{
  "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": 0.39305946230888367,
      "learning_rate": 0.000498046875,
      "loss": 1.780936598777771,
      "step": 256
    },
    {
      "epoch": 0.023647868458731697,
      "grad_norm": 0.4956166744232178,
      "learning_rate": 0.000998046875,
      "loss": 1.21787691116333,
      "step": 512
    },
    {
      "epoch": 0.03547180268809755,
      "grad_norm": 0.46151286363601685,
      "learning_rate": 0.000999640996023194,
      "loss": 1.1046085357666016,
      "step": 768
    },
    {
      "epoch": 0.047295736917463395,
      "grad_norm": 0.4608488082885742,
      "learning_rate": 0.0009985588674043958,
      "loss": 1.0648458003997803,
      "step": 1024
    },
    {
      "epoch": 0.047295736917463395,
      "eval_cos_loss": 0.3708630034096165,
      "eval_loss": 1.0313464179975258,
      "eval_mse_loss": 1.0313464179975258,
      "flow/cos_sim": 0.6291370091100806,
      "flow/improvement_ratio": 0.45305174915757895,
      "flow/mag_ratio_mean": 0.636674555197154,
      "flow/mag_ratio_std": 0.21184622356880745,
      "step": 1024
    },
    {
      "epoch": 0.047295736917463395,
      "eval_cos_loss": 0.3708630034096165,
      "eval_loss": 1.0313464179975258,
      "eval_mse_loss": 1.0313464179975258,
      "eval_runtime": 51.0215,
      "eval_samples_per_second": 548.652,
      "eval_steps_per_second": 8.585,
      "flow/cos_sim": 0.6291370091100806,
      "flow/improvement_ratio": 0.45305174915757895,
      "flow/mag_ratio_mean": 0.636674555197154,
      "flow/mag_ratio_std": 0.21184622356880745,
      "step": 1024
    },
    {
      "epoch": 0.05911967114682925,
      "grad_norm": 0.42826929688453674,
      "learning_rate": 0.0009967551747861387,
      "loss": 1.0448938608169556,
      "step": 1280
    },
    {
      "epoch": 0.0709436053761951,
      "grad_norm": 0.44176557660102844,
      "learning_rate": 0.000994232528651847,
      "loss": 1.0278342962265015,
      "step": 1536
    },
    {
      "epoch": 0.08276753960556095,
      "grad_norm": 0.4006131887435913,
      "learning_rate": 0.0009909945800260092,
      "loss": 1.0153039693832397,
      "step": 1792
    },
    {
      "epoch": 0.09459147383492679,
      "grad_norm": 0.5178146362304688,
      "learning_rate": 0.0009870460151900522,
      "loss": 1.0096927881240845,
      "step": 2048
    },
    {
      "epoch": 0.09459147383492679,
      "eval_cos_loss": 0.35073428046485605,
      "eval_loss": 0.9796474619543172,
      "eval_mse_loss": 0.9796474619543172,
      "flow/cos_sim": 0.6492657340280542,
      "flow/improvement_ratio": 0.45769495602067745,
      "flow/mag_ratio_mean": 0.6580993892395333,
      "flow/mag_ratio_std": 0.21975676535063138,
      "step": 2048
    },
    {
      "epoch": 0.09459147383492679,
      "eval_cos_loss": 0.35073428046485605,
      "eval_loss": 0.9796474619543172,
      "eval_mse_loss": 0.9796474619543172,
      "eval_runtime": 50.9087,
      "eval_samples_per_second": 549.867,
      "eval_steps_per_second": 8.604,
      "flow/cos_sim": 0.6492657340280542,
      "flow/improvement_ratio": 0.45769495602067745,
      "flow/mag_ratio_mean": 0.6580993892395333,
      "flow/mag_ratio_std": 0.21975676535063138,
      "step": 2048
    },
    {
      "epoch": 0.10641540806429264,
      "grad_norm": 0.4543623626232147,
      "learning_rate": 0.0009823925488998885,
      "loss": 1.0023009777069092,
      "step": 2304
    },
    {
      "epoch": 0.1182393422936585,
      "grad_norm": 0.41527411341667175,
      "learning_rate": 0.0009770409161149525,
      "loss": 0.9987744092941284,
      "step": 2560
    },
    {
      "epoch": 0.13006327652302435,
      "grad_norm": 0.37379226088523865,
      "learning_rate": 0.0009709988622506973,
      "loss": 0.9899244904518127,
      "step": 2816
    },
    {
      "epoch": 0.1418872107523902,
      "grad_norm": 0.4600990414619446,
      "learning_rate": 0.000964275131968659,
      "loss": 0.9836398959159851,
      "step": 3072
    },
    {
      "epoch": 0.1418872107523902,
      "eval_cos_loss": 0.33966573984383447,
      "eval_loss": 0.9576488193039481,
      "eval_mse_loss": 0.9576488193039481,
      "flow/cos_sim": 0.6603342864883545,
      "flow/improvement_ratio": 0.4705073194552774,
