Files
gemma-3-1b-it-xlsum-ua-sft/trainer_state.json

2899 lines
66 KiB
JSON
Raw Normal View History

{
"best_global_step": 4000,
"best_metric": 22.2348,
"best_model_checkpoint": "outputs/gemma3-1b-sft/checkpoints/checkpoint-4000",
"epoch": 1.4810220329568597,
"eval_steps": 500,
"global_step": 4000,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.0037030179596371044,
"grad_norm": 134.0,
"learning_rate": 1.1042944785276075e-06,
"loss": 3.1348,
"step": 10
},
{
"epoch": 0.007406035919274209,
"grad_norm": 95.5,
"learning_rate": 2.331288343558282e-06,
"loss": 2.7656,
"step": 20
},
{
"epoch": 0.011109053878911312,
"grad_norm": 50.75,
"learning_rate": 3.5582822085889574e-06,
"loss": 2.2816,
"step": 30
},
{
"epoch": 0.014812071838548418,
"grad_norm": 30.5,
"learning_rate": 4.785276073619632e-06,
"loss": 2.0351,
"step": 40
},
{
"epoch": 0.01851508979818552,
"grad_norm": 48.25,
"learning_rate": 6.012269938650307e-06,
"loss": 1.9907,
"step": 50
},
{
"epoch": 0.022218107757822625,
"grad_norm": 24.625,
"learning_rate": 7.239263803680983e-06,
"loss": 2.2951,
"step": 60
},
{
"epoch": 0.02592112571745973,
"grad_norm": 31.5,
"learning_rate": 8.466257668711658e-06,
"loss": 1.9474,
"step": 70
},
{
"epoch": 0.029624143677096836,
"grad_norm": 22.375,
"learning_rate": 9.693251533742331e-06,
"loss": 1.7471,
"step": 80
},
{
"epoch": 0.03332716163673394,
"grad_norm": 22.25,
"learning_rate": 1.0920245398773005e-05,
"loss": 1.6824,
"step": 90
},
{
"epoch": 0.03703017959637104,
"grad_norm": 40.5,
"learning_rate": 1.2147239263803683e-05,
"loss": 1.8535,
"step": 100
},
{
"epoch": 0.04073319755600815,
"grad_norm": 23.875,
"learning_rate": 1.3374233128834357e-05,
"loss": 2.2287,
"step": 110
},
{
"epoch": 0.04443621551564525,
"grad_norm": 19.75,
"learning_rate": 1.4601226993865032e-05,
"loss": 1.7886,
"step": 120
},
{
"epoch": 0.04813923347528235,
"grad_norm": 22.875,
"learning_rate": 1.5828220858895708e-05,
"loss": 1.6776,
"step": 130
},
{
"epoch": 0.05184225143491946,
"grad_norm": 20.125,
"learning_rate": 1.7055214723926382e-05,
"loss": 1.6208,
"step": 140
},
{
"epoch": 0.05554526939455656,
"grad_norm": 32.0,
"learning_rate": 1.828220858895706e-05,
"loss": 1.6916,
"step": 150
},
{
"epoch": 0.05924828735419367,
"grad_norm": 19.75,
"learning_rate": 1.9509202453987733e-05,
"loss": 2.1032,
"step": 160
},
{
"epoch": 0.06295130531383077,
"grad_norm": 20.875,
"learning_rate": 1.9999935274563924e-05,
"loss": 1.7711,
"step": 170
},
{
"epoch": 0.06665432327346787,
"grad_norm": 16.0,
"learning_rate": 1.999953973326663e-05,
"loss": 1.615,
"step": 180
},
{
"epoch": 0.07035734123310498,
"grad_norm": 17.375,
"learning_rate": 1.9998784623453477e-05,
"loss": 1.657,
"step": 190
},
{
"epoch": 0.07406035919274208,
"grad_norm": 25.75,
"learning_rate": 1.9997669972277077e-05,
"loss": 1.7599,
"step": 200
},
{
"epoch": 0.07776337715237919,
"grad_norm": 18.0,
"learning_rate": 1.9996195819818632e-05,
"loss": 2.0195,
"step": 210
},
{
"epoch": 0.0814663951120163,
"grad_norm": 16.625,
"learning_rate": 1.999436221908646e-05,
"loss": 1.7271,
"step": 220
},
{
"epoch": 0.08516941307165339,
"grad_norm": 16.875,
"learning_rate": 1.9992169236014103e-05,
"loss": 1.5971,
"step": 230
},
{
"epoch": 0.0888724310312905,
"grad_norm": 19.25,
"learning_rate": 1.9989616949457956e-05,
"loss": 1.5185,
"step": 240
},
{
"epoch": 0.09257544899092761,
"grad_norm": 27.375,
"learning_rate": 1.998670545119444e-05,
"loss": 1.6678,
"step": 250
},
{
"epoch": 0.0962784669505647,
"grad_norm": 15.25,
"learning_rate": 1.9983434845916675e-05,
"loss": 1.9963,
"step": 260
},
{
"epoch": 0.09998148491020181,
"grad_norm": 15.625,
"learning_rate": 1.9979805251230746e-05,
"loss": 1.6876,
"step": 270
},
{
"epoch": 0.10368450286983892,
"grad_norm": 20.0,
"learning_rate": 1.9975816797651458e-05,
"loss": 1.5567,
"step": 280
},
{
"epoch": 0.10738752082947603,
"grad_norm": 16.125,
"learning_rate": 1.997146962859765e-05,
"loss": 1.6051,
"step": 290
},
{
"epoch": 0.11109053878911312,
"grad_norm": 24.25,
"learning_rate": 1.9966763900387023e-05,
"loss": 1.6718,
"step": 300
},
{
"epoch": 0.11479355674875023,
"grad_norm": 18.125,
"learning_rate": 1.996169978223054e-05,
"loss": 1.9435,
"step": 310
},
{
"epoch": 0.11849657470838734,
"grad_norm": 37.5,
"learning_rate": 1.995627745622632e-05,
"loss": 1.7234,
"step": 320
},
{
"epoch": 0.12219959266802444,
"grad_norm": 16.875,
"learning_rate": 1.9950497117353118e-05,
"loss": 1.6168,
"step": 330
},
{
"epoch": 0.12590261062766153,
"grad_norm": 15.625,
"learning_rate": 1.9944358973463284e-05,
"loss": 1.5748,
"step": 340
},
{
"epoch": 0.12960562858729865,
"grad_norm": 32.0,
"learning_rate": 1.9937863245275303e-05,
"loss": 1.6313,
"step": 350
},
{
"epoch": 0.13330864654693575,
"grad_norm": 16.875,
"learning_rate": 1.9931010166365867e-05,
"loss": 2.0586,
"step": 360
},
{
"epoch": 0.13701166450657284,
"grad_norm": 15.4375,
"learning_rate": 1.9923799983161453e-05,
"loss": 1.7073,
"step": 370
},
{
"epoch": 0.14071468246620997,
"grad_norm": 15.375,
"learning_rate": 1.9916232954929483e-05,
"loss": 1.5613,
"step": 380
},
{
"epoch": 0.14441770042584706,
"grad_norm": 15.8125,
"learning_rate": 1.9908309353768994e-05,
"loss": 1.5504,
"step": 390
},
{
"epoch": 0.14812071838548416,
"grad_norm": 26.125,
"learning_rate": 1.9900029464600846e-05,
"loss": 1.6562,
"step": 400
},
{
"epoch": 0.15182373634512128,
"grad_norm": 14.1875,
"learning_rate": 1.98913935851575e-05,
"loss": 1.9998,
"step": 410
},
{
"epoch": 0.15552675430475837,
"grad_norm": 15.125,
"learning_rate": 1.988240202597227e-05,
"loss": 1.635,
"step": 420
},
{
"epoch": 0.15922977226439547,
"grad_norm": 13.875,
"learning_rate": 1.9873055110368208e-05,
"loss": 1.5194,
"step": 430
},
{
"epoch": 0.1629327902240326,
"grad_norm": 16.125,
"learning_rate": 1.986335317444643e-05,
"loss": 1.5388,
"step": 440
},
{
"epoch": 0.1666358081836697,
"grad_norm": 20.625,
"learning_rate": 1.9853296567074075e-05,
"loss": 1.6218,
"step": 450
},
{
"epoch": 0.17033882614330678,
"grad_norm": 15.5,
"learning_rate": 1.9842885649871716e-05,
"loss": 1.9735,
"step": 460
},
{
"epoch": 0.1740418441029439,
"grad_norm": 15.25,
"learning_rate": 1.9832120797200394e-05,
"loss": 1.6807,
"step": 470
},
{
"epoch": 0.177744862062581,
"grad_norm": 15.125,
"learning_rate": 1.9821002396148135e-05,
"loss": 1.5101,
"step": 480
},
{
"epoch": 0.1814478800222181,
"grad_norm": 13.375,
"learning_rate": 1.9809530846516033e-05,
"loss": 1.5068,
"step": 490
},
{
"epoch": 0.18515089798185522,
"grad_norm": 21.875,
