{
  "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": 0.1137767806649208,
      "learning_rate": 0.000498046875,
      "loss": 1.0040063858032227,
      "step": 256
    },
    {
      "epoch": 0.023647868458731697,
      "grad_norm": 5.475095272064209,
      "learning_rate": 0.000998046875,
      "loss": 0.7889968156814575,
      "step": 512
    },
    {
      "epoch": 0.03547180268809755,
      "grad_norm": 4.206150054931641,
      "learning_rate": 0.000999640996023194,
      "loss": 0.6388198137283325,
      "step": 768
    },
    {
      "epoch": 0.047295736917463395,
      "grad_norm": 4.658252716064453,
      "learning_rate": 0.0009985588674043958,
      "loss": 0.5990524888038635,
      "step": 1024
    },
    {
      "epoch": 0.047295736917463395,
      "eval_cos_loss": 0.3963755411372337,
      "eval_loss": 0.5656519263438439,
      "eval_mse_loss": 0.5656519263438439,
      "flow/cos_sim": 0.6036244705659614,
      "flow/improvement_ratio": 0.5193254046788499,
      "flow/mag_ratio_mean": 0.6099911789371543,
      "flow/mag_ratio_std": 0.2817785246935609,
      "step": 1024
    },
    {
      "epoch": 0.047295736917463395,
      "eval_cos_loss": 0.3963755411372337,
      "eval_loss": 0.5656519263438439,
      "eval_mse_loss": 0.5656519263438439,
      "eval_runtime": 37.956,
      "eval_samples_per_second": 737.513,
      "eval_steps_per_second": 11.54,
      "flow/cos_sim": 0.6036244705659614,
      "flow/improvement_ratio": 0.5193254046788499,
      "flow/mag_ratio_mean": 0.6099911789371543,
      "flow/mag_ratio_std": 0.2817785246935609,
      "step": 1024
    },
    {
      "epoch": 0.05911967114682925,
      "grad_norm": 0.6539129018783569,
      "learning_rate": 0.0009967551747861387,
      "loss": 0.5766449570655823,
      "step": 1280
    },
    {
      "epoch": 0.0709436053761951,
      "grad_norm": 0.4175584018230438,
      "learning_rate": 0.000994232528651847,
      "loss": 0.5719335675239563,
      "step": 1536
    },
    {
      "epoch": 0.08276753960556095,
      "grad_norm": 2.3471314907073975,
      "learning_rate": 0.0009909945800260092,
      "loss": 0.5546385645866394,
      "step": 1792
    },
    {
      "epoch": 0.09459147383492679,
      "grad_norm": 1.1768289804458618,
      "learning_rate": 0.0009870460151900522,
      "loss": 0.551981508731842,
      "step": 2048
    },
    {
      "epoch": 0.09459147383492679,
      "eval_cos_loss": 0.3284394587964228,
      "eval_loss": 0.5063902307046603,
      "eval_mse_loss": 0.5063902307046603,
      "flow/cos_sim": 0.6715605630450052,
      "flow/improvement_ratio": 0.4615051872245797,
      "flow/mag_ratio_mean": 0.618984971144428,
      "flow/mag_ratio_std": 0.2694709628956503,
      "step": 2048
    },
    {
      "epoch": 0.09459147383492679,
      "eval_cos_loss": 0.3284394587964228,
      "eval_loss": 0.5063902307046603,
      "eval_mse_loss": 0.5063902307046603,
      "eval_runtime": 37.859,
      "eval_samples_per_second": 739.401,
      "eval_steps_per_second": 11.569,
      "flow/cos_sim": 0.6715605630450052,
      "flow/improvement_ratio": 0.4615051872245797,
      "flow/mag_ratio_mean": 0.618984971144428,
      "flow/mag_ratio_std": 0.2694709628956503,
      "step": 2048
    },
    {
      "epoch": 0.10641540806429264,
      "grad_norm": 2.072291851043701,
      "learning_rate": 0.0009823925488998885,
      "loss": 0.5476918816566467,
      "step": 2304
    },
    {
      "epoch": 0.1182393422936585,
      "grad_norm": 0.6736424565315247,
      "learning_rate": 0.0009770409161149525,
      "loss": 0.5403198599815369,
      "step": 2560
    },
    {
      "epoch": 0.13006327652302435,
      "grad_norm": 4.842870235443115,
      "learning_rate": 0.0009709988622506973,
      "loss": 0.5355626940727234,
      "step": 2816
    },
    {
      "epoch": 0.1418872107523902,
      "grad_norm": 1.0661463737487793,
      "learning_rate": 0.000964275131968659,
      "loss": 0.5361711382865906,
      "step": 3072
    },
    {
      "epoch": 0.1418872107523902,
      "eval_cos_loss": 0.36135857718992453,
      "eval_loss": 0.5245981899570656,
      "eval_mse_loss": 0.5245981899570656,
      "flow/cos_sim": 0.638641438391655,
      "flow/improvement_ratio": 0.526839620718673,
      "flow/mag_ratio_mean": 0.6446732967139379,