      "flow/mag_ratio_mean": 0.6616723446269014,
      "flow/mag_ratio_std": 0.2178211697976883,
      "step": 3072
    },
    {
      "epoch": 0.1418872107523902,
      "eval_cos_loss": 0.33966573984383447,
      "eval_loss": 0.9576488193039481,
      "eval_mse_loss": 0.9576488193039481,
      "eval_runtime": 50.7658,
      "eval_samples_per_second": 551.415,
      "eval_steps_per_second": 8.628,
      "flow/cos_sim": 0.6603342864883545,
      "flow/improvement_ratio": 0.4705073194552774,
      "flow/mag_ratio_mean": 0.6616723446269014,
      "flow/mag_ratio_std": 0.2178211697976883,
      "step": 3072
    },
    {
      "epoch": 0.15371114498175603,
      "grad_norm": 0.4591522812843323,
      "learning_rate": 0.0009568794565203123,
      "loss": 0.9778329133987427,
      "step": 3328
    },
    {
      "epoch": 0.1655350792111219,
      "grad_norm": 0.506796658039093,
      "learning_rate": 0.0009488225396630347,
      "loss": 0.9751444458961487,
      "step": 3584
    },
    {
      "epoch": 0.17735901344048774,
      "grad_norm": 0.6946936845779419,
      "learning_rate": 0.0009401160421685646,
      "loss": 0.9650535583496094,
      "step": 3840
    },
    {
      "epoch": 0.18918294766985358,
      "grad_norm": 0.5334352254867554,
      "learning_rate": 0.0009307725649463714,
      "loss": 0.968015193939209,
      "step": 4096
    },
    {
      "epoch": 0.18918294766985358,
      "eval_cos_loss": 0.3295682058350681,
      "eval_loss": 0.9358410420210939,
      "eval_mse_loss": 0.9358410420210939,
      "flow/cos_sim": 0.6704318162785273,
      "flow/improvement_ratio": 0.4610280344883601,
      "flow/mag_ratio_mean": 0.6802759532514773,
      "flow/mag_ratio_std": 0.21443206633198753,
      "step": 4096
    },
    {
      "epoch": 0.18918294766985358,
      "eval_cos_loss": 0.3295682058350681,
      "eval_loss": 0.9358410420210939,
      "eval_mse_loss": 0.9358410420210939,
      "eval_runtime": 50.6775,
      "eval_samples_per_second": 552.375,
      "eval_steps_per_second": 8.643,
      "flow/cos_sim": 0.6704318162785273,
      "flow/improvement_ratio": 0.4610280344883601,
      "flow/mag_ratio_mean": 0.6802759532514773,
      "flow/mag_ratio_std": 0.21443206633198753,
      "step": 4096
    },
    {
      "epoch": 0.20100688189921945,
      "grad_norm": 0.23779992759227753,
      "learning_rate": 0.0009208056308063659,
      "loss": 0.9618788957595825,
      "step": 4352
    },
    {
      "epoch": 0.2128308161285853,
      "grad_norm": 0.3806602358818054,
      "learning_rate": 0.0009102296648873445,
      "loss": 0.9570742249488831,
      "step": 4608
    },
    {
      "epoch": 0.22465475035795113,
      "grad_norm": 0.41401633620262146,
      "learning_rate": 0.0008990599737794927,
      "loss": 0.9550731778144836,
      "step": 4864
    },
    {
      "epoch": 0.236478684587317,
      "grad_norm": 0.4475092589855194,
      "learning_rate": 0.0008873127233711644,
      "loss": 0.9473839998245239,
      "step": 5120
    },
    {
      "epoch": 0.236478684587317,
      "eval_cos_loss": 0.31993362566107486,
      "eval_loss": 0.9154947232710172,
      "eval_mse_loss": 0.9154947232710172,
      "flow/cos_sim": 0.6800663863142876,
      "flow/improvement_ratio": 0.4631300759778175,
      "flow/mag_ratio_mean": 0.6912006213240427,
      "flow/mag_ratio_std": 0.2171582194155754,
      "step": 5120
    },
    {
      "epoch": 0.236478684587317,
      "eval_cos_loss": 0.31993362566107486,
      "eval_loss": 0.9154947232710172,
      "eval_mse_loss": 0.9154947232710172,
      "eval_runtime": 50.7141,
      "eval_samples_per_second": 551.977,
      "eval_steps_per_second": 8.637,
      "flow/cos_sim": 0.6800663863142876,
      "flow/improvement_ratio": 0.4631300759778175,
      "flow/mag_ratio_mean": 0.6912006213240427,
      "flow/mag_ratio_std": 0.2171582194155754,
      "step": 5120
    },
    {
      "epoch": 0.24830261881668284,
      "grad_norm": 0.39829879999160767,
      "learning_rate": 0.0008750049154520011,
      "loss": 0.9461303949356079,
      "step": 5376
    },
    {
      "epoch": 0.2601265530460487,
      "grad_norm": 0.570689857006073,
      "learning_rate": 0.0008621543631062487,
      "loss": 0.9459335803985596,
      "step": 5632
    },
    {
      "epoch": 0.27195048727541454,
      "grad_norm": 0.49673837423324585,
