{
  "best_global_step": null,
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 0.9932104752667313,
  "eval_steps": 1024,
  "global_step": 21504,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.011823934229365849,
      "grad_norm": 1.9160995483398438,
      "learning_rate": 0.000498046875,
      "loss": 0.9917811751365662,
      "step": 256
    },
    {
      "epoch": 0.023647868458731697,
      "grad_norm": 4.747977256774902,
      "learning_rate": 0.000998046875,
      "loss": 0.7975242733955383,
      "step": 512
    },
    {
      "epoch": 0.03547180268809755,
      "grad_norm": 7.29379940032959,
      "learning_rate": 0.000999640996023194,
      "loss": 0.6437413692474365,
      "step": 768
    },
    {
      "epoch": 0.047295736917463395,
      "grad_norm": 9.162477493286133,
      "learning_rate": 0.0009985588674043958,
      "loss": 0.5476456880569458,
      "step": 1024
    },
    {
      "epoch": 0.047295736917463395,
      "eval_cos_loss": 0.39061775943051735,
      "eval_loss": 0.5246143476467698,
      "eval_mse_loss": 0.5246143476467698,
      "flow/cos_sim": 0.609382248904607,
      "flow/improvement_ratio": 0.6071757009448526,
      "flow/mag_ratio_mean": 0.6818114621182011,
      "flow/mag_ratio_std": 0.25719259335706224,
      "step": 1024
    },
    {
      "epoch": 0.047295736917463395,
      "eval_cos_loss": 0.39061775943051735,
      "eval_loss": 0.5246143476467698,
      "eval_mse_loss": 0.5246143476467698,
      "eval_runtime": 36.6551,
      "eval_samples_per_second": 763.687,
      "eval_steps_per_second": 11.949,
      "flow/cos_sim": 0.609382248904607,
      "flow/improvement_ratio": 0.6071757009448526,
      "flow/mag_ratio_mean": 0.6818114621182011,
      "flow/mag_ratio_std": 0.25719259335706224,
      "step": 1024
    },
    {
      "epoch": 0.05911967114682925,
      "grad_norm": 13.371840476989746,
      "learning_rate": 0.0009967551747861387,
      "loss": 0.5056764483451843,
      "step": 1280
    },
    {
      "epoch": 0.0709436053761951,
      "grad_norm": 7.342921733856201,
      "learning_rate": 0.000994232528651847,
      "loss": 0.4739161729812622,
      "step": 1536
    },
    {
      "epoch": 0.08276753960556095,
      "grad_norm": 9.060990333557129,
      "learning_rate": 0.0009909945800260092,
      "loss": 0.4664166569709778,
      "step": 1792
    },
    {
      "epoch": 0.09459147383492679,
      "grad_norm": 7.4525556564331055,
      "learning_rate": 0.0009870460151900522,
      "loss": 0.44106417894363403,
      "step": 2048
    },
    {
      "epoch": 0.09459147383492679,
      "eval_cos_loss": 0.3169162204344523,
      "eval_loss": 0.44313590847738255,
      "eval_mse_loss": 0.44313590847738255,
      "flow/cos_sim": 0.6830837719788835,
      "flow/improvement_ratio": 0.5721429720452932,
      "flow/mag_ratio_mean": 0.61852062170364,
      "flow/mag_ratio_std": 0.28984700054882867,
      "step": 2048
    },
    {
      "epoch": 0.09459147383492679,
      "eval_cos_loss": 0.3169162204344523,
      "eval_loss": 0.44313590847738255,
      "eval_mse_loss": 0.44313590847738255,
      "eval_runtime": 36.5832,
      "eval_samples_per_second": 765.188,
      "eval_steps_per_second": 11.973,
      "flow/cos_sim": 0.6830837719788835,
      "flow/improvement_ratio": 0.5721429720452932,
      "flow/mag_ratio_mean": 0.61852062170364,
      "flow/mag_ratio_std": 0.28984700054882867,
      "step": 2048
    },
    {
      "epoch": 0.10641540806429264,
      "grad_norm": 9.94619083404541,
      "learning_rate": 0.0009823925488998885,
      "loss": 0.42647701501846313,
      "step": 2304
    },
    {
      "epoch": 0.1182393422936585,
      "grad_norm": 3.24662446975708,
      "learning_rate": 0.0009770409161149525,
      "loss": 0.4398079514503479,
      "step": 2560
    },
    {
      "epoch": 0.13006327652302435,
      "grad_norm": 0.8419686555862427,
      "learning_rate": 0.0009709988622506973,
      "loss": 0.4244914650917053,
      "step": 2816
    },
    {
      "epoch": 0.1418872107523902,
      "grad_norm": 6.962368965148926,
      "learning_rate": 0.000964275131968659,
      "loss": 0.4100937843322754,
      "step": 3072
    },
    {
      "epoch": 0.1418872107523902,
      "eval_cos_loss": 0.2142252650571196,
      "eval_loss": 0.3256854459167071,
      "eval_mse_loss": 0.3256854459167071,
