{
  "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": 0.30032017827033997,
      "learning_rate": 0.000498046875,
      "loss": 1.8424009084701538,
      "step": 256
    },
    {
      "epoch": 0.023647868458731697,
      "grad_norm": 0.48476606607437134,
      "learning_rate": 0.000998046875,
      "loss": 1.3167253732681274,
      "step": 512
    },
    {
      "epoch": 0.03547180268809755,
      "grad_norm": 0.46512553095817566,
      "learning_rate": 0.000999640996023194,
      "loss": 1.0889508724212646,
      "step": 768
    },
    {
      "epoch": 0.047295736917463395,
      "grad_norm": 0.3862663209438324,
      "learning_rate": 0.0009985588674043958,
      "loss": 0.9328931570053101,
      "step": 1024
    },
    {
      "epoch": 0.047295736917463395,
      "eval_cos_loss": 0.3909751661972368,
      "eval_loss": 0.859363008171456,
      "eval_mse_loss": 0.859363008171456,
      "flow/cos_sim": 0.6090248374089803,
      "flow/improvement_ratio": 0.96680837529435,
      "flow/mag_ratio_mean": 0.6285397670312559,
      "flow/mag_ratio_std": 0.2260889009865996,
      "step": 1024
    },
    {
      "epoch": 0.047295736917463395,
      "eval_cos_loss": 0.3909751661972368,
      "eval_loss": 0.859363008171456,
      "eval_mse_loss": 0.859363008171456,
      "eval_runtime": 27.6707,
      "eval_samples_per_second": 1011.649,
      "eval_steps_per_second": 15.829,
      "flow/cos_sim": 0.6090248374089803,
      "flow/improvement_ratio": 0.96680837529435,
      "flow/mag_ratio_mean": 0.6285397670312559,
      "flow/mag_ratio_std": 0.2260889009865996,
      "step": 1024
    },
    {
      "epoch": 0.05911967114682925,
      "grad_norm": 0.3437928557395935,
      "learning_rate": 0.0009967551747861387,
      "loss": 0.7950779795646667,
      "step": 1280
    },
    {
      "epoch": 0.0709436053761951,
      "grad_norm": 0.4011635482311249,
      "learning_rate": 0.000994232528651847,
      "loss": 0.6695980429649353,
      "step": 1536
    },
    {
      "epoch": 0.08276753960556095,
      "grad_norm": 0.49726057052612305,
      "learning_rate": 0.0009909945800260092,
      "loss": 0.5587890148162842,
      "step": 1792
    },
    {
      "epoch": 0.09459147383492679,
      "grad_norm": 0.29605454206466675,
      "learning_rate": 0.0009870460151900522,
      "loss": 0.4556920826435089,
      "step": 2048
    },
    {
      "epoch": 0.09459147383492679,
      "eval_cos_loss": 0.21293426390107908,
      "eval_loss": 0.4076748799243474,
      "eval_mse_loss": 0.4076748799243474,
      "flow/cos_sim": 0.7870657346020006,
      "flow/improvement_ratio": 0.9822775999160662,
      "flow/mag_ratio_mean": 0.7956366545805648,
      "flow/mag_ratio_std": 0.18839361525425627,
      "step": 2048
    },
    {
      "epoch": 0.09459147383492679,
      "eval_cos_loss": 0.21293426390107908,
      "eval_loss": 0.4076748799243474,
      "eval_mse_loss": 0.4076748799243474,
      "eval_runtime": 27.9237,
      "eval_samples_per_second": 1002.48,
      "eval_steps_per_second": 15.686,
      "flow/cos_sim": 0.7870657346020006,
      "flow/improvement_ratio": 0.9822775999160662,
      "flow/mag_ratio_mean": 0.7956366545805648,
      "flow/mag_ratio_std": 0.18839361525425627,
      "step": 2048
    },
    {
      "epoch": 0.10641540806429264,
      "grad_norm": 0.37945282459259033,
      "learning_rate": 0.0009823925488998885,
      "loss": 0.3644663393497467,
      "step": 2304
    },
    {
      "epoch": 0.1182393422936585,
      "grad_norm": 0.2784593999385834,
      "learning_rate": 0.0009770409161149525,
      "loss": 0.2877335250377655,
      "step": 2560
    },
    {
      "epoch": 0.13006327652302435,
      "grad_norm": 0.28011366724967957,
      "learning_rate": 0.0009709988622506973,
      "loss": 0.22570422291755676,
      "step": 2816
    },
    {
      "epoch": 0.1418872107523902,
      "grad_norm": 0.16544577479362488,
      "learning_rate": 0.000964275131968659,
      "loss": 0.18250930309295654,
      "step": 3072
    },
    {
      "epoch": 0.1418872107523902,
      "eval_cos_loss": 0.08555437008675919,
      "eval_loss": 0.16474426010427953,
      "eval_mse_loss": 0.16474426010427953,
      "flow/cos_sim": 0.9144456284503414,
      "flow/improvement_ratio": 0.9908438801221108,
      "flow/mag_ratio_mean": 0.9158023052836117,
