Files
MPropositioneur-V2-large/trainer_state.json

4495 lines
104 KiB
JSON
Raw Normal View History

{
"best_global_step": 6032,
"best_metric": 0.11902373284101486,
"best_model_checkpoint": "models/Qwen3-4B-distilled/checkpoint-6032",
"epoch": 0.9632705206004472,
"eval_steps": 208,
"global_step": 6032,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"epoch": 0.00015969338869370808,
"grad_norm": 12.6875,
"learning_rate": 0.0,
"loss": 0.6257,
"step": 1
},
{
"epoch": 0.0015969338869370809,
"grad_norm": 12.625,
"learning_rate": 1.8000000000000001e-06,
"loss": 0.6647,
"step": 10
},
{
"epoch": 0.0031938677738741618,
"grad_norm": 7.78125,
"learning_rate": 3.8000000000000005e-06,
"loss": 0.5401,
"step": 20
},
{
"epoch": 0.0047908016608112424,
"grad_norm": 4.6875,
"learning_rate": 5.8e-06,
"loss": 0.3113,
"step": 30
},
{
"epoch": 0.0063877355477483235,
"grad_norm": 4.90625,
"learning_rate": 7.800000000000002e-06,
"loss": 0.222,
"step": 40
},
{
"epoch": 0.007984669434685404,
"grad_norm": 3.234375,
"learning_rate": 9.800000000000001e-06,
"loss": 0.2135,
"step": 50
},
{
"epoch": 0.009581603321622485,
"grad_norm": 4.46875,
"learning_rate": 1.18e-05,
"loss": 0.1899,
"step": 60
},
{
"epoch": 0.011178537208559566,
"grad_norm": 2.609375,
"learning_rate": 1.38e-05,
"loss": 0.2244,
"step": 70
},
{
"epoch": 0.012775471095496647,
"grad_norm": 2.5625,
"learning_rate": 1.58e-05,
"loss": 0.169,
"step": 80
},
{
"epoch": 0.014372404982433726,
"grad_norm": 2.984375,
"learning_rate": 1.7800000000000002e-05,
"loss": 0.1712,
"step": 90
},
{
"epoch": 0.015969338869370808,
"grad_norm": 4.90625,
"learning_rate": 1.98e-05,
"loss": 0.2008,
"step": 100
},
{
"epoch": 0.01756627275630789,
"grad_norm": 3.03125,
"learning_rate": 1.998551191242756e-05,
"loss": 0.1738,
"step": 110
},
{
"epoch": 0.01916320664324497,
"grad_norm": 2.59375,
"learning_rate": 1.9969414037347072e-05,
"loss": 0.1831,
"step": 120
},
{
"epoch": 0.02076014053018205,
"grad_norm": 3.109375,
"learning_rate": 1.9953316162266584e-05,
"loss": 0.2049,
"step": 130
},
{
"epoch": 0.022357074417119132,
"grad_norm": 4.9375,
"learning_rate": 1.9937218287186093e-05,
"loss": 0.1702,
"step": 140
},
{
"epoch": 0.023954008304056213,
"grad_norm": 4.125,
"learning_rate": 1.9921120412105604e-05,
"loss": 0.1455,
"step": 150
},
{
"epoch": 0.025550942190993294,
"grad_norm": 4.46875,
"learning_rate": 1.9905022537025113e-05,
"loss": 0.2209,
"step": 160
},
{
"epoch": 0.027147876077930375,
"grad_norm": 2.671875,
"learning_rate": 1.9888924661944625e-05,
"loss": 0.183,
"step": 170
},
{
"epoch": 0.028744809964867453,
"grad_norm": 1.8671875,
"learning_rate": 1.9872826786864133e-05,
"loss": 0.1661,
"step": 180
},
{
"epoch": 0.030341743851804534,
"grad_norm": 2.75,
"learning_rate": 1.9856728911783648e-05,
"loss": 0.1617,
"step": 190
},
{
"epoch": 0.031938677738741615,
"grad_norm": 10.5,
"learning_rate": 1.9840631036703157e-05,
"loss": 0.2303,
"step": 200
},
{
"epoch": 0.03321622484829128,
"eval_loss": 0.16591905057430267,
"eval_runtime": 27.9572,
"eval_samples_per_second": 9.085,
"eval_steps_per_second": 9.085,
"step": 208
},
{
"epoch": 0.033535611625678696,
"grad_norm": 2.90625,
"learning_rate": 1.982453316162267e-05,
"loss": 0.1539,
"step": 210
},
{
"epoch": 0.03513254551261578,
"grad_norm": 2.265625,
"learning_rate": 1.9808435286542177e-05,
"loss": 0.1612,
"step": 220
},
{
"epoch": 0.03672947939955286,
"grad_norm": 2.125,
"learning_rate": 1.979233741146169e-05,
"loss": 0.1427,
"step": 230
},
{
"epoch": 0.03832641328648994,
"grad_norm": 3.328125,
"learning_rate": 1.97762395363812e-05,
"loss": 0.1668,
"step": 240
},
{
"epoch": 0.03992334717342702,
"grad_norm": 2.09375,
"learning_rate": 1.976014166130071e-05,
"loss": 0.1711,
"step": 250
},
{
"epoch": 0.0415202810603641,
"grad_norm": 3.203125,
"learning_rate": 1.974404378622022e-05,
"loss": 0.1576,
"step": 260
},
{
"epoch": 0.04311721494730118,
"grad_norm": 2.84375,
"learning_rate": 1.9727945911139732e-05,
"loss": 0.1637,
"step": 270
},
{
"epoch": 0.044714148834238264,
"grad_norm": 2.359375,
"learning_rate": 1.971184803605924e-05,
"loss": 0.1687,
"step": 280
},
{
"epoch": 0.046311082721175345,
"grad_norm": 3.265625,
"learning_rate": 1.9695750160978753e-05,
"loss": 0.146,
"step": 290
},
{
"epoch": 0.047908016608112426,
"grad_norm": 2.109375,
"learning_rate": 1.9679652285898264e-05,
"loss": 0.1804,
"step": 300
},
{
"epoch": 0.04950495049504951,
"grad_norm": 4.21875,
"learning_rate": 1.9663554410817773e-05,
"loss": 0.1641,
"step": 310
},
{
"epoch": 0.05110188438198659,
"grad_norm": 2.296875,
"learning_rate": 1.9647456535737284e-05,
"loss": 0.1628,
"step": 320
},
{
"epoch": 0.05269881826892367,
"grad_norm": 3.078125,
"learning_rate": 1.9631358660656793e-05,
"loss": 0.1911,
"step": 330
},
{
"epoch": 0.05429575215586075,
"grad_norm": 3.515625,
"learning_rate": 1.9615260785576308e-05,
"loss": 0.2062,
"step": 340
},
{
"epoch": 0.055892686042797825,
"grad_norm": 1.84375,
"learning_rate": 1.9599162910495816e-05,
"loss": 0.1437,
"step": 350
},
{
"epoch": 0.057489619929734906,
"grad_norm": 2.40625,
"learning_rate": 1.9583065035415328e-05,
"loss": 0.1527,
"step": 360
},
{
"epoch": 0.05908655381667199,
"grad_norm": 1.1953125,
"learning_rate": 1.9566967160334837e-05,
"loss": 0.1472,
"step": 370
},
{
"epoch": 0.06068348770360907,
"grad_norm": 3.28125,
"learning_rate": 1.955086928525435e-05,
"loss": 0.1737,
"step": 380
},
{
"epoch": 0.06228042159054615,
"grad_norm": 2.75,
"learning_rate": 1.9534771410173857e-05,
"loss": 0.1426,
"step": 390
},
{
"epoch": 0.06387735547748323,
"grad_norm": 1.953125,
"learning_rate": 1.951867353509337e-05,
"loss": 0.1816,
"step": 400
},
{
"epoch": 0.06547428936442032,
"grad_norm": 2.046875,
"learning_rate": 1.950257566001288e-05,
"loss": 0.1518,
"step": 410
},
{
"epoch": 0.06643244969658256,
"eval_loss": 0.14999131858348846,
"eval_runtime": 24.8593,
"eval_samples_per_second": 10.218,
"eval_steps_per_second": 10.218,
"step": 416
},
{
"epoch": 0.06707122325135739,
"grad_norm": 2.40625,
"learning_rate": 1.9486477784932392e-05,
"loss": 0.1484,
"step": 420
},
{
"epoch": 0.06866815713829448,
"grad_norm": 2.34375,
"learning_rate": 1.94703799098519e-05,
"loss": 0.1488,
"step": 430
},
{
"epoch": 0.07026509102523155,
"grad_norm": 3.09375,
"learning_rate": 1.9454282034771412e-05,
"loss": 0.1691,
"step": 440
},
{
"epoch": 0.07186202491216864,
"grad_norm": 1.6796875,
"learning_rate": 1.9438184159690924e-05,
"loss": 0.1254,
"step": 450
},
{
"epoch": 0.07345895879910572,
"grad_norm": 14.875,
"learning_rate": 1.9422086284610433e-05,
"loss": 0.1505,
"step": 460
},
{
"epoch": 0.07505589268604279,
"grad_norm": 2.5,
"learning_rate": 1.9405988409529944e-05,
"loss": 0.2086,
"step": 470
},
{
"epoch": 0.07665282657297988,
"grad_norm": 4.5625,
"learning_rate": 1.9389890534449453e-05,
"loss": 0.2219,
"step": 480
},
{
"epoch": 0.07824976045991695,
"grad_norm": 2.03125,
"learning_rate": 1.9373792659368965e-05,
"loss": 0.1457,
"step": 490
},
{
"epoch": 0.07984669434685404,
"grad_norm": 2.53125,
"learning_rate": 1.9357694784288476e-05,
"loss": 0.1797,
"step": 500
},
{
"epoch": 0.08144362823379112,
"grad_norm": 2.28125,
"learning_rate": 1.9341596909207988e-05,
"loss": 0.1732,
"step": 510
},
{
"epoch": 0.0830405621207282,
"grad_norm": 2.25,
"learning_rate": 1.9325499034127497e-05,
"loss": 0.1713,
"step": 520
},
{
"epoch": 0.08463749600766528,
"grad_norm": 1.78125,
"learning_rate": 1.930940115904701e-05,
"loss": 0.1816,
"step": 530
},
{
"epoch": 0.08623442989460237,
"grad_norm": 4.625,
"learning_rate": 1.9293303283966517e-05,
"loss": 0.1367,
"step": 540
},
{
"epoch": 0.08783136378153944,
"grad_norm": 2.453125,
"learning_rate": 1.927720540888603e-05,
"loss": 0.133,
"step": 550
},
{
"epoch": 0.08942829766847653,
"grad_norm": 2.484375,
"learning_rate": 1.9261107533805537e-05,
"loss": 0.1537,
"step": 560
},
{
"epoch": 0.0910252315554136,
"grad_norm": 1.796875,
"learning_rate": 1.924500965872505e-05,
"loss": 0.1339,
"step": 570
},
{
"epoch": 0.09262216544235069,
"grad_norm": 3.015625,
"learning_rate": 1.922891178364456e-05,
"loss": 0.1954,
"step": 580
},
{
"epoch": 0.09421909932928776,
"grad_norm": 2.75,
"learning_rate": 1.9212813908564072e-05,
"loss": 0.1967,
"step": 590
},
{
"epoch": 0.09581603321622485,
"grad_norm": 1.8671875,
"learning_rate": 1.919671603348358e-05,
"loss": 0.1439,
"step": 600
},
{
"epoch": 0.09741296710316193,
"grad_norm": 2.65625,
"learning_rate": 1.9180618158403093e-05,
"loss": 0.1944,
"step": 610
},
{
"epoch": 0.09900990099009901,
"grad_norm": 3.265625,
"learning_rate": 1.9164520283322604e-05,
"loss": 0.187,
"step": 620
},
{
"epoch": 0.09964867454487385,
"eval_loss": 0.15051136910915375,
"eval_runtime": 24.887,
"eval_samples_per_second": 10.206,
"eval_steps_per_second": 10.206,
"step": 624
},
{
"epoch": 0.10060683487703609,
"grad_norm": 3.40625,
"learning_rate": 1.9148422408242113e-05,
"loss": 0.1642,
"step": 630
},
{
"epoch": 0.10220376876397318,
"grad_norm": 1.984375,
"learning_rate": 1.9132324533161625e-05,
"loss": 0.1487,
"step": 640
},
{
"epoch": 0.10380070265091025,
"grad_norm": 2.421875,
"learning_rate": 1.9116226658081133e-05,
"loss": 0.2103,
"step": 650
},
{
"epoch": 0.10539763653784734,
"grad_norm": 1.7734375,
"learning_rate": 1.9100128783000648e-05,
"loss": 0.1463,
"step": 660
},
{
"epoch": 0.10699457042478441,
"grad_norm": 2.671875,
"learning_rate": 1.9084030907920157e-05,
"loss": 0.1391,
"step": 670
},
{
"epoch": 0.1085915043117215,
"grad_norm": 2.875,
"learning_rate": 1.906793303283967e-05,
"loss": 0.1438,
"step": 680
},
{
"epoch": 0.11018843819865858,
"grad_norm": 2.671875,
"learning_rate": 1.9051835157759177e-05,
"loss": 0.1621,
"step": 690
},
{
"epoch": 0.11178537208559565,
"grad_norm": 2.21875,
"learning_rate": 1.903573728267869e-05,
"loss": 0.1676,
"step": 700
},
{
"epoch": 0.11338230597253274,
"grad_norm": 1.71875,
"learning_rate": 1.9019639407598197e-05,
"loss": 0.158,
"step": 710
},
{
"epoch": 0.11497923985946981,
"grad_norm": 2.40625,
"learning_rate": 1.900354153251771e-05,
"loss": 0.1683,
"step": 720
},
{
"epoch": 0.1165761737464069,
"grad_norm": 2.015625,
"learning_rate": 1.898744365743722e-05,
"loss": 0.14,
"step": 730
},
{
"epoch": 0.11817310763334397,
"grad_norm": 1.796875,
"learning_rate": 1.8971345782356732e-05,
"loss": 0.2339,
"step": 740
},
{
