{
  "best_global_step": null,
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 1.999950290798827,
  "eval_steps": 500,
  "global_step": 20117,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.000994184023462743,
      "grad_norm": 1.0385397672653198,
      "learning_rate": 1.8e-05,
      "loss": 0.604,
      "memory/device_reserved (GiB)": 21.62,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 10,
      "tokens_per_second_per_gpu": 347.46
    },
    {
      "epoch": 0.001988368046925486,
      "grad_norm": 0.556696891784668,
      "learning_rate": 3.8e-05,
      "loss": 0.4244,
      "memory/device_reserved (GiB)": 21.62,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 20,
      "tokens_per_second_per_gpu": 318.15
    },
    {
      "epoch": 0.002982552070388229,
      "grad_norm": 0.24665255844593048,
      "learning_rate": 5.8e-05,
      "loss": 0.3883,
      "memory/device_reserved (GiB)": 21.62,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 30,
      "tokens_per_second_per_gpu": 330.79
    },
    {
      "epoch": 0.003976736093850972,
      "grad_norm": 0.3350813090801239,
      "learning_rate": 7.800000000000001e-05,
      "loss": 0.4163,
      "memory/device_reserved (GiB)": 22.38,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 40,
      "tokens_per_second_per_gpu": 396.14
    },
    {
      "epoch": 0.0049709201173137145,
      "grad_norm": 0.42506587505340576,
      "learning_rate": 9.8e-05,
      "loss": 0.3811,
      "memory/device_reserved (GiB)": 22.42,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 50,
      "tokens_per_second_per_gpu": 402.54
    },
    {
      "epoch": 0.005965104140776458,
      "grad_norm": 0.5153183937072754,
      "learning_rate": 0.000118,
      "loss": 0.418,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 60,
      "tokens_per_second_per_gpu": 330.44
    },
    {
      "epoch": 0.0069592881642392005,
      "grad_norm": 0.3010534644126892,
      "learning_rate": 0.000138,
      "loss": 0.3671,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 70,
      "tokens_per_second_per_gpu": 344.47
    },
    {
      "epoch": 0.007953472187701944,
      "grad_norm": 0.46113327145576477,
      "learning_rate": 0.00015800000000000002,
      "loss": 0.3387,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 80,
      "tokens_per_second_per_gpu": 372.27
    },
    {
      "epoch": 0.008947656211164686,
      "grad_norm": 0.4268002212047577,
      "learning_rate": 0.00017800000000000002,
      "loss": 0.2999,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 90,
      "tokens_per_second_per_gpu": 344.49
    },
    {
      "epoch": 0.009941840234627429,
      "grad_norm": 0.5650917291641235,
      "learning_rate": 0.00019800000000000002,
      "loss": 0.3356,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 100,
      "tokens_per_second_per_gpu": 335.57
    },
    {
      "epoch": 0.010936024258090173,
      "grad_norm": 0.2521424889564514,
      "learning_rate": 0.00019999990023993625,
      "loss": 0.3025,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 110,
      "tokens_per_second_per_gpu": 307.78
    },
    {
      "epoch": 0.011930208281552916,
      "grad_norm": 0.34742406010627747,
      "learning_rate": 0.00019999955539058868,
      "loss": 0.351,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 120,
      "tokens_per_second_per_gpu": 355.93
    },
    {
      "epoch": 0.012924392305015658,
      "grad_norm": 0.2816642224788666,
      "learning_rate": 0.00019999896422120075,
      "loss": 0.4031,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 130,
      "tokens_per_second_per_gpu": 392.82
    },
    {
      "epoch": 0.013918576328478401,
      "grad_norm": 0.41705670952796936,
      "learning_rate": 0.0001999981267332287,
      "loss": 0.3481,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 140,
      "tokens_per_second_per_gpu": 404.67
    },
    {
      "epoch": 0.014912760351941143,
      "grad_norm": 0.5290879011154175,
      "learning_rate": 0.00019999704292873545,
      "loss": 0.3784,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 150,
      "tokens_per_second_per_gpu": 409.59
    },
    {
      "epoch": 0.015906944375403888,
      "grad_norm": 0.2704632878303528,
      "learning_rate": 0.0001999957128103906,
      "loss": 0.2029,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 160,
      "tokens_per_second_per_gpu": 309.25
    },
    {
      "epoch": 0.01690112839886663,
      "grad_norm": 0.3863286077976227,
      "learning_rate": 0.00019999413638147049,
      "loss": 0.3084,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 170,
      "tokens_per_second_per_gpu": 395.0
    },
    {
      "epoch": 0.017895312422329373,
      "grad_norm": 0.32178717851638794,
      "learning_rate": 0.00019999231364585827,
      "loss": 0.2713,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 180,
      "tokens_per_second_per_gpu": 343.31
    },
    {
      "epoch": 0.018889496445792117,
      "grad_norm": 0.3010699450969696,
      "learning_rate": 0.00019999024460804366,
      "loss": 0.353,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 190,
      "tokens_per_second_per_gpu": 353.97
    },
    {
      "epoch": 0.019883680469254858,
      "grad_norm": 0.2678498327732086,
      "learning_rate": 0.00019998792927312315,
      "loss": 0.1904,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 200,
      "tokens_per_second_per_gpu": 352.72
    },
    {
      "epoch": 0.020877864492717602,
      "grad_norm": 0.25298821926116943,
      "learning_rate": 0.00019998536764679993,
      "loss": 0.2397,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 210,
      "tokens_per_second_per_gpu": 336.24
    },
    {
      "epoch": 0.021872048516180347,
      "grad_norm": 0.4027327001094818,
      "learning_rate": 0.0001999825597353838,
      "loss": 0.3111,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 220,
      "tokens_per_second_per_gpu": 388.05
    },
    {
      "epoch": 0.022866232539643087,
      "grad_norm": 0.4060591757297516,
      "learning_rate": 0.00019997950554579124,
      "loss": 0.2578,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 230,
      "tokens_per_second_per_gpu": 354.57
    },
    {
      "epoch": 0.02386041656310583,
      "grad_norm": 0.29408156871795654,
      "learning_rate": 0.00019997620508554537,
      "loss": 0.2952,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 240,
      "tokens_per_second_per_gpu": 334.85
    },
    {
      "epoch": 0.024854600586568572,
      "grad_norm": 0.381528377532959,
      "learning_rate": 0.00019997265836277595,
      "loss": 0.2397,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 250,
      "tokens_per_second_per_gpu": 402.73
    },
    {
      "epoch": 0.025848784610031317,
      "grad_norm": 0.30223792791366577,
      "learning_rate": 0.00019996886538621925,
      "loss": 0.4017,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 260,
      "tokens_per_second_per_gpu": 430.9
    },
    {
      "epoch": 0.02684296863349406,
      "grad_norm": 0.3889918327331543,
      "learning_rate": 0.0001999648261652182,
      "loss": 0.2051,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 270,
      "tokens_per_second_per_gpu": 279.64
    },
    {
      "epoch": 0.027837152656956802,
      "grad_norm": 0.4357030391693115,
      "learning_rate": 0.00019996054070972225,
      "loss": 0.3332,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 280,
      "tokens_per_second_per_gpu": 362.33
    },
    {
      "epoch": 0.028831336680419546,
      "grad_norm": 0.3736005425453186,
      "learning_rate": 0.00019995600903028742,
      "loss": 0.3052,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 290,
      "tokens_per_second_per_gpu": 362.89
    },
    {
      "epoch": 0.029825520703882287,
      "grad_norm": 0.39748865365982056,
      "learning_rate": 0.00019995123113807615,
      "loss": 0.361,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 300,
      "tokens_per_second_per_gpu": 387.24
    },
    {
      "epoch": 0.03081970472734503,
      "grad_norm": 0.18977899849414825,
      "learning_rate": 0.00019994620704485741,
      "loss": 0.2449,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 310,
      "tokens_per_second_per_gpu": 339.16
    },
    {
      "epoch": 0.031813888750807776,
      "grad_norm": 0.3898354172706604,
      "learning_rate": 0.00019994093676300662,
      "loss": 0.266,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 320,
      "tokens_per_second_per_gpu": 360.62
    },
    {
      "epoch": 0.032808072774270516,
      "grad_norm": 0.3335312008857727,
      "learning_rate": 0.00019993542030550553,
      "loss": 0.2886,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 330,
      "tokens_per_second_per_gpu": 386.95
    },
    {
      "epoch": 0.03380225679773326,
      "grad_norm": 0.3043772280216217,
      "learning_rate": 0.00019992965768594244,
      "loss": 0.2542,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 340,
      "tokens_per_second_per_gpu": 400.18
    },
    {
      "epoch": 0.034796440821196005,
      "grad_norm": 0.35784608125686646,
      "learning_rate": 0.00019992364891851185,
      "loss": 0.2748,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.66,
      "memory/max_allocated (GiB)": 19.66,
      "step": 350,
      "tokens_per_second_per_gpu": 333.64
    },
    {
      "epoch": 0.035790624844658746,
      "grad_norm": 0.5068204998970032,
      "learning_rate": 0.00019991739401801464,
      "loss": 0.2705,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 360,
      "tokens_per_second_per_gpu": 367.07
    },
    {
      "epoch": 0.03678480886812149,
      "grad_norm": 0.44382113218307495,
      "learning_rate": 0.00019991089299985793,
      "loss": 0.2403,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 370,
      "tokens_per_second_per_gpu": 365.0
    },
    {
      "epoch": 0.037778992891584234,
      "grad_norm": 0.27181145548820496,
      "learning_rate": 0.0001999041458800551,
      "loss": 0.3065,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 380,
      "tokens_per_second_per_gpu": 372.54
    },
    {
      "epoch": 0.038773176915046975,
      "grad_norm": 0.28408923745155334,
      "learning_rate": 0.00019989715267522575,
      "loss": 0.2894,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 390,
      "tokens_per_second_per_gpu": 340.31
    },
    {
      "epoch": 0.039767360938509716,
      "grad_norm": 0.4882698357105255,
      "learning_rate": 0.00019988991340259563,
      "loss": 0.4061,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 400,
      "tokens_per_second_per_gpu": 358.94
    },
    {
      "epoch": 0.040761544961972464,
      "grad_norm": 0.2663392722606659,
      "learning_rate": 0.0001998824280799966,
      "loss": 0.3141,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 410,
      "tokens_per_second_per_gpu": 368.76
    },
    {
      "epoch": 0.041755728985435205,
      "grad_norm": 0.25356051325798035,
      "learning_rate": 0.00019987469672586654,
      "loss": 0.3374,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 420,
      "tokens_per_second_per_gpu": 402.99
    },
    {
      "epoch": 0.042749913008897945,
      "grad_norm": 0.4773045778274536,
      "learning_rate": 0.00019986671935924946,
      "loss": 0.2929,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 430,
      "tokens_per_second_per_gpu": 360.32
    },
    {
      "epoch": 0.04374409703236069,
      "grad_norm": 0.37164929509162903,
      "learning_rate": 0.0001998584959997953,
      "loss": 0.3106,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 440,
      "tokens_per_second_per_gpu": 415.21
    },
    {
      "epoch": 0.044738281055823434,
      "grad_norm": 0.3310747742652893,
      "learning_rate": 0.00019985002666775986,
      "loss": 0.2676,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 450,
      "tokens_per_second_per_gpu": 363.95
    },
    {
      "epoch": 0.045732465079286175,
      "grad_norm": 0.32523512840270996,
      "learning_rate": 0.000199841311384005,
      "loss": 0.3139,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 460,
      "tokens_per_second_per_gpu": 361.92
    },
    {
      "epoch": 0.046726649102748916,
      "grad_norm": 0.40525123476982117,
      "learning_rate": 0.00019983235016999827,
      "loss": 0.323,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 470,
      "tokens_per_second_per_gpu": 413.28
    },
    {
      "epoch": 0.04772083312621166,
      "grad_norm": 0.4233141541481018,
      "learning_rate": 0.000199823143047813,
      "loss": 0.2941,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 480,
      "tokens_per_second_per_gpu": 383.95
    },
    {
      "epoch": 0.048715017149674404,
      "grad_norm": 0.21106044948101044,
      "learning_rate": 0.0001998136900401283,
      "loss": 0.2835,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 490,
      "tokens_per_second_per_gpu": 368.99
    },
    {
      "epoch": 0.049709201173137145,
      "grad_norm": 0.34198832511901855,
      "learning_rate": 0.00019980399117022895,
      "loss": 0.3895,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 500,
      "tokens_per_second_per_gpu": 387.18
    },
    {
      "epoch": 0.05070338519659989,
      "grad_norm": 0.44045203924179077,
      "learning_rate": 0.00019979404646200527,
      "loss": 0.2854,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 510,
      "tokens_per_second_per_gpu": 336.09
    },
    {
      "epoch": 0.051697569220062634,
      "grad_norm": 0.33906373381614685,
      "learning_rate": 0.0001997838559399532,
      "loss": 0.3218,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 520,
      "tokens_per_second_per_gpu": 406.12
    },
    {
      "epoch": 0.052691753243525374,
      "grad_norm": 0.32613444328308105,
      "learning_rate": 0.00019977341962917414,
      "loss": 0.2803,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 530,
      "tokens_per_second_per_gpu": 407.45
    },
    {
      "epoch": 0.05368593726698812,
      "grad_norm": 0.3789099454879761,
      "learning_rate": 0.00019976273755537499,
      "loss": 0.3143,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 540,
      "tokens_per_second_per_gpu": 377.61
    },
    {
      "epoch": 0.05468012129045086,
      "grad_norm": 0.4602185785770416,
      "learning_rate": 0.00019975180974486786,
      "loss": 0.2434,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 550,
      "tokens_per_second_per_gpu": 349.16
    },
    {
      "epoch": 0.055674305313913604,
      "grad_norm": 0.4232983887195587,
      "learning_rate": 0.00019974063622457032,
      "loss": 0.3238,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 560,
      "tokens_per_second_per_gpu": 348.84
    },
    {
      "epoch": 0.05666848933737635,
      "grad_norm": 0.16223137080669403,
      "learning_rate": 0.0001997292170220051,
      "loss": 0.2722,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 570,
      "tokens_per_second_per_gpu": 340.57
    },
    {
      "epoch": 0.05766267336083909,
      "grad_norm": 0.4484419822692871,
      "learning_rate": 0.00019971755216530008,
      "loss": 0.2801,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 580,
      "tokens_per_second_per_gpu": 353.53
    },
    {
      "epoch": 0.05865685738430183,
      "grad_norm": 0.23834413290023804,
      "learning_rate": 0.0001997056416831883,
      "loss": 0.3015,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 590,
      "tokens_per_second_per_gpu": 365.09
    },
    {
      "epoch": 0.059651041407764574,
      "grad_norm": 0.4154009521007538,
      "learning_rate": 0.0001996934856050078,
      "loss": 0.2959,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 600,
      "tokens_per_second_per_gpu": 365.21
    },
    {
      "epoch": 0.06064522543122732,
      "grad_norm": 0.23120558261871338,
      "learning_rate": 0.00019968108396070157,
      "loss": 0.2563,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 610,
      "tokens_per_second_per_gpu": 397.04
    },
    {
      "epoch": 0.06163940945469006,
      "grad_norm": 0.4453487694263458,
      "learning_rate": 0.00019966843678081745,
      "loss": 0.3025,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 620,
      "tokens_per_second_per_gpu": 380.75
    },
    {
      "epoch": 0.06263359347815281,
      "grad_norm": 0.47098028659820557,
      "learning_rate": 0.0001996555440965081,
      "loss": 0.2248,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 630,
      "tokens_per_second_per_gpu": 367.41
    },
    {
      "epoch": 0.06362777750161555,
      "grad_norm": 0.2540164887905121,
      "learning_rate": 0.000199642405939531,
      "loss": 0.2597,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 640,
      "tokens_per_second_per_gpu": 361.69
    },
    {
      "epoch": 0.06462196152507829,
      "grad_norm": 0.30327877402305603,
      "learning_rate": 0.00019962902234224816,
      "loss": 0.2623,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 650,
      "tokens_per_second_per_gpu": 316.43
    },
    {
      "epoch": 0.06561614554854103,
      "grad_norm": 0.3211521804332733,
      "learning_rate": 0.00019961539333762622,
      "loss": 0.2571,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 660,
      "tokens_per_second_per_gpu": 352.04
    },
    {
      "epoch": 0.06661032957200377,
      "grad_norm": 0.19880682229995728,
      "learning_rate": 0.00019960151895923628,
      "loss": 0.2531,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 670,
      "tokens_per_second_per_gpu": 419.33
    },
    {
      "epoch": 0.06760451359546651,
      "grad_norm": 0.3732224702835083,
      "learning_rate": 0.0001995873992412539,
      "loss": 0.3275,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 680,
      "tokens_per_second_per_gpu": 428.82
    },
    {
      "epoch": 0.06859869761892927,
      "grad_norm": 0.2961219847202301,
      "learning_rate": 0.00019957303421845889,
      "loss": 0.2884,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 690,
      "tokens_per_second_per_gpu": 394.73
    },
    {
      "epoch": 0.06959288164239201,
      "grad_norm": 0.4014001488685608,
      "learning_rate": 0.00019955842392623539,
      "loss": 0.25,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 700,
      "tokens_per_second_per_gpu": 402.5
    },
    {
      "epoch": 0.07058706566585475,
      "grad_norm": 0.3465085029602051,
      "learning_rate": 0.0001995435684005716,
      "loss": 0.3208,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 710,
      "tokens_per_second_per_gpu": 418.41
    },
    {
      "epoch": 0.07158124968931749,
      "grad_norm": 0.302223265171051,
      "learning_rate": 0.0001995284676780598,
      "loss": 0.2538,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 720,
      "tokens_per_second_per_gpu": 303.27
    },
    {
      "epoch": 0.07257543371278023,
      "grad_norm": 0.27174416184425354,
      "learning_rate": 0.00019951312179589632,
      "loss": 0.2559,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 730,
      "tokens_per_second_per_gpu": 373.05
    },
    {
      "epoch": 0.07356961773624297,
      "grad_norm": 0.23477095365524292,
      "learning_rate": 0.00019949753079188124,
      "loss": 0.2655,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 740,
      "tokens_per_second_per_gpu": 334.0
    },
    {
      "epoch": 0.07456380175970573,
      "grad_norm": 0.35739773511886597,
      "learning_rate": 0.00019948169470441855,
      "loss": 0.2869,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 750,
      "tokens_per_second_per_gpu": 398.33
    },
    {
      "epoch": 0.07555798578316847,
      "grad_norm": 0.42109552025794983,
      "learning_rate": 0.0001994656135725159,
      "loss": 0.254,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.69,
      "memory/max_allocated (GiB)": 19.69,
      "step": 760,
      "tokens_per_second_per_gpu": 303.95
    },
    {
      "epoch": 0.07655216980663121,
      "grad_norm": 0.5730820298194885,
      "learning_rate": 0.00019944928743578446,
      "loss": 0.2718,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.09,
      "memory/max_allocated (GiB)": 19.09,
      "step": 770,
      "tokens_per_second_per_gpu": 379.86
    },
    {
      "epoch": 0.07754635383009395,
      "grad_norm": 0.3591574430465698,
      "learning_rate": 0.000199432716334439,
      "loss": 0.3039,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 780,
      "tokens_per_second_per_gpu": 387.31
    },
    {
      "epoch": 0.07854053785355669,
      "grad_norm": 0.2447095662355423,
      "learning_rate": 0.0001994159003092976,
      "loss": 0.3044,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 790,
      "tokens_per_second_per_gpu": 332.59
    },
    {
      "epoch": 0.07953472187701943,
      "grad_norm": 1.134704351425171,
      "learning_rate": 0.0001993988394017817,
      "loss": 0.21,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 800,
      "tokens_per_second_per_gpu": 376.02
    },
    {
      "epoch": 0.08052890590048217,
      "grad_norm": 0.4009522795677185,
      "learning_rate": 0.00019938153365391595,
      "loss": 0.3189,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 810,
      "tokens_per_second_per_gpu": 409.87
    },
    {
      "epoch": 0.08152308992394493,
      "grad_norm": 0.32234618067741394,
      "learning_rate": 0.00019936398310832802,
      "loss": 0.3242,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 820,
      "tokens_per_second_per_gpu": 429.4
    },
    {
      "epoch": 0.08251727394740767,
      "grad_norm": 0.42835402488708496,
      "learning_rate": 0.00019934618780824865,
      "loss": 0.2646,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 830,
      "tokens_per_second_per_gpu": 334.03
    },
    {
      "epoch": 0.08351145797087041,
      "grad_norm": 0.39109423756599426,
      "learning_rate": 0.00019932814779751143,
      "loss": 0.2891,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 840,
      "tokens_per_second_per_gpu": 386.65
    },
    {
      "epoch": 0.08450564199433315,
      "grad_norm": 0.30428260564804077,
      "learning_rate": 0.00019930986312055268,
      "loss": 0.2478,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 850,
      "tokens_per_second_per_gpu": 344.59
    },
    {
      "epoch": 0.08549982601779589,
      "grad_norm": 0.41264912486076355,
      "learning_rate": 0.00019929133382241146,
      "loss": 0.2942,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 860,
      "tokens_per_second_per_gpu": 315.94
    },
    {
      "epoch": 0.08649401004125863,
      "grad_norm": 0.3414939045906067,
      "learning_rate": 0.00019927255994872932,
      "loss": 0.2403,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 870,
      "tokens_per_second_per_gpu": 373.66
    },
    {
      "epoch": 0.08748819406472139,
      "grad_norm": 0.3265244662761688,
      "learning_rate": 0.00019925354154575028,
      "loss": 0.2024,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 880,
      "tokens_per_second_per_gpu": 304.2
    },
    {
      "epoch": 0.08848237808818413,
      "grad_norm": 0.45628246665000916,
      "learning_rate": 0.00019923427866032074,
      "loss": 0.2319,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 890,
      "tokens_per_second_per_gpu": 361.11
    },
    {
      "epoch": 0.08947656211164687,
      "grad_norm": 0.15768177807331085,
      "learning_rate": 0.00019921477133988917,
      "loss": 0.2944,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 900,
      "tokens_per_second_per_gpu": 342.86
    },
    {
      "epoch": 0.09047074613510961,
      "grad_norm": 0.31603825092315674,
      "learning_rate": 0.0001991950196325063,
      "loss": 0.2697,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 910,
      "tokens_per_second_per_gpu": 366.81
    },
    {
      "epoch": 0.09146493015857235,
      "grad_norm": 0.30974528193473816,
      "learning_rate": 0.00019917502358682474,
      "loss": 0.2915,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.23,
      "memory/max_allocated (GiB)": 19.23,
      "step": 920,
      "tokens_per_second_per_gpu": 333.81
    },
    {
      "epoch": 0.09245911418203509,
      "grad_norm": 0.5116154551506042,
      "learning_rate": 0.00019915478325209892,
      "loss": 0.2984,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 930,
      "tokens_per_second_per_gpu": 420.26
    },
    {
      "epoch": 0.09345329820549783,
      "grad_norm": 0.4105079174041748,
      "learning_rate": 0.00019913429867818517,
      "loss": 0.2456,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 940,
      "tokens_per_second_per_gpu": 297.2
    },
    {
      "epoch": 0.09444748222896059,
      "grad_norm": 0.32649049162864685,
      "learning_rate": 0.00019911356991554122,
      "loss": 0.2974,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 950,
      "tokens_per_second_per_gpu": 395.2
    },
    {
      "epoch": 0.09544166625242333,
      "grad_norm": 0.3354904353618622,
      "learning_rate": 0.00019909259701522645,
      "loss": 0.2627,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 960,
      "tokens_per_second_per_gpu": 324.97
    },
    {
      "epoch": 0.09643585027588607,
      "grad_norm": 0.33404502272605896,
      "learning_rate": 0.00019907138002890154,
      "loss": 0.2386,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 970,
      "tokens_per_second_per_gpu": 317.18
    },
    {
      "epoch": 0.09743003429934881,
      "grad_norm": 0.3161865472793579,
      "learning_rate": 0.0001990499190088284,
      "loss": 0.2362,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 980,
      "tokens_per_second_per_gpu": 310.78
    },
    {
      "epoch": 0.09842421832281155,
      "grad_norm": 0.33621177077293396,
      "learning_rate": 0.00019902821400787004,
      "loss": 0.2792,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 990,
      "tokens_per_second_per_gpu": 333.17
    },
    {
      "epoch": 0.09941840234627429,
      "grad_norm": 0.09930042922496796,
      "learning_rate": 0.00019900626507949053,
      "loss": 0.2622,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1000,
      "tokens_per_second_per_gpu": 365.14
    },
    {
      "epoch": 0.10041258636973704,
      "grad_norm": 0.3367346525192261,
      "learning_rate": 0.00019898407227775464,
      "loss": 0.2214,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 1010,
      "tokens_per_second_per_gpu": 342.03
    },
    {
      "epoch": 0.10140677039319979,
      "grad_norm": 0.35485416650772095,
      "learning_rate": 0.00019896163565732798,
      "loss": 0.3446,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 1020,
      "tokens_per_second_per_gpu": 378.53
    },
    {
      "epoch": 0.10240095441666253,
      "grad_norm": 0.31839531660079956,
      "learning_rate": 0.0001989389552734767,
      "loss": 0.2519,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 1030,
      "tokens_per_second_per_gpu": 390.74
    },
    {
      "epoch": 0.10339513844012527,
      "grad_norm": 0.36318239569664,
      "learning_rate": 0.0001989160311820673,
      "loss": 0.3,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1040,
      "tokens_per_second_per_gpu": 404.8
    },
    {
      "epoch": 0.10438932246358801,
      "grad_norm": 0.20552317798137665,
      "learning_rate": 0.00019889286343956677,
      "loss": 0.2531,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 1050,
      "tokens_per_second_per_gpu": 349.7
    },
    {
      "epoch": 0.10538350648705075,
      "grad_norm": 0.4065081477165222,
      "learning_rate": 0.00019886945210304208,
      "loss": 0.3196,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 1060,
      "tokens_per_second_per_gpu": 356.46
    },
    {
      "epoch": 0.10637769051051349,
      "grad_norm": 0.3974571228027344,
      "learning_rate": 0.00019884579723016037,
      "loss": 0.2585,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 1070,
      "tokens_per_second_per_gpu": 361.16
    },
    {
      "epoch": 0.10737187453397624,
      "grad_norm": 0.39827367663383484,
      "learning_rate": 0.0001988218988791885,
      "loss": 0.2807,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1080,
      "tokens_per_second_per_gpu": 399.5
    },
    {
      "epoch": 0.10836605855743899,
      "grad_norm": 0.38661155104637146,
      "learning_rate": 0.00019879775710899322,
      "loss": 0.262,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 1090,
      "tokens_per_second_per_gpu": 311.89
    },
    {
      "epoch": 0.10936024258090173,
      "grad_norm": 0.272942453622818,
      "learning_rate": 0.0001987733719790408,
      "loss": 0.1925,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1100,
      "tokens_per_second_per_gpu": 399.72
    },
    {
      "epoch": 0.11035442660436447,
      "grad_norm": 0.32272958755493164,
      "learning_rate": 0.00019874874354939697,
      "loss": 0.1643,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 1110,
      "tokens_per_second_per_gpu": 315.4
    },
    {
      "epoch": 0.11134861062782721,
      "grad_norm": 0.3583936095237732,
      "learning_rate": 0.00019872387188072673,
      "loss": 0.2834,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 1120,
      "tokens_per_second_per_gpu": 439.26
    },
    {
      "epoch": 0.11234279465128995,
      "grad_norm": 0.3295114040374756,
      "learning_rate": 0.00019869875703429433,
      "loss": 0.2157,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 1130,
      "tokens_per_second_per_gpu": 290.71
    },
    {
      "epoch": 0.1133369786747527,
      "grad_norm": 0.21794943511486053,
      "learning_rate": 0.00019867339907196283,
      "loss": 0.2848,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1140,
      "tokens_per_second_per_gpu": 382.48
    },
    {
      "epoch": 0.11433116269821544,
      "grad_norm": 0.47928282618522644,
      "learning_rate": 0.00019864779805619435,
      "loss": 0.2497,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 1150,
      "tokens_per_second_per_gpu": 351.65
    },
    {
      "epoch": 0.11532534672167818,
      "grad_norm": 0.4386768341064453,
      "learning_rate": 0.0001986219540500496,
      "loss": 0.2885,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 1160,
      "tokens_per_second_per_gpu": 351.08
    },
    {
      "epoch": 0.11631953074514093,
      "grad_norm": 0.46047040820121765,
      "learning_rate": 0.00019859586711718776,
      "loss": 0.325,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 1170,
      "tokens_per_second_per_gpu": 375.58
    },
    {
      "epoch": 0.11731371476860367,
      "grad_norm": 0.4252080023288727,
      "learning_rate": 0.00019856953732186653,
      "loss": 0.2923,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 1180,
      "tokens_per_second_per_gpu": 349.36
    },
    {
      "epoch": 0.11830789879206641,
      "grad_norm": 0.3643350899219513,
      "learning_rate": 0.00019854296472894168,
      "loss": 0.2315,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1190,
      "tokens_per_second_per_gpu": 390.28
    },
    {
      "epoch": 0.11930208281552915,
      "grad_norm": 0.4747346341609955,
      "learning_rate": 0.00019851614940386722,
      "loss": 0.3214,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.19,
      "memory/max_allocated (GiB)": 19.19,
      "step": 1200,
      "tokens_per_second_per_gpu": 414.56
    },
    {
      "epoch": 0.1202962668389919,
      "grad_norm": 0.32468438148498535,
      "learning_rate": 0.0001984890914126949,
      "loss": 0.2606,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 1210,
      "tokens_per_second_per_gpu": 393.59
    },
    {
      "epoch": 0.12129045086245464,
      "grad_norm": 0.27018123865127563,
      "learning_rate": 0.00019846179082207429,
      "loss": 0.2457,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 1220,
      "tokens_per_second_per_gpu": 357.71
    },
    {
      "epoch": 0.12228463488591738,
      "grad_norm": 0.39820268750190735,
      "learning_rate": 0.00019843424769925248,
      "loss": 0.2618,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 1230,
      "tokens_per_second_per_gpu": 377.03
    },
    {
      "epoch": 0.12327881890938013,
      "grad_norm": 0.4186467230319977,
      "learning_rate": 0.00019840646211207407,
      "loss": 0.2864,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 1240,
      "tokens_per_second_per_gpu": 415.58
    },
    {
      "epoch": 0.12427300293284287,
      "grad_norm": 0.3047218918800354,
      "learning_rate": 0.00019837843412898081,
      "loss": 0.1777,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 1250,
      "tokens_per_second_per_gpu": 376.07
    },
    {
      "epoch": 0.12526718695630562,
      "grad_norm": 0.3663698136806488,
      "learning_rate": 0.0001983501638190115,
      "loss": 0.2906,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1260,
      "tokens_per_second_per_gpu": 341.64
    },
    {
      "epoch": 0.12626137097976836,
      "grad_norm": 0.5897945761680603,
      "learning_rate": 0.00019832165125180194,
      "loss": 0.2498,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 1270,
      "tokens_per_second_per_gpu": 380.54
    },
    {
      "epoch": 0.1272555550032311,
      "grad_norm": 0.40836209058761597,
      "learning_rate": 0.0001982928964975846,
      "loss": 0.2722,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 1280,
      "tokens_per_second_per_gpu": 387.55
    },
    {
      "epoch": 0.12824973902669384,
      "grad_norm": 0.33597612380981445,
      "learning_rate": 0.00019826389962718848,
      "loss": 0.3202,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1290,
      "tokens_per_second_per_gpu": 313.03
    },
    {
      "epoch": 0.12924392305015658,
      "grad_norm": 0.44784456491470337,
      "learning_rate": 0.00019823466071203902,
      "loss": 0.2949,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1300,
      "tokens_per_second_per_gpu": 390.03
    },
    {
      "epoch": 0.13023810707361932,
      "grad_norm": 0.3199595510959625,
      "learning_rate": 0.0001982051798241579,
      "loss": 0.2323,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 1310,
      "tokens_per_second_per_gpu": 415.8
    },
    {
      "epoch": 0.13123229109708207,
      "grad_norm": 0.4944785535335541,
      "learning_rate": 0.0001981754570361627,
      "loss": 0.291,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 1320,
      "tokens_per_second_per_gpu": 347.08
    },
    {
      "epoch": 0.1322264751205448,
      "grad_norm": 0.379162073135376,
      "learning_rate": 0.00019814549242126698,
      "loss": 0.2631,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 1330,
      "tokens_per_second_per_gpu": 405.3
    },
    {
      "epoch": 0.13322065914400755,
      "grad_norm": 0.20690025389194489,
      "learning_rate": 0.00019811528605327992,
      "loss": 0.2099,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 1340,
      "tokens_per_second_per_gpu": 377.71
    },
    {
      "epoch": 0.1342148431674703,
      "grad_norm": 0.39738351106643677,
      "learning_rate": 0.00019808483800660612,
      "loss": 0.2486,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 1350,
      "tokens_per_second_per_gpu": 333.2
    },
    {
      "epoch": 0.13520902719093303,
      "grad_norm": 0.5237305164337158,
      "learning_rate": 0.00019805414835624566,
      "loss": 0.2407,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 1360,
      "tokens_per_second_per_gpu": 353.94
    },
    {
      "epoch": 0.1362032112143958,
      "grad_norm": 0.2773837447166443,
      "learning_rate": 0.00019802321717779354,
      "loss": 0.3119,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 1370,
      "tokens_per_second_per_gpu": 391.05
    },
    {
      "epoch": 0.13719739523785854,
      "grad_norm": 0.2825298011302948,
      "learning_rate": 0.00019799204454743987,
      "loss": 0.2812,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 1380,
      "tokens_per_second_per_gpu": 348.62
    },
    {
      "epoch": 0.13819157926132128,
      "grad_norm": 0.3622908592224121,
      "learning_rate": 0.00019796063054196937,
      "loss": 0.2506,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1390,
      "tokens_per_second_per_gpu": 380.11
    },
    {
      "epoch": 0.13918576328478402,
      "grad_norm": 0.3992385268211365,
      "learning_rate": 0.0001979289752387614,
      "loss": 0.2132,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1400,
      "tokens_per_second_per_gpu": 301.04
    },
    {
      "epoch": 0.14017994730824676,
      "grad_norm": 0.4148050546646118,
      "learning_rate": 0.00019789707871578966,
      "loss": 0.1813,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 1410,
      "tokens_per_second_per_gpu": 296.18
    },
    {
      "epoch": 0.1411741313317095,
      "grad_norm": 0.36811864376068115,
      "learning_rate": 0.000197864941051622,
      "loss": 0.2607,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 1420,
      "tokens_per_second_per_gpu": 376.48
    },
    {
      "epoch": 0.14216831535517224,
      "grad_norm": 0.33353865146636963,
      "learning_rate": 0.00019783256232542033,
      "loss": 0.2694,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 1430,
      "tokens_per_second_per_gpu": 352.36
    },
    {
      "epoch": 0.14316249937863498,
      "grad_norm": 0.4390527606010437,
      "learning_rate": 0.00019779994261694025,
      "loss": 0.2851,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 1440,
      "tokens_per_second_per_gpu": 332.85
    },
    {
      "epoch": 0.14415668340209772,
      "grad_norm": 0.4553990066051483,
      "learning_rate": 0.00019776708200653102,
      "loss": 0.3301,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 1450,
      "tokens_per_second_per_gpu": 400.73
    },
    {
      "epoch": 0.14515086742556046,
      "grad_norm": 0.3526112139225006,
      "learning_rate": 0.00019773398057513526,
      "loss": 0.2276,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 1460,
      "tokens_per_second_per_gpu": 342.73
    },
    {
      "epoch": 0.1461450514490232,
      "grad_norm": 0.4758242070674896,
      "learning_rate": 0.0001977006384042888,
      "loss": 0.2185,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 1470,
      "tokens_per_second_per_gpu": 343.6
    },
    {
      "epoch": 0.14713923547248595,
      "grad_norm": 0.4020686447620392,
      "learning_rate": 0.00019766705557612045,
      "loss": 0.2598,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 1480,
      "tokens_per_second_per_gpu": 288.04
    },
    {
      "epoch": 0.1481334194959487,
      "grad_norm": 0.44152265787124634,
      "learning_rate": 0.00019763323217335182,
      "loss": 0.3394,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 1490,
      "tokens_per_second_per_gpu": 408.81
    },
    {
      "epoch": 0.14912760351941146,
      "grad_norm": 0.31458431482315063,
      "learning_rate": 0.00019759916827929706,
      "loss": 0.2692,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 1500,
      "tokens_per_second_per_gpu": 359.64
    },
    {
      "epoch": 0.1501217875428742,
      "grad_norm": 0.48072609305381775,
      "learning_rate": 0.0001975648639778628,
      "loss": 0.3545,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 1510,
      "tokens_per_second_per_gpu": 374.78
    },
    {
      "epoch": 0.15111597156633694,
      "grad_norm": 0.30275699496269226,
      "learning_rate": 0.00019753031935354777,
      "loss": 0.2109,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 1520,
      "tokens_per_second_per_gpu": 371.06
    },
    {
      "epoch": 0.15211015558979968,
      "grad_norm": 0.5390923619270325,
      "learning_rate": 0.00019749553449144267,
      "loss": 0.2435,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 1530,
      "tokens_per_second_per_gpu": 368.76
    },
    {
      "epoch": 0.15310433961326242,
      "grad_norm": 0.28221625089645386,
      "learning_rate": 0.00019746050947722993,
      "loss": 0.2105,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1540,
      "tokens_per_second_per_gpu": 322.88
    },
    {
      "epoch": 0.15409852363672516,
      "grad_norm": 0.3471927046775818,
      "learning_rate": 0.00019742524439718363,
      "loss": 0.2761,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 17.4,
      "memory/max_allocated (GiB)": 17.4,
      "step": 1550,
      "tokens_per_second_per_gpu": 331.45
    },
    {
      "epoch": 0.1550927076601879,
      "grad_norm": 0.34601831436157227,
      "learning_rate": 0.0001973897393381691,
      "loss": 0.2419,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1560,
      "tokens_per_second_per_gpu": 330.91
    },
    {
      "epoch": 0.15608689168365064,
      "grad_norm": 0.4680122435092926,
      "learning_rate": 0.00019735399438764275,
      "loss": 0.2948,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 1570,
      "tokens_per_second_per_gpu": 369.5
    },
    {
      "epoch": 0.15708107570711338,
      "grad_norm": 0.35631850361824036,
      "learning_rate": 0.000197318009633652,
      "loss": 0.2865,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 1580,
      "tokens_per_second_per_gpu": 331.56
    },
    {
      "epoch": 0.15807525973057612,
      "grad_norm": 0.43517372012138367,
      "learning_rate": 0.0001972817851648349,
      "loss": 0.2912,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1590,
      "tokens_per_second_per_gpu": 334.38
    },
    {
      "epoch": 0.15906944375403886,
      "grad_norm": 0.3614802360534668,
      "learning_rate": 0.00019724532107041995,
      "loss": 0.2182,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 1600,
      "tokens_per_second_per_gpu": 330.71
    },
    {
      "epoch": 0.1600636277775016,
      "grad_norm": 0.3124898672103882,
      "learning_rate": 0.00019720861744022594,
      "loss": 0.1887,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1610,
      "tokens_per_second_per_gpu": 419.4
    },
    {
      "epoch": 0.16105781180096435,
      "grad_norm": 0.37882000207901,
      "learning_rate": 0.00019717167436466166,
      "loss": 0.3199,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 1620,
      "tokens_per_second_per_gpu": 352.77
    },
    {
      "epoch": 0.16205199582442711,
      "grad_norm": 0.47813880443573,
      "learning_rate": 0.00019713449193472572,
      "loss": 0.2644,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 1630,
      "tokens_per_second_per_gpu": 328.31
    },
    {
      "epoch": 0.16304617984788985,
      "grad_norm": 0.4390937387943268,
      "learning_rate": 0.00019709707024200633,
      "loss": 0.2157,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 1640,
      "tokens_per_second_per_gpu": 337.97
    },
    {
      "epoch": 0.1640403638713526,
      "grad_norm": 0.28492602705955505,
      "learning_rate": 0.00019705940937868096,
      "loss": 0.2301,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1650,
      "tokens_per_second_per_gpu": 377.43
    },
    {
      "epoch": 0.16503454789481534,
      "grad_norm": 0.41057339310646057,
      "learning_rate": 0.00019702150943751636,
      "loss": 0.2755,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 1660,
      "tokens_per_second_per_gpu": 364.78
    },
    {
      "epoch": 0.16602873191827808,
      "grad_norm": 0.36004287004470825,
      "learning_rate": 0.00019698337051186803,
      "loss": 0.2254,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 1670,
      "tokens_per_second_per_gpu": 348.57
    },
    {
      "epoch": 0.16702291594174082,
      "grad_norm": 0.36488133668899536,
      "learning_rate": 0.00019694499269568022,
      "loss": 0.2556,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 1680,
      "tokens_per_second_per_gpu": 338.09
    },
    {
      "epoch": 0.16801709996520356,
      "grad_norm": 0.1386333703994751,
      "learning_rate": 0.00019690637608348562,
      "loss": 0.2765,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 1690,
      "tokens_per_second_per_gpu": 279.22
    },
    {
      "epoch": 0.1690112839886663,
      "grad_norm": 0.30839794874191284,
      "learning_rate": 0.00019686752077040505,
      "loss": 0.2745,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 1700,
      "tokens_per_second_per_gpu": 363.03
    },
    {
      "epoch": 0.17000546801212904,
      "grad_norm": 0.48986828327178955,
      "learning_rate": 0.00019682842685214745,
      "loss": 0.2415,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 1710,
      "tokens_per_second_per_gpu": 321.69
    },
    {
      "epoch": 0.17099965203559178,
      "grad_norm": 0.25523653626441956,
      "learning_rate": 0.00019678909442500937,
      "loss": 0.2618,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 1720,
      "tokens_per_second_per_gpu": 312.41