      "flow/mag_ratio_std": 0.2603021453232526,
      "step": 3072
    },
    {
      "epoch": 0.1418872107523902,
      "eval_cos_loss": 0.36135857718992453,
      "eval_loss": 0.5245981899570656,
      "eval_mse_loss": 0.5245981899570656,
      "eval_runtime": 37.0701,
      "eval_samples_per_second": 755.137,
      "eval_steps_per_second": 11.815,
      "flow/cos_sim": 0.638641438391655,
      "flow/improvement_ratio": 0.526839620718673,
      "flow/mag_ratio_mean": 0.6446732967139379,
      "flow/mag_ratio_std": 0.2603021453232526,
      "step": 3072
    },
    {
      "epoch": 0.15371114498175603,
      "grad_norm": 1.6855143308639526,
      "learning_rate": 0.0009568794565203123,
      "loss": 0.5352926850318909,
      "step": 3328
    },
    {
      "epoch": 0.1655350792111219,
      "grad_norm": 3.7140069007873535,
      "learning_rate": 0.0009488225396630347,
      "loss": 0.5290294885635376,
      "step": 3584
    },
    {
      "epoch": 0.17735901344048774,
      "grad_norm": 4.1050310134887695,
      "learning_rate": 0.0009401160421685646,
      "loss": 0.5295140743255615,
      "step": 3840
    },
    {
      "epoch": 0.18918294766985358,
      "grad_norm": 5.94942569732666,
      "learning_rate": 0.0009307725649463714,
      "loss": 0.5271365642547607,
      "step": 4096
    },
    {
      "epoch": 0.18918294766985358,
      "eval_cos_loss": 0.31836620156895623,
      "eval_loss": 0.4890012886017969,
      "eval_mse_loss": 0.4890012886017969,
      "flow/cos_sim": 0.6816338111548663,
      "flow/improvement_ratio": 0.48559793387646,
      "flow/mag_ratio_mean": 0.6403457065151162,
      "flow/mag_ratio_std": 0.26559364904551747,
      "step": 4096
    },
    {
      "epoch": 0.18918294766985358,
      "eval_cos_loss": 0.31836620156895623,
      "eval_loss": 0.4890012886017969,
      "eval_mse_loss": 0.4890012886017969,
      "eval_runtime": 37.2681,
      "eval_samples_per_second": 751.126,
      "eval_steps_per_second": 11.753,
      "flow/cos_sim": 0.6816338111548663,
      "flow/improvement_ratio": 0.48559793387646,
      "flow/mag_ratio_mean": 0.6403457065151162,
      "flow/mag_ratio_std": 0.26559364904551747,
      "step": 4096
    },
    {
      "epoch": 0.20100688189921945,
      "grad_norm": 2.1582860946655273,
      "learning_rate": 0.0009208056308063659,
      "loss": 0.5225115418434143,
      "step": 4352
    },
    {
      "epoch": 0.2128308161285853,
      "grad_norm": 2.84121036529541,
      "learning_rate": 0.0009102296648873445,
      "loss": 0.5245823860168457,
      "step": 4608
    },
    {
      "epoch": 0.22465475035795113,
      "grad_norm": 2.509031057357788,
      "learning_rate": 0.0008990599737794927,
      "loss": 0.5176577568054199,
      "step": 4864
    },
    {
      "epoch": 0.236478684587317,
      "grad_norm": 1.5172899961471558,
      "learning_rate": 0.0008873127233711644,
      "loss": 0.5137428641319275,
      "step": 5120
    },
    {
      "epoch": 0.236478684587317,
      "eval_cos_loss": 0.31018479253467357,
      "eval_loss": 0.4786865841172057,
      "eval_mse_loss": 0.4786865841172057,
      "flow/cos_sim": 0.6898152210396719,
      "flow/improvement_ratio": 0.4824195757304152,
      "flow/mag_ratio_mean": 0.6519407997392628,
      "flow/mag_ratio_std": 0.2654882789202477,
      "step": 5120
    },
    {
      "epoch": 0.236478684587317,
      "eval_cos_loss": 0.31018479253467357,
      "eval_loss": 0.4786865841172057,
      "eval_mse_loss": 0.4786865841172057,
      "eval_runtime": 37.4766,
      "eval_samples_per_second": 746.946,
      "eval_steps_per_second": 11.687,
      "flow/cos_sim": 0.6898152210396719,
      "flow/improvement_ratio": 0.4824195757304152,
      "flow/mag_ratio_mean": 0.6519407997392628,
      "flow/mag_ratio_std": 0.2654882789202477,
      "step": 5120
    },
    {
      "epoch": 0.24830261881668284,
      "grad_norm": 2.9429166316986084,
      "learning_rate": 0.0008750049154520011,
      "loss": 0.5114176869392395,
      "step": 5376
    },
    {
      "epoch": 0.2601265530460487,
      "grad_norm": 2.600057363510132,
      "learning_rate": 0.0008621543631062487,
      "loss": 0.5167359113693237,
      "step": 5632
    },
    {
      "epoch": 0.27195048727541454,
      "grad_norm": 0.4475926458835602,