      "flow/mag_ratio_std": 0.12712571353196553,
      "step": 3072
    },
    {
      "epoch": 0.1418872107523902,
      "eval_cos_loss": 0.08555437008675919,
      "eval_loss": 0.16474426010427953,
      "eval_mse_loss": 0.16474426010427953,
      "eval_runtime": 27.5022,
      "eval_samples_per_second": 1017.847,
      "eval_steps_per_second": 15.926,
      "flow/cos_sim": 0.9144456284503414,
      "flow/improvement_ratio": 0.9908438801221108,
      "flow/mag_ratio_mean": 0.9158023052836117,
      "flow/mag_ratio_std": 0.12712571353196553,
      "step": 3072
    },
    {
      "epoch": 0.15371114498175603,
      "grad_norm": 0.6755663752555847,
      "learning_rate": 0.0009568794565203123,
      "loss": 0.15251992642879486,
      "step": 3328
    },
    {
      "epoch": 0.1655350792111219,
      "grad_norm": 1.6686415672302246,
      "learning_rate": 0.0009488225396630347,
      "loss": 0.13442441821098328,
      "step": 3584
    },
    {
      "epoch": 0.17735901344048774,
      "grad_norm": 0.1821625530719757,
      "learning_rate": 0.0009401160421685646,
      "loss": 0.12373481690883636,
      "step": 3840
    },
    {
      "epoch": 0.18918294766985358,
      "grad_norm": 0.14941175282001495,
      "learning_rate": 0.0009307725649463714,
      "loss": 0.11532580107450485,
      "step": 4096
    },
    {
      "epoch": 0.18918294766985358,
      "eval_cos_loss": 0.05926439563263225,
      "eval_loss": 0.11456974646816515,
      "eval_mse_loss": 0.11456974646816515,
      "flow/cos_sim": 0.9407356054815528,
      "flow/improvement_ratio": 0.9921626927645784,
      "flow/mag_ratio_mean": 0.942024191764936,
      "flow/mag_ratio_std": 0.09949292650778,
      "step": 4096
    },
    {
      "epoch": 0.18918294766985358,
      "eval_cos_loss": 0.05926439563263225,
      "eval_loss": 0.11456974646816515,
      "eval_mse_loss": 0.11456974646816515,
      "eval_runtime": 28.5456,
      "eval_samples_per_second": 980.64,
      "eval_steps_per_second": 15.344,
      "flow/cos_sim": 0.9407356054815528,
      "flow/improvement_ratio": 0.9921626927645784,
      "flow/mag_ratio_mean": 0.942024191764936,
      "flow/mag_ratio_std": 0.09949292650778,
      "step": 4096
    },
    {
      "epoch": 0.20100688189921945,
      "grad_norm": 0.18909402191638947,
      "learning_rate": 0.0009208056308063659,
      "loss": 0.10999151319265366,
      "step": 4352
    },
    {
      "epoch": 0.2128308161285853,
      "grad_norm": 0.40145087242126465,
      "learning_rate": 0.0009102296648873445,
      "loss": 0.10512682795524597,
      "step": 4608
    },
    {
      "epoch": 0.22465475035795113,
      "grad_norm": 0.19676516950130463,
      "learning_rate": 0.0008990599737794927,
      "loss": 0.10141732543706894,
      "step": 4864
    },
    {
      "epoch": 0.236478684587317,
      "grad_norm": 0.17175287008285522,
      "learning_rate": 0.0008873127233711644,
      "loss": 0.09812680631875992,
      "step": 5120
    },
    {
      "epoch": 0.236478684587317,
      "eval_cos_loss": 0.050695853061191565,
      "eval_loss": 0.09832691372325432,
      "eval_mse_loss": 0.09832691372325432,
      "flow/cos_sim": 0.9493041464455052,
      "flow/improvement_ratio": 0.9926231454496515,
      "flow/mag_ratio_mean": 0.9585911660161737,
      "flow/mag_ratio_std": 0.08434220563331152,
      "step": 5120
    },
    {
      "epoch": 0.236478684587317,
      "eval_cos_loss": 0.050695853061191565,
      "eval_loss": 0.09832691372325432,
      "eval_mse_loss": 0.09832691372325432,
      "eval_runtime": 27.3575,
      "eval_samples_per_second": 1023.228,
      "eval_steps_per_second": 16.01,
      "flow/cos_sim": 0.9493041464455052,
      "flow/improvement_ratio": 0.9926231454496515,
      "flow/mag_ratio_mean": 0.9585911660161737,
      "flow/mag_ratio_std": 0.08434220563331152,
      "step": 5120
    },
    {
      "epoch": 0.24830261881668284,
      "grad_norm": 0.13017332553863525,
      "learning_rate": 0.0008750049154520011,
      "loss": 0.09549019485712051,
      "step": 5376
    },
    {
      "epoch": 0.2601265530460487,
      "grad_norm": 0.12280620634555817,
      "learning_rate": 0.0008621543631062487,
      "loss": 0.09369617700576782,
      "step": 5632
    },
    {
      "epoch": 0.27195048727541454,
      "grad_norm": 0.1377657949924469,