"epoch": 0.11977004152028106,
"grad_norm": 1.5703125,
"learning_rate": 1.895524790727624e-05,
"loss": 0.1421,
"step": 750
},
{
"epoch": 0.12136697540721814,
"grad_norm": 3.125,
"learning_rate": 1.8939150032195753e-05,
"loss": 0.1366,
"step": 760
},
{
"epoch": 0.12296390929415522,
"grad_norm": 2.296875,
"learning_rate": 1.892305215711526e-05,
"loss": 0.1489,
"step": 770
},
{
"epoch": 0.1245608431810923,
"grad_norm": 2.078125,
"learning_rate": 1.8906954282034773e-05,
"loss": 0.1286,
"step": 780
},
{
"epoch": 0.1261577770680294,
"grad_norm": 3.765625,
"learning_rate": 1.8890856406954285e-05,
"loss": 0.2018,
"step": 790
},
{
"epoch": 0.12775471095496646,
"grad_norm": 2.515625,
"learning_rate": 1.8874758531873793e-05,
"loss": 0.1647,
"step": 800
},
{
"epoch": 0.12935164484190353,
"grad_norm": 3.0625,
"learning_rate": 1.8858660656793305e-05,
"loss": 0.17,
"step": 810
},
{
"epoch": 0.13094857872884064,
"grad_norm": 3.09375,
"learning_rate": 1.8842562781712817e-05,
"loss": 0.129,
"step": 820
},
{
"epoch": 0.1325455126157777,
"grad_norm": 2.171875,
"learning_rate": 1.882646490663233e-05,
"loss": 0.1215,
"step": 830
},
{
"epoch": 0.13286489939316512,
"eval_loss": 0.14854636788368225,
"eval_runtime": 25.1024,
"eval_samples_per_second": 10.119,
"eval_steps_per_second": 10.119,
"step": 832
},
{
"epoch": 0.13414244650271478,
"grad_norm": 2.390625,
"learning_rate": 1.8810367031551837e-05,
"loss": 0.2387,
"step": 840
},
{
"epoch": 0.13573938038965186,
"grad_norm": 2.53125,
"learning_rate": 1.879426915647135e-05,
"loss": 0.1388,
"step": 850
},
{
"epoch": 0.13733631427658896,
"grad_norm": 1.4140625,
"learning_rate": 1.8778171281390857e-05,
"loss": 0.1406,
"step": 860
},
{
"epoch": 0.13893324816352604,
"grad_norm": 2.578125,
"learning_rate": 1.876207340631037e-05,
"loss": 0.168,
"step": 870
},
{
"epoch": 0.1405301820504631,
"grad_norm": 2.453125,
"learning_rate": 1.8745975531229877e-05,
"loss": 0.1512,
"step": 880
},
{
"epoch": 0.14212711593740018,
"grad_norm": 2.875,
"learning_rate": 1.872987765614939e-05,
"loss": 0.175,
"step": 890
},
{
"epoch": 0.14372404982433729,
"grad_norm": 2.09375,
"learning_rate": 1.87137797810689e-05,
"loss": 0.1097,
"step": 900
},
{
"epoch": 0.14532098371127436,
"grad_norm": 1.0390625,
"learning_rate": 1.8697681905988413e-05,
"loss": 0.1612,
"step": 910
},
{
"epoch": 0.14691791759821143,
"grad_norm": 5.21875,
"learning_rate": 1.868158403090792e-05,
"loss": 0.1997,
"step": 920
},
{
"epoch": 0.1485148514851485,
"grad_norm": 2.34375,
"learning_rate": 1.8665486155827433e-05,
"loss": 0.1474,
"step": 930
},
{
"epoch": 0.15011178537208558,
"grad_norm": 3.234375,
"learning_rate": 1.8649388280746945e-05,
"loss": 0.1423,
"step": 940
},
{
"epoch": 0.15170871925902268,
"grad_norm": 3.296875,
"learning_rate": 1.8633290405666453e-05,
"loss": 0.1634,
"step": 950
},
{
"epoch": 0.15330565314595976,
"grad_norm": 1.9609375,
"learning_rate": 1.8617192530585965e-05,
"loss": 0.155,
"step": 960
},
{
"epoch": 0.15490258703289683,
"grad_norm": 1.734375,
"learning_rate": 1.8601094655505473e-05,
"loss": 0.1417,
"step": 970
},
{
"epoch": 0.1564995209198339,
"grad_norm": 2.171875,
"learning_rate": 1.8584996780424985e-05,
"loss": 0.1517,
"step": 980
},
{
"epoch": 0.158096454806771,
"grad_norm": 1.8828125,
"learning_rate": 1.8568898905344497e-05,
"loss": 0.1403,
"step": 990
},
{
"epoch": 0.15969338869370808,
"grad_norm": 2.15625,
"learning_rate": 1.855280103026401e-05,
"loss": 0.131,
"step": 1000
},
{
"epoch": 0.16129032258064516,
"grad_norm": 2.84375,
"learning_rate": 1.8536703155183517e-05,
"loss": 0.1523,
"step": 1010
},
{
"epoch": 0.16288725646758223,
"grad_norm": 2.484375,
"learning_rate": 1.852060528010303e-05,
"loss": 0.1746,
"step": 1020
},
{
"epoch": 0.16448419035451933,
"grad_norm": 1.4609375,
"learning_rate": 1.8504507405022537e-05,
"loss": 0.1512,
"step": 1030
},
{
"epoch": 0.1660811242414564,
"grad_norm": 2.59375,
"learning_rate": 1.848840952994205e-05,
"loss": 0.1909,
"step": 1040
},
{
"epoch": 0.1660811242414564,
"eval_loss": 0.14355012774467468,
"eval_runtime": 24.9519,
"eval_samples_per_second": 10.18,
"eval_steps_per_second": 10.18,
"step": 1040
},
{
"epoch": 0.16767805812839348,
"grad_norm": 1.65625,
"learning_rate": 1.8472311654861557e-05,
"loss": 0.136,
"step": 1050
},
{
"epoch": 0.16927499201533056,
"grad_norm": 2.5,
"learning_rate": 1.8456213779781072e-05,
"loss": 0.1841,
"step": 1060
},
{
"epoch": 0.17087192590226766,
"grad_norm": 2.328125,
"learning_rate": 1.844011590470058e-05,
"loss": 0.1442,
"step": 1070
},
{
"epoch": 0.17246885978920473,
"grad_norm": 2.828125,
"learning_rate": 1.8424018029620093e-05,
"loss": 0.1289,
"step": 1080
},
{
"epoch": 0.1740657936761418,
"grad_norm": 2.609375,
"learning_rate": 1.84079201545396e-05,
"loss": 0.1409,
"step": 1090
},
{
"epoch": 0.17566272756307888,
"grad_norm": 2.109375,
"learning_rate": 1.8391822279459113e-05,
"loss": 0.1502,
"step": 1100
},
{
"epoch": 0.17725966145001598,
"grad_norm": 2.25,
"learning_rate": 1.8375724404378625e-05,
"loss": 0.1325,
"step": 1110
},
{
"epoch": 0.17885659533695306,
"grad_norm": 1.3828125,
"learning_rate": 1.8359626529298133e-05,
"loss": 0.1756,
"step": 1120
},
{
"epoch": 0.18045352922389013,
"grad_norm": 2.734375,
"learning_rate": 1.8343528654217645e-05,
"loss": 0.1764,
"step": 1130
},
{
"epoch": 0.1820504631108272,
"grad_norm": 2.453125,
"learning_rate": 1.8327430779137157e-05,
"loss": 0.1409,
"step": 1140
},
{
"epoch": 0.1836473969977643,
"grad_norm": 1.625,
"learning_rate": 1.8311332904056665e-05,
"loss": 0.1539,
"step": 1150
},
{
"epoch": 0.18524433088470138,
"grad_norm": 2.8125,
"learning_rate": 1.8295235028976177e-05,
"loss": 0.2219,
"step": 1160
},
{
"epoch": 0.18684126477163845,
"grad_norm": 2.203125,
"learning_rate": 1.827913715389569e-05,
"loss": 0.148,
"step": 1170
},
{
"epoch": 0.18843819865857553,
"grad_norm": 2.15625,
"learning_rate": 1.8263039278815197e-05,
"loss": 0.1724,
"step": 1180
},
{
"epoch": 0.1900351325455126,
"grad_norm": 1.8203125,
"learning_rate": 1.824694140373471e-05,
"loss": 0.1386,
"step": 1190
},
{
"epoch": 0.1916320664324497,
"grad_norm": 1.1171875,
"learning_rate": 1.8230843528654217e-05,
"loss": 0.1208,
"step": 1200
},
{
"epoch": 0.19322900031938678,
"grad_norm": 1.953125,
"learning_rate": 1.821474565357373e-05,
"loss": 0.1354,
"step": 1210
},
{
"epoch": 0.19482593420632385,
"grad_norm": 2.609375,
"learning_rate": 1.819864777849324e-05,
"loss": 0.1645,
"step": 1220
},
{
"epoch": 0.19642286809326093,
"grad_norm": 2.21875,
"learning_rate": 1.8182549903412753e-05,
"loss": 0.1784,
"step": 1230
},
{
"epoch": 0.19801980198019803,
"grad_norm": 3.125,
"learning_rate": 1.816645202833226e-05,
"loss": 0.1448,
"step": 1240
},
{
"epoch": 0.1992973490897477,
"eval_loss": 0.1424371749162674,
"eval_runtime": 27.9818,
"eval_samples_per_second": 9.077,
"eval_steps_per_second": 9.077,
"step": 1248
},
{
"epoch": 0.1996167358671351,
"grad_norm": 2.140625,
"learning_rate": 1.8150354153251773e-05,
"loss": 0.1623,
"step": 1250
},
{
"epoch": 0.20121366975407218,
"grad_norm": 1.7578125,
"learning_rate": 1.813425627817128e-05,
"loss": 0.1535,
"step": 1260
},
{
"epoch": 0.20281060364100925,
"grad_norm": 2.390625,
"learning_rate": 1.8118158403090793e-05,
"loss": 0.1965,
"step": 1270
},
{
"epoch": 0.20440753752794635,
"grad_norm": 1.6171875,
"learning_rate": 1.8102060528010305e-05,
"loss": 0.1248,
"step": 1280
},
{
"epoch": 0.20600447141488343,
"grad_norm": 1.4140625,
"learning_rate": 1.8085962652929813e-05,
"loss": 0.1499,
"step": 1290
},
{
"epoch": 0.2076014053018205,
"grad_norm": 1.5390625,
"learning_rate": 1.8069864777849325e-05,
"loss": 0.1418,
"step": 1300
},
{
"epoch": 0.20919833918875758,
"grad_norm": 1.6796875,
"learning_rate": 1.8053766902768837e-05,
"loss": 0.1419,
"step": 1310
},
{
"epoch": 0.21079527307569468,
"grad_norm": 2.71875,
"learning_rate": 1.803766902768835e-05,
"loss": 0.1777,
"step": 1320
},
{
"epoch": 0.21239220696263175,
"grad_norm": 2.65625,
"learning_rate": 1.8021571152607857e-05,
"loss": 0.1479,
"step": 1330
},
{
"epoch": 0.21398914084956883,
"grad_norm": 3.046875,
"learning_rate": 1.800547327752737e-05,
"loss": 0.1404,
"step": 1340
},
{
"epoch": 0.2155860747365059,
"grad_norm": 1.8828125,
"learning_rate": 1.7989375402446877e-05,
"loss": 0.1308,
"step": 1350
},
{
"epoch": 0.217183008623443,
"grad_norm": 2.078125,
"learning_rate": 1.797327752736639e-05,
"loss": 0.1311,
"step": 1360
},
{
"epoch": 0.21877994251038008,
"grad_norm": 2.21875,
"learning_rate": 1.7957179652285897e-05,
"loss": 0.1878,
"step": 1370
},
{
"epoch": 0.22037687639731715,
"grad_norm": 2.359375,
"learning_rate": 1.7941081777205413e-05,
"loss": 0.157,
"step": 1380
},
{
"epoch": 0.22197381028425422,
"grad_norm": 2.921875,
"learning_rate": 1.792498390212492e-05,
"loss": 0.1354,
"step": 1390
},
{
"epoch": 0.2235707441711913,
"grad_norm": 3.328125,
"learning_rate": 1.7908886027044433e-05,
"loss": 0.1754,
"step": 1400
},
{
"epoch": 0.2251676780581284,
"grad_norm": 2.84375,
"learning_rate": 1.789278815196394e-05,
"loss": 0.165,
"step": 1410
},
{
"epoch": 0.22676461194506548,
"grad_norm": 2.0625,
"learning_rate": 1.7876690276883453e-05,
"loss": 0.1801,
"step": 1420
},
{
"epoch": 0.22836154583200255,
"grad_norm": 2.28125,
"learning_rate": 1.786059240180296e-05,
"loss": 0.1754,
"step": 1430
},
{
"epoch": 0.22995847971893962,
"grad_norm": 1.515625,
"learning_rate": 1.7844494526722473e-05,
"loss": 0.1357,
"step": 1440
},
{
"epoch": 0.23155541360587673,
"grad_norm": 2.75,
"learning_rate": 1.7828396651641985e-05,
"loss": 0.1382,
"step": 1450
},
{
"epoch": 0.23251357393803895,
"eval_loss": 0.1425570845603943,
"eval_runtime": 27.8196,
"eval_samples_per_second": 9.13,
"eval_steps_per_second": 9.13,
"step": 1456
},
{
"epoch": 0.2331523474928138,
"grad_norm": 2.234375,
"learning_rate": 1.7812298776561497e-05,
"loss": 0.1441,
"step": 1460
},
{
"epoch": 0.23474928137975087,
"grad_norm": 2.578125,
"learning_rate": 1.7796200901481005e-05,
"loss": 0.2088,
"step": 1470
},
{
"epoch": 0.23634621526668795,
"grad_norm": 1.7890625,
"learning_rate": 1.7780103026400517e-05,
"loss": 0.1483,
"step": 1480
},
{
"epoch": 0.23794314915362505,
"grad_norm": 3.640625,
"learning_rate": 1.776400515132003e-05,
"loss": 0.1538,
"step": 1490
},
{
"epoch": 0.23954008304056212,
"grad_norm": 2.828125,