      "learning_rate": 0.0008487796649318904,
      "loss": 0.9424968957901001,
      "step": 5888
    },
    {
      "epoch": 0.2837744215047804,
      "grad_norm": 0.267270028591156,
      "learning_rate": 0.0008349001781229053,
      "loss": 0.9410290122032166,
      "step": 6144
    },
    {
      "epoch": 0.2837744215047804,
      "eval_cos_loss": 0.31771982879671334,
      "eval_loss": 0.9094011069160618,
      "eval_mse_loss": 0.9094011069160618,
      "flow/cos_sim": 0.682280182430189,
      "flow/improvement_ratio": 0.46761273847867363,
      "flow/mag_ratio_mean": 0.6880964620472634,
      "flow/mag_ratio_std": 0.21481055140359218,
      "step": 6144
    },
    {
      "epoch": 0.2837744215047804,
      "eval_cos_loss": 0.31771982879671334,
      "eval_loss": 0.9094011069160618,
      "eval_mse_loss": 0.9094011069160618,
      "eval_runtime": 51.0025,
      "eval_samples_per_second": 548.856,
      "eval_steps_per_second": 8.588,
      "flow/cos_sim": 0.682280182430189,
      "flow/improvement_ratio": 0.46761273847867363,
      "flow/mag_ratio_mean": 0.6880964620472634,
      "flow/mag_ratio_std": 0.21481055140359218,
      "step": 6144
    },
    {
      "epoch": 0.2955983557341462,
      "grad_norm": 0.43064209818840027,
      "learning_rate": 0.0008205359904536107,
      "loss": 0.9332870841026306,
      "step": 6400
    },
    {
      "epoch": 0.30742228996351206,
      "grad_norm": 0.45424163341522217,
      "learning_rate": 0.0008057078912056363,
      "loss": 0.9331082105636597,
      "step": 6656
    },
    {
      "epoch": 0.3192462241928779,
      "grad_norm": 0.6565593481063843,
      "learning_rate": 0.0007904373410796086,
      "loss": 0.9313054084777832,
      "step": 6912
    },
    {
      "epoch": 0.3310701584222438,
      "grad_norm": 0.5485164523124695,
      "learning_rate": 0.0007747464411350876,
      "loss": 0.9296150207519531,
      "step": 7168
    },
    {
      "epoch": 0.3310701584222438,
      "eval_cos_loss": 0.3111404624839896,
      "eval_loss": 0.8942771547189042,
      "eval_mse_loss": 0.8942771547189042,
      "flow/cos_sim": 0.688859552281088,
      "flow/improvement_ratio": 0.46996809215578317,
      "flow/mag_ratio_mean": 0.6797864257472835,
      "flow/mag_ratio_std": 0.21222736455125896,
      "step": 7168
    },
    {
      "epoch": 0.3310701584222438,
      "eval_cos_loss": 0.3111404624839896,
      "eval_loss": 0.8942771547189042,
      "eval_mse_loss": 0.8942771547189042,
      "eval_runtime": 51.2111,
      "eval_samples_per_second": 546.619,
      "eval_steps_per_second": 8.553,
      "flow/cos_sim": 0.688859552281088,
      "flow/improvement_ratio": 0.46996809215578317,
      "flow/mag_ratio_mean": 0.6797864257472835,
      "flow/mag_ratio_std": 0.21222736455125896,
      "step": 7168
    },
    {
      "epoch": 0.34289409265160964,
      "grad_norm": 0.3187243938446045,
      "learning_rate": 0.000758657900803716,
      "loss": 0.928747832775116,
      "step": 7424
    },
    {
      "epoch": 0.3547180268809755,
      "grad_norm": 0.3337271213531494,
      "learning_rate": 0.000742195005021869,
      "loss": 0.925912082195282,
      "step": 7680
    },
    {
      "epoch": 0.3665419611103413,
      "grad_norm": 0.5179343223571777,
      "learning_rate": 0.0007253815805303786,
      "loss": 0.923928439617157,
      "step": 7936
    },
    {
      "epoch": 0.37836589533970716,
      "grad_norm": 0.3594360053539276,
      "learning_rate": 0.000708309515673374,
      "loss": 0.9226717948913574,
      "step": 8192
    },
    {
      "epoch": 0.37836589533970716,
      "eval_cos_loss": 0.3082785013059503,
      "eval_loss": 0.8866745081122063,
      "eval_mse_loss": 0.8866745081122063,
      "flow/cos_sim": 0.6917215108054958,
      "flow/improvement_ratio": 0.46622252083260174,
      "flow/mag_ratio_mean": 0.6977443612057325,
      "flow/mag_ratio_std": 0.21176587402412336,
      "step": 8192
    },
    {
      "epoch": 0.37836589533970716,
      "eval_cos_loss": 0.3082785013059503,
      "eval_loss": 0.8866745081122063,
      "eval_mse_loss": 0.8866745081122063,
      "eval_runtime": 50.453,
      "eval_samples_per_second": 554.833,
      "eval_steps_per_second": 8.681,
      "flow/cos_sim": 0.6917215108054958,