      "learning_rate": 0.0008487796649318904,
      "loss": 0.09171723574399948,
      "step": 5888
    },
    {
      "epoch": 0.2837744215047804,
      "grad_norm": 0.13840267062187195,
      "learning_rate": 0.0008349001781229053,
      "loss": 0.0928829088807106,
      "step": 6144
    },
    {
      "epoch": 0.2837744215047804,
      "eval_cos_loss": 0.04612090527058736,
      "eval_loss": 0.08980049764456814,
      "eval_mse_loss": 0.08980049764456814,
      "flow/cos_sim": 0.9538790959201447,
      "flow/improvement_ratio": 0.9942120166674052,
      "flow/mag_ratio_mean": 0.9548581241200503,
      "flow/mag_ratio_std": 0.07417129391812843,
      "step": 6144
    },
    {
      "epoch": 0.2837744215047804,
      "eval_cos_loss": 0.04612090527058736,
      "eval_loss": 0.08980049764456814,
      "eval_mse_loss": 0.08980049764456814,
      "eval_runtime": 27.8042,
      "eval_samples_per_second": 1006.79,
      "eval_steps_per_second": 15.753,
      "flow/cos_sim": 0.9538790959201447,
      "flow/improvement_ratio": 0.9942120166674052,
      "flow/mag_ratio_mean": 0.9548581241200503,
      "flow/mag_ratio_std": 0.07417129391812843,
      "step": 6144
    },
    {
      "epoch": 0.2955983557341462,
      "grad_norm": 0.10361608117818832,
      "learning_rate": 0.0008205359904536107,
      "loss": 0.08833561092615128,
      "step": 6400
    },
    {
      "epoch": 0.30742228996351206,
      "grad_norm": 0.40116435289382935,
      "learning_rate": 0.0008057078912056363,
      "loss": 0.08737602084875107,
      "step": 6656
    },
    {
      "epoch": 0.3192462241928779,
      "grad_norm": 0.12478947639465332,
      "learning_rate": 0.0007904373410796086,
      "loss": 0.08542584627866745,
      "step": 6912
    },
    {
      "epoch": 0.3310701584222438,
      "grad_norm": 0.4613255560398102,
      "learning_rate": 0.0007747464411350876,
      "loss": 0.08413137495517731,
      "step": 7168
    },
    {
      "epoch": 0.3310701584222438,
      "eval_cos_loss": 0.042676221596323734,
      "eval_loss": 0.08318097989785073,
      "eval_mse_loss": 0.08318097989785073,
      "flow/cos_sim": 0.9573237789820318,
      "flow/improvement_ratio": 0.9943706639553314,
      "flow/mag_ratio_mean": 0.9544149404519224,
      "flow/mag_ratio_std": 0.06704916739538652,
      "step": 7168
    },
    {
      "epoch": 0.3310701584222438,
      "eval_cos_loss": 0.042676221596323734,
      "eval_loss": 0.08318097989785073,
      "eval_mse_loss": 0.08318097989785073,
      "eval_runtime": 28.2024,
      "eval_samples_per_second": 992.576,
      "eval_steps_per_second": 15.531,
      "flow/cos_sim": 0.9573237789820318,
      "flow/improvement_ratio": 0.9943706639553314,
      "flow/mag_ratio_mean": 0.9544149404519224,
      "flow/mag_ratio_std": 0.06704916739538652,
      "step": 7168
    },
    {
      "epoch": 0.34289409265160964,
      "grad_norm": 0.3934769928455353,
      "learning_rate": 0.000758657900803716,
      "loss": 0.08310537040233612,
      "step": 7424
    },
    {
      "epoch": 0.3547180268809755,
      "grad_norm": 0.36915773153305054,
      "learning_rate": 0.000742195005021869,
      "loss": 0.08330061286687851,
      "step": 7680
    },
    {
      "epoch": 0.3665419611103413,
      "grad_norm": 0.11406463384628296,
      "learning_rate": 0.0007253815805303786,
      "loss": 0.08167832344770432,
      "step": 7936
    },
    {
      "epoch": 0.37836589533970716,
      "grad_norm": 0.10967294871807098,
      "learning_rate": 0.0007082419613901028,
      "loss": 0.08145447075366974,
      "step": 8192
    },
    {
      "epoch": 0.37836589533970716,
      "eval_cos_loss": 0.041529001344896885,
      "eval_loss": 0.08086973952648302,
      "eval_mse_loss": 0.08086973952648302,
      "flow/cos_sim": 0.9584709986976293,
      "flow/improvement_ratio": 0.9945931761232141,
      "flow/mag_ratio_mean": 0.9566228298563936,
      "flow/mag_ratio_std": 0.06226046868178942,
      "step": 8192
    },
    {
      "epoch": 0.37836589533970716,
      "eval_cos_loss": 0.041529001344896885,
      "eval_loss": 0.08086973952648302,
      "eval_mse_loss": 0.08086973952648302,
      "eval_runtime": 27.7752,
      "eval_samples_per_second": 1007.843,
      "eval_steps_per_second": 15.769,
      "flow/cos_sim": 0.9584709986976293,
      "flow/improvement_ratio": 0.9945931761232141,