"learning_rate": 1.9797706560803887e-05,
"loss": 1.5896,
"step": 500
},
{
"epoch": 0.18515089798185522,
"eval_loss": 1.5721455812454224,
"eval_runtime": 210.7323,
"eval_samples_per_second": 25.62,
"eval_steps_per_second": 3.203,
"step": 500
},
{
"epoch": 0.1888539159414923,
"grad_norm": 14.875,
"learning_rate": 1.9785529964195346e-05,
"loss": 1.9671,
"step": 510
},
{
"epoch": 0.1925569339011294,
"grad_norm": 13.6875,
"learning_rate": 1.977300149454265e-05,
"loss": 1.545,
"step": 520
},
{
"epoch": 0.19625995186076653,
"grad_norm": 13.75,
"learning_rate": 1.976012160235085e-05,
"loss": 1.4808,
"step": 530
},
{
"epoch": 0.19996296982040362,
"grad_norm": 13.75,
"learning_rate": 1.9746890750761637e-05,
"loss": 1.5404,
"step": 540
},
{
"epoch": 0.20366598778004075,
"grad_norm": 20.75,
"learning_rate": 1.973330941553668e-05,
"loss": 1.5283,
"step": 550
},
{
"epoch": 0.20736900573967784,
"grad_norm": 18.375,
"learning_rate": 1.97193780850405e-05,
"loss": 1.969,
"step": 560
},
{
"epoch": 0.21107202369931494,
"grad_norm": 15.5625,
"learning_rate": 1.9705097260222952e-05,
"loss": 1.6752,
"step": 570
},
{
"epoch": 0.21477504165895206,
"grad_norm": 14.4375,
"learning_rate": 1.969046745460116e-05,
"loss": 1.4525,
"step": 580
},
{
"epoch": 0.21847805961858915,
"grad_norm": 15.8125,
"learning_rate": 1.9675489194241092e-05,
"loss": 1.5357,
"step": 590
},
{
"epoch": 0.22218107757822625,
"grad_norm": 20.375,
"learning_rate": 1.966016301773862e-05,
"loss": 1.4364,
"step": 600
},
{
"epoch": 0.22588409553786337,
"grad_norm": 16.75,
"learning_rate": 1.9644489476200162e-05,
"loss": 2.0206,
"step": 610
},
{
"epoch": 0.22958711349750047,
"grad_norm": 15.875,
"learning_rate": 1.962846913322287e-05,
"loss": 1.6784,
"step": 620
},
{
"epoch": 0.23329013145713756,
"grad_norm": 15.125,
"learning_rate": 1.961210256487435e-05,
"loss": 1.5007,
"step": 630
},
{
"epoch": 0.23699314941677468,
"grad_norm": 13.0,
"learning_rate": 1.9595390359671956e-05,
"loss": 1.5176,
"step": 640
},
{
"epoch": 0.24069616737641178,
"grad_norm": 24.25,
"learning_rate": 1.9578333118561633e-05,
"loss": 1.5439,
"step": 650
},
{
"epoch": 0.24439918533604887,
"grad_norm": 14.625,
"learning_rate": 1.95609314548963e-05,
"loss": 1.8682,
"step": 660
},
{
"epoch": 0.248102203295686,
"grad_norm": 16.375,
"learning_rate": 1.9543185994413792e-05,
"loss": 1.6154,
"step": 670
},
{
"epoch": 0.25180522125532306,
"grad_norm": 14.5,
"learning_rate": 1.952509737521437e-05,
"loss": 1.5561,
"step": 680
},
{
"epoch": 0.2555082392149602,
"grad_norm": 16.875,
"learning_rate": 1.9506666247737768e-05,
"loss": 1.4455,
"step": 690
},
{
"epoch": 0.2592112571745973,
"grad_norm": 20.125,
"learning_rate": 1.9487893274739803e-05,
"loss": 1.5225,
"step": 700
},
{
"epoch": 0.2629142751342344,
"grad_norm": 14.375,
"learning_rate": 1.9468779131268553e-05,
"loss": 1.8288,
"step": 710
},
{
"epoch": 0.2666172930938715,
"grad_norm": 12.75,
"learning_rate": 1.9449324504640067e-05,
"loss": 1.5648,
"step": 720
},
{
"epoch": 0.2703203110535086,
"grad_norm": 14.875,
"learning_rate": 1.942953009441367e-05,
"loss": 1.4296,
"step": 730
},
{
"epoch": 0.2740233290131457,
"grad_norm": 12.75,
"learning_rate": 1.9409396612366796e-05,
"loss": 1.4813,
"step": 740
},
{
"epoch": 0.2777263469727828,
"grad_norm": 23.75,
"learning_rate": 1.9388924782469384e-05,
"loss": 1.5315,
"step": 750
},
{
"epoch": 0.28142936493241993,
"grad_norm": 15.0,
"learning_rate": 1.9368115340857875e-05,
"loss": 1.9072,
"step": 760
},
{
"epoch": 0.285132382892057,
"grad_norm": 13.9375,
"learning_rate": 1.9346969035808706e-05,
"loss": 1.5498,
"step": 770
},
{
"epoch": 0.2888354008516941,
"grad_norm": 14.125,
"learning_rate": 1.9325486627711433e-05,
"loss": 1.4234,
"step": 780
},
{
"epoch": 0.29253841881133125,
"grad_norm": 12.5625,
"learning_rate": 1.930366888904137e-05,
"loss": 1.4537,
"step": 790
},
{
"epoch": 0.2962414367709683,
"grad_norm": 25.875,
"learning_rate": 1.928151660433181e-05,
"loss": 1.5224,
"step": 800
},
{
"epoch": 0.29994445473060544,
"grad_norm": 12.3125,
"learning_rate": 1.9259030570145837e-05,
"loss": 1.9214,
"step": 810
},
{
"epoch": 0.30364747269024256,
"grad_norm": 13.875,
"learning_rate": 1.923621159504766e-05,
"loss": 1.6403,
"step": 820
},
{
"epoch": 0.3073504906498796,
"grad_norm": 13.3125,
"learning_rate": 1.9213060499573542e-05,
"loss": 1.4146,
"step": 830
},
{
"epoch": 0.31105350860951675,
"grad_norm": 13.625,
"learning_rate": 1.918957811620231e-05,
"loss": 1.4342,
"step": 840
},
{
"epoch": 0.31475652656915387,
"grad_norm": 24.5,
"learning_rate": 1.9165765289325394e-05,
"loss": 1.5787,
"step": 850
},
{
"epoch": 0.31845954452879094,
"grad_norm": 14.125,
"learning_rate": 1.9141622875216495e-05,
"loss": 1.9358,
"step": 860
},
{
"epoch": 0.32216256248842806,
"grad_norm": 14.125,
"learning_rate": 1.9117151742000767e-05,
"loss": 1.5976,
"step": 870
},
{
"epoch": 0.3258655804480652,
"grad_norm": 13.5625,
"learning_rate": 1.9092352769623618e-05,
"loss": 1.4429,
"step": 880
},
{
"epoch": 0.32956859840770225,
"grad_norm": 13.125,
"learning_rate": 1.9067226849819063e-05,
"loss": 1.4909,
"step": 890
},
{
"epoch": 0.3332716163673394,
"grad_norm": 28.875,
"learning_rate": 1.904177488607765e-05,
"loss": 1.5425,
"step": 900
},
{
"epoch": 0.3369746343269765,
"grad_norm": 13.0625,
"learning_rate": 1.901599779361399e-05,
"loss": 1.9504,
"step": 910
},
{
"epoch": 0.34067765228661356,
"grad_norm": 11.4375,
"learning_rate": 1.8989896499333825e-05,
"loss": 1.621,
"step": 920
},
{
"epoch": 0.3443806702462507,
"grad_norm": 15.875,
"learning_rate": 1.8963471941800735e-05,
"loss": 1.3815,
"step": 930
},
{
"epoch": 0.3480836882058878,
"grad_norm": 12.375,
"learning_rate": 1.8936725071202334e-05,
"loss": 1.4537,
"step": 940
},
{
"epoch": 0.3517867061655249,
"grad_norm": 20.25,
"learning_rate": 1.8909656849316154e-05,
"loss": 1.4894,
"step": 950
},
{
"epoch": 0.355489724125162,
"grad_norm": 16.0,
"learning_rate": 1.8882268249475028e-05,
"loss": 1.9715,
"step": 960
},
{
"epoch": 0.3591927420847991,
"grad_norm": 13.4375,
"learning_rate": 1.8854560256532098e-05,
"loss": 1.5505,
"step": 970
},
{
"epoch": 0.3628957600444362,
"grad_norm": 14.1875,
"learning_rate": 1.8826533866825424e-05,
"loss": 1.5242,
"step": 980
},
{
"epoch": 0.3665987780040733,
"grad_norm": 15.3125,
"learning_rate": 1.879819008814212e-05,
"loss": 1.4714,
"step": 990
},
{
"epoch": 0.37030179596371043,
"grad_norm": 20.625,
"learning_rate": 1.8769529939682146e-05,
"loss": 1.5105,
"step": 1000
},
{