      "flow/improvement_ratio": 0.46622252083260174,
      "flow/mag_ratio_mean": 0.6977443612057325,
      "flow/mag_ratio_std": 0.21176587402412336,
      "step": 8192
    },
    {
      "epoch": 0.390189829569073,
      "grad_norm": 0.2854546010494232,
      "learning_rate": 0.0006908696365085842,
      "loss": 0.9181273579597473,
      "step": 8448
    },
    {
      "epoch": 0.4020137637984389,
      "grad_norm": 0.6743373870849609,
      "learning_rate": 0.0006731535118143318,
      "loss": 0.9195088148117065,
      "step": 8704
    },
    {
      "epoch": 0.41383769802780473,
      "grad_norm": 0.31884175539016724,
      "learning_rate": 0.0006551867821290267,
      "loss": 0.9166384935379028,
      "step": 8960
    },
    {
      "epoch": 0.4256616322571706,
      "grad_norm": 0.377502977848053,
      "learning_rate": 0.0006369954506915572,
      "loss": 0.9148516654968262,
      "step": 9216
    },
    {
      "epoch": 0.4256616322571706,
      "eval_cos_loss": 0.3045589508803468,
      "eval_loss": 0.8784834154664654,
      "eval_mse_loss": 0.8784834154664654,
      "flow/cos_sim": 0.6954410576929241,
      "flow/improvement_ratio": 0.4660896160150772,
      "flow/mag_ratio_mean": 0.6896047796288581,
      "flow/mag_ratio_std": 0.20802477359363478,
      "step": 9216
    },
    {
      "epoch": 0.4256616322571706,
      "eval_cos_loss": 0.3045589508803468,
      "eval_loss": 0.8784834154664654,
      "eval_mse_loss": 0.8784834154664654,
      "eval_runtime": 50.5191,
      "eval_samples_per_second": 554.108,
      "eval_steps_per_second": 8.67,
      "flow/cos_sim": 0.6954410576929241,
      "flow/improvement_ratio": 0.4660896160150772,
      "flow/mag_ratio_mean": 0.6896047796288581,
      "flow/mag_ratio_std": 0.20802477359363478,
      "step": 9216
    },
    {
      "epoch": 0.4374855664865364,
      "grad_norm": 0.2935723066329956,
      "learning_rate": 0.0006186058458068149,
      "loss": 0.9158189296722412,
      "step": 9472
    },
    {
      "epoch": 0.44930950071590225,
      "grad_norm": 0.6002705097198486,
      "learning_rate": 0.0006000445827407526,
      "loss": 0.9137871265411377,
      "step": 9728
    },
    {
      "epoch": 0.4611334349452681,
      "grad_norm": 0.3263113498687744,
      "learning_rate": 0.0005813385252001232,
      "loss": 0.9110487103462219,
      "step": 9984
    },
    {
      "epoch": 0.472957369174634,
      "grad_norm": 0.3971521854400635,
      "learning_rate": 0.0005625884706462481,
      "loss": 0.9095574021339417,
      "step": 10240
    },
    {
      "epoch": 0.472957369174634,
      "eval_cos_loss": 0.30255618235564125,
      "eval_loss": 0.8742554381013461,
      "eval_mse_loss": 0.8742554381013461,
      "flow/cos_sim": 0.6974438190732373,
      "flow/improvement_ratio": 0.46927081490760525,
      "flow/mag_ratio_mean": 0.6958274860360306,
      "flow/mag_ratio_std": 0.21019614664795191,
      "step": 10240
    },
    {
      "epoch": 0.472957369174634,
      "eval_cos_loss": 0.30255618235564125,
      "eval_loss": 0.8742554381013461,
      "eval_mse_loss": 0.8742554381013461,
      "eval_runtime": 51.3022,
      "eval_samples_per_second": 545.649,
      "eval_steps_per_second": 8.538,
      "flow/cos_sim": 0.6974438190732373,
      "flow/improvement_ratio": 0.46927081490760525,
      "flow/mag_ratio_mean": 0.6958274860360306,
      "flow/mag_ratio_std": 0.21019614664795191,
      "step": 10240
    },
    {
      "epoch": 0.48478130340399983,
      "grad_norm": 0.3551290035247803,
      "learning_rate": 0.0005436745145807214,
      "loss": 0.9060850143432617,
      "step": 10496
    },
    {
      "epoch": 0.49660523763336567,
      "grad_norm": 0.4094627797603607,
      "learning_rate": 0.0005246973484120841,
      "loss": 0.9049650430679321,
      "step": 10752
    },
    {
      "epoch": 0.5084291718627315,
      "grad_norm": 0.4399157762527466,
      "learning_rate": 0.0005056844377834413,
      "loss": 0.9042121767997742,
      "step": 11008
    },
    {
      "epoch": 0.5202531060920974,
      "grad_norm": 0.5329852104187012,
      "learning_rate": 0.0004866633000708374,
      "loss": 0.9038464426994324,
      "step": 11264
    },
    {
      "epoch": 0.5202531060920974,
      "eval_cos_loss": 0.3004189523183592,
      "eval_loss": 0.8695836310789465,