      "flow/cos_sim": 0.7857747346026712,
      "flow/improvement_ratio": 0.5133735555219868,
      "flow/mag_ratio_mean": 0.754958403437105,
      "flow/mag_ratio_std": 0.28671529593124784,
      "step": 3072
    },
    {
      "epoch": 0.1418872107523902,
      "eval_cos_loss": 0.2142252650571196,
      "eval_loss": 0.3256854459167071,
      "eval_mse_loss": 0.3256854459167071,
      "eval_runtime": 37.4143,
      "eval_samples_per_second": 748.189,
      "eval_steps_per_second": 11.707,
      "flow/cos_sim": 0.7857747346026712,
      "flow/improvement_ratio": 0.5133735555219868,
      "flow/mag_ratio_mean": 0.754958403437105,
      "flow/mag_ratio_std": 0.28671529593124784,
      "step": 3072
    },
    {
      "epoch": 0.15371114498175603,
      "grad_norm": 13.790922164916992,
      "learning_rate": 0.0009568794565203123,
      "loss": 0.4049389958381653,
      "step": 3328
    },
    {
      "epoch": 0.1655350792111219,
      "grad_norm": 2.3658220767974854,
      "learning_rate": 0.0009488225396630347,
      "loss": 0.3951943516731262,
      "step": 3584
    },
    {
      "epoch": 0.17735901344048774,
      "grad_norm": 10.210602760314941,
      "learning_rate": 0.0009401160421685646,
      "loss": 0.3869022727012634,
      "step": 3840
    },
    {
      "epoch": 0.18918294766985358,
      "grad_norm": 8.834197998046875,
      "learning_rate": 0.0009307725649463714,
      "loss": 0.39980870485305786,
      "step": 4096
    },
    {
      "epoch": 0.18918294766985358,
      "eval_cos_loss": 0.19631782169007275,
      "eval_loss": 0.305890281791012,
      "eval_mse_loss": 0.305890281791012,
      "flow/cos_sim": 0.8036821900981747,
      "flow/improvement_ratio": 0.49340173255090847,
      "flow/mag_ratio_mean": 0.7550306366458875,
      "flow/mag_ratio_std": 0.2717885251526963,
      "step": 4096
    },
    {
      "epoch": 0.18918294766985358,
      "eval_cos_loss": 0.19631782169007275,
      "eval_loss": 0.305890281791012,
      "eval_mse_loss": 0.305890281791012,
      "eval_runtime": 37.2111,
      "eval_samples_per_second": 752.276,
      "eval_steps_per_second": 11.771,
      "flow/cos_sim": 0.8036821900981747,
      "flow/improvement_ratio": 0.49340173255090847,
      "flow/mag_ratio_mean": 0.7550306366458875,
      "flow/mag_ratio_std": 0.2717885251526963,
      "step": 4096
    },
    {
      "epoch": 0.20100688189921945,
      "grad_norm": 2.6557016372680664,
      "learning_rate": 0.0009208056308063659,
      "loss": 0.38342079520225525,
      "step": 4352
    },
    {
      "epoch": 0.2128308161285853,
      "grad_norm": 6.585168838500977,
      "learning_rate": 0.0009102296648873445,
      "loss": 0.38296523690223694,
      "step": 4608
    },
    {
      "epoch": 0.22465475035795113,
      "grad_norm": 11.361368179321289,
      "learning_rate": 0.0008990599737794927,
      "loss": 0.3765565752983093,
      "step": 4864
    },
    {
      "epoch": 0.236478684587317,
      "grad_norm": 19.619009017944336,
      "learning_rate": 0.0008873127233711644,
      "loss": 0.3799913823604584,
      "step": 5120
    },
    {
      "epoch": 0.236478684587317,
      "eval_cos_loss": 0.2768729927877313,
      "eval_loss": 0.3794649503955014,
      "eval_mse_loss": 0.3794649503955014,
      "flow/cos_sim": 0.7231270148329538,
      "flow/improvement_ratio": 0.3986093916305124,
      "flow/mag_ratio_mean": 0.7333052162438223,
      "flow/mag_ratio_std": 0.2822123572366423,
      "step": 5120
    },
    {
      "epoch": 0.236478684587317,
      "eval_cos_loss": 0.2768729927877313,
      "eval_loss": 0.3794649503955014,
      "eval_mse_loss": 0.3794649503955014,
      "eval_runtime": 36.7774,
      "eval_samples_per_second": 761.148,
      "eval_steps_per_second": 11.91,
      "flow/cos_sim": 0.7231270148329538,
      "flow/improvement_ratio": 0.3986093916305124,
      "flow/mag_ratio_mean": 0.7333052162438223,
      "flow/mag_ratio_std": 0.2822123572366423,
      "step": 5120
    },
    {
      "epoch": 0.24830261881668284,
      "grad_norm": 8.770722389221191,
      "learning_rate": 0.0008750049154520011,
      "loss": 0.369044691324234,
      "step": 5376
    },
    {
      "epoch": 0.2601265530460487,
      "grad_norm": 8.493452072143555,
      "learning_rate": 0.0008621543631062487,
      "loss": 0.3745132386684418,
      "step": 5632
    },
    {