      "learning_rate": 0.0008487796649318904,
      "loss": 0.5100853443145752,
      "step": 5888
    },
    {
      "epoch": 0.2837744215047804,
      "grad_norm": 0.7397291660308838,
      "learning_rate": 0.0008349001781229053,
      "loss": 0.5120869278907776,
      "step": 6144
    },
    {
      "epoch": 0.2837744215047804,
      "eval_cos_loss": 0.29584084784603554,
      "eval_loss": 0.46408402525126663,
      "eval_mse_loss": 0.46408402525126663,
      "flow/cos_sim": 0.7041591786902789,
      "flow/improvement_ratio": 0.47311876884334164,
      "flow/mag_ratio_mean": 0.6741217052283353,
      "flow/mag_ratio_std": 0.2542141846734095,
      "step": 6144
    },
    {
      "epoch": 0.2837744215047804,
      "eval_cos_loss": 0.29584084784603554,
      "eval_loss": 0.46408402525126663,
      "eval_mse_loss": 0.46408402525126663,
      "eval_runtime": 36.91,
      "eval_samples_per_second": 758.413,
      "eval_steps_per_second": 11.867,
      "flow/cos_sim": 0.7041591786902789,
      "flow/improvement_ratio": 0.47311876884334164,
      "flow/mag_ratio_mean": 0.6741217052283353,
      "flow/mag_ratio_std": 0.2542141846734095,
      "step": 6144
    },
    {
      "epoch": 0.2955983557341462,
      "grad_norm": 0.4129859507083893,
      "learning_rate": 0.0008205359904536107,
      "loss": 0.5060973763465881,
      "step": 6400
    },
    {
      "epoch": 0.30742228996351206,
      "grad_norm": 0.29374027252197266,
      "learning_rate": 0.0008057078912056363,
      "loss": 0.5066944360733032,
      "step": 6656
    },
    {
      "epoch": 0.3192462241928779,
      "grad_norm": 1.1076785326004028,
      "learning_rate": 0.0007904373410796086,
      "loss": 0.5057422518730164,
      "step": 6912
    },
    {
      "epoch": 0.3310701584222438,
      "grad_norm": 0.5511188507080078,
      "learning_rate": 0.0007747464411350876,
      "loss": 0.5001094341278076,
      "step": 7168
    },
    {
      "epoch": 0.3310701584222438,
      "eval_cos_loss": 0.30180720311321624,
      "eval_loss": 0.4666885096052466,
      "eval_mse_loss": 0.4666885096052466,
      "flow/cos_sim": 0.6981928144415764,
      "flow/improvement_ratio": 0.49193395667424483,
      "flow/mag_ratio_mean": 0.6697109201455225,
      "flow/mag_ratio_std": 0.25267424516073644,
      "step": 7168
    },
    {
      "epoch": 0.3310701584222438,
      "eval_cos_loss": 0.30180720311321624,
      "eval_loss": 0.4666885096052466,
      "eval_mse_loss": 0.4666885096052466,
      "eval_runtime": 37.002,
      "eval_samples_per_second": 756.528,
      "eval_steps_per_second": 11.837,
      "flow/cos_sim": 0.6981928144415764,
      "flow/improvement_ratio": 0.49193395667424483,
      "flow/mag_ratio_mean": 0.6697109201455225,
      "flow/mag_ratio_std": 0.25267424516073644,
      "step": 7168
    },
    {
      "epoch": 0.34289409265160964,
      "grad_norm": 1.092437505722046,
      "learning_rate": 0.000758657900803716,
      "loss": 0.5043932199478149,
      "step": 7424
    },
    {
      "epoch": 0.3547180268809755,
      "grad_norm": 1.3343017101287842,
      "learning_rate": 0.000742195005021869,
      "loss": 0.5025789737701416,
      "step": 7680
    },
    {
      "epoch": 0.3665419611103413,
      "grad_norm": 0.6918906569480896,
      "learning_rate": 0.0007253815805303786,
      "loss": 0.49820616841316223,
      "step": 7936
    },
    {
      "epoch": 0.37836589533970716,
      "grad_norm": 4.359390735626221,
      "learning_rate": 0.0007082419613901028,
      "loss": 0.5000379681587219,
      "step": 8192
    },
    {
      "epoch": 0.37836589533970716,
      "eval_cos_loss": 0.31474396868791754,
      "eval_loss": 0.47536985759865746,
      "eval_mse_loss": 0.47536985759865746,
      "flow/cos_sim": 0.6852560571339577,
      "flow/improvement_ratio": 0.5048932789667556,
      "flow/mag_ratio_mean": 0.6644382656437077,
      "flow/mag_ratio_std": 0.25931949406590094,
      "step": 8192
    },
    {
      "epoch": 0.37836589533970716,
      "eval_cos_loss": 0.31474396868791754,
      "eval_loss": 0.47536985759865746,
      "eval_mse_loss": 0.47536985759865746,
      "eval_runtime": 37.0849,
      "eval_samples_per_second": 754.836,
      "eval_steps_per_second": 11.811,
      "flow/cos_sim": 0.6852560571339577,
      "flow/improvement_ratio": 0.5048932789667556,