    },
    {
      "epoch": 0.17199383605905452,
      "grad_norm": 0.1990797519683838,
      "learning_rate": 0.00019674952358587488,
      "loss": 0.2569,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 1730,
      "tokens_per_second_per_gpu": 338.48
    },
    {
      "epoch": 0.17298802008251726,
      "grad_norm": 1.2966935634613037,
      "learning_rate": 0.00019670971443221528,
      "loss": 0.2789,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 1740,
      "tokens_per_second_per_gpu": 377.96
    },
    {
      "epoch": 0.17398220410598,
      "grad_norm": 0.4156191647052765,
      "learning_rate": 0.00019666966706208898,
      "loss": 0.2537,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1750,
      "tokens_per_second_per_gpu": 423.21
    },
    {
      "epoch": 0.17497638812944277,
      "grad_norm": 0.43119025230407715,
      "learning_rate": 0.00019662938157414113,
      "loss": 0.3316,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 1760,
      "tokens_per_second_per_gpu": 385.14
    },
    {
      "epoch": 0.1759705721529055,
      "grad_norm": 0.4492398798465729,
      "learning_rate": 0.00019658885806760336,
      "loss": 0.2969,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 1770,
      "tokens_per_second_per_gpu": 364.99
    },
    {
      "epoch": 0.17696475617636825,
      "grad_norm": 0.4664623737335205,
      "learning_rate": 0.00019654809664229364,
      "loss": 0.291,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1780,
      "tokens_per_second_per_gpu": 295.57
    },
    {
      "epoch": 0.177958940199831,
      "grad_norm": 0.46608006954193115,
      "learning_rate": 0.000196507097398616,
      "loss": 0.295,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1790,
      "tokens_per_second_per_gpu": 284.68
    },
    {
      "epoch": 0.17895312422329374,
      "grad_norm": 0.378513365983963,
      "learning_rate": 0.00019646586043756023,
      "loss": 0.2396,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 1800,
      "tokens_per_second_per_gpu": 325.71
    },
    {
      "epoch": 0.17994730824675648,
      "grad_norm": 0.36179885268211365,
      "learning_rate": 0.00019642438586070168,
      "loss": 0.2364,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 1810,
      "tokens_per_second_per_gpu": 349.38
    },
    {
      "epoch": 0.18094149227021922,
      "grad_norm": 0.31644207239151,
      "learning_rate": 0.000196382673770201,
      "loss": 0.223,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 1820,
      "tokens_per_second_per_gpu": 320.59
    },
    {
      "epoch": 0.18193567629368196,
      "grad_norm": 0.27562472224235535,
      "learning_rate": 0.00019634072426880382,
      "loss": 0.2641,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 1830,
      "tokens_per_second_per_gpu": 386.93
    },
    {
      "epoch": 0.1829298603171447,
      "grad_norm": 0.4514225423336029,
      "learning_rate": 0.00019629853745984076,
      "loss": 0.2167,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1840,
      "tokens_per_second_per_gpu": 350.66
    },
    {
      "epoch": 0.18392404434060744,
      "grad_norm": 0.43807438015937805,
      "learning_rate": 0.00019625611344722675,
      "loss": 0.2429,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1850,
      "tokens_per_second_per_gpu": 390.61
    },
    {
      "epoch": 0.18491822836407018,
      "grad_norm": 0.3817460536956787,
      "learning_rate": 0.00019621345233546115,
      "loss": 0.2565,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 1860,
      "tokens_per_second_per_gpu": 366.89
    },
    {
      "epoch": 0.18591241238753292,
      "grad_norm": 0.51050865650177,
      "learning_rate": 0.0001961705542296272,
      "loss": 0.2815,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 1870,
      "tokens_per_second_per_gpu": 314.07
    },
    {
      "epoch": 0.18690659641099566,
      "grad_norm": 0.34085533022880554,
      "learning_rate": 0.00019612741923539218,
      "loss": 0.2353,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 1880,
      "tokens_per_second_per_gpu": 366.93
    },
    {
      "epoch": 0.18790078043445843,
      "grad_norm": 0.38903746008872986,
      "learning_rate": 0.00019608404745900652,
      "loss": 0.2679,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 1890,
      "tokens_per_second_per_gpu": 407.24
    },
    {
      "epoch": 0.18889496445792117,
      "grad_norm": 0.3198868930339813,
      "learning_rate": 0.00019604043900730414,
      "loss": 0.299,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 1900,
      "tokens_per_second_per_gpu": 420.81
    },
    {
      "epoch": 0.1898891484813839,
      "grad_norm": 0.43067699670791626,
      "learning_rate": 0.0001959965939877019,
      "loss": 0.2957,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 1910,
      "tokens_per_second_per_gpu": 467.81
    },
    {
      "epoch": 0.19088333250484665,
      "grad_norm": 0.3319333493709564,
      "learning_rate": 0.00019595251250819932,
      "loss": 0.2512,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 17.12,
      "memory/max_allocated (GiB)": 17.12,
      "step": 1920,
      "tokens_per_second_per_gpu": 283.54
    },
    {
      "epoch": 0.1918775165283094,
      "grad_norm": 0.399617999792099,
      "learning_rate": 0.00019590819467737837,
      "loss": 0.2627,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 1930,
      "tokens_per_second_per_gpu": 314.92
    },
    {
      "epoch": 0.19287170055177214,
      "grad_norm": 0.31971487402915955,
      "learning_rate": 0.00019586364060440332,
      "loss": 0.2705,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 1940,
      "tokens_per_second_per_gpu": 371.9
    },
    {
      "epoch": 0.19386588457523488,
      "grad_norm": 0.22907792031764984,
      "learning_rate": 0.0001958188503990202,
      "loss": 0.2479,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 1950,
      "tokens_per_second_per_gpu": 329.97
    },
    {
      "epoch": 0.19486006859869762,
      "grad_norm": 0.3624865710735321,
      "learning_rate": 0.00019577382417155676,
      "loss": 0.309,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 1960,
      "tokens_per_second_per_gpu": 466.41
    },
    {
      "epoch": 0.19585425262216036,
      "grad_norm": 0.3285991847515106,
      "learning_rate": 0.00019572856203292215,
      "loss": 0.2188,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 1970,
      "tokens_per_second_per_gpu": 336.34
    },
    {
      "epoch": 0.1968484366456231,
      "grad_norm": 0.22952090203762054,
      "learning_rate": 0.00019568306409460654,
      "loss": 0.2277,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 1980,
      "tokens_per_second_per_gpu": 377.15
    },
    {
      "epoch": 0.19784262066908584,
      "grad_norm": 0.38868236541748047,
      "learning_rate": 0.000195637330468681,
      "loss": 0.299,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 1990,
      "tokens_per_second_per_gpu": 371.96
    },
    {
      "epoch": 0.19883680469254858,
      "grad_norm": 0.30058735609054565,
      "learning_rate": 0.0001955913612677971,
      "loss": 0.2993,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2000,
      "tokens_per_second_per_gpu": 418.93
    },
    {
      "epoch": 0.19983098871601132,
      "grad_norm": 0.25231263041496277,
      "learning_rate": 0.00019554515660518668,
      "loss": 0.2894,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 2010,
      "tokens_per_second_per_gpu": 372.68
    },
    {
      "epoch": 0.2008251727394741,
      "grad_norm": 0.5621690154075623,
      "learning_rate": 0.00019549871659466165,
      "loss": 0.2152,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 2020,
      "tokens_per_second_per_gpu": 358.38
    },
    {
      "epoch": 0.20181935676293683,
      "grad_norm": 0.3443728983402252,
      "learning_rate": 0.0001954520413506135,
      "loss": 0.2398,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2030,
      "tokens_per_second_per_gpu": 306.09
    },
    {
      "epoch": 0.20281354078639957,
      "grad_norm": 0.40964439511299133,
      "learning_rate": 0.0001954051309880133,
      "loss": 0.2259,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 2040,
      "tokens_per_second_per_gpu": 371.31
    },
    {
      "epoch": 0.2038077248098623,
      "grad_norm": 0.49535176157951355,
      "learning_rate": 0.0001953579856224111,
      "loss": 0.2681,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 2050,
      "tokens_per_second_per_gpu": 368.85
    },
    {
      "epoch": 0.20480190883332505,
      "grad_norm": 0.4733733832836151,
      "learning_rate": 0.00019531060536993598,
      "loss": 0.2309,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.47,
      "memory/max_allocated (GiB)": 21.47,
      "step": 2060,
      "tokens_per_second_per_gpu": 378.99
    },
    {
      "epoch": 0.2057960928567878,
      "grad_norm": 0.4018077552318573,
      "learning_rate": 0.00019526299034729544,
      "loss": 0.2717,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 2070,
      "tokens_per_second_per_gpu": 373.89
    },
    {
      "epoch": 0.20679027688025053,
      "grad_norm": 0.18390242755413055,
      "learning_rate": 0.0001952151406717754,
      "loss": 0.2604,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 2080,
      "tokens_per_second_per_gpu": 339.07
    },
    {
      "epoch": 0.20778446090371328,
      "grad_norm": 0.31284740567207336,
      "learning_rate": 0.0001951670564612397,
      "loss": 0.2528,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 2090,
      "tokens_per_second_per_gpu": 358.01
    },
    {
      "epoch": 0.20877864492717602,
      "grad_norm": 0.1479184776544571,
      "learning_rate": 0.0001951187378341299,
      "loss": 0.2753,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 2100,
      "tokens_per_second_per_gpu": 373.8
    },
    {
      "epoch": 0.20977282895063876,
      "grad_norm": 0.35812532901763916,
      "learning_rate": 0.00019507018490946503,
      "loss": 0.2799,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2110,
      "tokens_per_second_per_gpu": 394.97
    },
    {
      "epoch": 0.2107670129741015,
      "grad_norm": 0.39899322390556335,
      "learning_rate": 0.00019502139780684118,
      "loss": 0.2785,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 2120,
      "tokens_per_second_per_gpu": 371.01
    },
    {
      "epoch": 0.21176119699756424,
      "grad_norm": 0.30708786845207214,
      "learning_rate": 0.00019497237664643132,
      "loss": 0.2985,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 2130,
      "tokens_per_second_per_gpu": 393.88
    },
    {
      "epoch": 0.21275538102102698,
      "grad_norm": 0.280734658241272,
      "learning_rate": 0.00019492312154898488,
      "loss": 0.2661,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 2140,
      "tokens_per_second_per_gpu": 401.57
    },
    {
      "epoch": 0.21374956504448975,
      "grad_norm": 0.19114673137664795,
      "learning_rate": 0.00019487363263582765,
      "loss": 0.2197,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 2150,
      "tokens_per_second_per_gpu": 346.18
    },
    {
      "epoch": 0.2147437490679525,
      "grad_norm": 0.5506372451782227,
      "learning_rate": 0.00019482391002886122,
      "loss": 0.2724,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 2160,
      "tokens_per_second_per_gpu": 281.19
    },
    {
      "epoch": 0.21573793309141523,
      "grad_norm": 0.3187256157398224,
      "learning_rate": 0.0001947739538505629,
      "loss": 0.2758,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2170,
      "tokens_per_second_per_gpu": 405.51
    },
    {
      "epoch": 0.21673211711487797,
      "grad_norm": 0.4344545602798462,
      "learning_rate": 0.00019472376422398528,
      "loss": 0.2792,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2180,
      "tokens_per_second_per_gpu": 313.61
    },
    {
      "epoch": 0.2177263011383407,
      "grad_norm": 0.4564478099346161,
      "learning_rate": 0.00019467334127275606,
      "loss": 0.2474,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.43,
      "memory/max_allocated (GiB)": 20.43,
      "step": 2190,
      "tokens_per_second_per_gpu": 328.19
    },
    {
      "epoch": 0.21872048516180345,
      "grad_norm": 0.25979048013687134,
      "learning_rate": 0.00019462268512107766,
      "loss": 0.2877,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2200,
      "tokens_per_second_per_gpu": 360.33
    },
    {
      "epoch": 0.2197146691852662,
      "grad_norm": 0.3676232397556305,
      "learning_rate": 0.00019457179589372684,
      "loss": 0.3336,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2210,
      "tokens_per_second_per_gpu": 368.21
    },
    {
      "epoch": 0.22070885320872893,
      "grad_norm": 0.7130278944969177,
      "learning_rate": 0.0001945206737160545,
      "loss": 0.253,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 2220,
      "tokens_per_second_per_gpu": 305.12
    },
    {
      "epoch": 0.22170303723219167,
      "grad_norm": 2.660079002380371,
      "learning_rate": 0.0001944693187139854,
      "loss": 0.3069,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 2230,
      "tokens_per_second_per_gpu": 337.93
    },
    {
      "epoch": 0.22269722125565442,
      "grad_norm": 0.2822214663028717,
      "learning_rate": 0.00019441773101401777,
      "loss": 0.2744,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 2240,
      "tokens_per_second_per_gpu": 349.76
    },
    {
      "epoch": 0.22369140527911716,
      "grad_norm": 0.45128756761550903,
      "learning_rate": 0.00019436591074322302,
      "loss": 0.245,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 2250,
      "tokens_per_second_per_gpu": 390.72
    },
    {
      "epoch": 0.2246855893025799,
      "grad_norm": 0.2821448743343353,
      "learning_rate": 0.00019431385802924539,
      "loss": 0.2625,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2260,
      "tokens_per_second_per_gpu": 321.11
    },
    {
      "epoch": 0.22567977332604264,
      "grad_norm": 0.3787819445133209,
      "learning_rate": 0.00019426157300030176,
      "loss": 0.2116,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 2270,
      "tokens_per_second_per_gpu": 345.13
    },
    {
      "epoch": 0.2266739573495054,
      "grad_norm": 0.36587730050086975,
      "learning_rate": 0.0001942090557851812,
      "loss": 0.2618,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 2280,
      "tokens_per_second_per_gpu": 288.26
    },
    {
      "epoch": 0.22766814137296815,
      "grad_norm": 0.5559653639793396,
      "learning_rate": 0.0001941563065132447,
      "loss": 0.307,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2290,
      "tokens_per_second_per_gpu": 402.27
    },
    {
      "epoch": 0.2286623253964309,
      "grad_norm": 0.26463228464126587,
      "learning_rate": 0.0001941033253144249,
      "loss": 0.2583,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2300,
      "tokens_per_second_per_gpu": 388.95
    },
    {
      "epoch": 0.22965650941989363,
      "grad_norm": 0.330695778131485,
      "learning_rate": 0.0001940501123192256,
      "loss": 0.2671,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 2310,
      "tokens_per_second_per_gpu": 353.27
    },
    {
      "epoch": 0.23065069344335637,
      "grad_norm": 0.3159751892089844,
      "learning_rate": 0.00019399666765872176,
      "loss": 0.2023,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2320,
      "tokens_per_second_per_gpu": 347.05
    },
    {
      "epoch": 0.2316448774668191,
      "grad_norm": 0.4762006103992462,
      "learning_rate": 0.0001939429914645588,
      "loss": 0.2633,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 2330,
      "tokens_per_second_per_gpu": 292.93
    },
    {
      "epoch": 0.23263906149028185,
      "grad_norm": 0.47802531719207764,
      "learning_rate": 0.00019388908386895254,
      "loss": 0.2381,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 2340,
      "tokens_per_second_per_gpu": 366.87
    },
    {
      "epoch": 0.2336332455137446,
      "grad_norm": 0.4501783847808838,
      "learning_rate": 0.00019383494500468883,
      "loss": 0.3052,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 2350,
      "tokens_per_second_per_gpu": 412.56
    },
    {
      "epoch": 0.23462742953720733,
      "grad_norm": 0.3937671184539795,
      "learning_rate": 0.0001937805750051231,
      "loss": 0.2495,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2360,
      "tokens_per_second_per_gpu": 374.59
    },
    {
      "epoch": 0.23562161356067007,
      "grad_norm": 0.2139206975698471,
      "learning_rate": 0.00019372597400418019,
      "loss": 0.1679,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 2370,
      "tokens_per_second_per_gpu": 329.45
    },
    {
      "epoch": 0.23661579758413281,
      "grad_norm": 0.4658704102039337,
      "learning_rate": 0.00019367114213635382,
      "loss": 0.2242,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 2380,
      "tokens_per_second_per_gpu": 302.96
    },
    {
      "epoch": 0.23760998160759556,
      "grad_norm": 0.34023401141166687,
      "learning_rate": 0.00019361607953670654,
      "loss": 0.2632,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2390,
      "tokens_per_second_per_gpu": 387.14
    },
    {
      "epoch": 0.2386041656310583,
      "grad_norm": 0.4804230034351349,
      "learning_rate": 0.00019356078634086914,
      "loss": 0.251,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 2400,
      "tokens_per_second_per_gpu": 410.6
    },
    {
      "epoch": 0.23959834965452106,
      "grad_norm": 0.32980307936668396,
      "learning_rate": 0.00019350526268504048,
      "loss": 0.176,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 2410,
      "tokens_per_second_per_gpu": 372.52
    },
    {
      "epoch": 0.2405925336779838,
      "grad_norm": 0.25920480489730835,
      "learning_rate": 0.00019344950870598703,
      "loss": 0.2976,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.83,
      "memory/max_allocated (GiB)": 19.83,
      "step": 2420,
      "tokens_per_second_per_gpu": 330.47
    },
    {
      "epoch": 0.24158671770144655,
      "grad_norm": 0.455616295337677,
      "learning_rate": 0.00019339352454104264,
      "loss": 0.2976,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 2430,
      "tokens_per_second_per_gpu": 340.47
    },
    {
      "epoch": 0.2425809017249093,
      "grad_norm": 0.3018989562988281,
      "learning_rate": 0.00019333731032810812,
      "loss": 0.2732,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2440,
      "tokens_per_second_per_gpu": 462.35
    },
    {
      "epoch": 0.24357508574837203,
      "grad_norm": 0.4818798303604126,
      "learning_rate": 0.00019328086620565095,
      "loss": 0.2886,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 2450,
      "tokens_per_second_per_gpu": 323.25
    },
    {
      "epoch": 0.24456926977183477,
      "grad_norm": 0.4339105188846588,
      "learning_rate": 0.000193224192312705,
      "loss": 0.3325,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 2460,
      "tokens_per_second_per_gpu": 401.06
    },
    {
      "epoch": 0.2455634537952975,
      "grad_norm": 0.2657606303691864,
      "learning_rate": 0.00019316728878887,
      "loss": 0.2369,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2470,
      "tokens_per_second_per_gpu": 344.52
    },
    {
      "epoch": 0.24655763781876025,
      "grad_norm": 0.2487681657075882,
      "learning_rate": 0.0001931101557743113,
      "loss": 0.2481,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 2480,
      "tokens_per_second_per_gpu": 338.75
    },
    {
      "epoch": 0.247551821842223,
      "grad_norm": 0.4219423532485962,
      "learning_rate": 0.0001930527934097597,
      "loss": 0.2467,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.23,
      "memory/max_allocated (GiB)": 19.23,
      "step": 2490,
      "tokens_per_second_per_gpu": 357.65
    },
    {
      "epoch": 0.24854600586568573,
      "grad_norm": 0.3535912334918976,
      "learning_rate": 0.00019299520183651075,
      "loss": 0.2844,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 2500,
      "tokens_per_second_per_gpu": 341.4
    },
    {
      "epoch": 0.24954018988914847,
      "grad_norm": 0.28509521484375,
      "learning_rate": 0.0001929373811964247,
      "loss": 0.2335,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 2510,
      "tokens_per_second_per_gpu": 325.61
    },
    {
      "epoch": 0.25053437391261124,
      "grad_norm": 0.540902853012085,
      "learning_rate": 0.00019287933163192602,
      "loss": 0.314,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 2520,
      "tokens_per_second_per_gpu": 390.5
    },
    {
      "epoch": 0.251528557936074,
      "grad_norm": 0.2346281111240387,
      "learning_rate": 0.00019282105328600303,
      "loss": 0.2592,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 17.12,
      "memory/max_allocated (GiB)": 17.12,
      "step": 2530,
      "tokens_per_second_per_gpu": 335.14
    },
    {
      "epoch": 0.2525227419595367,
      "grad_norm": 0.4211244583129883,
      "learning_rate": 0.0001927625463022076,
      "loss": 0.2119,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2540,
      "tokens_per_second_per_gpu": 360.45
    },
    {
      "epoch": 0.25351692598299946,
      "grad_norm": 0.5125908851623535,
      "learning_rate": 0.00019270381082465483,
      "loss": 0.2628,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 2550,
      "tokens_per_second_per_gpu": 316.73
    },
    {
      "epoch": 0.2545111100064622,
      "grad_norm": 0.20704445242881775,
      "learning_rate": 0.00019264484699802262,
      "loss": 0.2393,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 2560,
      "tokens_per_second_per_gpu": 352.58
    },
    {
      "epoch": 0.25550529402992495,
      "grad_norm": 0.2666438817977905,
      "learning_rate": 0.00019258565496755128,
      "loss": 0.1621,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 2570,
      "tokens_per_second_per_gpu": 346.34
    },
    {
      "epoch": 0.2564994780533877,
      "grad_norm": 0.4072653353214264,
      "learning_rate": 0.00019252623487904335,
      "loss": 0.2066,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 17.11,
      "memory/max_allocated (GiB)": 17.11,
      "step": 2580,
      "tokens_per_second_per_gpu": 360.03
    },
    {
      "epoch": 0.2574936620768504,
      "grad_norm": 0.517437219619751,
      "learning_rate": 0.00019246658687886302,
      "loss": 0.256,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 2590,
      "tokens_per_second_per_gpu": 335.1
    },
    {
      "epoch": 0.25848784610031317,
      "grad_norm": 1.1084229946136475,
      "learning_rate": 0.00019240671111393597,
      "loss": 0.2437,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2600,
      "tokens_per_second_per_gpu": 336.03
    },
    {
      "epoch": 0.2594820301237759,
      "grad_norm": 0.3532284200191498,
      "learning_rate": 0.00019234660773174883,
      "loss": 0.2102,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 2610,
      "tokens_per_second_per_gpu": 304.29
    },
    {
      "epoch": 0.26047621414723865,
      "grad_norm": 0.4219340980052948,
      "learning_rate": 0.00019228627688034898,
      "loss": 0.3338,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.61,
      "memory/max_allocated (GiB)": 20.61,
      "step": 2620,
      "tokens_per_second_per_gpu": 350.23
    },
    {
      "epoch": 0.2614703981707014,
      "grad_norm": 0.3215112090110779,
      "learning_rate": 0.000192225718708344,
      "loss": 0.2466,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 2630,
      "tokens_per_second_per_gpu": 369.51
    },
    {
      "epoch": 0.26246458219416413,
      "grad_norm": 0.36818957328796387,
      "learning_rate": 0.00019216493336490152,
      "loss": 0.2839,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 2640,
      "tokens_per_second_per_gpu": 369.73
    },
    {
      "epoch": 0.26345876621762687,
      "grad_norm": 0.41190099716186523,
      "learning_rate": 0.0001921039209997486,
      "loss": 0.2231,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2650,
      "tokens_per_second_per_gpu": 334.38
    },
    {
      "epoch": 0.2644529502410896,
      "grad_norm": 0.3659015893936157,
      "learning_rate": 0.0001920426817631717,
      "loss": 0.162,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 2660,
      "tokens_per_second_per_gpu": 336.11
    },
    {
      "epoch": 0.26544713426455235,
      "grad_norm": 0.13166293501853943,
      "learning_rate": 0.00019198121580601596,
      "loss": 0.2313,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 2670,
      "tokens_per_second_per_gpu": 341.25
    },
    {
      "epoch": 0.2664413182880151,
      "grad_norm": 0.3052745759487152,
      "learning_rate": 0.00019191952327968497,
      "loss": 0.2887,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 2680,
      "tokens_per_second_per_gpu": 390.54
    },
    {
      "epoch": 0.26743550231147784,
      "grad_norm": 0.3094649910926819,
      "learning_rate": 0.00019185760433614054,
      "loss": 0.2272,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 2690,
      "tokens_per_second_per_gpu": 371.44
    },
    {
      "epoch": 0.2684296863349406,
      "grad_norm": 4.753208160400391,
      "learning_rate": 0.00019179545912790207,
      "loss": 0.2826,
      "memory/device_reserved (GiB)": 22.43,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 2700,
      "tokens_per_second_per_gpu": 374.86
    },
    {
      "epoch": 0.2694238703584033,
      "grad_norm": 0.3265855014324188,
      "learning_rate": 0.00019173308780804637,
      "loss": 0.2372,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 2710,
      "tokens_per_second_per_gpu": 320.39
    },
    {
      "epoch": 0.27041805438186606,
      "grad_norm": 0.3575897514820099,
      "learning_rate": 0.00019167049053020712,
      "loss": 0.247,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 2720,
      "tokens_per_second_per_gpu": 312.17
    },
    {
      "epoch": 0.27141223840532885,
      "grad_norm": 0.486605703830719,
      "learning_rate": 0.00019160766744857476,
      "loss": 0.2732,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 2730,
      "tokens_per_second_per_gpu": 388.44
    },
    {
      "epoch": 0.2724064224287916,
      "grad_norm": 0.3815803527832031,
      "learning_rate": 0.00019154461871789572,
      "loss": 0.2733,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.76,
      "memory/max_allocated (GiB)": 19.76,
      "step": 2740,
      "tokens_per_second_per_gpu": 338.38
    },
    {
      "epoch": 0.27340060645225434,
      "grad_norm": 0.3805699050426483,
      "learning_rate": 0.0001914813444934724,
      "loss": 0.2912,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.67,
      "memory/max_allocated (GiB)": 19.67,
      "step": 2750,
      "tokens_per_second_per_gpu": 385.54
    },
    {
      "epoch": 0.2743947904757171,
      "grad_norm": 0.5376086831092834,
      "learning_rate": 0.00019141784493116254,
      "loss": 0.3009,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2760,
      "tokens_per_second_per_gpu": 392.53
    },
    {
      "epoch": 0.2753889744991798,
      "grad_norm": 0.2709028720855713,
      "learning_rate": 0.000191354120187379,
      "loss": 0.3044,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 2770,
      "tokens_per_second_per_gpu": 363.03
    },
    {
      "epoch": 0.27638315852264256,
      "grad_norm": 0.24123673141002655,
      "learning_rate": 0.00019129017041908934,
      "loss": 0.2744,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 2780,
      "tokens_per_second_per_gpu": 351.75
    },
    {
      "epoch": 0.2773773425461053,
      "grad_norm": 0.4202309548854828,
      "learning_rate": 0.00019122599578381532,
      "loss": 0.2764,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2790,
      "tokens_per_second_per_gpu": 387.68
    },
    {
      "epoch": 0.27837152656956804,
      "grad_norm": 0.42621132731437683,
      "learning_rate": 0.00019116159643963262,
      "loss": 0.245,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 2800,
      "tokens_per_second_per_gpu": 351.56
    },
    {
      "epoch": 0.2793657105930308,
      "grad_norm": 0.30440858006477356,
      "learning_rate": 0.00019109697254517048,
      "loss": 0.2809,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2810,
      "tokens_per_second_per_gpu": 382.5
    },
    {
      "epoch": 0.2803598946164935,
      "grad_norm": 0.49908211827278137,
      "learning_rate": 0.00019103212425961111,
      "loss": 0.319,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 2820,
      "tokens_per_second_per_gpu": 402.18
    },
    {
      "epoch": 0.28135407863995626,
      "grad_norm": 0.38030481338500977,
      "learning_rate": 0.00019096705174268967,
      "loss": 0.2392,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 2830,
      "tokens_per_second_per_gpu": 319.29
    },
    {
      "epoch": 0.282348262663419,
      "grad_norm": 0.4159802198410034,
      "learning_rate": 0.00019090175515469344,
      "loss": 0.2962,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 2840,
      "tokens_per_second_per_gpu": 354.51
    },
    {
      "epoch": 0.28334244668688174,
      "grad_norm": 0.47257235646247864,
      "learning_rate": 0.00019083623465646172,
      "loss": 0.2523,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 2850,
      "tokens_per_second_per_gpu": 336.31
    },
    {
      "epoch": 0.2843366307103445,
      "grad_norm": 0.2591971158981323,
      "learning_rate": 0.0001907704904093854,
      "loss": 0.2605,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 2860,
      "tokens_per_second_per_gpu": 386.6
    },
    {
      "epoch": 0.2853308147338072,
      "grad_norm": 0.3341462314128876,
      "learning_rate": 0.00019070452257540638,
      "loss": 0.3024,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 2870,
      "tokens_per_second_per_gpu": 389.19
    },
    {
      "epoch": 0.28632499875726997,
      "grad_norm": 0.3934953808784485,
      "learning_rate": 0.00019063833131701744,
      "loss": 0.2648,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 2880,
      "tokens_per_second_per_gpu": 333.15
    },
    {
      "epoch": 0.2873191827807327,
      "grad_norm": 0.3615911304950714,
      "learning_rate": 0.00019057191679726162,
      "loss": 0.2232,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 2890,
      "tokens_per_second_per_gpu": 326.24
    },
    {
      "epoch": 0.28831336680419545,
      "grad_norm": 0.45740652084350586,
      "learning_rate": 0.00019050527917973192,
      "loss": 0.2467,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 2900,
      "tokens_per_second_per_gpu": 399.74
    },
    {
      "epoch": 0.2893075508276582,
      "grad_norm": 0.5232924818992615,
      "learning_rate": 0.00019043841862857088,
      "loss": 0.292,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 2910,
      "tokens_per_second_per_gpu": 385.0
    },
    {
      "epoch": 0.29030173485112093,
      "grad_norm": 0.3617020547389984,
      "learning_rate": 0.00019037133530847014,
      "loss": 0.2618,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 2920,
      "tokens_per_second_per_gpu": 380.42
    },
    {
      "epoch": 0.29129591887458367,
      "grad_norm": 0.4762667119503021,
      "learning_rate": 0.00019030402938467013,
      "loss": 0.281,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 2930,
      "tokens_per_second_per_gpu": 398.66
    },
    {
      "epoch": 0.2922901028980464,
      "grad_norm": 0.5239278078079224,
      "learning_rate": 0.00019023650102295957,
      "loss": 0.2205,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 2940,
      "tokens_per_second_per_gpu": 317.01
    },
    {
      "epoch": 0.29328428692150915,
      "grad_norm": 0.38952499628067017,
      "learning_rate": 0.00019016875038967507,
      "loss": 0.2618,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 2950,
      "tokens_per_second_per_gpu": 356.78
    },
    {
      "epoch": 0.2942784709449719,
      "grad_norm": 0.4657343924045563,
      "learning_rate": 0.00019010077765170072,
      "loss": 0.2508,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 2960,
      "tokens_per_second_per_gpu": 364.11
    },
    {
      "epoch": 0.29527265496843463,
      "grad_norm": 0.4659232497215271,
      "learning_rate": 0.0001900325829764678,
      "loss": 0.2148,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.69,
      "memory/max_allocated (GiB)": 19.69,
      "step": 2970,
      "tokens_per_second_per_gpu": 387.16
    },
    {
      "epoch": 0.2962668389918974,
      "grad_norm": 0.5370404124259949,
      "learning_rate": 0.0001899641665319542,
      "loss": 0.2857,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 2980,
      "tokens_per_second_per_gpu": 315.91
    },
    {
      "epoch": 0.29726102301536017,
      "grad_norm": 0.3220022916793823,
      "learning_rate": 0.00018989552848668406,
      "loss": 0.2989,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 2990,
      "tokens_per_second_per_gpu": 394.41
    },
    {
      "epoch": 0.2982552070388229,
      "grad_norm": 0.38055410981178284,
      "learning_rate": 0.0001898266690097274,
      "loss": 0.2354,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 3000,
      "tokens_per_second_per_gpu": 319.33
    },
    {
      "epoch": 0.29924939106228565,
      "grad_norm": 0.36663955450057983,
      "learning_rate": 0.00018975758827069968,
      "loss": 0.2437,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3010,
      "tokens_per_second_per_gpu": 326.18
    },
    {
      "epoch": 0.3002435750857484,
      "grad_norm": 0.3347165882587433,
      "learning_rate": 0.00018968828643976135,
      "loss": 0.2016,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3020,
      "tokens_per_second_per_gpu": 363.38
    },
    {
      "epoch": 0.30123775910921113,
      "grad_norm": 0.4647526144981384,
      "learning_rate": 0.0001896187636876175,
      "loss": 0.2421,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 3030,
      "tokens_per_second_per_gpu": 366.11
    },
    {
      "epoch": 0.3022319431326739,
      "grad_norm": 0.35668620467185974,
      "learning_rate": 0.00018954902018551728,
      "loss": 0.28,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3040,
      "tokens_per_second_per_gpu": 333.01
    },
    {
      "epoch": 0.3032261271561366,
      "grad_norm": 0.4070911109447479,
      "learning_rate": 0.00018947905610525374,
      "loss": 0.3329,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 3050,
      "tokens_per_second_per_gpu": 367.0
    },
    {
      "epoch": 0.30422031117959936,
      "grad_norm": 0.4529660642147064,
      "learning_rate": 0.00018940887161916317,
      "loss": 0.2532,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3060,
      "tokens_per_second_per_gpu": 438.51
    },
    {
      "epoch": 0.3052144952030621,
      "grad_norm": 0.323428213596344,
      "learning_rate": 0.0001893384669001248,
      "loss": 0.2717,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 3070,
      "tokens_per_second_per_gpu": 304.06
    },
    {
      "epoch": 0.30620867922652484,
      "grad_norm": 0.33601540327072144,
      "learning_rate": 0.00018926784212156038,
      "loss": 0.2575,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 17.29,
      "memory/max_allocated (GiB)": 17.29,
      "step": 3080,
      "tokens_per_second_per_gpu": 315.78
    },
    {
      "epoch": 0.3072028632499876,
      "grad_norm": 0.4147079586982727,
      "learning_rate": 0.0001891969974574336,
      "loss": 0.2606,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3090,
      "tokens_per_second_per_gpu": 357.02
    },
    {
      "epoch": 0.3081970472734503,
      "grad_norm": 0.5923524498939514,
      "learning_rate": 0.00018912593308224987,
      "loss": 0.2824,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 3100,
      "tokens_per_second_per_gpu": 360.95
    },
    {
      "epoch": 0.30919123129691306,
      "grad_norm": 0.2961815297603607,
      "learning_rate": 0.00018905464917105577,
      "loss": 0.3001,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3110,
      "tokens_per_second_per_gpu": 369.31
    },
    {
      "epoch": 0.3101854153203758,
      "grad_norm": 0.2864610254764557,
      "learning_rate": 0.00018898314589943862,
      "loss": 0.2441,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 3120,
      "tokens_per_second_per_gpu": 383.33
    },
    {
      "epoch": 0.31117959934383854,
      "grad_norm": 0.28341177105903625,
      "learning_rate": 0.00018891142344352611,
      "loss": 0.2402,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 3130,
      "tokens_per_second_per_gpu": 347.6
    },
    {
      "epoch": 0.3121737833673013,
      "grad_norm": 0.3956068158149719,
      "learning_rate": 0.0001888394819799858,
      "loss": 0.2318,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 3140,
      "tokens_per_second_per_gpu": 343.71
    },
    {
      "epoch": 0.313167967390764,
      "grad_norm": 0.3796366751194,
      "learning_rate": 0.00018876732168602472,
      "loss": 0.2047,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3150,
      "tokens_per_second_per_gpu": 353.35
    },
    {
      "epoch": 0.31416215141422676,
      "grad_norm": 0.34916970133781433,
      "learning_rate": 0.00018869494273938893,
      "loss": 0.3128,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 3160,
      "tokens_per_second_per_gpu": 316.54
    },
    {
      "epoch": 0.3151563354376895,
      "grad_norm": 0.3254728317260742,
      "learning_rate": 0.00018862234531836307,
      "loss": 0.3033,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 3170,
      "tokens_per_second_per_gpu": 377.71
    },
    {
      "epoch": 0.31615051946115225,
      "grad_norm": 0.43219825625419617,
      "learning_rate": 0.0001885495296017699,
      "loss": 0.2952,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 3180,
      "tokens_per_second_per_gpu": 314.3
    },
    {
      "epoch": 0.317144703484615,
      "grad_norm": 0.2606908082962036,
      "learning_rate": 0.00018847649576897,
      "loss": 0.2087,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3190,
      "tokens_per_second_per_gpu": 345.1
    },
    {
      "epoch": 0.31813888750807773,
      "grad_norm": 0.46498534083366394,
      "learning_rate": 0.00018840324399986105,
      "loss": 0.2665,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3200,
      "tokens_per_second_per_gpu": 313.16
    },
    {
      "epoch": 0.31913307153154047,
      "grad_norm": 0.35087594389915466,
      "learning_rate": 0.00018832977447487772,
      "loss": 0.2328,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3210,
      "tokens_per_second_per_gpu": 321.23
    },
    {
      "epoch": 0.3201272555550032,
      "grad_norm": 0.46294090151786804,
      "learning_rate": 0.00018825608737499088,
      "loss": 0.2233,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3220,
      "tokens_per_second_per_gpu": 366.88
    },
    {
      "epoch": 0.32112143957846595,
      "grad_norm": 0.35385075211524963,
      "learning_rate": 0.00018818218288170753,
      "loss": 0.2653,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3230,
      "tokens_per_second_per_gpu": 350.06
    },
    {
      "epoch": 0.3221156236019287,
      "grad_norm": 0.4959578812122345,
      "learning_rate": 0.00018810806117706998,
      "loss": 0.2921,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3240,
      "tokens_per_second_per_gpu": 415.69
    },
    {
      "epoch": 0.3231098076253915,
      "grad_norm": 0.4978230893611908,
      "learning_rate": 0.0001880337224436557,
      "loss": 0.3187,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 3250,
      "tokens_per_second_per_gpu": 377.08
    },
    {
      "epoch": 0.32410399164885423,
      "grad_norm": 0.45267730951309204,
      "learning_rate": 0.00018795916686457667,
      "loss": 0.3013,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3260,
      "tokens_per_second_per_gpu": 410.24
    },
    {
      "epoch": 0.32509817567231697,
      "grad_norm": 0.3263493478298187,
      "learning_rate": 0.00018788439462347908,
      "loss": 0.2876,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3270,
      "tokens_per_second_per_gpu": 446.75
    },
    {
      "epoch": 0.3260923596957797,
      "grad_norm": 0.4769454896450043,
      "learning_rate": 0.00018780940590454277,
      "loss": 0.3504,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 3280,
      "tokens_per_second_per_gpu": 301.09
    },
    {
      "epoch": 0.32708654371924245,
      "grad_norm": 0.21503253281116486,
      "learning_rate": 0.00018773420089248074,
      "loss": 0.2291,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 3290,
      "tokens_per_second_per_gpu": 375.18
    },
    {
      "epoch": 0.3280807277427052,
      "grad_norm": 0.49010154604911804,
      "learning_rate": 0.00018765877977253888,
      "loss": 0.2674,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3300,
      "tokens_per_second_per_gpu": 349.74
    },
    {
      "epoch": 0.32907491176616793,
      "grad_norm": 0.3158112168312073,
      "learning_rate": 0.00018758314273049532,
      "loss": 0.2975,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3310,
      "tokens_per_second_per_gpu": 336.25
    },
    {
      "epoch": 0.3300690957896307,
      "grad_norm": 0.428846150636673,
      "learning_rate": 0.0001875072899526601,
      "loss": 0.2618,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3320,
      "tokens_per_second_per_gpu": 284.44
    },
    {
      "epoch": 0.3310632798130934,
      "grad_norm": 0.5033756494522095,
      "learning_rate": 0.00018743122162587464,
      "loss": 0.2926,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 3330,
      "tokens_per_second_per_gpu": 419.59
    },
    {
      "epoch": 0.33205746383655615,
      "grad_norm": 0.17825426161289215,
      "learning_rate": 0.0001873549379375113,
      "loss": 0.2464,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 3340,
      "tokens_per_second_per_gpu": 342.16
    },
    {
      "epoch": 0.3330516478600189,
      "grad_norm": 0.34919679164886475,
      "learning_rate": 0.00018727843907547293,
      "loss": 0.2729,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 3350,
      "tokens_per_second_per_gpu": 390.88
    },
    {
      "epoch": 0.33404583188348164,
      "grad_norm": 0.37125077843666077,
      "learning_rate": 0.00018720172522819243,
      "loss": 0.3117,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3360,
      "tokens_per_second_per_gpu": 394.73
    },
    {
      "epoch": 0.3350400159069444,
      "grad_norm": 0.31769683957099915,
      "learning_rate": 0.00018712479658463215,
      "loss": 0.2702,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3370,
      "tokens_per_second_per_gpu": 424.16
    },
    {
      "epoch": 0.3360341999304071,
      "grad_norm": 0.398548424243927,
      "learning_rate": 0.00018704765333428367,
      "loss": 0.1966,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3380,
      "tokens_per_second_per_gpu": 310.88
    },
    {
      "epoch": 0.33702838395386986,
      "grad_norm": 0.24172648787498474,
      "learning_rate": 0.00018697029566716705,
      "loss": 0.2189,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3390,
      "tokens_per_second_per_gpu": 311.63
    },
    {
      "epoch": 0.3380225679773326,
      "grad_norm": 0.46132785081863403,
      "learning_rate": 0.00018689272377383064,
      "loss": 0.3093,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 3400,
      "tokens_per_second_per_gpu": 411.8
    },
    {
      "epoch": 0.33901675200079534,
      "grad_norm": 0.3627679944038391,
      "learning_rate": 0.00018681493784535036,
      "loss": 0.2558,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 3410,
      "tokens_per_second_per_gpu": 319.21
    },
    {
      "epoch": 0.3400109360242581,
      "grad_norm": 1.1992244720458984,
      "learning_rate": 0.00018673693807332945,
      "loss": 0.228,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3420,
      "tokens_per_second_per_gpu": 354.78
    },
    {
      "epoch": 0.3410051200477208,
      "grad_norm": 0.26419004797935486,
      "learning_rate": 0.00018665872464989773,
      "loss": 0.1874,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3430,
      "tokens_per_second_per_gpu": 369.02
    },
    {
      "epoch": 0.34199930407118356,
      "grad_norm": 0.3501751720905304,
      "learning_rate": 0.00018658029776771152,
      "loss": 0.2231,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 3440,