      "flow/mag_ratio_mean": 0.9566228298563936,
      "flow/mag_ratio_std": 0.06226046868178942,
      "step": 8192
    },
    {
      "epoch": 0.390189829569073,
      "grad_norm": 0.25795963406562805,
      "learning_rate": 0.0006908009537632514,
      "loss": 0.0798477828502655,
      "step": 8448
    },
    {
      "epoch": 0.4020137637984389,
      "grad_norm": 0.1671624779701233,
      "learning_rate": 0.0006730838000114403,
      "loss": 0.079444020986557,
      "step": 8704
    },
    {
      "epoch": 0.41383769802780473,
      "grad_norm": 0.48908066749572754,
      "learning_rate": 0.0006551161421624341,
      "loss": 0.0788843110203743,
      "step": 8960
    },
    {
      "epoch": 0.4256616322571706,
      "grad_norm": 0.4118134677410126,
      "learning_rate": 0.0006369239847984517,
      "loss": 0.07910227030515671,
      "step": 9216
    },
    {
      "epoch": 0.4256616322571706,
      "eval_cos_loss": 0.04017602187075299,
      "eval_loss": 0.07839778928111678,
      "eval_mse_loss": 0.07839778928111678,
      "flow/cos_sim": 0.9598239789542542,
      "flow/improvement_ratio": 0.9939218965023076,
      "flow/mag_ratio_mean": 0.9669629585797384,
      "flow/mag_ratio_std": 0.05801496024511448,
      "step": 9216
    },
    {
      "epoch": 0.4256616322571706,
      "eval_cos_loss": 0.04017602187075299,
      "eval_loss": 0.07839778928111678,
      "eval_mse_loss": 0.07839778928111678,
      "eval_runtime": 27.7311,
      "eval_samples_per_second": 1009.444,
      "eval_steps_per_second": 15.795,
      "flow/cos_sim": 0.9598239789542542,
      "flow/improvement_ratio": 0.9939218965023076,
      "flow/mag_ratio_mean": 0.9669629585797384,
      "flow/mag_ratio_std": 0.05801496024511448,
      "step": 9216
    },
    {
      "epoch": 0.4374855664865364,
      "grad_norm": 0.2151251882314682,
      "learning_rate": 0.0006185336574197479,
      "loss": 0.07854308187961578,
      "step": 9472
    },
    {
      "epoch": 0.44930950071590225,
      "grad_norm": 0.06261777877807617,
      "learning_rate": 0.0005999717763379407,
      "loss": 0.07761576026678085,
      "step": 9728
    },
    {
      "epoch": 0.4611334349452681,
      "grad_norm": 0.4810551404953003,
      "learning_rate": 0.0005812652061542363,
      "loss": 0.0762307196855545,
      "step": 9984
    },
    {
      "epoch": 0.472957369174634,
      "grad_norm": 0.2286372035741806,
      "learning_rate": 0.0005624410208783071,
      "loss": 0.07773169875144958,
      "step": 10240
    },
    {
      "epoch": 0.472957369174634,
      "eval_cos_loss": 0.039269575551492436,
      "eval_loss": 0.07653046936885408,
      "eval_mse_loss": 0.07653046936885408,
      "flow/cos_sim": 0.9607304255711978,
      "flow/improvement_ratio": 0.9943927899343238,
      "flow/mag_ratio_mean": 0.9607061044266235,
      "flow/mag_ratio_std": 0.055606617375391805,
      "step": 10240
    },
    {
      "epoch": 0.472957369174634,
      "eval_cos_loss": 0.039269575551492436,
      "eval_loss": 0.07653046936885408,
      "eval_mse_loss": 0.07653046936885408,
      "eval_runtime": 27.9635,
      "eval_samples_per_second": 1001.054,
      "eval_steps_per_second": 15.663,
      "flow/cos_sim": 0.9607304255711978,
      "flow/improvement_ratio": 0.9943927899343238,
      "flow/mag_ratio_mean": 0.9607061044266235,
      "flow/mag_ratio_std": 0.055606617375391805,
      "step": 10240
    },
    {
      "epoch": 0.48478130340399983,
      "grad_norm": 0.11119114607572556,
      "learning_rate": 0.0005435264647440881,
      "loss": 0.07176047563552856,
      "step": 10496
    },
    {
      "epoch": 0.49660523763336567,
      "grad_norm": 0.345055490732193,
      "learning_rate": 0.000524548912779213,
      "loss": 0.056456491351127625,
      "step": 10752
    },
    {
      "epoch": 0.5084291718627315,
      "grad_norm": 0.21917253732681274,
      "learning_rate": 0.0005055358311851499,
      "loss": 0.053561244159936905,
      "step": 11008
    },
    {
      "epoch": 0.5202531060920974,
      "grad_norm": 0.21707729995250702,
      "learning_rate": 0.0004865147375853812,
      "loss": 0.05172141268849373,
      "step": 11264
    },
    {
      "epoch": 0.5202531060920974,
      "eval_cos_loss": 0.026124844331821623,
      "eval_loss": 0.0521071719082251,