"epoch": 0.37030179596371043,
"eval_loss": 1.4920871257781982,
"eval_runtime": 202.3546,
"eval_samples_per_second": 26.681,
"eval_steps_per_second": 3.336,
"step": 1000
},
{
"epoch": 0.3740048139233475,
"grad_norm": 14.75,
"learning_rate": 1.874055445202163e-05,
"loss": 1.8687,
"step": 1010
},
{
"epoch": 0.3777078318829846,
"grad_norm": 12.8125,
"learning_rate": 1.8711264667075847e-05,
"loss": 1.5272,
"step": 1020
},
{
"epoch": 0.38141084984262175,
"grad_norm": 13.4375,
"learning_rate": 1.8681661638061718e-05,
"loss": 1.4739,
"step": 1030
},
{
"epoch": 0.3851138678022588,
"grad_norm": 13.625,
"learning_rate": 1.8651746429459957e-05,
"loss": 1.4486,
"step": 1040
},
{
"epoch": 0.38881688576189594,
"grad_norm": 25.375,
"learning_rate": 1.8621520116976798e-05,
"loss": 1.5157,
"step": 1050
},
{
"epoch": 0.39251990372153306,
"grad_norm": 13.5,
"learning_rate": 1.8590983787505293e-05,
"loss": 1.855,
"step": 1060
},
{
"epoch": 0.3962229216811701,
"grad_norm": 14.5625,
"learning_rate": 1.8560138539086248e-05,
"loss": 1.5695,
"step": 1070
},
{
"epoch": 0.39992593964080725,
"grad_norm": 12.125,
"learning_rate": 1.8528985480868736e-05,
"loss": 1.4379,
"step": 1080
},
{
"epoch": 0.40362895760044437,
"grad_norm": 14.375,
"learning_rate": 1.8497525733070206e-05,
"loss": 1.484,
"step": 1090
},
{
"epoch": 0.4073319755600815,
"grad_norm": 18.75,
"learning_rate": 1.8465760426936212e-05,
"loss": 1.4685,
"step": 1100
},
{
"epoch": 0.41103499351971856,
"grad_norm": 14.375,
"learning_rate": 1.8433690704699718e-05,
"loss": 1.8358,
"step": 1110
},
{
"epoch": 0.4147380114793557,
"grad_norm": 13.3125,
"learning_rate": 1.840131771954005e-05,
"loss": 1.5536,
"step": 1120
},
{
"epoch": 0.4184410294389928,
"grad_norm": 14.4375,
"learning_rate": 1.8368642635541408e-05,
"loss": 1.4037,
"step": 1130
},
{
"epoch": 0.4221440473986299,
"grad_norm": 12.3125,
"learning_rate": 1.8335666627651016e-05,
"loss": 1.4152,
"step": 1140
},
{
"epoch": 0.425847065358267,
"grad_norm": 23.875,
"learning_rate": 1.8302390881636868e-05,
"loss": 1.4929,
"step": 1150
},
{
"epoch": 0.4295500833179041,
"grad_norm": 15.375,
"learning_rate": 1.8268816594045108e-05,
"loss": 1.9374,
"step": 1160
},
{
"epoch": 0.4332531012775412,
"grad_norm": 13.75,
"learning_rate": 1.823494497215697e-05,
"loss": 1.5484,
"step": 1170
},
{
"epoch": 0.4369561192371783,
"grad_norm": 14.375,
"learning_rate": 1.8200777233945405e-05,
"loss": 1.3895,
"step": 1180
},
{
"epoch": 0.44065913719681543,
"grad_norm": 14.5,
"learning_rate": 1.8166314608031238e-05,
"loss": 1.3277,
"step": 1190
},
{
"epoch": 0.4443621551564525,
"grad_norm": 20.75,
"learning_rate": 1.8131558333639047e-05,
"loss": 1.5164,
"step": 1200
},
{
"epoch": 0.4480651731160896,
"grad_norm": 12.5625,
"learning_rate": 1.8096509660552552e-05,
"loss": 1.8795,
"step": 1210
},
{
"epoch": 0.45176819107572674,
"grad_norm": 14.5625,
"learning_rate": 1.806116984906969e-05,
"loss": 1.4731,
"step": 1220
},
{
"epoch": 0.4554712090353638,
"grad_norm": 12.625,
"learning_rate": 1.8025540169957315e-05,
"loss": 1.391,
"step": 1230
},
{
"epoch": 0.45917422699500093,
"grad_norm": 12.1875,
"learning_rate": 1.7989621904405472e-05,
"loss": 1.4385,
"step": 1240
},
{
"epoch": 0.46287724495463806,
"grad_norm": 21.125,
"learning_rate": 1.7953416343981354e-05,
"loss": 1.3887,
"step": 1250
},
{
"epoch": 0.4665802629142751,
"grad_norm": 14.875,
"learning_rate": 1.7916924790582853e-05,
"loss": 1.9236,
"step": 1260
},
{
"epoch": 0.47028328087391225,
"grad_norm": 14.25,
"learning_rate": 1.7880148556391724e-05,
"loss": 1.6421,
"step": 1270
},
{
"epoch": 0.47398629883354937,
"grad_norm": 12.3125,
"learning_rate": 1.7843088963826437e-05,
"loss": 1.3995,
"step": 1280
},
{
"epoch": 0.47768931679318644,
"grad_norm": 12.3125,
"learning_rate": 1.7805747345494585e-05,
"loss": 1.3695,
"step": 1290
},
{
"epoch": 0.48139233475282356,
"grad_norm": 19.125,
"learning_rate": 1.776812504414501e-05,
"loss": 1.4958,
"step": 1300
},
{
"epoch": 0.4850953527124607,
"grad_norm": 14.875,
"learning_rate": 1.7730223412619484e-05,
"loss": 1.8908,
"step": 1310
},
{
"epoch": 0.48879837067209775,
"grad_norm": 12.875,
"learning_rate": 1.769204381380407e-05,
"loss": 1.5785,
"step": 1320
},
{
"epoch": 0.49250138863173487,
"grad_norm": 13.4375,
"learning_rate": 1.7653587620580126e-05,
"loss": 1.4385,
"step": 1330
},
{
"epoch": 0.496204406591372,
"grad_norm": 15.0,
"learning_rate": 1.7614856215774938e-05,
"loss": 1.4015,
"step": 1340
},
{
"epoch": 0.49990742455100906,
"grad_norm": 20.375,
"learning_rate": 1.7575850992111973e-05,
"loss": 1.5091,
"step": 1350
},
{
"epoch": 0.5036104425106461,
"grad_norm": 12.375,
"learning_rate": 1.753657335216083e-05,
"loss": 1.7503,
"step": 1360
},
{
"epoch": 0.5073134604702833,
"grad_norm": 12.5,
"learning_rate": 1.749702470828679e-05,
"loss": 1.5914,
"step": 1370
},
{
"epoch": 0.5110164784299204,
"grad_norm": 16.25,
"learning_rate": 1.745720648260002e-05,
"loss": 1.4525,
"step": 1380
},
{
"epoch": 0.5147194963895575,
"grad_norm": 13.25,
"learning_rate": 1.7417120106904453e-05,
"loss": 1.3997,
"step": 1390
},
{
"epoch": 0.5184225143491946,
"grad_norm": 19.0,
"learning_rate": 1.7376767022646302e-05,
"loss": 1.417,
"step": 1400
},
{
"epoch": 0.5221255323088317,
"grad_norm": 12.8125,
"learning_rate": 1.7336148680862206e-05,
"loss": 1.8016,
"step": 1410
},
{
"epoch": 0.5258285502684688,
"grad_norm": 12.8125,
"learning_rate": 1.729526654212708e-05,
"loss": 1.5355,
"step": 1420
},
{
"epoch": 0.5295315682281059,
"grad_norm": 13.125,
"learning_rate": 1.7254122076501588e-05,
"loss": 1.3607,
"step": 1430
},
{
"epoch": 0.533234586187743,
"grad_norm": 12.25,
"learning_rate": 1.7212716763479265e-05,
"loss": 1.4062,
"step": 1440
},
{
"epoch": 0.5369376041473801,
"grad_norm": 20.5,
"learning_rate": 1.7171052091933332e-05,
"loss": 1.4949,
"step": 1450
},
{
"epoch": 0.5406406221070172,
"grad_norm": 13.375,
"learning_rate": 1.712912956006316e-05,
"loss": 1.7958,
"step": 1460
},
{
"epoch": 0.5443436400666544,
"grad_norm": 11.9375,
"learning_rate": 1.708695067534039e-05,
"loss": 1.4787,
"step": 1470
},
{
"epoch": 0.5480466580262914,
"grad_norm": 13.625,
"learning_rate": 1.7044516954454722e-05,
"loss": 1.3447,
"step": 1480
},
{
"epoch": 0.5517496759859285,
"grad_norm": 13.625,
"learning_rate": 1.70018299232594e-05,
"loss": 1.4111,
"step": 1490
},
{
"epoch": 0.5554526939455656,
"grad_norm": 19.5,
"learning_rate": 1.6958891116716315e-05,
"loss": 1.4568,
"step": 1500
},
{
"epoch": 0.5554526939455656,
"eval_loss": 1.4441919326782227,