      "flow/mag_ratio_mean": 0.6644382656437077,
      "flow/mag_ratio_std": 0.25931949406590094,
      "step": 8192
    },
    {
      "epoch": 0.390189829569073,
      "grad_norm": 1.0232681035995483,
      "learning_rate": 0.0006908009537632514,
      "loss": 0.5013407468795776,
      "step": 8448
    },
    {
      "epoch": 0.4020137637984389,
      "grad_norm": 0.7591996192932129,
      "learning_rate": 0.0006730838000114403,
      "loss": 0.4965692162513733,
      "step": 8704
    },
    {
      "epoch": 0.41383769802780473,
      "grad_norm": 1.4552847146987915,
      "learning_rate": 0.0006551161421624341,
      "loss": 0.49521973729133606,
      "step": 8960
    },
    {
      "epoch": 0.4256616322571706,
      "grad_norm": 2.3110909461975098,
      "learning_rate": 0.0006369239847984517,
      "loss": 0.49350598454475403,
      "step": 9216
    },
    {
      "epoch": 0.4256616322571706,
      "eval_cos_loss": 0.28651542172312194,
      "eval_loss": 0.4529013752801233,
      "eval_mse_loss": 0.4529013752801233,
      "flow/cos_sim": 0.7134845855573541,
      "flow/improvement_ratio": 0.4720227978953488,
      "flow/mag_ratio_mean": 0.6730685518484681,
      "flow/mag_ratio_std": 0.24056239671086613,
      "step": 9216
    },
    {
      "epoch": 0.4256616322571706,
      "eval_cos_loss": 0.28651542172312194,
      "eval_loss": 0.4529013752801233,
      "eval_mse_loss": 0.4529013752801233,
      "eval_runtime": 36.9629,
      "eval_samples_per_second": 757.326,
      "eval_steps_per_second": 11.85,
      "flow/cos_sim": 0.7134845855573541,
      "flow/improvement_ratio": 0.4720227978953488,
      "flow/mag_ratio_mean": 0.6730685518484681,
      "flow/mag_ratio_std": 0.24056239671086613,
      "step": 9216
    },
    {
      "epoch": 0.4374855664865364,
      "grad_norm": 0.4148007333278656,
      "learning_rate": 0.0006185336574197479,
      "loss": 0.49597373604774475,
      "step": 9472
    },
    {
      "epoch": 0.44930950071590225,
      "grad_norm": 0.5046154260635376,
      "learning_rate": 0.0005999717763379407,
      "loss": 0.49263298511505127,
      "step": 9728
    },
    {
      "epoch": 0.4611334349452681,
      "grad_norm": 1.6516804695129395,
      "learning_rate": 0.0005812652061542363,
      "loss": 0.4927175045013428,
      "step": 9984
    },
    {
      "epoch": 0.472957369174634,
      "grad_norm": 2.369363784790039,
      "learning_rate": 0.0005624410208783071,
      "loss": 0.49063339829444885,
      "step": 10240
    },
    {
      "epoch": 0.472957369174634,
      "eval_cos_loss": 0.32282865789110804,
      "eval_loss": 0.4822733311756561,
      "eval_mse_loss": 0.4822733311756561,
      "flow/cos_sim": 0.6771713574183041,
      "flow/improvement_ratio": 0.5181982032784588,
      "flow/mag_ratio_mean": 0.6684839627514146,
      "flow/mag_ratio_std": 0.2625512615458606,
      "step": 10240
    },
    {
      "epoch": 0.472957369174634,
      "eval_cos_loss": 0.32282865789110804,
      "eval_loss": 0.4822733311756561,
      "eval_mse_loss": 0.4822733311756561,
      "eval_runtime": 37.1141,
      "eval_samples_per_second": 754.242,
      "eval_steps_per_second": 11.801,
      "flow/cos_sim": 0.6771713574183041,
      "flow/improvement_ratio": 0.5181982032784588,
      "flow/mag_ratio_mean": 0.6684839627514146,
      "flow/mag_ratio_std": 0.2625512615458606,
      "step": 10240
    },
    {
      "epoch": 0.48478130340399983,
      "grad_norm": 0.24072349071502686,
      "learning_rate": 0.0005435264647440881,
      "loss": 0.4890187382698059,
      "step": 10496
    },
    {
      "epoch": 0.49660523763336567,
      "grad_norm": 5.027256965637207,
      "learning_rate": 0.000524548912779213,
      "loss": 0.48836761713027954,
      "step": 10752
    },
    {
      "epoch": 0.5084291718627315,
      "grad_norm": 0.5343558192253113,
      "learning_rate": 0.0005055358311851499,
      "loss": 0.4868377149105072,
      "step": 11008
    },
    {
      "epoch": 0.5202531060920974,
      "grad_norm": 2.5563480854034424,
      "learning_rate": 0.0004865147375853812,
      "loss": 0.4865650236606598,
      "step": 11264
    },
    {
      "epoch": 0.5202531060920974,
      "eval_cos_loss": 0.2832905833171383,
      "eval_loss": 0.4457947185866909,