      "eval_mse_loss": 0.0521071719082251,
      "flow/cos_sim": 0.9738751555958839,
      "flow/improvement_ratio": 0.9952182877281485,
      "flow/mag_ratio_mean": 0.9838547212620304,
      "flow/mag_ratio_std": 0.058393157618843256,
      "step": 11264
    },
    {
      "epoch": 0.5202531060920974,
      "eval_cos_loss": 0.026124844331821623,
      "eval_loss": 0.0521071719082251,
      "eval_mse_loss": 0.0521071719082251,
      "eval_runtime": 27.9732,
      "eval_samples_per_second": 1000.709,
      "eval_steps_per_second": 15.658,
      "flow/cos_sim": 0.9738751555958839,
      "flow/improvement_ratio": 0.9952182877281485,
      "flow/mag_ratio_mean": 0.9838547212620304,
      "flow/mag_ratio_std": 0.058393157618843256,
      "step": 11264
    },
    {
      "epoch": 0.5320770403214632,
      "grad_norm": 0.12577901780605316,
      "learning_rate": 0.0004675131611991607,
      "loss": 0.05032345652580261,
      "step": 11520
    },
    {
      "epoch": 0.5439009745508291,
      "grad_norm": 0.34826231002807617,
      "learning_rate": 0.0004485586029984899,
      "loss": 0.049038730561733246,
      "step": 11776
    },
    {
      "epoch": 0.5557249087801949,
      "grad_norm": 0.44762158393859863,
      "learning_rate": 0.00042967849590597266,
      "loss": 0.048636652529239655,
      "step": 12032
    },
    {
      "epoch": 0.5675488430095608,
      "grad_norm": 0.27833688259124756,
      "learning_rate": 0.0004109001650911621,
      "loss": 0.04764322564005852,
      "step": 12288
    },
    {
      "epoch": 0.5675488430095608,
      "eval_cos_loss": 0.024070425779818126,
      "eval_loss": 0.04753756304547939,
      "eval_mse_loss": 0.04753756304547939,
      "flow/cos_sim": 0.9759295767845084,
      "flow/improvement_ratio": 0.9942291939639609,
      "flow/mag_ratio_mean": 0.9780929656061408,
      "flow/mag_ratio_std": 0.05394861750202636,
      "step": 12288
    },
    {
      "epoch": 0.5675488430095608,
      "eval_cos_loss": 0.024070425779818126,
      "eval_loss": 0.04753756304547939,
      "eval_mse_loss": 0.04753756304547939,
      "eval_runtime": 27.6721,
      "eval_samples_per_second": 1011.595,
      "eval_steps_per_second": 15.828,
      "flow/cos_sim": 0.9759295767845084,
      "flow/improvement_ratio": 0.9942291939639609,
      "flow/mag_ratio_mean": 0.9780929656061408,
      "flow/mag_ratio_std": 0.05394861750202636,
      "step": 12288
    },
    {
      "epoch": 0.5793727772389267,
      "grad_norm": 0.17133653163909912,
      "learning_rate": 0.0003922507884228551,
      "loss": 0.04732891544699669,
      "step": 12544
    },
    {
      "epoch": 0.5911967114682924,
      "grad_norm": 0.08853697776794434,
      "learning_rate": 0.00037375735713457723,
      "loss": 0.04574156925082207,
      "step": 12800
    },
    {
      "epoch": 0.6030206456976583,
      "grad_norm": 0.14900043606758118,
      "learning_rate": 0.00035544663676018276,
      "loss": 0.0459522120654583,
      "step": 13056
    },
    {
      "epoch": 0.6148445799270241,
      "grad_norm": 0.10938149690628052,
      "learning_rate": 0.00033734512839611255,
      "loss": 0.04567023739218712,
      "step": 13312
    },
    {
      "epoch": 0.6148445799270241,
      "eval_cos_loss": 0.022640768034579276,
      "eval_loss": 0.044894084334373474,
      "eval_mse_loss": 0.044894084334373474,
      "flow/cos_sim": 0.977359231884621,
      "flow/improvement_ratio": 0.9948612654318004,
      "flow/mag_ratio_mean": 0.9749562089846014,
      "flow/mag_ratio_std": 0.049488119823543446,
      "step": 13312
    },
    {
      "epoch": 0.6148445799270241,
      "eval_cos_loss": 0.022640768034579276,
      "eval_loss": 0.044894084334373474,
      "eval_mse_loss": 0.044894084334373474,
      "eval_runtime": 28.6172,
      "eval_samples_per_second": 978.188,
      "eval_steps_per_second": 15.305,
      "flow/cos_sim": 0.977359231884621,
      "flow/improvement_ratio": 0.9948612654318004,
      "flow/mag_ratio_mean": 0.9749562089846014,
      "flow/mag_ratio_std": 0.049488119823543446,
      "step": 13312
    },
    {
      "epoch": 0.62666851415639,
      "grad_norm": 0.23983722925186157,
      "learning_rate": 0.0003194790303463687,
      "loss": 0.04453244432806969,
      "step": 13568
    },
    {
      "epoch": 0.6384924483857558,
      "grad_norm": 0.08603578805923462,
