0.25985663082437277, + "grad_norm": 0.41209507390817657, + "learning_rate": 0.00018484039833248085, + "loss": 0.2828, + "step": 870 + }, + { + "epoch": 0.2613500597371565, + "grad_norm": 0.4469642279365962, + "learning_rate": 0.0001845632758535036, + "loss": 0.332, + "step": 875 + }, + { + "epoch": 0.2628434886499403, + "grad_norm": 0.36473061276227664, + "learning_rate": 0.00018428385499202988, + "loss": 0.2647, + "step": 880 + }, + { + "epoch": 0.26433691756272404, + "grad_norm": 0.4433634956798964, + "learning_rate": 0.00018400214334256227, + "loss": 0.2952, + "step": 885 + }, + { + "epoch": 0.26583034647550774, + "grad_norm": 0.3494167236815255, + "learning_rate": 0.00018371814856186572, + "loss": 0.2825, + "step": 890 + }, + { + "epoch": 0.2673237753882915, + "grad_norm": 0.38005445773872815, + "learning_rate": 0.00018343187836875928, + "loss": 0.3096, + "step": 895 + }, + { + "epoch": 0.26881720430107525, + "grad_norm": 0.5133589136001121, + "learning_rate": 0.00018314334054390664, + "loss": 0.3078, + "step": 900 + }, + { + "epoch": 0.270310633213859, + "grad_norm": 0.37745450032232525, + "learning_rate": 0.00018285254292960433, + "loss": 0.2687, + "step": 905 + }, + { + "epoch": 0.27180406212664276, + "grad_norm": 0.40501108505960953, + "learning_rate": 0.00018255949342956863, + "loss": 0.2969, + "step": 910 + }, + { + "epoch": 0.2732974910394265, + "grad_norm": 0.4103922981810604, + "learning_rate": 0.000182264200008721, + "loss": 0.2635, + "step": 915 + }, + { + "epoch": 0.2747909199522103, + "grad_norm": 0.8269261010911084, + "learning_rate": 0.00018196667069297123, + "loss": 0.2894, + "step": 920 + }, + { + "epoch": 0.27628434886499403, + "grad_norm": 0.5797869922585875, + "learning_rate": 0.0001816669135689996, + "loss": 0.3175, + "step": 925 + }, + { + "epoch": 0.2777777777777778, + "grad_norm": 0.3960727461959874, + "learning_rate": 0.00018136493678403686, + "loss": 0.2122, + "step": 930 + }, + { + "epoch": 0.27927120669056155, + "grad_norm": 0.3281979427197208, + "learning_rate": 0.00018106074854564306, + "loss": 0.2772, + "step": 935 + }, + { + "epoch": 0.2807646356033453, + "grad_norm": 0.3180009593314954, + "learning_rate": 0.00018075435712148417, + "loss": 0.2494, + "step": 940 + }, + { + "epoch": 0.28225806451612906, + "grad_norm": 0.5630380699907694, + "learning_rate": 0.00018044577083910758, + "loss": 0.287, + "step": 945 + }, + { + "epoch": 0.28375149342891276, + "grad_norm": 0.5821492853630149, + "learning_rate": 0.00018013499808571567, + "loss": 0.3064, + "step": 950 + }, + { + "epoch": 0.2852449223416965, + "grad_norm": 0.34343056721676674, + "learning_rate": 0.00017982204730793795, + "loss": 0.2479, + "step": 955 + }, + { + "epoch": 0.2867383512544803, + "grad_norm": 0.4045824911297299, + "learning_rate": 0.0001795069270116013, + "loss": 0.3042, + "step": 960 + }, + { + "epoch": 0.28823178016726403, + "grad_norm": 0.4423077224611283, + "learning_rate": 0.000179189645761499, + "loss": 0.2587, + "step": 965 + }, + { + "epoch": 0.2897252090800478, + "grad_norm": 0.3881310780181844, + "learning_rate": 0.00017887021218115782, + "loss": 0.2883, + "step": 970 + }, + { + "epoch": 0.29121863799283154, + "grad_norm": 0.4157883620231867, + "learning_rate": 0.00017854863495260354, + "loss": 0.3163, + "step": 975 + }, + { + "epoch": 0.2927120669056153, + "grad_norm": 0.5286260329472577, + "learning_rate": 0.00017822492281612532, + "loss": 0.2784, + "step": 980 + }, + { + "epoch": 0.29420549581839905, + "grad_norm": 0.43556243885834467, + "learning_rate": 0.00017789908457003777, + "loss": 0.2934, + "step": 985 + }, + { + "epoch": 0.2956989247311828, + "grad_norm": 0.46879974456717627, + "learning_rate": 0.000177571129070442, + "loss": 0.273, + "step": 990 + }, + { + "epoch": 0.29719235364396657, + "grad_norm": 0.3913970910916109, + "learning_rate": 0.00017724106523098486, + "loss": 0.2864, + "step": 995 + }, + { + "epoch": 0.2986857825567503, + "grad_norm": 0.5166170033604143, + "learning_rate": 0.00017690890202261676, + "loss": 0.3116, + "step": 1000 + }, + { + "epoch": 0.2986857825567503, + "eval_loss": 0.27715662121772766, + "eval_runtime": 670.7329, + "eval_samples_per_second": 5.964, + "eval_steps_per_second": 0.745, + "step": 1000 + }, + { + "epoch": 0.300179211469534, + "grad_norm": 0.49927259244291966, + "learning_rate": 0.00017657464847334775, + "loss": 0.2295, + "step": 1005 + }, + { + "epoch": 0.3016726403823178, + "grad_norm": 0.4015981411197979, + "learning_rate": 0.0001762383136680022, + "loss": 0.2807, + "step": 1010 + }, + { + "epoch": 0.30316606929510154, + "grad_norm": 0.33536145313741844, + "learning_rate": 0.00017589990674797171, + "loss": 0.2683, + "step": 1015 + }, + { + "epoch": 0.3046594982078853, + "grad_norm": 0.4291404869332869, + "learning_rate": 0.00017555943691096706, + "loss": 0.2704, + "step": 1020 + }, + { + "epoch": 0.30615292712066905, + "grad_norm": 0.47290922277355507, + "learning_rate": 0.00017521691341076774, + "loss": 0.3136, + "step": 1025 + }, + { + "epoch": 0.3076463560334528, + "grad_norm": 0.4688378459680159, + "learning_rate": 0.00017487234555697072, + "loss": 0.272, + "step": 1030 + }, + { + "epoch": 0.30913978494623656, + "grad_norm": 0.36206038691433684, + "learning_rate": 0.0001745257427147374, + "loss": 0.2949, + "step": 1035 + }, + { + "epoch": 0.3106332138590203, + "grad_norm": 0.34718258949903247, + "learning_rate": 0.00017417711430453897, + "loss": 0.2724, + "step": 1040 + }, + { + "epoch": 0.3121266427718041, + "grad_norm": 0.35896609288857995, + "learning_rate": 0.00017382646980190048, + "loss": 0.2893, + "step": 1045 + }, + { + "epoch": 0.31362007168458783, + "grad_norm": 0.4371043853081932, + "learning_rate": 0.00017347381873714316, + "loss": 0.2874, + "step": 1050 + }, + { + "epoch": 0.3151135005973716, + "grad_norm": 0.33305502199179604, + "learning_rate": 0.00017311917069512555, + "loss": 0.2329, + "step": 1055 + }, + { + "epoch": 0.31660692951015534, + "grad_norm": 0.43013382325716004, + "learning_rate": 0.00017276253531498293, + "loss": 0.312, + "step": 1060 + }, + { + "epoch": 0.31810035842293904, + "grad_norm": 0.347929106798288, + "learning_rate": 0.00017240392228986518, + "loss": 0.2648, + "step": 1065 + }, + { + "epoch": 0.3195937873357228, + "grad_norm": 0.3492181786101714, + "learning_rate": 0.00017204334136667365, + "loss": 0.2821, + "step": 1070 + }, + { + "epoch": 0.32108721624850656, + "grad_norm": 0.45792503261610795, + "learning_rate": 0.0001716808023457959, + "loss": 0.3277, + "step": 1075 + }, + { + "epoch": 0.3225806451612903, + "grad_norm": 0.3654545092753955, + "learning_rate": 0.00017131631508083962, + "loss": 0.2675, + "step": 1080 + }, + { + "epoch": 0.32407407407407407, + "grad_norm": 0.37947758138241466, + "learning_rate": 0.0001709498894783646, + "loss": 0.2893, + "step": 1085 + }, + { + "epoch": 0.3255675029868578, + "grad_norm": 0.4068572595536966, + "learning_rate": 0.0001705815354976135, + "loss": 0.2435, + "step": 1090 + }, + { + "epoch": 0.3270609318996416, + "grad_norm": 0.4809381438376721, + "learning_rate": 0.00017021126315024145, + "loss": 0.2985, + "step": 1095 + }, + { + "epoch": 0.32855436081242534, + "grad_norm": 0.40059623315230986, + "learning_rate": 0.00016983908250004344, + "loss": 0.273, + "step": 1100 + }, + { + "epoch": 0.3300477897252091, + "grad_norm": 0.4052544383737238, + "learning_rate": 0.00016946500366268123, + "loss": 0.2433, + "step": 1105 + }, + { + "epoch": 0.33154121863799285, + "grad_norm": 0.3560399096095347, + "learning_rate": 0.0001690890368054082, + "loss": 0.2932, + "step": 1110 + }, + { + "epoch": 0.3330346475507766, + "grad_norm": 0.4156230656107325, + "learning_rate": 0.00016871119214679304, + "loss": 0.2685, + "step": 1115 + }, + { + "epoch": 0.3345280764635603, + "grad_norm": 0.3271097281515262, + "learning_rate": 0.00016833147995644202, + "loss": 0.3102, + "step": 1120 + }, + { + "epoch": 0.33602150537634407, + "grad_norm": 0.4217178620957521, + "learning_rate": 0.00016794991055471992, + "loss": 0.293, + "step": 1125 + }, + { + "epoch": 0.3375149342891278, + "grad_norm": 0.37416501140273456, + "learning_rate": 0.00016756649431246953, + "loss": 0.2479, + "step": 1130 + }, + { + "epoch": 0.3390083632019116, + "grad_norm": 0.45055007223505494, + "learning_rate": 0.00016718124165072953, + "loss": 0.2836, + "step": 1135 + }, + { + "epoch": 0.34050179211469533, + "grad_norm": 0.40163491862797124, + "learning_rate": 0.0001667941630404517, + "loss": 0.2484, + "step": 1140 + }, + { + "epoch": 0.3419952210274791, + "grad_norm": 0.41798757856275687, + "learning_rate": 0.00016640526900221593, + "loss": 0.2881, + "step": 1145 + }, + { + "epoch": 0.34348864994026285, + "grad_norm": 0.38258918057820673, + "learning_rate": 0.00016601457010594447, + "loss": 0.3107, + "step": 1150 + }, + { + "epoch": 0.3449820788530466, + "grad_norm": 0.3845197614655172, + "learning_rate": 0.0001656220769706146, + "loss": 0.2848, + "step": 1155 + }, + { + "epoch": 0.34647550776583036, + "grad_norm": 0.3985931675974656, + "learning_rate": 0.00016522780026397, + "loss": 0.2887, + "step": 1160 + }, + { + "epoch": 0.3479689366786141, + "grad_norm": 0.35770417481369043, + "learning_rate": 0.00016483175070223081, + "loss": 0.2504, + "step": 1165 + }, + { + "epoch": 0.34946236559139787, + "grad_norm": 0.4093106514843705, + "learning_rate": 0.00016443393904980242, + "loss": 0.2898, + "step": 1170 + }, + { + "epoch": 0.35095579450418163, + "grad_norm": 0.4539044394175168, + "learning_rate": 0.00016403437611898282, + "loss": 0.2809, + "step": 1175 + }, + { + "epoch": 0.35244922341696533, + "grad_norm": 0.36797578012073034, + "learning_rate": 0.0001636330727696688, + "loss": 0.2529, + "step": 1180 + }, + { + "epoch": 0.3539426523297491, + "grad_norm": 0.3182084185670439, + "learning_rate": 0.00016323003990906072, + "loss": 0.3142, + "step": 1185 + }, + { + "epoch": 0.35543608124253284, + "grad_norm": 0.33781054620192646, + "learning_rate": 0.00016282528849136612, + "loss": 0.2403, + "step": 1190 + }, + { + "epoch": 0.3569295101553166, + "grad_norm": 0.3880915127154374, + "learning_rate": 0.0001624188295175019, + "loss": 0.2999, + "step": 1195 + }, + { + "epoch": 0.35842293906810035, + "grad_norm": 0.359774759483394, + "learning_rate": 0.00016201067403479543, + "loss": 0.3101, + "step": 1200 + }, + { + "epoch": 0.35842293906810035, + "eval_loss": 0.2727881968021393, + "eval_runtime": 672.0857, + "eval_samples_per_second": 5.952, + "eval_steps_per_second": 0.744, + "step": 1200 + }, + { + "epoch": 0.3599163679808841, + "grad_norm": 0.388508285355121, + "learning_rate": 0.0001616008331366843, + "loss": 0.2564, + "step": 1205 + }, + { + "epoch": 0.36140979689366787, + "grad_norm": 0.36408140890196194, + "learning_rate": 0.00016118931796241457, + "loss": 0.3252, + "step": 1210 + }, + { + "epoch": 0.3629032258064516, + "grad_norm": 0.41167492762052504, + "learning_rate": 0.0001607761396967384, + "loss": 0.26, + "step": 1215 + }, + { + "epoch": 0.3643966547192354, + "grad_norm": 0.5248867549730605, + "learning_rate": 0.00016036130956960967, + "loss": 0.2655, + "step": 1220 + }, + { + "epoch": 0.36589008363201914, + "grad_norm": 0.49678835879333444, + "learning_rate": 0.00015994483885587902, + "loss": 0.3085, + "step": 1225 + }, + { + "epoch": 0.3673835125448029, + "grad_norm": 0.3988902390012257, + "learning_rate": 0.0001595267388749873, + "loss": 0.2607, + "step": 1230 + }, + { + "epoch": 0.3688769414575866, + "grad_norm": 0.3925797006591169, + "learning_rate": 0.0001591070209906579, + "loss": 0.2997, + "step": 1235 + }, + { + "epoch": 0.37037037037037035, + "grad_norm": 0.3107954625386247, + "learning_rate": 0.000158685696610588, + "loss": 0.2537, + "step": 1240 + }, + { + "epoch": 0.3718637992831541, + "grad_norm": 0.41712837863937413, + "learning_rate": 0.00015826277718613833, + "loss": 0.3003, + "step": 1245 + }, + { + "epoch": 0.37335722819593786, + "grad_norm": 0.3630150573236221, + "learning_rate": 0.00015783827421202214, + "loss": 0.3056, + "step": 1250 + }, + { + "epoch": 0.3748506571087216, + "grad_norm": 0.38258678014490427, + "learning_rate": 0.00015741219922599253, + "loss": 0.2641, + "step": 1255 + }, + { + "epoch": 0.3763440860215054, + "grad_norm": 0.42701857998693166, + "learning_rate": 0.00015698456380852915, + "loss": 0.3068, + "step": 1260 + }, + { + "epoch": 0.37783751493428913, + "grad_norm": 0.4380649167022928, + "learning_rate": 0.00015655537958252324, + "loss": 0.2576, + "step": 1265 + }, + { + "epoch": 0.3793309438470729, + "grad_norm": 0.4082068880960036, + "learning_rate": 0.00015612465821296175, + "loss": 0.2953, + "step": 1270 + }, + { + "epoch": 0.38082437275985664, + "grad_norm": 0.4190875138856109, + "learning_rate": 0.00015569241140661047, + "loss": 0.3032, + "step": 1275 + }, + { + "epoch": 0.3823178016726404, + "grad_norm": 0.36568309245691916, + "learning_rate": 0.0001552586509116955, + "loss": 0.2195, + "step": 1280 + }, + { + "epoch": 0.38381123058542416, + "grad_norm": 0.44216085615223133, + "learning_rate": 0.0001548233885175843, + "loss": 0.3073, + "step": 1285 + }, + { + "epoch": 0.38530465949820786, + "grad_norm": 0.24922873576347065, + "learning_rate": 0.00015438663605446507, + "loss": 0.2472, + "step": 1290 + }, + { + "epoch": 0.3867980884109916, + "grad_norm": 0.41662099994810825, + "learning_rate": 0.00015394840539302527, + "loss": 0.2868, + "step": 1295 + }, + { + "epoch": 0.38829151732377537, + "grad_norm": 0.39646098577140865, + "learning_rate": 0.0001535087084441289, + "loss": 0.2855, + "step": 1300 + }, + { + "epoch": 0.3897849462365591, + "grad_norm": 0.36239571418608835, + "learning_rate": 0.00015306755715849293, + "loss": 0.227, + "step": 1305 + }, + { + "epoch": 0.3912783751493429, + "grad_norm": 0.35793958618986693, + "learning_rate": 0.0001526249635263623, + "loss": 0.2888, + "step": 1310 + }, + { + "epoch": 0.39277180406212664, + "grad_norm": 0.3783233493544514, + "learning_rate": 0.00015218093957718415, + "loss": 0.2605, + "step": 1315 + }, + { + "epoch": 0.3942652329749104, + "grad_norm": 0.37987257135077446, + "learning_rate": 0.00015173549737928084, + "loss": 0.2854, + "step": 1320 + }, + { + "epoch": 0.39575866188769415, + "grad_norm": 0.3892605729869642, + "learning_rate": 0.0001512886490395219, + "loss": 0.2918, + "step": 1325 + }, + { + "epoch": 0.3972520908004779, + "grad_norm": 0.41196692553982217, + "learning_rate": 0.00015084040670299516, + "loss": 0.2557, + "step": 1330 + }, + { + "epoch": 0.39874551971326166, + "grad_norm": 0.3254462520604089, + "learning_rate": 0.00015039078255267628, + "loss": 0.3088, + "step": 1335 + }, + { + "epoch": 0.4002389486260454, + "grad_norm": 0.40781344675681797, + "learning_rate": 0.00014993978880909796, + "loss": 0.2805, + "step": 1340 + }, + { + "epoch": 0.4017323775388292, + "grad_norm": 0.39304465117259413, + "learning_rate": 0.00014948743773001772, + "loss": 0.2843, + "step": 1345 + }, + { + "epoch": 0.4032258064516129, + "grad_norm": 0.48556741767644995, + "learning_rate": 0.00014903374161008464, + "loss": 0.3051, + "step": 1350 + }, + { + "epoch": 0.40471923536439663, + "grad_norm": 0.3663553575319031, + "learning_rate": 0.0001485787127805052, + "loss": 0.2336, + "step": 1355 + }, + { + "epoch": 0.4062126642771804, + "grad_norm": 0.4541411950152405, + "learning_rate": 0.00014812236360870834, + "loss": 0.2854, + "step": 1360 + }, + { + "epoch": 0.40770609318996415, + "grad_norm": 0.34795001206833953, + "learning_rate": 0.00014766470649800904, + "loss": 0.26, + "step": 1365 + }, + { + "epoch": 0.4091995221027479, + "grad_norm": 0.35146295322347754, + "learning_rate": 0.00014720575388727132, + "loss": 0.266, + "step": 1370 + }, + { + "epoch": 0.41069295101553166, + "grad_norm": 0.44633677983363984, + "learning_rate": 0.00014674551825057024, + "loss": 0.2853, + "step": 1375 + }, + { + "epoch": 0.4121863799283154, + "grad_norm": 0.3830303196674387, + "learning_rate": 0.0001462840120968527, + "loss": 0.2515, + "step": 1380 + }, + { + "epoch": 0.4136798088410992, + "grad_norm": 0.39604955470206826, + "learning_rate": 0.00014582124796959765, + "loss": 0.2751, + "step": 1385 + }, + { + "epoch": 0.41517323775388293, + "grad_norm": 0.3633797549153069, + "learning_rate": 0.00014535723844647503, + "loss": 0.2531, + "step": 1390 + }, + { + "epoch": 0.4166666666666667, + "grad_norm": 0.6020608532163392, + "learning_rate": 0.00014489199613900385, + "loss": 0.2766, + "step": 1395 + }, + { + "epoch": 0.41816009557945044, + "grad_norm": 0.38363605454823274, + "learning_rate": 0.0001444255336922095, + "loss": 0.2913, + "step": 1400 + }, + { + "epoch": 0.41816009557945044, + "eval_loss": 0.2678522765636444, + "eval_runtime": 671.8286, + "eval_samples_per_second": 5.954, + "eval_steps_per_second": 0.744, + "step": 1400 + }, + { + "epoch": 0.41965352449223414, + "grad_norm": 0.34726800590023943, + "learning_rate": 0.00014395786378428033, + "loss": 0.2308, + "step": 1405 + }, + { + "epoch": 0.4211469534050179, + "grad_norm": 0.3445589635670954, + "learning_rate": 0.00014348899912622248, + "loss": 0.306, + "step": 1410 + }, + { + "epoch": 0.42264038231780165, + "grad_norm": 0.3573121569173274, + "learning_rate": 0.0001430189524615149, + "loss": 0.2669, + "step": 1415 + }, + { + "epoch": 0.4241338112305854, + "grad_norm": 0.40102571413373606, + "learning_rate": 0.0001425477365657628, + "loss": 0.2701, + "step": 1420 + }, + { + "epoch": 0.42562724014336917, + "grad_norm": 0.3591108084356434, + "learning_rate": 0.0001420753642463504, + "loss": 0.2866, + "step": 1425 + }, + { + "epoch": 0.4271206690561529, + "grad_norm": 0.38049883203567986, + "learning_rate": 0.00014160184834209296, + "loss": 0.267, + "step": 1430 + }, + { + "epoch": 0.4286140979689367, + "grad_norm": 0.34901096097918405, + "learning_rate": 0.0001411272017228876, + "loss": 0.2772, + "step": 1435 + }, + { + "epoch": 0.43010752688172044, + "grad_norm": 0.43110083401740446, + "learning_rate": 0.0001406514372893637, + "loss": 0.2773, + "step": 1440 + }, + { + "epoch": 0.4316009557945042, + "grad_norm": 0.49928896541185935, + "learning_rate": 0.0001401745679725323, + "loss": 0.3121, + "step": 1445 + }, + { + "epoch": 0.43309438470728795, + "grad_norm": 0.38753869075919367, + "learning_rate": 0.00013969660673343435, + "loss": 0.2809, + "step": 1450 + }, + { + "epoch": 0.4345878136200717, + "grad_norm": 0.431047229984796, + "learning_rate": 0.00013921756656278884, + "loss": 0.246, + "step": 1455 + }, + { + "epoch": 0.43608124253285546, + "grad_norm": 0.33787432704790127, + "learning_rate": 0.00013873746048063943, + "loss": 0.2905, + "step": 1460 + }, + { + "epoch": 0.43757467144563916, + "grad_norm": 0.5450443166622904, + "learning_rate": 0.00013825630153600058, + "loss": 0.2439, + "step": 1465 + }, + { + "epoch": 0.4390681003584229, + "grad_norm": 0.46991031442183334, + "learning_rate": 0.0001377741028065031, + "loss": 0.2916, + "step": 1470 + }, + { + "epoch": 0.4405615292712067, + "grad_norm": 0.43373701568669226, + "learning_rate": 0.00013729087739803846, + "loss": 0.3018, + "step": 1475 + }, + { + "epoch": 0.44205495818399043, + "grad_norm": 0.41610864937340164, + "learning_rate": 0.0001368066384444028, + "loss": 0.2252, + "step": 1480 + }, + { + "epoch": 0.4435483870967742, + "grad_norm": 0.3708263427774675, + "learning_rate": 0.0001363213991069397, + "loss": 0.2644, + "step": 1485 + }, + { + "epoch": 0.44504181600955794, + "grad_norm": 0.37850068431288725, + "learning_rate": 