"eval_runtime": 201.8638,
"eval_samples_per_second": 26.746,
"eval_steps_per_second": 3.344,
"step": 1500
},
{
"epoch": 0.5591557119052027,
"grad_norm": 13.4375,
"learning_rate": 1.6915702078840824e-05,
"loss": 1.77,
"step": 1510
},
{
"epoch": 0.5628587298648399,
"grad_norm": 12.125,
"learning_rate": 1.6872264362646236e-05,
"loss": 1.4758,
"step": 1520
},
{
"epoch": 0.566561747824477,
"grad_norm": 14.25,
"learning_rate": 1.6828579530087953e-05,
"loss": 1.4234,
"step": 1530
},
{
"epoch": 0.570264765784114,
"grad_norm": 10.875,
"learning_rate": 1.678464915200732e-05,
"loss": 1.3914,
"step": 1540
},
{
"epoch": 0.5739677837437511,
"grad_norm": 19.75,
"learning_rate": 1.6740474808075126e-05,
"loss": 1.4175,
"step": 1550
},
{
"epoch": 0.5776708017033882,
"grad_norm": 14.0625,
"learning_rate": 1.669605808673481e-05,
"loss": 1.7598,
"step": 1560
},
{
"epoch": 0.5813738196630254,
"grad_norm": 13.125,
"learning_rate": 1.6651400585145347e-05,
"loss": 1.5145,
"step": 1570
},
{
"epoch": 0.5850768376226625,
"grad_norm": 12.9375,
"learning_rate": 1.66065039091238e-05,
"loss": 1.4569,
"step": 1580
},
{
"epoch": 0.5887798555822996,
"grad_norm": 12.5,
"learning_rate": 1.6561369673087588e-05,
"loss": 1.3612,
"step": 1590
},
{
"epoch": 0.5924828735419366,
"grad_norm": 21.375,
"learning_rate": 1.6515999499996444e-05,
"loss": 1.434,
"step": 1600
},
{
"epoch": 0.5961858915015738,
"grad_norm": 13.875,
"learning_rate": 1.6470395021294036e-05,
"loss": 1.8501,
"step": 1610
},
{
"epoch": 0.5998889094612109,
"grad_norm": 12.75,
"learning_rate": 1.6424557876849308e-05,
"loss": 1.452,
"step": 1620
},
{
"epoch": 0.603591927420848,
"grad_norm": 12.5,
"learning_rate": 1.637848971489753e-05,
"loss": 1.3828,
"step": 1630
},
{
"epoch": 0.6072949453804851,
"grad_norm": 12.4375,
"learning_rate": 1.6332192191981008e-05,
"loss": 1.3617,
"step": 1640
},
{
"epoch": 0.6109979633401222,
"grad_norm": 20.875,
"learning_rate": 1.6285666972889527e-05,
"loss": 1.4193,
"step": 1650
},
{
"epoch": 0.6147009812997593,
"grad_norm": 13.3125,
"learning_rate": 1.6238915730600485e-05,
"loss": 1.7409,
"step": 1660
},
{
"epoch": 0.6184039992593964,
"grad_norm": 13.3125,
"learning_rate": 1.619194014621874e-05,
"loss": 1.4807,
"step": 1670
},
{
"epoch": 0.6221070172190335,
"grad_norm": 11.875,
"learning_rate": 1.614474190891615e-05,
"loss": 1.3946,
"step": 1680
},
{
"epoch": 0.6258100351786706,
"grad_norm": 13.4375,
"learning_rate": 1.609732271587085e-05,
"loss": 1.3165,
"step": 1690
},
{
"epoch": 0.6295130531383077,
"grad_norm": 17.625,
"learning_rate": 1.6049684272206207e-05,
"loss": 1.4339,
"step": 1700
},
{
"epoch": 0.6332160710979449,
"grad_norm": 18.25,
"learning_rate": 1.600182829092951e-05,
"loss": 1.8922,
"step": 1710
},
{
"epoch": 0.6369190890575819,
"grad_norm": 13.125,
"learning_rate": 1.5953756492870385e-05,
"loss": 1.5272,
"step": 1720
},
{
"epoch": 0.640622107017219,
"grad_norm": 13.0625,
"learning_rate": 1.5905470606618898e-05,
"loss": 1.3767,
"step": 1730
},
{
"epoch": 0.6443251249768561,
"grad_norm": 12.1875,
"learning_rate": 1.5856972368463405e-05,
"loss": 1.3218,
"step": 1740
},
{
"epoch": 0.6480281429364932,
"grad_norm": 22.5,
"learning_rate": 1.5808263522328137e-05,
"loss": 1.4272,
"step": 1750
},
{
"epoch": 0.6517311608961304,
"grad_norm": 15.125,
"learning_rate": 1.575934581971045e-05,
"loss": 1.8197,
"step": 1760
},
{
"epoch": 0.6554341788557675,
"grad_norm": 12.75,
"learning_rate": 1.571022101961788e-05,
"loss": 1.577,
"step": 1770
},
{
"epoch": 0.6591371968154045,
"grad_norm": 12.0625,
"learning_rate": 1.5660890888504884e-05,
"loss": 1.3704,
"step": 1780
},
{
"epoch": 0.6628402147750416,
"grad_norm": 13.8125,
"learning_rate": 1.5611357200209302e-05,
"loss": 1.3452,
"step": 1790
},
{
"epoch": 0.6665432327346787,
"grad_norm": 24.25,
"learning_rate": 1.55616217358886e-05,
"loss": 1.4719,
"step": 1800
},
{
"epoch": 0.6702462506943159,
"grad_norm": 13.3125,
"learning_rate": 1.5511686283955792e-05,
"loss": 1.8528,
"step": 1810
},
{
"epoch": 0.673949268653953,
"grad_norm": 12.5,
"learning_rate": 1.5461552640015164e-05,
"loss": 1.4783,
"step": 1820
},
{
"epoch": 0.6776522866135901,
"grad_norm": 12.75,
"learning_rate": 1.5411222606797686e-05,
"loss": 1.3574,
"step": 1830
},
{
"epoch": 0.6813553045732271,
"grad_norm": 12.125,
"learning_rate": 1.536069799409619e-05,
"loss": 1.353,
"step": 1840
},
{
"epoch": 0.6850583225328643,
"grad_norm": 21.625,
"learning_rate": 1.530998061870029e-05,
"loss": 1.4811,
"step": 1850
},
{
"epoch": 0.6887613404925014,
"grad_norm": 15.625,
"learning_rate": 1.525907230433108e-05,
"loss": 1.7574,
"step": 1860
},
{
"epoch": 0.6924643584521385,
"grad_norm": 13.0,
"learning_rate": 1.520797488157551e-05,
"loss": 1.5413,
"step": 1870
},
{
"epoch": 0.6961673764117756,
"grad_norm": 13.5,
"learning_rate": 1.5156690187820596e-05,
"loss": 1.3608,
"step": 1880
},
{
"epoch": 0.6998703943714127,
"grad_norm": 14.875,
"learning_rate": 1.510522006718734e-05,
"loss": 1.3702,
"step": 1890
},
{
"epoch": 0.7035734123310498,
"grad_norm": 19.375,
"learning_rate": 1.5053566370464416e-05,
"loss": 1.4014,
"step": 1900
},
{
"epoch": 0.7072764302906869,
"grad_norm": 14.6875,
"learning_rate": 1.5001730955041623e-05,
"loss": 1.8693,
"step": 1910
},
{
"epoch": 0.710979448250324,
"grad_norm": 13.5625,
"learning_rate": 1.4949715684843087e-05,
"loss": 1.6336,
"step": 1920
},
{
"epoch": 0.7146824662099611,
"grad_norm": 13.75,
"learning_rate": 1.489752243026025e-05,
"loss": 1.4099,
"step": 1930
},
{
"epoch": 0.7183854841695982,
"grad_norm": 10.875,
"learning_rate": 1.48451530680846e-05,
"loss": 1.3062,
"step": 1940
},
{
"epoch": 0.7220885021292354,
"grad_norm": 18.25,
"learning_rate": 1.4792609481440196e-05,
"loss": 1.4361,
"step": 1950
},
{
"epoch": 0.7257915200888724,
"grad_norm": 14.9375,
"learning_rate": 1.4739893559715945e-05,
"loss": 1.7371,
"step": 1960
},
{
"epoch": 0.7294945380485095,
"grad_norm": 12.4375,
"learning_rate": 1.4687007198497674e-05,
"loss": 1.4281,
"step": 1970
},
{
"epoch": 0.7331975560081466,
"grad_norm": 13.3125,
"learning_rate": 1.4633952299499947e-05,
"loss": 1.3712,
"step": 1980
},
{
"epoch": 0.7369005739677837,
"grad_norm": 12.4375,
"learning_rate": 1.4580730770497706e-05,
"loss": 1.3762,
"step": 1990
},
{
"epoch": 0.7406035919274209,
"grad_norm": 22.375,
"learning_rate": 1.4527344525257652e-05,
"loss": 1.3943,
"step": 2000
},
{
"epoch": 0.7406035919274209,
"eval_loss": 1.405870795249939,
"eval_runtime": 201.4686,
"eval_samples_per_second": 26.798,