      "learning_rate": 0.00030187420020572406,
      "loss": 0.04426493123173714,
      "step": 13824
    },
    {
      "epoch": 0.6503163826151217,
      "grad_norm": 0.39926600456237793,
      "learning_rate": 0.00028455611743603626,
      "loss": 0.04409552738070488,
      "step": 14080
    },
    {
      "epoch": 0.6621403168444876,
      "grad_norm": 0.6597372889518738,
      "learning_rate": 0.0002675498464898373,
      "loss": 0.04343748837709427,
      "step": 14336
    },
    {
      "epoch": 0.6621403168444876,
      "eval_cos_loss": 0.022019475217925633,
      "eval_loss": 0.04345123091346869,
      "eval_mse_loss": 0.04345123091346869,
      "flow/cos_sim": 0.9779805275675368,
      "flow/improvement_ratio": 0.9940928038396791,
      "flow/mag_ratio_mean": 0.9784686472862278,
      "flow/mag_ratio_std": 0.04926606593440929,
      "step": 14336
    },
    {
      "epoch": 0.6621403168444876,
      "eval_cos_loss": 0.022019475217925633,
      "eval_loss": 0.04345123091346869,
      "eval_mse_loss": 0.04345123091346869,
      "eval_runtime": 29.1857,
      "eval_samples_per_second": 959.135,
      "eval_steps_per_second": 15.007,
      "flow/cos_sim": 0.9779805275675368,
      "flow/improvement_ratio": 0.9940928038396791,
      "flow/mag_ratio_mean": 0.9784686472862278,
      "flow/mag_ratio_std": 0.04926606593440929,
      "step": 14336
    },
    {
      "epoch": 0.6739642510738534,
      "grad_norm": 0.5350319743156433,
      "learning_rate": 0.0002508800005345623,
      "loss": 0.042832110077142715,
      "step": 14592
    },
    {
      "epoch": 0.6857881853032193,
      "grad_norm": 0.18352548778057098,
      "learning_rate": 0.00023457070582992562,
      "loss": 0.042873408645391464,
      "step": 14848
    },
    {
      "epoch": 0.6976121195325851,
      "grad_norm": 0.06461436301469803,
      "learning_rate": 0.00021864556680999692,
      "loss": 0.0426117368042469,
      "step": 15104
    },
    {
      "epoch": 0.709436053761951,
      "grad_norm": 0.07462433725595474,
      "learning_rate": 0.0002031276319205152,
      "loss": 0.04218180850148201,
      "step": 15360
    },
    {
      "epoch": 0.709436053761951,
      "eval_cos_loss": 0.021368508580136517,
      "eval_loss": 0.04225278059743583,
      "eval_mse_loss": 0.04225278059743583,
      "flow/cos_sim": 0.9786314918025988,
      "flow/improvement_ratio": 0.9940550222516604,
      "flow/mag_ratio_mean": 0.9777582721895279,
      "flow/mag_ratio_std": 0.0456332682916835,
      "step": 15360
    },
    {
      "epoch": 0.709436053761951,
      "eval_cos_loss": 0.021368508580136517,
      "eval_loss": 0.04225278059743583,
      "eval_mse_loss": 0.04225278059743583,
      "eval_runtime": 28.5705,
      "eval_samples_per_second": 979.785,
      "eval_steps_per_second": 15.33,
      "flow/cos_sim": 0.9786314918025988,
      "flow/improvement_ratio": 0.9940550222516604,
      "flow/mag_ratio_mean": 0.9777582721895279,
      "flow/mag_ratio_std": 0.0456332682916835,
      "step": 15360
    },
    {
      "epoch": 0.7212599879913169,
      "grad_norm": 0.04815741628408432,
      "learning_rate": 0.00018803936026088542,
      "loss": 0.04143794998526573,
      "step": 15616
    },
    {
      "epoch": 0.7330839222206826,
      "grad_norm": 0.07569579780101776,
      "learning_rate": 0.00017340258907913464,
      "loss": 0.04184219613671303,
      "step": 15872
    },
    {
      "epoch": 0.7449078564500485,
      "grad_norm": 0.19264720380306244,
      "learning_rate": 0.0001592385021668743,
      "loss": 0.041615359485149384,
      "step": 16128
    },
    {
      "epoch": 0.7567317906794143,
      "grad_norm": 0.09954190999269485,
      "learning_rate": 0.0001455675992000087,
      "loss": 0.04146020486950874,
      "step": 16384
    },
    {
      "epoch": 0.7567317906794143,
      "eval_cos_loss": 0.020705404857250108,
      "eval_loss": 0.04112391300567481,
      "eval_mse_loss": 0.04112391300567481,
      "flow/cos_sim": 0.979294594292227,
      "flow/improvement_ratio": 0.9948036799964295,
      "flow/mag_ratio_mean": 0.9830839078720301,
      "flow/mag_ratio_std": 0.04493439051250345,
      "step": 16384
    },
    {
      "epoch": 0.7567317906794143,
      "eval_cos_loss": 0.020705404857250108,
      "eval_loss": 0.04112391300567481,