0.00013583517257418278, + "loss": 0.2641, + "step": 1490 + }, + { + "epoch": 0.4465352449223417, + "grad_norm": 0.4878849200428849, + "learning_rate": 0.000135347972061497, + "loss": 0.2722, + "step": 1495 + }, + { + "epoch": 0.44802867383512546, + "grad_norm": 0.3890166450566535, + "learning_rate": 0.00013485981081071949, + "loss": 0.2876, + "step": 1500 + }, + { + "epoch": 0.4495221027479092, + "grad_norm": 0.4698494336715817, + "learning_rate": 0.00013437070208979974, + "loss": 0.2618, + "step": 1505 + }, + { + "epoch": 0.45101553166069297, + "grad_norm": 0.42945392994548376, + "learning_rate": 0.000133880659192439, + "loss": 0.2887, + "step": 1510 + }, + { + "epoch": 0.4525089605734767, + "grad_norm": 0.36608609898718975, + "learning_rate": 0.00013338969543772892, + "loss": 0.2309, + "step": 1515 + }, + { + "epoch": 0.4540023894862604, + "grad_norm": 0.31719911951791674, + "learning_rate": 0.00013289782416978947, + "loss": 0.2669, + "step": 1520 + }, + { + "epoch": 0.4554958183990442, + "grad_norm": 0.44642248141095303, + "learning_rate": 0.0001324050587574063, + "loss": 0.2638, + "step": 1525 + }, + { + "epoch": 0.45698924731182794, + "grad_norm": 0.4094419878172264, + "learning_rate": 0.0001319114125936675, + "loss": 0.2647, + "step": 1530 + }, + { + "epoch": 0.4584826762246117, + "grad_norm": 0.48369368762161424, + "learning_rate": 0.00013141689909559943, + "loss": 0.298, + "step": 1535 + }, + { + "epoch": 0.45997610513739545, + "grad_norm": 0.4137701154348147, + "learning_rate": 0.00013092153170380212, + "loss": 0.2706, + "step": 1540 + }, + { + "epoch": 0.4614695340501792, + "grad_norm": 0.4125118524580438, + "learning_rate": 0.00013042532388208398, + "loss": 0.2769, + "step": 1545 + }, + { + "epoch": 0.46296296296296297, + "grad_norm": 0.3835331195333388, + "learning_rate": 0.0001299282891170958, + "loss": 0.29, + "step": 1550 + }, + { + "epoch": 0.4644563918757467, + "grad_norm": 0.4050806911084048, + "learning_rate": 0.00012943044091796418, + "loss": 0.2733, + "step": 1555 + }, + { + "epoch": 0.4659498207885305, + "grad_norm": 0.3686929748618163, + "learning_rate": 0.00012893179281592453, + "loss": 0.299, + "step": 1560 + }, + { + "epoch": 0.46744324970131423, + "grad_norm": 0.3713013220008561, + "learning_rate": 0.0001284323583639531, + "loss": 0.2514, + "step": 1565 + }, + { + "epoch": 0.468936678614098, + "grad_norm": 0.37363712594365256, + "learning_rate": 0.00012793215113639862, + "loss": 0.278, + "step": 1570 + }, + { + "epoch": 0.47043010752688175, + "grad_norm": 0.3759963829551066, + "learning_rate": 0.00012743118472861365, + "loss": 0.2819, + "step": 1575 + }, + { + "epoch": 0.47192353643966545, + "grad_norm": 0.38520117361685985, + "learning_rate": 0.00012692947275658455, + "loss": 0.2345, + "step": 1580 + }, + { + "epoch": 0.4734169653524492, + "grad_norm": 0.3075498245821686, + "learning_rate": 0.00012642702885656192, + "loss": 0.263, + "step": 1585 + }, + { + "epoch": 0.47491039426523296, + "grad_norm": 0.3841403461793403, + "learning_rate": 0.00012592386668468968, + "loss": 0.2913, + "step": 1590 + }, + { + "epoch": 0.4764038231780167, + "grad_norm": 0.3367807617111592, + "learning_rate": 0.00012541999991663388, + "loss": 0.2801, + "step": 1595 + }, + { + "epoch": 0.4778972520908005, + "grad_norm": 0.378016075471954, + "learning_rate": 0.00012491544224721136, + "loss": 0.2765, + "step": 1600 + }, + { + "epoch": 0.4778972520908005, + "eval_loss": 0.2625010311603546, + "eval_runtime": 671.1626, + "eval_samples_per_second": 5.96, + "eval_steps_per_second": 0.745, + "step": 1600 + }, + { + "epoch": 0.47939068100358423, + "grad_norm": 0.3691633610486676, + "learning_rate": 0.00012441020739001698, + "loss": 0.2793, + "step": 1605 + }, + { + "epoch": 0.480884109916368, + "grad_norm": 0.3264004788515136, + "learning_rate": 0.00012390430907705134, + "loss": 0.2683, + "step": 1610 + }, + { + "epoch": 0.48237753882915174, + "grad_norm": 0.35913080769203726, + "learning_rate": 0.00012339776105834744, + "loss": 0.229, + "step": 1615 + }, + { + "epoch": 0.4838709677419355, + "grad_norm": 0.45935643829710515, + "learning_rate": 0.00012289057710159672, + "loss": 0.2714, + "step": 1620 + }, + { + "epoch": 0.48536439665471925, + "grad_norm": 0.3879403177186618, + "learning_rate": 0.00012238277099177526, + "loss": 0.2829, + "step": 1625 + }, + { + "epoch": 0.486857825567503, + "grad_norm": 0.3514395181766118, + "learning_rate": 0.00012187435653076889, + "loss": 0.249, + "step": 1630 + }, + { + "epoch": 0.4883512544802867, + "grad_norm": 0.3914277195367492, + "learning_rate": 0.0001213653475369979, + "loss": 0.2862, + "step": 1635 + }, + { + "epoch": 0.48984468339307047, + "grad_norm": 0.468390381134599, + "learning_rate": 0.00012085575784504191, + "loss": 0.2307, + "step": 1640 + }, + { + "epoch": 0.4913381123058542, + "grad_norm": 0.42676427345496687, + "learning_rate": 0.0001203456013052634, + "loss": 0.2963, + "step": 1645 + }, + { + "epoch": 0.492831541218638, + "grad_norm": 0.4042777140715105, + "learning_rate": 0.0001198348917834315, + "loss": 0.3155, + "step": 1650 + }, + { + "epoch": 0.49432497013142174, + "grad_norm": 0.333904149664281, + "learning_rate": 0.00011932364316034514, + "loss": 0.2433, + "step": 1655 + }, + { + "epoch": 0.4958183990442055, + "grad_norm": 0.41506264824806743, + "learning_rate": 0.0001188118693314557, + "loss": 0.2806, + "step": 1660 + }, + { + "epoch": 0.49731182795698925, + "grad_norm": 0.4293221243565824, + "learning_rate": 0.0001182995842064893, + "loss": 0.2582, + "step": 1665 + }, + { + "epoch": 0.498805256869773, + "grad_norm": 0.39173839921627196, + "learning_rate": 0.00011778680170906888, + "loss": 0.2886, + "step": 1670 + }, + { + "epoch": 0.5002986857825568, + "grad_norm": 0.47362247526518175, + "learning_rate": 0.0001172735357763356, + "loss": 0.293, + "step": 1675 + }, + { + "epoch": 0.5017921146953405, + "grad_norm": 0.4176472433369269, + "learning_rate": 0.00011675980035857019, + "loss": 0.2091, + "step": 1680 + }, + { + "epoch": 0.5032855436081243, + "grad_norm": 0.39087371326818915, + "learning_rate": 0.00011624560941881371, + "loss": 0.3164, + "step": 1685 + }, + { + "epoch": 0.504778972520908, + "grad_norm": 0.3444686695813055, + "learning_rate": 0.00011573097693248805, + "loss": 0.2601, + "step": 1690 + }, + { + "epoch": 0.5062724014336918, + "grad_norm": 0.3629126474420232, + "learning_rate": 0.00011521591688701605, + "loss": 0.2677, + "step": 1695 + }, + { + "epoch": 0.5077658303464755, + "grad_norm": 0.3677730257235259, + "learning_rate": 0.00011470044328144143, + "loss": 0.2845, + "step": 1700 + }, + { + "epoch": 0.5092592592592593, + "grad_norm": 0.4484524597840905, + "learning_rate": 0.0001141845701260482, + "loss": 0.2535, + "step": 1705 + }, + { + "epoch": 0.510752688172043, + "grad_norm": 0.44520906295668633, + "learning_rate": 0.0001136683114419799, + "loss": 0.2923, + "step": 1710 + }, + { + "epoch": 0.5122461170848268, + "grad_norm": 0.3843678110888455, + "learning_rate": 0.00011315168126085857, + "loss": 0.2419, + "step": 1715 + }, + { + "epoch": 0.5137395459976105, + "grad_norm": 0.3424866175556353, + "learning_rate": 0.00011263469362440331, + "loss": 0.2722, + "step": 1720 + }, + { + "epoch": 0.5152329749103942, + "grad_norm": 0.3664503507774818, + "learning_rate": 0.00011211736258404864, + "loss": 0.2652, + "step": 1725 + }, + { + "epoch": 0.516726403823178, + "grad_norm": 0.3882647311598004, + "learning_rate": 0.00011159970220056262, + "loss": 0.2613, + "step": 1730 + }, + { + "epoch": 0.5182198327359617, + "grad_norm": 0.3934259702377591, + "learning_rate": 0.00011108172654366467, + "loss": 0.2744, + "step": 1735 + }, + { + "epoch": 0.5197132616487455, + "grad_norm": 0.38348245744708515, + "learning_rate": 0.00011056344969164317, + "loss": 0.2484, + "step": 1740 + }, + { + "epoch": 0.5212066905615292, + "grad_norm": 0.4552854124473697, + "learning_rate": 0.0001100448857309728, + "loss": 0.2559, + "step": 1745 + }, + { + "epoch": 0.522700119474313, + "grad_norm": 0.39439256103140385, + "learning_rate": 0.00010952604875593171, + "loss": 0.2724, + "step": 1750 + }, + { + "epoch": 0.5241935483870968, + "grad_norm": 0.42701555614159337, + "learning_rate": 0.00010900695286821843, + "loss": 0.2284, + "step": 1755 + }, + { + "epoch": 0.5256869772998806, + "grad_norm": 0.3685725297263498, + "learning_rate": 0.00010848761217656856, + "loss": 0.2897, + "step": 1760 + }, + { + "epoch": 0.5271804062126643, + "grad_norm": 0.40798420628174437, + "learning_rate": 0.00010796804079637137, + "loss": 0.2689, + "step": 1765 + }, + { + "epoch": 0.5286738351254481, + "grad_norm": 0.4141834988836477, + "learning_rate": 0.0001074482528492861, + "loss": 0.2833, + "step": 1770 + }, + { + "epoch": 0.5301672640382318, + "grad_norm": 0.391053467201034, + "learning_rate": 0.00010692826246285815, + "loss": 0.2783, + "step": 1775 + }, + { + "epoch": 0.5316606929510155, + "grad_norm": 0.3660707237272847, + "learning_rate": 0.00010640808377013509, + "loss": 0.2504, + "step": 1780 + }, + { + "epoch": 0.5331541218637993, + "grad_norm": 0.29960550920174905, + "learning_rate": 0.00010588773090928268, + "loss": 0.2773, + "step": 1785 + }, + { + "epoch": 0.534647550776583, + "grad_norm": 0.49371001212196336, + "learning_rate": 0.00010536721802320027, + "loss": 0.2543, + "step": 1790 + }, + { + "epoch": 0.5361409796893668, + "grad_norm": 0.381676599509551, + "learning_rate": 0.00010484655925913669, + "loss": 0.3047, + "step": 1795 + }, + { + "epoch": 0.5376344086021505, + "grad_norm": 0.4165684073566037, + "learning_rate": 0.00010432576876830576, + "loss": 0.2697, + "step": 1800 + }, + { + "epoch": 0.5376344086021505, + "eval_loss": 0.2601146101951599, + "eval_runtime": 