      "eval_mse_loss": 0.8695836310789465,
      "flow/cos_sim": 0.699581062548781,
      "flow/improvement_ratio": 0.47203471392529195,
      "flow/mag_ratio_mean": 0.6982838295093955,
      "flow/mag_ratio_std": 0.20910314640634137,
      "step": 11264
    },
    {
      "epoch": 0.5202531060920974,
      "eval_cos_loss": 0.3004189523183592,
      "eval_loss": 0.8695836310789465,
      "eval_mse_loss": 0.8695836310789465,
      "eval_runtime": 51.0932,
      "eval_samples_per_second": 547.881,
      "eval_steps_per_second": 8.573,
      "flow/cos_sim": 0.699581062548781,
      "flow/improvement_ratio": 0.47203471392529195,
      "flow/mag_ratio_mean": 0.6982838295093955,
      "flow/mag_ratio_std": 0.20910314640634137,
      "step": 11264
    },
    {
      "epoch": 0.5320770403214632,
      "grad_norm": 0.3345041275024414,
      "learning_rate": 0.00046766146455737116,
      "loss": 0.9016320109367371,
      "step": 11520
    },
    {
      "epoch": 0.5439009745508291,
      "grad_norm": 0.47475534677505493,
      "learning_rate": 0.00044870643259007823,
      "loss": 0.9061211347579956,
      "step": 11776
    },
    {
      "epoch": 0.5557249087801949,
      "grad_norm": 0.6051743030548096,
      "learning_rate": 0.000429825637777245,
      "loss": 0.903529167175293,
      "step": 12032
    },
    {
      "epoch": 0.5675488430095608,
      "grad_norm": 0.39291107654571533,
      "learning_rate": 0.0004111195298279063,
      "loss": 0.8996114730834961,
      "step": 12288
    },
    {
      "epoch": 0.5675488430095608,
      "eval_cos_loss": 0.29944087231540245,
      "eval_loss": 0.8666894671307307,
      "eval_mse_loss": 0.8666894671307307,
      "flow/cos_sim": 0.700559140476462,
      "flow/improvement_ratio": 0.4712014450058001,
      "flow/mag_ratio_mean": 0.6962402411247497,
      "flow/mag_ratio_std": 0.20598308850913288,
      "step": 12288
    },
    {
      "epoch": 0.5675488430095608,
      "eval_cos_loss": 0.29944087231540245,
      "eval_loss": 0.8666894671307307,
      "eval_mse_loss": 0.8666894671307307,
      "eval_runtime": 50.9516,
      "eval_samples_per_second": 549.403,
      "eval_steps_per_second": 8.596,
      "flow/cos_sim": 0.700559140476462,
      "flow/improvement_ratio": 0.4712014450058001,
      "flow/mag_ratio_mean": 0.6962402411247497,
      "flow/mag_ratio_std": 0.20598308850913288,
      "step": 12288
    },
    {
      "epoch": 0.5793727772389267,
      "grad_norm": 0.38517266511917114,
      "learning_rate": 0.0003924684852772684,
      "loss": 0.9009869694709778,
      "step": 12544
    },
    {
      "epoch": 0.5911967114682924,
      "grad_norm": 0.3089086413383484,
      "learning_rate": 0.00037404498123644504,
      "loss": 0.9021767377853394,
      "step": 12800
    },
    {
      "epoch": 0.6030206456976583,
      "grad_norm": 0.5309678912162781,
      "learning_rate": 0.0003557312014609784,
      "loss": 0.9018325209617615,
      "step": 13056
    },
    {
      "epoch": 0.6148445799270241,
      "grad_norm": 0.36101001501083374,
      "learning_rate": 0.00033762622184548686,
      "loss": 0.8969551920890808,
      "step": 13312
    },
    {
      "epoch": 0.6148445799270241,
      "eval_cos_loss": 0.30058121572346447,
      "eval_loss": 0.8693417448159222,
      "eval_mse_loss": 0.8693417448159222,
      "flow/cos_sim": 0.6994188143510253,
      "flow/improvement_ratio": 0.4627683243658989,
      "flow/mag_ratio_mean": 0.7132113126859273,
      "flow/mag_ratio_std": 0.21025482180728217,
      "step": 13312
    },
    {
      "epoch": 0.6148445799270241,
      "eval_cos_loss": 0.30058121572346447,
      "eval_loss": 0.8693417448159222,
      "eval_mse_loss": 0.8693417448159222,
      "eval_runtime": 50.5293,
      "eval_samples_per_second": 553.996,
      "eval_steps_per_second": 8.668,
      "flow/cos_sim": 0.6994188143510253,
      "flow/improvement_ratio": 0.4627683243658989,
      "flow/mag_ratio_mean": 0.7132113126859273,
      "flow/mag_ratio_std": 0.21025482180728217,
      "step": 13312
    },
    {
      "epoch": 0.62666851415639,
      "grad_norm": 0.4100685119628906,
      "learning_rate": 0.00031975624571791317,
      "loss": 0.8950372934341431,
      "step": 13568
    },
    {
      "epoch": 0.6384924483857558,
      "grad_norm": 0.2832315266132355,