      "epoch": 0.27195048727541454,
      "grad_norm": 2.5794949531555176,
      "learning_rate": 0.0008487796649318904,
      "loss": 0.36342066526412964,
      "step": 5888
    },
    {
      "epoch": 0.2837744215047804,
      "grad_norm": 7.236816883087158,
      "learning_rate": 0.0008349001781229053,
      "loss": 0.35946616530418396,
      "step": 6144
    },
    {
      "epoch": 0.2837744215047804,
      "eval_cos_loss": 0.18904457571299654,
      "eval_loss": 0.29877711084063197,
      "eval_mse_loss": 0.29877711084063197,
      "flow/cos_sim": 0.8109554353097802,
      "flow/improvement_ratio": 0.4896438498747403,
      "flow/mag_ratio_mean": 0.7580637161590192,
      "flow/mag_ratio_std": 0.2673305695936016,
      "step": 6144
    },
    {
      "epoch": 0.2837744215047804,
      "eval_cos_loss": 0.18904457571299654,
      "eval_loss": 0.29877711084063197,
      "eval_mse_loss": 0.29877711084063197,
      "eval_runtime": 36.4777,
      "eval_samples_per_second": 767.399,
      "eval_steps_per_second": 12.007,
      "flow/cos_sim": 0.8109554353097802,
      "flow/improvement_ratio": 0.4896438498747403,
      "flow/mag_ratio_mean": 0.7580637161590192,
      "flow/mag_ratio_std": 0.2673305695936016,
      "step": 6144
    },
    {
      "epoch": 0.2955983557341462,
      "grad_norm": 9.964309692382812,
      "learning_rate": 0.0008205359904536107,
      "loss": 0.36158525943756104,
      "step": 6400
    },
    {
      "epoch": 0.30742228996351206,
      "grad_norm": 2.096620559692383,
      "learning_rate": 0.0008057078912056363,
      "loss": 0.3546257019042969,
      "step": 6656
    },
    {
      "epoch": 0.3192462241928779,
      "grad_norm": 4.746336936950684,
      "learning_rate": 0.0007904373410796086,
      "loss": 0.359122633934021,
      "step": 6912
    },
    {
      "epoch": 0.3310701584222438,
      "grad_norm": 3.691810369491577,
      "learning_rate": 0.0007747464411350876,
      "loss": 0.3498728573322296,
      "step": 7168
    },
    {
      "epoch": 0.3310701584222438,
      "eval_cos_loss": 0.20516517899556247,
      "eval_loss": 0.3059897372099363,
      "eval_mse_loss": 0.3059897372099363,
      "flow/cos_sim": 0.794834830717409,
      "flow/improvement_ratio": 0.5194083058262524,
      "flow/mag_ratio_mean": 0.7830329138666527,
      "flow/mag_ratio_std": 0.2623120350380466,
      "step": 7168
    },
    {
      "epoch": 0.3310701584222438,
      "eval_cos_loss": 0.20516517899556247,
      "eval_loss": 0.3059897372099363,
      "eval_mse_loss": 0.3059897372099363,
      "eval_runtime": 36.6184,
      "eval_samples_per_second": 764.452,
      "eval_steps_per_second": 11.961,
      "flow/cos_sim": 0.794834830717409,
      "flow/improvement_ratio": 0.5194083058262524,
      "flow/mag_ratio_mean": 0.7830329138666527,
      "flow/mag_ratio_std": 0.2623120350380466,
      "step": 7168
    },
    {
      "epoch": 0.34289409265160964,
      "grad_norm": 8.637118339538574,
      "learning_rate": 0.000758657900803716,
      "loss": 0.35182619094848633,
      "step": 7424
    },
    {
      "epoch": 0.3547180268809755,
      "grad_norm": 1.3778982162475586,
      "learning_rate": 0.000742195005021869,
      "loss": 0.3482975363731384,
      "step": 7680
    },
    {
      "epoch": 0.3665419611103413,
      "grad_norm": 6.048179626464844,
      "learning_rate": 0.0007253815805303786,
      "loss": 0.3527331054210663,
      "step": 7936
    },
    {
      "epoch": 0.37836589533970716,
      "grad_norm": 22.744041442871094,
      "learning_rate": 0.0007082419613901028,
      "loss": 0.35066965222358704,
      "step": 8192
    },
    {
      "epoch": 0.37836589533970716,
      "eval_cos_loss": 0.2029232965722748,
      "eval_loss": 0.3061500937427016,
      "eval_mse_loss": 0.3061500937427016,
      "flow/cos_sim": 0.7970767074251828,
      "flow/improvement_ratio": 0.5205303045440483,
      "flow/mag_ratio_mean": 0.7948372063299293,
      "flow/mag_ratio_std": 0.24897354126793064,
      "step": 8192
    },
    {
      "epoch": 0.37836589533970716,
      "eval_cos_loss": 0.2029232965722748,
      "eval_loss": 0.3061500937427016,
      "eval_mse_loss": 0.3061500937427016,
      "eval_runtime": 36.7266,
      "eval_samples_per_second": 762.2,
      "eval_steps_per_second": 11.926,
      "flow/cos_sim": 0.7970767074251828,