671.0756, + "eval_samples_per_second": 5.961, + "eval_steps_per_second": 0.745, + "step": 1800 + }, + { + "epoch": 0.5391278375149343, + "grad_norm": 0.34164713060339763, + "learning_rate": 0.00010380486070550135, + "loss": 0.2613, + "step": 1805 + }, + { + "epoch": 0.540621266427718, + "grad_norm": 0.3608068205239928, + "learning_rate": 0.00010328384922871307, + "loss": 0.3023, + "step": 1810 + }, + { + "epoch": 0.5421146953405018, + "grad_norm": 0.4134220541761705, + "learning_rate": 0.0001027627484987412, + "loss": 0.2759, + "step": 1815 + }, + { + "epoch": 0.5436081242532855, + "grad_norm": 0.4380879213819027, + "learning_rate": 0.00010224157267881176, + "loss": 0.2999, + "step": 1820 + }, + { + "epoch": 0.5451015531660693, + "grad_norm": 0.41334705402523975, + "learning_rate": 0.00010172033593419184, + "loss": 0.2905, + "step": 1825 + }, + { + "epoch": 0.546594982078853, + "grad_norm": 0.33573837522797895, + "learning_rate": 0.00010119905243180432, + "loss": 0.2197, + "step": 1830 + }, + { + "epoch": 0.5480884109916367, + "grad_norm": 0.34530007854548944, + "learning_rate": 0.00010067773633984294, + "loss": 0.2838, + "step": 1835 + }, + { + "epoch": 0.5495818399044206, + "grad_norm": 0.39282261233632176, + "learning_rate": 0.00010015640182738733, + "loss": 0.2449, + "step": 1840 + }, + { + "epoch": 0.5510752688172043, + "grad_norm": 0.369443654894832, + "learning_rate": 9.963506306401753e-05, + "loss": 0.2641, + "step": 1845 + }, + { + "epoch": 0.5525686977299881, + "grad_norm": 0.3736606416875578, + "learning_rate": 9.911373421942939e-05, + "loss": 0.2877, + "step": 1850 + }, + { + "epoch": 0.5540621266427718, + "grad_norm": 0.4082824771004434, + "learning_rate": 9.859242946304903e-05, + "loss": 0.2462, + "step": 1855 + }, + { + "epoch": 0.5555555555555556, + "grad_norm": 0.37234188557058073, + "learning_rate": 9.807116296364783e-05, + "loss": 0.2647, + "step": 1860 + }, + { + "epoch": 0.5570489844683393, + "grad_norm": 0.3373920413107433, + "learning_rate": 9.754994888895744e-05, + "loss": 0.2481, + "step": 1865 + }, + { + "epoch": 0.5585424133811231, + "grad_norm": 0.3691166711599759, + "learning_rate": 9.702880140528466e-05, + "loss": 0.2651, + "step": 1870 + }, + { + "epoch": 0.5600358422939068, + "grad_norm": 0.4237027962737052, + "learning_rate": 9.650773467712632e-05, + "loss": 0.2976, + "step": 1875 + }, + { + "epoch": 0.5615292712066906, + "grad_norm": 0.3628296244980769, + "learning_rate": 9.598676286678434e-05, + "loss": 0.233, + "step": 1880 + }, + { + "epoch": 0.5630227001194743, + "grad_norm": 0.32168405334156497, + "learning_rate": 9.546590013398091e-05, + "loss": 0.2875, + "step": 1885 + }, + { + "epoch": 0.5645161290322581, + "grad_norm": 0.4011755917649591, + "learning_rate": 9.494516063547353e-05, + "loss": 0.2556, + "step": 1890 + }, + { + "epoch": 0.5660095579450418, + "grad_norm": 0.4531648719387392, + "learning_rate": 9.442455852467019e-05, + "loss": 0.2674, + "step": 1895 + }, + { + "epoch": 0.5675029868578255, + "grad_norm": 0.4603202269611572, + "learning_rate": 9.390410795124484e-05, + "loss": 0.2902, + "step": 1900 + }, + { + "epoch": 0.5689964157706093, + "grad_norm": 0.3873591567262209, + "learning_rate": 9.338382306075274e-05, + "loss": 0.2272, + "step": 1905 + }, + { + "epoch": 0.570489844683393, + "grad_norm": 0.41809794251121246, + "learning_rate": 9.286371799424584e-05, + "loss": 0.309, + "step": 1910 + }, + { + "epoch": 0.5719832735961768, + "grad_norm": 0.41735588072251223, + "learning_rate": 9.234380688788879e-05, + "loss": 0.2422, + "step": 1915 + }, + { + "epoch": 0.5734767025089605, + "grad_norm": 0.35987074653242784, + "learning_rate": 9.182410387257434e-05, + "loss": 0.2868, + "step": 1920 + }, + { + "epoch": 0.5749701314217444, + "grad_norm": 0.3381000166870348, + "learning_rate": 9.130462307353945e-05, + "loss": 0.2794, + "step": 1925 + }, + { + "epoch": 0.5764635603345281, + "grad_norm": 0.3412056255088943, + "learning_rate": 9.078537860998155e-05, + "loss": 0.2391, + "step": 1930 + }, + { + "epoch": 0.5779569892473119, + "grad_norm": 0.40246570101157014, + "learning_rate": 9.026638459467435e-05, + "loss": 0.2757, + "step": 1935 + }, + { + "epoch": 0.5794504181600956, + "grad_norm": 0.29830016511773766, + "learning_rate": 8.974765513358466e-05, + "loss": 0.2403, + "step": 1940 + }, + { + "epoch": 0.5809438470728794, + "grad_norm": 0.3411808872796586, + "learning_rate": 8.92292043254889e-05, + "loss": 0.2708, + "step": 1945 + }, + { + "epoch": 0.5824372759856631, + "grad_norm": 0.4133927108399512, + "learning_rate": 8.871104626158972e-05, + "loss": 0.2873, + "step": 1950 + }, + { + "epoch": 0.5839307048984468, + "grad_norm": 0.39290137528322705, + "learning_rate": 8.81931950251332e-05, + "loss": 0.2103, + "step": 1955 + }, + { + "epoch": 0.5854241338112306, + "grad_norm": 0.45501182269960977, + "learning_rate": 8.767566469102613e-05, + "loss": 0.2781, + "step": 1960 + }, + { + "epoch": 0.5869175627240143, + "grad_norm": 0.4189475808727597, + "learning_rate": 8.715846932545317e-05, + "loss": 0.2378, + "step": 1965 + }, + { + "epoch": 0.5884109916367981, + "grad_norm": 0.4085112647747901, + "learning_rate": 8.664162298549483e-05, + "loss": 0.2806, + "step": 1970 + }, + { + "epoch": 0.5899044205495818, + "grad_norm": 0.3732596924042426, + "learning_rate": 8.612513971874534e-05, + "loss": 0.2932, + "step": 1975 + }, + { + "epoch": 0.5913978494623656, + "grad_norm": 0.35250262119931725, + "learning_rate": 8.560903356293069e-05, + "loss": 0.2418, + "step": 1980 + }, + { + "epoch": 0.5928912783751493, + "grad_norm": 0.41366159645877715, + "learning_rate": 8.509331854552724e-05, + "loss": 0.2743, + "step": 1985 + }, + { + "epoch": 0.5943847072879331, + "grad_norm": 0.3464575423602374, + "learning_rate": 8.457800868338051e-05, + "loss": 0.2222, + "step": 1990 + }, + { + "epoch": 0.5958781362007168, + "grad_norm": 0.360808643210085, + "learning_rate": 8.406311798232408e-05, + "loss": 0.2625, + "step": 1995 + }, + { + "epoch": 0.5973715651135006, + "grad_norm": 0.4208234863436116, + "learning_rate": 8.354866043679887e-05, + "loss": 0.2759, + "step": 2000 + }, + { + "epoch": 0.5973715651135006, + "eval_loss": 0.25568902492523193, + "eval_runtime": 670.9224, + "eval_samples_per_second": 5.962, + "eval_steps_per_second": 0.745, + "step": 2000 + }, + { + "epoch": 0.5988649940262843, + "grad_norm": 0.5128576614444417, + "learning_rate": 8.303465002947302e-05, + "loss": 0.2265, + "step": 2005 + }, + { + "epoch": 0.600358422939068, + "grad_norm": 0.4246792251104683, + "learning_rate": 8.252110073086165e-05, + "loss": 0.2591, + "step": 2010 + }, + { + "epoch": 0.6018518518518519, + "grad_norm": 0.36587149536697655, + "learning_rate": 8.200802649894713e-05, + "loss": 0.2598, + "step": 2015 + }, + { + "epoch": 0.6033452807646356, + "grad_norm": 0.3560153774291601, + "learning_rate": 8.149544127879988e-05, + "loss": 0.26, + "step": 2020 + }, + { + "epoch": 0.6048387096774194, + "grad_norm": 0.3820268167566159, + "learning_rate": 8.098335900219929e-05, + "loss": 0.2812, + "step": 2025 + }, + { + "epoch": 0.6063321385902031, + "grad_norm": 0.4670838320418128, + "learning_rate": 8.047179358725487e-05, + "loss": 0.2398, + "step": 2030 + }, + { + "epoch": 0.6078255675029869, + "grad_norm": 0.4359410281219078, + "learning_rate": 7.99607589380282e-05, + "loss": 0.2973, + "step": 2035 + }, + { + "epoch": 0.6093189964157706, + "grad_norm": 0.3406561530367968, + "learning_rate": 7.945026894415504e-05, + "loss": 0.251, + "step": 2040 + }, + { + "epoch": 0.6108124253285544, + "grad_norm": 0.34818876087607525, + "learning_rate": 7.894033748046756e-05, + "loss": 0.2691, + "step": 2045 + }, + { + "epoch": 0.6123058542413381, + "grad_norm": 0.39942001844135094, + "learning_rate": 7.843097840661748e-05, + "loss": 0.2879, + "step": 2050 + }, + { + "epoch": 0.6137992831541219, + "grad_norm": 0.3618592441449212, + "learning_rate": 7.792220556669929e-05, + "loss": 0.223, + "step": 2055 + }, + { + "epoch": 0.6152927120669056, + "grad_norm": 0.4022080561354502, + "learning_rate": 7.741403278887397e-05, + "loss": 0.2785, + "step": 2060 + }, + { + "epoch": 0.6167861409796893, + "grad_norm": 0.26523185523371673, + "learning_rate": 7.690647388499305e-05, + "loss": 0.2482, + "step": 2065 + }, + { + "epoch": 0.6182795698924731, + "grad_norm": 0.33411219118251206, + "learning_rate": 7.639954265022349e-05, + "loss": 0.2703, + "step": 2070 + }, + { + "epoch": 0.6197729988052568, + "grad_norm": 0.35999133636969094, + "learning_rate": 7.589325286267241e-05, + "loss": 0.262, + "step": 2075 + }, + { + "epoch": 0.6212664277180406, + "grad_norm": 0.3853527518287356, + "learning_rate": 7.538761828301276e-05, + "loss": 0.2261, + "step": 2080 + }, + { + "epoch": 0.6227598566308243, + "grad_norm": 0.43335074980788474, + "learning_rate": 7.48826526541094e-05, + "loss": 0.254, + "step": 2085 + }, + { + "epoch": 0.6242532855436081, + "grad_norm": 0.3542418988641108, + "learning_rate": 7.437836970064545e-05, + "loss": 0.2334, + "step": 2090 + }, + { + "epoch": 0.6257467144563919, + "grad_norm": 0.398192032214991, + "learning_rate": 7.387478312874916e-05, + "loss": 0.2742, + "step": 2095 + }, + { + "epoch": 0.6272401433691757, + "grad_norm": 0.33547770368905083, + "learning_rate": 7.337190662562174e-05, + "loss": 0.2511, + "step": 2100 + }, + { + "epoch": 0.6287335722819594, + "grad_norm": 0.4179386434535002, + "learning_rate": 7.2869753859165e-05, + "loss": 0.2667, + "step": 2105 + }, + { + "epoch": 0.6302270011947432, + "grad_norm": 0.36708628994003195, + "learning_rate": 7.236833847760994e-05, + "loss": 0.2805, + "step": 2110 + }, + { + "epoch": 0.6317204301075269, + "grad_norm": 0.3128895549559772, + "learning_rate": 