"learning_rate": 1.7747907276239537e-05,
"loss": 0.1343,
"step": 1500
},
{
"epoch": 0.2411370169274992,
"grad_norm": 3.515625,
"learning_rate": 1.773180940115905e-05,
"loss": 0.1574,
"step": 1510
},
{
"epoch": 0.24273395081443627,
"grad_norm": 1.8359375,
"learning_rate": 1.7715711526078557e-05,
"loss": 0.1441,
"step": 1520
},
{
"epoch": 0.24433088470137337,
"grad_norm": 1.4609375,
"learning_rate": 1.769961365099807e-05,
"loss": 0.1294,
"step": 1530
},
{
"epoch": 0.24592781858831045,
"grad_norm": 1.7890625,
"learning_rate": 1.768351577591758e-05,
"loss": 0.1687,
"step": 1540
},
{
"epoch": 0.24752475247524752,
"grad_norm": 2.015625,
"learning_rate": 1.7667417900837093e-05,
"loss": 0.1324,
"step": 1550
},
{
"epoch": 0.2491216863621846,
"grad_norm": 2.625,
"learning_rate": 1.76513200257566e-05,
"loss": 0.1509,
"step": 1560
},
{
"epoch": 0.25071862024912167,
"grad_norm": 1.96875,
"learning_rate": 1.7635222150676113e-05,
"loss": 0.1416,
"step": 1570
},
{
"epoch": 0.2523155541360588,
"grad_norm": 1.1796875,
"learning_rate": 1.761912427559562e-05,
"loss": 0.1323,
"step": 1580
},
{
"epoch": 0.2539124880229959,
"grad_norm": 1.796875,
"learning_rate": 1.7603026400515133e-05,
"loss": 0.1327,
"step": 1590
},
{
"epoch": 0.2555094219099329,
"grad_norm": 1.9453125,
"learning_rate": 1.7586928525434645e-05,
"loss": 0.1575,
"step": 1600
},
{
"epoch": 0.25710635579687,
"grad_norm": 2.171875,
"learning_rate": 1.7570830650354153e-05,
"loss": 0.1841,
"step": 1610
},
{
"epoch": 0.25870328968380707,
"grad_norm": 3.03125,
"learning_rate": 1.7554732775273665e-05,
"loss": 0.1705,
"step": 1620
},
{
"epoch": 0.26030022357074417,
"grad_norm": 2.25,
"learning_rate": 1.7538634900193177e-05,
"loss": 0.1279,
"step": 1630
},
{
"epoch": 0.2618971574576813,
"grad_norm": 1.625,
"learning_rate": 1.7522537025112685e-05,
"loss": 0.1626,
"step": 1640
},
{
"epoch": 0.2634940913446183,
"grad_norm": 2.03125,
"learning_rate": 1.7506439150032197e-05,
"loss": 0.1391,
"step": 1650
},
{
"epoch": 0.2650910252315554,
"grad_norm": 3.90625,
"learning_rate": 1.749034127495171e-05,
"loss": 0.1776,
"step": 1660
},
{
"epoch": 0.26572979878633024,
"eval_loss": 0.14302664995193481,
"eval_runtime": 28.1289,
"eval_samples_per_second": 9.03,
"eval_steps_per_second": 9.03,
"step": 1664
},
{
"epoch": 0.26668795911849247,
"grad_norm": 2.015625,
"learning_rate": 1.7474243399871217e-05,
"loss": 0.1512,
"step": 1670
},
{
"epoch": 0.26828489300542957,
"grad_norm": 1.3515625,
"learning_rate": 1.745814552479073e-05,
"loss": 0.1247,
"step": 1680
},
{
"epoch": 0.26988182689236667,
"grad_norm": 2.421875,
"learning_rate": 1.7442047649710238e-05,
"loss": 0.1974,
"step": 1690
},
{
"epoch": 0.2714787607793037,
"grad_norm": 3.796875,
"learning_rate": 1.7425949774629753e-05,
"loss": 0.1485,
"step": 1700
},
{
"epoch": 0.2730756946662408,
"grad_norm": 2.734375,
"learning_rate": 1.740985189954926e-05,
"loss": 0.1421,
"step": 1710
},
{
"epoch": 0.2746726285531779,
"grad_norm": 2.5625,
"learning_rate": 1.7393754024468773e-05,
"loss": 0.1561,
"step": 1720
},
{
"epoch": 0.27626956244011497,
"grad_norm": 2.578125,
"learning_rate": 1.737765614938828e-05,
"loss": 0.1491,
"step": 1730
},
{
"epoch": 0.27786649632705207,
"grad_norm": 2.046875,
"learning_rate": 1.7361558274307793e-05,
"loss": 0.1582,
"step": 1740
},
{
"epoch": 0.2794634302139891,
"grad_norm": 1.46875,
"learning_rate": 1.73454603992273e-05,
"loss": 0.1485,
"step": 1750
},
{
"epoch": 0.2810603641009262,
"grad_norm": 2.109375,
"learning_rate": 1.7329362524146813e-05,
"loss": 0.147,
"step": 1760
},
{
"epoch": 0.2826572979878633,
"grad_norm": 3.28125,
"learning_rate": 1.7313264649066325e-05,
"loss": 0.1601,
"step": 1770
},
{
"epoch": 0.28425423187480037,
"grad_norm": 1.640625,
"learning_rate": 1.7297166773985837e-05,
"loss": 0.1896,
"step": 1780
},
{
"epoch": 0.28585116576173747,
"grad_norm": 1.203125,
"learning_rate": 1.7281068898905345e-05,
"loss": 0.1295,
"step": 1790
},
{
"epoch": 0.28744809964867457,
"grad_norm": 1.8203125,
"learning_rate": 1.7264971023824857e-05,
"loss": 0.1552,
"step": 1800
},
{
"epoch": 0.2890450335356116,
"grad_norm": 1.9140625,
"learning_rate": 1.724887314874437e-05,
"loss": 0.1165,
"step": 1810
},
{
"epoch": 0.2906419674225487,
"grad_norm": 2.53125,
"learning_rate": 1.7232775273663877e-05,
"loss": 0.1691,
"step": 1820
},
{
"epoch": 0.29223890130948577,
"grad_norm": 2.34375,
"learning_rate": 1.721667739858339e-05,
"loss": 0.1648,
"step": 1830
},
{
"epoch": 0.29383583519642287,
"grad_norm": 1.5546875,
"learning_rate": 1.7200579523502897e-05,
"loss": 0.1802,
"step": 1840
},
{
"epoch": 0.29543276908335997,
"grad_norm": 1.859375,
"learning_rate": 1.718448164842241e-05,
"loss": 0.134,
"step": 1850
},
{
"epoch": 0.297029702970297,
"grad_norm": 4.71875,
"learning_rate": 1.716838377334192e-05,
"loss": 0.1923,
"step": 1860
},
{
"epoch": 0.2986266368572341,
"grad_norm": 1.6484375,
"learning_rate": 1.7152285898261433e-05,
"loss": 0.1559,
"step": 1870
},
{
"epoch": 0.2989460236346215,
"eval_loss": 0.13906948268413544,
"eval_runtime": 24.877,
"eval_samples_per_second": 10.21,
"eval_steps_per_second": 10.21,
"step": 1872
},
{
"epoch": 0.30022357074417116,
"grad_norm": 0.8984375,
"learning_rate": 1.713618802318094e-05,
"loss": 0.1411,
"step": 1880
},
{
"epoch": 0.30182050463110827,
"grad_norm": 4.625,
"learning_rate": 1.7120090148100453e-05,
"loss": 0.1293,
"step": 1890
},
{
"epoch": 0.30341743851804537,
"grad_norm": 2.046875,
"learning_rate": 1.710399227301996e-05,
"loss": 0.132,
"step": 1900
},
{
"epoch": 0.3050143724049824,
"grad_norm": 5.09375,
"learning_rate": 1.7087894397939473e-05,
"loss": 0.2111,
"step": 1910
},
{
"epoch": 0.3066113062919195,
"grad_norm": 1.578125,
"learning_rate": 1.707179652285898e-05,
"loss": 0.1398,
"step": 1920
},
{
"epoch": 0.3082082401788566,
"grad_norm": 1.3125,
"learning_rate": 1.7055698647778493e-05,
"loss": 0.1533,
"step": 1930
},
{
"epoch": 0.30980517406579366,
"grad_norm": 1.5546875,
"learning_rate": 1.7039600772698005e-05,
"loss": 0.249,
"step": 1940
},
{
"epoch": 0.31140210795273077,
"grad_norm": 2.671875,
"learning_rate": 1.7023502897617517e-05,
"loss": 0.1657,
"step": 1950
},
{
"epoch": 0.3129990418396678,
"grad_norm": 2.421875,
"learning_rate": 1.7007405022537025e-05,
"loss": 0.1804,
"step": 1960
},
{
"epoch": 0.3145959757266049,
"grad_norm": 2.5,
"learning_rate": 1.6991307147456537e-05,
"loss": 0.1369,
"step": 1970
},
{
"epoch": 0.316192909613542,
"grad_norm": 3.34375,
"learning_rate": 1.697520927237605e-05,
"loss": 0.148,
"step": 1980
},
{
"epoch": 0.31778984350047906,
"grad_norm": 1.515625,
"learning_rate": 1.6959111397295557e-05,
"loss": 0.1566,
"step": 1990
},
{
"epoch": 0.31938677738741617,
"grad_norm": 2.4375,
"learning_rate": 1.694301352221507e-05,
"loss": 0.1336,
"step": 2000
},
{
"epoch": 0.32098371127435327,
"grad_norm": 2.53125,
"learning_rate": 1.6926915647134578e-05,
"loss": 0.1196,
"step": 2010
},
{
"epoch": 0.3225806451612903,
"grad_norm": 2.5625,
"learning_rate": 1.6910817772054093e-05,
"loss": 0.13,
"step": 2020
},
{
"epoch": 0.3241775790482274,
"grad_norm": 1.765625,
"learning_rate": 1.68947198969736e-05,
"loss": 0.1239,
"step": 2030
},
{
"epoch": 0.32577451293516446,
"grad_norm": 1.09375,
"learning_rate": 1.6878622021893113e-05,
"loss": 0.1133,
"step": 2040
},
{
"epoch": 0.32737144682210156,
"grad_norm": 3.015625,
"learning_rate": 1.686252414681262e-05,
"loss": 0.1246,
"step": 2050
},
{
"epoch": 0.32896838070903867,
"grad_norm": 2.75,
"learning_rate": 1.6846426271732133e-05,
"loss": 0.1276,
"step": 2060
},
{
"epoch": 0.3305653145959757,
"grad_norm": 1.3671875,
"learning_rate": 1.683032839665164e-05,
"loss": 0.1565,
"step": 2070
},
{
"epoch": 0.3321622484829128,
"grad_norm": 1.1328125,
"learning_rate": 1.6814230521571153e-05,
"loss": 0.143,
"step": 2080
},
{
"epoch": 0.3321622484829128,
"eval_loss": 0.13785186409950256,
"eval_runtime": 25.0062,
"eval_samples_per_second": 10.157,
"eval_steps_per_second": 10.157,
"step": 2080
},
{
"epoch": 0.33375918236984986,
"grad_norm": 2.03125,
"learning_rate": 1.6798132646490665e-05,
"loss": 0.1548,
"step": 2090
},
{
"epoch": 0.33535611625678696,
"grad_norm": 1.125,
"learning_rate": 1.6782034771410177e-05,
"loss": 0.1193,
"step": 2100
},
{
"epoch": 0.33695305014372406,
"grad_norm": 2.75,
"learning_rate": 1.6765936896329685e-05,
"loss": 0.179,
"step": 2110
},
{
"epoch": 0.3385499840306611,
"grad_norm": 2.875,
"learning_rate": 1.6749839021249197e-05,
"loss": 0.1515,
"step": 2120
},
{
"epoch": 0.3401469179175982,
"grad_norm": 1.4453125,
"learning_rate": 1.6733741146168706e-05,
"loss": 0.1918,
"step": 2130
},
{
"epoch": 0.3417438518045353,
"grad_norm": 1.5234375,
"learning_rate": 1.6717643271088217e-05,
"loss": 0.1387,
"step": 2140
},
{
"epoch": 0.34334078569147236,
"grad_norm": 1.53125,
"learning_rate": 1.670154539600773e-05,
"loss": 0.1727,
"step": 2150
},
{
"epoch": 0.34493771957840946,
"grad_norm": 2.46875,
"learning_rate": 1.6685447520927238e-05,
"loss": 0.1655,
"step": 2160
},
{
"epoch": 0.3465346534653465,
"grad_norm": 2.234375,
"learning_rate": 1.666934964584675e-05,
"loss": 0.1519,
"step": 2170
},
{
"epoch": 0.3481315873522836,
"grad_norm": 2.453125,
"learning_rate": 1.665325177076626e-05,
"loss": 0.1636,
"step": 2180
},
{
"epoch": 0.3497285212392207,
"grad_norm": 1.9609375,
"learning_rate": 1.6637153895685773e-05,
"loss": 0.1773,
"step": 2190
},
{
"epoch": 0.35132545512615776,
"grad_norm": 2.765625,
"learning_rate": 1.662105602060528e-05,
"loss": 0.124,
"step": 2200
},
{
"epoch": 0.35292238901309486,
"grad_norm": 2.0625,
"learning_rate": 1.6604958145524793e-05,
"loss": 0.153,
"step": 2210
},
{
"epoch": 0.35451932290003196,
"grad_norm": 2.03125,
"learning_rate": 1.65888602704443e-05,
"loss": 0.126,
"step": 2220
},
{
"epoch": 0.356116256786969,
"grad_norm": 3.03125,
"learning_rate": 1.6572762395363813e-05,
"loss": 0.1571,
"step": 2230
},
{
"epoch": 0.3577131906739061,
"grad_norm": 2.515625,
"learning_rate": 1.6556664520283322e-05,
"loss": 0.1614,
"step": 2240
},
{
"epoch": 0.35931012456084316,
"grad_norm": 2.578125,
"learning_rate": 1.6540566645202834e-05,
"loss": 0.1292,
"step": 2250
},
{
"epoch": 0.36090705844778026,
"grad_norm": 1.6171875,
"learning_rate": 1.6524468770122345e-05,
"loss": 0.1246,
"step": 2260
},
{
"epoch": 0.36250399233471736,
"grad_norm": 2.46875,