      "tokens_per_second_per_gpu": 358.64
    },
    {
      "epoch": 0.3429934880946463,
      "grad_norm": 0.4123583137989044,
      "learning_rate": 0.0001865016576199527,
      "loss": 0.2456,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3450,
      "tokens_per_second_per_gpu": 399.31
    },
    {
      "epoch": 0.34398767211810904,
      "grad_norm": 0.4507691264152527,
      "learning_rate": 0.00018642280440032863,
      "loss": 0.2716,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 3460,
      "tokens_per_second_per_gpu": 384.1
    },
    {
      "epoch": 0.3449818561415718,
      "grad_norm": 0.43500733375549316,
      "learning_rate": 0.00018634373830307146,
      "loss": 0.2352,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 3470,
      "tokens_per_second_per_gpu": 319.67
    },
    {
      "epoch": 0.3459760401650345,
      "grad_norm": 0.40590760111808777,
      "learning_rate": 0.00018626445952293766,
      "loss": 0.2623,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 3480,
      "tokens_per_second_per_gpu": 335.37
    },
    {
      "epoch": 0.34697022418849727,
      "grad_norm": 0.2494644969701767,
      "learning_rate": 0.00018618496825520767,
      "loss": 0.2245,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3490,
      "tokens_per_second_per_gpu": 347.95
    },
    {
      "epoch": 0.34796440821196,
      "grad_norm": 0.47100207209587097,
      "learning_rate": 0.00018610526469568526,
      "loss": 0.2775,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 3500,
      "tokens_per_second_per_gpu": 393.01
    },
    {
      "epoch": 0.3489585922354228,
      "grad_norm": 0.5600543022155762,
      "learning_rate": 0.00018602534904069712,
      "loss": 0.3007,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 3510,
      "tokens_per_second_per_gpu": 399.51
    },
    {
      "epoch": 0.34995277625888555,
      "grad_norm": 0.25791943073272705,
      "learning_rate": 0.00018594522148709244,
      "loss": 0.2134,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 3520,
      "tokens_per_second_per_gpu": 341.75
    },
    {
      "epoch": 0.3509469602823483,
      "grad_norm": 0.2849276661872864,
      "learning_rate": 0.00018586488223224228,
      "loss": 0.1919,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 3530,
      "tokens_per_second_per_gpu": 369.12
    },
    {
      "epoch": 0.351941144305811,
      "grad_norm": 0.32681041955947876,
      "learning_rate": 0.00018578433147403925,
      "loss": 0.2192,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 3540,
      "tokens_per_second_per_gpu": 309.14
    },
    {
      "epoch": 0.35293532832927377,
      "grad_norm": 0.5691526532173157,
      "learning_rate": 0.00018570356941089686,
      "loss": 0.2775,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3550,
      "tokens_per_second_per_gpu": 357.84
    },
    {
      "epoch": 0.3539295123527365,
      "grad_norm": 0.35383832454681396,
      "learning_rate": 0.00018562259624174915,
      "loss": 0.2285,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3560,
      "tokens_per_second_per_gpu": 341.0
    },
    {
      "epoch": 0.35492369637619925,
      "grad_norm": 0.4128180742263794,
      "learning_rate": 0.00018554141216605016,
      "loss": 0.2216,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3570,
      "tokens_per_second_per_gpu": 310.04
    },
    {
      "epoch": 0.355917880399662,
      "grad_norm": 0.4811583459377289,
      "learning_rate": 0.00018546001738377338,
      "loss": 0.3354,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3580,
      "tokens_per_second_per_gpu": 446.97
    },
    {
      "epoch": 0.35691206442312473,
      "grad_norm": 0.4148445725440979,
      "learning_rate": 0.0001853784120954114,
      "loss": 0.216,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 3590,
      "tokens_per_second_per_gpu": 327.38
    },
    {
      "epoch": 0.35790624844658747,
      "grad_norm": 0.5083706378936768,
      "learning_rate": 0.0001852965965019753,
      "loss": 0.2762,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3600,
      "tokens_per_second_per_gpu": 371.89
    },
    {
      "epoch": 0.3589004324700502,
      "grad_norm": 0.4946528673171997,
      "learning_rate": 0.00018521457080499418,
      "loss": 0.2455,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 3610,
      "tokens_per_second_per_gpu": 361.51
    },
    {
      "epoch": 0.35989461649351295,
      "grad_norm": 0.4880548417568207,
      "learning_rate": 0.00018513233520651466,
      "loss": 0.2299,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3620,
      "tokens_per_second_per_gpu": 310.24
    },
    {
      "epoch": 0.3608888005169757,
      "grad_norm": 0.18661662936210632,
      "learning_rate": 0.00018504988990910036,
      "loss": 0.2325,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 3630,
      "tokens_per_second_per_gpu": 360.45
    },
    {
      "epoch": 0.36188298454043843,
      "grad_norm": 0.49652183055877686,
      "learning_rate": 0.00018496723511583153,
      "loss": 0.2312,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 3640,
      "tokens_per_second_per_gpu": 347.3
    },
    {
      "epoch": 0.3628771685639012,
      "grad_norm": 0.35343873500823975,
      "learning_rate": 0.0001848843710303044,
      "loss": 0.154,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3650,
      "tokens_per_second_per_gpu": 334.16
    },
    {
      "epoch": 0.3638713525873639,
      "grad_norm": 0.5269297361373901,
      "learning_rate": 0.0001848012978566307,
      "loss": 0.2677,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 3660,
      "tokens_per_second_per_gpu": 343.01
    },
    {
      "epoch": 0.36486553661082666,
      "grad_norm": 0.4809168875217438,
      "learning_rate": 0.00018471801579943717,
      "loss": 0.3083,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3670,
      "tokens_per_second_per_gpu": 359.61
    },
    {
      "epoch": 0.3658597206342894,
      "grad_norm": 0.4312402904033661,
      "learning_rate": 0.0001846345250638652,
      "loss": 0.2711,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 3680,
      "tokens_per_second_per_gpu": 343.63
    },
    {
      "epoch": 0.36685390465775214,
      "grad_norm": 0.44654685258865356,
      "learning_rate": 0.0001845508258555701,
      "loss": 0.2629,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3690,
      "tokens_per_second_per_gpu": 380.56
    },
    {
      "epoch": 0.3678480886812149,
      "grad_norm": 0.19989164173603058,
      "learning_rate": 0.00018446691838072067,
      "loss": 0.2451,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 3700,
      "tokens_per_second_per_gpu": 316.7
    },
    {
      "epoch": 0.3688422727046776,
      "grad_norm": 0.268655925989151,
      "learning_rate": 0.00018438280284599877,
      "loss": 0.2172,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 3710,
      "tokens_per_second_per_gpu": 270.99
    },
    {
      "epoch": 0.36983645672814036,
      "grad_norm": 0.3657344579696655,
      "learning_rate": 0.00018429847945859872,
      "loss": 0.2505,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 3720,
      "tokens_per_second_per_gpu": 319.93
    },
    {
      "epoch": 0.3708306407516031,
      "grad_norm": 0.40177711844444275,
      "learning_rate": 0.00018421394842622695,
      "loss": 0.2462,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3730,
      "tokens_per_second_per_gpu": 344.24
    },
    {
      "epoch": 0.37182482477506584,
      "grad_norm": 0.48767533898353577,
      "learning_rate": 0.00018412920995710113,
      "loss": 0.2827,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3740,
      "tokens_per_second_per_gpu": 380.45
    },
    {
      "epoch": 0.3728190087985286,
      "grad_norm": 0.45828619599342346,
      "learning_rate": 0.00018404426425995007,
      "loss": 0.2355,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 3750,
      "tokens_per_second_per_gpu": 388.78
    },
    {
      "epoch": 0.3738131928219913,
      "grad_norm": 0.49931567907333374,
      "learning_rate": 0.000183959111544013,
      "loss": 0.2813,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 3760,
      "tokens_per_second_per_gpu": 361.55
    },
    {
      "epoch": 0.3748073768454541,
      "grad_norm": 0.3232674300670624,
      "learning_rate": 0.00018387375201903903,
      "loss": 0.2488,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3770,
      "tokens_per_second_per_gpu": 326.01
    },
    {
      "epoch": 0.37580156086891686,
      "grad_norm": 0.41870149970054626,
      "learning_rate": 0.0001837881858952867,
      "loss": 0.3117,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3780,
      "tokens_per_second_per_gpu": 356.89
    },
    {
      "epoch": 0.3767957448923796,
      "grad_norm": 0.396383672952652,
      "learning_rate": 0.00018370241338352348,
      "loss": 0.3046,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 3790,
      "tokens_per_second_per_gpu": 397.36
    },
    {
      "epoch": 0.37778992891584234,
      "grad_norm": 0.33363988995552063,
      "learning_rate": 0.00018361643469502517,
      "loss": 0.2074,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3800,
      "tokens_per_second_per_gpu": 320.36
    },
    {
      "epoch": 0.3787841129393051,
      "grad_norm": 0.34591570496559143,
      "learning_rate": 0.00018353025004157552,
      "loss": 0.2449,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 3810,
      "tokens_per_second_per_gpu": 370.18
    },
    {
      "epoch": 0.3797782969627678,
      "grad_norm": 0.4369080066680908,
      "learning_rate": 0.00018344385963546547,
      "loss": 0.2017,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 3820,
      "tokens_per_second_per_gpu": 346.43
    },
    {
      "epoch": 0.38077248098623057,
      "grad_norm": 0.4190782308578491,
      "learning_rate": 0.00018335726368949286,
      "loss": 0.2987,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3830,
      "tokens_per_second_per_gpu": 381.05
    },
    {
      "epoch": 0.3817666650096933,
      "grad_norm": 0.4989373981952667,
      "learning_rate": 0.00018327046241696184,
      "loss": 0.2992,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 3840,
      "tokens_per_second_per_gpu": 363.52
    },
    {
      "epoch": 0.38276084903315605,
      "grad_norm": 0.3604322671890259,
      "learning_rate": 0.00018318345603168226,
      "loss": 0.2311,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 3850,
      "tokens_per_second_per_gpu": 286.25
    },
    {
      "epoch": 0.3837550330566188,
      "grad_norm": 0.303365021944046,
      "learning_rate": 0.00018309624474796926,
      "loss": 0.1952,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 3860,
      "tokens_per_second_per_gpu": 381.31
    },
    {
      "epoch": 0.38474921708008153,
      "grad_norm": 0.5674360990524292,
      "learning_rate": 0.00018300882878064266,
      "loss": 0.2694,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.82,
      "memory/max_allocated (GiB)": 18.82,
      "step": 3870,
      "tokens_per_second_per_gpu": 346.22
    },
    {
      "epoch": 0.38574340110354427,
      "grad_norm": 0.4402889609336853,
      "learning_rate": 0.00018292120834502643,
      "loss": 0.2825,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 3880,
      "tokens_per_second_per_gpu": 339.23
    },
    {
      "epoch": 0.386737585127007,
      "grad_norm": 0.3922783136367798,
      "learning_rate": 0.00018283338365694825,
      "loss": 0.2294,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3890,
      "tokens_per_second_per_gpu": 289.83
    },
    {
      "epoch": 0.38773176915046975,
      "grad_norm": 0.6003592014312744,
      "learning_rate": 0.00018274535493273893,
      "loss": 0.2244,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3900,
      "tokens_per_second_per_gpu": 350.12
    },
    {
      "epoch": 0.3887259531739325,
      "grad_norm": 0.436212956905365,
      "learning_rate": 0.00018265712238923175,
      "loss": 0.2341,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3910,
      "tokens_per_second_per_gpu": 343.17
    },
    {
      "epoch": 0.38972013719739523,
      "grad_norm": 0.2501852810382843,
      "learning_rate": 0.00018256868624376215,
      "loss": 0.2647,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 3920,
      "tokens_per_second_per_gpu": 396.07
    },
    {
      "epoch": 0.390714321220858,
      "grad_norm": 0.36171767115592957,
      "learning_rate": 0.00018248004671416704,
      "loss": 0.2664,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.69,
      "memory/max_allocated (GiB)": 19.69,
      "step": 3930,
      "tokens_per_second_per_gpu": 426.89
    },
    {
      "epoch": 0.3917085052443207,
      "grad_norm": 0.47077932953834534,
      "learning_rate": 0.00018239120401878432,
      "loss": 0.3584,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 3940,
      "tokens_per_second_per_gpu": 452.63
    },
    {
      "epoch": 0.39270268926778346,
      "grad_norm": 0.413924902677536,
      "learning_rate": 0.00018230215837645232,
      "loss": 0.2715,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3950,
      "tokens_per_second_per_gpu": 424.3
    },
    {
      "epoch": 0.3936968732912462,
      "grad_norm": 0.40877413749694824,
      "learning_rate": 0.00018221291000650928,
      "loss": 0.2855,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 3960,
      "tokens_per_second_per_gpu": 406.29
    },
    {
      "epoch": 0.39469105731470894,
      "grad_norm": 0.4080711007118225,
      "learning_rate": 0.0001821234591287928,
      "loss": 0.1893,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 3970,
      "tokens_per_second_per_gpu": 347.04
    },
    {
      "epoch": 0.3956852413381717,
      "grad_norm": 0.2622958719730377,
      "learning_rate": 0.00018203380596363932,
      "loss": 0.2328,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 3980,
      "tokens_per_second_per_gpu": 323.13
    },
    {
      "epoch": 0.3966794253616344,
      "grad_norm": 0.32758989930152893,
      "learning_rate": 0.0001819439507318835,
      "loss": 0.196,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 3990,
      "tokens_per_second_per_gpu": 255.43
    },
    {
      "epoch": 0.39767360938509716,
      "grad_norm": 0.4135094881057739,
      "learning_rate": 0.00018185389365485774,
      "loss": 0.2874,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 4000,
      "tokens_per_second_per_gpu": 352.42
    },
    {
      "epoch": 0.3986677934085599,
      "grad_norm": 0.4753275215625763,
      "learning_rate": 0.00018176363495439173,
      "loss": 0.2796,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 4010,
      "tokens_per_second_per_gpu": 300.74
    },
    {
      "epoch": 0.39966197743202264,
      "grad_norm": 0.41000860929489136,
      "learning_rate": 0.00018167317485281168,
      "loss": 0.3278,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 4020,
      "tokens_per_second_per_gpu": 381.42
    },
    {
      "epoch": 0.40065616145548544,
      "grad_norm": 0.3132636845111847,
      "learning_rate": 0.00018158251357293996,
      "loss": 0.2514,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4030,
      "tokens_per_second_per_gpu": 340.93
    },
    {
      "epoch": 0.4016503454789482,
      "grad_norm": 0.3330332338809967,
      "learning_rate": 0.00018149165133809442,
      "loss": 0.219,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 4040,
      "tokens_per_second_per_gpu": 331.95
    },
    {
      "epoch": 0.4026445295024109,
      "grad_norm": 0.5818430781364441,
      "learning_rate": 0.000181400588372088,
      "loss": 0.3451,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 4050,
      "tokens_per_second_per_gpu": 433.43
    },
    {
      "epoch": 0.40363871352587366,
      "grad_norm": 0.4116646945476532,
      "learning_rate": 0.00018130932489922804,
      "loss": 0.1907,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 4060,
      "tokens_per_second_per_gpu": 326.32
    },
    {
      "epoch": 0.4046328975493364,
      "grad_norm": 0.31805434823036194,
      "learning_rate": 0.0001812178611443157,
      "loss": 0.18,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 4070,
      "tokens_per_second_per_gpu": 284.15
    },
    {
      "epoch": 0.40562708157279914,
      "grad_norm": 0.5397758483886719,
      "learning_rate": 0.0001811261973326456,
      "loss": 0.2596,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 4080,
      "tokens_per_second_per_gpu": 369.06
    },
    {
      "epoch": 0.4066212655962619,
      "grad_norm": 0.4468703866004944,
      "learning_rate": 0.00018103433369000502,
      "loss": 0.2464,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 4090,
      "tokens_per_second_per_gpu": 343.56
    },
    {
      "epoch": 0.4076154496197246,
      "grad_norm": 0.3228696286678314,
      "learning_rate": 0.0001809422704426736,
      "loss": 0.2,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 4100,
      "tokens_per_second_per_gpu": 301.0
    },
    {
      "epoch": 0.40860963364318736,
      "grad_norm": 0.5811059474945068,
      "learning_rate": 0.00018085000781742252,
      "loss": 0.2642,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 4110,
      "tokens_per_second_per_gpu": 366.31
    },
    {
      "epoch": 0.4096038176666501,
      "grad_norm": 0.6858221292495728,
      "learning_rate": 0.00018075754604151415,
      "loss": 0.2658,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.68,
      "memory/max_allocated (GiB)": 19.68,
      "step": 4120,
      "tokens_per_second_per_gpu": 340.12
    },
    {
      "epoch": 0.41059800169011285,
      "grad_norm": 0.3095184862613678,
      "learning_rate": 0.00018066488534270142,
      "loss": 0.2542,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 4130,
      "tokens_per_second_per_gpu": 399.71
    },
    {
      "epoch": 0.4115921857135756,
      "grad_norm": 0.5910835266113281,
      "learning_rate": 0.0001805720259492271,
      "loss": 0.289,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 4140,
      "tokens_per_second_per_gpu": 366.05
    },
    {
      "epoch": 0.4125863697370383,
      "grad_norm": 0.3704458773136139,
      "learning_rate": 0.00018047896808982364,
      "loss": 0.2581,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 4150,
      "tokens_per_second_per_gpu": 348.57
    },
    {
      "epoch": 0.41358055376050107,
      "grad_norm": 0.4034087359905243,
      "learning_rate": 0.00018038571199371215,
      "loss": 0.2207,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 4160,
      "tokens_per_second_per_gpu": 323.67
    },
    {
      "epoch": 0.4145747377839638,
      "grad_norm": 0.5003114342689514,
      "learning_rate": 0.0001802922578906021,
      "loss": 0.2927,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 4170,
      "tokens_per_second_per_gpu": 325.75
    },
    {
      "epoch": 0.41556892180742655,
      "grad_norm": 0.29613539576530457,
      "learning_rate": 0.0001801986060106907,
      "loss": 0.2906,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4180,
      "tokens_per_second_per_gpu": 335.06
    },
    {
      "epoch": 0.4165631058308893,
      "grad_norm": 0.49740076065063477,
      "learning_rate": 0.00018010475658466235,
      "loss": 0.2853,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4190,
      "tokens_per_second_per_gpu": 420.46
    },
    {
      "epoch": 0.41755728985435203,
      "grad_norm": 0.32316842675209045,
      "learning_rate": 0.000180010709843688,
      "loss": 0.1658,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 4200,
      "tokens_per_second_per_gpu": 288.98
    },
    {
      "epoch": 0.4185514738778148,
      "grad_norm": 0.29272300004959106,
      "learning_rate": 0.00017991646601942467,
      "loss": 0.2719,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.23,
      "memory/max_allocated (GiB)": 19.23,
      "step": 4210,
      "tokens_per_second_per_gpu": 364.92
    },
    {
      "epoch": 0.4195456579012775,
      "grad_norm": 0.5684819221496582,
      "learning_rate": 0.0001798220253440148,
      "loss": 0.2755,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 4220,
      "tokens_per_second_per_gpu": 317.91
    },
    {
      "epoch": 0.42053984192474025,
      "grad_norm": 0.4488314986228943,
      "learning_rate": 0.00017972738805008574,
      "loss": 0.2131,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 4230,
      "tokens_per_second_per_gpu": 328.07
    },
    {
      "epoch": 0.421534025948203,
      "grad_norm": 0.3249848484992981,
      "learning_rate": 0.0001796325543707491,
      "loss": 0.2987,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 4240,
      "tokens_per_second_per_gpu": 421.46
    },
    {
      "epoch": 0.42252820997166574,
      "grad_norm": 0.6481621265411377,
      "learning_rate": 0.00017953752453960038,
      "loss": 0.2498,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.43,
      "memory/max_allocated (GiB)": 20.43,
      "step": 4250,
      "tokens_per_second_per_gpu": 363.83
    },
    {
      "epoch": 0.4235223939951285,
      "grad_norm": 0.3045104146003723,
      "learning_rate": 0.00017944229879071806,
      "loss": 0.2295,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 4260,
      "tokens_per_second_per_gpu": 398.66
    },
    {
      "epoch": 0.4245165780185912,
      "grad_norm": 0.32762956619262695,
      "learning_rate": 0.0001793468773586633,
      "loss": 0.2406,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 4270,
      "tokens_per_second_per_gpu": 399.72
    },
    {
      "epoch": 0.42551076204205396,
      "grad_norm": 0.6278170347213745,
      "learning_rate": 0.00017925126047847924,
      "loss": 0.2523,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 4280,
      "tokens_per_second_per_gpu": 324.65
    },
    {
      "epoch": 0.42650494606551675,
      "grad_norm": 0.45905986428260803,
      "learning_rate": 0.00017915544838569052,
      "loss": 0.2615,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4290,
      "tokens_per_second_per_gpu": 353.66
    },
    {
      "epoch": 0.4274991300889795,
      "grad_norm": 0.48581770062446594,
      "learning_rate": 0.00017905944131630253,
      "loss": 0.2519,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 4300,
      "tokens_per_second_per_gpu": 321.73
    },
    {
      "epoch": 0.42849331411244224,
      "grad_norm": 0.49877023696899414,
      "learning_rate": 0.00017896323950680098,
      "loss": 0.2382,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 4310,
      "tokens_per_second_per_gpu": 353.68
    },
    {
      "epoch": 0.429487498135905,
      "grad_norm": 0.580008327960968,
      "learning_rate": 0.00017886684319415127,
      "loss": 0.2478,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 4320,
      "tokens_per_second_per_gpu": 368.72
    },
    {
      "epoch": 0.4304816821593677,
      "grad_norm": 0.3998342454433441,
      "learning_rate": 0.00017877025261579788,
      "loss": 0.2202,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 4330,
      "tokens_per_second_per_gpu": 395.43
    },
    {
      "epoch": 0.43147586618283046,
      "grad_norm": 0.38088393211364746,
      "learning_rate": 0.00017867346800966383,
      "loss": 0.2521,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 4340,
      "tokens_per_second_per_gpu": 427.5
    },
    {
      "epoch": 0.4324700502062932,
      "grad_norm": 0.3859814703464508,
      "learning_rate": 0.00017857648961415004,
      "loss": 0.2353,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 4350,
      "tokens_per_second_per_gpu": 376.77
    },
    {
      "epoch": 0.43346423422975594,
      "grad_norm": 0.34923896193504333,
      "learning_rate": 0.00017847931766813482,
      "loss": 0.2567,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 4360,
      "tokens_per_second_per_gpu": 368.67
    },
    {
      "epoch": 0.4344584182532187,
      "grad_norm": 0.34750089049339294,
      "learning_rate": 0.0001783819524109732,
      "loss": 0.2291,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 4370,
      "tokens_per_second_per_gpu": 379.66
    },
    {
      "epoch": 0.4354526022766814,
      "grad_norm": 0.36663779616355896,
      "learning_rate": 0.0001782843940824964,
      "loss": 0.2866,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 4380,
      "tokens_per_second_per_gpu": 411.45
    },
    {
      "epoch": 0.43644678630014416,
      "grad_norm": 0.3929060697555542,
      "learning_rate": 0.00017818664292301118,
      "loss": 0.2563,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 4390,
      "tokens_per_second_per_gpu": 381.12
    },
    {
      "epoch": 0.4374409703236069,
      "grad_norm": 0.4182446599006653,
      "learning_rate": 0.0001780886991732993,
      "loss": 0.2268,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 4400,
      "tokens_per_second_per_gpu": 297.67
    },
    {
      "epoch": 0.43843515434706964,
      "grad_norm": 0.5998858213424683,
      "learning_rate": 0.00017799056307461696,
      "loss": 0.2629,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 4410,
      "tokens_per_second_per_gpu": 344.26
    },
    {
      "epoch": 0.4394293383705324,
      "grad_norm": 0.4282694160938263,
      "learning_rate": 0.0001778922348686941,
      "loss": 0.2523,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 4420,
      "tokens_per_second_per_gpu": 328.11
    },
    {
      "epoch": 0.4404235223939951,
      "grad_norm": 0.5925072431564331,
      "learning_rate": 0.00017779371479773382,
      "loss": 0.27,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 4430,
      "tokens_per_second_per_gpu": 336.68
    },
    {
      "epoch": 0.44141770641745787,
      "grad_norm": 0.5149052739143372,
      "learning_rate": 0.00017769500310441192,
      "loss": 0.3033,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 4440,
      "tokens_per_second_per_gpu": 370.47
    },
    {
      "epoch": 0.4424118904409206,
      "grad_norm": 0.418197363615036,
      "learning_rate": 0.00017759610003187617,
      "loss": 0.2193,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 4450,
      "tokens_per_second_per_gpu": 352.55
    },
    {
      "epoch": 0.44340607446438335,
      "grad_norm": 0.4415562152862549,
      "learning_rate": 0.00017749700582374574,
      "loss": 0.1978,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 4460,
      "tokens_per_second_per_gpu": 343.08
    },
    {
      "epoch": 0.4444002584878461,
      "grad_norm": 0.32262691855430603,
      "learning_rate": 0.0001773977207241106,
      "loss": 0.2448,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 4470,
      "tokens_per_second_per_gpu": 319.4
    },
    {
      "epoch": 0.44539444251130883,
      "grad_norm": 0.49002590775489807,
      "learning_rate": 0.00017729824497753093,
      "loss": 0.2772,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 4480,
      "tokens_per_second_per_gpu": 331.65
    },
    {
      "epoch": 0.44638862653477157,
      "grad_norm": 0.4270131587982178,
      "learning_rate": 0.0001771985788290365,
      "loss": 0.2557,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 4490,
      "tokens_per_second_per_gpu": 300.31
    },
    {
      "epoch": 0.4473828105582343,
      "grad_norm": 0.5524002909660339,
      "learning_rate": 0.00017709872252412616,
      "loss": 0.2696,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4500,
      "tokens_per_second_per_gpu": 348.17
    },
    {
      "epoch": 0.44837699458169705,
      "grad_norm": 0.32532012462615967,
      "learning_rate": 0.00017699867630876703,
      "loss": 0.1997,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 4510,
      "tokens_per_second_per_gpu": 315.46
    },
    {
      "epoch": 0.4493711786051598,
      "grad_norm": 0.49136292934417725,
      "learning_rate": 0.0001768984404293941,
      "loss": 0.2824,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 4520,
      "tokens_per_second_per_gpu": 391.14
    },
    {
      "epoch": 0.45036536262862253,
      "grad_norm": 0.43822959065437317,
      "learning_rate": 0.00017679801513290956,
      "loss": 0.2931,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 4530,
      "tokens_per_second_per_gpu": 382.92
    },
    {
      "epoch": 0.4513595466520853,
      "grad_norm": 0.448585569858551,
      "learning_rate": 0.00017669740066668214,
      "loss": 0.2444,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 4540,
      "tokens_per_second_per_gpu": 364.48
    },
    {
      "epoch": 0.45235373067554807,
      "grad_norm": 0.3125511407852173,
      "learning_rate": 0.0001765965972785465,
      "loss": 0.2227,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 4550,
      "tokens_per_second_per_gpu": 360.51
    },
    {
      "epoch": 0.4533479146990108,
      "grad_norm": 0.6492682695388794,
      "learning_rate": 0.00017649560521680266,
      "loss": 0.3157,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 4560,
      "tokens_per_second_per_gpu": 361.61
    },
    {
      "epoch": 0.45434209872247355,
      "grad_norm": 0.23529411852359772,
      "learning_rate": 0.0001763944247302155,
      "loss": 0.2644,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 4570,
      "tokens_per_second_per_gpu": 333.05
    },
    {
      "epoch": 0.4553362827459363,
      "grad_norm": 0.37334245443344116,
      "learning_rate": 0.00017629305606801387,
      "loss": 0.1995,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4580,
      "tokens_per_second_per_gpu": 380.12
    },
    {
      "epoch": 0.45633046676939903,
      "grad_norm": 0.26320332288742065,
      "learning_rate": 0.00017619149947989028,
      "loss": 0.201,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 4590,
      "tokens_per_second_per_gpu": 333.8
    },
    {
      "epoch": 0.4573246507928618,
      "grad_norm": 0.4711815416812897,
      "learning_rate": 0.000176089755216,
      "loss": 0.3037,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4600,
      "tokens_per_second_per_gpu": 398.88
    },
    {
      "epoch": 0.4583188348163245,
      "grad_norm": 0.3597434461116791,
      "learning_rate": 0.0001759878235269607,
      "loss": 0.2393,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4610,
      "tokens_per_second_per_gpu": 340.42
    },
    {
      "epoch": 0.45931301883978726,
      "grad_norm": 0.5157446265220642,
      "learning_rate": 0.00017588570466385166,
      "loss": 0.314,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 4620,
      "tokens_per_second_per_gpu": 374.07
    },
    {
      "epoch": 0.46030720286325,
      "grad_norm": 0.4747403562068939,
      "learning_rate": 0.0001757833988782132,
      "loss": 0.2606,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.47,
      "memory/max_allocated (GiB)": 21.47,
      "step": 4630,
      "tokens_per_second_per_gpu": 333.75
    },
    {
      "epoch": 0.46130138688671274,
      "grad_norm": 0.45278453826904297,
      "learning_rate": 0.00017568090642204612,
      "loss": 0.2106,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 4640,
      "tokens_per_second_per_gpu": 352.11
    },
    {
      "epoch": 0.4622955709101755,
      "grad_norm": 0.7563058137893677,
      "learning_rate": 0.00017557822754781102,
      "loss": 0.2457,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 4650,
      "tokens_per_second_per_gpu": 370.21
    },
    {
      "epoch": 0.4632897549336382,
      "grad_norm": 0.27436432242393494,
      "learning_rate": 0.00017547536250842765,
      "loss": 0.2659,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 4660,
      "tokens_per_second_per_gpu": 347.64
    },
    {
      "epoch": 0.46428393895710096,
      "grad_norm": 0.3469400107860565,
      "learning_rate": 0.00017537231155727428,
      "loss": 0.2744,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 4670,
      "tokens_per_second_per_gpu": 354.35
    },
    {
      "epoch": 0.4652781229805637,
      "grad_norm": 0.29021984338760376,
      "learning_rate": 0.0001752690749481873,
      "loss": 0.236,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 4680,
      "tokens_per_second_per_gpu": 357.93
    },
    {
      "epoch": 0.46627230700402644,
      "grad_norm": 0.28982868790626526,
      "learning_rate": 0.00017516565293546025,
      "loss": 0.2694,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 4690,
      "tokens_per_second_per_gpu": 433.34
    },
    {
      "epoch": 0.4672664910274892,
      "grad_norm": 0.39744624495506287,
      "learning_rate": 0.00017506204577384337,
      "loss": 0.2209,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 4700,
      "tokens_per_second_per_gpu": 380.58
    },
    {
      "epoch": 0.4682606750509519,
      "grad_norm": 0.14510154724121094,
      "learning_rate": 0.00017495825371854302,
      "loss": 0.2147,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 4710,
      "tokens_per_second_per_gpu": 342.34
    },
    {
      "epoch": 0.46925485907441467,
      "grad_norm": 0.26011091470718384,
      "learning_rate": 0.000174854277025221,
      "loss": 0.2411,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 4720,
      "tokens_per_second_per_gpu": 395.32
    },
    {
      "epoch": 0.4702490430978774,
      "grad_norm": 0.5020186901092529,
      "learning_rate": 0.00017475011594999385,
      "loss": 0.2466,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 4730,
      "tokens_per_second_per_gpu": 385.96
    },
    {
      "epoch": 0.47124322712134015,
      "grad_norm": 0.5160934925079346,
      "learning_rate": 0.0001746457707494323,
      "loss": 0.3094,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 4740,
      "tokens_per_second_per_gpu": 343.47
    },
    {
      "epoch": 0.4722374111448029,
      "grad_norm": 0.43567317724227905,
      "learning_rate": 0.00017454124168056066,
      "loss": 0.2324,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 4750,
      "tokens_per_second_per_gpu": 346.39
    },
    {
      "epoch": 0.47323159516826563,
      "grad_norm": 0.39488813281059265,
      "learning_rate": 0.0001744365290008561,
      "loss": 0.1983,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 4760,
      "tokens_per_second_per_gpu": 363.53
    },
    {
      "epoch": 0.47422577919172837,
      "grad_norm": 0.2595832943916321,
      "learning_rate": 0.00017433163296824808,
      "loss": 0.2783,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 4770,
      "tokens_per_second_per_gpu": 431.69
    },
    {
      "epoch": 0.4752199632151911,
      "grad_norm": 0.3657257556915283,
      "learning_rate": 0.00017422655384111772,
      "loss": 0.2223,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 4780,
      "tokens_per_second_per_gpu": 333.82
    },
    {
      "epoch": 0.47621414723865385,
      "grad_norm": 0.3269219994544983,
      "learning_rate": 0.00017412129187829712,
      "loss": 0.2042,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 4790,
      "tokens_per_second_per_gpu": 356.0
    },
    {
      "epoch": 0.4772083312621166,
      "grad_norm": 0.35581299662590027,
      "learning_rate": 0.00017401584733906872,
      "loss": 0.216,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 4800,
      "tokens_per_second_per_gpu": 306.32
    },
    {
      "epoch": 0.4782025152855794,
      "grad_norm": 0.5693733096122742,
      "learning_rate": 0.00017391022048316476,
      "loss": 0.3306,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.19,
      "memory/max_allocated (GiB)": 19.19,
      "step": 4810,
      "tokens_per_second_per_gpu": 359.5
    },
    {
      "epoch": 0.47919669930904213,
      "grad_norm": 0.33154231309890747,
      "learning_rate": 0.00017380441157076643,
      "loss": 0.2469,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 4820,
      "tokens_per_second_per_gpu": 363.91
    },
    {
      "epoch": 0.48019088333250487,
      "grad_norm": 0.417501837015152,
      "learning_rate": 0.00017369842086250347,
      "loss": 0.2286,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 4830,
      "tokens_per_second_per_gpu": 261.89
    },
    {
      "epoch": 0.4811850673559676,
      "grad_norm": 0.2794663608074188,
      "learning_rate": 0.00017359224861945345,
      "loss": 0.2415,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 4840,
      "tokens_per_second_per_gpu": 350.79
    },
    {
      "epoch": 0.48217925137943035,
      "grad_norm": 0.31123244762420654,
      "learning_rate": 0.00017348589510314096,
      "loss": 0.2396,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 4850,
      "tokens_per_second_per_gpu": 380.51
    },
    {
      "epoch": 0.4831734354028931,
      "grad_norm": 0.21615763008594513,
      "learning_rate": 0.00017337936057553726,
      "loss": 0.2286,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 4860,
      "tokens_per_second_per_gpu": 430.28
    },
    {
      "epoch": 0.48416761942635583,
      "grad_norm": 0.38201475143432617,
      "learning_rate": 0.0001732726452990594,
      "loss": 0.2213,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 4870,
      "tokens_per_second_per_gpu": 337.65
    },
    {
      "epoch": 0.4851618034498186,
      "grad_norm": 0.4545513987541199,
      "learning_rate": 0.00017316574953656958,
      "loss": 0.2696,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 4880,
      "tokens_per_second_per_gpu": 388.83
    },
    {
      "epoch": 0.4861559874732813,
      "grad_norm": 0.2672022879123688,
      "learning_rate": 0.00017305867355137475,
      "loss": 0.1962,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 4890,
      "tokens_per_second_per_gpu": 318.92
    },
    {
      "epoch": 0.48715017149674406,
      "grad_norm": 0.4752904772758484,
      "learning_rate": 0.00017295141760722567,
      "loss": 0.2107,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 4900,
      "tokens_per_second_per_gpu": 361.56
    },
    {
      "epoch": 0.4881443555202068,
      "grad_norm": 0.4459490180015564,
      "learning_rate": 0.0001728439819683164,
      "loss": 0.2648,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 4910,
      "tokens_per_second_per_gpu": 302.7
    },
    {
      "epoch": 0.48913853954366954,
      "grad_norm": 0.6155937314033508,
      "learning_rate": 0.00017273636689928357,
      "loss": 0.2714,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 4920,
      "tokens_per_second_per_gpu": 379.02
    },
    {
      "epoch": 0.4901327235671323,
      "grad_norm": 0.4557485282421112,
      "learning_rate": 0.00017262857266520595,
      "loss": 0.1966,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 4930,
      "tokens_per_second_per_gpu": 369.72
    },
    {
      "epoch": 0.491126907590595,
      "grad_norm": 0.4636807143688202,
      "learning_rate": 0.0001725205995316034,
      "loss": 0.2803,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 4940,
      "tokens_per_second_per_gpu": 373.68
    },
    {
      "epoch": 0.49212109161405776,
      "grad_norm": 0.45894595980644226,
      "learning_rate": 0.00017241244776443666,
      "loss": 0.3439,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 4950,
      "tokens_per_second_per_gpu": 426.86
    },
    {
      "epoch": 0.4931152756375205,
      "grad_norm": 0.39921844005584717,
      "learning_rate": 0.0001723041176301063,
      "loss": 0.3035,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 4960,
      "tokens_per_second_per_gpu": 375.09
    },
    {
      "epoch": 0.49410945966098324,
      "grad_norm": 0.28210845589637756,
      "learning_rate": 0.00017219560939545246,
      "loss": 0.2043,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 4970,
      "tokens_per_second_per_gpu": 320.21
    },
    {
      "epoch": 0.495103643684446,
      "grad_norm": 0.5301778316497803,
      "learning_rate": 0.00017208692332775375,
      "loss": 0.2293,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 4980,
      "tokens_per_second_per_gpu": 327.97
    },
    {
      "epoch": 0.4960978277079087,
      "grad_norm": 0.40421542525291443,
      "learning_rate": 0.000171978059694727,
      "loss": 0.3101,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 4990,
      "tokens_per_second_per_gpu": 395.91
    },
    {
      "epoch": 0.49709201173137146,
      "grad_norm": 0.383989155292511,
      "learning_rate": 0.0001718690187645263,
      "loss": 0.2447,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 5000,
      "tokens_per_second_per_gpu": 344.59
    },
    {
      "epoch": 0.4980861957548342,
      "grad_norm": 0.559518039226532,
      "learning_rate": 0.00017175980080574247,
      "loss": 0.3176,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 5010,
      "tokens_per_second_per_gpu": 378.55
    },
    {
      "epoch": 0.49908037977829695,
      "grad_norm": 0.46650487184524536,
      "learning_rate": 0.00017165040608740255,
      "loss": 0.2006,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 5020,
      "tokens_per_second_per_gpu": 334.42
    },
    {
      "epoch": 0.5000745638017597,
      "grad_norm": 0.3541397750377655,
      "learning_rate": 0.00017154083487896872,
      "loss": 0.2542,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 5030,
      "tokens_per_second_per_gpu": 360.31
    },
    {
      "epoch": 0.5010687478252225,
      "grad_norm": 0.48084208369255066,
      "learning_rate": 0.00017143108745033811,
      "loss": 0.2133,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 5040,
      "tokens_per_second_per_gpu": 281.78
    },
    {
      "epoch": 0.5020629318486852,
      "grad_norm": 0.28656521439552307,
      "learning_rate": 0.0001713211640718418,
      "loss": 0.246,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 5050,
      "tokens_per_second_per_gpu": 334.53
    },
    {
      "epoch": 0.503057115872148,
      "grad_norm": 0.48624783754348755,
      "learning_rate": 0.0001712110650142443,
      "loss": 0.2497,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.65,
      "memory/max_allocated (GiB)": 20.65,
      "step": 5060,
      "tokens_per_second_per_gpu": 397.02
    },
    {
      "epoch": 0.5040512998956107,
      "grad_norm": 0.3334798216819763,
      "learning_rate": 0.00017110079054874288,
      "loss": 0.2366,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 5070,
      "tokens_per_second_per_gpu": 371.25
    },
    {
      "epoch": 0.5050454839190734,
      "grad_norm": 0.4432489275932312,
      "learning_rate": 0.00017099034094696685,
      "loss": 0.2104,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5080,
      "tokens_per_second_per_gpu": 384.5
    },
    {
      "epoch": 0.5060396679425362,
      "grad_norm": 0.5348425507545471,
      "learning_rate": 0.00017087971648097693,
      "loss": 0.2292,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5090,
      "tokens_per_second_per_gpu": 311.48
    },
    {
      "epoch": 0.5070338519659989,
      "grad_norm": 0.4587235748767853,
      "learning_rate": 0.00017076891742326452,
      "loss": 0.2297,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 5100,
      "tokens_per_second_per_gpu": 371.52
    },
    {
      "epoch": 0.5080280359894617,
      "grad_norm": 0.3222121298313141,
      "learning_rate": 0.00017065794404675112,
      "loss": 0.2447,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 5110,
      "tokens_per_second_per_gpu": 346.06
    },
    {
      "epoch": 0.5090222200129244,
      "grad_norm": 0.38898178935050964,
      "learning_rate": 0.0001705467966247877,
      "loss": 0.192,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5120,
      "tokens_per_second_per_gpu": 388.93
    },
    {
      "epoch": 0.5100164040363871,
      "grad_norm": 0.49159225821495056,
      "learning_rate": 0.00017043547543115373,
      "loss": 0.2604,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 5130,
      "tokens_per_second_per_gpu": 348.77
    },
    {
      "epoch": 0.5110105880598499,
      "grad_norm": 0.34262073040008545,
      "learning_rate": 0.0001703239807400569,
      "loss": 0.1843,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 5140,
      "tokens_per_second_per_gpu": 361.68
    },
    {
      "epoch": 0.5120047720833126,
      "grad_norm": 0.5016794800758362,
      "learning_rate": 0.00017021231282613223,
      "loss": 0.2527,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 5150,
      "tokens_per_second_per_gpu": 385.36
    },
    {
      "epoch": 0.5129989561067754,
      "grad_norm": 0.44959282875061035,
      "learning_rate": 0.00017010047196444137,
      "loss": 0.2349,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 5160,
      "tokens_per_second_per_gpu": 319.58
    },
    {
      "epoch": 0.5139931401302381,
      "grad_norm": 0.27737611532211304,
      "learning_rate": 0.00016998845843047193,