"eval_steps_per_second": 3.35,
"step": 2000
},
{
"epoch": 0.744306609887058,
"grad_norm": 13.125,
"learning_rate": 1.4473795483469442e-05,
"loss": 1.8168,
"step": 2010
},
{
"epoch": 0.748009627846695,
"grad_norm": 12.75,
"learning_rate": 1.4420085570676646e-05,
"loss": 1.4723,
"step": 2020
},
{
"epoch": 0.7517126458063321,
"grad_norm": 12.8125,
"learning_rate": 1.4366216718207527e-05,
"loss": 1.3657,
"step": 2030
},
{
"epoch": 0.7554156637659692,
"grad_norm": 14.9375,
"learning_rate": 1.4312190863105564e-05,
"loss": 1.3663,
"step": 2040
},
{
"epoch": 0.7591186817256064,
"grad_norm": 19.25,
"learning_rate": 1.4258009948059835e-05,
"loss": 1.3707,
"step": 2050
},
{
"epoch": 0.7628216996852435,
"grad_norm": 12.8125,
"learning_rate": 1.4203675921335127e-05,
"loss": 1.7181,
"step": 2060
},
{
"epoch": 0.7665247176448806,
"grad_norm": 12.5,
"learning_rate": 1.4149190736701906e-05,
"loss": 1.4467,
"step": 2070
},
{
"epoch": 0.7702277356045176,
"grad_norm": 12.875,
"learning_rate": 1.4094556353366042e-05,
"loss": 1.2888,
"step": 2080
},
{
"epoch": 0.7739307535641547,
"grad_norm": 10.875,
"learning_rate": 1.403977473589837e-05,
"loss": 1.3861,
"step": 2090
},
{
"epoch": 0.7776337715237919,
"grad_norm": 20.0,
"learning_rate": 1.398484785416405e-05,
"loss": 1.388,
"step": 2100
},
{
"epoch": 0.781336789483429,
"grad_norm": 14.125,
"learning_rate": 1.3929777683251718e-05,
"loss": 1.79,
"step": 2110
},
{
"epoch": 0.7850398074430661,
"grad_norm": 12.4375,
"learning_rate": 1.387456620340248e-05,
"loss": 1.4829,
"step": 2120
},
{
"epoch": 0.7887428254027032,
"grad_norm": 13.375,
"learning_rate": 1.3819215399938706e-05,
"loss": 1.3329,
"step": 2130
},
{
"epoch": 0.7924458433623403,
"grad_norm": 11.3125,
"learning_rate": 1.3763727263192626e-05,
"loss": 1.2607,
"step": 2140
},
{
"epoch": 0.7961488613219774,
"grad_norm": 21.5,
"learning_rate": 1.3708103788434773e-05,
"loss": 1.3877,
"step": 2150
},
{
"epoch": 0.7998518792816145,
"grad_norm": 20.625,
"learning_rate": 1.3652346975802235e-05,
"loss": 1.6969,
"step": 2160
},
{
"epoch": 0.8035548972412516,
"grad_norm": 12.625,
"learning_rate": 1.3596458830226727e-05,
"loss": 1.3679,
"step": 2170
},
{
"epoch": 0.8072579152008887,
"grad_norm": 13.75,
"learning_rate": 1.3540441361362506e-05,
"loss": 1.3718,
"step": 2180
},
{
"epoch": 0.8109609331605259,
"grad_norm": 13.125,
"learning_rate": 1.3484296583514094e-05,
"loss": 1.2996,
"step": 2190
},
{
"epoch": 0.814663951120163,
"grad_norm": 20.375,
"learning_rate": 1.3428026515563861e-05,
"loss": 1.4657,
"step": 2200
},
{
"epoch": 0.8183669690798,
"grad_norm": 12.9375,
"learning_rate": 1.3371633180899417e-05,
"loss": 1.8187,
"step": 2210
},
{
"epoch": 0.8220699870394371,
"grad_norm": 12.25,
"learning_rate": 1.331511860734086e-05,
"loss": 1.4919,
"step": 2220
},
{
"epoch": 0.8257730049990742,
"grad_norm": 13.9375,
"learning_rate": 1.3258484827067864e-05,
"loss": 1.3168,
"step": 2230
},
{
"epoch": 0.8294760229587114,
"grad_norm": 13.0,
"learning_rate": 1.3201733876546583e-05,
"loss": 1.3238,
"step": 2240
},
{
"epoch": 0.8331790409183485,
"grad_norm": 21.25,
"learning_rate": 1.3144867796456459e-05,
"loss": 1.4547,
"step": 2250
},
{
"epoch": 0.8368820588779856,
"grad_norm": 11.625,
"learning_rate": 1.3087888631616801e-05,
"loss": 1.7292,
"step": 2260
},
{
"epoch": 0.8405850768376226,
"grad_norm": 13.625,
"learning_rate": 1.3030798430913289e-05,
"loss": 1.5482,
"step": 2270
},
{
"epoch": 0.8442880947972597,
"grad_norm": 11.9375,
"learning_rate": 1.2973599247224288e-05,
"loss": 1.3514,
"step": 2280
},
{
"epoch": 0.8479911127568969,
"grad_norm": 14.0,
"learning_rate": 1.2916293137347019e-05,
"loss": 1.2866,
"step": 2290
},
{
"epoch": 0.851694130716534,
"grad_norm": 16.125,
"learning_rate": 1.2858882161923618e-05,
"loss": 1.3992,
"step": 2300
},
{
"epoch": 0.8553971486761711,
"grad_norm": 14.875,
"learning_rate": 1.2801368385367021e-05,
"loss": 1.7692,
"step": 2310
},
{
"epoch": 0.8591001666358082,
"grad_norm": 12.875,
"learning_rate": 1.274375387578675e-05,
"loss": 1.5921,
"step": 2320
},
{
"epoch": 0.8628031845954452,
"grad_norm": 12.4375,
"learning_rate": 1.2686040704914527e-05,
"loss": 1.3274,
"step": 2330
},
{
"epoch": 0.8665062025550824,
"grad_norm": 10.875,
"learning_rate": 1.2628230948029789e-05,
"loss": 1.353,
"step": 2340
},
{
"epoch": 0.8702092205147195,
"grad_norm": 23.125,
"learning_rate": 1.2570326683885066e-05,
"loss": 1.4732,
"step": 2350
},
{
"epoch": 0.8739122384743566,
"grad_norm": 15.0,
"learning_rate": 1.2512329994631218e-05,
"loss": 1.7471,
"step": 2360
},
{
"epoch": 0.8776152564339937,
"grad_norm": 11.375,
"learning_rate": 1.2454242965742587e-05,
"loss": 1.4108,
"step": 2370
},
{
"epoch": 0.8813182743936309,
"grad_norm": 12.875,
"learning_rate": 1.2396067685941983e-05,
"loss": 1.362,
"step": 2380
},
{
"epoch": 0.8850212923532679,
"grad_norm": 11.8125,
"learning_rate": 1.2337806247125595e-05,
"loss": 1.2812,
"step": 2390
},
{
"epoch": 0.888724310312905,
"grad_norm": 22.625,
"learning_rate": 1.2279460744287755e-05,
"loss": 1.4155,
"step": 2400
},
{
"epoch": 0.8924273282725421,
"grad_norm": 14.5,
"learning_rate": 1.2221033275445611e-05,
"loss": 1.7539,
"step": 2410
},
{
"epoch": 0.8961303462321792,
"grad_norm": 13.1875,
"learning_rate": 1.216252594156369e-05,
"loss": 1.4012,
"step": 2420
},
{
"epoch": 0.8998333641918164,
"grad_norm": 13.3125,
"learning_rate": 1.210394084647835e-05,
"loss": 1.2873,
"step": 2430
},
{
"epoch": 0.9035363821514535,
"grad_norm": 12.8125,
"learning_rate": 1.2045280096822119e-05,
"loss": 1.2891,
"step": 2440
},
{
"epoch": 0.9072394001110905,
"grad_norm": 19.25,
"learning_rate": 1.1986545801947953e-05,
"loss": 1.4092,
"step": 2450
},
{
"epoch": 0.9109424180707276,
"grad_norm": 12.5625,
"learning_rate": 1.1927740073853382e-05,
"loss": 1.8059,
"step": 2460
},
{
"epoch": 0.9146454360303647,
"grad_norm": 16.25,
"learning_rate": 1.1868865027104573e-05,
"loss": 1.4751,
"step": 2470
},
{
"epoch": 0.9183484539900019,
"grad_norm": 12.125,
"learning_rate": 1.1809922778760283e-05,
"loss": 1.3628,
"step": 2480
},
{
"epoch": 0.922051471949639,
"grad_norm": 11.4375,
"learning_rate": 1.175091544829574e-05,
"loss": 1.2731,
"step": 2490
},
{
"epoch": 0.9257544899092761,
"grad_norm": 21.375,
"learning_rate": 1.1691845157526434e-05,
"loss": 1.4433,
"step": 2500
},
{
"epoch": 0.9257544899092761,
"eval_loss": 1.375000238418579,
"eval_runtime": 201.6753,
"eval_samples_per_second": 26.771,
"eval_steps_per_second": 3.347,
"step": 2500
},
{
"epoch": 0.9294575078689131,