"learning_rate": 1.6508370895041857e-05,
"loss": 0.165,
"step": 2270
},
{
"epoch": 0.3641009262216544,
"grad_norm": 2.0625,
"learning_rate": 1.6492273019961366e-05,
"loss": 0.1455,
"step": 2280
},
{
"epoch": 0.3653784733312041,
"eval_loss": 0.14108920097351074,
"eval_runtime": 25.0713,
"eval_samples_per_second": 10.131,
"eval_steps_per_second": 10.131,
"step": 2288
},
{
"epoch": 0.3656978601085915,
"grad_norm": 2.296875,
"learning_rate": 1.6476175144880877e-05,
"loss": 0.1509,
"step": 2290
},
{
"epoch": 0.3672947939955286,
"grad_norm": 1.53125,
"learning_rate": 1.6460077269800386e-05,
"loss": 0.1715,
"step": 2300
},
{
"epoch": 0.36889172788246566,
"grad_norm": 2.078125,
"learning_rate": 1.6443979394719898e-05,
"loss": 0.1118,
"step": 2310
},
{
"epoch": 0.37048866176940276,
"grad_norm": 1.234375,
"learning_rate": 1.642788151963941e-05,
"loss": 0.1263,
"step": 2320
},
{
"epoch": 0.3720855956563398,
"grad_norm": 1.6171875,
"learning_rate": 1.6411783644558918e-05,
"loss": 0.1441,
"step": 2330
},
{
"epoch": 0.3736825295432769,
"grad_norm": 2.46875,
"learning_rate": 1.639568576947843e-05,
"loss": 0.2059,
"step": 2340
},
{
"epoch": 0.375279463430214,
"grad_norm": 1.7890625,
"learning_rate": 1.637958789439794e-05,
"loss": 0.1239,
"step": 2350
},
{
"epoch": 0.37687639731715106,
"grad_norm": 2.484375,
"learning_rate": 1.6363490019317453e-05,
"loss": 0.1804,
"step": 2360
},
{
"epoch": 0.37847333120408816,
"grad_norm": 2.015625,
"learning_rate": 1.634739214423696e-05,
"loss": 0.1626,
"step": 2370
},
{
"epoch": 0.3800702650910252,
"grad_norm": 2.390625,
"learning_rate": 1.6331294269156473e-05,
"loss": 0.1737,
"step": 2380
},
{
"epoch": 0.3816671989779623,
"grad_norm": 1.9375,
"learning_rate": 1.6315196394075982e-05,
"loss": 0.1315,
"step": 2390
},
{
"epoch": 0.3832641328648994,
"grad_norm": 0.9609375,
"learning_rate": 1.6299098518995494e-05,
"loss": 0.1401,
"step": 2400
},
{
"epoch": 0.38486106675183646,
"grad_norm": 1.6015625,
"learning_rate": 1.6283000643915002e-05,
"loss": 0.1151,
"step": 2410
},
{
"epoch": 0.38645800063877356,
"grad_norm": 2.359375,
"learning_rate": 1.6266902768834517e-05,
"loss": 0.1481,
"step": 2420
},
{
"epoch": 0.38805493452571066,
"grad_norm": 2.375,
"learning_rate": 1.6250804893754026e-05,
"loss": 0.1366,
"step": 2430
},
{
"epoch": 0.3896518684126477,
"grad_norm": 3.515625,
"learning_rate": 1.6234707018673537e-05,
"loss": 0.1557,
"step": 2440
},
{
"epoch": 0.3912488022995848,
"grad_norm": 1.9921875,
"learning_rate": 1.6218609143593046e-05,
"loss": 0.144,
"step": 2450
},
{
"epoch": 0.39284573618652185,
"grad_norm": 2.515625,
"learning_rate": 1.6202511268512558e-05,
"loss": 0.143,
"step": 2460
},
{
"epoch": 0.39444267007345896,
"grad_norm": 1.8125,
"learning_rate": 1.618641339343207e-05,
"loss": 0.1555,
"step": 2470
},
{
"epoch": 0.39603960396039606,
"grad_norm": 1.609375,
"learning_rate": 1.6170315518351578e-05,
"loss": 0.1234,
"step": 2480
},
{
"epoch": 0.3976365378473331,
"grad_norm": 2.265625,
"learning_rate": 1.615421764327109e-05,
"loss": 0.1162,
"step": 2490
},
{
"epoch": 0.3985946981794954,
"eval_loss": 0.1328212171792984,
"eval_runtime": 25.1323,
"eval_samples_per_second": 10.107,
"eval_steps_per_second": 10.107,
"step": 2496
},
{
"epoch": 0.3992334717342702,
"grad_norm": 1.4765625,
"learning_rate": 1.61381197681906e-05,
"loss": 0.154,
"step": 2500
},
{
"epoch": 0.4008304056212073,
"grad_norm": 2.21875,
"learning_rate": 1.612202189311011e-05,
"loss": 0.1388,
"step": 2510
},
{
"epoch": 0.40242733950814435,
"grad_norm": 1.46875,
"learning_rate": 1.610592401802962e-05,
"loss": 0.1303,
"step": 2520
},
{
"epoch": 0.40402427339508146,
"grad_norm": 2.046875,
"learning_rate": 1.6089826142949133e-05,
"loss": 0.1318,
"step": 2530
},
{
"epoch": 0.4056212072820185,
"grad_norm": 2.40625,
"learning_rate": 1.6073728267868642e-05,
"loss": 0.1207,
"step": 2540
},
{
"epoch": 0.4072181411689556,
"grad_norm": 1.8125,
"learning_rate": 1.6057630392788153e-05,
"loss": 0.1542,
"step": 2550
},
{
"epoch": 0.4088150750558927,
"grad_norm": 2.59375,
"learning_rate": 1.6041532517707662e-05,
"loss": 0.156,
"step": 2560
},
{
"epoch": 0.41041200894282975,
"grad_norm": 1.453125,
"learning_rate": 1.6025434642627177e-05,
"loss": 0.1515,
"step": 2570
},
{
"epoch": 0.41200894282976686,
"grad_norm": 1.65625,
"learning_rate": 1.6009336767546685e-05,
"loss": 0.0988,
"step": 2580
},
{
"epoch": 0.4136058767167039,
"grad_norm": 2.40625,
"learning_rate": 1.5993238892466197e-05,
"loss": 0.1374,
"step": 2590
},
{
"epoch": 0.415202810603641,
"grad_norm": 1.34375,
"learning_rate": 1.5977141017385706e-05,
"loss": 0.1441,
"step": 2600
},
{
"epoch": 0.4167997444905781,
"grad_norm": 2.15625,
"learning_rate": 1.5961043142305217e-05,
"loss": 0.1585,
"step": 2610
},
{
"epoch": 0.41839667837751515,
"grad_norm": 2.984375,
"learning_rate": 1.5944945267224726e-05,
"loss": 0.1806,
"step": 2620
},
{
"epoch": 0.41999361226445225,
"grad_norm": 1.140625,
"learning_rate": 1.5928847392144238e-05,
"loss": 0.1272,
"step": 2630
},
{
"epoch": 0.42159054615138936,
"grad_norm": 1.359375,
"learning_rate": 1.591274951706375e-05,
"loss": 0.148,
"step": 2640
},
{
"epoch": 0.4231874800383264,
"grad_norm": 2.125,
"learning_rate": 1.589665164198326e-05,
"loss": 0.1252,
"step": 2650
},
{
"epoch": 0.4247844139252635,
"grad_norm": 2.03125,
"learning_rate": 1.588055376690277e-05,
"loss": 0.1311,
"step": 2660
},
{
"epoch": 0.42638134781220055,
"grad_norm": 1.625,
"learning_rate": 1.586445589182228e-05,
"loss": 0.1186,
"step": 2670
},
{
"epoch": 0.42797828169913765,
"grad_norm": 2.8125,
"learning_rate": 1.5848358016741793e-05,
"loss": 0.1481,
"step": 2680
},
{
"epoch": 0.42957521558607475,
"grad_norm": 2.125,
"learning_rate": 1.58322601416613e-05,
"loss": 0.1402,
"step": 2690
},
{
"epoch": 0.4311721494730118,
"grad_norm": 2.671875,
"learning_rate": 1.5816162266580813e-05,
"loss": 0.1451,
"step": 2700
},
{
"epoch": 0.4318109230277867,
"eval_loss": 0.1328616440296173,
"eval_runtime": 25.2649,
"eval_samples_per_second": 10.053,
"eval_steps_per_second": 10.053,
"step": 2704
},
{
"epoch": 0.4327690833599489,
"grad_norm": 2.578125,
"learning_rate": 1.5800064391500322e-05,
"loss": 0.1409,
"step": 2710
},
{
"epoch": 0.434366017246886,
"grad_norm": 3.28125,
"learning_rate": 1.5783966516419834e-05,
"loss": 0.1774,
"step": 2720
},
{
"epoch": 0.43596295113382305,
"grad_norm": 1.625,
"learning_rate": 1.5767868641339345e-05,
"loss": 0.1868,
"step": 2730
},
{
"epoch": 0.43755988502076015,
"grad_norm": 1.6328125,
"learning_rate": 1.5751770766258857e-05,
"loss": 0.152,
"step": 2740
},
{
"epoch": 0.4391568189076972,
"grad_norm": 1.59375,
"learning_rate": 1.5735672891178366e-05,
"loss": 0.1483,
"step": 2750
},
{
"epoch": 0.4407537527946343,
"grad_norm": 1.796875,
"learning_rate": 1.5719575016097877e-05,
"loss": 0.1548,
"step": 2760
},
{
"epoch": 0.4423506866815714,
"grad_norm": 2.703125,
"learning_rate": 1.5703477141017386e-05,
"loss": 0.1445,
"step": 2770
},
{
"epoch": 0.44394762056850845,
"grad_norm": 1.6640625,
"learning_rate": 1.5687379265936898e-05,
"loss": 0.2079,
"step": 2780
},
{
"epoch": 0.44554455445544555,
"grad_norm": 1.5,
"learning_rate": 1.5671281390856406e-05,
"loss": 0.139,
"step": 2790
},
{
"epoch": 0.4471414883423826,
"grad_norm": 2.125,
"learning_rate": 1.5655183515775918e-05,
"loss": 0.1595,
"step": 2800
},
{
"epoch": 0.4487384222293197,
"grad_norm": 1.328125,
"learning_rate": 1.563908564069543e-05,
"loss": 0.158,
"step": 2810
},
{
"epoch": 0.4503353561162568,
"grad_norm": 2.328125,
"learning_rate": 1.562298776561494e-05,
"loss": 0.1602,
"step": 2820
},
{
"epoch": 0.45193229000319385,
"grad_norm": 2.09375,
"learning_rate": 1.560688989053445e-05,
"loss": 0.145,
"step": 2830
},
{
"epoch": 0.45352922389013095,
"grad_norm": 2.34375,
"learning_rate": 1.559079201545396e-05,
"loss": 0.1366,
"step": 2840
},
{
"epoch": 0.45512615777706805,
"grad_norm": 3.09375,
"learning_rate": 1.5574694140373473e-05,
"loss": 0.1524,
"step": 2850
},
{
"epoch": 0.4567230916640051,
"grad_norm": 1.484375,
"learning_rate": 1.5558596265292982e-05,
"loss": 0.1155,
"step": 2860
},
{
"epoch": 0.4583200255509422,
"grad_norm": 1.3359375,
"learning_rate": 1.5542498390212494e-05,
"loss": 0.1611,
"step": 2870
},
{
"epoch": 0.45991695943787925,
"grad_norm": 1.7421875,
"learning_rate": 1.5526400515132002e-05,
"loss": 0.1441,
"step": 2880
},
{
"epoch": 0.46151389332481635,
"grad_norm": 1.078125,
"learning_rate": 1.5510302640051517e-05,
"loss": 0.1247,
"step": 2890
},
{
"epoch": 0.46311082721175345,
"grad_norm": 2.96875,
"learning_rate": 1.5494204764971026e-05,
"loss": 0.2078,
"step": 2900
},
{
"epoch": 0.4647077610986905,
"grad_norm": 1.234375,
"learning_rate": 1.5478106889890537e-05,
"loss": 0.1292,
"step": 2910
},
{
"epoch": 0.4650271478760779,
"eval_loss": 0.13005879521369934,
"eval_runtime": 25.4154,
"eval_samples_per_second": 9.994,
"eval_steps_per_second": 9.994,
"step": 2912
},
{
"epoch": 0.4663046949856276,
"grad_norm": 1.5078125,
"learning_rate": 1.5462009014810046e-05,
"loss": 0.1592,
"step": 2920
},
{
"epoch": 0.4679016288725647,
"grad_norm": 1.0,
"learning_rate": 1.5445911139729558e-05,
"loss": 0.1287,
"step": 2930
},
{
"epoch": 0.46949856275950175,
"grad_norm": 1.3828125,
"learning_rate": 1.5429813264649066e-05,
"loss": 0.1398,
"step": 2940
},
{
"epoch": 0.47109549664643885,
"grad_norm": 2.15625,
"learning_rate": 1.5413715389568578e-05,
"loss": 0.1419,
"step": 2950
},
{
"epoch": 0.4726924305333759,
"grad_norm": 1.3515625,
"learning_rate": 1.539761751448809e-05,
"loss": 0.1385,
"step": 2960
},
{
"epoch": 0.474289364420313,
"grad_norm": 2.375,
"learning_rate": 1.53815196394076e-05,
"loss": 0.1451,
"step": 2970
},
{
"epoch": 0.4758862983072501,
"grad_norm": 2.53125,
"learning_rate": 1.536542176432711e-05,
"loss": 0.1369,
"step": 2980
},
{
"epoch": 0.47748323219418715,
"grad_norm": 2.859375,
"learning_rate": 1.534932388924662e-05,
"loss": 0.1297,
"step": 2990
},
{
"epoch": 0.47908016608112425,
"grad_norm": 1.5234375,
"learning_rate": 1.533322601416613e-05,
"loss": 0.1112,
"step": 3000
},
{
"epoch": 0.48067709996806135,
"grad_norm": 4.65625,
"learning_rate": 1.5317128139085642e-05,
"loss": 0.1922,
"step": 3010
},
{
"epoch": 0.4822740338549984,
"grad_norm": 1.8203125,
"learning_rate": 1.5301030264005154e-05,
"loss": 0.1742,