      "flow/improvement_ratio": 0.5205303045440483,
      "flow/mag_ratio_mean": 0.7948372063299293,
      "flow/mag_ratio_std": 0.24897354126793064,
      "step": 8192
    },
    {
      "epoch": 0.390189829569073,
      "grad_norm": 5.6811418533325195,
      "learning_rate": 0.0006908009537632514,
      "loss": 0.3552817404270172,
      "step": 8448
    },
    {
      "epoch": 0.4020137637984389,
      "grad_norm": 22.05255126953125,
      "learning_rate": 0.0006730838000114403,
      "loss": 0.3447907269001007,
      "step": 8704
    },
    {
      "epoch": 0.41383769802780473,
      "grad_norm": 2.4954335689544678,
      "learning_rate": 0.0006551161421624341,
      "loss": 0.3462049663066864,
      "step": 8960
    },
    {
      "epoch": 0.4256616322571706,
      "grad_norm": 10.035924911499023,
      "learning_rate": 0.0006369239847984517,
      "loss": 0.3400329649448395,
      "step": 9216
    },
    {
      "epoch": 0.4256616322571706,
      "eval_cos_loss": 0.18391099199652672,
      "eval_loss": 0.28414961062882044,
      "eval_mse_loss": 0.28414961062882044,
      "flow/cos_sim": 0.816089018839135,
      "flow/improvement_ratio": 0.49869729156635667,
      "flow/mag_ratio_mean": 0.8094866498964562,
      "flow/mag_ratio_std": 0.256197469147373,
      "step": 9216
    },
    {
      "epoch": 0.4256616322571706,
      "eval_cos_loss": 0.18391099199652672,
      "eval_loss": 0.28414961062882044,
      "eval_mse_loss": 0.28414961062882044,
      "eval_runtime": 36.6985,
      "eval_samples_per_second": 762.783,
      "eval_steps_per_second": 11.935,
      "flow/cos_sim": 0.816089018839135,
      "flow/improvement_ratio": 0.49869729156635667,
      "flow/mag_ratio_mean": 0.8094866498964562,
      "flow/mag_ratio_std": 0.256197469147373,
      "step": 9216
    },
    {
      "epoch": 0.4374855664865364,
      "grad_norm": 7.901397705078125,
      "learning_rate": 0.0006185336574197479,
      "loss": 0.3395940065383911,
      "step": 9472
    },
    {
      "epoch": 0.44930950071590225,
      "grad_norm": 6.7792510986328125,
      "learning_rate": 0.0005999717763379407,
      "loss": 0.33901962637901306,
      "step": 9728
    },
    {
      "epoch": 0.4611334349452681,
      "grad_norm": 2.6271603107452393,
      "learning_rate": 0.0005812652061542363,
      "loss": 0.34404292702674866,
      "step": 9984
    },
    {
      "epoch": 0.472957369174634,
      "grad_norm": 4.694519996643066,
      "learning_rate": 0.0005624410208783071,
      "loss": 0.33998560905456543,
      "step": 10240
    },
    {
      "epoch": 0.472957369174634,
      "eval_cos_loss": 0.19037291955321892,
      "eval_loss": 0.29135872845508193,
      "eval_mse_loss": 0.29135872845508193,
      "flow/cos_sim": 0.809627095041754,
      "flow/improvement_ratio": 0.5105767372536333,
      "flow/mag_ratio_mean": 0.8087259782775896,
      "flow/mag_ratio_std": 0.26619178907239816,
      "step": 10240
    },
    {
      "epoch": 0.472957369174634,
      "eval_cos_loss": 0.19037291955321892,
      "eval_loss": 0.29135872845508193,
      "eval_mse_loss": 0.29135872845508193,
      "eval_runtime": 36.2962,
      "eval_samples_per_second": 771.237,
      "eval_steps_per_second": 12.067,
      "flow/cos_sim": 0.809627095041754,
      "flow/improvement_ratio": 0.5105767372536333,
      "flow/mag_ratio_mean": 0.8087259782775896,
      "flow/mag_ratio_std": 0.26619178907239816,
      "step": 10240
    },
    {
      "epoch": 0.48478130340399983,
      "grad_norm": 6.458459377288818,
      "learning_rate": 0.0005435264647440881,
      "loss": 0.33370938897132874,
      "step": 10496
    },
    {
      "epoch": 0.49660523763336567,
      "grad_norm": 4.395213603973389,
      "learning_rate": 0.000524548912779213,
      "loss": 0.3374914824962616,
      "step": 10752
    },
    {
      "epoch": 0.5084291718627315,
      "grad_norm": 14.4677152633667,
      "learning_rate": 0.0005055358311851499,
      "loss": 0.3334283232688904,
      "step": 11008
    },
    {
      "epoch": 0.5202531060920974,
      "grad_norm": 9.418996810913086,
      "learning_rate": 0.0004865147375853812,
      "loss": 0.33191806077957153,
      "step": 11264
    },
    {
      "epoch": 0.5202531060920974,
      "eval_cos_loss": 0.19875142282751052,
      "eval_loss": 0.2990237715990032,
      "eval_mse_loss": 0.2990237715990032,