"grad_norm": 15.3125,
"learning_rate": 1.1632714030531797e-05,
"loss": 1.7861,
"step": 2510
},
{
"epoch": 0.9331605258285502,
"grad_norm": 13.4375,
"learning_rate": 1.1573524193578863e-05,
"loss": 1.4002,
"step": 2520
},
{
"epoch": 0.9368635437881874,
"grad_norm": 15.3125,
"learning_rate": 1.1514277775045768e-05,
"loss": 1.3174,
"step": 2530
},
{
"epoch": 0.9405665617478245,
"grad_norm": 14.0625,
"learning_rate": 1.1454976905345249e-05,
"loss": 1.3128,
"step": 2540
},
{
"epoch": 0.9442695797074616,
"grad_norm": 22.375,
"learning_rate": 1.1395623716848025e-05,
"loss": 1.3928,
"step": 2550
},
{
"epoch": 0.9479725976670987,
"grad_norm": 12.6875,
"learning_rate": 1.1336220343806114e-05,
"loss": 1.6909,
"step": 2560
},
{
"epoch": 0.9516756156267357,
"grad_norm": 12.625,
"learning_rate": 1.1276768922276103e-05,
"loss": 1.4757,
"step": 2570
},
{
"epoch": 0.9553786335863729,
"grad_norm": 14.3125,
"learning_rate": 1.1217271590042329e-05,
"loss": 1.2579,
"step": 2580
},
{
"epoch": 0.95908165154601,
"grad_norm": 13.3125,
"learning_rate": 1.1157730486540004e-05,
"loss": 1.3886,
"step": 2590
},
{
"epoch": 0.9627846695056471,
"grad_norm": 20.25,
"learning_rate": 1.1098147752778297e-05,
"loss": 1.4061,
"step": 2600
},
{
"epoch": 0.9664876874652842,
"grad_norm": 12.9375,
"learning_rate": 1.1038525531263331e-05,
"loss": 1.7928,
"step": 2610
},
{
"epoch": 0.9701907054249214,
"grad_norm": 12.125,
"learning_rate": 1.0978865965921159e-05,
"loss": 1.5137,
"step": 2620
},
{
"epoch": 0.9738937233845584,
"grad_norm": 12.1875,
"learning_rate": 1.0919171202020652e-05,
"loss": 1.3493,
"step": 2630
},
{
"epoch": 0.9775967413441955,
"grad_norm": 12.875,
"learning_rate": 1.0859443386096382e-05,
"loss": 1.2636,
"step": 2640
},
{
"epoch": 0.9812997593038326,
"grad_norm": 20.0,
"learning_rate": 1.0799684665871404e-05,
"loss": 1.3472,
"step": 2650
},
{
"epoch": 0.9850027772634697,
"grad_norm": 14.25,
"learning_rate": 1.0739897190180066e-05,
"loss": 1.7834,
"step": 2660
},
{
"epoch": 0.9887057952231069,
"grad_norm": 12.1875,
"learning_rate": 1.06800831088907e-05,
"loss": 1.4765,
"step": 2670
},
{
"epoch": 0.992408813182744,
"grad_norm": 12.3125,
"learning_rate": 1.0620244572828352e-05,
"loss": 1.3083,
"step": 2680
},
{
"epoch": 0.996111831142381,
"grad_norm": 11.5,
"learning_rate": 1.0560383733697415e-05,
"loss": 1.2821,
"step": 2690
},
{
"epoch": 0.9998148491020181,
"grad_norm": 22.125,
"learning_rate": 1.050050274400427e-05,
"loss": 1.3596,
"step": 2700
},
{
"epoch": 1.0033327161636734,
"grad_norm": 11.0625,
"learning_rate": 1.0440603756979885e-05,
"loss": 1.565,
"step": 2710
},
{
"epoch": 1.0070357341233105,
"grad_norm": 11.0625,
"learning_rate": 1.0380688926502377e-05,
"loss": 1.1739,
"step": 2720
},
{
"epoch": 1.0107387520829476,
"grad_norm": 11.5625,
"learning_rate": 1.0320760407019584e-05,
"loss": 1.0515,
"step": 2730
},
{
"epoch": 1.0144417700425847,
"grad_norm": 11.4375,
"learning_rate": 1.0260820353471563e-05,
"loss": 1.0064,
"step": 2740
},
{
"epoch": 1.0181447880022219,
"grad_norm": 12.625,
"learning_rate": 1.0200870921213134e-05,
"loss": 1.0337,
"step": 2750
},
{
"epoch": 1.021847805961859,
"grad_norm": 13.6875,
"learning_rate": 1.0140914265936346e-05,
"loss": 1.5197,
"step": 2760
},
{
"epoch": 1.025550823921496,
"grad_norm": 11.8125,
"learning_rate": 1.0080952543592992e-05,
"loss": 1.2048,
"step": 2770
},
{
"epoch": 1.0292538418811332,
"grad_norm": 11.6875,
"learning_rate": 1.0020987910317066e-05,
"loss": 1.1073,
"step": 2780
},
{
"epoch": 1.0329568598407701,
"grad_norm": 11.4375,
"learning_rate": 9.961022522347226e-06,
"loss": 1.0716,
"step": 2790
},
{
"epoch": 1.0366598778004072,
"grad_norm": 17.25,
"learning_rate": 9.901058535949273e-06,
"loss": 1.078,
"step": 2800
},
{
"epoch": 1.0403628957600444,
"grad_norm": 14.9375,
"learning_rate": 9.841098107338613e-06,
"loss": 1.4483,
"step": 2810
},
{
"epoch": 1.0440659137196815,
"grad_norm": 12.8125,
"learning_rate": 9.781143392602717e-06,
"loss": 1.2545,
"step": 2820
},
{
"epoch": 1.0477689316793186,
"grad_norm": 12.625,
"learning_rate": 9.721196547623585e-06,
"loss": 1.1239,
"step": 2830
},
{
"epoch": 1.0514719496389557,
"grad_norm": 12.0,
"learning_rate": 9.66125972800025e-06,
"loss": 1.0841,
"step": 2840
},
{
"epoch": 1.0551749675985929,
"grad_norm": 14.6875,
"learning_rate": 9.601335088971234e-06,
"loss": 1.0894,
"step": 2850
},
{
"epoch": 1.05887798555823,
"grad_norm": 15.0625,
"learning_rate": 9.541424785337069e-06,
"loss": 1.5052,
"step": 2860
},
{
"epoch": 1.062581003517867,
"grad_norm": 12.25,
"learning_rate": 9.481530971382796e-06,
"loss": 1.1952,
"step": 2870
},
{
"epoch": 1.0662840214775042,
"grad_norm": 12.9375,
"learning_rate": 9.421655800800536e-06,
"loss": 1.0996,
"step": 2880
},
{
"epoch": 1.0699870394371414,
"grad_norm": 12.1875,
"learning_rate": 9.361801426612e-06,
"loss": 1.0772,
"step": 2890
},
{
"epoch": 1.0736900573967785,
"grad_norm": 14.0,
"learning_rate": 9.301970001091097e-06,
"loss": 1.0725,
"step": 2900
},
{
"epoch": 1.0773930753564154,
"grad_norm": 13.75,
"learning_rate": 9.242163675686545e-06,
"loss": 1.5707,
"step": 2910
},
{
"epoch": 1.0810960933160525,
"grad_norm": 11.9375,
"learning_rate": 9.182384600944494e-06,
"loss": 1.2892,
"step": 2920
},
{
"epoch": 1.0847991112756896,
"grad_norm": 13.0625,
"learning_rate": 9.12263492643119e-06,
"loss": 1.1303,
"step": 2930
},
{
"epoch": 1.0885021292353267,
"grad_norm": 12.8125,
"learning_rate": 9.06291680065571e-06,
"loss": 1.0225,
"step": 2940
},
{
"epoch": 1.0922051471949639,
"grad_norm": 14.9375,
"learning_rate": 9.003232370992673e-06,
"loss": 1.0733,
"step": 2950
},
{
"epoch": 1.095908165154601,
"grad_norm": 12.9375,
"learning_rate": 8.943583783605039e-06,
"loss": 1.5533,
"step": 2960
},
{
"epoch": 1.099611183114238,
"grad_norm": 12.5625,
"learning_rate": 8.88397318336693e-06,
"loss": 1.2088,
"step": 2970
},
{
"epoch": 1.1033142010738752,
"grad_norm": 12.875,
"learning_rate": 8.824402713786516e-06,
"loss": 1.0778,
"step": 2980
},
{
"epoch": 1.1070172190335124,
"grad_norm": 12.125,
"learning_rate": 8.764874516928922e-06,
"loss": 1.0287,
"step": 2990
},
{
"epoch": 1.1107202369931495,
"grad_norm": 14.625,
"learning_rate": 8.705390733339204e-06,
"loss": 1.0832,
"step": 3000
},
{
"epoch": 1.1107202369931495,
"eval_loss": 1.3891929388046265,
"eval_runtime": 201.7109,
"eval_samples_per_second": 26.766,
"eval_steps_per_second": 3.346,
"step": 3000
},
{
"epoch": 1.1144232549527866,
"grad_norm": 15.75,
"learning_rate": 8.645953501965397e-06,