"step": 3020
},
{
"epoch": 0.4838709677419355,
"grad_norm": 1.4609375,
"learning_rate": 1.5284932388924662e-05,
"loss": 0.1401,
"step": 3030
},
{
"epoch": 0.48546790162887254,
"grad_norm": 3.40625,
"learning_rate": 1.5268834513844174e-05,
"loss": 0.1571,
"step": 3040
},
{
"epoch": 0.48706483551580965,
"grad_norm": 2.359375,
"learning_rate": 1.5252736638763686e-05,
"loss": 0.1487,
"step": 3050
},
{
"epoch": 0.48866176940274675,
"grad_norm": 2.609375,
"learning_rate": 1.5236638763683196e-05,
"loss": 0.1471,
"step": 3060
},
{
"epoch": 0.4902587032896838,
"grad_norm": 1.9140625,
"learning_rate": 1.5220540888602706e-05,
"loss": 0.1618,
"step": 3070
},
{
"epoch": 0.4918556371766209,
"grad_norm": 3.203125,
"learning_rate": 1.5204443013522216e-05,
"loss": 0.1377,
"step": 3080
},
{
"epoch": 0.49345257106355794,
"grad_norm": 1.3125,
"learning_rate": 1.5188345138441728e-05,
"loss": 0.1422,
"step": 3090
},
{
"epoch": 0.49504950495049505,
"grad_norm": 2.15625,
"learning_rate": 1.5172247263361238e-05,
"loss": 0.1698,
"step": 3100
},
{
"epoch": 0.49664643883743215,
"grad_norm": 1.6171875,
"learning_rate": 1.5156149388280748e-05,
"loss": 0.1568,
"step": 3110
},
{
"epoch": 0.4982433727243692,
"grad_norm": 1.75,
"learning_rate": 1.5140051513200258e-05,
"loss": 0.1324,
"step": 3120
},
{
"epoch": 0.4982433727243692,
"eval_loss": 0.13277089595794678,
"eval_runtime": 25.7821,
"eval_samples_per_second": 9.852,
"eval_steps_per_second": 9.852,
"step": 3120
},
{
"epoch": 0.4998403066113063,
"grad_norm": 2.015625,
"learning_rate": 1.512395363811977e-05,
"loss": 0.1619,
"step": 3130
},
{
"epoch": 0.5014372404982433,
"grad_norm": 2.125,
"learning_rate": 1.5107855763039282e-05,
"loss": 0.1421,
"step": 3140
},
{
"epoch": 0.5030341743851805,
"grad_norm": 1.609375,
"learning_rate": 1.5091757887958792e-05,
"loss": 0.1275,
"step": 3150
},
{
"epoch": 0.5046311082721175,
"grad_norm": 2.5,
"learning_rate": 1.5075660012878302e-05,
"loss": 0.2466,
"step": 3160
},
{
"epoch": 0.5062280421590546,
"grad_norm": 3.875,
"learning_rate": 1.5059562137797812e-05,
"loss": 0.1598,
"step": 3170
},
{
"epoch": 0.5078249760459917,
"grad_norm": 2.0,
"learning_rate": 1.5043464262717322e-05,
"loss": 0.134,
"step": 3180
},
{
"epoch": 0.5094219099329288,
"grad_norm": 1.7421875,
"learning_rate": 1.5027366387636832e-05,
"loss": 0.1328,
"step": 3190
},
{
"epoch": 0.5110188438198658,
"grad_norm": 2.171875,
"learning_rate": 1.5011268512556342e-05,
"loss": 0.1298,
"step": 3200
},
{
"epoch": 0.5126157777068029,
"grad_norm": 1.328125,
"learning_rate": 1.4995170637475856e-05,
"loss": 0.1553,
"step": 3210
},
{
"epoch": 0.51421271159374,
"grad_norm": 1.2890625,
"learning_rate": 1.4979072762395366e-05,
"loss": 0.1397,
"step": 3220
},
{
"epoch": 0.5158096454806771,
"grad_norm": 1.1953125,
"learning_rate": 1.4962974887314876e-05,
"loss": 0.1658,
"step": 3230
},
{
"epoch": 0.5174065793676141,
"grad_norm": 2.28125,
"learning_rate": 1.4946877012234386e-05,
"loss": 0.1195,
"step": 3240
},
{
"epoch": 0.5190035132545513,
"grad_norm": 1.90625,
"learning_rate": 1.4930779137153896e-05,
"loss": 0.1354,
"step": 3250
},
{
"epoch": 0.5206004471414883,
"grad_norm": 3.015625,
"learning_rate": 1.4914681262073408e-05,
"loss": 0.1477,
"step": 3260
},
{
"epoch": 0.5221973810284254,
"grad_norm": 2.03125,
"learning_rate": 1.4898583386992918e-05,
"loss": 0.127,
"step": 3270
},
{
"epoch": 0.5237943149153625,
"grad_norm": 5.25,
"learning_rate": 1.4882485511912428e-05,
"loss": 0.1244,
"step": 3280
},
{
"epoch": 0.5253912488022996,
"grad_norm": 2.015625,
"learning_rate": 1.486638763683194e-05,
"loss": 0.132,
"step": 3290
},
{
"epoch": 0.5269881826892366,
"grad_norm": 1.328125,
"learning_rate": 1.4850289761751452e-05,
"loss": 0.1319,
"step": 3300
},
{
"epoch": 0.5285851165761738,
"grad_norm": 2.28125,
"learning_rate": 1.4834191886670962e-05,
"loss": 0.1394,
"step": 3310
},
{
"epoch": 0.5301820504631108,
"grad_norm": 3.25,
"learning_rate": 1.4818094011590472e-05,
"loss": 0.1454,
"step": 3320
},
{
"epoch": 0.5314595975726605,
"eval_loss": 0.1289980709552765,
"eval_runtime": 25.2531,
"eval_samples_per_second": 10.058,
"eval_steps_per_second": 10.058,
"step": 3328
},
{
"epoch": 0.5317789843500479,
"grad_norm": 2.546875,
"learning_rate": 1.4801996136509982e-05,
"loss": 0.1556,
"step": 3330
},
{
"epoch": 0.5333759182369849,
"grad_norm": 4.84375,
"learning_rate": 1.4785898261429492e-05,
"loss": 0.1866,
"step": 3340
},
{
"epoch": 0.5349728521239221,
"grad_norm": 2.5,
"learning_rate": 1.4769800386349002e-05,
"loss": 0.1561,
"step": 3350
},
{
"epoch": 0.5365697860108591,
"grad_norm": 1.96875,
"learning_rate": 1.4753702511268512e-05,
"loss": 0.1372,
"step": 3360
},
{
"epoch": 0.5381667198977962,
"grad_norm": 1.2265625,
"learning_rate": 1.4737604636188026e-05,
"loss": 0.1428,
"step": 3370
},
{
"epoch": 0.5397636537847333,
"grad_norm": 1.5,
"learning_rate": 1.4721506761107536e-05,
"loss": 0.1355,
"step": 3380
},
{
"epoch": 0.5413605876716704,
"grad_norm": 2.328125,
"learning_rate": 1.4705408886027046e-05,
"loss": 0.158,
"step": 3390
},
{
"epoch": 0.5429575215586074,
"grad_norm": 5.25,
"learning_rate": 1.4689311010946556e-05,
"loss": 0.1121,
"step": 3400
},
{
"epoch": 0.5445544554455446,
"grad_norm": 1.4765625,
"learning_rate": 1.4673213135866066e-05,
"loss": 0.1573,
"step": 3410
},
{
"epoch": 0.5461513893324816,
"grad_norm": 1.1875,
"learning_rate": 1.4657115260785578e-05,
"loss": 0.1385,
"step": 3420
},
{
"epoch": 0.5477483232194187,
"grad_norm": 2.109375,
"learning_rate": 1.4641017385705088e-05,
"loss": 0.1428,
"step": 3430
},
{
"epoch": 0.5493452571063558,
"grad_norm": 1.9375,
"learning_rate": 1.4624919510624598e-05,
"loss": 0.1714,
"step": 3440
},
{
"epoch": 0.5509421909932929,
"grad_norm": 2.953125,
"learning_rate": 1.460882163554411e-05,
"loss": 0.1099,
"step": 3450
},
{
"epoch": 0.5525391248802299,
"grad_norm": 1.5234375,
"learning_rate": 1.459272376046362e-05,
"loss": 0.1648,
"step": 3460
},
{
"epoch": 0.5541360587671671,
"grad_norm": 2.328125,
"learning_rate": 1.4576625885383132e-05,
"loss": 0.1139,
"step": 3470
},
{
"epoch": 0.5557329926541041,
"grad_norm": 2.359375,
"learning_rate": 1.4560528010302642e-05,
"loss": 0.1266,
"step": 3480
},
{
"epoch": 0.5573299265410412,
"grad_norm": 2.0625,
"learning_rate": 1.4544430135222152e-05,
"loss": 0.1183,
"step": 3490
},
{
"epoch": 0.5589268604279782,
"grad_norm": 2.703125,
"learning_rate": 1.4528332260141662e-05,
"loss": 0.127,
"step": 3500
},
{
"epoch": 0.5605237943149154,
"grad_norm": 3.65625,
"learning_rate": 1.4512234385061172e-05,
"loss": 0.1141,
"step": 3510
},
{
"epoch": 0.5621207282018524,
"grad_norm": 2.046875,
"learning_rate": 1.4496136509980682e-05,
"loss": 0.1337,
"step": 3520
},
{
"epoch": 0.5637176620887895,
"grad_norm": 1.7265625,
"learning_rate": 1.4480038634900196e-05,
"loss": 0.1548,
"step": 3530
},
{
"epoch": 0.5646758224209518,
"eval_loss": 0.12811633944511414,
"eval_runtime": 27.5834,
"eval_samples_per_second": 9.208,
"eval_steps_per_second": 9.208,
"step": 3536
},
{
"epoch": 0.5653145959757266,
"grad_norm": 2.4375,
"learning_rate": 1.4463940759819706e-05,
"loss": 0.1533,
"step": 3540
},
{
"epoch": 0.5669115298626637,
"grad_norm": 1.84375,
"learning_rate": 1.4447842884739216e-05,
"loss": 0.1228,
"step": 3550
},
{
"epoch": 0.5685084637496007,
"grad_norm": 2.15625,
"learning_rate": 1.4431745009658726e-05,
"loss": 0.1468,
"step": 3560
},
{
"epoch": 0.5701053976365379,
"grad_norm": 1.609375,
"learning_rate": 1.4415647134578236e-05,
"loss": 0.1475,
"step": 3570
},
{
"epoch": 0.5717023315234749,
"grad_norm": 1.65625,
"learning_rate": 1.4399549259497748e-05,
"loss": 0.1353,
"step": 3580
},
{
"epoch": 0.573299265410412,
"grad_norm": 3.0625,
"learning_rate": 1.4383451384417258e-05,
"loss": 0.1528,
"step": 3590
},
{
"epoch": 0.5748961992973491,
"grad_norm": 3.484375,
"learning_rate": 1.4367353509336768e-05,
"loss": 0.1383,
"step": 3600
},
{
"epoch": 0.5764931331842862,
"grad_norm": 1.3984375,
"learning_rate": 1.435125563425628e-05,
"loss": 0.1507,
"step": 3610
},
{
"epoch": 0.5780900670712232,
"grad_norm": 3.375,
"learning_rate": 1.433515775917579e-05,
"loss": 0.1202,
"step": 3620
},
{
"epoch": 0.5796870009581603,
"grad_norm": 1.5859375,
"learning_rate": 1.4319059884095302e-05,
"loss": 0.105,
"step": 3630
},
{
"epoch": 0.5812839348450974,
"grad_norm": 1.71875,
"learning_rate": 1.4302962009014812e-05,
"loss": 0.1253,
"step": 3640
},
{
"epoch": 0.5828808687320345,
"grad_norm": 1.3828125,
"learning_rate": 1.4286864133934322e-05,
"loss": 0.131,
"step": 3650
},
{
"epoch": 0.5844778026189715,
"grad_norm": 4.84375,
"learning_rate": 1.4270766258853832e-05,
"loss": 0.1822,
"step": 3660
},
{
"epoch": 0.5860747365059087,
"grad_norm": 1.453125,
"learning_rate": 1.4254668383773342e-05,
"loss": 0.1355,
"step": 3670
},
{
"epoch": 0.5876716703928457,
"grad_norm": 1.8828125,
"learning_rate": 1.4238570508692852e-05,
"loss": 0.1315,
"step": 3680
},
{
"epoch": 0.5892686042797828,
"grad_norm": 1.421875,
"learning_rate": 1.4222472633612366e-05,
"loss": 0.1427,
"step": 3690
},
{
"epoch": 0.5908655381667199,
"grad_norm": 1.3359375,
"learning_rate": 1.4206374758531876e-05,
"loss": 0.126,
"step": 3700
},
{
"epoch": 0.592462472053657,
"grad_norm": 2.015625,
"learning_rate": 1.4190276883451386e-05,
"loss": 0.1193,
"step": 3710
},
{
"epoch": 0.594059405940594,
"grad_norm": 2.28125,
"learning_rate": 1.4174179008370896e-05,
"loss": 0.1738,
"step": 3720
},
{
"epoch": 0.5956563398275312,
"grad_norm": 2.234375,
"learning_rate": 1.4158081133290406e-05,
"loss": 0.1103,
"step": 3730
},
{
"epoch": 0.5972532737144682,
"grad_norm": 2.859375,
"learning_rate": 1.4141983258209916e-05,
"loss": 0.1268,
"step": 3740
},
{
"epoch": 0.597892047269243,
"eval_loss": 0.1265590339899063,
"eval_runtime": 25.3406,
"eval_samples_per_second": 10.023,
"eval_steps_per_second": 10.023,
"step": 3744
},
{
"epoch": 0.5988502076014053,
"grad_norm": 1.25,
"learning_rate": 1.4125885383129428e-05,
"loss": 0.1321,
"step": 3750
},
{
"epoch": 0.6004471414883423,
"grad_norm": 1.578125,
"learning_rate": 1.4109787508048938e-05,
"loss": 0.1295,
"step": 3760
},
{
"epoch": 0.6020440753752795,
"grad_norm": 1.65625,
"learning_rate": 1.409368963296845e-05,
"loss": 0.1457,
"step": 3770
},
{
"epoch": 0.6036410092622165,
"grad_norm": 1.65625,