7.186767410914601e-05, + "loss": 0.2477, + "step": 2115 + }, + { + "epoch": 0.6332138590203107, + "grad_norm": 0.3308033239038678, + "learning_rate": 7.136777436155041e-05, + "loss": 0.2559, + "step": 2120 + }, + { + "epoch": 0.6347072879330944, + "grad_norm": 0.36157519220837353, + "learning_rate": 7.086865282181841e-05, + "loss": 0.2747, + "step": 2125 + }, + { + "epoch": 0.6362007168458781, + "grad_norm": 0.38491703355424195, + "learning_rate": 7.037032305579409e-05, + "loss": 0.202, + "step": 2130 + }, + { + "epoch": 0.6376941457586619, + "grad_norm": 0.29165681376515235, + "learning_rate": 6.987279860780148e-05, + "loss": 0.286, + "step": 2135 + }, + { + "epoch": 0.6391875746714456, + "grad_norm": 0.3691280781000132, + "learning_rate": 6.93760930002765e-05, + "loss": 0.2628, + "step": 2140 + }, + { + "epoch": 0.6406810035842294, + "grad_norm": 0.33721048070187576, + "learning_rate": 6.888021973339958e-05, + "loss": 0.2629, + "step": 2145 + }, + { + "epoch": 0.6421744324970131, + "grad_norm": 0.40230756965561154, + "learning_rate": 6.838519228472845e-05, + "loss": 0.2919, + "step": 2150 + }, + { + "epoch": 0.6436678614097969, + "grad_norm": 0.3832722542456181, + "learning_rate": 6.789102410883201e-05, + "loss": 0.2258, + "step": 2155 + }, + { + "epoch": 0.6451612903225806, + "grad_norm": 0.3679304149313773, + "learning_rate": 6.739772863692471e-05, + "loss": 0.2684, + "step": 2160 + }, + { + "epoch": 0.6466547192353644, + "grad_norm": 0.40756568956863426, + "learning_rate": 6.690531927650126e-05, + "loss": 0.2583, + "step": 2165 + }, + { + "epoch": 0.6481481481481481, + "grad_norm": 0.3980654356736509, + "learning_rate": 6.641380941097237e-05, + "loss": 0.26, + "step": 2170 + }, + { + "epoch": 0.649641577060932, + "grad_norm": 0.4297937372635352, + "learning_rate": 6.592321239930112e-05, + "loss": 0.2736, + "step": 2175 + }, + { + "epoch": 0.6511350059737157, + "grad_norm": 0.358991865502727, + "learning_rate": 6.54335415756396e-05, + "loss": 0.2458, + "step": 2180 + }, + { + "epoch": 0.6526284348864994, + "grad_norm": 0.35194743625956787, + "learning_rate": 6.49448102489666e-05, + "loss": 0.2731, + "step": 2185 + }, + { + "epoch": 0.6541218637992832, + "grad_norm": 0.3344682852694039, + "learning_rate": 6.445703170272603e-05, + "loss": 0.2392, + "step": 2190 + }, + { + "epoch": 0.6556152927120669, + "grad_norm": 0.36819816176424924, + "learning_rate": 6.39702191944657e-05, + "loss": 0.2742, + "step": 2195 + }, + { + "epoch": 0.6571087216248507, + "grad_norm": 0.41481385734175824, + "learning_rate": 6.348438595547705e-05, + "loss": 0.264, + "step": 2200 + }, + { + "epoch": 0.6571087216248507, + "eval_loss": 0.25236326456069946, + "eval_runtime": 670.9223, + "eval_samples_per_second": 5.962, + "eval_steps_per_second": 0.745, + "step": 2200 + }, + { + "epoch": 0.6586021505376344, + "grad_norm": 0.37828046273666066, + "learning_rate": 6.299954519043552e-05, + "loss": 0.2045, + "step": 2205 + }, + { + "epoch": 0.6600955794504182, + "grad_norm": 0.3647111671687454, + "learning_rate": 6.25157100770417e-05, + "loss": 0.2757, + "step": 2210 + }, + { + "epoch": 0.6615890083632019, + "grad_norm": 0.333924367566417, + "learning_rate": 6.203289376566307e-05, + "loss": 0.2334, + "step": 2215 + }, + { + "epoch": 0.6630824372759857, + "grad_norm": 0.3731260608605994, + "learning_rate": 6.155110937897667e-05, + "loss": 0.2359, + "step": 2220 + }, + { + "epoch": 0.6645758661887694, + "grad_norm": 0.36399635762642046, + "learning_rate": 6.107037001161248e-05, + "loss": 0.2509, + "step": 2225 + }, + { + "epoch": 0.6660692951015532, + "grad_norm": 0.3990653767812247, + "learning_rate": 6.0590688729797295e-05, + "loss": 0.2521, + "step": 2230 + }, + { + "epoch": 0.6675627240143369, + "grad_norm": 0.3608936095400094, + "learning_rate": 6.011207857099983e-05, + "loss": 0.2543, + "step": 2235 + }, + { + "epoch": 0.6690561529271206, + "grad_norm": 0.31260618340047636, + "learning_rate": 5.963455254357631e-05, + "loss": 0.2318, + "step": 2240 + }, + { + "epoch": 0.6705495818399044, + "grad_norm": 0.4642758261006183, + "learning_rate": 5.9158123626416794e-05, + "loss": 0.2754, + "step": 2245 + }, + { + "epoch": 0.6720430107526881, + "grad_norm": 0.37892791122014974, + "learning_rate": 5.868280476859249e-05, + "loss": 0.2821, + "step": 2250 + }, + { + "epoch": 0.6735364396654719, + "grad_norm": 0.317919026225424, + "learning_rate": 5.820860888900392e-05, + "loss": 0.2021, + "step": 2255 + }, + { + "epoch": 0.6750298685782556, + "grad_norm": 0.3638464340699625, + "learning_rate": 5.7735548876029597e-05, + "loss": 0.2758, + "step": 2260 + }, + { + "epoch": 0.6765232974910395, + "grad_norm": 0.4089378874713291, + "learning_rate": 5.7263637587175836e-05, + "loss": 0.2319, + "step": 2265 + }, + { + "epoch": 0.6780167264038232, + "grad_norm": 0.36255601718231956, + "learning_rate": 5.679288784872727e-05, + "loss": 0.2736, + "step": 2270 + }, + { + "epoch": 0.679510155316607, + "grad_norm": 0.4450766858081928, + "learning_rate": 5.632331245539826e-05, + "loss": 0.2682, + "step": 2275 + }, + { + "epoch": 0.6810035842293907, + "grad_norm": 0.38190771534156615, + "learning_rate": 5.585492416998511e-05, + "loss": 0.2289, + "step": 2280 + }, + { + "epoch": 0.6824970131421745, + "grad_norm": 0.4405779247739025, + "learning_rate": 5.5387735723019207e-05, + "loss": 0.2928, + "step": 2285 + }, + { + "epoch": 0.6839904420549582, + "grad_norm": 0.3774472213099317, + "learning_rate": 5.492175981242097e-05, + "loss": 0.2514, + "step": 2290 + }, + { + "epoch": 0.6854838709677419, + "grad_norm": 0.3377801375518953, + "learning_rate": 5.445700910315481e-05, + "loss": 0.2655, + "step": 2295 + }, + { + "epoch": 0.6869772998805257, + "grad_norm": 0.39587982015232115, + "learning_rate": 5.399349622688479e-05, + "loss": 0.2946, + "step": 2300 + }, + { + "epoch": 0.6884707287933094, + "grad_norm": 0.43119763327491156, + "learning_rate": 5.353123378163143e-05, + "loss": 0.2407, + "step": 2305 + }, + { + "epoch": 0.6899641577060932, + "grad_norm": 0.36132290047057575, + "learning_rate": 5.307023433142919e-05, + "loss": 0.2807, + "step": 2310 + }, + { + "epoch": 0.6914575866188769, + "grad_norm": 0.38966391503202974, + "learning_rate": 5.2610510405985034e-05, + "loss": 0.2187, + "step": 2315 + }, + { + "epoch": 0.6929510155316607, + "grad_norm": 0.36997141842452064, + "learning_rate": 5.21520745003379e-05, + "loss": 0.2772, + "step": 2320 + }, + { + "epoch": 0.6944444444444444, + "grad_norm": 0.4830585130129818, + "learning_rate": 5.169493907451906e-05, + "loss": 0.2846, + "step": 2325 + }, + { + "epoch": 0.6959378733572282, + "grad_norm": 0.4663739708408958, + "learning_rate": 5.123911655321345e-05, + "loss": 0.2334, + "step": 2330 + }, + { + "epoch": 0.6974313022700119, + "grad_norm": 0.38317110709020197, + "learning_rate": 5.078461932542204e-05, + "loss": 0.2903, + "step": 2335 + }, + { + "epoch": 0.6989247311827957, + "grad_norm": 0.44204152745518926, + "learning_rate": 5.033145974412502e-05, + "loss": 0.2418, + "step": 2340 + }, + { + "epoch": 0.7004181600955794, + "grad_norm": 0.3562540524906317, + "learning_rate": 4.987965012594612e-05, + "loss": 0.2429, + "step": 2345 + }, + { + "epoch": 0.7019115890083633, + "grad_norm": 0.415882439063641, + "learning_rate": 4.942920275081784e-05, + "loss": 0.2716, + "step": 2350 + }, + { + "epoch": 0.703405017921147, + "grad_norm": 0.4349967578566546, + "learning_rate": 4.898012986164764e-05, + "loss": 0.227, + "step": 2355 + }, + { + "epoch": 0.7048984468339307, + "grad_norm": 0.38888281415830517, + "learning_rate": 4.853244366398524e-05, + "loss": 0.2727, + "step": 2360 + }, + { + "epoch": 0.7063918757467145, + "grad_norm": 0.48748379250803325, + "learning_rate": 4.808615632569089e-05, + "loss": 0.2237, + "step": 2365 + }, + { + "epoch": 0.7078853046594982, + "grad_norm": 0.33545458271479295, + "learning_rate": 4.764127997660457e-05, + "loss": 0.2512, + "step": 2370 + }, + { + "epoch": 0.709378733572282, + "grad_norm": 0.4038148334793521, + "learning_rate": 4.719782670821642e-05, + "loss": 0.2729, + "step": 2375 + }, + { + "epoch": 0.7108721624850657, + "grad_norm": 0.34079737026356866, + "learning_rate": 4.675580857333799e-05, + "loss": 0.2256, + "step": 2380 + }, + { + "epoch": 0.7123655913978495, + "grad_norm": 0.46179757394318344, + "learning_rate": 4.631523758577475e-05, + "loss": 0.2722, + "step": 2385 + }, + { + "epoch": 0.7138590203106332, + "grad_norm": 0.39463185516160715, + "learning_rate": 4.5876125719999474e-05, + "loss": 0.2632, + "step": 2390 + }, + { + "epoch": 0.715352449223417, + "grad_norm": 0.370808339928303, + "learning_rate": 4.543848491082687e-05, + "loss": 0.2381, + "step": 2395 + }, + { + "epoch": 0.7168458781362007, + "grad_norm": 0.3791124095428876, + "learning_rate": 4.500232705308911e-05, + "loss": 0.2705, + "step": 2400 + }, + { + "epoch": 0.7168458781362007, + "eval_loss": 0.24903884530067444, + "eval_runtime": 670.4879, + "eval_samples_per_second": 5.966, + "eval_steps_per_second": 0.746, + "step": 2400 + }, + { + "epoch": 0.7183393070489845, + "grad_norm": 0.3775218725937434, + "learning_rate": 4.45676640013126e-05, + "loss": 0.2268, + "step": 2405 + }, + { + "epoch": 0.7198327359617682, + "grad_norm": 0.3827463104057589, + "learning_rate": 4.413450756939574e-05, + "loss": 0.2536, + "step": 2410 + }, + { + "epoch": 0.7213261648745519, + "grad_norm": 0.3648532710176232, + "learning_rate": 4.370286953028787e-05, + "loss": 0.2504, + "step": 2415 + }, + { + "epoch": 0.7228195937873357, + "grad_norm": 0.3055219924416998, + "learning_rate": 4.32727616156692e-05, + "loss": 0.2324, + "step": 2420 + }, + { + "epoch": 0.7243130227001194, + "grad_norm": 0.4438604649063271, + "learning_rate": 