"loss": 1.4455,
"step": 3010
},
{
"epoch": 1.1181262729124237,
"grad_norm": 13.25,
"learning_rate": 8.586564960081574e-06,
"loss": 1.2385,
"step": 3020
},
{
"epoch": 1.1218292908720606,
"grad_norm": 12.1875,
"learning_rate": 8.527227243211012e-06,
"loss": 1.071,
"step": 3030
},
{
"epoch": 1.1255323088316977,
"grad_norm": 12.125,
"learning_rate": 8.467942485049403e-06,
"loss": 1.0758,
"step": 3040
},
{
"epoch": 1.1292353267913349,
"grad_norm": 15.6875,
"learning_rate": 8.408712817388113e-06,
"loss": 1.1003,
"step": 3050
},
{
"epoch": 1.132938344750972,
"grad_norm": 13.5,
"learning_rate": 8.34954037003754e-06,
"loss": 1.4862,
"step": 3060
},
{
"epoch": 1.1366413627106091,
"grad_norm": 12.875,
"learning_rate": 8.290427270750519e-06,
"loss": 1.1732,
"step": 3070
},
{
"epoch": 1.1403443806702462,
"grad_norm": 12.0,
"learning_rate": 8.231375645145833e-06,
"loss": 1.1398,
"step": 3080
},
{
"epoch": 1.1440473986298834,
"grad_norm": 12.375,
"learning_rate": 8.17238761663175e-06,
"loss": 1.017,
"step": 3090
},
{
"epoch": 1.1477504165895205,
"grad_norm": 15.5,
"learning_rate": 8.11346530632968e-06,
"loss": 1.0629,
"step": 3100
},
{
"epoch": 1.1514534345491576,
"grad_norm": 11.25,
"learning_rate": 8.054610832997916e-06,
"loss": 1.45,
"step": 3110
},
{
"epoch": 1.1551564525087947,
"grad_norm": 11.625,
"learning_rate": 7.995826312955423e-06,
"loss": 1.2192,
"step": 3120
},
{
"epoch": 1.1588594704684319,
"grad_norm": 12.5625,
"learning_rate": 7.937113860005754e-06,
"loss": 1.0502,
"step": 3130
},
{
"epoch": 1.162562488428069,
"grad_norm": 12.625,
"learning_rate": 7.878475585361045e-06,
"loss": 1.0645,
"step": 3140
},
{
"epoch": 1.166265506387706,
"grad_norm": 14.0,
"learning_rate": 7.819913597566078e-06,
"loss": 1.065,
"step": 3150
},
{
"epoch": 1.169968524347343,
"grad_norm": 12.875,
"learning_rate": 7.761430002422482e-06,
"loss": 1.4785,
"step": 3160
},
{
"epoch": 1.1736715423069801,
"grad_norm": 13.8125,
"learning_rate": 7.703026902912998e-06,
"loss": 1.2364,
"step": 3170
},
{
"epoch": 1.1773745602666172,
"grad_norm": 13.8125,
"learning_rate": 7.644706399125871e-06,
"loss": 1.0796,
"step": 3180
},
{
"epoch": 1.1810775782262544,
"grad_norm": 13.8125,
"learning_rate": 7.586470588179324e-06,
"loss": 1.0625,
"step": 3190
},
{
"epoch": 1.1847805961858915,
"grad_norm": 12.875,
"learning_rate": 7.528321564146146e-06,
"loss": 1.0216,
"step": 3200
},
{
"epoch": 1.1884836141455286,
"grad_norm": 12.5,
"learning_rate": 7.470261417978411e-06,
"loss": 1.4834,
"step": 3210
},
{
"epoch": 1.1921866321051657,
"grad_norm": 11.3125,
"learning_rate": 7.412292237432266e-06,
"loss": 1.2316,
"step": 3220
},
{
"epoch": 1.1958896500648029,
"grad_norm": 13.4375,
"learning_rate": 7.354416106992872e-06,
"loss": 1.0373,
"step": 3230
},
{
"epoch": 1.19959266802444,
"grad_norm": 11.625,
"learning_rate": 7.296635107799454e-06,
"loss": 1.1084,
"step": 3240
},
{
"epoch": 1.203295685984077,
"grad_norm": 12.8125,
"learning_rate": 7.2389513175704555e-06,
"loss": 1.0169,
"step": 3250
},
{
"epoch": 1.2069987039437142,
"grad_norm": 12.625,
"learning_rate": 7.181366810528827e-06,
"loss": 1.4541,
"step": 3260
},
{
"epoch": 1.2107017219033511,
"grad_norm": 11.375,
"learning_rate": 7.1238836573274485e-06,
"loss": 1.1976,
"step": 3270
},
{
"epoch": 1.2144047398629882,
"grad_norm": 14.1875,
"learning_rate": 7.066503924974663e-06,
"loss": 1.099,
"step": 3280
},
{
"epoch": 1.2181077578226254,
"grad_norm": 10.375,
"learning_rate": 7.009229676759963e-06,
"loss": 1.0454,
"step": 3290
},
{
"epoch": 1.2218107757822625,
"grad_norm": 15.5,
"learning_rate": 6.952062972179778e-06,
"loss": 1.0566,
"step": 3300
},
{
"epoch": 1.2255137937418996,
"grad_norm": 13.75,
"learning_rate": 6.895005866863439e-06,
"loss": 1.4755,
"step": 3310
},
{
"epoch": 1.2292168117015367,
"grad_norm": 11.5625,
"learning_rate": 6.83806041249924e-06,
"loss": 1.2692,
"step": 3320
},
{
"epoch": 1.2329198296611739,
"grad_norm": 14.1875,
"learning_rate": 6.781228656760683e-06,
"loss": 1.1042,
"step": 3330
},
{
"epoch": 1.236622847620811,
"grad_norm": 12.0625,
"learning_rate": 6.724512643232834e-06,
"loss": 1.0694,
"step": 3340
},
{
"epoch": 1.240325865580448,
"grad_norm": 15.6875,
"learning_rate": 6.667914411338843e-06,
"loss": 1.0776,
"step": 3350
},
{
"epoch": 1.2440288835400852,
"grad_norm": 13.375,
"learning_rate": 6.611435996266605e-06,
"loss": 1.4464,
"step": 3360
},
{
"epoch": 1.2477319014997224,
"grad_norm": 13.5,
"learning_rate": 6.555079428895584e-06,
"loss": 1.1789,
"step": 3370
},
{
"epoch": 1.2514349194593595,
"grad_norm": 12.3125,
"learning_rate": 6.4988467357237805e-06,
"loss": 1.0525,
"step": 3380
},
{
"epoch": 1.2551379374189966,
"grad_norm": 12.875,
"learning_rate": 6.442739938794867e-06,
"loss": 1.0489,
"step": 3390
},
{
"epoch": 1.2588409553786337,
"grad_norm": 15.1875,
"learning_rate": 6.386761055625472e-06,
"loss": 1.0531,
"step": 3400
},
{
"epoch": 1.2625439733382706,
"grad_norm": 13.125,
"learning_rate": 6.33091209913264e-06,
"loss": 1.469,
"step": 3410
},
{
"epoch": 1.2662469912979077,
"grad_norm": 12.3125,
"learning_rate": 6.275195077561441e-06,
"loss": 1.1747,
"step": 3420
},
{
"epoch": 1.2699500092575449,
"grad_norm": 12.0,
"learning_rate": 6.219611994412762e-06,
"loss": 1.0746,
"step": 3430
},
{
"epoch": 1.273653027217182,
"grad_norm": 13.75,
"learning_rate": 6.1641648483712755e-06,
"loss": 1.0788,
"step": 3440
},
{
"epoch": 1.277356045176819,
"grad_norm": 14.375,
"learning_rate": 6.108855633233546e-06,
"loss": 1.0313,
"step": 3450
},
{
"epoch": 1.2810590631364562,
"grad_norm": 14.1875,
"learning_rate": 6.053686337836352e-06,
"loss": 1.5169,
"step": 3460
},
{
"epoch": 1.2847620810960934,
"grad_norm": 12.75,
"learning_rate": 5.998658945985165e-06,
"loss": 1.2318,
"step": 3470
},
{
"epoch": 1.2884650990557305,
"grad_norm": 12.0,
"learning_rate": 5.94377543638283e-06,
"loss": 1.0356,
"step": 3480
},
{
"epoch": 1.2921681170153676,
"grad_norm": 12.375,
"learning_rate": 5.889037782558384e-06,
"loss": 1.0159,
"step": 3490
},
{
"epoch": 1.2958711349750045,
"grad_norm": 16.125,
"learning_rate": 5.834447952796114e-06,
"loss": 1.0402,
"step": 3500
},
{
"epoch": 1.2958711349750045,
"eval_loss": 1.3764488697052002,
"eval_runtime": 202.0067,
"eval_samples_per_second": 26.727,
"eval_steps_per_second": 3.341,
"step": 3500
},
{
"epoch": 1.2995741529346416,
"grad_norm": 14.75,
"learning_rate": 5.780007910064782e-06,
"loss": 1.456,
"step": 3510
},
{
"epoch": 1.3032771708942787,
"grad_norm": 13.3125,