      "eval_mse_loss": 0.04112391300567481,
      "eval_runtime": 27.7293,
      "eval_samples_per_second": 1009.511,
      "eval_steps_per_second": 15.796,
      "flow/cos_sim": 0.979294594292227,
      "flow/improvement_ratio": 0.9948036799964295,
      "flow/mag_ratio_mean": 0.9830839078720301,
      "flow/mag_ratio_std": 0.04493439051250345,
      "step": 16384
    },
    {
      "epoch": 0.7685557249087802,
      "grad_norm": 0.08436273038387299,
      "learning_rate": 0.000132409666069565,
      "loss": 0.041200295090675354,
      "step": 16640
    },
    {
      "epoch": 0.780379659138146,
      "grad_norm": 0.15963269770145416,
      "learning_rate": 0.0001197837462455823,
      "loss": 0.04084719717502594,
      "step": 16896
    },
    {
      "epoch": 0.7922035933675119,
      "grad_norm": 0.11329685896635056,
      "learning_rate": 0.00010770811321550749,
      "loss": 0.04111889377236366,
      "step": 17152
    },
    {
      "epoch": 0.8040275275968778,
      "grad_norm": 0.06041397154331207,
      "learning_rate": 9.620024403698591e-05,
      "loss": 0.04088206961750984,
      "step": 17408
    },
    {
      "epoch": 0.8040275275968778,
      "eval_cos_loss": 0.02060152174369113,
      "eval_loss": 0.040920868566999696,
      "eval_mse_loss": 0.040920868566999696,
      "flow/cos_sim": 0.9793984800168912,
      "flow/improvement_ratio": 0.9943572009262973,
      "flow/mag_ratio_mean": 0.9805345156965735,
      "flow/mag_ratio_std": 0.04472058159235406,
      "step": 17408
    },
    {
      "epoch": 0.8040275275968778,
      "eval_cos_loss": 0.02060152174369113,
      "eval_loss": 0.040920868566999696,
      "eval_mse_loss": 0.040920868566999696,
      "eval_runtime": 28.2966,
      "eval_samples_per_second": 989.272,
      "eval_steps_per_second": 15.479,
      "flow/cos_sim": 0.9793984800168912,
      "flow/improvement_ratio": 0.9943572009262973,
      "flow/mag_ratio_mean": 0.9805345156965735,
      "flow/mag_ratio_std": 0.04472058159235406,
      "step": 17408
    },
    {
      "epoch": 0.8158514618262436,
      "grad_norm": 0.412060409784317,
      "learning_rate": 8.527679404332429e-05,
      "loss": 0.04045189172029495,
      "step": 17664
    },
    {
      "epoch": 0.8276753960556095,
      "grad_norm": 0.0535939484834671,
      "learning_rate": 7.495357273823544e-05,
      "loss": 0.040322255343198776,
      "step": 17920
    },
    {
      "epoch": 0.8394993302849753,
      "grad_norm": 0.05492197349667549,
      "learning_rate": 6.524552091475183e-05,
      "loss": 0.04015512391924858,
      "step": 18176
    },
    {
      "epoch": 0.8513232645143411,
      "grad_norm": 1.343531847000122,
      "learning_rate": 5.6166689031422024e-05,
      "loss": 0.04034010320901871,
      "step": 18432
    },
    {
      "epoch": 0.8513232645143411,
      "eval_cos_loss": 0.02017251738096209,
      "eval_loss": 0.040136819041959225,
      "eval_mse_loss": 0.040136819041959225,
      "flow/cos_sim": 0.9798274840394111,
      "flow/improvement_ratio": 0.9949594466653588,
      "flow/mag_ratio_mean": 0.9830197368038299,
      "flow/mag_ratio_std": 0.04276561933106075,
      "step": 18432
    },
    {
      "epoch": 0.8513232645143411,
      "eval_cos_loss": 0.02017251738096209,
      "eval_loss": 0.040136819041959225,
      "eval_mse_loss": 0.040136819041959225,
      "eval_runtime": 28.9254,
      "eval_samples_per_second": 967.764,
      "eval_steps_per_second": 15.142,
      "flow/cos_sim": 0.9798274840394111,
      "flow/improvement_ratio": 0.9949594466653588,
      "flow/mag_ratio_mean": 0.9830197368038299,
      "flow/mag_ratio_std": 0.04276561933106075,
      "step": 18432
    },
    {
      "epoch": 0.8631471987437069,
      "grad_norm": 0.07045256346464157,
      "learning_rate": 4.773021687709067e-05,
      "loss": 0.04022738337516785,
      "step": 18688
    },
    {
      "epoch": 0.8749711329730728,
      "grad_norm": 0.04534813016653061,
      "learning_rate": 3.994831455368719e-05,
      "loss": 0.040090736001729965,
      "step": 18944
    },
    {
      "epoch": 0.8867950672024387,
      "grad_norm": 0.12301895767450333,
      "learning_rate": 3.283224480455282e-05,
      "loss": 0.03993465378880501,
      "step": 19200
    },
    {
      "epoch": 0.8986190014318045,
      "grad_norm": 0.030255919322371483,