      "eval_mse_loss": 0.4457947185866909,
      "flow/cos_sim": 0.7167094211055808,
      "flow/improvement_ratio": 0.48482539637448036,
      "flow/mag_ratio_mean": 0.6898882619049995,
      "flow/mag_ratio_std": 0.2549947457702737,
      "step": 11264
    },
    {
      "epoch": 0.5202531060920974,
      "eval_cos_loss": 0.2832905833171383,
      "eval_loss": 0.4457947185866909,
      "eval_mse_loss": 0.4457947185866909,
      "eval_runtime": 37.7857,
      "eval_samples_per_second": 740.835,
      "eval_steps_per_second": 11.592,
      "flow/cos_sim": 0.7167094211055808,
      "flow/improvement_ratio": 0.48482539637448036,
      "flow/mag_ratio_mean": 0.6898882619049995,
      "flow/mag_ratio_std": 0.2549947457702737,
      "step": 11264
    },
    {
      "epoch": 0.5320770403214632,
      "grad_norm": 0.5450732707977295,
      "learning_rate": 0.0004675131611991607,
      "loss": 0.4828563332557678,
      "step": 11520
    },
    {
      "epoch": 0.5439009745508291,
      "grad_norm": 2.4773805141448975,
      "learning_rate": 0.0004485586029984899,
      "loss": 0.4860672950744629,
      "step": 11776
    },
    {
      "epoch": 0.5557249087801949,
      "grad_norm": 3.7647409439086914,
      "learning_rate": 0.00042967849590597266,
      "loss": 0.48244670033454895,
      "step": 12032
    },
    {
      "epoch": 0.5675488430095608,
      "grad_norm": 2.507927656173706,
      "learning_rate": 0.0004109001650911621,
      "loss": 0.48438385128974915,
      "step": 12288
    },
    {
      "epoch": 0.5675488430095608,
      "eval_cos_loss": 0.2910717453246247,
      "eval_loss": 0.45217968108447176,
      "eval_mse_loss": 0.45217968108447176,
      "flow/cos_sim": 0.7089282686579718,
      "flow/improvement_ratio": 0.49513497451941174,
      "flow/mag_ratio_mean": 0.6819437842935188,
      "flow/mag_ratio_std": 0.2516717043233244,
      "step": 12288
    },
    {
      "epoch": 0.5675488430095608,
      "eval_cos_loss": 0.2910717453246247,
      "eval_loss": 0.45217968108447176,
      "eval_mse_loss": 0.45217968108447176,
      "eval_runtime": 37.0211,
      "eval_samples_per_second": 756.135,
      "eval_steps_per_second": 11.831,
      "flow/cos_sim": 0.7089282686579718,
      "flow/improvement_ratio": 0.49513497451941174,
      "flow/mag_ratio_mean": 0.6819437842935188,
      "flow/mag_ratio_std": 0.2516717043233244,
      "step": 12288
    },
    {
      "epoch": 0.5793727772389267,
      "grad_norm": 1.5222080945968628,
      "learning_rate": 0.0003922507884228551,
      "loss": 0.48588937520980835,
      "step": 12544
    },
    {
      "epoch": 0.5911967114682924,
      "grad_norm": 0.5074782967567444,
      "learning_rate": 0.00037375735713457723,
      "loss": 0.48188111186027527,
      "step": 12800
    },
    {
      "epoch": 0.6030206456976583,
      "grad_norm": 1.3766505718231201,
      "learning_rate": 0.00035544663676018276,
      "loss": 0.48240211606025696,
      "step": 13056
    },
    {
      "epoch": 0.6148445799270241,
      "grad_norm": 1.290895938873291,
      "learning_rate": 0.00033734512839611255,
      "loss": 0.4810515344142914,
      "step": 13312
    },
    {
      "epoch": 0.6148445799270241,
      "eval_cos_loss": 0.2804271564499973,
      "eval_loss": 0.4426349750528597,
      "eval_mse_loss": 0.4426349750528597,
      "flow/cos_sim": 0.7195728606285026,
      "flow/improvement_ratio": 0.48104941613597957,
      "flow/mag_ratio_mean": 0.6869833815043376,
      "flow/mag_ratio_std": 0.25096172298470587,
      "step": 13312
    },
    {
      "epoch": 0.6148445799270241,
      "eval_cos_loss": 0.2804271564499973,
      "eval_loss": 0.4426349750528597,
      "eval_mse_loss": 0.4426349750528597,
      "eval_runtime": 37.3386,
      "eval_samples_per_second": 749.706,
      "eval_steps_per_second": 11.73,
      "flow/cos_sim": 0.7195728606285026,
      "flow/improvement_ratio": 0.48104941613597957,
      "flow/mag_ratio_mean": 0.6869833815043376,
      "flow/mag_ratio_std": 0.25096172298470587,
      "step": 13312
    },
    {
      "epoch": 0.62666851415639,
      "grad_norm": 0.6742284297943115,
      "learning_rate": 0.0003194790303463687,
      "loss": 0.4774361848831177,
      "step": 13568
    },
    {
      "epoch": 0.6384924483857558,
      "grad_norm": 1.8099576234817505,