      "learning_rate": 0.00030214713628576984,
      "loss": 0.8969645500183105,
      "step": 13824
    },
    {
      "epoch": 0.6503163826151217,
      "grad_norm": 0.39231061935424805,
      "learning_rate": 0.00028482437920433144,
      "loss": 0.8969537019729614,
      "step": 14080
    },
    {
      "epoch": 0.6621403168444876,
      "grad_norm": 0.7917575240135193,
      "learning_rate": 0.00026781304569125866,
      "loss": 0.8955793976783752,
      "step": 14336
    },
    {
      "epoch": 0.6621403168444876,
      "eval_cos_loss": 0.29705161533263175,
      "eval_loss": 0.8606252675731433,
      "eval_mse_loss": 0.8606252675731433,
      "flow/cos_sim": 0.7029484053180642,
      "flow/improvement_ratio": 0.46662241846459096,
      "flow/mag_ratio_mean": 0.7076406654429762,
      "flow/mag_ratio_std": 0.20891345561095023,
      "step": 14336
    },
    {
      "epoch": 0.6621403168444876,
      "eval_cos_loss": 0.29705161533263175,
      "eval_loss": 0.8606252675731433,
      "eval_mse_loss": 0.8606252675731433,
      "eval_runtime": 50.7312,
      "eval_samples_per_second": 551.791,
      "eval_steps_per_second": 8.634,
      "flow/cos_sim": 0.7029484053180642,
      "flow/improvement_ratio": 0.46662241846459096,
      "flow/mag_ratio_mean": 0.7076406654429762,
      "flow/mag_ratio_std": 0.20891345561095023,
      "step": 14336
    },
    {
      "epoch": 0.6739642510738534,
      "grad_norm": 0.2073170691728592,
      "learning_rate": 0.0002511377562410373,
      "loss": 0.8957792520523071,
      "step": 14592
    },
    {
      "epoch": 0.6857881853032193,
      "grad_norm": 0.6534483432769775,
      "learning_rate": 0.00023482264499174915,
      "loss": 0.8970192074775696,
      "step": 14848
    },
    {
      "epoch": 0.6976121195325851,
      "grad_norm": 0.24640800058841705,
      "learning_rate": 0.00021889132479574586,
      "loss": 0.8956112861633301,
      "step": 15104
    },
    {
      "epoch": 0.709436053761951,
      "grad_norm": 0.35241010785102844,
      "learning_rate": 0.0002033668530447801,
      "loss": 0.8931179642677307,
      "step": 15360
    },
    {
      "epoch": 0.709436053761951,
      "eval_cos_loss": 0.3005856423889665,
      "eval_loss": 0.8699537216256198,
      "eval_mse_loss": 0.8699537216256198,
      "flow/cos_sim": 0.6994143777514157,
      "flow/improvement_ratio": 0.47257714858066,
      "flow/mag_ratio_mean": 0.7026257012804894,
      "flow/mag_ratio_std": 0.20547213071965736,
      "step": 15360
    },
    {
      "epoch": 0.709436053761951,
      "eval_cos_loss": 0.3005856423889665,
      "eval_loss": 0.8699537216256198,
      "eval_mse_loss": 0.8699537216256198,
      "eval_runtime": 50.6626,
      "eval_samples_per_second": 552.538,
      "eval_steps_per_second": 8.645,
      "flow/cos_sim": 0.6994143777514157,
      "flow/improvement_ratio": 0.47257714858066,
      "flow/mag_ratio_mean": 0.7026257012804894,
      "flow/mag_ratio_std": 0.20547213071965736,
      "step": 15360
    },
    {
      "epoch": 0.7212599879913169,
      "grad_norm": 0.4916954040527344,
      "learning_rate": 0.0001882716982990524,
      "loss": 0.8969737887382507,
      "step": 15616
    },
    {
      "epoch": 0.7330839222206826,
      "grad_norm": 0.5466639399528503,
      "learning_rate": 0.00017362770776847765,
      "loss": 0.8959704637527466,
      "step": 15872
    },
    {
      "epoch": 0.7449078564500485,
      "grad_norm": 0.45104148983955383,
      "learning_rate": 0.00015951048788154866,
      "loss": 0.8920841813087463,
      "step": 16128
    },
    {
      "epoch": 0.7567317906794143,
      "grad_norm": 0.23488114774227142,
      "learning_rate": 0.0001458297605985633,
      "loss": 0.8922313451766968,
      "step": 16384
    },
    {
      "epoch": 0.7567317906794143,
      "eval_cos_loss": 0.2997721447928311,
      "eval_loss": 0.8672782906658574,
      "eval_mse_loss": 0.8672782906658574,
      "flow/cos_sim": 0.7002278570442984,
      "flow/improvement_ratio": 0.4662386941719273,
      "flow/mag_ratio_mean": 0.7100178600718442,
      "flow/mag_ratio_std": 0.20784400301436856,
      "step": 16384
    },
    {
      "epoch": 0.7567317906794143,
      "eval_cos_loss": 0.2997721447928311,
      "eval_loss": 0.8672782906658574,