      "loss": 0.2564,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 5170,
      "tokens_per_second_per_gpu": 325.49
    },
    {
      "epoch": 0.5149873241537009,
      "grad_norm": 0.5175918340682983,
      "learning_rate": 0.00016987627250013702,
      "loss": 0.2375,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 5180,
      "tokens_per_second_per_gpu": 334.38
    },
    {
      "epoch": 0.5159815081771636,
      "grad_norm": 0.5541722178459167,
      "learning_rate": 0.00016976391444977425,
      "loss": 0.2478,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 5190,
      "tokens_per_second_per_gpu": 316.46
    },
    {
      "epoch": 0.5169756922006263,
      "grad_norm": 1.0252865552902222,
      "learning_rate": 0.00016965138455614525,
      "loss": 0.2371,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 5200,
      "tokens_per_second_per_gpu": 339.79
    },
    {
      "epoch": 0.5179698762240891,
      "grad_norm": 0.582073450088501,
      "learning_rate": 0.00016953868309643491,
      "loss": 0.2311,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 5210,
      "tokens_per_second_per_gpu": 359.2
    },
    {
      "epoch": 0.5189640602475518,
      "grad_norm": 0.5147042870521545,
      "learning_rate": 0.0001694258103482508,
      "loss": 0.2693,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 5220,
      "tokens_per_second_per_gpu": 379.55
    },
    {
      "epoch": 0.5199582442710146,
      "grad_norm": 0.6816319823265076,
      "learning_rate": 0.0001693127665896223,
      "loss": 0.2228,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 5230,
      "tokens_per_second_per_gpu": 296.31
    },
    {
      "epoch": 0.5209524282944773,
      "grad_norm": 0.765450656414032,
      "learning_rate": 0.00016919955209900012,
      "loss": 0.2764,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 5240,
      "tokens_per_second_per_gpu": 389.55
    },
    {
      "epoch": 0.52194661231794,
      "grad_norm": 0.3426934778690338,
      "learning_rate": 0.00016908616715525544,
      "loss": 0.3197,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 5250,
      "tokens_per_second_per_gpu": 404.47
    },
    {
      "epoch": 0.5229407963414028,
      "grad_norm": 0.45074665546417236,
      "learning_rate": 0.0001689726120376794,
      "loss": 0.2624,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 5260,
      "tokens_per_second_per_gpu": 362.3
    },
    {
      "epoch": 0.5239349803648655,
      "grad_norm": 0.4062357246875763,
      "learning_rate": 0.00016885888702598218,
      "loss": 0.2068,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.52,
      "memory/max_allocated (GiB)": 20.52,
      "step": 5270,
      "tokens_per_second_per_gpu": 383.64
    },
    {
      "epoch": 0.5249291643883283,
      "grad_norm": 0.4385395050048828,
      "learning_rate": 0.00016874499240029253,
      "loss": 0.2886,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 5280,
      "tokens_per_second_per_gpu": 332.1
    },
    {
      "epoch": 0.525923348411791,
      "grad_norm": 0.4379644989967346,
      "learning_rate": 0.0001686309284411571,
      "loss": 0.2305,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.08,
      "memory/max_allocated (GiB)": 19.08,
      "step": 5290,
      "tokens_per_second_per_gpu": 328.75
    },
    {
      "epoch": 0.5269175324352537,
      "grad_norm": 0.5357609987258911,
      "learning_rate": 0.00016851669542953935,
      "loss": 0.2526,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 5300,
      "tokens_per_second_per_gpu": 396.9
    },
    {
      "epoch": 0.5279117164587165,
      "grad_norm": 0.5790386199951172,
      "learning_rate": 0.00016840229364681948,
      "loss": 0.193,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5310,
      "tokens_per_second_per_gpu": 313.5
    },
    {
      "epoch": 0.5289059004821792,
      "grad_norm": 0.4063149690628052,
      "learning_rate": 0.00016828772337479318,
      "loss": 0.2071,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 5320,
      "tokens_per_second_per_gpu": 290.65
    },
    {
      "epoch": 0.529900084505642,
      "grad_norm": 0.3840952515602112,
      "learning_rate": 0.00016817298489567127,
      "loss": 0.2086,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 5330,
      "tokens_per_second_per_gpu": 372.36
    },
    {
      "epoch": 0.5308942685291047,
      "grad_norm": 0.36974212527275085,
      "learning_rate": 0.0001680580784920789,
      "loss": 0.2556,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 5340,
      "tokens_per_second_per_gpu": 369.63
    },
    {
      "epoch": 0.5318884525525674,
      "grad_norm": 0.5495437979698181,
      "learning_rate": 0.00016794300444705477,
      "loss": 0.2667,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 5350,
      "tokens_per_second_per_gpu": 327.36
    },
    {
      "epoch": 0.5328826365760302,
      "grad_norm": 0.5131349563598633,
      "learning_rate": 0.0001678277630440506,
      "loss": 0.2311,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 5360,
      "tokens_per_second_per_gpu": 357.39
    },
    {
      "epoch": 0.5338768205994929,
      "grad_norm": 0.28221410512924194,
      "learning_rate": 0.00016771235456693035,
      "loss": 0.2532,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 5370,
      "tokens_per_second_per_gpu": 385.44
    },
    {
      "epoch": 0.5348710046229557,
      "grad_norm": 0.5543202757835388,
      "learning_rate": 0.0001675967792999695,
      "loss": 0.2999,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.19,
      "memory/max_allocated (GiB)": 19.19,
      "step": 5380,
      "tokens_per_second_per_gpu": 393.99
    },
    {
      "epoch": 0.5358651886464184,
      "grad_norm": 0.3077482283115387,
      "learning_rate": 0.00016748103752785426,
      "loss": 0.2071,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 5390,
      "tokens_per_second_per_gpu": 372.33
    },
    {
      "epoch": 0.5368593726698812,
      "grad_norm": 0.6371411681175232,
      "learning_rate": 0.00016736512953568117,
      "loss": 0.1986,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.43,
      "memory/max_allocated (GiB)": 20.43,
      "step": 5400,
      "tokens_per_second_per_gpu": 325.81
    },
    {
      "epoch": 0.5378535566933439,
      "grad_norm": 0.4577232003211975,
      "learning_rate": 0.0001672490556089561,
      "loss": 0.253,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5410,
      "tokens_per_second_per_gpu": 388.89
    },
    {
      "epoch": 0.5388477407168066,
      "grad_norm": 0.2831481099128723,
      "learning_rate": 0.00016713281603359366,
      "loss": 0.1994,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 5420,
      "tokens_per_second_per_gpu": 373.61
    },
    {
      "epoch": 0.5398419247402694,
      "grad_norm": 0.45899850130081177,
      "learning_rate": 0.00016701641109591648,
      "loss": 0.1997,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 5430,
      "tokens_per_second_per_gpu": 386.1
    },
    {
      "epoch": 0.5408361087637321,
      "grad_norm": 0.2898833453655243,
      "learning_rate": 0.0001668998410826545,
      "loss": 0.276,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 5440,
      "tokens_per_second_per_gpu": 374.49
    },
    {
      "epoch": 0.5418302927871949,
      "grad_norm": 0.37996944785118103,
      "learning_rate": 0.00016678310628094438,
      "loss": 0.2529,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 5450,
      "tokens_per_second_per_gpu": 359.7
    },
    {
      "epoch": 0.5428244768106577,
      "grad_norm": 0.255087673664093,
      "learning_rate": 0.0001666662069783285,
      "loss": 0.2682,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 5460,
      "tokens_per_second_per_gpu": 346.66
    },
    {
      "epoch": 0.5438186608341204,
      "grad_norm": 0.4574085474014282,
      "learning_rate": 0.00016654914346275466,
      "loss": 0.2516,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 5470,
      "tokens_per_second_per_gpu": 350.42
    },
    {
      "epoch": 0.5448128448575832,
      "grad_norm": 0.5208483934402466,
      "learning_rate": 0.00016643191602257496,
      "loss": 0.2549,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 5480,
      "tokens_per_second_per_gpu": 320.05
    },
    {
      "epoch": 0.5458070288810459,
      "grad_norm": 0.31626808643341064,
      "learning_rate": 0.00016631452494654541,
      "loss": 0.2151,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 5490,
      "tokens_per_second_per_gpu": 335.31
    },
    {
      "epoch": 0.5468012129045087,
      "grad_norm": 0.28286507725715637,
      "learning_rate": 0.000166196970523825,
      "loss": 0.2847,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 5500,
      "tokens_per_second_per_gpu": 406.77
    },
    {
      "epoch": 0.5477953969279714,
      "grad_norm": 0.3647201955318451,
      "learning_rate": 0.00016607925304397517,
      "loss": 0.1912,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 5510,
      "tokens_per_second_per_gpu": 334.31
    },
    {
      "epoch": 0.5487895809514342,
      "grad_norm": 0.48184719681739807,
      "learning_rate": 0.0001659613727969589,
      "loss": 0.2336,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5520,
      "tokens_per_second_per_gpu": 355.21
    },
    {
      "epoch": 0.5497837649748969,
      "grad_norm": 0.5385854244232178,
      "learning_rate": 0.00016584333007314017,
      "loss": 0.2764,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 5530,
      "tokens_per_second_per_gpu": 384.64
    },
    {
      "epoch": 0.5507779489983596,
      "grad_norm": 0.2823385000228882,
      "learning_rate": 0.00016572512516328317,
      "loss": 0.3002,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.82,
      "memory/max_allocated (GiB)": 18.82,
      "step": 5540,
      "tokens_per_second_per_gpu": 417.12
    },
    {
      "epoch": 0.5517721330218224,
      "grad_norm": 0.4133371114730835,
      "learning_rate": 0.0001656067583585516,
      "loss": 0.2066,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 5550,
      "tokens_per_second_per_gpu": 314.99
    },
    {
      "epoch": 0.5527663170452851,
      "grad_norm": 0.2247128188610077,
      "learning_rate": 0.00016548822995050787,
      "loss": 0.2582,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 5560,
      "tokens_per_second_per_gpu": 341.74
    },
    {
      "epoch": 0.5537605010687479,
      "grad_norm": 0.5088145136833191,
      "learning_rate": 0.0001653695402311125,
      "loss": 0.3169,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 5570,
      "tokens_per_second_per_gpu": 385.48
    },
    {
      "epoch": 0.5547546850922106,
      "grad_norm": 0.600646436214447,
      "learning_rate": 0.0001652506894927234,
      "loss": 0.3391,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 5580,
      "tokens_per_second_per_gpu": 338.0
    },
    {
      "epoch": 0.5557488691156733,
      "grad_norm": 0.34408292174339294,
      "learning_rate": 0.00016513167802809502,
      "loss": 0.2003,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5590,
      "tokens_per_second_per_gpu": 341.35
    },
    {
      "epoch": 0.5567430531391361,
      "grad_norm": 0.44201189279556274,
      "learning_rate": 0.0001650125061303778,
      "loss": 0.2494,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 5600,
      "tokens_per_second_per_gpu": 386.01
    },
    {
      "epoch": 0.5577372371625988,
      "grad_norm": 0.5484967827796936,
      "learning_rate": 0.00016489317409311717,
      "loss": 0.2881,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 5610,
      "tokens_per_second_per_gpu": 327.21
    },
    {
      "epoch": 0.5587314211860616,
      "grad_norm": 0.45674219727516174,
      "learning_rate": 0.00016477368221025333,
      "loss": 0.2377,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 5620,
      "tokens_per_second_per_gpu": 365.25
    },
    {
      "epoch": 0.5597256052095243,
      "grad_norm": 0.3383237421512604,
      "learning_rate": 0.00016465403077612001,
      "loss": 0.2251,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 5630,
      "tokens_per_second_per_gpu": 357.99
    },
    {
      "epoch": 0.560719789232987,
      "grad_norm": 0.3748398423194885,
      "learning_rate": 0.00016453422008544388,
      "loss": 0.2279,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 5640,
      "tokens_per_second_per_gpu": 335.12
    },
    {
      "epoch": 0.5617139732564498,
      "grad_norm": 0.4504337012767792,
      "learning_rate": 0.00016441425043334413,
      "loss": 0.261,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 5650,
      "tokens_per_second_per_gpu": 336.53
    },
    {
      "epoch": 0.5627081572799125,
      "grad_norm": 0.3367341458797455,
      "learning_rate": 0.00016429412211533127,
      "loss": 0.1855,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5660,
      "tokens_per_second_per_gpu": 347.1
    },
    {
      "epoch": 0.5637023413033753,
      "grad_norm": 0.3390282988548279,
      "learning_rate": 0.00016417383542730675,
      "loss": 0.2428,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 5670,
      "tokens_per_second_per_gpu": 385.32
    },
    {
      "epoch": 0.564696525326838,
      "grad_norm": 0.6450189352035522,
      "learning_rate": 0.00016405339066556212,
      "loss": 0.3651,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 5680,
      "tokens_per_second_per_gpu": 427.22
    },
    {
      "epoch": 0.5656907093503007,
      "grad_norm": 0.5655054450035095,
      "learning_rate": 0.0001639327881267783,
      "loss": 0.2204,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 5690,
      "tokens_per_second_per_gpu": 310.84
    },
    {
      "epoch": 0.5666848933737635,
      "grad_norm": 0.4713475704193115,
      "learning_rate": 0.00016381202810802483,
      "loss": 0.2294,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.47,
      "memory/max_allocated (GiB)": 21.47,
      "step": 5700,
      "tokens_per_second_per_gpu": 363.21
    },
    {
      "epoch": 0.5676790773972262,
      "grad_norm": 0.19377703964710236,
      "learning_rate": 0.00016369111090675916,
      "loss": 0.2522,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 5710,
      "tokens_per_second_per_gpu": 412.63
    },
    {
      "epoch": 0.568673261420689,
      "grad_norm": 0.5475621819496155,
      "learning_rate": 0.0001635700368208259,
      "loss": 0.3132,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 5720,
      "tokens_per_second_per_gpu": 367.47
    },
    {
      "epoch": 0.5696674454441517,
      "grad_norm": 0.3860287368297577,
      "learning_rate": 0.00016344880614845608,
      "loss": 0.2623,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 5730,
      "tokens_per_second_per_gpu": 381.97
    },
    {
      "epoch": 0.5706616294676145,
      "grad_norm": 0.3809770345687866,
      "learning_rate": 0.00016332741918826654,
      "loss": 0.2365,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 5740,
      "tokens_per_second_per_gpu": 326.73
    },
    {
      "epoch": 0.5716558134910772,
      "grad_norm": 0.5006048083305359,
      "learning_rate": 0.00016320587623925895,
      "loss": 0.2661,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 5750,
      "tokens_per_second_per_gpu": 334.44
    },
    {
      "epoch": 0.5726499975145399,
      "grad_norm": 0.2004530131816864,
      "learning_rate": 0.00016308417760081936,
      "loss": 0.1923,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 5760,
      "tokens_per_second_per_gpu": 377.56
    },
    {
      "epoch": 0.5736441815380027,
      "grad_norm": 0.30563804507255554,
      "learning_rate": 0.00016296232357271718,
      "loss": 0.2089,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 5770,
      "tokens_per_second_per_gpu": 351.29
    },
    {
      "epoch": 0.5746383655614654,
      "grad_norm": 0.3142051100730896,
      "learning_rate": 0.00016284031445510465,
      "loss": 0.1931,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 5780,
      "tokens_per_second_per_gpu": 327.41
    },
    {
      "epoch": 0.5756325495849282,
      "grad_norm": 0.4678092300891876,
      "learning_rate": 0.000162718150548516,
      "loss": 0.3011,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 5790,
      "tokens_per_second_per_gpu": 367.12
    },
    {
      "epoch": 0.5766267336083909,
      "grad_norm": 0.44859957695007324,
      "learning_rate": 0.00016259583215386675,
      "loss": 0.2855,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5800,
      "tokens_per_second_per_gpu": 377.51
    },
    {
      "epoch": 0.5776209176318536,
      "grad_norm": 0.3087036609649658,
      "learning_rate": 0.00016247335957245303,
      "loss": 0.2181,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.21,
      "memory/max_allocated (GiB)": 18.21,
      "step": 5810,
      "tokens_per_second_per_gpu": 292.23
    },
    {
      "epoch": 0.5786151016553164,
      "grad_norm": 0.29946303367614746,
      "learning_rate": 0.00016235073310595058,
      "loss": 0.203,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 5820,
      "tokens_per_second_per_gpu": 294.26
    },
    {
      "epoch": 0.5796092856787791,
      "grad_norm": 0.41978803277015686,
      "learning_rate": 0.0001622279530564144,
      "loss": 0.2186,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 5830,
      "tokens_per_second_per_gpu": 332.7
    },
    {
      "epoch": 0.5806034697022419,
      "grad_norm": 0.5330750942230225,
      "learning_rate": 0.00016210501972627764,
      "loss": 0.2726,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 5840,
      "tokens_per_second_per_gpu": 372.14
    },
    {
      "epoch": 0.5815976537257046,
      "grad_norm": 0.4443431794643402,
      "learning_rate": 0.0001619819334183511,
      "loss": 0.2121,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 5850,
      "tokens_per_second_per_gpu": 329.51
    },
    {
      "epoch": 0.5825918377491673,
      "grad_norm": 0.4602350890636444,
      "learning_rate": 0.00016185869443582237,
      "loss": 0.2845,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.85,
      "memory/max_allocated (GiB)": 18.85,
      "step": 5860,
      "tokens_per_second_per_gpu": 334.35
    },
    {
      "epoch": 0.5835860217726301,
      "grad_norm": 0.2751925587654114,
      "learning_rate": 0.00016173530308225513,
      "loss": 0.1812,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 5870,
      "tokens_per_second_per_gpu": 397.13
    },
    {
      "epoch": 0.5845802057960928,
      "grad_norm": 0.5334510803222656,
      "learning_rate": 0.00016161175966158834,
      "loss": 0.2088,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 5880,
      "tokens_per_second_per_gpu": 297.42
    },
    {
      "epoch": 0.5855743898195556,
      "grad_norm": 0.36908817291259766,
      "learning_rate": 0.00016148806447813553,
      "loss": 0.2197,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 5890,
      "tokens_per_second_per_gpu": 294.32
    },
    {
      "epoch": 0.5865685738430183,
      "grad_norm": 0.24384719133377075,
      "learning_rate": 0.00016136421783658416,
      "loss": 0.2757,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.81,
      "memory/max_allocated (GiB)": 18.81,
      "step": 5900,
      "tokens_per_second_per_gpu": 315.54
    },
    {
      "epoch": 0.587562757866481,
      "grad_norm": 0.41709259152412415,
      "learning_rate": 0.0001612402200419946,
      "loss": 0.2858,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 5910,
      "tokens_per_second_per_gpu": 385.92
    },
    {
      "epoch": 0.5885569418899438,
      "grad_norm": 0.5187587738037109,
      "learning_rate": 0.00016111607139979967,
      "loss": 0.2347,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 5920,
      "tokens_per_second_per_gpu": 377.8
    },
    {
      "epoch": 0.5895511259134065,
      "grad_norm": 0.29502683877944946,
      "learning_rate": 0.00016099177221580373,
      "loss": 0.2563,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 5930,
      "tokens_per_second_per_gpu": 335.51
    },
    {
      "epoch": 0.5905453099368693,
      "grad_norm": 0.360385000705719,
      "learning_rate": 0.00016086732279618188,
      "loss": 0.1982,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 5940,
      "tokens_per_second_per_gpu": 315.07
    },
    {
      "epoch": 0.591539493960332,
      "grad_norm": 0.35591524839401245,
      "learning_rate": 0.0001607427234474794,
      "loss": 0.2535,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 5950,
      "tokens_per_second_per_gpu": 379.0
    },
    {
      "epoch": 0.5925336779837947,
      "grad_norm": 0.49772289395332336,
      "learning_rate": 0.0001606179744766108,
      "loss": 0.3105,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 5960,
      "tokens_per_second_per_gpu": 403.62
    },
    {
      "epoch": 0.5935278620072575,
      "grad_norm": 0.3001823425292969,
      "learning_rate": 0.00016049307619085915,
      "loss": 0.2011,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 5970,
      "tokens_per_second_per_gpu": 289.33
    },
    {
      "epoch": 0.5945220460307203,
      "grad_norm": 0.593662679195404,
      "learning_rate": 0.00016036802889787536,
      "loss": 0.2728,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 5980,
      "tokens_per_second_per_gpu": 397.95
    },
    {
      "epoch": 0.5955162300541831,
      "grad_norm": 0.38041049242019653,
      "learning_rate": 0.00016024283290567732,
      "loss": 0.2016,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 5990,
      "tokens_per_second_per_gpu": 310.64
    },
    {
      "epoch": 0.5965104140776458,
      "grad_norm": 0.28381025791168213,
      "learning_rate": 0.0001601174885226492,
      "loss": 0.2,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 6000,
      "tokens_per_second_per_gpu": 368.07
    },
    {
      "epoch": 0.5975045981011086,
      "grad_norm": 0.4913434088230133,
      "learning_rate": 0.0001599919960575407,
      "loss": 0.2653,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 6010,
      "tokens_per_second_per_gpu": 365.97
    },
    {
      "epoch": 0.5984987821245713,
      "grad_norm": 0.4141637086868286,
      "learning_rate": 0.00015986635581946638,
      "loss": 0.2687,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 6020,
      "tokens_per_second_per_gpu": 363.69
    },
    {
      "epoch": 0.599492966148034,
      "grad_norm": 0.45988166332244873,
      "learning_rate": 0.00015974056811790462,
      "loss": 0.2625,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 6030,
      "tokens_per_second_per_gpu": 402.4
    },
    {
      "epoch": 0.6004871501714968,
      "grad_norm": 0.5327023267745972,
      "learning_rate": 0.0001596146332626971,
      "loss": 0.2462,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 6040,
      "tokens_per_second_per_gpu": 382.41
    },
    {
      "epoch": 0.6014813341949595,
      "grad_norm": 0.2608936131000519,
      "learning_rate": 0.00015948855156404802,
      "loss": 0.2171,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 6050,
      "tokens_per_second_per_gpu": 369.97
    },
    {
      "epoch": 0.6024755182184223,
      "grad_norm": 0.5496955513954163,
      "learning_rate": 0.00015936232333252327,
      "loss": 0.3213,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 6060,
      "tokens_per_second_per_gpu": 365.39
    },
    {
      "epoch": 0.603469702241885,
      "grad_norm": 0.40054014325141907,
      "learning_rate": 0.00015923594887904964,
      "loss": 0.1949,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 6070,
      "tokens_per_second_per_gpu": 293.13
    },
    {
      "epoch": 0.6044638862653477,
      "grad_norm": 0.7320500612258911,
      "learning_rate": 0.0001591094285149141,
      "loss": 0.2662,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 6080,
      "tokens_per_second_per_gpu": 336.97
    },
    {
      "epoch": 0.6054580702888105,
      "grad_norm": 0.484331876039505,
      "learning_rate": 0.00015898276255176303,
      "loss": 0.2487,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 6090,
      "tokens_per_second_per_gpu": 359.19
    },
    {
      "epoch": 0.6064522543122732,
      "grad_norm": 0.4304184019565582,
      "learning_rate": 0.00015885595130160155,
      "loss": 0.2546,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 6100,
      "tokens_per_second_per_gpu": 329.56
    },
    {
      "epoch": 0.607446438335736,
      "grad_norm": 0.49904513359069824,
      "learning_rate": 0.00015872899507679252,
      "loss": 0.2622,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 6110,
      "tokens_per_second_per_gpu": 381.18
    },
    {
      "epoch": 0.6084406223591987,
      "grad_norm": 0.525607168674469,
      "learning_rate": 0.00015860189419005595,
      "loss": 0.2424,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 6120,
      "tokens_per_second_per_gpu": 331.0
    },
    {
      "epoch": 0.6094348063826615,
      "grad_norm": 0.3618375360965729,
      "learning_rate": 0.0001584746489544682,
      "loss": 0.3099,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 6130,
      "tokens_per_second_per_gpu": 435.27
    },
    {
      "epoch": 0.6104289904061242,
      "grad_norm": 0.41309383511543274,
      "learning_rate": 0.00015834725968346116,
      "loss": 0.2337,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 6140,
      "tokens_per_second_per_gpu": 381.0
    },
    {
      "epoch": 0.6114231744295869,
      "grad_norm": 0.37885645031929016,
      "learning_rate": 0.00015821972669082156,
      "loss": 0.3318,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 6150,
      "tokens_per_second_per_gpu": 365.41
    },
    {
      "epoch": 0.6124173584530497,
      "grad_norm": 0.4314253032207489,
      "learning_rate": 0.0001580920502906901,
      "loss": 0.2213,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6160,
      "tokens_per_second_per_gpu": 343.04
    },
    {
      "epoch": 0.6134115424765124,
      "grad_norm": 0.5435411930084229,
      "learning_rate": 0.00015796423079756074,
      "loss": 0.243,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 6170,
      "tokens_per_second_per_gpu": 365.75
    },
    {
      "epoch": 0.6144057264999752,
      "grad_norm": 0.31716790795326233,
      "learning_rate": 0.00015783626852627992,
      "loss": 0.2484,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 6180,
      "tokens_per_second_per_gpu": 435.76
    },
    {
      "epoch": 0.6153999105234379,
      "grad_norm": 0.3328079879283905,
      "learning_rate": 0.0001577081637920457,
      "loss": 0.223,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6190,
      "tokens_per_second_per_gpu": 310.86
    },
    {
      "epoch": 0.6163940945469006,
      "grad_norm": 0.3827805519104004,
      "learning_rate": 0.00015757991691040722,
      "loss": 0.2311,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 6200,
      "tokens_per_second_per_gpu": 333.74
    },
    {
      "epoch": 0.6173882785703634,
      "grad_norm": 0.3756648004055023,
      "learning_rate": 0.00015745152819726356,
      "loss": 0.229,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 6210,
      "tokens_per_second_per_gpu": 352.29
    },
    {
      "epoch": 0.6183824625938261,
      "grad_norm": 0.4345835745334625,
      "learning_rate": 0.0001573229979688633,
      "loss": 0.2694,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 6220,
      "tokens_per_second_per_gpu": 411.14
    },
    {
      "epoch": 0.6193766466172889,
      "grad_norm": 0.23183055222034454,
      "learning_rate": 0.00015719432654180357,
      "loss": 0.1925,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6230,
      "tokens_per_second_per_gpu": 363.23
    },
    {
      "epoch": 0.6203708306407516,
      "grad_norm": 0.2867846190929413,
      "learning_rate": 0.00015706551423302925,
      "loss": 0.1506,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6240,
      "tokens_per_second_per_gpu": 275.12
    },
    {
      "epoch": 0.6213650146642143,
      "grad_norm": 0.42951786518096924,
      "learning_rate": 0.00015693656135983233,
      "loss": 0.2867,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 6250,
      "tokens_per_second_per_gpu": 383.43
    },
    {
      "epoch": 0.6223591986876771,
      "grad_norm": 0.5975165963172913,
      "learning_rate": 0.00015680746823985094,
      "loss": 0.263,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 6260,
      "tokens_per_second_per_gpu": 304.9
    },
    {
      "epoch": 0.6233533827111398,
      "grad_norm": 0.48659709095954895,
      "learning_rate": 0.00015667823519106873,
      "loss": 0.2079,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 6270,
      "tokens_per_second_per_gpu": 340.54
    },
    {
      "epoch": 0.6243475667346026,
      "grad_norm": 0.461224764585495,
      "learning_rate": 0.00015654886253181402,
      "loss": 0.2537,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 6280,
      "tokens_per_second_per_gpu": 321.0
    },
    {
      "epoch": 0.6253417507580653,
      "grad_norm": 0.36294886469841003,
      "learning_rate": 0.00015641935058075904,
      "loss": 0.2009,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 6290,
      "tokens_per_second_per_gpu": 308.81
    },
    {
      "epoch": 0.626335934781528,
      "grad_norm": 0.4274291396141052,
      "learning_rate": 0.0001562896996569191,
      "loss": 0.2343,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 6300,
      "tokens_per_second_per_gpu": 319.11
    },
    {
      "epoch": 0.6273301188049908,
      "grad_norm": 0.47336748242378235,
      "learning_rate": 0.00015615991007965176,
      "loss": 0.211,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 6310,
      "tokens_per_second_per_gpu": 355.64
    },
    {
      "epoch": 0.6283243028284535,
      "grad_norm": 0.4946894347667694,
      "learning_rate": 0.00015602998216865624,
      "loss": 0.2492,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 6320,
      "tokens_per_second_per_gpu": 360.15
    },
    {
      "epoch": 0.6293184868519163,
      "grad_norm": 0.38327473402023315,
      "learning_rate": 0.00015589991624397244,
      "loss": 0.2308,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.61,
      "memory/max_allocated (GiB)": 20.61,
      "step": 6330,
      "tokens_per_second_per_gpu": 301.77
    },
    {
      "epoch": 0.630312670875379,
      "grad_norm": 0.46532171964645386,
      "learning_rate": 0.00015576971262598024,
      "loss": 0.2812,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 6340,
      "tokens_per_second_per_gpu": 344.05
    },
    {
      "epoch": 0.6313068548988418,
      "grad_norm": 0.3749048709869385,
      "learning_rate": 0.00015563937163539862,
      "loss": 0.2415,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 6350,
      "tokens_per_second_per_gpu": 348.41
    },
    {
      "epoch": 0.6323010389223045,
      "grad_norm": 0.2943509519100189,
      "learning_rate": 0.000155508893593285,
      "loss": 0.2228,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 6360,
      "tokens_per_second_per_gpu": 288.63
    },
    {
      "epoch": 0.6332952229457672,
      "grad_norm": 0.5494127869606018,
      "learning_rate": 0.00015537827882103442,
      "loss": 0.2499,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 6370,
      "tokens_per_second_per_gpu": 356.07
    },
    {
      "epoch": 0.63428940696923,
      "grad_norm": 0.48988381028175354,
      "learning_rate": 0.0001552475276403786,
      "loss": 0.2142,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 6380,
      "tokens_per_second_per_gpu": 343.8
    },
    {
      "epoch": 0.6352835909926927,
      "grad_norm": 0.3422715365886688,
      "learning_rate": 0.00015511664037338538,
      "loss": 0.2364,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 6390,
      "tokens_per_second_per_gpu": 409.18
    },
    {
      "epoch": 0.6362777750161555,
      "grad_norm": 0.6021102070808411,
      "learning_rate": 0.00015498561734245776,
      "loss": 0.2392,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 6400,
      "tokens_per_second_per_gpu": 306.59
    },
    {
      "epoch": 0.6372719590396182,
      "grad_norm": 0.6073122620582581,
      "learning_rate": 0.00015485445887033317,
      "loss": 0.2798,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 6410,
      "tokens_per_second_per_gpu": 342.93
    },
    {
      "epoch": 0.6382661430630809,
      "grad_norm": 0.3407362401485443,
      "learning_rate": 0.0001547231652800826,
      "loss": 0.2477,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 6420,
      "tokens_per_second_per_gpu": 367.05
    },
    {
      "epoch": 0.6392603270865437,
      "grad_norm": 0.506592869758606,
      "learning_rate": 0.00015459173689510994,
      "loss": 0.2399,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 6430,
      "tokens_per_second_per_gpu": 412.24
    },
    {
      "epoch": 0.6402545111100064,
      "grad_norm": 0.5419439077377319,
      "learning_rate": 0.0001544601740391511,
      "loss": 0.1948,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 6440,
      "tokens_per_second_per_gpu": 329.23
    },
    {
      "epoch": 0.6412486951334692,
      "grad_norm": 0.48251059651374817,
      "learning_rate": 0.00015432847703627316,
      "loss": 0.2146,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 6450,
      "tokens_per_second_per_gpu": 317.65
    },
    {
      "epoch": 0.6422428791569319,
      "grad_norm": 0.22626249492168427,
      "learning_rate": 0.0001541966462108737,
      "loss": 0.2593,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 6460,
      "tokens_per_second_per_gpu": 331.91
    },
    {
      "epoch": 0.6432370631803946,
      "grad_norm": 0.5113493204116821,
      "learning_rate": 0.0001540646818876799,
      "loss": 0.2162,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 6470,
      "tokens_per_second_per_gpu": 311.21
    },
    {
      "epoch": 0.6442312472038574,
      "grad_norm": 0.3097884953022003,
      "learning_rate": 0.0001539325843917478,
      "loss": 0.1879,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 6480,
      "tokens_per_second_per_gpu": 320.21
    },
    {
      "epoch": 0.6452254312273201,
      "grad_norm": 0.32837173342704773,
      "learning_rate": 0.0001538003540484614,
      "loss": 0.217,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.68,
      "memory/max_allocated (GiB)": 19.68,
      "step": 6490,
      "tokens_per_second_per_gpu": 351.68
    },
    {
      "epoch": 0.646219615250783,
      "grad_norm": 0.519063413143158,
      "learning_rate": 0.00015366799118353202,
      "loss": 0.2531,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 6500,
      "tokens_per_second_per_gpu": 379.3
    },
    {
      "epoch": 0.6472137992742457,
      "grad_norm": 0.3581913113594055,
      "learning_rate": 0.0001535354961229974,
      "loss": 0.291,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 6510,
      "tokens_per_second_per_gpu": 389.02
    },
    {
      "epoch": 0.6482079832977085,
      "grad_norm": 0.3630671799182892,
      "learning_rate": 0.0001534028691932208,
      "loss": 0.2409,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 6520,
      "tokens_per_second_per_gpu": 313.47
    },
    {
      "epoch": 0.6492021673211712,
      "grad_norm": 0.3670892119407654,
      "learning_rate": 0.00015327011072089044,
      "loss": 0.2133,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 6530,
      "tokens_per_second_per_gpu": 290.75
    },
    {
      "epoch": 0.6501963513446339,
      "grad_norm": 0.40198561549186707,
      "learning_rate": 0.00015313722103301852,
      "loss": 0.27,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.85,
      "memory/max_allocated (GiB)": 18.85,
      "step": 6540,
      "tokens_per_second_per_gpu": 429.99
    },
    {
      "epoch": 0.6511905353680967,
      "grad_norm": 0.3494684398174286,
      "learning_rate": 0.00015300420045694034,
      "loss": 0.1676,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6550,
      "tokens_per_second_per_gpu": 368.89
    },
    {
      "epoch": 0.6521847193915594,
      "grad_norm": 0.42560404539108276,
      "learning_rate": 0.00015287104932031374,
      "loss": 0.2585,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 6560,
      "tokens_per_second_per_gpu": 317.83
    },
    {
      "epoch": 0.6531789034150222,
      "grad_norm": 0.511513352394104,
      "learning_rate": 0.00015273776795111813,
      "loss": 0.2129,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 6570,
      "tokens_per_second_per_gpu": 327.26
    },
    {
      "epoch": 0.6541730874384849,
      "grad_norm": 0.3022279441356659,
      "learning_rate": 0.00015260435667765364,
      "loss": 0.2674,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 6580,
      "tokens_per_second_per_gpu": 332.65
    },
    {
      "epoch": 0.6551672714619476,
      "grad_norm": 0.3808051347732544,
      "learning_rate": 0.00015247081582854053,
      "loss": 0.2512,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 6590,
      "tokens_per_second_per_gpu": 379.8
    },
    {
      "epoch": 0.6561614554854104,
      "grad_norm": 0.4839475154876709,
      "learning_rate": 0.00015233714573271802,
      "loss": 0.2376,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 6600,
      "tokens_per_second_per_gpu": 329.54
    },
    {
      "epoch": 0.6571556395088731,
      "grad_norm": 0.7145663499832153,
      "learning_rate": 0.0001522033467194439,
      "loss": 0.2289,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 6610,
      "tokens_per_second_per_gpu": 327.02
    },
    {
      "epoch": 0.6581498235323359,
      "grad_norm": 0.4483419358730316,
      "learning_rate": 0.00015206941911829336,
      "loss": 0.2619,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 6620,
      "tokens_per_second_per_gpu": 344.26
    },
    {
      "epoch": 0.6591440075557986,
      "grad_norm": 0.7042835354804993,
      "learning_rate": 0.00015193536325915842,
      "loss": 0.3162,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 6630,
      "tokens_per_second_per_gpu": 325.39
    },
    {
      "epoch": 0.6601381915792613,
      "grad_norm": 0.44085246324539185,
      "learning_rate": 0.00015180117947224698,
      "loss": 0.1955,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 6640,
      "tokens_per_second_per_gpu": 356.35
    },
    {
      "epoch": 0.6611323756027241,
      "grad_norm": 0.32135269045829773,
      "learning_rate": 0.00015166686808808208,
      "loss": 0.2302,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 6650,
      "tokens_per_second_per_gpu": 401.75
    },
    {
      "epoch": 0.6621265596261868,
      "grad_norm": 0.5171180367469788,
      "learning_rate": 0.00015153242943750103,
      "loss": 0.251,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 6660,
      "tokens_per_second_per_gpu": 328.4
    },
    {
      "epoch": 0.6631207436496496,
      "grad_norm": 0.5205950140953064,
      "learning_rate": 0.00015139786385165462,
      "loss": 0.2186,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 6670,
      "tokens_per_second_per_gpu": 325.18
    },
    {
      "epoch": 0.6641149276731123,
      "grad_norm": 0.31780245900154114,
      "learning_rate": 0.0001512631716620064,
      "loss": 0.1604,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 6680,
      "tokens_per_second_per_gpu": 319.84
    },
    {
      "epoch": 0.665109111696575,
      "grad_norm": 0.29278233647346497,
      "learning_rate": 0.00015112835320033163,
      "loss": 0.266,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 6690,
      "tokens_per_second_per_gpu": 408.19
    },
    {
      "epoch": 0.6661032957200378,
      "grad_norm": 0.47382065653800964,
      "learning_rate": 0.00015099340879871668,
      "loss": 0.1933,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 6700,
      "tokens_per_second_per_gpu": 302.45
    },
    {
      "epoch": 0.6670974797435005,
      "grad_norm": 0.3947311043739319,
      "learning_rate": 0.00015085833878955823,
      "loss": 0.2225,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 6710,
      "tokens_per_second_per_gpu": 418.2
    },
    {
      "epoch": 0.6680916637669633,
      "grad_norm": 0.5490260720252991,
      "learning_rate": 0.00015072314350556213,
      "loss": 0.2056,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 6720,
      "tokens_per_second_per_gpu": 319.35
    },
    {
      "epoch": 0.669085847790426,
      "grad_norm": 0.412194162607193,
      "learning_rate": 0.000150587823279743,
      "loss": 0.2377,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 6730,
      "tokens_per_second_per_gpu": 304.73
    },
    {
      "epoch": 0.6700800318138888,
      "grad_norm": 0.40393805503845215,
      "learning_rate": 0.00015045237844542317,
      "loss": 0.2622,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 6740,
      "tokens_per_second_per_gpu": 335.77
    },
    {
      "epoch": 0.6710742158373515,
      "grad_norm": 0.5896100401878357,
      "learning_rate": 0.00015031680933623188,
      "loss": 0.3129,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 6750,
      "tokens_per_second_per_gpu": 304.37
    },
    {
      "epoch": 0.6720683998608142,
      "grad_norm": 0.5198945999145508,
      "learning_rate": 0.00015018111628610446,
      "loss": 0.2704,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 6760,
      "tokens_per_second_per_gpu": 380.02
    },
    {
      "epoch": 0.673062583884277,
      "grad_norm": 0.32067760825157166,
      "learning_rate": 0.00015004529962928164,
      "loss": 0.2495,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 6770,
      "tokens_per_second_per_gpu": 369.15
    },
    {
      "epoch": 0.6740567679077397,
      "grad_norm": 0.49704423546791077,
      "learning_rate": 0.0001499093597003085,
      "loss": 0.2095,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 6780,
      "tokens_per_second_per_gpu": 324.06
    },
    {
      "epoch": 0.6750509519312025,
      "grad_norm": 0.42155733704566956,
      "learning_rate": 0.00014977329683403385,
      "loss": 0.1743,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 6790,
      "tokens_per_second_per_gpu": 309.09
    },
    {
      "epoch": 0.6760451359546652,
      "grad_norm": 0.5538848638534546,
      "learning_rate": 0.00014963711136560924,
      "loss": 0.3424,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6800,
      "tokens_per_second_per_gpu": 347.86
    },
    {
      "epoch": 0.6770393199781279,
      "grad_norm": 0.3429434299468994,
      "learning_rate": 0.00014950080363048833,
      "loss": 0.2047,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 6810,
      "tokens_per_second_per_gpu": 341.24
    },
    {
      "epoch": 0.6780335040015907,
      "grad_norm": 0.39259403944015503,
      "learning_rate": 0.0001493643739644258,
      "loss": 0.187,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 6820,
      "tokens_per_second_per_gpu": 367.18
    },
    {
      "epoch": 0.6790276880250534,
      "grad_norm": 0.37642526626586914,
      "learning_rate": 0.00014922782270347686,
      "loss": 0.236,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 6830,
      "tokens_per_second_per_gpu": 324.51
    },
    {
      "epoch": 0.6800218720485162,
      "grad_norm": 0.5826324820518494,
      "learning_rate": 0.00014909115018399603,
      "loss": 0.2494,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.68,
      "memory/max_allocated (GiB)": 19.68,
      "step": 6840,
      "tokens_per_second_per_gpu": 343.26
    },
    {
      "epoch": 0.6810160560719789,
      "grad_norm": 0.39206662774086,
      "learning_rate": 0.00014895435674263662,
      "loss": 0.2522,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6850,
      "tokens_per_second_per_gpu": 330.16
    },
    {
      "epoch": 0.6820102400954416,
      "grad_norm": 0.21635837852954865,
      "learning_rate": 0.00014881744271634986,
      "loss": 0.2534,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 6860,
      "tokens_per_second_per_gpu": 319.9
    },
    {
      "epoch": 0.6830044241189044,
      "grad_norm": 0.25813058018684387,
      "learning_rate": 0.00014868040844238386,
      "loss": 0.2255,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 6870,
      "tokens_per_second_per_gpu": 352.96
    },
    {
      "epoch": 0.6839986081423671,
      "grad_norm": 0.46098119020462036,
      "learning_rate": 0.00014854325425828305,
      "loss": 0.2135,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 6880,
      "tokens_per_second_per_gpu": 278.19
    },
    {
      "epoch": 0.6849927921658299,
      "grad_norm": 0.45799604058265686,
      "learning_rate": 0.00014840598050188715,
      "loss": 0.2283,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6890,