      "learning_rate": 0.00030187420020572406,
      "loss": 0.48075973987579346,
      "step": 13824
    },
    {
      "epoch": 0.6503163826151217,
      "grad_norm": 2.060770034790039,
      "learning_rate": 0.00028455611743603626,
      "loss": 0.4758567214012146,
      "step": 14080
    },
    {
      "epoch": 0.6621403168444876,
      "grad_norm": 4.473404884338379,
      "learning_rate": 0.0002675498464898373,
      "loss": 0.47404515743255615,
      "step": 14336
    },
    {
      "epoch": 0.6621403168444876,
      "eval_cos_loss": 0.28930224678968186,
      "eval_loss": 0.44853738038779395,
      "eval_mse_loss": 0.44853738038779395,
      "flow/cos_sim": 0.710697771887801,
      "flow/improvement_ratio": 0.49212116812734297,
      "flow/mag_ratio_mean": 0.6857217257697833,
      "flow/mag_ratio_std": 0.25427674481721774,
      "step": 14336
    },
    {
      "epoch": 0.6621403168444876,
      "eval_cos_loss": 0.28930224678968186,
      "eval_loss": 0.44853738038779395,
      "eval_mse_loss": 0.44853738038779395,
      "eval_runtime": 36.974,
      "eval_samples_per_second": 757.099,
      "eval_steps_per_second": 11.846,
      "flow/cos_sim": 0.710697771887801,
      "flow/improvement_ratio": 0.49212116812734297,
      "flow/mag_ratio_mean": 0.6857217257697833,
      "flow/mag_ratio_std": 0.25427674481721774,
      "step": 14336
    },
    {
      "epoch": 0.6739642510738534,
      "grad_norm": 0.17484501004219055,
      "learning_rate": 0.0002508800005345623,
      "loss": 0.47856900095939636,
      "step": 14592
    },
    {
      "epoch": 0.6857881853032193,
      "grad_norm": 1.5286951065063477,
      "learning_rate": 0.00023457070582992562,
      "loss": 0.47592371702194214,
      "step": 14848
    },
    {
      "epoch": 0.6976121195325851,
      "grad_norm": 3.2949068546295166,
      "learning_rate": 0.00021864556680999692,
      "loss": 0.4775611162185669,
      "step": 15104
    },
    {
      "epoch": 0.709436053761951,
      "grad_norm": 0.25448572635650635,
      "learning_rate": 0.0002031276319205152,
      "loss": 0.47541379928588867,
      "step": 15360
    },
    {
      "epoch": 0.709436053761951,
      "eval_cos_loss": 0.2735697959792124,
      "eval_loss": 0.4344245920034304,
      "eval_mse_loss": 0.4344245920034304,
      "flow/cos_sim": 0.7264302035444947,
      "flow/improvement_ratio": 0.4827755716157286,
      "flow/mag_ratio_mean": 0.6983764671027388,
      "flow/mag_ratio_std": 0.24496402853427957,
      "step": 15360
    },
    {
      "epoch": 0.709436053761951,
      "eval_cos_loss": 0.2735697959792124,
      "eval_loss": 0.4344245920034304,
      "eval_mse_loss": 0.4344245920034304,
      "eval_runtime": 37.1576,
      "eval_samples_per_second": 753.358,
      "eval_steps_per_second": 11.788,
      "flow/cos_sim": 0.7264302035444947,
      "flow/improvement_ratio": 0.4827755716157286,
      "flow/mag_ratio_mean": 0.6983764671027388,
      "flow/mag_ratio_std": 0.24496402853427957,
      "step": 15360
    },
    {
      "epoch": 0.7212599879913169,
      "grad_norm": 0.7958180904388428,
      "learning_rate": 0.00018803936026088542,
      "loss": 0.47444477677345276,
      "step": 15616
    },
    {
      "epoch": 0.7330839222206826,
      "grad_norm": 2.6243398189544678,
      "learning_rate": 0.00017340258907913464,
      "loss": 0.4730243980884552,
      "step": 15872
    },
    {
      "epoch": 0.7449078564500485,
      "grad_norm": 0.6033198833465576,
      "learning_rate": 0.0001592385021668743,
      "loss": 0.4736136496067047,
      "step": 16128
    },
    {
      "epoch": 0.7567317906794143,
      "grad_norm": 0.8605419993400574,
      "learning_rate": 0.0001455675992000087,
      "loss": 0.4723820984363556,
      "step": 16384
    },
    {
      "epoch": 0.7567317906794143,
      "eval_cos_loss": 0.2758947659505012,
      "eval_loss": 0.4379414171660872,
      "eval_mse_loss": 0.4379414171660872,
      "flow/cos_sim": 0.7241052517063542,
      "flow/improvement_ratio": 0.47650186486167995,
      "flow/mag_ratio_mean": 0.6886805119035451,
      "flow/mag_ratio_std": 0.25369694692903455,
      "step": 16384
    },
    {
      "epoch": 0.7567317906794143,
      "eval_cos_loss": 0.2758947659505012,
      "eval_loss": 0.4379414171660872,