"learning_rate": 5.725719611947021e-06,
"loss": 1.2757,
"step": 3520
},
{
"epoch": 1.3069801888539159,
"grad_norm": 11.625,
"learning_rate": 5.6715850105689585e-06,
"loss": 1.1012,
"step": 3530
},
{
"epoch": 1.310683206813553,
"grad_norm": 13.375,
"learning_rate": 5.617606052530022e-06,
"loss": 1.0308,
"step": 3540
},
{
"epoch": 1.3143862247731901,
"grad_norm": 14.5625,
"learning_rate": 5.563784678832933e-06,
"loss": 1.0085,
"step": 3550
},
{
"epoch": 1.3180892427328272,
"grad_norm": 13.5,
"learning_rate": 5.510122824813909e-06,
"loss": 1.5282,
"step": 3560
},
{
"epoch": 1.3217922606924644,
"grad_norm": 13.125,
"learning_rate": 5.456622420073084e-06,
"loss": 1.2315,
"step": 3570
},
{
"epoch": 1.3254952786521015,
"grad_norm": 13.125,
"learning_rate": 5.403285388405122e-06,
"loss": 1.0705,
"step": 3580
},
{
"epoch": 1.3291982966117386,
"grad_norm": 12.5625,
"learning_rate": 5.350113647730029e-06,
"loss": 1.0469,
"step": 3590
},
{
"epoch": 1.3329013145713757,
"grad_norm": 14.125,
"learning_rate": 5.297109110024189e-06,
"loss": 1.0423,
"step": 3600
},
{
"epoch": 1.3366043325310129,
"grad_norm": 13.5625,
"learning_rate": 5.244273681251633e-06,
"loss": 1.4896,
"step": 3610
},
{
"epoch": 1.34030735049065,
"grad_norm": 12.625,
"learning_rate": 5.19160926129547e-06,
"loss": 1.1438,
"step": 3620
},
{
"epoch": 1.344010368450287,
"grad_norm": 12.5,
"learning_rate": 5.139117743889592e-06,
"loss": 1.0885,
"step": 3630
},
{
"epoch": 1.3477133864099242,
"grad_norm": 10.75,
"learning_rate": 5.086801016550585e-06,
"loss": 1.005,
"step": 3640
},
{
"epoch": 1.3514164043695611,
"grad_norm": 17.0,
"learning_rate": 5.034660960509832e-06,
"loss": 1.0636,
"step": 3650
},
{
"epoch": 1.3551194223291982,
"grad_norm": 13.5,
"learning_rate": 4.982699450645882e-06,
"loss": 1.4411,
"step": 3660
},
{
"epoch": 1.3588224402888354,
"grad_norm": 11.6875,
"learning_rate": 4.930918355417033e-06,
"loss": 1.144,
"step": 3670
},
{
"epoch": 1.3625254582484725,
"grad_norm": 14.3125,
"learning_rate": 4.879319536794144e-06,
"loss": 1.0613,
"step": 3680
},
{
"epoch": 1.3662284762081096,
"grad_norm": 11.375,
"learning_rate": 4.827904850193675e-06,
"loss": 1.0702,
"step": 3690
},
{
"epoch": 1.3699314941677467,
"grad_norm": 14.125,
"learning_rate": 4.776676144410973e-06,
"loss": 1.0377,
"step": 3700
},
{
"epoch": 1.3736345121273839,
"grad_norm": 11.625,
"learning_rate": 4.7256352615538e-06,
"loss": 1.3927,
"step": 3710
},
{
"epoch": 1.377337530087021,
"grad_norm": 11.6875,
"learning_rate": 4.67478403697608e-06,
"loss": 1.2107,
"step": 3720
},
{
"epoch": 1.381040548046658,
"grad_norm": 11.375,
"learning_rate": 4.624124299211903e-06,
"loss": 1.0376,
"step": 3730
},
{
"epoch": 1.384743566006295,
"grad_norm": 13.625,
"learning_rate": 4.573657869909796e-06,
"loss": 1.1002,
"step": 3740
},
{
"epoch": 1.3884465839659321,
"grad_norm": 16.75,
"learning_rate": 4.523386563767183e-06,
"loss": 1.1041,
"step": 3750
},
{
"epoch": 1.3921496019255692,
"grad_norm": 12.375,
"learning_rate": 4.4733121884651665e-06,
"loss": 1.4119,
"step": 3760
},
{
"epoch": 1.3958526198852064,
"grad_norm": 13.0625,
"learning_rate": 4.423436544603494e-06,
"loss": 1.1938,
"step": 3770
},
{
"epoch": 1.3995556378448435,
"grad_norm": 12.0625,
"learning_rate": 4.373761425635846e-06,
"loss": 1.0435,
"step": 3780
},
{
"epoch": 1.4032586558044806,
"grad_norm": 12.8125,
"learning_rate": 4.324288617805315e-06,
"loss": 1.0658,
"step": 3790
},
{
"epoch": 1.4069616737641177,
"grad_norm": 14.4375,
"learning_rate": 4.275019900080183e-06,
"loss": 1.0571,
"step": 3800
},
{
"epoch": 1.4106646917237549,
"grad_norm": 13.125,
"learning_rate": 4.225957044089973e-06,
"loss": 1.4838,
"step": 3810
},
{
"epoch": 1.414367709683392,
"grad_norm": 13.5,
"learning_rate": 4.1771018140617125e-06,
"loss": 1.2512,
"step": 3820
},
{
"epoch": 1.418070727643029,
"grad_norm": 12.75,
"learning_rate": 4.128455966756512e-06,
"loss": 1.0584,
"step": 3830
},
{
"epoch": 1.4217737456026662,
"grad_norm": 12.6875,
"learning_rate": 4.080021251406403e-06,
"loss": 1.0343,
"step": 3840
},
{
"epoch": 1.4254767635623034,
"grad_norm": 14.5,
"learning_rate": 4.031799409651413e-06,
"loss": 1.0836,
"step": 3850
},
{
"epoch": 1.4291797815219405,
"grad_norm": 11.625,
"learning_rate": 3.98379217547696e-06,
"loss": 1.4869,
"step": 3860
},
{
"epoch": 1.4328827994815776,
"grad_norm": 13.0625,
"learning_rate": 3.936001275151489e-06,
"loss": 1.2013,
"step": 3870
},
{
"epoch": 1.4365858174412147,
"grad_norm": 12.25,
"learning_rate": 3.888428427164413e-06,
"loss": 1.0757,
"step": 3880
},
{
"epoch": 1.4402888354008516,
"grad_norm": 11.6875,
"learning_rate": 3.8410753421643e-06,
"loss": 1.0011,
"step": 3890
},
{
"epoch": 1.4439918533604887,
"grad_norm": 15.6875,
"learning_rate": 3.7939437228973664e-06,
"loss": 1.0911,
"step": 3900
},
{
"epoch": 1.4476948713201259,
"grad_norm": 13.25,
"learning_rate": 3.747035264146264e-06,
"loss": 1.4666,
"step": 3910
},
{
"epoch": 1.451397889279763,
"grad_norm": 13.8125,
"learning_rate": 3.7003516526691142e-06,
"loss": 1.2087,
"step": 3920
},
{
"epoch": 1.4551009072394,
"grad_norm": 13.75,
"learning_rate": 3.653894567138867e-06,
"loss": 1.0964,
"step": 3930
},
{
"epoch": 1.4588039251990372,
"grad_norm": 12.25,
"learning_rate": 3.607665678082948e-06,
"loss": 1.0539,
"step": 3940
},
{
"epoch": 1.4625069431586744,
"grad_norm": 14.25,
"learning_rate": 3.561666647823165e-06,
"loss": 1.0532,
"step": 3950
},
{
"epoch": 1.4662099611183115,
"grad_norm": 13.875,
"learning_rate": 3.5158991304159572e-06,
"loss": 1.5298,
"step": 3960
},
{
"epoch": 1.4699129790779486,
"grad_norm": 12.25,
"learning_rate": 3.4703647715928977e-06,
"loss": 1.2145,
"step": 3970
},
{
"epoch": 1.4736159970375855,
"grad_norm": 12.1875,
"learning_rate": 3.425065208701537e-06,
"loss": 1.0814,
"step": 3980
},
{
"epoch": 1.4773190149972226,
"grad_norm": 12.0,
"learning_rate": 3.380002070646503e-06,
"loss": 1.0524,
"step": 3990
},
{
"epoch": 1.4810220329568597,
"grad_norm": 14.5625,
"learning_rate": 3.335176977830945e-06,
"loss": 1.0256,
"step": 4000
},
{
"epoch": 1.4810220329568597,
"eval_loss": 1.37130868434906,
"eval_runtime": 201.4424,
"eval_samples_per_second": 26.802,
"eval_steps_per_second": 3.351,
"step": 4000
}
],
"logging_steps": 10,
"max_steps": 5402,
"num_input_tokens_seen": 0,
"num_train_epochs": 2,
"save_steps": 500,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": false
},
"attributes": {}
}
},
"total_flos": 2.5816335729356006e+17,
"train_batch_size": 8,
"trial_name": null,
"trial_params": null
}