      "flow/cos_sim": 0.8012485821225327,
      "flow/improvement_ratio": 0.50358519822223,
      "flow/mag_ratio_mean": 0.7997565068066392,
      "flow/mag_ratio_std": 0.2594736643373694,
      "step": 11264
    },
    {
      "epoch": 0.5202531060920974,
      "eval_cos_loss": 0.19875142282751052,
      "eval_loss": 0.2990237715990032,
      "eval_mse_loss": 0.2990237715990032,
      "eval_runtime": 36.1011,
      "eval_samples_per_second": 775.405,
      "eval_steps_per_second": 12.133,
      "flow/cos_sim": 0.8012485821225327,
      "flow/improvement_ratio": 0.50358519822223,
      "flow/mag_ratio_mean": 0.7997565068066392,
      "flow/mag_ratio_std": 0.2594736643373694,
      "step": 11264
    },
    {
      "epoch": 0.5320770403214632,
      "grad_norm": 12.210289001464844,
      "learning_rate": 0.0004675131611991607,
      "loss": 0.33119097352027893,
      "step": 11520
    },
    {
      "epoch": 0.5439009745508291,
      "grad_norm": 5.221109867095947,
      "learning_rate": 0.0004485586029984899,
      "loss": 0.3348545432090759,
      "step": 11776
    },
    {
      "epoch": 0.5557249087801949,
      "grad_norm": 9.297412872314453,
      "learning_rate": 0.00042967849590597266,
      "loss": 0.3332735300064087,
      "step": 12032
    },
    {
      "epoch": 0.5675488430095608,
      "grad_norm": 11.681222915649414,
      "learning_rate": 0.0004109001650911621,
      "loss": 0.32956069707870483,
      "step": 12288
    },
    {
      "epoch": 0.5675488430095608,
      "eval_cos_loss": 0.19949678523831715,
      "eval_loss": 0.2990707095083036,
      "eval_mse_loss": 0.2990707095083036,
      "flow/cos_sim": 0.800503221821023,
      "flow/improvement_ratio": 0.45373746521396724,
      "flow/mag_ratio_mean": 0.8073946686640178,
      "flow/mag_ratio_std": 0.24824306055835393,
      "step": 12288
    },
    {
      "epoch": 0.5675488430095608,
      "eval_cos_loss": 0.19949678523831715,
      "eval_loss": 0.2990707095083036,
      "eval_mse_loss": 0.2990707095083036,
      "eval_runtime": 36.2453,
      "eval_samples_per_second": 772.321,
      "eval_steps_per_second": 12.084,
      "flow/cos_sim": 0.800503221821023,
      "flow/improvement_ratio": 0.45373746521396724,
      "flow/mag_ratio_mean": 0.8073946686640178,
      "flow/mag_ratio_std": 0.24824306055835393,
      "step": 12288
    },
    {
      "epoch": 0.5793727772389267,
      "grad_norm": 11.020252227783203,
      "learning_rate": 0.0003922507884228551,
      "loss": 0.3290621340274811,
      "step": 12544
    },
    {
      "epoch": 0.5911967114682924,
      "grad_norm": 12.451437950134277,
      "learning_rate": 0.00037375735713457723,
      "loss": 0.32443854212760925,
      "step": 12800
    },
    {
      "epoch": 0.6030206456976583,
      "grad_norm": 1.0615566968917847,
      "learning_rate": 0.00035544663676018276,
      "loss": 0.32868245244026184,
      "step": 13056
    },
    {
      "epoch": 0.6148445799270241,
      "grad_norm": 12.848718643188477,
      "learning_rate": 0.00033734512839611255,
      "loss": 0.32849442958831787,
      "step": 13312
    },
    {
      "epoch": 0.6148445799270241,
      "eval_cos_loss": 0.22722740034194297,
      "eval_loss": 0.3263611817332708,
      "eval_mse_loss": 0.3263611817332708,
      "flow/cos_sim": 0.7727726012060087,
      "flow/improvement_ratio": 0.5387330570993902,
      "flow/mag_ratio_mean": 0.8061602507552056,
      "flow/mag_ratio_std": 0.24406742666886277,
      "step": 13312
    },
    {
      "epoch": 0.6148445799270241,
      "eval_cos_loss": 0.22722740034194297,
      "eval_loss": 0.3263611817332708,
      "eval_mse_loss": 0.3263611817332708,
      "eval_runtime": 36.2088,
      "eval_samples_per_second": 773.1,
      "eval_steps_per_second": 12.097,
      "flow/cos_sim": 0.7727726012060087,
      "flow/improvement_ratio": 0.5387330570993902,
      "flow/mag_ratio_mean": 0.8061602507552056,
      "flow/mag_ratio_std": 0.24406742666886277,
      "step": 13312
    },
    {
      "epoch": 0.62666851415639,
      "grad_norm": 3.315863609313965,
      "learning_rate": 0.0003194790303463687,
      "loss": 0.32513225078582764,
      "step": 13568
    },
    {
      "epoch": 0.6384924483857558,
      "grad_norm": 0.8015891909599304,
      "learning_rate": 0.00030187420020572406,
      "loss": 0.32288485765457153,
      "step": 13824
    },