"learning_rate": 1.407759175788796e-05,
"loss": 0.1066,
"step": 3780
},
{
"epoch": 0.6052379431491536,
"grad_norm": 2.015625,
"learning_rate": 1.406149388280747e-05,
"loss": 0.1567,
"step": 3790
},
{
"epoch": 0.6068348770360907,
"grad_norm": 2.84375,
"learning_rate": 1.4045396007726982e-05,
"loss": 0.145,
"step": 3800
},
{
"epoch": 0.6084318109230278,
"grad_norm": 2.09375,
"learning_rate": 1.4029298132646492e-05,
"loss": 0.105,
"step": 3810
},
{
"epoch": 0.6100287448099648,
"grad_norm": 2.078125,
"learning_rate": 1.4013200257566002e-05,
"loss": 0.1298,
"step": 3820
},
{
"epoch": 0.611625678696902,
"grad_norm": 2.984375,
"learning_rate": 1.3997102382485512e-05,
"loss": 0.1185,
"step": 3830
},
{
"epoch": 0.613222612583839,
"grad_norm": 1.921875,
"learning_rate": 1.3981004507405022e-05,
"loss": 0.1493,
"step": 3840
},
{
"epoch": 0.6148195464707761,
"grad_norm": 1.3515625,
"learning_rate": 1.3964906632324536e-05,
"loss": 0.1487,
"step": 3850
},
{
"epoch": 0.6164164803577132,
"grad_norm": 1.9140625,
"learning_rate": 1.3948808757244046e-05,
"loss": 0.1053,
"step": 3860
},
{
"epoch": 0.6180134142446503,
"grad_norm": 0.86328125,
"learning_rate": 1.3932710882163556e-05,
"loss": 0.1096,
"step": 3870
},
{
"epoch": 0.6196103481315873,
"grad_norm": 2.15625,
"learning_rate": 1.3916613007083066e-05,
"loss": 0.114,
"step": 3880
},
{
"epoch": 0.6212072820185245,
"grad_norm": 2.265625,
"learning_rate": 1.3900515132002576e-05,
"loss": 0.1182,
"step": 3890
},
{
"epoch": 0.6228042159054615,
"grad_norm": 2.65625,
"learning_rate": 1.3884417256922086e-05,
"loss": 0.1458,
"step": 3900
},
{
"epoch": 0.6244011497923986,
"grad_norm": 2.265625,
"learning_rate": 1.3868319381841598e-05,
"loss": 0.1225,
"step": 3910
},
{
"epoch": 0.6259980836793356,
"grad_norm": 1.84375,
"learning_rate": 1.3852221506761108e-05,
"loss": 0.1264,
"step": 3920
},
{
"epoch": 0.6275950175662728,
"grad_norm": 2.21875,
"learning_rate": 1.383612363168062e-05,
"loss": 0.1707,
"step": 3930
},
{
"epoch": 0.6291919514532098,
"grad_norm": 2.234375,
"learning_rate": 1.382002575660013e-05,
"loss": 0.1633,
"step": 3940
},
{
"epoch": 0.6307888853401469,
"grad_norm": 2.796875,
"learning_rate": 1.380392788151964e-05,
"loss": 0.1451,
"step": 3950
},
{
"epoch": 0.6311082721175343,
"eval_loss": 0.1257615089416504,
"eval_runtime": 25.3833,
"eval_samples_per_second": 10.007,
"eval_steps_per_second": 10.007,
"step": 3952
},
{
"epoch": 0.632385819227084,
"grad_norm": 2.359375,
"learning_rate": 1.3787830006439152e-05,
"loss": 0.1251,
"step": 3960
},
{
"epoch": 0.6339827531140211,
"grad_norm": 1.984375,
"learning_rate": 1.3771732131358662e-05,
"loss": 0.1389,
"step": 3970
},
{
"epoch": 0.6355796870009581,
"grad_norm": 1.9296875,
"learning_rate": 1.3755634256278172e-05,
"loss": 0.1485,
"step": 3980
},
{
"epoch": 0.6371766208878953,
"grad_norm": 3.34375,
"learning_rate": 1.3739536381197682e-05,
"loss": 0.1155,
"step": 3990
},
{
"epoch": 0.6387735547748323,
"grad_norm": 1.0703125,
"learning_rate": 1.3723438506117192e-05,
"loss": 0.1389,
"step": 4000
},
{
"epoch": 0.6403704886617694,
"grad_norm": 1.546875,
"learning_rate": 1.3707340631036706e-05,
"loss": 0.1305,
"step": 4010
},
{
"epoch": 0.6419674225487065,
"grad_norm": 2.390625,
"learning_rate": 1.3691242755956216e-05,
"loss": 0.1184,
"step": 4020
},
{
"epoch": 0.6435643564356436,
"grad_norm": 1.90625,
"learning_rate": 1.3675144880875726e-05,
"loss": 0.1333,
"step": 4030
},
{
"epoch": 0.6451612903225806,
"grad_norm": 1.7265625,
"learning_rate": 1.3659047005795236e-05,
"loss": 0.128,
"step": 4040
},
{
"epoch": 0.6467582242095177,
"grad_norm": 2.421875,
"learning_rate": 1.3642949130714746e-05,
"loss": 0.1281,
"step": 4050
},
{
"epoch": 0.6483551580964548,
"grad_norm": 4.28125,
"learning_rate": 1.3626851255634256e-05,
"loss": 0.1335,
"step": 4060
},
{
"epoch": 0.6499520919833919,
"grad_norm": 1.828125,
"learning_rate": 1.3610753380553766e-05,
"loss": 0.1343,
"step": 4070
},
{
"epoch": 0.6515490258703289,
"grad_norm": 2.703125,
"learning_rate": 1.3594655505473278e-05,
"loss": 0.1464,
"step": 4080
},
{
"epoch": 0.6531459597572661,
"grad_norm": 1.4296875,
"learning_rate": 1.357855763039279e-05,
"loss": 0.1378,
"step": 4090
},
{
"epoch": 0.6547428936442031,
"grad_norm": 1.8984375,
"learning_rate": 1.35624597553123e-05,
"loss": 0.1351,
"step": 4100
},
{
"epoch": 0.6563398275311402,
"grad_norm": 2.390625,
"learning_rate": 1.354636188023181e-05,
"loss": 0.1307,
"step": 4110
},
{
"epoch": 0.6579367614180773,
"grad_norm": 1.28125,
"learning_rate": 1.353026400515132e-05,
"loss": 0.1244,
"step": 4120
},
{
"epoch": 0.6595336953050144,
"grad_norm": 2.8125,
"learning_rate": 1.3514166130070832e-05,
"loss": 0.1622,
"step": 4130
},
{
"epoch": 0.6611306291919514,
"grad_norm": 1.359375,
"learning_rate": 1.3498068254990342e-05,
"loss": 0.1371,
"step": 4140
},
{
"epoch": 0.6627275630788886,
"grad_norm": 2.0,
"learning_rate": 1.3481970379909852e-05,
"loss": 0.1224,
"step": 4150
},
{
"epoch": 0.6643244969658256,
"grad_norm": 1.5078125,
"learning_rate": 1.3465872504829362e-05,
"loss": 0.125,
"step": 4160
},
{
"epoch": 0.6643244969658256,
"eval_loss": 0.1270388513803482,
"eval_runtime": 25.0112,
"eval_samples_per_second": 10.155,
"eval_steps_per_second": 10.155,
"step": 4160
},
{
"epoch": 0.6659214308527627,
"grad_norm": 3.734375,
"learning_rate": 1.3449774629748876e-05,
"loss": 0.1624,
"step": 4170
},
{
"epoch": 0.6675183647396997,
"grad_norm": 1.828125,
"learning_rate": 1.3433676754668386e-05,
"loss": 0.1357,
"step": 4180
},
{
"epoch": 0.6691152986266369,
"grad_norm": 1.8046875,
"learning_rate": 1.3417578879587896e-05,
"loss": 0.0979,
"step": 4190
},
{
"epoch": 0.6707122325135739,
"grad_norm": 2.34375,
"learning_rate": 1.3401481004507406e-05,
"loss": 0.1556,
"step": 4200
},
{
"epoch": 0.672309166400511,
"grad_norm": 1.546875,
"learning_rate": 1.3385383129426916e-05,
"loss": 0.1352,
"step": 4210
},
{
"epoch": 0.6739061002874481,
"grad_norm": 5.59375,
"learning_rate": 1.3369285254346426e-05,
"loss": 0.138,
"step": 4220
},
{
"epoch": 0.6755030341743852,
"grad_norm": 1.4609375,
"learning_rate": 1.3353187379265937e-05,
"loss": 0.1133,
"step": 4230
},
{
"epoch": 0.6770999680613222,
"grad_norm": 2.203125,
"learning_rate": 1.3337089504185448e-05,
"loss": 0.1212,
"step": 4240
},
{
"epoch": 0.6786969019482594,
"grad_norm": 3.28125,
"learning_rate": 1.332099162910496e-05,
"loss": 0.1561,
"step": 4250
},
{
"epoch": 0.6802938358351964,
"grad_norm": 1.6640625,
"learning_rate": 1.330489375402447e-05,
"loss": 0.184,
"step": 4260
},
{
"epoch": 0.6818907697221335,
"grad_norm": 1.6484375,
"learning_rate": 1.328879587894398e-05,
"loss": 0.1242,
"step": 4270
},
{
"epoch": 0.6834877036090706,
"grad_norm": 3.65625,
"learning_rate": 1.327269800386349e-05,
"loss": 0.1422,
"step": 4280
},
{
"epoch": 0.6850846374960077,
"grad_norm": 2.359375,
"learning_rate": 1.3256600128783002e-05,
"loss": 0.1209,
"step": 4290
},
{
"epoch": 0.6866815713829447,
"grad_norm": 0.67578125,
"learning_rate": 1.3240502253702512e-05,
"loss": 0.1407,
"step": 4300
},
{
"epoch": 0.6882785052698819,
"grad_norm": 1.4453125,
"learning_rate": 1.3224404378622022e-05,
"loss": 0.1062,
"step": 4310
},
{
"epoch": 0.6898754391568189,
"grad_norm": 3.421875,
"learning_rate": 1.3208306503541533e-05,
"loss": 0.126,
"step": 4320
},
{
"epoch": 0.691472373043756,
"grad_norm": 1.421875,
"learning_rate": 1.3192208628461044e-05,
"loss": 0.1319,
"step": 4330
},
{
"epoch": 0.693069306930693,
"grad_norm": 2.0625,
"learning_rate": 1.3176110753380556e-05,
"loss": 0.1666,
"step": 4340
},
{
"epoch": 0.6946662408176302,
"grad_norm": 1.5390625,
"learning_rate": 1.3160012878300066e-05,
"loss": 0.1079,
"step": 4350
},
{
"epoch": 0.6962631747045672,
"grad_norm": 1.53125,
"learning_rate": 1.3143915003219576e-05,
"loss": 0.1378,
"step": 4360
},
{
"epoch": 0.6975407218141169,
"eval_loss": 0.1270260214805603,
"eval_runtime": 25.1624,
"eval_samples_per_second": 10.094,
"eval_steps_per_second": 10.094,
"step": 4368
},
{
"epoch": 0.6978601085915043,
"grad_norm": 1.53125,
"learning_rate": 1.3127817128139086e-05,
"loss": 0.1365,
"step": 4370
},
{
"epoch": 0.6994570424784414,
"grad_norm": 2.484375,
"learning_rate": 1.3111719253058597e-05,
"loss": 0.0956,
"step": 4380
},
{
"epoch": 0.7010539763653785,
"grad_norm": 2.421875,
"learning_rate": 1.3095621377978107e-05,
"loss": 0.1255,
"step": 4390
},
{
"epoch": 0.7026509102523155,
"grad_norm": 2.734375,
"learning_rate": 1.3079523502897617e-05,
"loss": 0.197,
"step": 4400
},
{
"epoch": 0.7042478441392527,
"grad_norm": 2.140625,
"learning_rate": 1.306342562781713e-05,
"loss": 0.1328,
"step": 4410
},
{
"epoch": 0.7058447780261897,
"grad_norm": 2.84375,
"learning_rate": 1.304732775273664e-05,
"loss": 0.124,
"step": 4420
},
{
"epoch": 0.7074417119131268,
"grad_norm": 0.98828125,
"learning_rate": 1.303122987765615e-05,
"loss": 0.1296,
"step": 4430
},
{
"epoch": 0.7090386458000639,
"grad_norm": 1.9375,
"learning_rate": 1.301513200257566e-05,
"loss": 0.1263,
"step": 4440
},
{
"epoch": 0.710635579687001,
"grad_norm": 2.03125,
"learning_rate": 1.2999034127495172e-05,
"loss": 0.1873,
"step": 4450
},
{
"epoch": 0.712232513573938,
"grad_norm": 2.03125,
"learning_rate": 1.2982936252414682e-05,
"loss": 0.1535,
"step": 4460
},
{
"epoch": 0.7138294474608751,
"grad_norm": 1.4296875,
"learning_rate": 1.2966838377334192e-05,
"loss": 0.1236,
"step": 4470
},
{
"epoch": 0.7154263813478122,
"grad_norm": 2.4375,
"learning_rate": 1.2950740502253703e-05,
"loss": 0.1293,
"step": 4480
},
{
"epoch": 0.7170233152347493,
"grad_norm": 1.46875,
"learning_rate": 1.2934642627173214e-05,
"loss": 0.1763,
"step": 4490
},
{
"epoch": 0.7186202491216863,
"grad_norm": 1.921875,
"learning_rate": 1.2918544752092726e-05,
"loss": 0.1237,
"step": 4500
},
{
"epoch": 0.7202171830086235,
"grad_norm": 1.609375,
"learning_rate": 1.2902446877012236e-05,
"loss": 0.1537,
"step": 4510
},
{
"epoch": 0.7218141168955605,
"grad_norm": 0.81640625,
"learning_rate": 1.2886349001931746e-05,
"loss": 0.1226,
"step": 4520
},
{
"epoch": 0.7234110507824976,
"grad_norm": 2.265625,
"learning_rate": 1.2870251126851256e-05,
"loss": 0.1707,
"step": 4530
},
{
"epoch": 0.7250079846694347,
"grad_norm": 1.375,
"learning_rate": 1.2854153251770767e-05,
"loss": 0.164,
"step": 4540
},
{
"epoch": 0.7266049185563718,
"grad_norm": 8.9375,