      "eval_mse_loss": 0.4379414171660872,
      "eval_runtime": 37.1262,
      "eval_samples_per_second": 753.995,
      "eval_steps_per_second": 11.798,
      "flow/cos_sim": 0.7241052517063542,
      "flow/improvement_ratio": 0.47650186486167995,
      "flow/mag_ratio_mean": 0.6886805119035451,
      "flow/mag_ratio_std": 0.25369694692903455,
      "step": 16384
    },
    {
      "epoch": 0.7685557249087802,
      "grad_norm": 0.1524878591299057,
      "learning_rate": 0.000132409666069565,
      "loss": 0.4706062972545624,
      "step": 16640
    },
    {
      "epoch": 0.780379659138146,
      "grad_norm": 0.3120117783546448,
      "learning_rate": 0.0001197837462455823,
      "loss": 0.47130054235458374,
      "step": 16896
    },
    {
      "epoch": 0.7922035933675119,
      "grad_norm": 1.8674224615097046,
      "learning_rate": 0.00010770811321550749,
      "loss": 0.4707598090171814,
      "step": 17152
    },
    {
      "epoch": 0.8040275275968778,
      "grad_norm": 1.0960615873336792,
      "learning_rate": 9.620024403698591e-05,
      "loss": 0.472503662109375,
      "step": 17408
    },
    {
      "epoch": 0.8040275275968778,
      "eval_cos_loss": 0.2721319894676339,
      "eval_loss": 0.4329949348620628,
      "eval_mse_loss": 0.4329949348620628,
      "flow/cos_sim": 0.7278679995776312,
      "flow/improvement_ratio": 0.4802333515256507,
      "flow/mag_ratio_mean": 0.6976211946032363,
      "flow/mag_ratio_std": 0.24612450511216027,
      "step": 17408
    },
    {
      "epoch": 0.8040275275968778,
      "eval_cos_loss": 0.2721319894676339,
      "eval_loss": 0.4329949348620628,
      "eval_mse_loss": 0.4329949348620628,
      "eval_runtime": 37.0362,
      "eval_samples_per_second": 755.828,
      "eval_steps_per_second": 11.826,
      "flow/cos_sim": 0.7278679995776312,
      "flow/improvement_ratio": 0.4802333515256507,
      "flow/mag_ratio_mean": 0.6976211946032363,
      "flow/mag_ratio_std": 0.24612450511216027,
      "step": 17408
    },
    {
      "epoch": 0.8158514618262436,
      "grad_norm": 0.16665872931480408,
      "learning_rate": 8.527679404332429e-05,
      "loss": 0.4697099030017853,
      "step": 17664
    },
    {
      "epoch": 0.8276753960556095,
      "grad_norm": 1.170737862586975,
      "learning_rate": 7.495357273823544e-05,
      "loss": 0.4676206409931183,
      "step": 17920
    },
    {
      "epoch": 0.8394993302849753,
      "grad_norm": 2.653127670288086,
      "learning_rate": 6.524552091475183e-05,
      "loss": 0.46873748302459717,
      "step": 18176
    },
    {
      "epoch": 0.8513232645143411,
      "grad_norm": 0.2692463994026184,
      "learning_rate": 5.6166689031422024e-05,
      "loss": 0.4690783619880676,
      "step": 18432
    },
    {
      "epoch": 0.8513232645143411,
      "eval_cos_loss": 0.2771332638516818,
      "eval_loss": 0.4356902488698698,
      "eval_mse_loss": 0.4356902488698698,
      "flow/cos_sim": 0.7228667485659526,
      "flow/improvement_ratio": 0.48473947968112824,
      "flow/mag_ratio_mean": 0.7041790871859686,
      "flow/mag_ratio_std": 0.24285391320104469,
      "step": 18432
    },
    {
      "epoch": 0.8513232645143411,
      "eval_cos_loss": 0.2771332638516818,
      "eval_loss": 0.4356902488698698,
      "eval_mse_loss": 0.4356902488698698,
      "eval_runtime": 37.2383,
      "eval_samples_per_second": 751.727,
      "eval_steps_per_second": 11.762,
      "flow/cos_sim": 0.7228667485659526,
      "flow/improvement_ratio": 0.48473947968112824,
      "flow/mag_ratio_mean": 0.7041790871859686,
      "flow/mag_ratio_std": 0.24285391320104469,
      "step": 18432
    },
    {
      "epoch": 0.8631471987437069,
      "grad_norm": 0.7530527114868164,
      "learning_rate": 4.773021687709067e-05,
      "loss": 0.4693593978881836,
      "step": 18688
    },
    {
      "epoch": 0.8749711329730728,
      "grad_norm": 3.6022565364837646,
      "learning_rate": 3.994831455368719e-05,
      "loss": 0.4689854681491852,
      "step": 18944
    },
    {
      "epoch": 0.8867950672024387,
      "grad_norm": 0.15695664286613464,
      "learning_rate": 3.283224480455282e-05,
      "loss": 0.4696367084980011,
      "step": 19200
    },
    {
      "epoch": 0.8986190014318045,
      "grad_norm": 0.37616676092147827,