      "tokens_per_second_per_gpu": 358.38
    },
    {
      "epoch": 0.6859869761892926,
      "grad_norm": 0.6016408205032349,
      "learning_rate": 0.00014826858751133042,
      "loss": 0.2261,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 6900,
      "tokens_per_second_per_gpu": 368.61
    },
    {
      "epoch": 0.6869811602127553,
      "grad_norm": 0.488506555557251,
      "learning_rate": 0.00014813107562504084,
      "loss": 0.2799,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 6910,
      "tokens_per_second_per_gpu": 367.4
    },
    {
      "epoch": 0.6879753442362181,
      "grad_norm": 0.6327788829803467,
      "learning_rate": 0.00014799344518173928,
      "loss": 0.1868,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 6920,
      "tokens_per_second_per_gpu": 300.97
    },
    {
      "epoch": 0.6889695282596808,
      "grad_norm": 0.4955579340457916,
      "learning_rate": 0.00014785569652043856,
      "loss": 0.2496,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 6930,
      "tokens_per_second_per_gpu": 343.13
    },
    {
      "epoch": 0.6899637122831436,
      "grad_norm": 0.5724585652351379,
      "learning_rate": 0.0001477178299804428,
      "loss": 0.2611,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.82,
      "memory/max_allocated (GiB)": 18.82,
      "step": 6940,
      "tokens_per_second_per_gpu": 326.77
    },
    {
      "epoch": 0.6909578963066063,
      "grad_norm": 0.2057613730430603,
      "learning_rate": 0.00014757984590134642,
      "loss": 0.1107,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 6950,
      "tokens_per_second_per_gpu": 293.99
    },
    {
      "epoch": 0.691952080330069,
      "grad_norm": 0.3206622004508972,
      "learning_rate": 0.00014744174462303334,
      "loss": 0.2379,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 6960,
      "tokens_per_second_per_gpu": 371.29
    },
    {
      "epoch": 0.6929462643535318,
      "grad_norm": 0.3926986753940582,
      "learning_rate": 0.00014730352648567623,
      "loss": 0.2558,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 6970,
      "tokens_per_second_per_gpu": 377.71
    },
    {
      "epoch": 0.6939404483769945,
      "grad_norm": 0.34591636061668396,
      "learning_rate": 0.00014716519182973552,
      "loss": 0.2601,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 6980,
      "tokens_per_second_per_gpu": 352.12
    },
    {
      "epoch": 0.6949346324004573,
      "grad_norm": 0.5908513069152832,
      "learning_rate": 0.00014702674099595876,
      "loss": 0.2027,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 6990,
      "tokens_per_second_per_gpu": 345.62
    },
    {
      "epoch": 0.69592881642392,
      "grad_norm": 0.3830493986606598,
      "learning_rate": 0.00014688817432537962,
      "loss": 0.1987,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 7000,
      "tokens_per_second_per_gpu": 359.97
    },
    {
      "epoch": 0.6969230004473828,
      "grad_norm": 0.444762647151947,
      "learning_rate": 0.00014674949215931707,
      "loss": 0.2059,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 7010,
      "tokens_per_second_per_gpu": 289.86
    },
    {
      "epoch": 0.6979171844708456,
      "grad_norm": 0.31576088070869446,
      "learning_rate": 0.00014661069483937458,
      "loss": 0.2115,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 7020,
      "tokens_per_second_per_gpu": 339.22
    },
    {
      "epoch": 0.6989113684943083,
      "grad_norm": 0.4755282700061798,
      "learning_rate": 0.00014647178270743932,
      "loss": 0.265,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 7030,
      "tokens_per_second_per_gpu": 329.94
    },
    {
      "epoch": 0.6999055525177711,
      "grad_norm": 0.4698229134082794,
      "learning_rate": 0.00014633275610568123,
      "loss": 0.2492,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 7040,
      "tokens_per_second_per_gpu": 315.54
    },
    {
      "epoch": 0.7008997365412338,
      "grad_norm": 0.3248315453529358,
      "learning_rate": 0.00014619361537655215,
      "loss": 0.2412,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 7050,
      "tokens_per_second_per_gpu": 294.7
    },
    {
      "epoch": 0.7018939205646966,
      "grad_norm": 0.48639553785324097,
      "learning_rate": 0.0001460543608627852,
      "loss": 0.2356,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 7060,
      "tokens_per_second_per_gpu": 280.17
    },
    {
      "epoch": 0.7028881045881593,
      "grad_norm": 0.5937051773071289,
      "learning_rate": 0.00014591499290739362,
      "loss": 0.1679,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 7070,
      "tokens_per_second_per_gpu": 336.7
    },
    {
      "epoch": 0.703882288611622,
      "grad_norm": 0.3488394021987915,
      "learning_rate": 0.00014577551185367013,
      "loss": 0.2474,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7080,
      "tokens_per_second_per_gpu": 400.34
    },
    {
      "epoch": 0.7048764726350848,
      "grad_norm": 0.4485851526260376,
      "learning_rate": 0.0001456359180451861,
      "loss": 0.2709,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 7090,
      "tokens_per_second_per_gpu": 392.29
    },
    {
      "epoch": 0.7058706566585475,
      "grad_norm": 0.4746951758861542,
      "learning_rate": 0.00014549621182579055,
      "loss": 0.2353,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 7100,
      "tokens_per_second_per_gpu": 330.73
    },
    {
      "epoch": 0.7068648406820103,
      "grad_norm": 0.5027205944061279,
      "learning_rate": 0.00014535639353960942,
      "loss": 0.2576,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 7110,
      "tokens_per_second_per_gpu": 379.65
    },
    {
      "epoch": 0.707859024705473,
      "grad_norm": 0.449788361787796,
      "learning_rate": 0.00014521646353104472,
      "loss": 0.2186,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 7120,
      "tokens_per_second_per_gpu": 269.74
    },
    {
      "epoch": 0.7088532087289358,
      "grad_norm": 0.31661751866340637,
      "learning_rate": 0.00014507642214477362,
      "loss": 0.2481,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 7130,
      "tokens_per_second_per_gpu": 302.22
    },
    {
      "epoch": 0.7098473927523985,
      "grad_norm": 0.3295125663280487,
      "learning_rate": 0.00014493626972574765,
      "loss": 0.2284,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7140,
      "tokens_per_second_per_gpu": 350.96
    },
    {
      "epoch": 0.7108415767758612,
      "grad_norm": 0.5383651256561279,
      "learning_rate": 0.0001447960066191919,
      "loss": 0.2427,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 7150,
      "tokens_per_second_per_gpu": 345.67
    },
    {
      "epoch": 0.711835760799324,
      "grad_norm": 0.3970474898815155,
      "learning_rate": 0.00014465563317060394,
      "loss": 0.2434,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 7160,
      "tokens_per_second_per_gpu": 374.3
    },
    {
      "epoch": 0.7128299448227867,
      "grad_norm": 0.16766348481178284,
      "learning_rate": 0.00014451514972575332,
      "loss": 0.1649,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 7170,
      "tokens_per_second_per_gpu": 304.14
    },
    {
      "epoch": 0.7138241288462495,
      "grad_norm": 0.4426742196083069,
      "learning_rate": 0.00014437455663068042,
      "loss": 0.2633,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 7180,
      "tokens_per_second_per_gpu": 374.18
    },
    {
      "epoch": 0.7148183128697122,
      "grad_norm": 0.4757481515407562,
      "learning_rate": 0.00014423385423169575,
      "loss": 0.2584,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 7190,
      "tokens_per_second_per_gpu": 302.88
    },
    {
      "epoch": 0.7158124968931749,
      "grad_norm": 0.4964188039302826,
      "learning_rate": 0.00014409304287537906,
      "loss": 0.2386,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 7200,
      "tokens_per_second_per_gpu": 325.69
    },
    {
      "epoch": 0.7168066809166377,
      "grad_norm": 0.5026222467422485,
      "learning_rate": 0.0001439521229085785,
      "loss": 0.2161,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 7210,
      "tokens_per_second_per_gpu": 332.92
    },
    {
      "epoch": 0.7178008649401004,
      "grad_norm": 0.41850724816322327,
      "learning_rate": 0.00014381109467840976,
      "loss": 0.2157,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 7220,
      "tokens_per_second_per_gpu": 328.35
    },
    {
      "epoch": 0.7187950489635632,
      "grad_norm": 0.3922070264816284,
      "learning_rate": 0.00014366995853225514,
      "loss": 0.2112,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 7230,
      "tokens_per_second_per_gpu": 263.36
    },
    {
      "epoch": 0.7197892329870259,
      "grad_norm": 0.5679214000701904,
      "learning_rate": 0.0001435287148177628,
      "loss": 0.2715,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7240,
      "tokens_per_second_per_gpu": 434.11
    },
    {
      "epoch": 0.7207834170104886,
      "grad_norm": 0.5302831530570984,
      "learning_rate": 0.0001433873638828458,
      "loss": 0.252,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 7250,
      "tokens_per_second_per_gpu": 358.75
    },
    {
      "epoch": 0.7217776010339514,
      "grad_norm": 0.49475687742233276,
      "learning_rate": 0.00014324590607568149,
      "loss": 0.2613,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 7260,
      "tokens_per_second_per_gpu": 377.3
    },
    {
      "epoch": 0.7227717850574141,
      "grad_norm": 0.4263441264629364,
      "learning_rate": 0.00014310434174471024,
      "loss": 0.288,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7270,
      "tokens_per_second_per_gpu": 378.27
    },
    {
      "epoch": 0.7237659690808769,
      "grad_norm": 0.4663153886795044,
      "learning_rate": 0.000142962671238635,
      "loss": 0.2572,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7280,
      "tokens_per_second_per_gpu": 390.5
    },
    {
      "epoch": 0.7247601531043396,
      "grad_norm": 0.3563691973686218,
      "learning_rate": 0.0001428208949064201,
      "loss": 0.2024,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 7290,
      "tokens_per_second_per_gpu": 330.99
    },
    {
      "epoch": 0.7257543371278024,
      "grad_norm": 0.2805791199207306,
      "learning_rate": 0.00014267901309729066,
      "loss": 0.2371,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 7300,
      "tokens_per_second_per_gpu": 410.4
    },
    {
      "epoch": 0.7267485211512651,
      "grad_norm": 0.30967897176742554,
      "learning_rate": 0.00014253702616073155,
      "loss": 0.231,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 7310,
      "tokens_per_second_per_gpu": 302.94
    },
    {
      "epoch": 0.7277427051747278,
      "grad_norm": 0.353834867477417,
      "learning_rate": 0.00014239493444648658,
      "loss": 0.1885,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 7320,
      "tokens_per_second_per_gpu": 296.75
    },
    {
      "epoch": 0.7287368891981906,
      "grad_norm": 0.32480210065841675,
      "learning_rate": 0.00014225273830455773,
      "loss": 0.2713,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7330,
      "tokens_per_second_per_gpu": 389.69
    },
    {
      "epoch": 0.7297310732216533,
      "grad_norm": 0.6818671226501465,
      "learning_rate": 0.00014211043808520405,
      "loss": 0.3248,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 7340,
      "tokens_per_second_per_gpu": 355.32
    },
    {
      "epoch": 0.730725257245116,
      "grad_norm": 0.5786187648773193,
      "learning_rate": 0.0001419680341389412,
      "loss": 0.2262,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7350,
      "tokens_per_second_per_gpu": 321.29
    },
    {
      "epoch": 0.7317194412685788,
      "grad_norm": 0.5133084058761597,
      "learning_rate": 0.0001418255268165401,
      "loss": 0.2653,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 7360,
      "tokens_per_second_per_gpu": 348.43
    },
    {
      "epoch": 0.7327136252920415,
      "grad_norm": 0.4247760474681854,
      "learning_rate": 0.0001416829164690264,
      "loss": 0.2312,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 7370,
      "tokens_per_second_per_gpu": 360.14
    },
    {
      "epoch": 0.7337078093155043,
      "grad_norm": 0.32232165336608887,
      "learning_rate": 0.00014154020344767955,
      "loss": 0.2825,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 7380,
      "tokens_per_second_per_gpu": 379.26
    },
    {
      "epoch": 0.734701993338967,
      "grad_norm": 0.4452918767929077,
      "learning_rate": 0.0001413973881040319,
      "loss": 0.2205,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7390,
      "tokens_per_second_per_gpu": 359.25
    },
    {
      "epoch": 0.7356961773624298,
      "grad_norm": 0.3855791985988617,
      "learning_rate": 0.0001412544707898678,
      "loss": 0.2868,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 7400,
      "tokens_per_second_per_gpu": 410.81
    },
    {
      "epoch": 0.7366903613858925,
      "grad_norm": 0.42609113454818726,
      "learning_rate": 0.00014111145185722283,
      "loss": 0.2523,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 7410,
      "tokens_per_second_per_gpu": 331.0
    },
    {
      "epoch": 0.7376845454093552,
      "grad_norm": 0.47836732864379883,
      "learning_rate": 0.00014096833165838283,
      "loss": 0.2962,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 7420,
      "tokens_per_second_per_gpu": 344.4
    },
    {
      "epoch": 0.738678729432818,
      "grad_norm": 0.508818507194519,
      "learning_rate": 0.0001408251105458831,
      "loss": 0.3254,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7430,
      "tokens_per_second_per_gpu": 389.2
    },
    {
      "epoch": 0.7396729134562807,
      "grad_norm": 0.3887844681739807,
      "learning_rate": 0.00014068178887250752,
      "loss": 0.2353,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 7440,
      "tokens_per_second_per_gpu": 380.02
    },
    {
      "epoch": 0.7406670974797435,
      "grad_norm": 0.41547468304634094,
      "learning_rate": 0.00014053836699128765,
      "loss": 0.2424,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7450,
      "tokens_per_second_per_gpu": 376.26
    },
    {
      "epoch": 0.7416612815032062,
      "grad_norm": 0.5015019178390503,
      "learning_rate": 0.00014039484525550186,
      "loss": 0.2329,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 7460,
      "tokens_per_second_per_gpu": 351.38
    },
    {
      "epoch": 0.742655465526669,
      "grad_norm": 0.43546929955482483,
      "learning_rate": 0.0001402512240186746,
      "loss": 0.2186,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 7470,
      "tokens_per_second_per_gpu": 303.33
    },
    {
      "epoch": 0.7436496495501317,
      "grad_norm": 0.5051418542861938,
      "learning_rate": 0.0001401075036345753,
      "loss": 0.2439,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.19,
      "memory/max_allocated (GiB)": 19.19,
      "step": 7480,
      "tokens_per_second_per_gpu": 375.31
    },
    {
      "epoch": 0.7446438335735944,
      "grad_norm": 0.35766085982322693,
      "learning_rate": 0.0001399636844572176,
      "loss": 0.277,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 7490,
      "tokens_per_second_per_gpu": 343.63
    },
    {
      "epoch": 0.7456380175970572,
      "grad_norm": 0.5930467247962952,
      "learning_rate": 0.0001398197668408586,
      "loss": 0.2474,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 7500,
      "tokens_per_second_per_gpu": 396.18
    },
    {
      "epoch": 0.7466322016205199,
      "grad_norm": 0.4920576810836792,
      "learning_rate": 0.00013967575113999777,
      "loss": 0.2408,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 7510,
      "tokens_per_second_per_gpu": 382.95
    },
    {
      "epoch": 0.7476263856439826,
      "grad_norm": 0.44312262535095215,
      "learning_rate": 0.0001395316377093762,
      "loss": 0.2249,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 7520,
      "tokens_per_second_per_gpu": 387.85
    },
    {
      "epoch": 0.7486205696674454,
      "grad_norm": 0.4043440818786621,
      "learning_rate": 0.00013938742690397575,
      "loss": 0.2141,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 7530,
      "tokens_per_second_per_gpu": 360.87
    },
    {
      "epoch": 0.7496147536909082,
      "grad_norm": 0.3910767138004303,
      "learning_rate": 0.00013924311907901813,
      "loss": 0.1528,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 7540,
      "tokens_per_second_per_gpu": 308.96
    },
    {
      "epoch": 0.750608937714371,
      "grad_norm": 0.3407839238643646,
      "learning_rate": 0.00013909871458996399,
      "loss": 0.2192,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 7550,
      "tokens_per_second_per_gpu": 316.74
    },
    {
      "epoch": 0.7516031217378337,
      "grad_norm": 0.316240519285202,
      "learning_rate": 0.00013895421379251207,
      "loss": 0.2317,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 7560,
      "tokens_per_second_per_gpu": 332.89
    },
    {
      "epoch": 0.7525973057612965,
      "grad_norm": 0.49255135655403137,
      "learning_rate": 0.00013880961704259846,
      "loss": 0.2413,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7570,
      "tokens_per_second_per_gpu": 352.94
    },
    {
      "epoch": 0.7535914897847592,
      "grad_norm": 0.4979618489742279,
      "learning_rate": 0.0001386649246963955,
      "loss": 0.2434,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 7580,
      "tokens_per_second_per_gpu": 307.59
    },
    {
      "epoch": 0.754585673808222,
      "grad_norm": 0.2949107885360718,
      "learning_rate": 0.00013852013711031095,
      "loss": 0.2112,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 7590,
      "tokens_per_second_per_gpu": 363.41
    },
    {
      "epoch": 0.7555798578316847,
      "grad_norm": 0.3708727955818176,
      "learning_rate": 0.0001383752546409873,
      "loss": 0.2232,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 7600,
      "tokens_per_second_per_gpu": 386.01
    },
    {
      "epoch": 0.7565740418551474,
      "grad_norm": 0.6432907581329346,
      "learning_rate": 0.00013823027764530067,
      "loss": 0.2707,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 7610,
      "tokens_per_second_per_gpu": 313.62
    },
    {
      "epoch": 0.7575682258786102,
      "grad_norm": 0.3710288405418396,
      "learning_rate": 0.00013808520648036005,
      "loss": 0.2353,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 7620,
      "tokens_per_second_per_gpu": 432.88
    },
    {
      "epoch": 0.7585624099020729,
      "grad_norm": 0.3577297031879425,
      "learning_rate": 0.00013794004150350636,
      "loss": 0.212,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 7630,
      "tokens_per_second_per_gpu": 323.27
    },
    {
      "epoch": 0.7595565939255357,
      "grad_norm": 0.4883553385734558,
      "learning_rate": 0.00013779478307231164,
      "loss": 0.2747,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7640,
      "tokens_per_second_per_gpu": 292.24
    },
    {
      "epoch": 0.7605507779489984,
      "grad_norm": 0.19372917711734772,
      "learning_rate": 0.00013764943154457812,
      "loss": 0.233,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7650,
      "tokens_per_second_per_gpu": 375.4
    },
    {
      "epoch": 0.7615449619724611,
      "grad_norm": 0.46450668573379517,
      "learning_rate": 0.00013750398727833735,
      "loss": 0.219,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 7660,
      "tokens_per_second_per_gpu": 351.49
    },
    {
      "epoch": 0.7625391459959239,
      "grad_norm": 0.3964915871620178,
      "learning_rate": 0.00013735845063184921,
      "loss": 0.2376,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7670,
      "tokens_per_second_per_gpu": 309.02
    },
    {
      "epoch": 0.7635333300193866,
      "grad_norm": 0.6207079887390137,
      "learning_rate": 0.00013721282196360127,
      "loss": 0.2547,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7680,
      "tokens_per_second_per_gpu": 338.05
    },
    {
      "epoch": 0.7645275140428494,
      "grad_norm": 0.2084685117006302,
      "learning_rate": 0.00013706710163230773,
      "loss": 0.2504,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 7690,
      "tokens_per_second_per_gpu": 294.32
    },
    {
      "epoch": 0.7655216980663121,
      "grad_norm": 0.4136933386325836,
      "learning_rate": 0.0001369212899969086,
      "loss": 0.1809,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7700,
      "tokens_per_second_per_gpu": 314.54
    },
    {
      "epoch": 0.7665158820897748,
      "grad_norm": 0.529629111289978,
      "learning_rate": 0.0001367753874165687,
      "loss": 0.255,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7710,
      "tokens_per_second_per_gpu": 357.08
    },
    {
      "epoch": 0.7675100661132376,
      "grad_norm": 0.36684682965278625,
      "learning_rate": 0.0001366293942506769,
      "loss": 0.2128,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 7720,
      "tokens_per_second_per_gpu": 318.04
    },
    {
      "epoch": 0.7685042501367003,
      "grad_norm": 0.40612316131591797,
      "learning_rate": 0.00013648331085884527,
      "loss": 0.2159,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 7730,
      "tokens_per_second_per_gpu": 364.09
    },
    {
      "epoch": 0.7694984341601631,
      "grad_norm": 0.13119497895240784,
      "learning_rate": 0.0001363371376009081,
      "loss": 0.2255,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 7740,
      "tokens_per_second_per_gpu": 336.24
    },
    {
      "epoch": 0.7704926181836258,
      "grad_norm": 0.5006715655326843,
      "learning_rate": 0.00013619087483692099,
      "loss": 0.2595,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7750,
      "tokens_per_second_per_gpu": 338.82
    },
    {
      "epoch": 0.7714868022070885,
      "grad_norm": 0.3994678258895874,
      "learning_rate": 0.00013604452292716003,
      "loss": 0.203,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 7760,
      "tokens_per_second_per_gpu": 327.9
    },
    {
      "epoch": 0.7724809862305513,
      "grad_norm": 0.17447052896022797,
      "learning_rate": 0.00013589808223212087,
      "loss": 0.2537,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 7770,
      "tokens_per_second_per_gpu": 376.09
    },
    {
      "epoch": 0.773475170254014,
      "grad_norm": 0.5262983441352844,
      "learning_rate": 0.000135751553112518,
      "loss": 0.2112,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 7780,
      "tokens_per_second_per_gpu": 338.13
    },
    {
      "epoch": 0.7744693542774768,
      "grad_norm": 0.32633262872695923,
      "learning_rate": 0.00013560493592928356,
      "loss": 0.2235,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 7790,
      "tokens_per_second_per_gpu": 331.86
    },
    {
      "epoch": 0.7754635383009395,
      "grad_norm": 0.4296337068080902,
      "learning_rate": 0.00013545823104356663,
      "loss": 0.297,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7800,
      "tokens_per_second_per_gpu": 370.8
    },
    {
      "epoch": 0.7764577223244022,
      "grad_norm": 0.5057851672172546,
      "learning_rate": 0.00013531143881673237,
      "loss": 0.1952,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 7810,
      "tokens_per_second_per_gpu": 321.52
    },
    {
      "epoch": 0.777451906347865,
      "grad_norm": 0.49617013335227966,
      "learning_rate": 0.00013516455961036104,
      "loss": 0.2589,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 7820,
      "tokens_per_second_per_gpu": 299.57
    },
    {
      "epoch": 0.7784460903713277,
      "grad_norm": 0.3173094689846039,
      "learning_rate": 0.00013501759378624722,
      "loss": 0.2328,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7830,
      "tokens_per_second_per_gpu": 349.07
    },
    {
      "epoch": 0.7794402743947905,
      "grad_norm": 0.4631012976169586,
      "learning_rate": 0.00013487054170639877,
      "loss": 0.2472,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 7840,
      "tokens_per_second_per_gpu": 361.13
    },
    {
      "epoch": 0.7804344584182532,
      "grad_norm": 0.3672430217266083,
      "learning_rate": 0.000134723403733036,
      "loss": 0.2183,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 7850,
      "tokens_per_second_per_gpu": 324.51
    },
    {
      "epoch": 0.781428642441716,
      "grad_norm": 0.5141401886940002,
      "learning_rate": 0.00013457618022859092,
      "loss": 0.3104,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.47,
      "memory/max_allocated (GiB)": 21.47,
      "step": 7860,
      "tokens_per_second_per_gpu": 384.61
    },
    {
      "epoch": 0.7824228264651787,
      "grad_norm": 0.43661215901374817,
      "learning_rate": 0.00013442887155570607,
      "loss": 0.2228,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 7870,
      "tokens_per_second_per_gpu": 359.28
    },
    {
      "epoch": 0.7834170104886414,
      "grad_norm": 0.375987708568573,
      "learning_rate": 0.00013428147807723387,
      "loss": 0.2215,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 7880,
      "tokens_per_second_per_gpu": 327.47
    },
    {
      "epoch": 0.7844111945121042,
      "grad_norm": 0.2800423204898834,
      "learning_rate": 0.00013413400015623562,
      "loss": 0.263,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 7890,
      "tokens_per_second_per_gpu": 367.65
    },
    {
      "epoch": 0.7854053785355669,
      "grad_norm": 0.44610151648521423,
      "learning_rate": 0.00013398643815598063,
      "loss": 0.2533,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7900,
      "tokens_per_second_per_gpu": 364.29
    },
    {
      "epoch": 0.7863995625590297,
      "grad_norm": 0.6232859492301941,
      "learning_rate": 0.0001338387924399452,
      "loss": 0.2273,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 7910,
      "tokens_per_second_per_gpu": 335.13
    },
    {
      "epoch": 0.7873937465824924,
      "grad_norm": 0.3155955374240875,
      "learning_rate": 0.00013369106337181202,
      "loss": 0.2007,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 7920,
      "tokens_per_second_per_gpu": 357.92
    },
    {
      "epoch": 0.7883879306059551,
      "grad_norm": 0.47753843665122986,
      "learning_rate": 0.00013354325131546902,
      "loss": 0.1722,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 7930,
      "tokens_per_second_per_gpu": 350.28
    },
    {
      "epoch": 0.7893821146294179,
      "grad_norm": 0.6098092198371887,
      "learning_rate": 0.0001333953566350085,
      "loss": 0.2172,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 7940,
      "tokens_per_second_per_gpu": 328.34
    },
    {
      "epoch": 0.7903762986528806,
      "grad_norm": 0.40892454981803894,
      "learning_rate": 0.00013324737969472628,
      "loss": 0.2365,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 7950,
      "tokens_per_second_per_gpu": 376.03
    },
    {
      "epoch": 0.7913704826763434,
      "grad_norm": 0.6622501015663147,
      "learning_rate": 0.00013309932085912092,
      "loss": 0.265,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 7960,
      "tokens_per_second_per_gpu": 401.88
    },
    {
      "epoch": 0.7923646666998061,
      "grad_norm": 0.5111701488494873,
      "learning_rate": 0.00013295118049289255,
      "loss": 0.2164,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 7970,
      "tokens_per_second_per_gpu": 359.69
    },
    {
      "epoch": 0.7933588507232688,
      "grad_norm": 0.5144445300102234,
      "learning_rate": 0.00013280295896094224,
      "loss": 0.2567,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 7980,
      "tokens_per_second_per_gpu": 301.89
    },
    {
      "epoch": 0.7943530347467316,
      "grad_norm": 0.5570478439331055,
      "learning_rate": 0.00013265465662837093,
      "loss": 0.1934,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 7990,
      "tokens_per_second_per_gpu": 336.55
    },
    {
      "epoch": 0.7953472187701943,
      "grad_norm": 0.33518052101135254,
      "learning_rate": 0.00013250627386047866,
      "loss": 0.2247,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 8000,
      "tokens_per_second_per_gpu": 335.15
    },
    {
      "epoch": 0.7963414027936571,
      "grad_norm": 0.508229672908783,
      "learning_rate": 0.0001323578110227635,
      "loss": 0.1587,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 8010,
      "tokens_per_second_per_gpu": 351.3
    },
    {
      "epoch": 0.7973355868171198,
      "grad_norm": 0.3688840866088867,
      "learning_rate": 0.0001322092684809208,
      "loss": 0.1929,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 8020,
      "tokens_per_second_per_gpu": 289.68
    },
    {
      "epoch": 0.7983297708405825,
      "grad_norm": 0.31160444021224976,
      "learning_rate": 0.00013206064660084227,
      "loss": 0.2318,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8030,
      "tokens_per_second_per_gpu": 351.62
    },
    {
      "epoch": 0.7993239548640453,
      "grad_norm": 0.46203359961509705,
      "learning_rate": 0.000131911945748615,
      "loss": 0.2808,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 8040,
      "tokens_per_second_per_gpu": 360.06
    },
    {
      "epoch": 0.800318138887508,
      "grad_norm": 0.4416126608848572,
      "learning_rate": 0.00013176316629052054,
      "loss": 0.2065,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 8050,
      "tokens_per_second_per_gpu": 312.23
    },
    {
      "epoch": 0.8013123229109709,
      "grad_norm": 0.16777446866035461,
      "learning_rate": 0.00013161430859303427,
      "loss": 0.1713,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 8060,
      "tokens_per_second_per_gpu": 347.86
    },
    {
      "epoch": 0.8023065069344336,
      "grad_norm": 0.3447447121143341,
      "learning_rate": 0.0001314653730228241,
      "loss": 0.1954,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8070,
      "tokens_per_second_per_gpu": 372.63
    },
    {
      "epoch": 0.8033006909578964,
      "grad_norm": 0.4045270085334778,
      "learning_rate": 0.0001313163599467498,
      "loss": 0.1929,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 8080,
      "tokens_per_second_per_gpu": 355.89
    },
    {
      "epoch": 0.8042948749813591,
      "grad_norm": 0.462365984916687,
      "learning_rate": 0.00013116726973186208,
      "loss": 0.2551,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 8090,
      "tokens_per_second_per_gpu": 361.83
    },
    {
      "epoch": 0.8052890590048218,
      "grad_norm": 0.5786636471748352,
      "learning_rate": 0.00013101810274540168,
      "loss": 0.2499,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 8100,
      "tokens_per_second_per_gpu": 300.27
    },
    {
      "epoch": 0.8062832430282846,
      "grad_norm": 0.3487481474876404,
      "learning_rate": 0.0001308688593547984,
      "loss": 0.2186,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 17.11,
      "memory/max_allocated (GiB)": 17.11,
      "step": 8110,
      "tokens_per_second_per_gpu": 312.52
    },
    {
      "epoch": 0.8072774270517473,
      "grad_norm": 0.3610248863697052,
      "learning_rate": 0.00013071953992767015,
      "loss": 0.2167,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 8120,
      "tokens_per_second_per_gpu": 328.39
    },
    {
      "epoch": 0.8082716110752101,
      "grad_norm": 0.37153443694114685,
      "learning_rate": 0.00013057014483182242,
      "loss": 0.241,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 8130,
      "tokens_per_second_per_gpu": 344.6
    },
    {
      "epoch": 0.8092657950986728,
      "grad_norm": 0.3705120086669922,
      "learning_rate": 0.00013042067443524681,
      "loss": 0.2749,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 8140,
      "tokens_per_second_per_gpu": 332.87
    },
    {
      "epoch": 0.8102599791221355,
      "grad_norm": 0.3014324903488159,
      "learning_rate": 0.00013027112910612052,
      "loss": 0.1438,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 8150,
      "tokens_per_second_per_gpu": 308.05
    },
    {
      "epoch": 0.8112541631455983,
      "grad_norm": 0.5011573433876038,
      "learning_rate": 0.00013012150921280527,
      "loss": 0.2032,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 8160,
      "tokens_per_second_per_gpu": 296.89
    },
    {
      "epoch": 0.812248347169061,
      "grad_norm": 0.3287215530872345,
      "learning_rate": 0.00012997181512384653,
      "loss": 0.2055,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 8170,
      "tokens_per_second_per_gpu": 405.75
    },
    {
      "epoch": 0.8132425311925238,
      "grad_norm": 0.7308348417282104,
      "learning_rate": 0.00012982204720797245,
      "loss": 0.2805,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 8180,
      "tokens_per_second_per_gpu": 443.15
    },
    {
      "epoch": 0.8142367152159865,
      "grad_norm": 0.5808447599411011,
      "learning_rate": 0.00012967220583409304,
      "loss": 0.2066,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 8190,
      "tokens_per_second_per_gpu": 401.43
    },
    {
      "epoch": 0.8152308992394492,
      "grad_norm": 0.31127744913101196,
      "learning_rate": 0.0001295222913712993,
      "loss": 0.2464,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 8200,
      "tokens_per_second_per_gpu": 343.59
    },
    {
      "epoch": 0.816225083262912,
      "grad_norm": 0.2803351581096649,
      "learning_rate": 0.00012937230418886224,
      "loss": 0.1986,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 8210,
      "tokens_per_second_per_gpu": 401.58
    },
    {
      "epoch": 0.8172192672863747,
      "grad_norm": 0.45312055945396423,
      "learning_rate": 0.000129222244656232,
      "loss": 0.1904,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 8220,
      "tokens_per_second_per_gpu": 290.19
    },
    {
      "epoch": 0.8182134513098375,
      "grad_norm": 0.4169121980667114,
      "learning_rate": 0.0001290721131430369,
      "loss": 0.2074,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 8230,
      "tokens_per_second_per_gpu": 320.72
    },
    {
      "epoch": 0.8192076353333002,
      "grad_norm": 0.5725305080413818,
      "learning_rate": 0.0001289219100190826,
      "loss": 0.2809,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 8240,
      "tokens_per_second_per_gpu": 334.56
    },
    {
      "epoch": 0.820201819356763,
      "grad_norm": 0.4698241055011749,
      "learning_rate": 0.00012877163565435114,
      "loss": 0.1873,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.66,
      "memory/max_allocated (GiB)": 19.66,
      "step": 8250,
      "tokens_per_second_per_gpu": 319.13
    },
    {
      "epoch": 0.8211960033802257,
      "grad_norm": 0.33453720808029175,
      "learning_rate": 0.000128621290419,
      "loss": 0.2172,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 8260,
      "tokens_per_second_per_gpu": 306.65
    },
    {
      "epoch": 0.8221901874036884,
      "grad_norm": 0.20304298400878906,
      "learning_rate": 0.00012847087468336135,
      "loss": 0.2102,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8270,
      "tokens_per_second_per_gpu": 311.33
    },
    {
      "epoch": 0.8231843714271512,
      "grad_norm": 0.27367445826530457,
      "learning_rate": 0.00012832038881794086,
      "loss": 0.2437,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 8280,
      "tokens_per_second_per_gpu": 417.4
    },
    {
      "epoch": 0.8241785554506139,
      "grad_norm": 0.3914351463317871,
      "learning_rate": 0.00012816983319341712,
      "loss": 0.2692,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 8290,
      "tokens_per_second_per_gpu": 342.88
    },
    {
      "epoch": 0.8251727394740767,
      "grad_norm": 0.5103800296783447,
      "learning_rate": 0.00012801920818064034,
      "loss": 0.2341,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 8300,
      "tokens_per_second_per_gpu": 362.66
    },
    {
      "epoch": 0.8261669234975394,
      "grad_norm": 0.1988827884197235,
      "learning_rate": 0.00012786851415063185,
      "loss": 0.2141,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 8310,
      "tokens_per_second_per_gpu": 404.61
    },
    {
      "epoch": 0.8271611075210021,
      "grad_norm": 0.482526570558548,
      "learning_rate": 0.00012771775147458288,
      "loss": 0.2341,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8320,
      "tokens_per_second_per_gpu": 366.94
    },
    {
      "epoch": 0.8281552915444649,
      "grad_norm": 0.5179364085197449,
      "learning_rate": 0.0001275669205238537,
      "loss": 0.2458,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 8330,
      "tokens_per_second_per_gpu": 308.79
    },
    {
      "epoch": 0.8291494755679276,
      "grad_norm": 0.4961225390434265,
      "learning_rate": 0.00012741602166997288,
      "loss": 0.2324,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 8340,
      "tokens_per_second_per_gpu": 312.63
    },
    {
      "epoch": 0.8301436595913904,
      "grad_norm": 0.6281317472457886,
      "learning_rate": 0.0001272650552846362,
      "loss": 0.2808,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 8350,
      "tokens_per_second_per_gpu": 365.38
    },
    {
      "epoch": 0.8311378436148531,
      "grad_norm": 0.3268338739871979,
      "learning_rate": 0.00012711402173970574,
      "loss": 0.2125,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8360,
      "tokens_per_second_per_gpu": 381.45
    },
    {
      "epoch": 0.8321320276383158,
      "grad_norm": 0.5050214529037476,
      "learning_rate": 0.00012696292140720907,
      "loss": 0.3039,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.67,
      "memory/max_allocated (GiB)": 19.67,
      "step": 8370,
      "tokens_per_second_per_gpu": 394.88
    },
    {
      "epoch": 0.8331262116617786,
      "grad_norm": 0.3680170178413391,
      "learning_rate": 0.00012681175465933822,
      "loss": 0.1876,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 8380,
      "tokens_per_second_per_gpu": 310.52
    },
    {
      "epoch": 0.8341203956852413,
      "grad_norm": 0.6084402799606323,
      "learning_rate": 0.00012666052186844883,
      "loss": 0.2137,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 8390,
      "tokens_per_second_per_gpu": 308.84
    },
    {
      "epoch": 0.8351145797087041,
      "grad_norm": 0.41174978017807007,
      "learning_rate": 0.00012650922340705925,
      "loss": 0.2423,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 8400,
      "tokens_per_second_per_gpu": 389.04
    },
    {
      "epoch": 0.8361087637321668,
      "grad_norm": 0.3895050883293152,
      "learning_rate": 0.0001263578596478496,
      "loss": 0.2144,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.99,
      "memory/max_allocated (GiB)": 19.99,
      "step": 8410,
      "tokens_per_second_per_gpu": 325.56
    },
    {
      "epoch": 0.8371029477556295,
      "grad_norm": 0.2492019683122635,
      "learning_rate": 0.00012620643096366077,
      "loss": 0.2292,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 8420,
      "tokens_per_second_per_gpu": 312.76
    },
    {
      "epoch": 0.8380971317790923,
      "grad_norm": 0.5276951789855957,
      "learning_rate": 0.0001260549377274936,
      "loss": 0.2566,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 8430,
      "tokens_per_second_per_gpu": 341.75
    },
    {
      "epoch": 0.839091315802555,
      "grad_norm": 0.3949599266052246,
      "learning_rate": 0.00012590338031250796,
      "loss": 0.2108,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 8440,
      "tokens_per_second_per_gpu": 370.88
    },
    {
      "epoch": 0.8400854998260178,
      "grad_norm": 0.3476475179195404,
      "learning_rate": 0.00012575175909202186,
      "loss": 0.1811,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 8450,
      "tokens_per_second_per_gpu": 336.59
    },
    {
      "epoch": 0.8410796838494805,
      "grad_norm": 0.26099368929862976,
      "learning_rate": 0.00012560007443951032,
      "loss": 0.2144,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 8460,
      "tokens_per_second_per_gpu": 353.23
    },
    {
      "epoch": 0.8420738678729432,
      "grad_norm": 0.4799298346042633,
      "learning_rate": 0.00012544832672860474,
      "loss": 0.1781,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8470,
      "tokens_per_second_per_gpu": 360.11
    },
    {
      "epoch": 0.843068051896406,
      "grad_norm": 0.46986958384513855,
      "learning_rate": 0.0001252965163330918,
      "loss": 0.209,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 8480,
      "tokens_per_second_per_gpu": 393.35
    },
    {
      "epoch": 0.8440622359198687,
      "grad_norm": 1.179408311843872,
      "learning_rate": 0.00012514464362691258,
      "loss": 0.2061,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8490,
      "tokens_per_second_per_gpu": 377.29
    },
    {
      "epoch": 0.8450564199433315,
      "grad_norm": 0.2782179117202759,
      "learning_rate": 0.0001249927089841617,
      "loss": 0.2643,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8500,
      "tokens_per_second_per_gpu": 352.65
    },
    {
      "epoch": 0.8460506039667942,
      "grad_norm": 0.5103908181190491,
      "learning_rate": 0.00012484071277908622,
      "loss": 0.2086,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 8510,
      "tokens_per_second_per_gpu": 313.7
    },
    {
      "epoch": 0.847044787990257,
      "grad_norm": 0.34362590312957764,
      "learning_rate": 0.000124688655386085,
      "loss": 0.2545,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8520,
      "tokens_per_second_per_gpu": 357.23
    },
    {
      "epoch": 0.8480389720137197,
      "grad_norm": 0.5372098088264465,
      "learning_rate": 0.00012453653717970747,
      "loss": 0.2191,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 8530,
      "tokens_per_second_per_gpu": 275.82
    },
    {
      "epoch": 0.8490331560371824,
      "grad_norm": 0.21182872354984283,
      "learning_rate": 0.00012438435853465296,
      "loss": 0.2291,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 8540,
      "tokens_per_second_per_gpu": 394.59
    },
    {
      "epoch": 0.8500273400606452,
      "grad_norm": 0.4921363890171051,
      "learning_rate": 0.0001242321198257696,
      "loss": 0.2142,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 8550,
      "tokens_per_second_per_gpu": 342.56
    },
    {
      "epoch": 0.8510215240841079,
      "grad_norm": 0.3340144157409668,
      "learning_rate": 0.00012407982142805356,
      "loss": 0.2034,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 8560,
      "tokens_per_second_per_gpu": 381.5
    },
    {
      "epoch": 0.8520157081075707,
      "grad_norm": 0.45434701442718506,
      "learning_rate": 0.00012392746371664797,
      "loss": 0.2031,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 8570,
      "tokens_per_second_per_gpu": 337.22
    },
    {
      "epoch": 0.8530098921310335,
      "grad_norm": 0.2646021544933319,
      "learning_rate": 0.00012377504706684206,
      "loss": 0.1807,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8580,
      "tokens_per_second_per_gpu": 369.87
    },
    {
      "epoch": 0.8540040761544962,
      "grad_norm": 0.4125272333621979,
      "learning_rate": 0.00012362257185407022,
      "loss": 0.2258,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 8590,
      "tokens_per_second_per_gpu": 374.23
    },
    {
      "epoch": 0.854998260177959,
      "grad_norm": 0.3133401572704315,
      "learning_rate": 0.00012347003845391118,
      "loss": 0.1624,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 19.98,
      "memory/max_allocated (GiB)": 19.98,
      "step": 8600,
      "tokens_per_second_per_gpu": 336.61
    },
    {
      "epoch": 0.8559924442014217,
      "grad_norm": 0.23082919418811798,
      "learning_rate": 0.00012331744724208694,
      "loss": 0.207,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 8610,
      "tokens_per_second_per_gpu": 316.78
    },
    {
      "epoch": 0.8569866282248845,
      "grad_norm": 0.45513761043548584,
      "learning_rate": 0.00012316479859446187,
      "loss": 0.2465,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 8620,
      "tokens_per_second_per_gpu": 355.67
    },