    {
      "epoch": 0.6503163826151217,
      "grad_norm": 8.767995834350586,
      "learning_rate": 0.00028455611743603626,
      "loss": 0.31935569643974304,
      "step": 14080
    },
    {
      "epoch": 0.6621403168444876,
      "grad_norm": 2.7436001300811768,
      "learning_rate": 0.0002675498464898373,
      "loss": 0.3218556344509125,
      "step": 14336
    },
    {
      "epoch": 0.6621403168444876,
      "eval_cos_loss": 0.17437796600877423,
      "eval_loss": 0.27404612850516896,
      "eval_mse_loss": 0.27404612850516896,
      "flow/cos_sim": 0.8256220499130145,
      "flow/improvement_ratio": 0.4894021302461624,
      "flow/mag_ratio_mean": 0.8008586715345514,
      "flow/mag_ratio_std": 0.2502900305199841,
      "step": 14336
    },
    {
      "epoch": 0.6621403168444876,
      "eval_cos_loss": 0.17437796600877423,
      "eval_loss": 0.27404612850516896,
      "eval_mse_loss": 0.27404612850516896,
      "eval_runtime": 36.4787,
      "eval_samples_per_second": 767.38,
      "eval_steps_per_second": 12.007,
      "flow/cos_sim": 0.8256220499130145,
      "flow/improvement_ratio": 0.4894021302461624,
      "flow/mag_ratio_mean": 0.8008586715345514,
      "flow/mag_ratio_std": 0.2502900305199841,
      "step": 14336
    },
    {
      "epoch": 0.6739642510738534,
      "grad_norm": 1.7940927743911743,
      "learning_rate": 0.0002508800005345623,
      "loss": 0.32087063789367676,
      "step": 14592
    },
    {
      "epoch": 0.6857881853032193,
      "grad_norm": 3.976600170135498,
      "learning_rate": 0.00023457070582992562,
      "loss": 0.31643542647361755,
      "step": 14848
    },
    {
      "epoch": 0.6976121195325851,
      "grad_norm": 14.614996910095215,
      "learning_rate": 0.00021864556680999692,
      "loss": 0.31989988684654236,
      "step": 15104
    },
    {
      "epoch": 0.709436053761951,
      "grad_norm": 1.3015860319137573,
      "learning_rate": 0.0002031276319205152,
      "loss": 0.319489985704422,
      "step": 15360
    },
    {
      "epoch": 0.709436053761951,
      "eval_cos_loss": 0.1751825540883628,
      "eval_loss": 0.2740960814841262,
      "eval_mse_loss": 0.2740960814841262,
      "flow/cos_sim": 0.8248174480379444,
      "flow/improvement_ratio": 0.5053689104917387,
      "flow/mag_ratio_mean": 0.7900927232825048,
      "flow/mag_ratio_std": 0.24186882635229798,
      "step": 15360
    },
    {
      "epoch": 0.709436053761951,
      "eval_cos_loss": 0.1751825540883628,
      "eval_loss": 0.2740960814841262,
      "eval_mse_loss": 0.2740960814841262,
      "eval_runtime": 36.3752,
      "eval_samples_per_second": 769.563,
      "eval_steps_per_second": 12.041,
      "flow/cos_sim": 0.8248174480379444,
      "flow/improvement_ratio": 0.5053689104917387,
      "flow/mag_ratio_mean": 0.7900927232825048,
      "flow/mag_ratio_std": 0.24186882635229798,
      "step": 15360
    },
    {
      "epoch": 0.7212599879913169,
      "grad_norm": 0.6580540537834167,
      "learning_rate": 0.00018803936026088542,
      "loss": 0.31635990738868713,
      "step": 15616
    },
    {
      "epoch": 0.7330839222206826,
      "grad_norm": 4.1114654541015625,
      "learning_rate": 0.00017340258907913464,
      "loss": 0.3187476694583893,
      "step": 15872
    },
    {
      "epoch": 0.7449078564500485,
      "grad_norm": 12.402987480163574,
      "learning_rate": 0.0001592385021668743,
      "loss": 0.3155273497104645,
      "step": 16128
    },
    {
      "epoch": 0.7567317906794143,
      "grad_norm": 1.0991050004959106,
      "learning_rate": 0.0001455675992000087,
      "loss": 0.31325414776802063,
      "step": 16384
    },
    {
      "epoch": 0.7567317906794143,
      "eval_cos_loss": 0.17473208807306748,
      "eval_loss": 0.2738769624263184,
      "eval_mse_loss": 0.2738769624263184,
      "flow/cos_sim": 0.8252679411679098,
      "flow/improvement_ratio": 0.48874352702267093,
      "flow/mag_ratio_mean": 0.7912447798197673,
      "flow/mag_ratio_std": 0.25162723619643956,
      "step": 16384
    },
    {
      "epoch": 0.7567317906794143,
      "eval_cos_loss": 0.17473208807306748,
      "eval_loss": 0.2738769624263184,
      "eval_mse_loss": 0.2738769624263184,
      "eval_runtime": 36.1853,
      "eval_samples_per_second": 773.602,
      "eval_steps_per_second": 12.104,
      "flow/cos_sim": 0.8252679411679098,