4.2844195515632166e-05, + "loss": 0.2589, + "step": 2425 + }, + { + "epoch": 0.7258064516129032, + "grad_norm": 0.3795959005263556, + "learning_rate": 4.2417182878363336e-05, + "loss": 0.2476, + "step": 2430 + }, + { + "epoch": 0.727299880525687, + "grad_norm": 0.3427145661133711, + "learning_rate": 4.199173530982715e-05, + "loss": 0.2829, + "step": 2435 + }, + { + "epoch": 0.7287933094384708, + "grad_norm": 0.39445938386306895, + "learning_rate": 4.156786437345044e-05, + "loss": 0.2703, + "step": 2440 + }, + { + "epoch": 0.7302867383512545, + "grad_norm": 0.39923700769389914, + "learning_rate": 4.114558158980785e-05, + "loss": 0.285, + "step": 2445 + }, + { + "epoch": 0.7317801672640383, + "grad_norm": 0.4038674564871101, + "learning_rate": 4.072489843630903e-05, + "loss": 0.2814, + "step": 2450 + }, + { + "epoch": 0.733273596176822, + "grad_norm": 0.36737797848087894, + "learning_rate": 4.030582634688669e-05, + "loss": 0.2397, + "step": 2455 + }, + { + "epoch": 0.7347670250896058, + "grad_norm": 0.33021542644527924, + "learning_rate": 3.98883767116855e-05, + "loss": 0.2415, + "step": 2460 + }, + { + "epoch": 0.7362604540023895, + "grad_norm": 0.4047149773056364, + "learning_rate": 3.947256087675286e-05, + "loss": 0.2636, + "step": 2465 + }, + { + "epoch": 0.7377538829151732, + "grad_norm": 0.418965527019863, + "learning_rate": 3.905839014373047e-05, + "loss": 0.2818, + "step": 2470 + }, + { + "epoch": 0.739247311827957, + "grad_norm": 0.44764561078371773, + "learning_rate": 3.864587576954688e-05, + "loss": 0.2781, + "step": 2475 + }, + { + "epoch": 0.7407407407407407, + "grad_norm": 0.39101875860430557, + "learning_rate": 3.823502896611184e-05, + "loss": 0.2577, + "step": 2480 + }, + { + "epoch": 0.7422341696535245, + "grad_norm": 0.3922118709138838, + "learning_rate": 3.782586090001155e-05, + "loss": 0.2563, + "step": 2485 + }, + { + "epoch": 0.7437275985663082, + "grad_norm": 0.379882936966773, + "learning_rate": 3.741838269220496e-05, + "loss": 0.2517, + "step": 2490 + }, + { + "epoch": 0.745221027479092, + "grad_norm": 0.37880769907144246, + "learning_rate": 3.7012605417721534e-05, + "loss": 0.2908, + "step": 2495 + }, + { + "epoch": 0.7467144563918757, + "grad_norm": 0.4030754847132898, + "learning_rate": 3.660854010536055e-05, + "loss": 0.2634, + "step": 2500 + }, + { + "epoch": 0.7482078853046595, + "grad_norm": 0.32788742027641565, + "learning_rate": 3.620619773739097e-05, + "loss": 0.2183, + "step": 2505 + }, + { + "epoch": 0.7497013142174432, + "grad_norm": 0.3263923214186051, + "learning_rate": 3.580558924925305e-05, + "loss": 0.2751, + "step": 2510 + }, + { + "epoch": 0.751194743130227, + "grad_norm": 0.450636347100308, + "learning_rate": 3.540672552926122e-05, + "loss": 0.2625, + "step": 2515 + }, + { + "epoch": 0.7526881720430108, + "grad_norm": 0.33757741246501627, + "learning_rate": 3.500961741830821e-05, + "loss": 0.2575, + "step": 2520 + }, + { + "epoch": 0.7541816009557945, + "grad_norm": 0.4249647628134897, + "learning_rate": 3.4614275709570033e-05, + "loss": 0.2619, + "step": 2525 + }, + { + "epoch": 0.7556750298685783, + "grad_norm": 0.37310284633404023, + "learning_rate": 3.422071114821304e-05, + "loss": 0.2466, + "step": 2530 + }, + { + "epoch": 0.757168458781362, + "grad_norm": 0.3827899754640218, + "learning_rate": 3.382893443110175e-05, + "loss": 0.2808, + "step": 2535 + }, + { + "epoch": 0.7586618876941458, + "grad_norm": 0.34434815179683986, + "learning_rate": 3.343895620650793e-05, + "loss": 0.2369, + "step": 2540 + }, + { + "epoch": 0.7601553166069295, + "grad_norm": 0.38063545879124566, + "learning_rate": 3.305078707382138e-05, + "loss": 0.277, + "step": 2545 + }, + { + "epoch": 0.7616487455197133, + "grad_norm": 0.39371915129642, + "learning_rate": 3.26644375832619e-05, + "loss": 0.2721, + "step": 2550 + }, + { + "epoch": 0.763142174432497, + "grad_norm": 0.4068353272903545, + "learning_rate": 3.2279918235592346e-05, + "loss": 0.23, + "step": 2555 + }, + { + "epoch": 0.7646356033452808, + "grad_norm": 0.40612317566802253, + "learning_rate": 3.189723948183322e-05, + "loss": 0.2665, + "step": 2560 + }, + { + "epoch": 0.7661290322580645, + "grad_norm": 0.36414950817333475, + "learning_rate": 3.151641172297891e-05, + "loss": 0.2534, + "step": 2565 + }, + { + "epoch": 0.7676224611708483, + "grad_norm": 0.36001677147408806, + "learning_rate": 3.1137445309714705e-05, + "loss": 0.2537, + "step": 2570 + }, + { + "epoch": 0.769115890083632, + "grad_norm": 0.4064294097626817, + "learning_rate": 3.076035054213548e-05, + "loss": 0.2628, + "step": 2575 + }, + { + "epoch": 0.7706093189964157, + "grad_norm": 0.38634424971456655, + "learning_rate": 3.0385137669465967e-05, + "loss": 0.2475, + "step": 2580 + }, + { + "epoch": 0.7721027479091995, + "grad_norm": 0.3868359752050659, + "learning_rate": 3.001181688978203e-05, + "loss": 0.257, + "step": 2585 + }, + { + "epoch": 0.7735961768219832, + "grad_norm": 0.31847898193899454, + "learning_rate": 2.9640398349733334e-05, + "loss": 0.243, + "step": 2590 + }, + { + "epoch": 0.775089605734767, + "grad_norm": 0.4155060821736252, + "learning_rate": 2.9270892144267993e-05, + "loss": 0.2739, + "step": 2595 + }, + { + "epoch": 0.7765830346475507, + "grad_norm": 0.36416964732293394, + "learning_rate": 2.890330831635778e-05, + "loss": 0.2694, + "step": 2600 + }, + { + "epoch": 0.7765830346475507, + "eval_loss": 0.24662606418132782, + "eval_runtime": 671.3061, + "eval_samples_per_second": 5.959, + "eval_steps_per_second": 0.745, + "step": 2600 + }, + { + "epoch": 0.7780764635603346, + "grad_norm": 0.34102575217523634, + "learning_rate": 2.853765685672528e-05, + "loss": 0.2345, + "step": 2605 + }, + { + "epoch": 0.7795698924731183, + "grad_norm": 0.3812471975597926, + "learning_rate": 2.8173947703572546e-05, + "loss": 0.2619, + "step": 2610 + }, + { + "epoch": 0.7810633213859021, + "grad_norm": 0.4709017717954452, + "learning_rate": 2.7812190742310707e-05, + "loss": 0.2359, + "step": 2615 + }, + { + "epoch": 0.7825567502986858, + "grad_norm": 0.3815586898993409, + "learning_rate": 2.7452395805291463e-05, + "loss": 0.2485, + "step": 2620 + }, + { + "epoch": 0.7840501792114696, + "grad_norm": 0.44121044789192165, + "learning_rate": 2.7094572671539765e-05, + "loss": 0.2615, + "step": 2625 + }, + { + "epoch": 0.7855436081242533, + "grad_norm": 0.35382301353141093, + "learning_rate": 2.6738731066488075e-05, + "loss": 0.2154, + "step": 2630 + }, + { + "epoch": 0.7870370370370371, + "grad_norm": 0.33814033552150285, + "learning_rate": 2.638488066171201e-05, + "loss": 0.2569, + "step": 2635 + }, + { + "epoch": 0.7885304659498208, + "grad_norm": 0.37849934442478683, + "learning_rate": 2.603303107466747e-05, + "loss": 0.2616, + "step": 2640 + }, + { + "epoch": 0.7900238948626045, + "grad_norm": 0.3904880202778289, + "learning_rate": 2.5683191868429247e-05, + "loss": 0.2587, + "step": 2645 + }, + { + "epoch": 0.7915173237753883, + "grad_norm": 0.3646861718649452, + "learning_rate": 2.5335372551431125e-05, + "loss": 0.2766, + "step": 2650 + }, + { + "epoch": 0.793010752688172, + "grad_norm": 0.38220642098483965, + "learning_rate": 2.4989582577207417e-05, + "loss": 0.2506, + "step": 2655 + }, + { + "epoch": 0.7945041816009558, + "grad_norm": 0.34734076514892087, + "learning_rate": 2.4645831344136037e-05, + "loss": 0.2676, + "step": 2660 + }, + { + "epoch": 0.7959976105137395, + "grad_norm": 0.33364705048661314, + "learning_rate": 2.4304128195183086e-05, + "loss": 0.2144, + "step": 2665 + }, + { + "epoch": 0.7974910394265233, + "grad_norm": 0.41830275352951174, + "learning_rate": 2.396448241764877e-05, + "loss": 0.281, + "step": 2670 + }, + { + "epoch": 0.798984468339307, + "grad_norm": 0.4095216224292848, + "learning_rate": 2.3626903242915267e-05, + "loss": 0.2855, + "step": 2675 + }, + { + "epoch": 0.8004778972520908, + "grad_norm": 0.3532672344479507, + "learning_rate": 2.329139984619553e-05, + "loss": 0.2391, + "step": 2680 + }, + { + "epoch": 0.8019713261648745, + "grad_norm": 0.3758073226768294, + "learning_rate": 2.295798134628404e-05, + "loss": 0.2929, + "step": 2685 + }, + { + "epoch": 0.8034647550776584, + "grad_norm": 0.35018181270060134, + "learning_rate": 2.2626656805309e-05, + "loss": 0.2357, + "step": 2690 + }, + { + "epoch": 0.804958183990442, + "grad_norm": 0.4220541464364436, + "learning_rate": 2.2297435228485918e-05, + "loss": 0.246, + "step": 2695 + }, + { + "epoch": 0.8064516129032258, + "grad_norm": 0.3870739284799057, + "learning_rate": 2.197032556387295e-05, + "loss": 0.2766, + "step": 2700 + }, + { + "epoch": 0.8079450418160096, + "grad_norm": 0.41053303775154976, + "learning_rate": 2.164533670212766e-05, + "loss": 0.2407, + "step": 2705 + }, + { + "epoch": 0.8094384707287933, + "grad_norm": 0.3731772239188146, + "learning_rate": 2.1322477476265367e-05, + "loss": 0.2864, + "step": 2710 + }, + { + "epoch": 0.8109318996415771, + "grad_norm": 0.37634222348253304, + "learning_rate": 2.1001756661419093e-05, + "loss": 0.2424, + "step": 2715 + }, + { + "epoch": 0.8124253285543608, + "grad_norm": 0.3644559156551827, + "learning_rate": 2.068318297460102e-05, + "loss": 0.2839, + "step": 2720 + }, + { + "epoch": 0.8139187574671446, + "grad_norm": 0.39313980741067645, + "learning_rate": 2.036676507446562e-05, + "loss": 0.2642, + "step": 2725 + }, + { + "epoch": 0.8154121863799283, + "grad_norm": 0.3850789535231269, + "learning_rate": 2.005251156107426e-05, + "loss": 0.2104, + "step": 2730 + }, + { + "epoch": 0.8169056152927121, + "grad_norm": 0.39312531551082797, + "learning_rate": 1.9740430975661528e-05, + "loss": 0.264, + "step": 2735 + }, + { + "epoch": 0.8183990442054958, + "grad_norm": 0.36295621136819534, + "learning_rate": 