    {
      "epoch": 0.8579808122483472,
      "grad_norm": 0.4129338562488556,
      "learning_rate": 0.00012301209288704184,
      "loss": 0.2563,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 8630,
      "tokens_per_second_per_gpu": 323.02
    },
    {
      "epoch": 0.85897499627181,
      "grad_norm": 0.37343230843544006,
      "learning_rate": 0.00012285933049597335,
      "loss": 0.154,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8640,
      "tokens_per_second_per_gpu": 326.66
    },
    {
      "epoch": 0.8599691802952727,
      "grad_norm": 0.32739633321762085,
      "learning_rate": 0.00012270651179754243,
      "loss": 0.2135,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8650,
      "tokens_per_second_per_gpu": 340.36
    },
    {
      "epoch": 0.8609633643187354,
      "grad_norm": 0.5008605718612671,
      "learning_rate": 0.0001225536371681738,
      "loss": 0.2221,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 8660,
      "tokens_per_second_per_gpu": 295.75
    },
    {
      "epoch": 0.8619575483421982,
      "grad_norm": 2.8313498497009277,
      "learning_rate": 0.00012240070698443,
      "loss": 0.2402,
      "memory/device_reserved (GiB)": 22.45,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8670,
      "tokens_per_second_per_gpu": 343.07
    },
    {
      "epoch": 0.8629517323656609,
      "grad_norm": 3.499013662338257,
      "learning_rate": 0.00012224772162301042,
      "loss": 0.2588,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.58,
      "memory/max_allocated (GiB)": 21.58,
      "step": 8680,
      "tokens_per_second_per_gpu": 348.94
    },
    {
      "epoch": 0.8639459163891237,
      "grad_norm": 0.4069509208202362,
      "learning_rate": 0.0001220946814607503,
      "loss": 0.2485,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 8690,
      "tokens_per_second_per_gpu": 365.05
    },
    {
      "epoch": 0.8649401004125864,
      "grad_norm": 0.5977867841720581,
      "learning_rate": 0.00012194158687461992,
      "loss": 0.2119,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 8700,
      "tokens_per_second_per_gpu": 326.01
    },
    {
      "epoch": 0.8659342844360491,
      "grad_norm": 6.747653961181641,
      "learning_rate": 0.00012178843824172361,
      "loss": 0.2719,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 8710,
      "tokens_per_second_per_gpu": 390.15
    },
    {
      "epoch": 0.8669284684595119,
      "grad_norm": 0.46154770255088806,
      "learning_rate": 0.00012163523593929884,
      "loss": 0.1836,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 8720,
      "tokens_per_second_per_gpu": 343.33
    },
    {
      "epoch": 0.8679226524829746,
      "grad_norm": 0.272504985332489,
      "learning_rate": 0.00012148198034471524,
      "loss": 0.2419,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 8730,
      "tokens_per_second_per_gpu": 347.62
    },
    {
      "epoch": 0.8689168365064374,
      "grad_norm": 0.4848499596118927,
      "learning_rate": 0.00012132867183547372,
      "loss": 0.2379,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 8740,
      "tokens_per_second_per_gpu": 315.42
    },
    {
      "epoch": 0.8699110205299001,
      "grad_norm": 0.42282259464263916,
      "learning_rate": 0.00012117531078920556,
      "loss": 0.2358,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 8750,
      "tokens_per_second_per_gpu": 404.19
    },
    {
      "epoch": 0.8709052045533628,
      "grad_norm": 0.5138429999351501,
      "learning_rate": 0.00012102189758367142,
      "loss": 0.2602,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 8760,
      "tokens_per_second_per_gpu": 350.33
    },
    {
      "epoch": 0.8718993885768256,
      "grad_norm": 0.2951951324939728,
      "learning_rate": 0.00012086843259676041,
      "loss": 0.303,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8770,
      "tokens_per_second_per_gpu": 396.42
    },
    {
      "epoch": 0.8728935726002883,
      "grad_norm": 0.4597817361354828,
      "learning_rate": 0.00012071491620648934,
      "loss": 0.2519,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.68,
      "memory/max_allocated (GiB)": 19.68,
      "step": 8780,
      "tokens_per_second_per_gpu": 338.59
    },
    {
      "epoch": 0.8738877566237511,
      "grad_norm": 0.13618378341197968,
      "learning_rate": 0.00012056134879100138,
      "loss": 0.2235,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.43,
      "memory/max_allocated (GiB)": 20.43,
      "step": 8790,
      "tokens_per_second_per_gpu": 325.39
    },
    {
      "epoch": 0.8748819406472138,
      "grad_norm": 0.5435792207717896,
      "learning_rate": 0.00012040773072856566,
      "loss": 0.2563,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8800,
      "tokens_per_second_per_gpu": 346.95
    },
    {
      "epoch": 0.8758761246706765,
      "grad_norm": 0.24607907235622406,
      "learning_rate": 0.00012025406239757588,
      "loss": 0.1721,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8810,
      "tokens_per_second_per_gpu": 305.91
    },
    {
      "epoch": 0.8768703086941393,
      "grad_norm": 0.3879210650920868,
      "learning_rate": 0.00012010034417654962,
      "loss": 0.2026,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 8820,
      "tokens_per_second_per_gpu": 305.29
    },
    {
      "epoch": 0.877864492717602,
      "grad_norm": 0.4256599545478821,
      "learning_rate": 0.00011994657644412734,
      "loss": 0.1985,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 8830,
      "tokens_per_second_per_gpu": 330.52
    },
    {
      "epoch": 0.8788586767410648,
      "grad_norm": 0.2512848973274231,
      "learning_rate": 0.00011979275957907146,
      "loss": 0.2153,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 8840,
      "tokens_per_second_per_gpu": 321.37
    },
    {
      "epoch": 0.8798528607645275,
      "grad_norm": 0.38337549567222595,
      "learning_rate": 0.00011963889396026547,
      "loss": 0.2383,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 8850,
      "tokens_per_second_per_gpu": 359.62
    },
    {
      "epoch": 0.8808470447879903,
      "grad_norm": 0.3447147309780121,
      "learning_rate": 0.00011948497996671286,
      "loss": 0.2304,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8860,
      "tokens_per_second_per_gpu": 332.81
    },
    {
      "epoch": 0.881841228811453,
      "grad_norm": 0.40846845507621765,
      "learning_rate": 0.00011933101797753637,
      "loss": 0.2297,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 8870,
      "tokens_per_second_per_gpu": 384.1
    },
    {
      "epoch": 0.8828354128349157,
      "grad_norm": 0.5454410910606384,
      "learning_rate": 0.0001191770083719769,
      "loss": 0.2242,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 8880,
      "tokens_per_second_per_gpu": 387.99
    },
    {
      "epoch": 0.8838295968583785,
      "grad_norm": 0.4706827700138092,
      "learning_rate": 0.00011902295152939262,
      "loss": 0.2381,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 8890,
      "tokens_per_second_per_gpu": 304.64
    },
    {
      "epoch": 0.8848237808818412,
      "grad_norm": 0.6073552370071411,
      "learning_rate": 0.00011886884782925816,
      "loss": 0.2417,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 8900,
      "tokens_per_second_per_gpu": 315.52
    },
    {
      "epoch": 0.885817964905304,
      "grad_norm": 0.3200027644634247,
      "learning_rate": 0.00011871469765116346,
      "loss": 0.2117,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 8910,
      "tokens_per_second_per_gpu": 306.02
    },
    {
      "epoch": 0.8868121489287667,
      "grad_norm": 0.42333486676216125,
      "learning_rate": 0.00011856050137481301,
      "loss": 0.2552,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 8920,
      "tokens_per_second_per_gpu": 355.23
    },
    {
      "epoch": 0.8878063329522294,
      "grad_norm": 0.2578692138195038,
      "learning_rate": 0.00011840625938002481,
      "loss": 0.1743,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 8930,
      "tokens_per_second_per_gpu": 356.78
    },
    {
      "epoch": 0.8888005169756922,
      "grad_norm": 0.3321487605571747,
      "learning_rate": 0.00011825197204672952,
      "loss": 0.2637,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 8940,
      "tokens_per_second_per_gpu": 424.07
    },
    {
      "epoch": 0.8897947009991549,
      "grad_norm": 0.3663140833377838,
      "learning_rate": 0.00011809763975496944,
      "loss": 0.2272,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 8950,
      "tokens_per_second_per_gpu": 360.15
    },
    {
      "epoch": 0.8907888850226177,
      "grad_norm": 0.2619111239910126,
      "learning_rate": 0.00011794326288489761,
      "loss": 0.2723,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.03,
      "memory/max_allocated (GiB)": 20.03,
      "step": 8960,
      "tokens_per_second_per_gpu": 357.64
    },
    {
      "epoch": 0.8917830690460804,
      "grad_norm": 0.4359273612499237,
      "learning_rate": 0.0001177888418167769,
      "loss": 0.2633,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 8970,
      "tokens_per_second_per_gpu": 406.63
    },
    {
      "epoch": 0.8927772530695431,
      "grad_norm": 0.29716983437538147,
      "learning_rate": 0.00011763437693097903,
      "loss": 0.2789,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 8980,
      "tokens_per_second_per_gpu": 386.97
    },
    {
      "epoch": 0.8937714370930059,
      "grad_norm": 0.6009801626205444,
      "learning_rate": 0.00011747986860798368,
      "loss": 0.2057,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.43,
      "memory/max_allocated (GiB)": 20.43,
      "step": 8990,
      "tokens_per_second_per_gpu": 318.77
    },
    {
      "epoch": 0.8947656211164686,
      "grad_norm": 0.4606180191040039,
      "learning_rate": 0.0001173253172283775,
      "loss": 0.2232,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9000,
      "tokens_per_second_per_gpu": 283.3
    },
    {
      "epoch": 0.8957598051399314,
      "grad_norm": 0.3887878358364105,
      "learning_rate": 0.00011717072317285318,
      "loss": 0.2175,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9010,
      "tokens_per_second_per_gpu": 313.78
    },
    {
      "epoch": 0.8967539891633941,
      "grad_norm": 0.3322821855545044,
      "learning_rate": 0.0001170160868222086,
      "loss": 0.2491,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9020,
      "tokens_per_second_per_gpu": 339.47
    },
    {
      "epoch": 0.8977481731868568,
      "grad_norm": 0.5119565725326538,
      "learning_rate": 0.00011686140855734571,
      "loss": 0.2568,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.66,
      "memory/max_allocated (GiB)": 19.66,
      "step": 9030,
      "tokens_per_second_per_gpu": 322.62
    },
    {
      "epoch": 0.8987423572103196,
      "grad_norm": 0.32013964653015137,
      "learning_rate": 0.00011670668875926982,
      "loss": 0.3019,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9040,
      "tokens_per_second_per_gpu": 356.77
    },
    {
      "epoch": 0.8997365412337823,
      "grad_norm": 0.19132547080516815,
      "learning_rate": 0.00011655192780908849,
      "loss": 0.1927,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.16,
      "memory/max_allocated (GiB)": 18.16,
      "step": 9050,
      "tokens_per_second_per_gpu": 306.79
    },
    {
      "epoch": 0.9007307252572451,
      "grad_norm": 0.4459245502948761,
      "learning_rate": 0.00011639712608801059,
      "loss": 0.2013,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 9060,
      "tokens_per_second_per_gpu": 369.24
    },
    {
      "epoch": 0.9017249092807078,
      "grad_norm": 0.3540112376213074,
      "learning_rate": 0.00011624228397734556,
      "loss": 0.1513,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 9070,
      "tokens_per_second_per_gpu": 323.78
    },
    {
      "epoch": 0.9027190933041706,
      "grad_norm": 0.3951474130153656,
      "learning_rate": 0.00011608740185850219,
      "loss": 0.2055,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9080,
      "tokens_per_second_per_gpu": 358.92
    },
    {
      "epoch": 0.9037132773276333,
      "grad_norm": 0.3915760815143585,
      "learning_rate": 0.00011593248011298791,
      "loss": 0.2148,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9090,
      "tokens_per_second_per_gpu": 397.61
    },
    {
      "epoch": 0.9047074613510961,
      "grad_norm": 0.49453213810920715,
      "learning_rate": 0.00011577751912240771,
      "loss": 0.187,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9100,
      "tokens_per_second_per_gpu": 379.24
    },
    {
      "epoch": 0.9057016453745589,
      "grad_norm": 0.5185866951942444,
      "learning_rate": 0.00011562251926846326,
      "loss": 0.219,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9110,
      "tokens_per_second_per_gpu": 307.18
    },
    {
      "epoch": 0.9066958293980216,
      "grad_norm": 0.44589415192604065,
      "learning_rate": 0.00011546748093295195,
      "loss": 0.2127,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 9120,
      "tokens_per_second_per_gpu": 360.35
    },
    {
      "epoch": 0.9076900134214844,
      "grad_norm": 0.4594690501689911,
      "learning_rate": 0.00011531240449776594,
      "loss": 0.2057,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9130,
      "tokens_per_second_per_gpu": 351.45
    },
    {
      "epoch": 0.9086841974449471,
      "grad_norm": 0.4069642722606659,
      "learning_rate": 0.00011515729034489133,
      "loss": 0.2213,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 9140,
      "tokens_per_second_per_gpu": 331.15
    },
    {
      "epoch": 0.9096783814684098,
      "grad_norm": 0.5103911757469177,
      "learning_rate": 0.00011500213885640705,
      "loss": 0.2258,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 9150,
      "tokens_per_second_per_gpu": 323.85
    },
    {
      "epoch": 0.9106725654918726,
      "grad_norm": 0.5270497798919678,
      "learning_rate": 0.00011484695041448399,
      "loss": 0.2709,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 9160,
      "tokens_per_second_per_gpu": 363.69
    },
    {
      "epoch": 0.9116667495153353,
      "grad_norm": 0.48765021562576294,
      "learning_rate": 0.00011469172540138407,
      "loss": 0.1935,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 9170,
      "tokens_per_second_per_gpu": 329.09
    },
    {
      "epoch": 0.9126609335387981,
      "grad_norm": 0.4249947667121887,
      "learning_rate": 0.00011453646419945934,
      "loss": 0.2296,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 9180,
      "tokens_per_second_per_gpu": 412.86
    },
    {
      "epoch": 0.9136551175622608,
      "grad_norm": 0.36251431703567505,
      "learning_rate": 0.00011438116719115089,
      "loss": 0.2361,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 9190,
      "tokens_per_second_per_gpu": 343.29
    },
    {
      "epoch": 0.9146493015857236,
      "grad_norm": 0.3858913481235504,
      "learning_rate": 0.00011422583475898814,
      "loss": 0.2446,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 9200,
      "tokens_per_second_per_gpu": 331.61
    },
    {
      "epoch": 0.9156434856091863,
      "grad_norm": 0.43422338366508484,
      "learning_rate": 0.00011407046728558768,
      "loss": 0.1683,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 9210,
      "tokens_per_second_per_gpu": 303.63
    },
    {
      "epoch": 0.916637669632649,
      "grad_norm": 0.43746358156204224,
      "learning_rate": 0.00011391506515365245,
      "loss": 0.1423,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 9220,
      "tokens_per_second_per_gpu": 337.04
    },
    {
      "epoch": 0.9176318536561118,
      "grad_norm": 0.30726879835128784,
      "learning_rate": 0.00011375962874597073,
      "loss": 0.179,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9230,
      "tokens_per_second_per_gpu": 312.89
    },
    {
      "epoch": 0.9186260376795745,
      "grad_norm": 0.5088397264480591,
      "learning_rate": 0.00011360415844541523,
      "loss": 0.2571,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9240,
      "tokens_per_second_per_gpu": 359.77
    },
    {
      "epoch": 0.9196202217030373,
      "grad_norm": 0.545219898223877,
      "learning_rate": 0.00011344865463494219,
      "loss": 0.2228,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 9250,
      "tokens_per_second_per_gpu": 337.28
    },
    {
      "epoch": 0.9206144057265,
      "grad_norm": 0.5700411796569824,
      "learning_rate": 0.00011329311769759035,
      "loss": 0.2236,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 9260,
      "tokens_per_second_per_gpu": 340.23
    },
    {
      "epoch": 0.9216085897499627,
      "grad_norm": 0.3975035548210144,
      "learning_rate": 0.00011313754801648003,
      "loss": 0.2487,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 9270,
      "tokens_per_second_per_gpu": 345.97
    },
    {
      "epoch": 0.9226027737734255,
      "grad_norm": 0.5616829991340637,
      "learning_rate": 0.00011298194597481226,
      "loss": 0.2511,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9280,
      "tokens_per_second_per_gpu": 331.41
    },
    {
      "epoch": 0.9235969577968882,
      "grad_norm": 0.38982534408569336,
      "learning_rate": 0.00011282631195586777,
      "loss": 0.2809,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 9290,
      "tokens_per_second_per_gpu": 361.76
    },
    {
      "epoch": 0.924591141820351,
      "grad_norm": 0.531318187713623,
      "learning_rate": 0.00011267064634300603,
      "loss": 0.2608,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9300,
      "tokens_per_second_per_gpu": 393.15
    },
    {
      "epoch": 0.9255853258438137,
      "grad_norm": 0.5956375002861023,
      "learning_rate": 0.00011251494951966437,
      "loss": 0.2229,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 9310,
      "tokens_per_second_per_gpu": 284.92
    },
    {
      "epoch": 0.9265795098672764,
      "grad_norm": 0.42666998505592346,
      "learning_rate": 0.0001123592218693569,
      "loss": 0.1985,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 9320,
      "tokens_per_second_per_gpu": 362.3
    },
    {
      "epoch": 0.9275736938907392,
      "grad_norm": 0.4182765781879425,
      "learning_rate": 0.00011220346377567381,
      "loss": 0.2535,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 9330,
      "tokens_per_second_per_gpu": 348.83
    },
    {
      "epoch": 0.9285678779142019,
      "grad_norm": 0.5867879390716553,
      "learning_rate": 0.00011204767562228017,
      "loss": 0.2309,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 9340,
      "tokens_per_second_per_gpu": 334.46
    },
    {
      "epoch": 0.9295620619376647,
      "grad_norm": 0.27041056752204895,
      "learning_rate": 0.00011189185779291515,
      "loss": 0.232,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 9350,
      "tokens_per_second_per_gpu": 331.56
    },
    {
      "epoch": 0.9305562459611274,
      "grad_norm": 0.5081501603126526,
      "learning_rate": 0.00011173601067139099,
      "loss": 0.2399,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9360,
      "tokens_per_second_per_gpu": 336.84
    },
    {
      "epoch": 0.9315504299845901,
      "grad_norm": 0.7966908812522888,
      "learning_rate": 0.00011158013464159208,
      "loss": 0.2606,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 9370,
      "tokens_per_second_per_gpu": 313.66
    },
    {
      "epoch": 0.9325446140080529,
      "grad_norm": 0.45302364230155945,
      "learning_rate": 0.00011142423008747403,
      "loss": 0.1581,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 9380,
      "tokens_per_second_per_gpu": 322.72
    },
    {
      "epoch": 0.9335387980315156,
      "grad_norm": 0.5959784984588623,
      "learning_rate": 0.00011126829739306271,
      "loss": 0.2115,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.1,
      "memory/max_allocated (GiB)": 19.1,
      "step": 9390,
      "tokens_per_second_per_gpu": 304.85
    },
    {
      "epoch": 0.9345329820549784,
      "grad_norm": 0.5530646443367004,
      "learning_rate": 0.00011111233694245328,
      "loss": 0.1854,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 9400,
      "tokens_per_second_per_gpu": 347.73
    },
    {
      "epoch": 0.9355271660784411,
      "grad_norm": 0.4127480387687683,
      "learning_rate": 0.00011095634911980933,
      "loss": 0.2307,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9410,
      "tokens_per_second_per_gpu": 325.27
    },
    {
      "epoch": 0.9365213501019038,
      "grad_norm": 0.4038192629814148,
      "learning_rate": 0.0001108003343093618,
      "loss": 0.1953,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9420,
      "tokens_per_second_per_gpu": 327.3
    },
    {
      "epoch": 0.9375155341253666,
      "grad_norm": 0.4245215654373169,
      "learning_rate": 0.00011064429289540821,
      "loss": 0.2505,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9430,
      "tokens_per_second_per_gpu": 337.4
    },
    {
      "epoch": 0.9385097181488293,
      "grad_norm": 0.4768611192703247,
      "learning_rate": 0.00011048822526231148,
      "loss": 0.1584,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 9440,
      "tokens_per_second_per_gpu": 314.23
    },
    {
      "epoch": 0.9395039021722921,
      "grad_norm": 0.3262840807437897,
      "learning_rate": 0.00011033213179449917,
      "loss": 0.2287,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 9450,
      "tokens_per_second_per_gpu": 358.18
    },
    {
      "epoch": 0.9404980861957548,
      "grad_norm": 0.5882810354232788,
      "learning_rate": 0.00011017601287646251,
      "loss": 0.206,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9460,
      "tokens_per_second_per_gpu": 405.83
    },
    {
      "epoch": 0.9414922702192176,
      "grad_norm": 0.4905533492565155,
      "learning_rate": 0.0001100198688927554,
      "loss": 0.2209,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 9470,
      "tokens_per_second_per_gpu": 327.29
    },
    {
      "epoch": 0.9424864542426803,
      "grad_norm": 0.5608656406402588,
      "learning_rate": 0.00010986370022799346,
      "loss": 0.2418,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 9480,
      "tokens_per_second_per_gpu": 335.39
    },
    {
      "epoch": 0.943480638266143,
      "grad_norm": 0.4696904718875885,
      "learning_rate": 0.00010970750726685309,
      "loss": 0.2742,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 9490,
      "tokens_per_second_per_gpu": 399.98
    },
    {
      "epoch": 0.9444748222896058,
      "grad_norm": 0.423969030380249,
      "learning_rate": 0.00010955129039407062,
      "loss": 0.2259,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9500,
      "tokens_per_second_per_gpu": 336.84
    },
    {
      "epoch": 0.9454690063130685,
      "grad_norm": 0.438812792301178,
      "learning_rate": 0.0001093950499944412,
      "loss": 0.1855,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 9510,
      "tokens_per_second_per_gpu": 318.59
    },
    {
      "epoch": 0.9464631903365313,
      "grad_norm": 0.5361111164093018,
      "learning_rate": 0.00010923878645281794,
      "loss": 0.2713,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9520,
      "tokens_per_second_per_gpu": 354.0
    },
    {
      "epoch": 0.947457374359994,
      "grad_norm": 0.39556553959846497,
      "learning_rate": 0.000109082500154111,
      "loss": 0.2539,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 9530,
      "tokens_per_second_per_gpu": 342.96
    },
    {
      "epoch": 0.9484515583834567,
      "grad_norm": 0.42129939794540405,
      "learning_rate": 0.00010892619148328654,
      "loss": 0.2282,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9540,
      "tokens_per_second_per_gpu": 375.18
    },
    {
      "epoch": 0.9494457424069195,
      "grad_norm": 0.49444761872291565,
      "learning_rate": 0.00010876986082536584,
      "loss": 0.2342,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.68,
      "memory/max_allocated (GiB)": 19.68,
      "step": 9550,
      "tokens_per_second_per_gpu": 394.82
    },
    {
      "epoch": 0.9504399264303822,
      "grad_norm": 0.3940083682537079,
      "learning_rate": 0.0001086135085654244,
      "loss": 0.1993,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 9560,
      "tokens_per_second_per_gpu": 346.47
    },
    {
      "epoch": 0.951434110453845,
      "grad_norm": 0.40422961115837097,
      "learning_rate": 0.00010845713508859088,
      "loss": 0.2479,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 9570,
      "tokens_per_second_per_gpu": 338.94
    },
    {
      "epoch": 0.9524282944773077,
      "grad_norm": 0.5008521676063538,
      "learning_rate": 0.00010830074078004615,
      "loss": 0.2217,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 9580,
      "tokens_per_second_per_gpu": 390.26
    },
    {
      "epoch": 0.9534224785007704,
      "grad_norm": 0.37146544456481934,
      "learning_rate": 0.00010814432602502246,
      "loss": 0.2178,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 9590,
      "tokens_per_second_per_gpu": 347.55
    },
    {
      "epoch": 0.9544166625242332,
      "grad_norm": 0.4229485094547272,
      "learning_rate": 0.00010798789120880246,
      "loss": 0.2183,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 9600,
      "tokens_per_second_per_gpu": 311.66
    },
    {
      "epoch": 0.9554108465476959,
      "grad_norm": 0.38394778966903687,
      "learning_rate": 0.00010783143671671813,
      "loss": 0.2663,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 9610,
      "tokens_per_second_per_gpu": 329.51
    },
    {
      "epoch": 0.9564050305711588,
      "grad_norm": 0.4587366580963135,
      "learning_rate": 0.00010767496293414996,
      "loss": 0.2776,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9620,
      "tokens_per_second_per_gpu": 464.41
    },
    {
      "epoch": 0.9573992145946215,
      "grad_norm": 0.5589081645011902,
      "learning_rate": 0.0001075184702465259,
      "loss": 0.2387,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.47,
      "memory/max_allocated (GiB)": 21.47,
      "step": 9630,
      "tokens_per_second_per_gpu": 358.44
    },
    {
      "epoch": 0.9583933986180843,
      "grad_norm": 0.6210941672325134,
      "learning_rate": 0.0001073619590393206,
      "loss": 0.2587,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9640,
      "tokens_per_second_per_gpu": 373.16
    },
    {
      "epoch": 0.959387582641547,
      "grad_norm": 0.6241095066070557,
      "learning_rate": 0.0001072054296980542,
      "loss": 0.2378,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9650,
      "tokens_per_second_per_gpu": 385.04
    },
    {
      "epoch": 0.9603817666650097,
      "grad_norm": 0.40504932403564453,
      "learning_rate": 0.00010704888260829156,
      "loss": 0.2767,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9660,
      "tokens_per_second_per_gpu": 384.99
    },
    {
      "epoch": 0.9613759506884725,
      "grad_norm": 0.4956580400466919,
      "learning_rate": 0.0001068923181556412,
      "loss": 0.2388,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 9670,
      "tokens_per_second_per_gpu": 378.96
    },
    {
      "epoch": 0.9623701347119352,
      "grad_norm": 0.2848256230354309,
      "learning_rate": 0.00010673573672575454,
      "loss": 0.2238,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 9680,
      "tokens_per_second_per_gpu": 359.23
    },
    {
      "epoch": 0.963364318735398,
      "grad_norm": 0.38930168747901917,
      "learning_rate": 0.00010657913870432468,
      "loss": 0.2305,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 9690,
      "tokens_per_second_per_gpu": 309.48
    },
    {
      "epoch": 0.9643585027588607,
      "grad_norm": 0.4527674913406372,
      "learning_rate": 0.00010642252447708563,
      "loss": 0.2731,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9700,
      "tokens_per_second_per_gpu": 349.69
    },
    {
      "epoch": 0.9653526867823234,
      "grad_norm": 0.4257194697856903,
      "learning_rate": 0.00010626589442981138,
      "loss": 0.2635,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 9710,
      "tokens_per_second_per_gpu": 414.72
    },
    {
      "epoch": 0.9663468708057862,
      "grad_norm": 0.3962489068508148,
      "learning_rate": 0.00010610924894831483,
      "loss": 0.2862,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 9720,
      "tokens_per_second_per_gpu": 365.4
    },
    {
      "epoch": 0.9673410548292489,
      "grad_norm": 0.29012176394462585,
      "learning_rate": 0.00010595258841844688,
      "loss": 0.2353,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 9730,
      "tokens_per_second_per_gpu": 364.81
    },
    {
      "epoch": 0.9683352388527117,
      "grad_norm": 0.5301884412765503,
      "learning_rate": 0.00010579591322609559,
      "loss": 0.1947,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.61,
      "memory/max_allocated (GiB)": 20.61,
      "step": 9740,
      "tokens_per_second_per_gpu": 369.66
    },
    {
      "epoch": 0.9693294228761744,
      "grad_norm": 0.41399112343788147,
      "learning_rate": 0.000105639223757185,
      "loss": 0.1859,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.98,
      "memory/max_allocated (GiB)": 19.98,
      "step": 9750,
      "tokens_per_second_per_gpu": 314.55
    },
    {
      "epoch": 0.9703236068996371,
      "grad_norm": 0.3453201353549957,
      "learning_rate": 0.00010548252039767443,
      "loss": 0.1971,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9760,
      "tokens_per_second_per_gpu": 368.71
    },
    {
      "epoch": 0.9713177909230999,
      "grad_norm": 0.5453227162361145,
      "learning_rate": 0.00010532580353355734,
      "loss": 0.3006,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.47,
      "memory/max_allocated (GiB)": 21.47,
      "step": 9770,
      "tokens_per_second_per_gpu": 420.65
    },
    {
      "epoch": 0.9723119749465626,
      "grad_norm": 0.42438754439353943,
      "learning_rate": 0.00010516907355086055,
      "loss": 0.2027,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 9780,
      "tokens_per_second_per_gpu": 396.2
    },
    {
      "epoch": 0.9733061589700254,
      "grad_norm": 0.5116370916366577,
      "learning_rate": 0.00010501233083564306,
      "loss": 0.1968,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 9790,
      "tokens_per_second_per_gpu": 345.38
    },
    {
      "epoch": 0.9743003429934881,
      "grad_norm": 0.23473972082138062,
      "learning_rate": 0.00010485557577399536,
      "loss": 0.1676,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 9800,
      "tokens_per_second_per_gpu": 386.72
    },
    {
      "epoch": 0.9752945270169509,
      "grad_norm": 0.5427513718605042,
      "learning_rate": 0.00010469880875203827,
      "loss": 0.2066,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 9810,
      "tokens_per_second_per_gpu": 372.28
    },
    {
      "epoch": 0.9762887110404136,
      "grad_norm": 0.4717273712158203,
      "learning_rate": 0.00010454203015592214,
      "loss": 0.2097,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9820,
      "tokens_per_second_per_gpu": 317.9
    },
    {
      "epoch": 0.9772828950638763,
      "grad_norm": 0.2797994017601013,
      "learning_rate": 0.00010438524037182573,
      "loss": 0.3073,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9830,
      "tokens_per_second_per_gpu": 371.67
    },
    {
      "epoch": 0.9782770790873391,
      "grad_norm": 0.3694850504398346,
      "learning_rate": 0.00010422843978595542,
      "loss": 0.2309,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 9840,
      "tokens_per_second_per_gpu": 426.94
    },
    {
      "epoch": 0.9792712631108018,
      "grad_norm": 0.3863958716392517,
      "learning_rate": 0.00010407162878454423,
      "loss": 0.2697,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 9850,
      "tokens_per_second_per_gpu": 338.41
    },
    {
      "epoch": 0.9802654471342646,
      "grad_norm": 0.22021318972110748,
      "learning_rate": 0.00010391480775385078,
      "loss": 0.1866,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 9860,
      "tokens_per_second_per_gpu": 331.0
    },
    {
      "epoch": 0.9812596311577273,
      "grad_norm": 0.3698722720146179,
      "learning_rate": 0.00010375797708015844,
      "loss": 0.1992,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 9870,
      "tokens_per_second_per_gpu": 328.05
    },
    {
      "epoch": 0.98225381518119,
      "grad_norm": 0.45455843210220337,
      "learning_rate": 0.00010360113714977428,
      "loss": 0.2508,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 9880,
      "tokens_per_second_per_gpu": 409.92
    },
    {
      "epoch": 0.9832479992046528,
      "grad_norm": 0.3562302887439728,
      "learning_rate": 0.00010344428834902822,
      "loss": 0.187,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 9890,
      "tokens_per_second_per_gpu": 262.76
    },
    {
      "epoch": 0.9842421832281155,
      "grad_norm": 0.5735921263694763,
      "learning_rate": 0.00010328743106427197,
      "loss": 0.2517,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 9900,
      "tokens_per_second_per_gpu": 338.79
    },
    {
      "epoch": 0.9852363672515783,
      "grad_norm": 0.47819098830223083,
      "learning_rate": 0.00010313056568187818,
      "loss": 0.2298,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 9910,
      "tokens_per_second_per_gpu": 351.88
    },
    {
      "epoch": 0.986230551275041,
      "grad_norm": 0.29087749123573303,
      "learning_rate": 0.00010297369258823948,
      "loss": 0.1617,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 9920,
      "tokens_per_second_per_gpu": 311.57
    },
    {
      "epoch": 0.9872247352985037,
      "grad_norm": 0.5746156573295593,
      "learning_rate": 0.00010281681216976742,
      "loss": 0.2215,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9930,
      "tokens_per_second_per_gpu": 360.68
    },
    {
      "epoch": 0.9882189193219665,
      "grad_norm": 0.5306881666183472,
      "learning_rate": 0.00010265992481289164,
      "loss": 0.2165,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 9940,
      "tokens_per_second_per_gpu": 350.01
    },
    {
      "epoch": 0.9892131033454292,
      "grad_norm": 0.5628572106361389,
      "learning_rate": 0.00010250303090405886,
      "loss": 0.2246,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 9950,
      "tokens_per_second_per_gpu": 331.08
    },
    {
      "epoch": 0.990207287368892,
      "grad_norm": 0.38296106457710266,
      "learning_rate": 0.00010234613082973195,
      "loss": 0.2294,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 9960,
      "tokens_per_second_per_gpu": 310.12
    },
    {
      "epoch": 0.9912014713923547,
      "grad_norm": 0.5048671960830688,
      "learning_rate": 0.00010218922497638893,
      "loss": 0.2263,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 9970,
      "tokens_per_second_per_gpu": 360.03
    },
    {
      "epoch": 0.9921956554158174,
      "grad_norm": 0.2620450258255005,
      "learning_rate": 0.00010203231373052205,
      "loss": 0.2391,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 9980,
      "tokens_per_second_per_gpu": 397.95
    },
    {
      "epoch": 0.9931898394392802,
      "grad_norm": 0.37807098031044006,
      "learning_rate": 0.00010187539747863693,
      "loss": 0.2654,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 9990,
      "tokens_per_second_per_gpu": 413.64
    },
    {
      "epoch": 0.9941840234627429,
      "grad_norm": 0.3059203624725342,
      "learning_rate": 0.00010171847660725147,
      "loss": 0.2081,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 10000,
      "tokens_per_second_per_gpu": 317.3
    },
    {
      "epoch": 0.9951782074862057,
      "grad_norm": 0.38892289996147156,
      "learning_rate": 0.0001015615515028949,
      "loss": 0.2826,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10010,
      "tokens_per_second_per_gpu": 379.68
    },
    {
      "epoch": 0.9961723915096684,
      "grad_norm": 0.22751818597316742,
      "learning_rate": 0.00010140462255210696,
      "loss": 0.2674,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 10020,
      "tokens_per_second_per_gpu": 362.08
    },
    {
      "epoch": 0.9971665755331311,
      "grad_norm": 0.3578786551952362,
      "learning_rate": 0.00010124769014143678,
      "loss": 0.2125,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 10030,
      "tokens_per_second_per_gpu": 367.81
    },
    {
      "epoch": 0.9981607595565939,
      "grad_norm": 0.32971230149269104,
      "learning_rate": 0.00010109075465744208,
      "loss": 0.1599,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 10040,
      "tokens_per_second_per_gpu": 336.7
    },
    {
      "epoch": 0.9991549435800566,
      "grad_norm": 0.4897179901599884,
      "learning_rate": 0.00010093381648668813,
      "loss": 0.2485,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 10050,
      "tokens_per_second_per_gpu": 303.41
    },
    {
      "epoch": 1.0000994184023462,
      "grad_norm": 0.23543600738048553,
      "learning_rate": 0.00010077687601574678,
      "loss": 0.2974,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 10060,
      "tokens_per_second_per_gpu": 335.91
    },
    {
      "epoch": 1.001093602425809,
      "grad_norm": 0.30785781145095825,
      "learning_rate": 0.00010061993363119566,
      "loss": 0.155,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 10070,
      "tokens_per_second_per_gpu": 313.67
    },
    {
      "epoch": 1.0020877864492717,
      "grad_norm": 0.36587706208229065,
      "learning_rate": 0.00010046298971961695,
      "loss": 0.2054,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10080,
      "tokens_per_second_per_gpu": 388.65
    },
    {
      "epoch": 1.0030819704727345,
      "grad_norm": 0.5172699093818665,
      "learning_rate": 0.0001003060446675967,
      "loss": 0.1678,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 10090,
      "tokens_per_second_per_gpu": 356.25
    },
    {
      "epoch": 1.0040761544961971,
      "grad_norm": 0.3362785279750824,
      "learning_rate": 0.00010014909886172377,
      "loss": 0.1414,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10100,
      "tokens_per_second_per_gpu": 317.06
    },
    {
      "epoch": 1.00507033851966,
      "grad_norm": 0.5255736112594604,
      "learning_rate": 9.99921526885888e-05,
      "loss": 0.2086,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 10110,
      "tokens_per_second_per_gpu": 428.98
    },
    {
      "epoch": 1.0060645225431228,
      "grad_norm": 0.4546601474285126,
      "learning_rate": 9.983520653478343e-05,
      "loss": 0.1895,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 10120,
      "tokens_per_second_per_gpu": 304.97
    },
    {
      "epoch": 1.0070587065665855,
      "grad_norm": 0.33736681938171387,
      "learning_rate": 9.967826078689919e-05,
      "loss": 0.1643,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.67,
      "memory/max_allocated (GiB)": 19.67,
      "step": 10130,
      "tokens_per_second_per_gpu": 337.84
    },
    {
      "epoch": 1.0080528905900483,
      "grad_norm": 0.2856805622577667,
      "learning_rate": 9.952131583152665e-05,
      "loss": 0.1171,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 10140,
      "tokens_per_second_per_gpu": 320.63
    },
    {
      "epoch": 1.009047074613511,
      "grad_norm": 0.35713866353034973,
      "learning_rate": 9.936437205525437e-05,
      "loss": 0.1597,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 10150,
      "tokens_per_second_per_gpu": 365.04
    },
    {
      "epoch": 1.0100412586369738,
      "grad_norm": 0.4092197120189667,
      "learning_rate": 9.920742984466809e-05,
      "loss": 0.2034,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 10160,
      "tokens_per_second_per_gpu": 331.18
    },
    {
      "epoch": 1.0110354426604364,
      "grad_norm": 0.36163678765296936,
      "learning_rate": 9.905048958634958e-05,
      "loss": 0.1466,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 10170,
      "tokens_per_second_per_gpu": 416.08
    },
    {
      "epoch": 1.0120296266838993,
      "grad_norm": 0.5861209630966187,
      "learning_rate": 9.889355166687593e-05,
      "loss": 0.1191,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 10180,
      "tokens_per_second_per_gpu": 317.4
    },
    {
      "epoch": 1.013023810707362,
      "grad_norm": 0.4099638760089874,
      "learning_rate": 9.873661647281836e-05,
      "loss": 0.1707,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 10190,
      "tokens_per_second_per_gpu": 380.48
    },
    {
      "epoch": 1.0140179947308248,
      "grad_norm": 0.5533301830291748,
      "learning_rate": 9.857968439074142e-05,
      "loss": 0.2481,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 10200,
      "tokens_per_second_per_gpu": 355.99
    },
    {
      "epoch": 1.0150121787542874,
      "grad_norm": 0.4641655385494232,
      "learning_rate": 9.842275580720205e-05,
      "loss": 0.1997,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10210,
      "tokens_per_second_per_gpu": 392.6
    },
    {
      "epoch": 1.0160063627777502,
      "grad_norm": 0.5175334215164185,
      "learning_rate": 9.826583110874847e-05,
      "loss": 0.1672,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 10220,
      "tokens_per_second_per_gpu": 338.18
    },
    {
      "epoch": 1.0170005468012129,
      "grad_norm": 0.2824496924877167,
      "learning_rate": 9.810891068191942e-05,
      "loss": 0.152,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 10230,
      "tokens_per_second_per_gpu": 384.17
    },
    {
      "epoch": 1.0179947308246757,
      "grad_norm": 0.5829957127571106,
      "learning_rate": 9.795199491324302e-05,
      "loss": 0.1287,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 10240,
      "tokens_per_second_per_gpu": 306.4
    },
    {
      "epoch": 1.0189889148481384,
      "grad_norm": 0.33510005474090576,
      "learning_rate": 9.779508418923604e-05,
      "loss": 0.1509,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 10250,
      "tokens_per_second_per_gpu": 368.11
    },
    {
      "epoch": 1.0199830988716012,
      "grad_norm": 0.4509134888648987,
      "learning_rate": 9.763817889640267e-05,
      "loss": 0.1762,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 10260,
      "tokens_per_second_per_gpu": 370.17
    },
    {
      "epoch": 1.0209772828950638,
      "grad_norm": 0.5804489254951477,
      "learning_rate": 9.74812794212339e-05,
      "loss": 0.1569,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 10270,
      "tokens_per_second_per_gpu": 321.78
    },
    {
      "epoch": 1.0219714669185267,
      "grad_norm": 0.5096241235733032,
      "learning_rate": 9.732438615020623e-05,
      "loss": 0.1669,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10280,
      "tokens_per_second_per_gpu": 333.99
    },
    {
      "epoch": 1.0229656509419893,
      "grad_norm": 0.6617136001586914,
      "learning_rate": 9.716749946978102e-05,
      "loss": 0.1857,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 10290,
      "tokens_per_second_per_gpu": 260.5
    },
    {
      "epoch": 1.0239598349654522,
      "grad_norm": 0.5293213725090027,
      "learning_rate": 9.701061976640323e-05,
      "loss": 0.1787,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 10300,
      "tokens_per_second_per_gpu": 382.76
    },
    {
      "epoch": 1.0249540189889148,
      "grad_norm": 0.6056047081947327,
      "learning_rate": 9.685374742650083e-05,
      "loss": 0.1412,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 10310,
      "tokens_per_second_per_gpu": 350.37
    },
    {
      "epoch": 1.0259482030123777,
      "grad_norm": 0.2792462110519409,
      "learning_rate": 9.669688283648344e-05,
      "loss": 0.1149,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 10320,
      "tokens_per_second_per_gpu": 269.87
    },
    {
      "epoch": 1.0269423870358403,
      "grad_norm": 0.4464641511440277,
      "learning_rate": 9.654002638274176e-05,
      "loss": 0.1829,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 10330,
      "tokens_per_second_per_gpu": 411.8
    },
    {
      "epoch": 1.0279365710593031,
      "grad_norm": 0.4715180993080139,
      "learning_rate": 9.638317845164639e-05,