      "flow/improvement_ratio": 0.48874352702267093,
      "flow/mag_ratio_mean": 0.7912447798197673,
      "flow/mag_ratio_std": 0.25162723619643956,
      "step": 16384
    },
    {
      "epoch": 0.7685557249087802,
      "grad_norm": 1.9538664817810059,
      "learning_rate": 0.000132409666069565,
      "loss": 0.31205466389656067,
      "step": 16640
    },
    {
      "epoch": 0.780379659138146,
      "grad_norm": 3.74383544921875,
      "learning_rate": 0.0001197837462455823,
      "loss": 0.3163967728614807,
      "step": 16896
    },
    {
      "epoch": 0.7922035933675119,
      "grad_norm": 2.0866622924804688,
      "learning_rate": 0.00010770811321550749,
      "loss": 0.3133208453655243,
      "step": 17152
    },
    {
      "epoch": 0.8040275275968778,
      "grad_norm": 3.845749855041504,
      "learning_rate": 9.620024403698591e-05,
      "loss": 0.31291234493255615,
      "step": 17408
    },
    {
      "epoch": 0.8040275275968778,
      "eval_cos_loss": 0.17509721335210757,
      "eval_loss": 0.273063686272325,
      "eval_mse_loss": 0.273063686272325,
      "flow/cos_sim": 0.824902795357247,
      "flow/improvement_ratio": 0.5037120235020711,
      "flow/mag_ratio_mean": 0.8056292350161566,
      "flow/mag_ratio_std": 0.23707074216103444,
      "step": 17408
    },
    {
      "epoch": 0.8040275275968778,
      "eval_cos_loss": 0.17509721335210757,
      "eval_loss": 0.273063686272325,
      "eval_mse_loss": 0.273063686272325,
      "eval_runtime": 36.2474,
      "eval_samples_per_second": 772.277,
      "eval_steps_per_second": 12.084,
      "flow/cos_sim": 0.824902795357247,
      "flow/improvement_ratio": 0.5037120235020711,
      "flow/mag_ratio_mean": 0.8056292350161566,
      "flow/mag_ratio_std": 0.23707074216103444,
      "step": 17408
    },
    {
      "epoch": 0.8158514618262436,
      "grad_norm": 4.972906112670898,
      "learning_rate": 8.527679404332429e-05,
      "loss": 0.3136768639087677,
      "step": 17664
    },
    {
      "epoch": 0.8276753960556095,
      "grad_norm": 0.5566576719284058,
      "learning_rate": 7.495357273823544e-05,
      "loss": 0.30899107456207275,
      "step": 17920
    },
    {
      "epoch": 0.8394993302849753,
      "grad_norm": 2.5373237133026123,
      "learning_rate": 6.524552091475183e-05,
      "loss": 0.30978360772132874,
      "step": 18176
    },
    {
      "epoch": 0.8513232645143411,
      "grad_norm": 8.832274436950684,
      "learning_rate": 5.6166689031422024e-05,
      "loss": 0.3114849030971527,
      "step": 18432
    },
    {
      "epoch": 0.8513232645143411,
      "eval_cos_loss": 0.179012101160609,
      "eval_loss": 0.27659866115137866,
      "eval_mse_loss": 0.27659866115137866,
      "flow/cos_sim": 0.8209879062219297,
      "flow/improvement_ratio": 0.5032780424236707,
      "flow/mag_ratio_mean": 0.8021674478707248,
      "flow/mag_ratio_std": 0.2417376903502364,
      "step": 18432
    },
    {
      "epoch": 0.8513232645143411,
      "eval_cos_loss": 0.179012101160609,
      "eval_loss": 0.27659866115137866,
      "eval_mse_loss": 0.27659866115137866,
      "eval_runtime": 36.438,
      "eval_samples_per_second": 768.236,
      "eval_steps_per_second": 12.02,
      "flow/cos_sim": 0.8209879062219297,
      "flow/improvement_ratio": 0.5032780424236707,
      "flow/mag_ratio_mean": 0.8021674478707248,
      "flow/mag_ratio_std": 0.2417376903502364,
      "step": 18432
    },
    {
      "epoch": 0.8631471987437069,
      "grad_norm": 6.2862749099731445,
      "learning_rate": 4.773021687709067e-05,
      "loss": 0.30685460567474365,
      "step": 18688
    },
    {
      "epoch": 0.8749711329730728,
      "grad_norm": 1.9315192699432373,
      "learning_rate": 3.994831455368719e-05,
      "loss": 0.31047487258911133,
      "step": 18944
    },
    {
      "epoch": 0.8867950672024387,
      "grad_norm": 8.073254585266113,
      "learning_rate": 3.283224480455282e-05,
      "loss": 0.3102303445339203,
      "step": 19200
    },
    {
      "epoch": 0.8986190014318045,
      "grad_norm": 0.3247526288032532,
      "learning_rate": 2.639230671387627e-05,
      "loss": 0.3080306351184845,
      "step": 19456
    },
    {
      "epoch": 0.8986190014318045,
      "eval_cos_loss": 0.17426143144363682,
      "eval_loss": 0.2714149275258796,