"learning_rate": 1.2838055376690277e-05,
"loss": 0.1219,
"step": 4550
},
{
"epoch": 0.7282018524433088,
"grad_norm": 2.125,
"learning_rate": 1.2821957501609787e-05,
"loss": 0.1223,
"step": 4560
},
{
"epoch": 0.729798786330246,
"grad_norm": 2.890625,
"learning_rate": 1.28058596265293e-05,
"loss": 0.1766,
"step": 4570
},
{
"epoch": 0.7307569466624082,
"eval_loss": 0.12562747299671173,
"eval_runtime": 25.1975,
"eval_samples_per_second": 10.08,
"eval_steps_per_second": 10.08,
"step": 4576
},
{
"epoch": 0.731395720217183,
"grad_norm": 1.421875,
"learning_rate": 1.278976175144881e-05,
"loss": 0.1353,
"step": 4580
},
{
"epoch": 0.7329926541041201,
"grad_norm": 1.609375,
"learning_rate": 1.277366387636832e-05,
"loss": 0.1739,
"step": 4590
},
{
"epoch": 0.7345895879910572,
"grad_norm": 2.578125,
"learning_rate": 1.275756600128783e-05,
"loss": 0.1318,
"step": 4600
},
{
"epoch": 0.7361865218779943,
"grad_norm": 1.5390625,
"learning_rate": 1.274146812620734e-05,
"loss": 0.1075,
"step": 4610
},
{
"epoch": 0.7377834557649313,
"grad_norm": 1.5234375,
"learning_rate": 1.2725370251126852e-05,
"loss": 0.1584,
"step": 4620
},
{
"epoch": 0.7393803896518684,
"grad_norm": 1.3359375,
"learning_rate": 1.2709272376046363e-05,
"loss": 0.1175,
"step": 4630
},
{
"epoch": 0.7409773235388055,
"grad_norm": 2.125,
"learning_rate": 1.2693174500965873e-05,
"loss": 0.1571,
"step": 4640
},
{
"epoch": 0.7425742574257426,
"grad_norm": 1.40625,
"learning_rate": 1.2677076625885384e-05,
"loss": 0.1187,
"step": 4650
},
{
"epoch": 0.7441711913126796,
"grad_norm": 2.71875,
"learning_rate": 1.2660978750804895e-05,
"loss": 0.1353,
"step": 4660
},
{
"epoch": 0.7457681251996168,
"grad_norm": 2.078125,
"learning_rate": 1.2644880875724406e-05,
"loss": 0.1094,
"step": 4670
},
{
"epoch": 0.7473650590865538,
"grad_norm": 1.9375,
"learning_rate": 1.2628783000643916e-05,
"loss": 0.1443,
"step": 4680
},
{
"epoch": 0.7489619929734909,
"grad_norm": 3.90625,
"learning_rate": 1.2612685125563427e-05,
"loss": 0.1159,
"step": 4690
},
{
"epoch": 0.750558926860428,
"grad_norm": 1.8125,
"learning_rate": 1.2596587250482937e-05,
"loss": 0.1184,
"step": 4700
},
{
"epoch": 0.7521558607473651,
"grad_norm": 0.91796875,
"learning_rate": 1.2580489375402447e-05,
"loss": 0.1138,
"step": 4710
},
{
"epoch": 0.7537527946343021,
"grad_norm": 2.21875,
"learning_rate": 1.2564391500321957e-05,
"loss": 0.1536,
"step": 4720
},
{
"epoch": 0.7553497285212393,
"grad_norm": 2.359375,
"learning_rate": 1.254829362524147e-05,
"loss": 0.1191,
"step": 4730
},
{
"epoch": 0.7569466624081763,
"grad_norm": 1.7265625,
"learning_rate": 1.253219575016098e-05,
"loss": 0.1293,
"step": 4740
},
{
"epoch": 0.7585435962951134,
"grad_norm": 1.7421875,
"learning_rate": 1.251609787508049e-05,
"loss": 0.1512,
"step": 4750
},
{
"epoch": 0.7601405301820504,
"grad_norm": 0.953125,
"learning_rate": 1.25e-05,
"loss": 0.1154,
"step": 4760
},
{
"epoch": 0.7617374640689876,
"grad_norm": 3.421875,
"learning_rate": 1.248390212491951e-05,
"loss": 0.1252,
"step": 4770
},
{
"epoch": 0.7633343979559246,
"grad_norm": 1.609375,
"learning_rate": 1.2467804249839022e-05,
"loss": 0.0981,
"step": 4780
},
{
"epoch": 0.7639731715106994,
"eval_loss": 0.12583701312541962,
"eval_runtime": 25.1086,
"eval_samples_per_second": 10.116,
"eval_steps_per_second": 10.116,
"step": 4784
},
{
"epoch": 0.7649313318428617,
"grad_norm": 2.171875,
"learning_rate": 1.2451706374758533e-05,
"loss": 0.1523,
"step": 4790
},
{
"epoch": 0.7665282657297988,
"grad_norm": 1.78125,
"learning_rate": 1.2435608499678044e-05,
"loss": 0.1389,
"step": 4800
},
{
"epoch": 0.7681251996167359,
"grad_norm": 1.2109375,
"learning_rate": 1.2419510624597554e-05,
"loss": 0.1287,
"step": 4810
},
{
"epoch": 0.7697221335036729,
"grad_norm": 1.7734375,
"learning_rate": 1.2403412749517065e-05,
"loss": 0.134,
"step": 4820
},
{
"epoch": 0.7713190673906101,
"grad_norm": 2.140625,
"learning_rate": 1.2387314874436576e-05,
"loss": 0.1244,
"step": 4830
},
{
"epoch": 0.7729160012775471,
"grad_norm": 3.859375,
"learning_rate": 1.2371216999356086e-05,
"loss": 0.142,
"step": 4840
},
{
"epoch": 0.7745129351644842,
"grad_norm": 1.0546875,
"learning_rate": 1.2355119124275597e-05,
"loss": 0.1281,
"step": 4850
},
{
"epoch": 0.7761098690514213,
"grad_norm": 1.6796875,
"learning_rate": 1.2339021249195107e-05,
"loss": 0.1214,
"step": 4860
},
{
"epoch": 0.7777068029383584,
"grad_norm": 1.6015625,
"learning_rate": 1.2322923374114617e-05,
"loss": 0.129,
"step": 4870
},
{
"epoch": 0.7793037368252954,
"grad_norm": 1.8671875,
"learning_rate": 1.230682549903413e-05,
"loss": 0.141,
"step": 4880
},
{
"epoch": 0.7809006707122325,
"grad_norm": 0.96484375,
"learning_rate": 1.229072762395364e-05,
"loss": 0.1237,
"step": 4890
},
{
"epoch": 0.7824976045991696,
"grad_norm": 1.5859375,
"learning_rate": 1.227462974887315e-05,
"loss": 0.1217,
"step": 4900
},
{
"epoch": 0.7840945384861067,
"grad_norm": 1.8984375,
"learning_rate": 1.225853187379266e-05,
"loss": 0.1371,
"step": 4910
},
{
"epoch": 0.7856914723730437,
"grad_norm": 1.4140625,
"learning_rate": 1.224243399871217e-05,
"loss": 0.0956,
"step": 4920
},
{
"epoch": 0.7872884062599809,
"grad_norm": 2.109375,
"learning_rate": 1.222633612363168e-05,
"loss": 0.1277,
"step": 4930
},
{
"epoch": 0.7888853401469179,
"grad_norm": 1.84375,
"learning_rate": 1.2210238248551191e-05,
"loss": 0.1142,
"step": 4940
},
{
"epoch": 0.790482274033855,
"grad_norm": 1.859375,
"learning_rate": 1.2194140373470703e-05,
"loss": 0.1223,
"step": 4950
},
{
"epoch": 0.7920792079207921,
"grad_norm": 1.640625,
"learning_rate": 1.2178042498390214e-05,
"loss": 0.1444,
"step": 4960
},
{
"epoch": 0.7936761418077292,
"grad_norm": 1.796875,
"learning_rate": 1.2161944623309725e-05,
"loss": 0.098,
"step": 4970
},
{
"epoch": 0.7952730756946662,
"grad_norm": 1.578125,
"learning_rate": 1.2145846748229235e-05,
"loss": 0.1263,
"step": 4980
},
{
"epoch": 0.7968700095816034,
"grad_norm": 1.546875,
"learning_rate": 1.2129748873148746e-05,
"loss": 0.1723,
"step": 4990
},
{
"epoch": 0.7971893963589908,
"eval_loss": 0.1247447282075882,
"eval_runtime": 25.2078,
"eval_samples_per_second": 10.076,
"eval_steps_per_second": 10.076,
"step": 4992
},
{
"epoch": 0.7984669434685404,
"grad_norm": 1.7265625,
"learning_rate": 1.2113650998068257e-05,
"loss": 0.1256,
"step": 5000
},
{
"epoch": 0.8000638773554775,
"grad_norm": 2.078125,
"learning_rate": 1.2097553122987767e-05,
"loss": 0.171,
"step": 5010
},
{
"epoch": 0.8016608112424146,
"grad_norm": 1.6640625,
"learning_rate": 1.2081455247907277e-05,
"loss": 0.1306,
"step": 5020
},
{
"epoch": 0.8032577451293517,
"grad_norm": 0.8984375,
"learning_rate": 1.2065357372826787e-05,
"loss": 0.1248,
"step": 5030
},
{
"epoch": 0.8048546790162887,
"grad_norm": 1.9375,
"learning_rate": 1.20492594977463e-05,
"loss": 0.1152,
"step": 5040
},
{
"epoch": 0.8064516129032258,
"grad_norm": 2.046875,
"learning_rate": 1.203316162266581e-05,
"loss": 0.1551,
"step": 5050
},
{
"epoch": 0.8080485467901629,
"grad_norm": 1.4140625,
"learning_rate": 1.201706374758532e-05,
"loss": 0.1772,
"step": 5060
},
{
"epoch": 0.8096454806771,
"grad_norm": 2.328125,
"learning_rate": 1.200096587250483e-05,
"loss": 0.1431,
"step": 5070
},
{
"epoch": 0.811242414564037,
"grad_norm": 2.328125,
"learning_rate": 1.198486799742434e-05,
"loss": 0.1204,
"step": 5080
},
{
"epoch": 0.8128393484509742,
"grad_norm": 1.984375,
"learning_rate": 1.196877012234385e-05,
"loss": 0.1232,
"step": 5090
},
{
"epoch": 0.8144362823379112,
"grad_norm": 2.203125,
"learning_rate": 1.1952672247263361e-05,
"loss": 0.1272,
"step": 5100
},
{
"epoch": 0.8160332162248483,
"grad_norm": 2.390625,
"learning_rate": 1.1936574372182873e-05,
"loss": 0.1192,
"step": 5110
},
{
"epoch": 0.8176301501117854,
"grad_norm": 2.21875,
"learning_rate": 1.1920476497102384e-05,
"loss": 0.1352,
"step": 5120
},
{
"epoch": 0.8192270839987225,
"grad_norm": 4.96875,
"learning_rate": 1.1904378622021895e-05,
"loss": 0.1213,
"step": 5130
},
{
"epoch": 0.8208240178856595,
"grad_norm": 1.75,
"learning_rate": 1.1888280746941405e-05,
"loss": 0.1181,
"step": 5140
},
{
"epoch": 0.8224209517725967,
"grad_norm": 1.984375,
"learning_rate": 1.1872182871860915e-05,
"loss": 0.1416,
"step": 5150
},
{
"epoch": 0.8240178856595337,
"grad_norm": 3.515625,
"learning_rate": 1.1856084996780427e-05,
"loss": 0.1216,
"step": 5160
},
{
"epoch": 0.8256148195464708,
"grad_norm": 3.78125,
"learning_rate": 1.1839987121699937e-05,
"loss": 0.1637,
"step": 5170
},
{
"epoch": 0.8272117534334078,
"grad_norm": 1.3984375,
"learning_rate": 1.1823889246619447e-05,
"loss": 0.119,
"step": 5180
},
{
"epoch": 0.828808687320345,
"grad_norm": 2.140625,
"learning_rate": 1.1807791371538957e-05,
"loss": 0.1551,
"step": 5190
},
{
"epoch": 0.830405621207282,
"grad_norm": 1.4765625,
"learning_rate": 1.1791693496458469e-05,
"loss": 0.1357,
"step": 5200
},
{
"epoch": 0.830405621207282,
"eval_loss": 0.1232486441731453,
"eval_runtime": 25.0692,
"eval_samples_per_second": 10.132,
"eval_steps_per_second": 10.132,
"step": 5200
},
{
"epoch": 0.832002555094219,
"grad_norm": 1.8046875,
"learning_rate": 1.177559562137798e-05,
"loss": 0.1239,
"step": 5210
},
{
"epoch": 0.8335994889811562,
"grad_norm": 1.921875,
"learning_rate": 1.175949774629749e-05,
"loss": 0.1207,
"step": 5220
},
{
"epoch": 0.8351964228680933,
"grad_norm": 1.8125,
"learning_rate": 1.1743399871217e-05,
"loss": 0.1317,
"step": 5230
},
{
"epoch": 0.8367933567550303,
"grad_norm": 1.9140625,
"learning_rate": 1.172730199613651e-05,
"loss": 0.1796,
"step": 5240
},
{
"epoch": 0.8383902906419675,
"grad_norm": 1.7734375,
"learning_rate": 1.1711204121056021e-05,
"loss": 0.1412,
"step": 5250
},
{
"epoch": 0.8399872245289045,
"grad_norm": 0.98828125,
"learning_rate": 1.1695106245975531e-05,
"loss": 0.155,
"step": 5260
},
{
"epoch": 0.8415841584158416,
"grad_norm": 2.625,
"learning_rate": 1.1679008370895043e-05,
"loss": 0.0997,
"step": 5270
},
{
"epoch": 0.8431810923027787,
"grad_norm": 1.640625,
"learning_rate": 1.1662910495814555e-05,
"loss": 0.1321,
"step": 5280
},
{
"epoch": 0.8447780261897158,
"grad_norm": 1.203125,
"learning_rate": 1.1646812620734065e-05,
"loss": 0.1221,
"step": 5290
},
{
"epoch": 0.8463749600766528,
"grad_norm": 2.421875,
"learning_rate": 1.1630714745653575e-05,
"loss": 0.129,
"step": 5300