      "loss": 0.1695,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10340,
      "tokens_per_second_per_gpu": 271.86
    },
    {
      "epoch": 1.0289307550827658,
      "grad_norm": 0.43288466334342957,
      "learning_rate": 9.622633942954693e-05,
      "loss": 0.1405,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10350,
      "tokens_per_second_per_gpu": 371.24
    },
    {
      "epoch": 1.0299249391062286,
      "grad_norm": 0.4133097529411316,
      "learning_rate": 9.606950970277106e-05,
      "loss": 0.1613,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10360,
      "tokens_per_second_per_gpu": 335.26
    },
    {
      "epoch": 1.0309191231296913,
      "grad_norm": 0.36935463547706604,
      "learning_rate": 9.591268965762348e-05,
      "loss": 0.1809,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 10370,
      "tokens_per_second_per_gpu": 363.08
    },
    {
      "epoch": 1.031913307153154,
      "grad_norm": 0.6092913746833801,
      "learning_rate": 9.57558796803852e-05,
      "loss": 0.1507,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 10380,
      "tokens_per_second_per_gpu": 296.26
    },
    {
      "epoch": 1.0329074911766167,
      "grad_norm": 0.24895969033241272,
      "learning_rate": 9.559908015731223e-05,
      "loss": 0.1674,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 10390,
      "tokens_per_second_per_gpu": 306.56
    },
    {
      "epoch": 1.0339016752000796,
      "grad_norm": 0.3319539725780487,
      "learning_rate": 9.544229147463502e-05,
      "loss": 0.1683,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 10400,
      "tokens_per_second_per_gpu": 365.18
    },
    {
      "epoch": 1.0348958592235422,
      "grad_norm": 0.3731245994567871,
      "learning_rate": 9.528551401855718e-05,
      "loss": 0.1347,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 10410,
      "tokens_per_second_per_gpu": 304.91
    },
    {
      "epoch": 1.035890043247005,
      "grad_norm": 0.42913174629211426,
      "learning_rate": 9.512874817525474e-05,
      "loss": 0.1558,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 10420,
      "tokens_per_second_per_gpu": 368.64
    },
    {
      "epoch": 1.0368842272704677,
      "grad_norm": 0.5695579648017883,
      "learning_rate": 9.49719943308751e-05,
      "loss": 0.1816,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 10430,
      "tokens_per_second_per_gpu": 382.9
    },
    {
      "epoch": 1.0378784112939305,
      "grad_norm": 0.659853458404541,
      "learning_rate": 9.481525287153616e-05,
      "loss": 0.1885,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10440,
      "tokens_per_second_per_gpu": 362.22
    },
    {
      "epoch": 1.0388725953173932,
      "grad_norm": 0.5355104207992554,
      "learning_rate": 9.465852418332518e-05,
      "loss": 0.192,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 10450,
      "tokens_per_second_per_gpu": 422.56
    },
    {
      "epoch": 1.039866779340856,
      "grad_norm": 0.6618303060531616,
      "learning_rate": 9.450180865229807e-05,
      "loss": 0.1468,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 10460,
      "tokens_per_second_per_gpu": 331.93
    },
    {
      "epoch": 1.0408609633643187,
      "grad_norm": 0.4398798644542694,
      "learning_rate": 9.434510666447838e-05,
      "loss": 0.1703,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 10470,
      "tokens_per_second_per_gpu": 332.46
    },
    {
      "epoch": 1.0418551473877815,
      "grad_norm": 0.25756803154945374,
      "learning_rate": 9.41884186058561e-05,
      "loss": 0.1676,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 10480,
      "tokens_per_second_per_gpu": 332.3
    },
    {
      "epoch": 1.0428493314112441,
      "grad_norm": 0.2897964417934418,
      "learning_rate": 9.403174486238714e-05,
      "loss": 0.1372,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10490,
      "tokens_per_second_per_gpu": 315.04
    },
    {
      "epoch": 1.043843515434707,
      "grad_norm": 0.2930709719657898,
      "learning_rate": 9.387508581999197e-05,
      "loss": 0.1843,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10500,
      "tokens_per_second_per_gpu": 364.97
    },
    {
      "epoch": 1.0448376994581696,
      "grad_norm": 0.3903422951698303,
      "learning_rate": 9.371844186455501e-05,
      "loss": 0.2168,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 10510,
      "tokens_per_second_per_gpu": 377.16
    },
    {
      "epoch": 1.0458318834816325,
      "grad_norm": 0.3957577049732208,
      "learning_rate": 9.356181338192332e-05,
      "loss": 0.1676,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 10520,
      "tokens_per_second_per_gpu": 306.27
    },
    {
      "epoch": 1.046826067505095,
      "grad_norm": 0.5264449119567871,
      "learning_rate": 9.340520075790606e-05,
      "loss": 0.1574,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 10530,
      "tokens_per_second_per_gpu": 322.3
    },
    {
      "epoch": 1.047820251528558,
      "grad_norm": 0.5075907111167908,
      "learning_rate": 9.324860437827312e-05,
      "loss": 0.1808,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 10540,
      "tokens_per_second_per_gpu": 375.27
    },
    {
      "epoch": 1.0488144355520206,
      "grad_norm": 0.28377336263656616,
      "learning_rate": 9.309202462875457e-05,
      "loss": 0.1528,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 10550,
      "tokens_per_second_per_gpu": 356.22
    },
    {
      "epoch": 1.0498086195754834,
      "grad_norm": 0.4088289439678192,
      "learning_rate": 9.293546189503938e-05,
      "loss": 0.1717,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 10560,
      "tokens_per_second_per_gpu": 355.64
    },
    {
      "epoch": 1.050802803598946,
      "grad_norm": 0.1233486533164978,
      "learning_rate": 9.27789165627747e-05,
      "loss": 0.1417,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10570,
      "tokens_per_second_per_gpu": 378.24
    },
    {
      "epoch": 1.051796987622409,
      "grad_norm": 0.54539555311203,
      "learning_rate": 9.26223890175647e-05,
      "loss": 0.131,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 10580,
      "tokens_per_second_per_gpu": 321.17
    },
    {
      "epoch": 1.0527911716458715,
      "grad_norm": 0.3036106526851654,
      "learning_rate": 9.246587964496984e-05,
      "loss": 0.1601,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 10590,
      "tokens_per_second_per_gpu": 327.08
    },
    {
      "epoch": 1.0537853556693344,
      "grad_norm": 0.27965182065963745,
      "learning_rate": 9.230938883050581e-05,
      "loss": 0.1604,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 10600,
      "tokens_per_second_per_gpu": 350.41
    },
    {
      "epoch": 1.054779539692797,
      "grad_norm": 0.33622997999191284,
      "learning_rate": 9.215291695964252e-05,
      "loss": 0.1158,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10610,
      "tokens_per_second_per_gpu": 317.77
    },
    {
      "epoch": 1.0557737237162599,
      "grad_norm": 0.35641202330589294,
      "learning_rate": 9.199646441780332e-05,
      "loss": 0.1452,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 10620,
      "tokens_per_second_per_gpu": 362.54
    },
    {
      "epoch": 1.0567679077397227,
      "grad_norm": 0.612156331539154,
      "learning_rate": 9.184003159036379e-05,
      "loss": 0.2526,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 10630,
      "tokens_per_second_per_gpu": 406.46
    },
    {
      "epoch": 1.0577620917631854,
      "grad_norm": 0.38296809792518616,
      "learning_rate": 9.168361886265113e-05,
      "loss": 0.162,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 10640,
      "tokens_per_second_per_gpu": 319.6
    },
    {
      "epoch": 1.0587562757866482,
      "grad_norm": 0.35586997866630554,
      "learning_rate": 9.15272266199429e-05,
      "loss": 0.1507,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 10650,
      "tokens_per_second_per_gpu": 336.67
    },
    {
      "epoch": 1.0597504598101108,
      "grad_norm": 0.2866521179676056,
      "learning_rate": 9.13708552474663e-05,
      "loss": 0.1771,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10660,
      "tokens_per_second_per_gpu": 376.96
    },
    {
      "epoch": 1.0607446438335737,
      "grad_norm": 0.3502759635448456,
      "learning_rate": 9.1214505130397e-05,
      "loss": 0.1399,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 10670,
      "tokens_per_second_per_gpu": 341.29
    },
    {
      "epoch": 1.0617388278570363,
      "grad_norm": 0.3619881570339203,
      "learning_rate": 9.105817665385846e-05,
      "loss": 0.1584,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 10680,
      "tokens_per_second_per_gpu": 336.33
    },
    {
      "epoch": 1.0627330118804992,
      "grad_norm": 0.6374160051345825,
      "learning_rate": 9.090187020292068e-05,
      "loss": 0.2043,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 10690,
      "tokens_per_second_per_gpu": 381.46
    },
    {
      "epoch": 1.0637271959039618,
      "grad_norm": 0.486200749874115,
      "learning_rate": 9.074558616259954e-05,
      "loss": 0.1564,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 10700,
      "tokens_per_second_per_gpu": 337.08
    },
    {
      "epoch": 1.0647213799274247,
      "grad_norm": 0.8452271223068237,
      "learning_rate": 9.058932491785564e-05,
      "loss": 0.1719,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 10710,
      "tokens_per_second_per_gpu": 342.56
    },
    {
      "epoch": 1.0657155639508873,
      "grad_norm": 0.5657733082771301,
      "learning_rate": 9.043308685359344e-05,
      "loss": 0.184,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 10720,
      "tokens_per_second_per_gpu": 331.5
    },
    {
      "epoch": 1.0667097479743501,
      "grad_norm": 0.3643028438091278,
      "learning_rate": 9.027687235466038e-05,
      "loss": 0.1483,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 10730,
      "tokens_per_second_per_gpu": 335.73
    },
    {
      "epoch": 1.0677039319978128,
      "grad_norm": 0.6992392539978027,
      "learning_rate": 9.012068180584569e-05,
      "loss": 0.1743,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10740,
      "tokens_per_second_per_gpu": 370.01
    },
    {
      "epoch": 1.0686981160212756,
      "grad_norm": 0.5826465487480164,
      "learning_rate": 8.996451559187981e-05,
      "loss": 0.1825,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 10750,
      "tokens_per_second_per_gpu": 345.47
    },
    {
      "epoch": 1.0696923000447383,
      "grad_norm": 0.3841560184955597,
      "learning_rate": 8.980837409743304e-05,
      "loss": 0.205,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 10760,
      "tokens_per_second_per_gpu": 396.05
    },
    {
      "epoch": 1.070686484068201,
      "grad_norm": 0.6366297006607056,
      "learning_rate": 8.965225770711493e-05,
      "loss": 0.1274,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 10770,
      "tokens_per_second_per_gpu": 318.68
    },
    {
      "epoch": 1.0716806680916637,
      "grad_norm": 0.3577064275741577,
      "learning_rate": 8.94961668054731e-05,
      "loss": 0.1986,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 10780,
      "tokens_per_second_per_gpu": 380.85
    },
    {
      "epoch": 1.0726748521151266,
      "grad_norm": 0.4104318618774414,
      "learning_rate": 8.934010177699252e-05,
      "loss": 0.1747,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 10790,
      "tokens_per_second_per_gpu": 301.88
    },
    {
      "epoch": 1.0736690361385892,
      "grad_norm": 0.6288489103317261,
      "learning_rate": 8.918406300609424e-05,
      "loss": 0.1995,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 10800,
      "tokens_per_second_per_gpu": 334.51
    },
    {
      "epoch": 1.074663220162052,
      "grad_norm": 0.4445495009422302,
      "learning_rate": 8.902805087713482e-05,
      "loss": 0.1503,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10810,
      "tokens_per_second_per_gpu": 300.06
    },
    {
      "epoch": 1.0756574041855147,
      "grad_norm": 0.4280208945274353,
      "learning_rate": 8.887206577440502e-05,
      "loss": 0.1843,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 10820,
      "tokens_per_second_per_gpu": 371.92
    },
    {
      "epoch": 1.0766515882089776,
      "grad_norm": 0.5971205830574036,
      "learning_rate": 8.871610808212918e-05,
      "loss": 0.202,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 10830,
      "tokens_per_second_per_gpu": 361.5
    },
    {
      "epoch": 1.0776457722324402,
      "grad_norm": 0.20293696224689484,
      "learning_rate": 8.856017818446402e-05,
      "loss": 0.191,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.46,
      "memory/max_allocated (GiB)": 20.46,
      "step": 10840,
      "tokens_per_second_per_gpu": 292.31
    },
    {
      "epoch": 1.078639956255903,
      "grad_norm": 0.5575738549232483,
      "learning_rate": 8.840427646549788e-05,
      "loss": 0.1699,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 10850,
      "tokens_per_second_per_gpu": 313.54
    },
    {
      "epoch": 1.0796341402793657,
      "grad_norm": 0.5824712514877319,
      "learning_rate": 8.824840330924959e-05,
      "loss": 0.191,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 10860,
      "tokens_per_second_per_gpu": 344.34
    },
    {
      "epoch": 1.0806283243028285,
      "grad_norm": 0.6753024458885193,
      "learning_rate": 8.809255909966771e-05,
      "loss": 0.1255,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 10870,
      "tokens_per_second_per_gpu": 298.62
    },
    {
      "epoch": 1.0816225083262911,
      "grad_norm": 0.6008898019790649,
      "learning_rate": 8.793674422062949e-05,
      "loss": 0.1856,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 10880,
      "tokens_per_second_per_gpu": 399.03
    },
    {
      "epoch": 1.082616692349754,
      "grad_norm": 0.507643461227417,
      "learning_rate": 8.778095905593986e-05,
      "loss": 0.1631,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 10890,
      "tokens_per_second_per_gpu": 359.61
    },
    {
      "epoch": 1.0836108763732166,
      "grad_norm": 0.43057945370674133,
      "learning_rate": 8.762520398933065e-05,
      "loss": 0.1861,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 10900,
      "tokens_per_second_per_gpu": 359.02
    },
    {
      "epoch": 1.0846050603966795,
      "grad_norm": 0.20500341057777405,
      "learning_rate": 8.746947940445946e-05,
      "loss": 0.1294,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 10910,
      "tokens_per_second_per_gpu": 333.37
    },
    {
      "epoch": 1.085599244420142,
      "grad_norm": 0.5553747415542603,
      "learning_rate": 8.73137856849089e-05,
      "loss": 0.1633,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10920,
      "tokens_per_second_per_gpu": 269.33
    },
    {
      "epoch": 1.086593428443605,
      "grad_norm": 0.4564213156700134,
      "learning_rate": 8.715812321418546e-05,
      "loss": 0.1754,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 10930,
      "tokens_per_second_per_gpu": 367.99
    },
    {
      "epoch": 1.0875876124670676,
      "grad_norm": 0.3952537178993225,
      "learning_rate": 8.700249237571879e-05,
      "loss": 0.1829,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 10940,
      "tokens_per_second_per_gpu": 370.04
    },
    {
      "epoch": 1.0885817964905304,
      "grad_norm": 0.45536085963249207,
      "learning_rate": 8.684689355286045e-05,
      "loss": 0.1488,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 10950,
      "tokens_per_second_per_gpu": 357.96
    },
    {
      "epoch": 1.089575980513993,
      "grad_norm": 0.45513296127319336,
      "learning_rate": 8.669132712888328e-05,
      "loss": 0.187,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10960,
      "tokens_per_second_per_gpu": 376.08
    },
    {
      "epoch": 1.090570164537456,
      "grad_norm": 0.44525572657585144,
      "learning_rate": 8.653579348698021e-05,
      "loss": 0.142,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 10970,
      "tokens_per_second_per_gpu": 307.76
    },
    {
      "epoch": 1.0915643485609186,
      "grad_norm": 0.5557878613471985,
      "learning_rate": 8.638029301026351e-05,
      "loss": 0.2135,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.82,
      "memory/max_allocated (GiB)": 19.82,
      "step": 10980,
      "tokens_per_second_per_gpu": 408.3
    },
    {
      "epoch": 1.0925585325843814,
      "grad_norm": 0.3728674054145813,
      "learning_rate": 8.622482608176374e-05,
      "loss": 0.174,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 10990,
      "tokens_per_second_per_gpu": 309.34
    },
    {
      "epoch": 1.093552716607844,
      "grad_norm": 0.38084620237350464,
      "learning_rate": 8.606939308442877e-05,
      "loss": 0.1752,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.24,
      "memory/max_allocated (GiB)": 19.24,
      "step": 11000,
      "tokens_per_second_per_gpu": 280.4
    },
    {
      "epoch": 1.0945469006313069,
      "grad_norm": 0.47729748487472534,
      "learning_rate": 8.591399440112296e-05,
      "loss": 0.1787,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 11010,
      "tokens_per_second_per_gpu": 344.64
    },
    {
      "epoch": 1.0955410846547695,
      "grad_norm": 0.567870020866394,
      "learning_rate": 8.575863041462603e-05,
      "loss": 0.176,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 11020,
      "tokens_per_second_per_gpu": 353.03
    },
    {
      "epoch": 1.0965352686782324,
      "grad_norm": 0.37033653259277344,
      "learning_rate": 8.560330150763243e-05,
      "loss": 0.1765,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11030,
      "tokens_per_second_per_gpu": 329.79
    },
    {
      "epoch": 1.097529452701695,
      "grad_norm": 0.6800894141197205,
      "learning_rate": 8.544800806274998e-05,
      "loss": 0.2017,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 11040,
      "tokens_per_second_per_gpu": 303.81
    },
    {
      "epoch": 1.0985236367251578,
      "grad_norm": 0.6861316561698914,
      "learning_rate": 8.529275046249934e-05,
      "loss": 0.245,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 11050,
      "tokens_per_second_per_gpu": 400.94
    },
    {
      "epoch": 1.0995178207486205,
      "grad_norm": 0.5849536657333374,
      "learning_rate": 8.513752908931273e-05,
      "loss": 0.1653,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 11060,
      "tokens_per_second_per_gpu": 383.81
    },
    {
      "epoch": 1.1005120047720833,
      "grad_norm": 0.43626952171325684,
      "learning_rate": 8.498234432553328e-05,
      "loss": 0.1368,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 11070,
      "tokens_per_second_per_gpu": 332.97
    },
    {
      "epoch": 1.101506188795546,
      "grad_norm": 0.5107852816581726,
      "learning_rate": 8.482719655341374e-05,
      "loss": 0.1739,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.54,
      "memory/max_allocated (GiB)": 21.54,
      "step": 11080,
      "tokens_per_second_per_gpu": 333.13
    },
    {
      "epoch": 1.1025003728190088,
      "grad_norm": 0.4791623651981354,
      "learning_rate": 8.467208615511599e-05,
      "loss": 0.1172,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 11090,
      "tokens_per_second_per_gpu": 310.57
    },
    {
      "epoch": 1.1034945568424714,
      "grad_norm": 0.4793543219566345,
      "learning_rate": 8.451701351270965e-05,
      "loss": 0.168,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 11100,
      "tokens_per_second_per_gpu": 280.35
    },
    {
      "epoch": 1.1044887408659343,
      "grad_norm": 0.41162610054016113,
      "learning_rate": 8.436197900817145e-05,
      "loss": 0.1585,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 11110,
      "tokens_per_second_per_gpu": 368.7
    },
    {
      "epoch": 1.105482924889397,
      "grad_norm": 0.7751170992851257,
      "learning_rate": 8.420698302338407e-05,
      "loss": 0.165,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 11120,
      "tokens_per_second_per_gpu": 345.65
    },
    {
      "epoch": 1.1064771089128598,
      "grad_norm": 0.2921695113182068,
      "learning_rate": 8.405202594013546e-05,
      "loss": 0.1529,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 11130,
      "tokens_per_second_per_gpu": 338.72
    },
    {
      "epoch": 1.1074712929363226,
      "grad_norm": 0.5489593744277954,
      "learning_rate": 8.389710814011764e-05,
      "loss": 0.201,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 11140,
      "tokens_per_second_per_gpu": 401.46
    },
    {
      "epoch": 1.1084654769597853,
      "grad_norm": 0.5984140634536743,
      "learning_rate": 8.37422300049259e-05,
      "loss": 0.1832,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.46,
      "memory/max_allocated (GiB)": 20.46,
      "step": 11150,
      "tokens_per_second_per_gpu": 314.29
    },
    {
      "epoch": 1.1094596609832479,
      "grad_norm": 0.4428146779537201,
      "learning_rate": 8.358739191605783e-05,
      "loss": 0.1515,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 11160,
      "tokens_per_second_per_gpu": 295.44
    },
    {
      "epoch": 1.1104538450067107,
      "grad_norm": 0.4830903112888336,
      "learning_rate": 8.343259425491234e-05,
      "loss": 0.1363,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11170,
      "tokens_per_second_per_gpu": 313.11
    },
    {
      "epoch": 1.1114480290301736,
      "grad_norm": 0.8339301943778992,
      "learning_rate": 8.327783740278882e-05,
      "loss": 0.1691,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11180,
      "tokens_per_second_per_gpu": 309.19
    },
    {
      "epoch": 1.1124422130536362,
      "grad_norm": 0.30678462982177734,
      "learning_rate": 8.312312174088606e-05,
      "loss": 0.1616,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 11190,
      "tokens_per_second_per_gpu": 349.77
    },
    {
      "epoch": 1.113436397077099,
      "grad_norm": 0.1882571429014206,
      "learning_rate": 8.296844765030147e-05,
      "loss": 0.1807,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 11200,
      "tokens_per_second_per_gpu": 331.03
    },
    {
      "epoch": 1.1144305811005617,
      "grad_norm": 0.5480899214744568,
      "learning_rate": 8.281381551203e-05,
      "loss": 0.1784,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 11210,
      "tokens_per_second_per_gpu": 367.42
    },
    {
      "epoch": 1.1154247651240246,
      "grad_norm": 0.3690171241760254,
      "learning_rate": 8.265922570696336e-05,
      "loss": 0.1421,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11220,
      "tokens_per_second_per_gpu": 338.8
    },
    {
      "epoch": 1.1164189491474872,
      "grad_norm": 0.3193921744823456,
      "learning_rate": 8.250467861588879e-05,
      "loss": 0.1304,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 11230,
      "tokens_per_second_per_gpu": 316.67
    },
    {
      "epoch": 1.11741313317095,
      "grad_norm": 0.4590331017971039,
      "learning_rate": 8.235017461948858e-05,
      "loss": 0.1901,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 11240,
      "tokens_per_second_per_gpu": 371.35
    },
    {
      "epoch": 1.1184073171944127,
      "grad_norm": 0.5001057386398315,
      "learning_rate": 8.219571409833862e-05,
      "loss": 0.1895,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 11250,
      "tokens_per_second_per_gpu": 379.59
    },
    {
      "epoch": 1.1194015012178755,
      "grad_norm": 0.5432204008102417,
      "learning_rate": 8.204129743290783e-05,
      "loss": 0.1579,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11260,
      "tokens_per_second_per_gpu": 294.03
    },
    {
      "epoch": 1.1203956852413381,
      "grad_norm": 0.4477890431880951,
      "learning_rate": 8.188692500355716e-05,
      "loss": 0.23,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 11270,
      "tokens_per_second_per_gpu": 348.24
    },
    {
      "epoch": 1.121389869264801,
      "grad_norm": 0.5133992433547974,
      "learning_rate": 8.173259719053847e-05,
      "loss": 0.1724,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 11280,
      "tokens_per_second_per_gpu": 379.58
    },
    {
      "epoch": 1.1223840532882636,
      "grad_norm": 0.369495153427124,
      "learning_rate": 8.157831437399383e-05,
      "loss": 0.1225,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 11290,
      "tokens_per_second_per_gpu": 341.16
    },
    {
      "epoch": 1.1233782373117265,
      "grad_norm": 0.30792126059532166,
      "learning_rate": 8.142407693395438e-05,
      "loss": 0.1445,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 11300,
      "tokens_per_second_per_gpu": 301.05
    },
    {
      "epoch": 1.124372421335189,
      "grad_norm": 0.3198167085647583,
      "learning_rate": 8.126988525033958e-05,
      "loss": 0.1675,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 11310,
      "tokens_per_second_per_gpu": 310.65
    },
    {
      "epoch": 1.125366605358652,
      "grad_norm": 0.6837435960769653,
      "learning_rate": 8.111573970295607e-05,
      "loss": 0.2455,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 11320,
      "tokens_per_second_per_gpu": 354.62
    },
    {
      "epoch": 1.1263607893821146,
      "grad_norm": 0.5170619487762451,
      "learning_rate": 8.096164067149701e-05,
      "loss": 0.1768,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 11330,
      "tokens_per_second_per_gpu": 287.06
    },
    {
      "epoch": 1.1273549734055774,
      "grad_norm": 0.4704976975917816,
      "learning_rate": 8.080758853554075e-05,
      "loss": 0.1772,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 11340,
      "tokens_per_second_per_gpu": 308.38
    },
    {
      "epoch": 1.12834915742904,
      "grad_norm": 0.5932011008262634,
      "learning_rate": 8.065358367455038e-05,
      "loss": 0.1593,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 11350,
      "tokens_per_second_per_gpu": 348.84
    },
    {
      "epoch": 1.129343341452503,
      "grad_norm": 0.6632476449012756,
      "learning_rate": 8.049962646787235e-05,
      "loss": 0.1394,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 11360,
      "tokens_per_second_per_gpu": 250.63
    },
    {
      "epoch": 1.1303375254759656,
      "grad_norm": 0.5370444655418396,
      "learning_rate": 8.034571729473587e-05,
      "loss": 0.1832,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 11370,
      "tokens_per_second_per_gpu": 326.76
    },
    {
      "epoch": 1.1313317094994284,
      "grad_norm": 0.42281845211982727,
      "learning_rate": 8.019185653425168e-05,
      "loss": 0.1373,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11380,
      "tokens_per_second_per_gpu": 307.3
    },
    {
      "epoch": 1.132325893522891,
      "grad_norm": 0.27983081340789795,
      "learning_rate": 8.00380445654114e-05,
      "loss": 0.1607,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 11390,
      "tokens_per_second_per_gpu": 330.95
    },
    {
      "epoch": 1.1333200775463539,
      "grad_norm": 0.6440351009368896,
      "learning_rate": 7.988428176708644e-05,
      "loss": 0.2021,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 11400,
      "tokens_per_second_per_gpu": 397.31
    },
    {
      "epoch": 1.1343142615698165,
      "grad_norm": 0.4090474843978882,
      "learning_rate": 7.9730568518027e-05,
      "loss": 0.2105,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 11410,
      "tokens_per_second_per_gpu": 342.86
    },
    {
      "epoch": 1.1353084455932794,
      "grad_norm": 0.2903016209602356,
      "learning_rate": 7.957690519686137e-05,
      "loss": 0.1956,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 11420,
      "tokens_per_second_per_gpu": 362.76
    },
    {
      "epoch": 1.136302629616742,
      "grad_norm": 0.5340930223464966,
      "learning_rate": 7.942329218209474e-05,
      "loss": 0.1982,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11430,
      "tokens_per_second_per_gpu": 312.98
    },
    {
      "epoch": 1.1372968136402049,
      "grad_norm": 0.515466570854187,
      "learning_rate": 7.926972985210848e-05,
      "loss": 0.1499,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11440,
      "tokens_per_second_per_gpu": 289.34
    },
    {
      "epoch": 1.1382909976636675,
      "grad_norm": 0.29830217361450195,
      "learning_rate": 7.911621858515901e-05,
      "loss": 0.1676,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 11450,
      "tokens_per_second_per_gpu": 360.42
    },
    {
      "epoch": 1.1392851816871303,
      "grad_norm": 0.4327114224433899,
      "learning_rate": 7.896275875937709e-05,
      "loss": 0.1715,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.68,
      "memory/max_allocated (GiB)": 19.68,
      "step": 11460,
      "tokens_per_second_per_gpu": 334.12
    },
    {
      "epoch": 1.140279365710593,
      "grad_norm": 0.5234288573265076,
      "learning_rate": 7.880935075276663e-05,
      "loss": 0.1645,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 11470,
      "tokens_per_second_per_gpu": 263.98
    },
    {
      "epoch": 1.1412735497340558,
      "grad_norm": 0.35436221957206726,
      "learning_rate": 7.865599494320402e-05,
      "loss": 0.1548,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 11480,
      "tokens_per_second_per_gpu": 328.71
    },
    {
      "epoch": 1.1422677337575184,
      "grad_norm": 0.5308985710144043,
      "learning_rate": 7.850269170843702e-05,
      "loss": 0.1843,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.23,
      "memory/max_allocated (GiB)": 19.23,
      "step": 11490,
      "tokens_per_second_per_gpu": 341.98
    },
    {
      "epoch": 1.1432619177809813,
      "grad_norm": 0.4634507894515991,
      "learning_rate": 7.834944142608394e-05,
      "loss": 0.2258,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 11500,
      "tokens_per_second_per_gpu": 409.58
    },
    {
      "epoch": 1.144256101804444,
      "grad_norm": 0.5410236716270447,
      "learning_rate": 7.819624447363252e-05,
      "loss": 0.1471,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 11510,
      "tokens_per_second_per_gpu": 252.15
    },
    {
      "epoch": 1.1452502858279068,
      "grad_norm": 0.3912404477596283,
      "learning_rate": 7.80431012284393e-05,
      "loss": 0.1358,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 11520,
      "tokens_per_second_per_gpu": 310.18
    },
    {
      "epoch": 1.1462444698513694,
      "grad_norm": 0.2692343592643738,
      "learning_rate": 7.789001206772849e-05,
      "loss": 0.1441,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 11530,
      "tokens_per_second_per_gpu": 230.07
    },
    {
      "epoch": 1.1472386538748323,
      "grad_norm": 0.44046950340270996,
      "learning_rate": 7.773697736859098e-05,
      "loss": 0.1424,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 11540,
      "tokens_per_second_per_gpu": 351.3
    },
    {
      "epoch": 1.148232837898295,
      "grad_norm": 0.48867931962013245,
      "learning_rate": 7.758399750798364e-05,
      "loss": 0.194,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 11550,
      "tokens_per_second_per_gpu": 280.59
    },
    {
      "epoch": 1.1492270219217577,
      "grad_norm": 0.3554067313671112,
      "learning_rate": 7.743107286272812e-05,
      "loss": 0.1481,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 11560,
      "tokens_per_second_per_gpu": 367.0
    },
    {
      "epoch": 1.1502212059452204,
      "grad_norm": 0.4472406506538391,
      "learning_rate": 7.727820380951022e-05,
      "loss": 0.1755,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11570,
      "tokens_per_second_per_gpu": 291.81
    },
    {
      "epoch": 1.1512153899686832,
      "grad_norm": 0.47527438402175903,
      "learning_rate": 7.712539072487867e-05,
      "loss": 0.18,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.99,
      "memory/max_allocated (GiB)": 19.99,
      "step": 11580,
      "tokens_per_second_per_gpu": 312.04
    },
    {
      "epoch": 1.1522095739921459,
      "grad_norm": 0.4945676624774933,
      "learning_rate": 7.697263398524448e-05,
      "loss": 0.1479,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 11590,
      "tokens_per_second_per_gpu": 271.43
    },
    {
      "epoch": 1.1532037580156087,
      "grad_norm": 0.3966761529445648,
      "learning_rate": 7.681993396687968e-05,
      "loss": 0.1455,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 11600,
      "tokens_per_second_per_gpu": 324.63
    },
    {
      "epoch": 1.1541979420390716,
      "grad_norm": 0.34350907802581787,
      "learning_rate": 7.666729104591678e-05,
      "loss": 0.1397,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11610,
      "tokens_per_second_per_gpu": 292.42
    },
    {
      "epoch": 1.1551921260625342,
      "grad_norm": 0.5656160712242126,
      "learning_rate": 7.651470559834747e-05,
      "loss": 0.1825,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11620,
      "tokens_per_second_per_gpu": 367.89
    },
    {
      "epoch": 1.1561863100859968,
      "grad_norm": 0.4410739541053772,
      "learning_rate": 7.636217800002203e-05,
      "loss": 0.1718,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11630,
      "tokens_per_second_per_gpu": 365.13
    },
    {
      "epoch": 1.1571804941094597,
      "grad_norm": 0.5351715087890625,
      "learning_rate": 7.620970862664811e-05,
      "loss": 0.1638,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 11640,
      "tokens_per_second_per_gpu": 301.44
    },
    {
      "epoch": 1.1581746781329225,
      "grad_norm": 0.5122233033180237,
      "learning_rate": 7.605729785379005e-05,
      "loss": 0.2124,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.46,
      "memory/max_allocated (GiB)": 20.46,
      "step": 11650,
      "tokens_per_second_per_gpu": 339.63
    },
    {
      "epoch": 1.1591688621563851,
      "grad_norm": 0.49694183468818665,
      "learning_rate": 7.590494605686781e-05,
      "loss": 0.1758,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 11660,
      "tokens_per_second_per_gpu": 332.59
    },
    {
      "epoch": 1.1601630461798478,
      "grad_norm": 0.5573239922523499,
      "learning_rate": 7.5752653611156e-05,
      "loss": 0.1506,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 11670,
      "tokens_per_second_per_gpu": 326.32
    },
    {
      "epoch": 1.1611572302033106,
      "grad_norm": 0.4641178250312805,
      "learning_rate": 7.560042089178319e-05,
      "loss": 0.1212,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 11680,
      "tokens_per_second_per_gpu": 253.53
    },
    {
      "epoch": 1.1621514142267735,
      "grad_norm": 0.2381938248872757,
      "learning_rate": 7.544824827373064e-05,
      "loss": 0.1489,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 11690,
      "tokens_per_second_per_gpu": 326.27
    },
    {
      "epoch": 1.1631455982502361,
      "grad_norm": 0.5554631948471069,
      "learning_rate": 7.529613613183174e-05,
      "loss": 0.2159,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 11700,
      "tokens_per_second_per_gpu": 373.55
    },
    {
      "epoch": 1.1641397822736987,
      "grad_norm": 0.3771873414516449,
      "learning_rate": 7.514408484077081e-05,
      "loss": 0.1905,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 11710,
      "tokens_per_second_per_gpu": 313.67
    },
    {
      "epoch": 1.1651339662971616,
      "grad_norm": 0.5581318736076355,
      "learning_rate": 7.499209477508238e-05,
      "loss": 0.1409,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 11720,
      "tokens_per_second_per_gpu": 259.67
    },
    {
      "epoch": 1.1661281503206244,
      "grad_norm": 0.4785107970237732,
      "learning_rate": 7.484016630915003e-05,
      "loss": 0.1236,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 11730,
      "tokens_per_second_per_gpu": 309.21
    },
    {
      "epoch": 1.167122334344087,
      "grad_norm": 0.48833346366882324,
      "learning_rate": 7.468829981720574e-05,
      "loss": 0.1259,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.41,
      "memory/max_allocated (GiB)": 21.41,
      "step": 11740,
      "tokens_per_second_per_gpu": 348.37
    },
    {
      "epoch": 1.16811651836755,
      "grad_norm": 0.3307853639125824,
      "learning_rate": 7.453649567332871e-05,
      "loss": 0.1447,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 11750,
      "tokens_per_second_per_gpu": 261.54
    },
    {
      "epoch": 1.1691107023910126,
      "grad_norm": 0.6457106471061707,
      "learning_rate": 7.438475425144469e-05,
      "loss": 0.1734,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 11760,
      "tokens_per_second_per_gpu": 323.76
    },
    {
      "epoch": 1.1701048864144754,
      "grad_norm": 0.4975489675998688,
      "learning_rate": 7.423307592532484e-05,
      "loss": 0.1597,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 11770,
      "tokens_per_second_per_gpu": 333.31
    },
    {
      "epoch": 1.171099070437938,
      "grad_norm": 0.5595365166664124,
      "learning_rate": 7.408146106858496e-05,
      "loss": 0.1448,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11780,
      "tokens_per_second_per_gpu": 302.61
    },
    {
      "epoch": 1.172093254461401,
      "grad_norm": 0.5762649178504944,
      "learning_rate": 7.392991005468449e-05,
      "loss": 0.1829,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 11790,
      "tokens_per_second_per_gpu": 323.36
    },
    {
      "epoch": 1.1730874384848635,
      "grad_norm": 0.4875398278236389,
      "learning_rate": 7.377842325692557e-05,
      "loss": 0.1615,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.54,
      "memory/max_allocated (GiB)": 21.54,
      "step": 11800,
      "tokens_per_second_per_gpu": 304.43
    },
    {
      "epoch": 1.1740816225083264,
      "grad_norm": 0.7832923531532288,
      "learning_rate": 7.362700104845226e-05,
      "loss": 0.1935,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11810,
      "tokens_per_second_per_gpu": 332.32
    },
    {
      "epoch": 1.175075806531789,
      "grad_norm": 0.39051079750061035,
      "learning_rate": 7.34756438022494e-05,
      "loss": 0.1337,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11820,
      "tokens_per_second_per_gpu": 287.23
    },
    {
      "epoch": 1.1760699905552519,
      "grad_norm": 0.3700825273990631,
      "learning_rate": 7.332435189114194e-05,
      "loss": 0.1535,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11830,
      "tokens_per_second_per_gpu": 341.88
    },
    {
      "epoch": 1.1770641745787145,
      "grad_norm": 0.6411652565002441,
      "learning_rate": 7.317312568779375e-05,
      "loss": 0.1586,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 11840,
      "tokens_per_second_per_gpu": 334.43
    },
    {
      "epoch": 1.1780583586021773,
      "grad_norm": 0.46753886342048645,
      "learning_rate": 7.302196556470701e-05,
      "loss": 0.1639,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.66,
      "memory/max_allocated (GiB)": 19.66,
      "step": 11850,
      "tokens_per_second_per_gpu": 329.98
    },
    {
      "epoch": 1.17905254262564,
      "grad_norm": 0.7362424731254578,
      "learning_rate": 7.287087189422099e-05,
      "loss": 0.2674,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 11860,
      "tokens_per_second_per_gpu": 360.74
    },
    {
      "epoch": 1.1800467266491028,
      "grad_norm": 0.5120696425437927,
      "learning_rate": 7.271984504851141e-05,
      "loss": 0.166,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 11870,
      "tokens_per_second_per_gpu": 321.01
    },
    {
      "epoch": 1.1810409106725654,
      "grad_norm": 0.5142634510993958,
      "learning_rate": 7.256888539958923e-05,
      "loss": 0.1663,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 11880,
      "tokens_per_second_per_gpu": 348.0
    },
    {
      "epoch": 1.1820350946960283,
      "grad_norm": 0.5077565908432007,
      "learning_rate": 7.241799331930006e-05,
      "loss": 0.2012,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 11890,
      "tokens_per_second_per_gpu": 333.88
    },
    {
      "epoch": 1.183029278719491,
      "grad_norm": 0.5016006231307983,
      "learning_rate": 7.226716917932289e-05,
      "loss": 0.1986,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11900,
      "tokens_per_second_per_gpu": 349.37
    },
    {
      "epoch": 1.1840234627429538,
      "grad_norm": 0.5038326382637024,
      "learning_rate": 7.21164133511695e-05,
      "loss": 0.1379,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 11910,
      "tokens_per_second_per_gpu": 331.98
    },
    {
      "epoch": 1.1850176467664164,
      "grad_norm": 0.44399407505989075,
      "learning_rate": 7.196572620618336e-05,
      "loss": 0.1479,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 11920,
      "tokens_per_second_per_gpu": 295.3
    },
    {
      "epoch": 1.1860118307898793,
      "grad_norm": 0.5076953172683716,
      "learning_rate": 7.181510811553874e-05,
      "loss": 0.2141,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11930,
      "tokens_per_second_per_gpu": 342.74
    },
    {
      "epoch": 1.187006014813342,
      "grad_norm": 0.39150622487068176,
      "learning_rate": 7.166455945023989e-05,
      "loss": 0.1896,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11940,
      "tokens_per_second_per_gpu": 341.43
    },
    {
      "epoch": 1.1880001988368047,
      "grad_norm": 0.305494099855423,
      "learning_rate": 7.151408058111991e-05,
      "loss": 0.1467,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 11950,
      "tokens_per_second_per_gpu": 271.68
    },
    {
      "epoch": 1.1889943828602674,
      "grad_norm": 0.5669624209403992,
      "learning_rate": 7.136367187884014e-05,
      "loss": 0.1725,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 11960,
      "tokens_per_second_per_gpu": 305.75
    },
    {
      "epoch": 1.1899885668837302,
      "grad_norm": 0.7032368183135986,
      "learning_rate": 7.121333371388889e-05,
      "loss": 0.164,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 11970,
      "tokens_per_second_per_gpu": 247.59
    },
    {
      "epoch": 1.1909827509071929,
      "grad_norm": 0.5729434490203857,
      "learning_rate": 7.106306645658095e-05,
      "loss": 0.1808,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 11980,
      "tokens_per_second_per_gpu": 358.52
    },
    {
      "epoch": 1.1919769349306557,
      "grad_norm": 0.6809588074684143,
      "learning_rate": 7.091287047705626e-05,
      "loss": 0.1459,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 11990,
      "tokens_per_second_per_gpu": 316.28
    },
    {
      "epoch": 1.1929711189541183,
      "grad_norm": 0.34328803420066833,
      "learning_rate": 7.076274614527934e-05,
      "loss": 0.1697,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 12000,
      "tokens_per_second_per_gpu": 338.71
    },
    {
      "epoch": 1.1939653029775812,
      "grad_norm": 0.7516032457351685,
      "learning_rate": 7.061269383103804e-05,
      "loss": 0.1699,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 12010,
      "tokens_per_second_per_gpu": 342.16
    },
    {
      "epoch": 1.1949594870010438,
      "grad_norm": 0.4296916425228119,
      "learning_rate": 7.046271390394303e-05,
      "loss": 0.1585,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 12020,
      "tokens_per_second_per_gpu": 367.62
    },
    {
      "epoch": 1.1959536710245067,
      "grad_norm": 0.2905307412147522,
      "learning_rate": 7.031280673342648e-05,
      "loss": 0.1239,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 12030,
      "tokens_per_second_per_gpu": 303.39
    },
    {
      "epoch": 1.1969478550479693,
      "grad_norm": 0.2632843554019928,
      "learning_rate": 7.016297268874152e-05,
      "loss": 0.1541,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12040,
      "tokens_per_second_per_gpu": 354.37
    },
    {
      "epoch": 1.1979420390714322,
      "grad_norm": 0.41677621006965637,
      "learning_rate": 7.001321213896099e-05,
      "loss": 0.125,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 12050,
      "tokens_per_second_per_gpu": 372.01
    },
    {
      "epoch": 1.1989362230948948,
      "grad_norm": 0.6199091672897339,
      "learning_rate": 6.98635254529768e-05,
      "loss": 0.1617,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12060,
      "tokens_per_second_per_gpu": 336.05