1.943053180040302e-05, + "loss": 0.2307, + "step": 2740 + }, + { + "epoch": 0.8198924731182796, + "grad_norm": 0.3725954567916029, + "learning_rate": 1.9122822458184807e-05, + "loss": 0.2479, + "step": 2745 + }, + { + "epoch": 0.8213859020310633, + "grad_norm": 0.3713829913873105, + "learning_rate": 1.8817311312374564e-05, + "loss": 0.2575, + "step": 2750 + }, + { + "epoch": 0.822879330943847, + "grad_norm": 0.3202055385441327, + "learning_rate": 1.8514006666594197e-05, + "loss": 0.2308, + "step": 2755 + }, + { + "epoch": 0.8243727598566308, + "grad_norm": 0.3833958130308794, + "learning_rate": 1.821291676449417e-05, + "loss": 0.2713, + "step": 2760 + }, + { + "epoch": 0.8258661887694145, + "grad_norm": 0.37070939699033145, + "learning_rate": 1.7914049789529465e-05, + "loss": 0.2253, + "step": 2765 + }, + { + "epoch": 0.8273596176821983, + "grad_norm": 0.3091049252789486, + "learning_rate": 1.7617413864737153e-05, + "loss": 0.2491, + "step": 2770 + }, + { + "epoch": 0.828853046594982, + "grad_norm": 0.3863395305062906, + "learning_rate": 1.73230170525156e-05, + "loss": 0.3097, + "step": 2775 + }, + { + "epoch": 0.8303464755077659, + "grad_norm": 0.3398351066167249, + "learning_rate": 1.7030867354405354e-05, + "loss": 0.2175, + "step": 2780 + }, + { + "epoch": 0.8318399044205496, + "grad_norm": 0.353170388342167, + "learning_rate": 1.674097271087165e-05, + "loss": 0.2526, + "step": 2785 + }, + { + "epoch": 0.8333333333333334, + "grad_norm": 0.37416682070994395, + "learning_rate": 1.6453341001088572e-05, + "loss": 0.2155, + "step": 2790 + }, + { + "epoch": 0.8348267622461171, + "grad_norm": 0.38759381091463324, + "learning_rate": 1.616798004272506e-05, + "loss": 0.2516, + "step": 2795 + }, + { + "epoch": 0.8363201911589009, + "grad_norm": 0.3774042044993102, + "learning_rate": 1.5884897591732127e-05, + "loss": 0.2639, + "step": 2800 + }, + { + "epoch": 0.8363201911589009, + "eval_loss": 0.24498099088668823, + "eval_runtime": 670.7113, + "eval_samples_per_second": 5.964, + "eval_steps_per_second": 0.745, + "step": 2800 + }, + { + "epoch": 0.8378136200716846, + "grad_norm": 0.47282083906303013, + "learning_rate": 1.5604101342132306e-05, + "loss": 0.1951, + "step": 2805 + }, + { + "epoch": 0.8393070489844683, + "grad_norm": 0.3464605159651398, + "learning_rate": 1.5325598925810548e-05, + "loss": 0.265, + "step": 2810 + }, + { + "epoch": 0.8408004778972521, + "grad_norm": 0.4146312634378869, + "learning_rate": 1.5049397912306518e-05, + "loss": 0.2574, + "step": 2815 + }, + { + "epoch": 0.8422939068100358, + "grad_norm": 0.40351451569319263, + "learning_rate": 1.4775505808609191e-05, + "loss": 0.2733, + "step": 2820 + }, + { + "epoch": 0.8437873357228196, + "grad_norm": 0.3716555969690491, + "learning_rate": 1.4503930058952586e-05, + "loss": 0.277, + "step": 2825 + }, + { + "epoch": 0.8452807646356033, + "grad_norm": 0.34726179203272095, + "learning_rate": 1.423467804461357e-05, + "loss": 0.2247, + "step": 2830 + }, + { + "epoch": 0.8467741935483871, + "grad_norm": 0.3357554714579106, + "learning_rate": 1.396775708371113e-05, + "loss": 0.2656, + "step": 2835 + }, + { + "epoch": 0.8482676224611708, + "grad_norm": 0.3575152949170555, + "learning_rate": 1.3703174431007559e-05, + "loss": 0.2352, + "step": 2840 + }, + { + "epoch": 0.8497610513739546, + "grad_norm": 0.3501080536067048, + "learning_rate": 1.344093727771124e-05, + "loss": 0.2406, + "step": 2845 + }, + { + "epoch": 0.8512544802867383, + "grad_norm": 0.5043952724631786, + "learning_rate": 1.3181052751281197e-05, + "loss": 0.2768, + "step": 2850 + }, + { + "epoch": 0.8527479091995221, + "grad_norm": 0.3575895422258797, + "learning_rate": 1.2923527915233336e-05, + "loss": 0.2125, + "step": 2855 + }, + { + "epoch": 0.8542413381123058, + "grad_norm": 0.3407245335177286, + "learning_rate": 1.2668369768948608e-05, + "loss": 0.2661, + "step": 2860 + }, + { + "epoch": 0.8557347670250897, + "grad_norm": 0.4458153062323654, + "learning_rate": 1.2415585247482498e-05, + "loss": 0.2226, + "step": 2865 + }, + { + "epoch": 0.8572281959378734, + "grad_norm": 0.4369000451331262, + "learning_rate": 1.2165181221376787e-05, + "loss": 0.272, + "step": 2870 + }, + { + "epoch": 0.8587216248506571, + "grad_norm": 0.48118794169539114, + "learning_rate": 1.1917164496472799e-05, + "loss": 0.2616, + "step": 2875 + }, + { + "epoch": 0.8602150537634409, + "grad_norm": 0.3744622943326254, + "learning_rate": 1.1671541813726194e-05, + "loss": 0.264, + "step": 2880 + }, + { + "epoch": 0.8617084826762246, + "grad_norm": 0.4536685590862434, + "learning_rate": 1.1428319849023984e-05, + "loss": 0.2557, + "step": 2885 + }, + { + "epoch": 0.8632019115890084, + "grad_norm": 0.34146777737976525, + "learning_rate": 1.1187505213003068e-05, + "loss": 0.2228, + "step": 2890 + }, + { + "epoch": 0.8646953405017921, + "grad_norm": 0.4000563885902401, + "learning_rate": 1.0949104450870384e-05, + "loss": 0.2672, + "step": 2895 + }, + { + "epoch": 0.8661887694145759, + "grad_norm": 0.380767424516121, + "learning_rate": 1.0713124042225165e-05, + "loss": 0.2704, + "step": 2900 + }, + { + "epoch": 0.8676821983273596, + "grad_norm": 0.3742084189949366, + "learning_rate": 1.0479570400882888e-05, + "loss": 0.2166, + "step": 2905 + }, + { + "epoch": 0.8691756272401434, + "grad_norm": 0.4568051868923601, + "learning_rate": 1.0248449874700705e-05, + "loss": 0.2639, + "step": 2910 + }, + { + "epoch": 0.8706690561529271, + "grad_norm": 0.3959326669620433, + "learning_rate": 1.0019768745405122e-05, + "loss": 0.224, + "step": 2915 + }, + { + "epoch": 0.8721624850657109, + "grad_norm": 0.3843360418655597, + "learning_rate": 9.79353322842127e-06, + "loss": 0.2596, + "step": 2920 + }, + { + "epoch": 0.8736559139784946, + "grad_norm": 0.3404605327204112, + "learning_rate": 9.569749472703816e-06, + "loss": 0.2532, + "step": 2925 + }, + { + "epoch": 0.8751493428912783, + "grad_norm": 0.4023542619848836, + "learning_rate": 9.34842356056993e-06, + "loss": 0.2461, + "step": 2930 + }, + { + "epoch": 0.8766427718040621, + "grad_norm": 0.34874999168812615, + "learning_rate": 9.129561507534046e-06, + "loss": 0.2625, + "step": 2935 + }, + { + "epoch": 0.8781362007168458, + "grad_norm": 0.363113367835939, + "learning_rate": 8.913169262144239e-06, + "loss": 0.2298, + "step": 2940 + }, + { + "epoch": 0.8796296296296297, + "grad_norm": 0.4366858825613128, + "learning_rate": 8.699252705820526e-06, + "loss": 0.266, + "step": 2945 + }, + { + "epoch": 0.8811230585424134, + "grad_norm": 0.3631698452397834, + "learning_rate": 8.487817652695229e-06, + "loss": 0.2884, + "step": 2950 + }, + { + "epoch": 0.8826164874551972, + "grad_norm": 0.3576538561239022, + "learning_rate": 8.278869849454718e-06, + "loss": 0.242, + "step": 2955 + }, + { + "epoch": 0.8841099163679809, + "grad_norm": 0.4230742530870855, + "learning_rate": 8.07241497518324e-06, + "loss": 0.2733, + "step": 2960 + }, + { + "epoch": 0.8856033452807647, + "grad_norm": 0.4193451166331605, + "learning_rate": 7.868458641208765e-06, + "loss": 0.2314, + "step": 2965 + }, + { + "epoch": 0.8870967741935484, + "grad_norm": 0.3701152603399041, + "learning_rate": 7.667006390950248e-06, + "loss": 0.2767, + "step": 2970 + }, + { + "epoch": 0.8885902031063322, + "grad_norm": 0.47090020438598185, + "learning_rate": 7.468063699767081e-06, + "loss": 0.2731, + "step": 2975 + }, + { + "epoch": 0.8900836320191159, + "grad_norm": 0.4420262562340979, + "learning_rate": 7.27163597481022e-06, + "loss": 0.2428, + "step": 2980 + }, + { + "epoch": 0.8915770609318996, + "grad_norm": 0.4555285504311469, + "learning_rate": 7.077728554875263e-06, + "loss": 0.2637, + "step": 2985 + }, + { + "epoch": 0.8930704898446834, + "grad_norm": 0.32544659259467257, + "learning_rate": 6.886346710257363e-06, + "loss": 0.2332, + "step": 2990 + }, + { + "epoch": 0.8945639187574671, + "grad_norm": 0.39917120432545034, + "learning_rate": 6.697495642607854e-06, + "loss": 0.2774, + "step": 2995 + }, + { + "epoch": 0.8960573476702509, + "grad_norm": 0.4322326823522948, + "learning_rate": 6.511180484793067e-06, + "loss": 0.2598, + "step": 3000 + }, + { + "epoch": 0.8960573476702509, + "eval_loss": 0.24354791641235352, + "eval_runtime": 670.7371, + "eval_samples_per_second": 5.964, + "eval_steps_per_second": 0.745, + "step": 3000 + }, + { + "epoch": 0.8975507765830346, + "grad_norm": 0.41924813736657407, + "learning_rate": 6.3274063007547125e-06, + "loss": 0.249, + "step": 3005 + }, + { + "epoch": 0.8990442054958184, + "grad_norm": 0.31419280668927857, + "learning_rate": 6.146178085372156e-06, + "loss": 0.2753, + "step": 3010 + }, + { + "epoch": 0.9005376344086021, + "grad_norm": 0.36068884644950366, + "learning_rate": 5.9675007643269054e-06, + "loss": 0.2412, + "step": 3015 + }, + { + "epoch": 0.9020310633213859, + "grad_norm": 0.3391790680684511, + "learning_rate": 5.791379193968505e-06, + "loss": 0.2583, + "step": 3020 + }, + { + "epoch": 0.9035244922341696, + "grad_norm": 0.362307062919707, + "learning_rate": 5.617818161182586e-06, + "loss": 0.2867, + "step": 3025 + }, + { + "epoch": 0.9050179211469535, + "grad_norm": 0.38255580305837367, + "learning_rate": 5.446822383260908e-06, + "loss": 0.2392, + "step": 3030 + }, + { + "epoch": 0.9065113500597372, + "grad_norm": 0.41055570601524083, + "learning_rate": 5.2783965077729755e-06, + "loss": 0.2414, + "step": 3035 + }, + { + "epoch": 0.9080047789725209, + "grad_norm": 0.4885743038657098, + "learning_rate": 5.112545112439782e-06, + "loss": 0.2534, + "step": 3040 + }, + { + "epoch": 0.9094982078853047, + "grad_norm": 0.42278986461993495, + "learning_rate": 4.949272705009411e-06, + "loss": 0.2591, + "step": 3045 + }, + { + "epoch": 0.9109916367980884, + "grad_norm": 0.4025383892648368, + "learning_rate": 