      "eval_mse_loss": 0.8672782906658574,
      "eval_runtime": 50.9632,
      "eval_samples_per_second": 549.278,
      "eval_steps_per_second": 8.594,
      "flow/cos_sim": 0.7002278570442984,
      "flow/improvement_ratio": 0.4662386941719273,
      "flow/mag_ratio_mean": 0.7100178600718442,
      "flow/mag_ratio_std": 0.20784400301436856,
      "step": 16384
    },
    {
      "epoch": 0.7685557249087802,
      "grad_norm": 0.25124895572662354,
      "learning_rate": 0.00013266162372593872,
      "loss": 0.8903353810310364,
      "step": 16640
    },
    {
      "epoch": 0.780379659138146,
      "grad_norm": 0.2108362913131714,
      "learning_rate": 0.00012002513550158511,
      "loss": 0.8916131854057312,
      "step": 16896
    },
    {
      "epoch": 0.7922035933675119,
      "grad_norm": 0.22269435226917267,
      "learning_rate": 0.00010793858470858986,
      "loss": 0.8932616114616394,
      "step": 17152
    },
    {
      "epoch": 0.8040275275968778,
      "grad_norm": 0.2587853968143463,
      "learning_rate": 9.641946420587128e-05,
      "loss": 0.8924716114997864,
      "step": 17408
    },
    {
      "epoch": 0.8040275275968778,
      "eval_cos_loss": 0.29718101851335943,
      "eval_loss": 0.8611624605579463,
      "eval_mse_loss": 0.8611624605579463,
      "flow/cos_sim": 0.7028190088054361,
      "flow/improvement_ratio": 0.465975399158861,
      "flow/mag_ratio_mean": 0.7077710173990084,
      "flow/mag_ratio_std": 0.20818141616506664,
      "step": 17408
    },
    {
      "epoch": 0.8040275275968778,
      "eval_cos_loss": 0.29718101851335943,
      "eval_loss": 0.8611624605579463,
      "eval_mse_loss": 0.8611624605579463,
      "eval_runtime": 50.5279,
      "eval_samples_per_second": 554.011,
      "eval_steps_per_second": 8.668,
      "flow/cos_sim": 0.7028190088054361,
      "flow/improvement_ratio": 0.465975399158861,
      "flow/mag_ratio_mean": 0.7077710173990084,
      "flow/mag_ratio_std": 0.20818141616506664,
      "step": 17408
    },
    {
      "epoch": 0.8158514618262436,
      "grad_norm": 0.22211408615112305,
      "learning_rate": 8.548444561077174e-05,
      "loss": 0.8929438591003418,
      "step": 17664
    },
    {
      "epoch": 0.8276753960556095,
      "grad_norm": 0.16869360208511353,
      "learning_rate": 7.514935517023558e-05,
      "loss": 0.8896693587303162,
      "step": 17920
    },
    {
      "epoch": 0.8394993302849753,
      "grad_norm": 0.2596553862094879,
      "learning_rate": 6.542915085548828e-05,
      "loss": 0.8902697563171387,
      "step": 18176
    },
    {
      "epoch": 0.8513232645143411,
      "grad_norm": 0.23781056702136993,
      "learning_rate": 5.6337900713373745e-05,
      "loss": 0.8896790742874146,
      "step": 18432
    },
    {
      "epoch": 0.8513232645143411,
      "eval_cos_loss": 0.29792979842723777,
      "eval_loss": 0.8626222945239446,
      "eval_mse_loss": 0.8626222945239446,
      "flow/cos_sim": 0.7020702216451027,
      "flow/improvement_ratio": 0.4633185115034722,
      "flow/mag_ratio_mean": 0.7091497124330094,
      "flow/mag_ratio_std": 0.20735928929967967,
      "step": 18432
    },
    {
      "epoch": 0.8513232645143411,
      "eval_cos_loss": 0.29792979842723777,
      "eval_loss": 0.8626222945239446,
      "eval_mse_loss": 0.8626222945239446,
      "eval_runtime": 50.6008,
      "eval_samples_per_second": 553.212,
      "eval_steps_per_second": 8.656,
      "flow/cos_sim": 0.7020702216451027,
      "flow/improvement_ratio": 0.4633185115034722,
      "flow/mag_ratio_mean": 0.7091497124330094,
      "flow/mag_ratio_std": 0.20735928929967967,
      "step": 18432
    },
    {
      "epoch": 0.8631471987437069,
      "grad_norm": 0.24920311570167542,
      "learning_rate": 4.78887625056757e-05,
      "loss": 0.8902055621147156,
      "step": 18688
    },
    {
      "epoch": 0.8749711329730728,
      "grad_norm": 0.29849973320961,
      "learning_rate": 4.009396466589682e-05,
      "loss": 0.8859462141990662,
      "step": 18944
    },
    {
      "epoch": 0.8867950672024387,
      "grad_norm": 0.2714053690433502,
      "learning_rate": 3.296478860105229e-05,
      "loss": 0.890446662902832,
      "step": 19200
    },
    {
      "epoch": 0.8986190014318045,
      "grad_norm": 0.3177722096443176,