      "learning_rate": 2.639230671387627e-05,
      "loss": 0.03986474499106407,
      "step": 19456
    },
    {
      "epoch": 0.8986190014318045,
      "eval_cos_loss": 0.01997290319894955,
      "eval_loss": 0.03975143301466556,
      "eval_mse_loss": 0.03975143301466556,
      "flow/cos_sim": 0.9800270978174254,
      "flow/improvement_ratio": 0.9948155708508949,
      "flow/mag_ratio_mean": 0.9813638247855722,
      "flow/mag_ratio_std": 0.0424358811586687,
      "step": 19456
    },
    {
      "epoch": 0.8986190014318045,
      "eval_cos_loss": 0.01997290319894955,
      "eval_loss": 0.03975143301466556,
      "eval_mse_loss": 0.03975143301466556,
      "eval_runtime": 29.2292,
      "eval_samples_per_second": 957.708,
      "eval_steps_per_second": 14.985,
      "flow/cos_sim": 0.9800270978174254,
      "flow/improvement_ratio": 0.9948155708508949,
      "flow/mag_ratio_mean": 0.9813638247855722,
      "flow/mag_ratio_std": 0.0424358811586687,
      "step": 19456
    },
    {
      "epoch": 0.9104429356611704,
      "grad_norm": 2.355804681777954,
      "learning_rate": 2.063782080083576e-05,
      "loss": 0.03972519189119339,
      "step": 19712
    },
    {
      "epoch": 0.9222668698905362,
      "grad_norm": 0.06069854274392128,
      "learning_rate": 1.557711553001523e-05,
      "loss": 0.039879463613033295,
      "step": 19968
    },
    {
      "epoch": 0.9340908041199021,
      "grad_norm": 0.029576338827610016,
      "learning_rate": 1.1217515257622269e-05,
      "loss": 0.039681050926446915,
      "step": 20224
    },
    {
      "epoch": 0.945914738349268,
      "grad_norm": 0.05208945646882057,
      "learning_rate": 7.565329630950746e-06,
      "loss": 0.03949534893035889,
      "step": 20480
    },
    {
      "epoch": 0.945914738349268,
      "eval_cos_loss": 0.019900511831181234,
      "eval_loss": 0.03963183302041058,
      "eval_mse_loss": 0.03963183302041058,
      "flow/cos_sim": 0.9800994916049313,
      "flow/improvement_ratio": 0.9955979625111846,
      "flow/mag_ratio_mean": 0.9811528873498037,
      "flow/mag_ratio_std": 0.04229976768397033,
      "step": 20480
    },
    {
      "epoch": 0.945914738349268,
      "eval_cos_loss": 0.019900511831181234,
      "eval_loss": 0.03963183302041058,
      "eval_mse_loss": 0.03963183302041058,
      "eval_runtime": 29.2534,
      "eval_samples_per_second": 956.914,
      "eval_steps_per_second": 14.973,
      "flow/cos_sim": 0.9800994916049313,
      "flow/improvement_ratio": 0.9955979625111846,
      "flow/mag_ratio_mean": 0.9811528873498037,
      "flow/mag_ratio_std": 0.04229976768397033,
      "step": 20480
    },
    {
      "epoch": 0.9577386725786338,
      "grad_norm": 0.08807196468114853,
      "learning_rate": 4.62584445643166e-06,
      "loss": 0.039531927555799484,
      "step": 20736
    },
    {
      "epoch": 0.9695626068079997,
      "grad_norm": 0.026826005429029465,
      "learning_rate": 2.40331404948807e-06,
      "loss": 0.03951640799641609,
      "step": 20992
    },
    {
      "epoch": 0.9813865410373654,
      "grad_norm": 0.06786994636058807,
      "learning_rate": 9.009550772663965e-07,
      "loss": 0.03962159901857376,
      "step": 21248
    },
    {
      "epoch": 0.9932104752667313,
      "grad_norm": 0.05299762263894081,
      "learning_rate": 1.2094190315575791e-07,
      "loss": 0.03955233097076416,
      "step": 21504
    },
    {
      "epoch": 0.9932104752667313,
      "eval_cos_loss": 0.019904947206037773,
      "eval_loss": 0.03966651372101209,
      "eval_mse_loss": 0.03966651372101209,
      "flow/cos_sim": 0.9800950555496564,
      "flow/improvement_ratio": 0.9943993097026599,
      "flow/mag_ratio_mean": 0.9814930751443454,
      "flow/mag_ratio_std": 0.042273162626002204,
      "step": 21504
    },
    {
      "epoch": 0.9932104752667313,
      "eval_cos_loss": 0.019904947206037773,
      "eval_loss": 0.03966651372101209,
      "eval_mse_loss": 0.03966651372101209,
      "eval_runtime": 29.0571,
      "eval_samples_per_second": 963.378,
      "eval_steps_per_second": 15.074,
      "flow/cos_sim": 0.9800950555496564,
      "flow/improvement_ratio": 0.9943993097026599,
      "flow/mag_ratio_mean": 0.9814930751443454,
      "flow/mag_ratio_std": 0.042273162626002204,
      "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
}