4.788583723134498e-06, + "loss": 0.2887, + "step": 3050 + }, + { + "epoch": 0.9124850657108722, + "grad_norm": 0.42940556785118494, + "learning_rate": 4.630482534251601e-06, + "loss": 0.2239, + "step": 3055 + }, + { + "epoch": 0.9139784946236559, + "grad_norm": 0.4252373447840496, + "learning_rate": 4.474973435462526e-06, + "loss": 0.2648, + "step": 3060 + }, + { + "epoch": 0.9154719235364397, + "grad_norm": 0.34717731405681657, + "learning_rate": 4.322060653417525e-06, + "loss": 0.2359, + "step": 3065 + }, + { + "epoch": 0.9169653524492234, + "grad_norm": 0.3932159130540263, + "learning_rate": 4.171748344200399e-06, + "loss": 0.2844, + "step": 3070 + }, + { + "epoch": 0.9184587813620072, + "grad_norm": 0.3818285602145142, + "learning_rate": 4.0240405932155725e-06, + "loss": 0.2601, + "step": 3075 + }, + { + "epoch": 0.9199522102747909, + "grad_norm": 0.3967329457656309, + "learning_rate": 3.87894141507702e-06, + "loss": 0.2485, + "step": 3080 + }, + { + "epoch": 0.9214456391875747, + "grad_norm": 0.3646492393972917, + "learning_rate": 3.7364547534991745e-06, + "loss": 0.2921, + "step": 3085 + }, + { + "epoch": 0.9229390681003584, + "grad_norm": 0.31319503804493787, + "learning_rate": 3.5965844811897243e-06, + "loss": 0.2294, + "step": 3090 + }, + { + "epoch": 0.9244324970131422, + "grad_norm": 0.36230176443854195, + "learning_rate": 3.459334399744374e-06, + "loss": 0.2616, + "step": 3095 + }, + { + "epoch": 0.9259259259259259, + "grad_norm": 0.46214491992499196, + "learning_rate": 3.3247082395434835e-06, + "loss": 0.2821, + "step": 3100 + }, + { + "epoch": 0.9274193548387096, + "grad_norm": 0.32464884452568865, + "learning_rate": 3.19270965965075e-06, + "loss": 0.2249, + "step": 3105 + }, + { + "epoch": 0.9289127837514934, + "grad_norm": 0.38961006991574, + "learning_rate": 3.063342247713652e-06, + "loss": 0.2814, + "step": 3110 + }, + { + "epoch": 0.9304062126642771, + "grad_norm": 0.3430373709286391, + "learning_rate": 2.9366095198660292e-06, + "loss": 0.2424, + "step": 3115 + }, + { + "epoch": 0.931899641577061, + "grad_norm": 0.3754679213082568, + "learning_rate": 2.812514920632481e-06, + "loss": 0.2458, + "step": 3120 + }, + { + "epoch": 0.9333930704898447, + "grad_norm": 0.32857648616715374, + "learning_rate": 2.6910618228347397e-06, + "loss": 0.2648, + "step": 3125 + }, + { + "epoch": 0.9348864994026285, + "grad_norm": 0.38704797315639966, + "learning_rate": 2.5722535275000014e-06, + "loss": 0.2576, + "step": 3130 + }, + { + "epoch": 0.9363799283154122, + "grad_norm": 0.347816629229733, + "learning_rate": 2.456093263771242e-06, + "loss": 0.2579, + "step": 3135 + }, + { + "epoch": 0.937873357228196, + "grad_norm": 0.3285229980093725, + "learning_rate": 2.3425841888193744e-06, + "loss": 0.226, + "step": 3140 + }, + { + "epoch": 0.9393667861409797, + "grad_norm": 0.35095116254795605, + "learning_rate": 2.2317293877575195e-06, + "loss": 0.2715, + "step": 3145 + }, + { + "epoch": 0.9408602150537635, + "grad_norm": 0.43133915586490196, + "learning_rate": 2.1235318735571164e-06, + "loss": 0.2743, + "step": 3150 + }, + { + "epoch": 0.9423536439665472, + "grad_norm": 0.39808144981035765, + "learning_rate": 2.0179945869660098e-06, + "loss": 0.2133, + "step": 3155 + }, + { + "epoch": 0.9438470728793309, + "grad_norm": 0.37858205437619763, + "learning_rate": 1.9151203964285936e-06, + "loss": 0.2744, + "step": 3160 + }, + { + "epoch": 0.9453405017921147, + "grad_norm": 0.33007798696691204, + "learning_rate": 1.8149120980077595e-06, + "loss": 0.2364, + "step": 3165 + }, + { + "epoch": 0.9468339307048984, + "grad_norm": 0.40641752996751496, + "learning_rate": 1.7173724153089598e-06, + "loss": 0.2546, + "step": 3170 + }, + { + "epoch": 0.9483273596176822, + "grad_norm": 0.41533410990816155, + "learning_rate": 1.6225039994061552e-06, + "loss": 0.2551, + "step": 3175 + }, + { + "epoch": 0.9498207885304659, + "grad_norm": 0.31219456691145986, + "learning_rate": 1.5303094287697938e-06, + "loss": 0.2152, + "step": 3180 + }, + { + "epoch": 0.9513142174432497, + "grad_norm": 0.394461459007535, + "learning_rate": 1.4407912091966902e-06, + "loss": 0.2989, + "step": 3185 + }, + { + "epoch": 0.9528076463560334, + "grad_norm": 0.3577386265843814, + "learning_rate": 1.353951773741935e-06, + "loss": 0.2229, + "step": 3190 + }, + { + "epoch": 0.9543010752688172, + "grad_norm": 0.3868747591156057, + "learning_rate": 1.2697934826527701e-06, + "loss": 0.249, + "step": 3195 + }, + { + "epoch": 0.955794504181601, + "grad_norm": 0.3780574063379163, + "learning_rate": 1.1883186233044186e-06, + "loss": 0.2483, + "step": 3200 + }, + { + "epoch": 0.955794504181601, + "eval_loss": 0.2432253062725067, + "eval_runtime": 671.4179, + "eval_samples_per_second": 5.958, + "eval_steps_per_second": 0.745, + "step": 3200 + }, + { + "epoch": 0.9572879330943848, + "grad_norm": 0.3892026722877345, + "learning_rate": 1.109529410137966e-06, + "loss": 0.2305, + "step": 3205 + }, + { + "epoch": 0.9587813620071685, + "grad_norm": 0.3931487469870653, + "learning_rate": 1.0334279846001106e-06, + "loss": 0.2544, + "step": 3210 + }, + { + "epoch": 0.9602747909199522, + "grad_norm": 0.36069640184936463, + "learning_rate": 9.600164150849854e-07, + "loss": 0.2575, + "step": 3215 + }, + { + "epoch": 0.961768219832736, + "grad_norm": 0.44849903301516003, + "learning_rate": 8.892966968779615e-07, + "loss": 0.2503, + "step": 3220 + }, + { + "epoch": 0.9632616487455197, + "grad_norm": 0.411995160828815, + "learning_rate": 8.212707521013774e-07, + "loss": 0.2612, + "step": 3225 + }, + { + "epoch": 0.9647550776583035, + "grad_norm": 0.5604597904188017, + "learning_rate": 7.559404296623495e-07, + "loss": 0.2106, + "step": 3230 + }, + { + "epoch": 0.9662485065710872, + "grad_norm": 0.3789859551503982, + "learning_rate": 6.933075052024562e-07, + "loss": 0.2597, + "step": 3235 + }, + { + "epoch": 0.967741935483871, + "grad_norm": 0.3029358775615665, + "learning_rate": 6.333736810495095e-07, + "loss": 0.22, + "step": 3240 + }, + { + "epoch": 0.9692353643966547, + "grad_norm": 0.43863865163510735, + "learning_rate": 5.761405861713142e-07, + "loss": 0.2888, + "step": 3245 + }, + { + "epoch": 0.9707287933094385, + "grad_norm": 0.3410374697833037, + "learning_rate": 5.216097761313377e-07, + "loss": 0.2615, + "step": 3250 + }, + { + "epoch": 0.9722222222222222, + "grad_norm": 0.3972032404026882, + "learning_rate": 4.6978273304646434e-07, + "loss": 0.2084, + "step": 3255 + }, + { + "epoch": 0.973715651135006, + "grad_norm": 0.43978255054644894, + "learning_rate": 4.206608655467403e-07, + "loss": 0.2626, + "step": 3260 + }, + { + "epoch": 0.9752090800477897, + "grad_norm": 0.4054251372400272, + "learning_rate": 3.7424550873699227e-07, + "loss": 0.2104, + "step": 3265 + }, + { + "epoch": 0.9767025089605734, + "grad_norm": 0.3250043163561357, + "learning_rate": 3.305379241606343e-07, + "loss": 0.2575, + "step": 3270 + }, + { + "epoch": 0.9781959378733572, + "grad_norm": 0.34288193953849005, + "learning_rate": 2.8953929976536233e-07, + "loss": 0.2521, + "step": 3275 + }, + { + "epoch": 0.9796893667861409, + "grad_norm": 0.3438606265427078, + "learning_rate": 2.51250749870835e-07, + "loss": 0.2223, + "step": 3280 + }, + { + "epoch": 0.9811827956989247, + "grad_norm": 0.45991832962727547, + "learning_rate": 2.156733151383872e-07, + "loss": 0.2615, + "step": 3285 + }, + { + "epoch": 0.9826762246117084, + "grad_norm": 0.34236518853477144, + "learning_rate": 1.8280796254279698e-07, + "loss": 0.2332, + "step": 3290 + }, + { + "epoch": 0.9841696535244923, + "grad_norm": 0.3358771867133346, + "learning_rate": 1.5265558534592883e-07, + "loss": 0.2436, + "step": 3295 + }, + { + "epoch": 0.985663082437276, + "grad_norm": 0.393065156149057, + "learning_rate": 1.252170030725308e-07, + "loss": 0.2707, + "step": 3300 + }, + { + "epoch": 0.9871565113500598, + "grad_norm": 0.3901577896476038, + "learning_rate": 1.0049296148790799e-07, + "loss": 0.2241, + "step": 3305 + }, + { + "epoch": 0.9886499402628435, + "grad_norm": 0.39775264512368114, + "learning_rate": 7.848413257766085e-08, + "loss": 0.2768, + "step": 3310 + }, + { + "epoch": 0.9901433691756273, + "grad_norm": 0.4277804217503546, + "learning_rate": 5.9191114529433266e-08, + "loss": 0.2351, + "step": 3315 + }, + { + "epoch": 0.991636798088411, + "grad_norm": 0.326337320725509, + "learning_rate": 4.26144317166699e-08, + "loss": 0.2526, + "step": 3320 + }, + { + "epoch": 0.9931302270011948, + "grad_norm": 0.4455257895279916, + "learning_rate": 2.8754534684316547e-08, + "loss": 0.267, + "step": 3325 + }, + { + "epoch": 0.9946236559139785, + "grad_norm": 0.3562424876586232, + "learning_rate": 1.7611800136618783e-08, + "loss": 0.2358, + "step": 3330 + }, + { + "epoch": 0.9961170848267622, + "grad_norm": 0.36120057011336665, + "learning_rate": 9.186530926874604e-09, + "loss": 0.2675, + "step": 3335 + }, + { + "epoch": 0.997610513739546, + "grad_norm": 0.3421927721985385, + "learning_rate": 3.4789560491743643e-09, + "loss": 0.2207, + "step": 3340 + }, + { + "epoch": 0.9991039426523297, + "grad_norm": 0.3603187953155133, + "learning_rate": 4.892306322057394e-10, + "loss": 0.2668, + "step": 3345 + }, + { + "epoch": 1.0, + "step": 3348, + "total_flos": 1.6414748941746176e+16, + "train_loss": 0.27713136451503567, + "train_runtime": 31532.8493, + "train_samples_per_second": 1.699, + "train_steps_per_second": 0.106 + } + ], + "logging_steps": 5, + "max_steps": 3348, + "num_input_tokens_seen": 0, + "num_train_epochs": 1, + "save_steps": 200, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": true + }, + "attributes": {} + } + }, + "total_flos": 1.6414748941746176e+16, + "train_batch_size": 2, + "trial_name": null, + "trial_params": null +}