},
{
"epoch": 0.84797189396359,
"grad_norm": 1.6640625,
"learning_rate": 1.1614616870573085e-05,
"loss": 0.1445,
"step": 5310
},
{
"epoch": 0.849568827850527,
"grad_norm": 3.640625,
"learning_rate": 1.1598518995492597e-05,
"loss": 0.1316,
"step": 5320
},
{
"epoch": 0.851165761737464,
"grad_norm": 1.953125,
"learning_rate": 1.1582421120412107e-05,
"loss": 0.1228,
"step": 5330
},
{
"epoch": 0.8527626956244011,
"grad_norm": 3.71875,
"learning_rate": 1.1566323245331617e-05,
"loss": 0.0922,
"step": 5340
},
{
"epoch": 0.8543596295113383,
"grad_norm": 1.2109375,
"learning_rate": 1.1550225370251127e-05,
"loss": 0.1355,
"step": 5350
},
{
"epoch": 0.8559565633982753,
"grad_norm": 2.015625,
"learning_rate": 1.1534127495170639e-05,
"loss": 0.1704,
"step": 5360
},
{
"epoch": 0.8575534972852124,
"grad_norm": 2.6875,
"learning_rate": 1.151802962009015e-05,
"loss": 0.1314,
"step": 5370
},
{
"epoch": 0.8591504311721495,
"grad_norm": 2.796875,
"learning_rate": 1.150193174500966e-05,
"loss": 0.1491,
"step": 5380
},
{
"epoch": 0.8607473650590866,
"grad_norm": 2.234375,
"learning_rate": 1.148583386992917e-05,
"loss": 0.1296,
"step": 5390
},
{
"epoch": 0.8623442989460236,
"grad_norm": 2.9375,
"learning_rate": 1.146973599484868e-05,
"loss": 0.1301,
"step": 5400
},
{
"epoch": 0.8636218460555734,
"eval_loss": 0.12359596788883209,
"eval_runtime": 25.006,
"eval_samples_per_second": 10.158,
"eval_steps_per_second": 10.158,
"step": 5408
},
{
"epoch": 0.8639412328329608,
"grad_norm": 2.5,
"learning_rate": 1.1453638119768191e-05,
"loss": 0.1279,
"step": 5410
},
{
"epoch": 0.8655381667198978,
"grad_norm": 1.328125,
"learning_rate": 1.1437540244687701e-05,
"loss": 0.107,
"step": 5420
},
{
"epoch": 0.8671351006068349,
"grad_norm": 1.828125,
"learning_rate": 1.1421442369607211e-05,
"loss": 0.1226,
"step": 5430
},
{
"epoch": 0.868732034493772,
"grad_norm": 1.546875,
"learning_rate": 1.1405344494526725e-05,
"loss": 0.1261,
"step": 5440
},
{
"epoch": 0.8703289683807091,
"grad_norm": 1.8828125,
"learning_rate": 1.1389246619446235e-05,
"loss": 0.0931,
"step": 5450
},
{
"epoch": 0.8719259022676461,
"grad_norm": 2.71875,
"learning_rate": 1.1373148744365745e-05,
"loss": 0.1389,
"step": 5460
},
{
"epoch": 0.8735228361545831,
"grad_norm": 2.21875,
"learning_rate": 1.1357050869285255e-05,
"loss": 0.1269,
"step": 5470
},
{
"epoch": 0.8751197700415203,
"grad_norm": 1.6328125,
"learning_rate": 1.1340952994204765e-05,
"loss": 0.1117,
"step": 5480
},
{
"epoch": 0.8767167039284574,
"grad_norm": 2.140625,
"learning_rate": 1.1324855119124277e-05,
"loss": 0.1359,
"step": 5490
},
{
"epoch": 0.8783136378153944,
"grad_norm": 2.984375,
"learning_rate": 1.1308757244043787e-05,
"loss": 0.1418,
"step": 5500
},
{
"epoch": 0.8799105717023316,
"grad_norm": 2.828125,
"learning_rate": 1.1292659368963297e-05,
"loss": 0.1419,
"step": 5510
},
{
"epoch": 0.8815075055892686,
"grad_norm": 2.734375,
"learning_rate": 1.1276561493882809e-05,
"loss": 0.1354,
"step": 5520
},
{
"epoch": 0.8831044394762056,
"grad_norm": 1.5078125,
"learning_rate": 1.126046361880232e-05,
"loss": 0.1097,
"step": 5530
},
{
"epoch": 0.8847013733631428,
"grad_norm": 1.359375,
"learning_rate": 1.124436574372183e-05,
"loss": 0.0984,
"step": 5540
},
{
"epoch": 0.8862983072500799,
"grad_norm": 2.15625,
"learning_rate": 1.122826786864134e-05,
"loss": 0.0946,
"step": 5550
},
{
"epoch": 0.8878952411370169,
"grad_norm": 1.421875,
"learning_rate": 1.1212169993560851e-05,
"loss": 0.1278,
"step": 5560
},
{
"epoch": 0.8894921750239541,
"grad_norm": 0.96875,
"learning_rate": 1.1196072118480361e-05,
"loss": 0.1015,
"step": 5570
},
{
"epoch": 0.8910891089108911,
"grad_norm": 2.875,
"learning_rate": 1.1179974243399871e-05,
"loss": 0.135,
"step": 5580
},
{
"epoch": 0.8926860427978281,
"grad_norm": 1.765625,
"learning_rate": 1.1163876368319381e-05,
"loss": 0.1138,
"step": 5590
},
{
"epoch": 0.8942829766847652,
"grad_norm": 1.9609375,
"learning_rate": 1.1147778493238895e-05,
"loss": 0.1208,
"step": 5600
},
{
"epoch": 0.8958799105717024,
"grad_norm": 1.890625,
"learning_rate": 1.1131680618158405e-05,
"loss": 0.1028,
"step": 5610
},
{
"epoch": 0.8968380709038646,
"eval_loss": 0.1210414469242096,
"eval_runtime": 24.7689,
"eval_samples_per_second": 10.255,
"eval_steps_per_second": 10.255,
"step": 5616
},
{
"epoch": 0.8974768444586394,
"grad_norm": 1.9921875,
"learning_rate": 1.1115582743077915e-05,
"loss": 0.1076,
"step": 5620
},
{
"epoch": 0.8990737783455764,
"grad_norm": 1.484375,
"learning_rate": 1.1099484867997425e-05,
"loss": 0.1764,
"step": 5630
},
{
"epoch": 0.9006707122325136,
"grad_norm": 1.4765625,
"learning_rate": 1.1083386992916935e-05,
"loss": 0.1076,
"step": 5640
},
{
"epoch": 0.9022676461194507,
"grad_norm": 1.3828125,
"learning_rate": 1.1067289117836447e-05,
"loss": 0.1136,
"step": 5650
},
{
"epoch": 0.9038645800063877,
"grad_norm": 1.2265625,
"learning_rate": 1.1051191242755957e-05,
"loss": 0.106,
"step": 5660
},
{
"epoch": 0.9054615138933249,
"grad_norm": 1.890625,
"learning_rate": 1.1035093367675467e-05,
"loss": 0.111,
"step": 5670
},
{
"epoch": 0.9070584477802619,
"grad_norm": 1.6796875,
"learning_rate": 1.1018995492594979e-05,
"loss": 0.1285,
"step": 5680
},
{
"epoch": 0.908655381667199,
"grad_norm": 1.65625,
"learning_rate": 1.1002897617514489e-05,
"loss": 0.1101,
"step": 5690
},
{
"epoch": 0.9102523155541361,
"grad_norm": 1.796875,
"learning_rate": 1.0986799742434e-05,
"loss": 0.1303,
"step": 5700
},
{
"epoch": 0.9118492494410732,
"grad_norm": 2.734375,
"learning_rate": 1.097070186735351e-05,
"loss": 0.1638,
"step": 5710
},
{
"epoch": 0.9134461833280102,
"grad_norm": 1.2265625,
"learning_rate": 1.0954603992273021e-05,
"loss": 0.1201,
"step": 5720
},
{
"epoch": 0.9150431172149474,
"grad_norm": 1.53125,
"learning_rate": 1.0938506117192531e-05,
"loss": 0.0981,
"step": 5730
},
{
"epoch": 0.9166400511018844,
"grad_norm": 1.421875,
"learning_rate": 1.0922408242112041e-05,
"loss": 0.1256,
"step": 5740
},
{
"epoch": 0.9182369849888214,
"grad_norm": 1.703125,
"learning_rate": 1.0906310367031551e-05,
"loss": 0.152,
"step": 5750
},
{
"epoch": 0.9198339188757585,
"grad_norm": 1.46875,
"learning_rate": 1.0890212491951065e-05,
"loss": 0.1254,
"step": 5760
},
{
"epoch": 0.9214308527626957,
"grad_norm": 2.859375,
"learning_rate": 1.0874114616870575e-05,
"loss": 0.1593,
"step": 5770
},
{
"epoch": 0.9230277866496327,
"grad_norm": 1.4375,
"learning_rate": 1.0858016741790085e-05,
"loss": 0.1093,
"step": 5780
},
{
"epoch": 0.9246247205365697,
"grad_norm": 1.1484375,
"learning_rate": 1.0841918866709595e-05,
"loss": 0.1461,
"step": 5790
},
{
"epoch": 0.9262216544235069,
"grad_norm": 1.96875,
"learning_rate": 1.0825820991629105e-05,
"loss": 0.1092,
"step": 5800
},
{
"epoch": 0.927818588310444,
"grad_norm": 1.765625,
"learning_rate": 1.0809723116548615e-05,
"loss": 0.1215,
"step": 5810
},
{
"epoch": 0.929415522197381,
"grad_norm": 1.3125,
"learning_rate": 1.0793625241468127e-05,
"loss": 0.1543,
"step": 5820
},
{
"epoch": 0.9300542957521558,
"eval_loss": 0.1198168620467186,
"eval_runtime": 25.0926,
"eval_samples_per_second": 10.123,
"eval_steps_per_second": 10.123,
"step": 5824
},
{
"epoch": 0.9310124560843182,
"grad_norm": 1.96875,
"learning_rate": 1.0777527366387637e-05,
"loss": 0.1329,
"step": 5830
},
{
"epoch": 0.9326093899712552,
"grad_norm": 2.203125,
"learning_rate": 1.0761429491307149e-05,
"loss": 0.1128,
"step": 5840
},
{
"epoch": 0.9342063238581922,
"grad_norm": 1.515625,
"learning_rate": 1.0745331616226659e-05,
"loss": 0.129,
"step": 5850
},
{
"epoch": 0.9358032577451294,
"grad_norm": 1.765625,
"learning_rate": 1.072923374114617e-05,
"loss": 0.1141,
"step": 5860
},
{
"epoch": 0.9374001916320664,
"grad_norm": 2.8125,
"learning_rate": 1.0713135866065681e-05,
"loss": 0.1216,
"step": 5870
},
{
"epoch": 0.9389971255190035,
"grad_norm": 2.296875,
"learning_rate": 1.0697037990985191e-05,
"loss": 0.1501,
"step": 5880
},
{
"epoch": 0.9405940594059405,
"grad_norm": 1.84375,
"learning_rate": 1.0680940115904701e-05,
"loss": 0.1357,
"step": 5890
},
{
"epoch": 0.9421909932928777,
"grad_norm": 3.28125,
"learning_rate": 1.0664842240824211e-05,
"loss": 0.1511,
"step": 5900
},
{
"epoch": 0.9437879271798147,
"grad_norm": 1.7890625,
"learning_rate": 1.0648744365743721e-05,
"loss": 0.1752,
"step": 5910
},
{
"epoch": 0.9453848610667518,
"grad_norm": 1.3046875,
"learning_rate": 1.0632646490663235e-05,
"loss": 0.1318,
"step": 5920
},
{
"epoch": 0.946981794953689,
"grad_norm": 2.171875,
"learning_rate": 1.0616548615582745e-05,
"loss": 0.0973,
"step": 5930
},
{
"epoch": 0.948578728840626,
"grad_norm": 1.34375,
"learning_rate": 1.0600450740502255e-05,
"loss": 0.1397,
"step": 5940
},
{
"epoch": 0.950175662727563,
"grad_norm": 1.8125,
"learning_rate": 1.0584352865421765e-05,
"loss": 0.1132,
"step": 5950
},
{
"epoch": 0.9517725966145002,
"grad_norm": 1.4765625,
"learning_rate": 1.0568254990341275e-05,
"loss": 0.1176,
"step": 5960
},
{
"epoch": 0.9533695305014372,
"grad_norm": 1.7109375,
"learning_rate": 1.0552157115260785e-05,
"loss": 0.1493,
"step": 5970
},
{
"epoch": 0.9549664643883743,
"grad_norm": 1.4453125,
"learning_rate": 1.0536059240180297e-05,
"loss": 0.1141,
"step": 5980
},
{
"epoch": 0.9565633982753114,
"grad_norm": 2.296875,
"learning_rate": 1.0519961365099807e-05,
"loss": 0.1387,
"step": 5990
},
{
"epoch": 0.9581603321622485,
"grad_norm": 1.5390625,
"learning_rate": 1.0503863490019319e-05,
"loss": 0.1245,
"step": 6000
},
{
"epoch": 0.9597572660491855,
"grad_norm": 1.53125,
"learning_rate": 1.0487765614938829e-05,
"loss": 0.1307,
"step": 6010
},
{
"epoch": 0.9613541999361227,
"grad_norm": 3.546875,
"learning_rate": 1.047166773985834e-05,
"loss": 0.1479,
"step": 6020
},
{
"epoch": 0.9629511338230597,
"grad_norm": 1.9921875,
"learning_rate": 1.0455569864777851e-05,
"loss": 0.1244,
"step": 6030
},
{
"epoch": 0.9632705206004472,
"eval_loss": 0.11902373284101486,
"eval_runtime": 27.8248,
"eval_samples_per_second": 9.129,
"eval_steps_per_second": 9.129,
"step": 6032
}
],
"logging_steps": 10,
"max_steps": 12524,
"num_input_tokens_seen": 0,
"num_train_epochs": 2,
"save_steps": 208,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": false
},
"attributes": {}
}
},
"total_flos": 2.4880534024760525e+17,
"train_batch_size": 1,
"trial_name": null,
"trial_params": null
}