      "learning_rate": 2.639230671387627e-05,
      "loss": 0.46513649821281433,
      "step": 19456
    },
    {
      "epoch": 0.8986190014318045,
      "eval_cos_loss": 0.27949059135429394,
      "eval_loss": 0.43712959850215477,
      "eval_mse_loss": 0.43712959850215477,
      "flow/cos_sim": 0.7205094044611334,
      "flow/improvement_ratio": 0.49371096900064654,
      "flow/mag_ratio_mean": 0.7003211331421926,
      "flow/mag_ratio_std": 0.2446217237202,
      "step": 19456
    },
    {
      "epoch": 0.8986190014318045,
      "eval_cos_loss": 0.27949059135429394,
      "eval_loss": 0.43712959850215477,
      "eval_mse_loss": 0.43712959850215477,
      "eval_runtime": 38.2808,
      "eval_samples_per_second": 731.255,
      "eval_steps_per_second": 11.442,
      "flow/cos_sim": 0.7205094044611334,
      "flow/improvement_ratio": 0.49371096900064654,
      "flow/mag_ratio_mean": 0.7003211331421926,
      "flow/mag_ratio_std": 0.2446217237202,
      "step": 19456
    },
    {
      "epoch": 0.9104429356611704,
      "grad_norm": 0.8353758454322815,
      "learning_rate": 2.063782080083576e-05,
      "loss": 0.4667496979236603,
      "step": 19712
    },
    {
      "epoch": 0.9222668698905362,
      "grad_norm": 1.2311574220657349,
      "learning_rate": 1.557711553001523e-05,
      "loss": 0.46688342094421387,
      "step": 19968
    },
    {
      "epoch": 0.9340908041199021,
      "grad_norm": 0.6781339645385742,
      "learning_rate": 1.1217515257622269e-05,
      "loss": 0.46820777654647827,
      "step": 20224
    },
    {
      "epoch": 0.945914738349268,
      "grad_norm": 1.2881850004196167,
      "learning_rate": 7.565329630950746e-06,
      "loss": 0.4700950086116791,
      "step": 20480
    },
    {
      "epoch": 0.945914738349268,
      "eval_cos_loss": 0.2731730710698045,
      "eval_loss": 0.4320683794342764,
      "eval_mse_loss": 0.4320683794342764,
      "flow/cos_sim": 0.7268269342374584,
      "flow/improvement_ratio": 0.4821734848223865,
      "flow/mag_ratio_mean": 0.7012590013410402,
      "flow/mag_ratio_std": 0.24438341768228844,
      "step": 20480
    },
    {
      "epoch": 0.945914738349268,
      "eval_cos_loss": 0.2731730710698045,
      "eval_loss": 0.4320683794342764,
      "eval_mse_loss": 0.4320683794342764,
      "eval_runtime": 37.2037,
      "eval_samples_per_second": 752.425,
      "eval_steps_per_second": 11.773,
      "flow/cos_sim": 0.7268269342374584,
      "flow/improvement_ratio": 0.4821734848223865,
      "flow/mag_ratio_mean": 0.7012590013410402,
      "flow/mag_ratio_std": 0.24438341768228844,
      "step": 20480
    },
    {
      "epoch": 0.9577386725786338,
      "grad_norm": 0.4809494614601135,
      "learning_rate": 4.62584445643166e-06,
      "loss": 0.4665885865688324,
      "step": 20736
    },
    {
      "epoch": 0.9695626068079997,
      "grad_norm": 0.3120735287666321,
      "learning_rate": 2.40331404948807e-06,
      "loss": 0.46703049540519714,
      "step": 20992
    },
    {
      "epoch": 0.9813865410373654,
      "grad_norm": 0.4376528561115265,
      "learning_rate": 9.009550772663965e-07,
      "loss": 0.46802768111228943,
      "step": 21248
    },
    {
      "epoch": 0.9932104752667313,
      "grad_norm": 1.0960811376571655,
      "learning_rate": 1.2094190315575791e-07,
      "loss": 0.468431293964386,
      "step": 21504
    },
    {
      "epoch": 0.9932104752667313,
      "eval_cos_loss": 0.2702979859832215,
      "eval_loss": 0.4292263891462866,
      "eval_mse_loss": 0.4292263891462866,
      "flow/cos_sim": 0.7297020165343263,
      "flow/improvement_ratio": 0.4884998352015944,
      "flow/mag_ratio_mean": 0.7036541160111014,
      "flow/mag_ratio_std": 0.24063446641512656,
      "step": 21504
    },
    {
      "epoch": 0.9932104752667313,
      "eval_cos_loss": 0.2702979859832215,
      "eval_loss": 0.4292263891462866,
      "eval_mse_loss": 0.4292263891462866,
      "eval_runtime": 37.5499,
      "eval_samples_per_second": 745.488,
      "eval_steps_per_second": 11.664,
      "flow/cos_sim": 0.7297020165343263,
      "flow/improvement_ratio": 0.4884998352015944,
      "flow/mag_ratio_mean": 0.7036541160111014,
      "flow/mag_ratio_std": 0.24063446641512656,
      "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
}