      "eval_mse_loss": 0.2714149275258796,
      "flow/cos_sim": 0.8257385824368969,
      "flow/improvement_ratio": 0.5061814522362191,
      "flow/mag_ratio_mean": 0.8065648428653474,
      "flow/mag_ratio_std": 0.23905256662738922,
      "step": 19456
    },
    {
      "epoch": 0.8986190014318045,
      "eval_cos_loss": 0.17426143144363682,
      "eval_loss": 0.2714149275258796,
      "eval_mse_loss": 0.2714149275258796,
      "eval_runtime": 36.6997,
      "eval_samples_per_second": 762.759,
      "eval_steps_per_second": 11.935,
      "flow/cos_sim": 0.8257385824368969,
      "flow/improvement_ratio": 0.5061814522362191,
      "flow/mag_ratio_mean": 0.8065648428653474,
      "flow/mag_ratio_std": 0.23905256662738922,
      "step": 19456
    },
    {
      "epoch": 0.9104429356611704,
      "grad_norm": 7.262045383453369,
      "learning_rate": 2.063782080083576e-05,
      "loss": 0.30695223808288574,
      "step": 19712
    },
    {
      "epoch": 0.9222668698905362,
      "grad_norm": 0.8199298977851868,
      "learning_rate": 1.557711553001523e-05,
      "loss": 0.3086055517196655,
      "step": 19968
    },
    {
      "epoch": 0.9340908041199021,
      "grad_norm": 3.6796634197235107,
      "learning_rate": 1.1217515257622269e-05,
      "loss": 0.30838918685913086,
      "step": 20224
    },
    {
      "epoch": 0.945914738349268,
      "grad_norm": 0.3800135850906372,
      "learning_rate": 7.565329630950746e-06,
      "loss": 0.3107503354549408,
      "step": 20480
    },
    {
      "epoch": 0.945914738349268,
      "eval_cos_loss": 0.16941133098991495,
      "eval_loss": 0.26742714126360473,
      "eval_mse_loss": 0.26742714126360473,
      "flow/cos_sim": 0.8305886884530386,
      "flow/improvement_ratio": 0.49329238075371745,
      "flow/mag_ratio_mean": 0.8075202496628783,
      "flow/mag_ratio_std": 0.24095512161107913,
      "step": 20480
    },
    {
      "epoch": 0.945914738349268,
      "eval_cos_loss": 0.16941133098991495,
      "eval_loss": 0.26742714126360473,
      "eval_mse_loss": 0.26742714126360473,
      "eval_runtime": 37.4646,
      "eval_samples_per_second": 747.185,
      "eval_steps_per_second": 11.691,
      "flow/cos_sim": 0.8305886884530386,
      "flow/improvement_ratio": 0.49329238075371745,
      "flow/mag_ratio_mean": 0.8075202496628783,
      "flow/mag_ratio_std": 0.24095512161107913,
      "step": 20480
    },
    {
      "epoch": 0.9577386725786338,
      "grad_norm": 5.383033752441406,
      "learning_rate": 4.62584445643166e-06,
      "loss": 0.30672797560691833,
      "step": 20736
    },
    {
      "epoch": 0.9695626068079997,
      "grad_norm": 1.579878807067871,
      "learning_rate": 2.40331404948807e-06,
      "loss": 0.30701693892478943,
      "step": 20992
    },
    {
      "epoch": 0.9813865410373654,
      "grad_norm": 4.654466152191162,
      "learning_rate": 9.009550772663965e-07,
      "loss": 0.30958274006843567,
      "step": 21248
    },
    {
      "epoch": 0.9932104752667313,
      "grad_norm": 0.809504508972168,
      "learning_rate": 1.2094190315575791e-07,
      "loss": 0.3090161085128784,
      "step": 21504
    },
    {
      "epoch": 0.9932104752667313,
      "eval_cos_loss": 0.16669115675910967,
      "eval_loss": 0.2628253948402731,
      "eval_mse_loss": 0.2628253948402731,
      "flow/cos_sim": 0.8333088479357768,
      "flow/improvement_ratio": 0.5023197360915136,
      "flow/mag_ratio_mean": 0.8090449382453204,
      "flow/mag_ratio_std": 0.23580366417289325,
      "step": 21504
    },
    {
      "epoch": 0.9932104752667313,
      "eval_cos_loss": 0.16669115675910967,
      "eval_loss": 0.2628253948402731,
      "eval_mse_loss": 0.2628253948402731,
      "eval_runtime": 37.5609,
      "eval_samples_per_second": 745.271,
      "eval_steps_per_second": 11.661,
      "flow/cos_sim": 0.8333088479357768,
      "flow/improvement_ratio": 0.5023197360915136,
      "flow/mag_ratio_mean": 0.8090449382453204,
      "flow/mag_ratio_std": 0.23580366417289325,
      "step": 21504
    }
  ],
  "logging_steps": 256,
  "max_steps": 21651,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 1,
  "save_steps": 1024,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": false
      },
      "attributes": {}
    }
  },
  "total_flos": 0.0,
  "train_batch_size": 64,
  "trial_name": null,
  "trial_params": null
}