      "learning_rate": 2.6511552364095358e-05,
      "loss": 0.8924580216407776,
      "step": 19456
    },
    {
      "epoch": 0.8986190014318045,
      "eval_cos_loss": 0.2985236220504051,
      "eval_loss": 0.8639824096470663,
      "eval_mse_loss": 0.8639824096470663,
      "flow/cos_sim": 0.7014764100993605,
      "flow/improvement_ratio": 0.46989300261893774,
      "flow/mag_ratio_mean": 0.7079281760677355,
      "flow/mag_ratio_std": 0.20886976322899126,
      "step": 19456
    },
    {
      "epoch": 0.8986190014318045,
      "eval_cos_loss": 0.2985236220504051,
      "eval_loss": 0.8639824096470663,
      "eval_mse_loss": 0.8639824096470663,
      "eval_runtime": 51.0434,
      "eval_samples_per_second": 548.416,
      "eval_steps_per_second": 8.581,
      "flow/cos_sim": 0.7014764100993605,
      "flow/improvement_ratio": 0.46989300261893774,
      "flow/mag_ratio_mean": 0.7079281760677355,
      "flow/mag_ratio_std": 0.20886976322899126,
      "step": 19456
    },
    {
      "epoch": 0.9104429356611704,
      "grad_norm": 0.2206149697303772,
      "learning_rate": 2.074359572060447e-05,
      "loss": 0.888733446598053,
      "step": 19712
    },
    {
      "epoch": 0.9222668698905362,
      "grad_norm": 0.1809428334236145,
      "learning_rate": 1.5669266631345104e-05,
      "loss": 0.8881487250328064,
      "step": 19968
    },
    {
      "epoch": 0.9340908041199021,
      "grad_norm": 0.16622242331504822,
      "learning_rate": 1.1311620335770879e-05,
      "loss": 0.8930599689483643,
      "step": 20224
    },
    {
      "epoch": 0.945914738349268,
      "grad_norm": 0.25432291626930237,
      "learning_rate": 7.642790173984836e-06,
      "loss": 0.8947161436080933,
      "step": 20480
    },
    {
      "epoch": 0.945914738349268,
      "eval_cos_loss": 0.2954880066862389,
      "eval_loss": 0.8570421781714104,
      "eval_mse_loss": 0.8570421781714104,
      "flow/cos_sim": 0.7045120175026324,
      "flow/improvement_ratio": 0.4633371506107452,
      "flow/mag_ratio_mean": 0.7112098344384807,
      "flow/mag_ratio_std": 0.20906852869546577,
      "step": 20480
    },
    {
      "epoch": 0.945914738349268,
      "eval_cos_loss": 0.2954880066862389,
      "eval_loss": 0.8570421781714104,
      "eval_mse_loss": 0.8570421781714104,
      "eval_runtime": 50.6225,
      "eval_samples_per_second": 552.976,
      "eval_steps_per_second": 8.652,
      "flow/cos_sim": 0.7045120175026324,
      "flow/improvement_ratio": 0.4633371506107452,
      "flow/mag_ratio_mean": 0.7112098344384807,
      "flow/mag_ratio_std": 0.20906852869546577,
      "step": 20480
    },
    {
      "epoch": 0.9577386725786338,
      "grad_norm": 0.21038714051246643,
      "learning_rate": 4.686548355746001e-06,
      "loss": 0.8879244923591614,
      "step": 20736
    },
    {
      "epoch": 0.9695626068079997,
      "grad_norm": 0.24928613007068634,
      "learning_rate": 2.447173448359541e-06,
      "loss": 0.889561653137207,
      "step": 20992
    },
    {
      "epoch": 0.9813865410373654,
      "grad_norm": 0.21087926626205444,
      "learning_rate": 9.279064980089457e-07,
      "loss": 0.8902660608291626,
      "step": 21248
    },
    {
      "epoch": 0.9932104752667313,
      "grad_norm": 0.18456599116325378,
      "learning_rate": 1.3094633899163322e-07,
      "loss": 0.8902599811553955,
      "step": 21504
    },
    {
      "epoch": 0.9932104752667313,
      "eval_cos_loss": 0.29621158380350565,
      "eval_loss": 0.858656403544831,
      "eval_mse_loss": 0.858656403544831,
      "flow/cos_sim": 0.7037884391606126,
      "flow/improvement_ratio": 0.47204427734085413,
      "flow/mag_ratio_mean": 0.7089065290205011,
      "flow/mag_ratio_std": 0.20877777332584607,
      "step": 21504
    },
    {
      "epoch": 0.9932104752667313,
      "eval_cos_loss": 0.29621158380350565,
      "eval_loss": 0.858656403544831,
      "eval_mse_loss": 0.858656403544831,
      "eval_runtime": 50.8696,
      "eval_samples_per_second": 550.289,
      "eval_steps_per_second": 8.61,
      "flow/cos_sim": 0.7037884391606126,
      "flow/improvement_ratio": 0.47204427734085413,
      "flow/mag_ratio_mean": 0.7089065290205011,
      "flow/mag_ratio_std": 0.20877777332584607,
      "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
}