    },
    {
      "epoch": 1.1999304071183576,
      "grad_norm": 0.6814326643943787,
      "learning_rate": 6.971391299949895e-05,
      "loss": 0.1585,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 12070,
      "tokens_per_second_per_gpu": 285.66
    },
    {
      "epoch": 1.2009245911418203,
      "grad_norm": 0.5412986874580383,
      "learning_rate": 6.956437514705447e-05,
      "loss": 0.2202,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.07,
      "memory/max_allocated (GiB)": 18.07,
      "step": 12080,
      "tokens_per_second_per_gpu": 333.0
    },
    {
      "epoch": 1.2019187751652831,
      "grad_norm": 0.6325846314430237,
      "learning_rate": 6.941491226398675e-05,
      "loss": 0.1899,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 12090,
      "tokens_per_second_per_gpu": 333.08
    },
    {
      "epoch": 1.2029129591887457,
      "grad_norm": 0.5769904255867004,
      "learning_rate": 6.926552471845439e-05,
      "loss": 0.1916,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 12100,
      "tokens_per_second_per_gpu": 341.54
    },
    {
      "epoch": 1.2039071432122086,
      "grad_norm": 0.5062241554260254,
      "learning_rate": 6.911621287843058e-05,
      "loss": 0.1523,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 12110,
      "tokens_per_second_per_gpu": 317.78
    },
    {
      "epoch": 1.2049013272356712,
      "grad_norm": 0.6168234348297119,
      "learning_rate": 6.896697711170183e-05,
      "loss": 0.1967,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 12120,
      "tokens_per_second_per_gpu": 339.37
    },
    {
      "epoch": 1.205895511259134,
      "grad_norm": 0.3159475326538086,
      "learning_rate": 6.881781778586745e-05,
      "loss": 0.1494,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 12130,
      "tokens_per_second_per_gpu": 314.84
    },
    {
      "epoch": 1.2068896952825967,
      "grad_norm": 0.697409451007843,
      "learning_rate": 6.866873526833838e-05,
      "loss": 0.1419,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12140,
      "tokens_per_second_per_gpu": 286.26
    },
    {
      "epoch": 1.2078838793060596,
      "grad_norm": 0.4400809407234192,
      "learning_rate": 6.851972992633636e-05,
      "loss": 0.1646,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 12150,
      "tokens_per_second_per_gpu": 276.17
    },
    {
      "epoch": 1.2088780633295224,
      "grad_norm": 0.4484960436820984,
      "learning_rate": 6.837080212689302e-05,
      "loss": 0.1717,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12160,
      "tokens_per_second_per_gpu": 275.23
    },
    {
      "epoch": 1.209872247352985,
      "grad_norm": 0.24361641705036163,
      "learning_rate": 6.822195223684906e-05,
      "loss": 0.1457,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 12170,
      "tokens_per_second_per_gpu": 309.09
    },
    {
      "epoch": 1.2108664313764477,
      "grad_norm": 0.4752653241157532,
      "learning_rate": 6.807318062285314e-05,
      "loss": 0.1669,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12180,
      "tokens_per_second_per_gpu": 310.15
    },
    {
      "epoch": 1.2118606153999105,
      "grad_norm": 0.437656968832016,
      "learning_rate": 6.792448765136124e-05,
      "loss": 0.1572,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 12190,
      "tokens_per_second_per_gpu": 283.06
    },
    {
      "epoch": 1.2128547994233734,
      "grad_norm": 0.8095329999923706,
      "learning_rate": 6.777587368863558e-05,
      "loss": 0.1974,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 12200,
      "tokens_per_second_per_gpu": 345.42
    },
    {
      "epoch": 1.213848983446836,
      "grad_norm": 0.48734039068222046,
      "learning_rate": 6.762733910074372e-05,
      "loss": 0.1423,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 12210,
      "tokens_per_second_per_gpu": 289.48
    },
    {
      "epoch": 1.2148431674702986,
      "grad_norm": 0.6078053712844849,
      "learning_rate": 6.747888425355783e-05,
      "loss": 0.1729,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 12220,
      "tokens_per_second_per_gpu": 326.08
    },
    {
      "epoch": 1.2158373514937615,
      "grad_norm": 0.4170425832271576,
      "learning_rate": 6.733050951275347e-05,
      "loss": 0.1939,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12230,
      "tokens_per_second_per_gpu": 326.45
    },
    {
      "epoch": 1.2168315355172243,
      "grad_norm": 0.6094039678573608,
      "learning_rate": 6.71822152438091e-05,
      "loss": 0.1701,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 12240,
      "tokens_per_second_per_gpu": 344.33
    },
    {
      "epoch": 1.217825719540687,
      "grad_norm": 0.5931222438812256,
      "learning_rate": 6.703400181200472e-05,
      "loss": 0.168,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 12250,
      "tokens_per_second_per_gpu": 343.43
    },
    {
      "epoch": 1.2188199035641498,
      "grad_norm": 0.5588046908378601,
      "learning_rate": 6.688586958242144e-05,
      "loss": 0.1449,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 12260,
      "tokens_per_second_per_gpu": 323.04
    },
    {
      "epoch": 1.2198140875876125,
      "grad_norm": 0.7040362358093262,
      "learning_rate": 6.673781891994018e-05,
      "loss": 0.1925,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 12270,
      "tokens_per_second_per_gpu": 280.89
    },
    {
      "epoch": 1.2208082716110753,
      "grad_norm": 0.46226373314857483,
      "learning_rate": 6.658985018924104e-05,
      "loss": 0.1817,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 12280,
      "tokens_per_second_per_gpu": 363.62
    },
    {
      "epoch": 1.221802455634538,
      "grad_norm": 0.6106900572776794,
      "learning_rate": 6.644196375480228e-05,
      "loss": 0.1602,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 12290,
      "tokens_per_second_per_gpu": 242.36
    },
    {
      "epoch": 1.2227966396580008,
      "grad_norm": 0.3931345045566559,
      "learning_rate": 6.629415998089947e-05,
      "loss": 0.2175,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12300,
      "tokens_per_second_per_gpu": 319.03
    },
    {
      "epoch": 1.2237908236814634,
      "grad_norm": 0.5207247734069824,
      "learning_rate": 6.61464392316045e-05,
      "loss": 0.1477,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12310,
      "tokens_per_second_per_gpu": 324.44
    },
    {
      "epoch": 1.2247850077049263,
      "grad_norm": 0.709812343120575,
      "learning_rate": 6.599880187078479e-05,
      "loss": 0.1891,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 12320,
      "tokens_per_second_per_gpu": 357.92
    },
    {
      "epoch": 1.225779191728389,
      "grad_norm": 0.7825199961662292,
      "learning_rate": 6.585124826210245e-05,
      "loss": 0.1395,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 12330,
      "tokens_per_second_per_gpu": 265.35
    },
    {
      "epoch": 1.2267733757518517,
      "grad_norm": 0.6053150296211243,
      "learning_rate": 6.570377876901311e-05,
      "loss": 0.1813,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 12340,
      "tokens_per_second_per_gpu": 374.3
    },
    {
      "epoch": 1.2277675597753144,
      "grad_norm": 0.6844688653945923,
      "learning_rate": 6.555639375476532e-05,
      "loss": 0.2049,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 12350,
      "tokens_per_second_per_gpu": 318.45
    },
    {
      "epoch": 1.2287617437987772,
      "grad_norm": 0.5875769853591919,
      "learning_rate": 6.540909358239954e-05,
      "loss": 0.1821,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 12360,
      "tokens_per_second_per_gpu": 312.63
    },
    {
      "epoch": 1.2297559278222399,
      "grad_norm": 0.4794519543647766,
      "learning_rate": 6.526187861474727e-05,
      "loss": 0.1581,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 12370,
      "tokens_per_second_per_gpu": 344.97
    },
    {
      "epoch": 1.2307501118457027,
      "grad_norm": 0.4545304775238037,
      "learning_rate": 6.511474921442997e-05,
      "loss": 0.1482,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 12380,
      "tokens_per_second_per_gpu": 307.3
    },
    {
      "epoch": 1.2317442958691653,
      "grad_norm": 0.4457061290740967,
      "learning_rate": 6.496770574385858e-05,
      "loss": 0.1688,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 12390,
      "tokens_per_second_per_gpu": 357.22
    },
    {
      "epoch": 1.2327384798926282,
      "grad_norm": 0.5178716778755188,
      "learning_rate": 6.482074856523215e-05,
      "loss": 0.1487,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 12400,
      "tokens_per_second_per_gpu": 317.6
    },
    {
      "epoch": 1.2337326639160908,
      "grad_norm": 0.2689170837402344,
      "learning_rate": 6.467387804053731e-05,
      "loss": 0.208,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 12410,
      "tokens_per_second_per_gpu": 330.76
    },
    {
      "epoch": 1.2347268479395537,
      "grad_norm": 0.5642197132110596,
      "learning_rate": 6.45270945315472e-05,
      "loss": 0.1667,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 12420,
      "tokens_per_second_per_gpu": 299.73
    },
    {
      "epoch": 1.2357210319630163,
      "grad_norm": 0.38734033703804016,
      "learning_rate": 6.438039839982066e-05,
      "loss": 0.1317,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 12430,
      "tokens_per_second_per_gpu": 298.98
    },
    {
      "epoch": 1.2367152159864792,
      "grad_norm": 0.39504683017730713,
      "learning_rate": 6.42337900067012e-05,
      "loss": 0.1357,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 12440,
      "tokens_per_second_per_gpu": 359.51
    },
    {
      "epoch": 1.2377094000099418,
      "grad_norm": 0.38826897740364075,
      "learning_rate": 6.408726971331631e-05,
      "loss": 0.1919,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 12450,
      "tokens_per_second_per_gpu": 327.89
    },
    {
      "epoch": 1.2387035840334046,
      "grad_norm": 0.6274026036262512,
      "learning_rate": 6.39408378805765e-05,
      "loss": 0.1613,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 12460,
      "tokens_per_second_per_gpu": 325.25
    },
    {
      "epoch": 1.2396977680568673,
      "grad_norm": 0.5444367527961731,
      "learning_rate": 6.379449486917421e-05,
      "loss": 0.1976,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12470,
      "tokens_per_second_per_gpu": 398.28
    },
    {
      "epoch": 1.2406919520803301,
      "grad_norm": 0.40487316250801086,
      "learning_rate": 6.364824103958331e-05,
      "loss": 0.1205,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 12480,
      "tokens_per_second_per_gpu": 326.23
    },
    {
      "epoch": 1.2416861361037927,
      "grad_norm": 0.55726557970047,
      "learning_rate": 6.350207675205781e-05,
      "loss": 0.1472,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12490,
      "tokens_per_second_per_gpu": 319.36
    },
    {
      "epoch": 1.2426803201272556,
      "grad_norm": 0.5732585787773132,
      "learning_rate": 6.335600236663131e-05,
      "loss": 0.173,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12500,
      "tokens_per_second_per_gpu": 323.39
    },
    {
      "epoch": 1.2436745041507182,
      "grad_norm": 0.47033485770225525,
      "learning_rate": 6.321001824311583e-05,
      "loss": 0.1712,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12510,
      "tokens_per_second_per_gpu": 300.06
    },
    {
      "epoch": 1.244668688174181,
      "grad_norm": 0.44385817646980286,
      "learning_rate": 6.306412474110122e-05,
      "loss": 0.143,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 12520,
      "tokens_per_second_per_gpu": 312.49
    },
    {
      "epoch": 1.2456628721976437,
      "grad_norm": 0.5871978402137756,
      "learning_rate": 6.291832221995388e-05,
      "loss": 0.1772,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 12530,
      "tokens_per_second_per_gpu": 289.93
    },
    {
      "epoch": 1.2466570562211066,
      "grad_norm": 0.6150623559951782,
      "learning_rate": 6.277261103881638e-05,
      "loss": 0.134,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 12540,
      "tokens_per_second_per_gpu": 312.62
    },
    {
      "epoch": 1.2476512402445692,
      "grad_norm": 0.6367026567459106,
      "learning_rate": 6.262699155660601e-05,
      "loss": 0.2192,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12550,
      "tokens_per_second_per_gpu": 334.54
    },
    {
      "epoch": 1.248645424268032,
      "grad_norm": 0.29298296570777893,
      "learning_rate": 6.248146413201444e-05,
      "loss": 0.1669,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 12560,
      "tokens_per_second_per_gpu": 313.23
    },
    {
      "epoch": 1.2496396082914947,
      "grad_norm": 0.27840226888656616,
      "learning_rate": 6.233602912350639e-05,
      "loss": 0.1134,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12570,
      "tokens_per_second_per_gpu": 333.22
    },
    {
      "epoch": 1.2506337923149575,
      "grad_norm": 0.49214980006217957,
      "learning_rate": 6.219068688931908e-05,
      "loss": 0.1726,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 12580,
      "tokens_per_second_per_gpu": 394.03
    },
    {
      "epoch": 1.2516279763384204,
      "grad_norm": 0.5485401153564453,
      "learning_rate": 6.20454377874612e-05,
      "loss": 0.1014,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 12590,
      "tokens_per_second_per_gpu": 262.31
    },
    {
      "epoch": 1.252622160361883,
      "grad_norm": 0.5006467700004578,
      "learning_rate": 6.190028217571186e-05,
      "loss": 0.1883,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 12600,
      "tokens_per_second_per_gpu": 350.5
    },
    {
      "epoch": 1.2536163443853456,
      "grad_norm": 0.4242802560329437,
      "learning_rate": 6.175522041162016e-05,
      "loss": 0.1559,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 12610,
      "tokens_per_second_per_gpu": 319.2
    },
    {
      "epoch": 1.2546105284088085,
      "grad_norm": 0.532721996307373,
      "learning_rate": 6.161025285250373e-05,
      "loss": 0.1551,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 12620,
      "tokens_per_second_per_gpu": 342.02
    },
    {
      "epoch": 1.2556047124322713,
      "grad_norm": 0.6736690998077393,
      "learning_rate": 6.146537985544843e-05,
      "loss": 0.1887,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 12630,
      "tokens_per_second_per_gpu": 312.69
    },
    {
      "epoch": 1.256598896455734,
      "grad_norm": 0.4985780417919159,
      "learning_rate": 6.132060177730698e-05,
      "loss": 0.1409,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12640,
      "tokens_per_second_per_gpu": 330.02
    },
    {
      "epoch": 1.2575930804791966,
      "grad_norm": 49.2108268737793,
      "learning_rate": 6.117591897469847e-05,
      "loss": 0.1823,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 12650,
      "tokens_per_second_per_gpu": 319.61
    },
    {
      "epoch": 1.2585872645026595,
      "grad_norm": 0.4028538465499878,
      "learning_rate": 6.1031331804007154e-05,
      "loss": 0.2102,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 12660,
      "tokens_per_second_per_gpu": 355.74
    },
    {
      "epoch": 1.2595814485261223,
      "grad_norm": 0.47767624258995056,
      "learning_rate": 6.0886840621381856e-05,
      "loss": 0.153,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 12670,
      "tokens_per_second_per_gpu": 381.38
    },
    {
      "epoch": 1.260575632549585,
      "grad_norm": 0.3719147741794586,
      "learning_rate": 6.0742445782734825e-05,
      "loss": 0.15,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12680,
      "tokens_per_second_per_gpu": 351.13
    },
    {
      "epoch": 1.2615698165730476,
      "grad_norm": 0.32970771193504333,
      "learning_rate": 6.0598147643741124e-05,
      "loss": 0.144,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 12690,
      "tokens_per_second_per_gpu": 285.76
    },
    {
      "epoch": 1.2625640005965104,
      "grad_norm": 0.40551966428756714,
      "learning_rate": 6.045394655983753e-05,
      "loss": 0.149,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 12700,
      "tokens_per_second_per_gpu": 275.71
    },
    {
      "epoch": 1.2635581846199733,
      "grad_norm": 0.40349385142326355,
      "learning_rate": 6.0309842886221826e-05,
      "loss": 0.1615,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.16,
      "memory/max_allocated (GiB)": 18.16,
      "step": 12710,
      "tokens_per_second_per_gpu": 374.56
    },
    {
      "epoch": 1.264552368643436,
      "grad_norm": 0.5158276557922363,
      "learning_rate": 6.0165836977851796e-05,
      "loss": 0.1828,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 12720,
      "tokens_per_second_per_gpu": 404.51
    },
    {
      "epoch": 1.2655465526668985,
      "grad_norm": 0.42637962102890015,
      "learning_rate": 6.0021929189444416e-05,
      "loss": 0.1535,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12730,
      "tokens_per_second_per_gpu": 268.86
    },
    {
      "epoch": 1.2665407366903614,
      "grad_norm": 0.3736035227775574,
      "learning_rate": 5.987811987547504e-05,
      "loss": 0.1774,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 12740,
      "tokens_per_second_per_gpu": 410.2
    },
    {
      "epoch": 1.2675349207138242,
      "grad_norm": 0.479159414768219,
      "learning_rate": 5.9734409390176315e-05,
      "loss": 0.1869,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 12750,
      "tokens_per_second_per_gpu": 290.14
    },
    {
      "epoch": 1.2685291047372869,
      "grad_norm": 0.4126437306404114,
      "learning_rate": 5.959079808753765e-05,
      "loss": 0.1731,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12760,
      "tokens_per_second_per_gpu": 335.34
    },
    {
      "epoch": 1.2695232887607495,
      "grad_norm": 0.5344352722167969,
      "learning_rate": 5.944728632130392e-05,
      "loss": 0.1612,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 12770,
      "tokens_per_second_per_gpu": 304.8
    },
    {
      "epoch": 1.2705174727842123,
      "grad_norm": 0.4766976833343506,
      "learning_rate": 5.9303874444975005e-05,
      "loss": 0.1296,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 12780,
      "tokens_per_second_per_gpu": 292.47
    },
    {
      "epoch": 1.2715116568076752,
      "grad_norm": 0.42096179723739624,
      "learning_rate": 5.9160562811804644e-05,
      "loss": 0.181,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 12790,
      "tokens_per_second_per_gpu": 368.43
    },
    {
      "epoch": 1.2725058408311378,
      "grad_norm": 0.41255730390548706,
      "learning_rate": 5.901735177479972e-05,
      "loss": 0.1565,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.23,
      "memory/max_allocated (GiB)": 19.23,
      "step": 12800,
      "tokens_per_second_per_gpu": 291.39
    },
    {
      "epoch": 1.2735000248546005,
      "grad_norm": 0.4876402020454407,
      "learning_rate": 5.8874241686719234e-05,
      "loss": 0.1288,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 12810,
      "tokens_per_second_per_gpu": 297.94
    },
    {
      "epoch": 1.2744942088780633,
      "grad_norm": 0.47068750858306885,
      "learning_rate": 5.873123290007363e-05,
      "loss": 0.1144,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 12820,
      "tokens_per_second_per_gpu": 293.08
    },
    {
      "epoch": 1.2754883929015262,
      "grad_norm": 0.5882211327552795,
      "learning_rate": 5.8588325767123694e-05,
      "loss": 0.1776,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 12830,
      "tokens_per_second_per_gpu": 338.75
    },
    {
      "epoch": 1.2764825769249888,
      "grad_norm": 0.43064388632774353,
      "learning_rate": 5.844552063987997e-05,
      "loss": 0.1555,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 12840,
      "tokens_per_second_per_gpu": 342.44
    },
    {
      "epoch": 1.2774767609484516,
      "grad_norm": 0.6673071980476379,
      "learning_rate": 5.830281787010166e-05,
      "loss": 0.1722,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 12850,
      "tokens_per_second_per_gpu": 338.48
    },
    {
      "epoch": 1.2784709449719143,
      "grad_norm": 0.6424853801727295,
      "learning_rate": 5.8160217809295826e-05,
      "loss": 0.1647,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 12860,
      "tokens_per_second_per_gpu": 354.7
    },
    {
      "epoch": 1.2794651289953771,
      "grad_norm": 0.5545419454574585,
      "learning_rate": 5.801772080871659e-05,
      "loss": 0.1304,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 12870,
      "tokens_per_second_per_gpu": 326.79
    },
    {
      "epoch": 1.2804593130188398,
      "grad_norm": 0.5145922899246216,
      "learning_rate": 5.787532721936413e-05,
      "loss": 0.1984,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 12880,
      "tokens_per_second_per_gpu": 360.98
    },
    {
      "epoch": 1.2814534970423026,
      "grad_norm": 0.4928934574127197,
      "learning_rate": 5.7733037391984024e-05,
      "loss": 0.1239,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 12890,
      "tokens_per_second_per_gpu": 291.25
    },
    {
      "epoch": 1.2824476810657652,
      "grad_norm": 0.5820671916007996,
      "learning_rate": 5.759085167706611e-05,
      "loss": 0.1584,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12900,
      "tokens_per_second_per_gpu": 324.13
    },
    {
      "epoch": 1.283441865089228,
      "grad_norm": 0.42122918367385864,
      "learning_rate": 5.7448770424843926e-05,
      "loss": 0.1653,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 12910,
      "tokens_per_second_per_gpu": 316.56
    },
    {
      "epoch": 1.2844360491126907,
      "grad_norm": 0.30348268151283264,
      "learning_rate": 5.730679398529355e-05,
      "loss": 0.1152,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 12920,
      "tokens_per_second_per_gpu": 338.17
    },
    {
      "epoch": 1.2854302331361536,
      "grad_norm": 0.4230218529701233,
      "learning_rate": 5.716492270813305e-05,
      "loss": 0.1603,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 12930,
      "tokens_per_second_per_gpu": 371.96
    },
    {
      "epoch": 1.2864244171596162,
      "grad_norm": 0.4649568200111389,
      "learning_rate": 5.7023156942821274e-05,
      "loss": 0.1618,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12940,
      "tokens_per_second_per_gpu": 365.8
    },
    {
      "epoch": 1.287418601183079,
      "grad_norm": 0.7270090579986572,
      "learning_rate": 5.688149703855732e-05,
      "loss": 0.1442,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 12950,
      "tokens_per_second_per_gpu": 326.17
    },
    {
      "epoch": 1.2884127852065417,
      "grad_norm": 0.4988739490509033,
      "learning_rate": 5.6739943344279455e-05,
      "loss": 0.2068,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 12960,
      "tokens_per_second_per_gpu": 347.14
    },
    {
      "epoch": 1.2894069692300045,
      "grad_norm": 0.5430411100387573,
      "learning_rate": 5.65984962086644e-05,
      "loss": 0.2245,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 12970,
      "tokens_per_second_per_gpu": 428.47
    },
    {
      "epoch": 1.2904011532534672,
      "grad_norm": 0.6165452003479004,
      "learning_rate": 5.645715598012626e-05,
      "loss": 0.1454,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 12980,
      "tokens_per_second_per_gpu": 316.19
    },
    {
      "epoch": 1.29139533727693,
      "grad_norm": 0.525731086730957,
      "learning_rate": 5.631592300681593e-05,
      "loss": 0.186,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 12990,
      "tokens_per_second_per_gpu": 324.42
    },
    {
      "epoch": 1.2923895213003926,
      "grad_norm": 0.843223512172699,
      "learning_rate": 5.617479763662011e-05,
      "loss": 0.1512,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13000,
      "tokens_per_second_per_gpu": 288.5
    },
    {
      "epoch": 1.2933837053238555,
      "grad_norm": 0.557815432548523,
      "learning_rate": 5.6033780217160346e-05,
      "loss": 0.1475,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 13010,
      "tokens_per_second_per_gpu": 306.45
    },
    {
      "epoch": 1.2943778893473181,
      "grad_norm": 0.4504510462284088,
      "learning_rate": 5.589287109579242e-05,
      "loss": 0.1323,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13020,
      "tokens_per_second_per_gpu": 353.31
    },
    {
      "epoch": 1.295372073370781,
      "grad_norm": 0.6401469707489014,
      "learning_rate": 5.575207061960519e-05,
      "loss": 0.1525,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.67,
      "memory/max_allocated (GiB)": 19.67,
      "step": 13030,
      "tokens_per_second_per_gpu": 341.06
    },
    {
      "epoch": 1.2963662573942436,
      "grad_norm": 0.603769838809967,
      "learning_rate": 5.561137913542008e-05,
      "loss": 0.2351,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 13040,
      "tokens_per_second_per_gpu": 391.65
    },
    {
      "epoch": 1.2973604414177065,
      "grad_norm": 0.2735394239425659,
      "learning_rate": 5.5470796989789874e-05,
      "loss": 0.1351,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 13050,
      "tokens_per_second_per_gpu": 321.21
    },
    {
      "epoch": 1.298354625441169,
      "grad_norm": 0.46087950468063354,
      "learning_rate": 5.533032452899818e-05,
      "loss": 0.204,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 13060,
      "tokens_per_second_per_gpu": 353.8
    },
    {
      "epoch": 1.299348809464632,
      "grad_norm": 0.48193469643592834,
      "learning_rate": 5.518996209905829e-05,
      "loss": 0.1539,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 13070,
      "tokens_per_second_per_gpu": 293.56
    },
    {
      "epoch": 1.3003429934880946,
      "grad_norm": 0.4664101302623749,
      "learning_rate": 5.5049710045712596e-05,
      "loss": 0.1775,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 13080,
      "tokens_per_second_per_gpu": 283.89
    },
    {
      "epoch": 1.3013371775115574,
      "grad_norm": 0.4421910047531128,
      "learning_rate": 5.490956871443149e-05,
      "loss": 0.1428,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 13090,
      "tokens_per_second_per_gpu": 332.54
    },
    {
      "epoch": 1.3023313615350203,
      "grad_norm": 0.5099435448646545,
      "learning_rate": 5.4769538450412706e-05,
      "loss": 0.2018,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 13100,
      "tokens_per_second_per_gpu": 366.63
    },
    {
      "epoch": 1.303325545558483,
      "grad_norm": 0.4641667604446411,
      "learning_rate": 5.462961959858042e-05,
      "loss": 0.1445,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13110,
      "tokens_per_second_per_gpu": 363.8
    },
    {
      "epoch": 1.3043197295819455,
      "grad_norm": 0.3798070251941681,
      "learning_rate": 5.448981250358429e-05,
      "loss": 0.1464,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13120,
      "tokens_per_second_per_gpu": 301.06
    },
    {
      "epoch": 1.3053139136054084,
      "grad_norm": 0.49534204602241516,
      "learning_rate": 5.435011750979881e-05,
      "loss": 0.2212,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13130,
      "tokens_per_second_per_gpu": 337.64
    },
    {
      "epoch": 1.3063080976288712,
      "grad_norm": 0.586558997631073,
      "learning_rate": 5.421053496132218e-05,
      "loss": 0.1626,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13140,
      "tokens_per_second_per_gpu": 334.07
    },
    {
      "epoch": 1.3073022816523339,
      "grad_norm": 0.2836264967918396,
      "learning_rate": 5.40710652019758e-05,
      "loss": 0.1488,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 13150,
      "tokens_per_second_per_gpu": 335.48
    },
    {
      "epoch": 1.3082964656757965,
      "grad_norm": 0.6480442881584167,
      "learning_rate": 5.3931708575303096e-05,
      "loss": 0.1667,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13160,
      "tokens_per_second_per_gpu": 322.57
    },
    {
      "epoch": 1.3092906496992593,
      "grad_norm": 0.5337100625038147,
      "learning_rate": 5.379246542456897e-05,
      "loss": 0.1342,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 13170,
      "tokens_per_second_per_gpu": 349.13
    },
    {
      "epoch": 1.3102848337227222,
      "grad_norm": 0.4983210265636444,
      "learning_rate": 5.365333609275864e-05,
      "loss": 0.1469,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 13180,
      "tokens_per_second_per_gpu": 329.8
    },
    {
      "epoch": 1.3112790177461848,
      "grad_norm": 0.4759822487831116,
      "learning_rate": 5.351432092257716e-05,
      "loss": 0.1507,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 13190,
      "tokens_per_second_per_gpu": 282.24
    },
    {
      "epoch": 1.3122732017696475,
      "grad_norm": 0.2128848135471344,
      "learning_rate": 5.3375420256448175e-05,
      "loss": 0.1901,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13200,
      "tokens_per_second_per_gpu": 351.2
    },
    {
      "epoch": 1.3132673857931103,
      "grad_norm": 0.23509669303894043,
      "learning_rate": 5.323663443651345e-05,
      "loss": 0.1259,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 13210,
      "tokens_per_second_per_gpu": 364.3
    },
    {
      "epoch": 1.3142615698165732,
      "grad_norm": 0.4961312711238861,
      "learning_rate": 5.309796380463174e-05,
      "loss": 0.1589,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13220,
      "tokens_per_second_per_gpu": 288.96
    },
    {
      "epoch": 1.3152557538400358,
      "grad_norm": 0.6080997586250305,
      "learning_rate": 5.295940870237817e-05,
      "loss": 0.2004,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13230,
      "tokens_per_second_per_gpu": 350.85
    },
    {
      "epoch": 1.3162499378634984,
      "grad_norm": 0.4995987117290497,
      "learning_rate": 5.2820969471043204e-05,
      "loss": 0.1845,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 13240,
      "tokens_per_second_per_gpu": 406.17
    },
    {
      "epoch": 1.3172441218869613,
      "grad_norm": 0.4047294557094574,
      "learning_rate": 5.2682646451631945e-05,
      "loss": 0.1338,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13250,
      "tokens_per_second_per_gpu": 330.72
    },
    {
      "epoch": 1.3182383059104241,
      "grad_norm": 0.6713951826095581,
      "learning_rate": 5.254443998486327e-05,
      "loss": 0.2004,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 13260,
      "tokens_per_second_per_gpu": 366.71
    },
    {
      "epoch": 1.3192324899338868,
      "grad_norm": 0.3636043667793274,
      "learning_rate": 5.240635041116884e-05,
      "loss": 0.1612,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13270,
      "tokens_per_second_per_gpu": 379.85
    },
    {
      "epoch": 1.3202266739573494,
      "grad_norm": 0.6122763156890869,
      "learning_rate": 5.226837807069251e-05,
      "loss": 0.1599,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 13280,
      "tokens_per_second_per_gpu": 301.88
    },
    {
      "epoch": 1.3212208579808122,
      "grad_norm": 0.3268527686595917,
      "learning_rate": 5.213052330328929e-05,
      "loss": 0.152,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 13290,
      "tokens_per_second_per_gpu": 316.71
    },
    {
      "epoch": 1.322215042004275,
      "grad_norm": 0.27024024724960327,
      "learning_rate": 5.199278644852464e-05,
      "loss": 0.2408,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 13300,
      "tokens_per_second_per_gpu": 338.21
    },
    {
      "epoch": 1.3232092260277377,
      "grad_norm": 0.34981921315193176,
      "learning_rate": 5.18551678456735e-05,
      "loss": 0.1638,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 13310,
      "tokens_per_second_per_gpu": 320.81
    },
    {
      "epoch": 1.3242034100512003,
      "grad_norm": 0.44818684458732605,
      "learning_rate": 5.1717667833719627e-05,
      "loss": 0.1738,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13320,
      "tokens_per_second_per_gpu": 352.42
    },
    {
      "epoch": 1.3251975940746632,
      "grad_norm": 0.37409475445747375,
      "learning_rate": 5.1580286751354545e-05,
      "loss": 0.1738,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 13330,
      "tokens_per_second_per_gpu": 319.43
    },
    {
      "epoch": 1.326191778098126,
      "grad_norm": 0.48389047384262085,
      "learning_rate": 5.144302493697697e-05,
      "loss": 0.1665,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 13340,
      "tokens_per_second_per_gpu": 402.78
    },
    {
      "epoch": 1.3271859621215887,
      "grad_norm": 0.3536114990711212,
      "learning_rate": 5.13058827286917e-05,
      "loss": 0.131,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 13350,
      "tokens_per_second_per_gpu": 343.79
    },
    {
      "epoch": 1.3281801461450515,
      "grad_norm": 0.40804916620254517,
      "learning_rate": 5.116886046430903e-05,
      "loss": 0.1576,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13360,
      "tokens_per_second_per_gpu": 297.51
    },
    {
      "epoch": 1.3291743301685142,
      "grad_norm": 0.5419421792030334,
      "learning_rate": 5.10319584813437e-05,
      "loss": 0.1632,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 13370,
      "tokens_per_second_per_gpu": 309.55
    },
    {
      "epoch": 1.330168514191977,
      "grad_norm": 0.3658435642719269,
      "learning_rate": 5.089517711701426e-05,
      "loss": 0.1439,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 13380,
      "tokens_per_second_per_gpu": 300.73
    },
    {
      "epoch": 1.3311626982154396,
      "grad_norm": 0.3239598572254181,
      "learning_rate": 5.075851670824212e-05,
      "loss": 0.1575,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 13390,
      "tokens_per_second_per_gpu": 334.4
    },
    {
      "epoch": 1.3321568822389025,
      "grad_norm": 0.5001199245452881,
      "learning_rate": 5.0621977591650773e-05,
      "loss": 0.1593,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.43,
      "memory/max_allocated (GiB)": 20.43,
      "step": 13400,
      "tokens_per_second_per_gpu": 355.87
    },
    {
      "epoch": 1.3331510662623651,
      "grad_norm": 0.5077064633369446,
      "learning_rate": 5.048556010356491e-05,
      "loss": 0.1502,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13410,
      "tokens_per_second_per_gpu": 320.65
    },
    {
      "epoch": 1.334145250285828,
      "grad_norm": 0.573063313961029,
      "learning_rate": 5.0349264580009616e-05,
      "loss": 0.16,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 13420,
      "tokens_per_second_per_gpu": 365.59
    },
    {
      "epoch": 1.3351394343092906,
      "grad_norm": 0.6947848200798035,
      "learning_rate": 5.021309135670959e-05,
      "loss": 0.1975,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.69,
      "memory/max_allocated (GiB)": 19.69,
      "step": 13430,
      "tokens_per_second_per_gpu": 366.35
    },
    {
      "epoch": 1.3361336183327535,
      "grad_norm": 0.6331676244735718,
      "learning_rate": 5.007704076908825e-05,
      "loss": 0.1626,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 13440,
      "tokens_per_second_per_gpu": 377.99
    },
    {
      "epoch": 1.337127802356216,
      "grad_norm": 0.6537986397743225,
      "learning_rate": 4.994111315226697e-05,
      "loss": 0.1846,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13450,
      "tokens_per_second_per_gpu": 409.47
    },
    {
      "epoch": 1.338121986379679,
      "grad_norm": 0.27798891067504883,
      "learning_rate": 4.980530884106416e-05,
      "loss": 0.1483,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13460,
      "tokens_per_second_per_gpu": 297.57
    },
    {
      "epoch": 1.3391161704031416,
      "grad_norm": 0.455538272857666,
      "learning_rate": 4.9669628169994586e-05,
      "loss": 0.1544,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 13470,
      "tokens_per_second_per_gpu": 331.23
    },
    {
      "epoch": 1.3401103544266044,
      "grad_norm": 0.8112931251525879,
      "learning_rate": 4.9534071473268375e-05,
      "loss": 0.1629,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.43,
      "memory/max_allocated (GiB)": 20.43,
      "step": 13480,
      "tokens_per_second_per_gpu": 326.59
    },
    {
      "epoch": 1.341104538450067,
      "grad_norm": 0.541522204875946,
      "learning_rate": 4.939863908479037e-05,
      "loss": 0.1238,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13490,
      "tokens_per_second_per_gpu": 304.61
    },
    {
      "epoch": 1.34209872247353,
      "grad_norm": 0.574373185634613,
      "learning_rate": 4.9263331338159105e-05,
      "loss": 0.1664,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 13500,
      "tokens_per_second_per_gpu": 341.93
    },
    {
      "epoch": 1.3430929064969925,
      "grad_norm": 0.35874566435813904,
      "learning_rate": 4.9128148566666186e-05,
      "loss": 0.1402,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 13510,
      "tokens_per_second_per_gpu": 328.7
    },
    {
      "epoch": 1.3440870905204554,
      "grad_norm": 0.2937113046646118,
      "learning_rate": 4.899309110329541e-05,
      "loss": 0.1482,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 13520,
      "tokens_per_second_per_gpu": 328.58
    },
    {
      "epoch": 1.345081274543918,
      "grad_norm": 0.6201068162918091,
      "learning_rate": 4.885815928072176e-05,
      "loss": 0.14,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 13530,
      "tokens_per_second_per_gpu": 336.28
    },
    {
      "epoch": 1.3460754585673809,
      "grad_norm": 0.5328193306922913,
      "learning_rate": 4.872335343131088e-05,
      "loss": 0.1536,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 13540,
      "tokens_per_second_per_gpu": 337.94
    },
    {
      "epoch": 1.3470696425908435,
      "grad_norm": 0.6596990823745728,
      "learning_rate": 4.8588673887118054e-05,
      "loss": 0.1956,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 13550,
      "tokens_per_second_per_gpu": 333.18
    },
    {
      "epoch": 1.3480638266143063,
      "grad_norm": 0.7473700642585754,
      "learning_rate": 4.845412097988752e-05,
      "loss": 0.1762,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13560,
      "tokens_per_second_per_gpu": 345.86
    },
    {
      "epoch": 1.349058010637769,
      "grad_norm": 0.544845700263977,
      "learning_rate": 4.831969504105145e-05,
      "loss": 0.169,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 13570,
      "tokens_per_second_per_gpu": 366.86
    },
    {
      "epoch": 1.3500521946612318,
      "grad_norm": 0.48342615365982056,
      "learning_rate": 4.818539640172941e-05,
      "loss": 0.1178,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 13580,
      "tokens_per_second_per_gpu": 348.38
    },
    {
      "epoch": 1.3510463786846945,
      "grad_norm": 0.5576662421226501,
      "learning_rate": 4.805122539272725e-05,
      "loss": 0.1605,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 13590,
      "tokens_per_second_per_gpu": 341.46
    },
    {
      "epoch": 1.3520405627081573,
      "grad_norm": 0.6246042847633362,
      "learning_rate": 4.791718234453663e-05,
      "loss": 0.2546,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 13600,
      "tokens_per_second_per_gpu": 406.86
    },
    {
      "epoch": 1.35303474673162,
      "grad_norm": 0.31434324383735657,
      "learning_rate": 4.7783267587333794e-05,
      "loss": 0.1727,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13610,
      "tokens_per_second_per_gpu": 382.72
    },
    {
      "epoch": 1.3540289307550828,
      "grad_norm": 0.38281211256980896,
      "learning_rate": 4.764948145097919e-05,
      "loss": 0.1753,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 13620,
      "tokens_per_second_per_gpu": 302.55
    },
    {
      "epoch": 1.3550231147785454,
      "grad_norm": 0.3556180000305176,
      "learning_rate": 4.7515824265016276e-05,
      "loss": 0.1443,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.66,
      "memory/max_allocated (GiB)": 20.66,
      "step": 13630,
      "tokens_per_second_per_gpu": 341.41
    },
    {
      "epoch": 1.3560172988020083,
      "grad_norm": 0.29845160245895386,
      "learning_rate": 4.7382296358670976e-05,
      "loss": 0.1323,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 13640,
      "tokens_per_second_per_gpu": 281.42
    },
    {
      "epoch": 1.3570114828254711,
      "grad_norm": 0.4179192781448364,
      "learning_rate": 4.724889806085079e-05,
      "loss": 0.1176,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.68,
      "memory/max_allocated (GiB)": 19.68,
      "step": 13650,
      "tokens_per_second_per_gpu": 337.01
    },
    {
      "epoch": 1.3580056668489338,
      "grad_norm": 0.46176135540008545,
      "learning_rate": 4.711562970014384e-05,
      "loss": 0.1159,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13660,
      "tokens_per_second_per_gpu": 366.9
    },
    {
      "epoch": 1.3589998508723964,
      "grad_norm": 0.34104305505752563,
      "learning_rate": 4.6982491604818314e-05,
      "loss": 0.1194,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 13670,
      "tokens_per_second_per_gpu": 306.48
    },
    {
      "epoch": 1.3599940348958592,
      "grad_norm": 0.8108484148979187,
      "learning_rate": 4.684948410282146e-05,
      "loss": 0.2221,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 13680,
      "tokens_per_second_per_gpu": 379.12
    },
    {
      "epoch": 1.360988218919322,
      "grad_norm": 0.5723569393157959,
      "learning_rate": 4.671660752177892e-05,
      "loss": 0.1656,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 13690,
      "tokens_per_second_per_gpu": 380.88
    },
    {
      "epoch": 1.3619824029427847,
      "grad_norm": 0.5756139755249023,
      "learning_rate": 4.658386218899371e-05,
      "loss": 0.145,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13700,
      "tokens_per_second_per_gpu": 406.13
    },
    {
      "epoch": 1.3629765869662473,
      "grad_norm": 0.3543594777584076,
      "learning_rate": 4.645124843144574e-05,
      "loss": 0.1011,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 13710,
      "tokens_per_second_per_gpu": 308.42
    },
    {
      "epoch": 1.3639707709897102,
      "grad_norm": 0.210982084274292,
      "learning_rate": 4.631876657579062e-05,
      "loss": 0.1649,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 13720,
      "tokens_per_second_per_gpu": 339.26
    },
    {
      "epoch": 1.364964955013173,
      "grad_norm": 0.4334322214126587,
      "learning_rate": 4.6186416948359256e-05,
      "loss": 0.1152,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13730,
      "tokens_per_second_per_gpu": 384.14
    },
    {
      "epoch": 1.3659591390366357,
      "grad_norm": 0.7411201000213623,
      "learning_rate": 4.6054199875156665e-05,
      "loss": 0.1034,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 13740,
      "tokens_per_second_per_gpu": 318.13
    },
    {
      "epoch": 1.3669533230600983,
      "grad_norm": 0.6421767473220825,
      "learning_rate": 4.5922115681861536e-05,
      "loss": 0.1494,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 13750,
      "tokens_per_second_per_gpu": 343.47
    },
    {
      "epoch": 1.3679475070835612,
      "grad_norm": 0.6301234364509583,
      "learning_rate": 4.579016469382505e-05,
      "loss": 0.1698,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 13760,
      "tokens_per_second_per_gpu": 351.88
    },
    {
      "epoch": 1.368941691107024,
      "grad_norm": 0.5952703952789307,
      "learning_rate": 4.5658347236070445e-05,
      "loss": 0.1712,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 13770,
      "tokens_per_second_per_gpu": 380.8
    },
    {
      "epoch": 1.3699358751304866,
      "grad_norm": 0.37993893027305603,
      "learning_rate": 4.5526663633292e-05,
      "loss": 0.1618,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 13780,
      "tokens_per_second_per_gpu": 278.0
    },
    {
      "epoch": 1.3709300591539493,
      "grad_norm": 0.5828644633293152,
      "learning_rate": 4.5395114209854195e-05,
      "loss": 0.1806,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 13790,
      "tokens_per_second_per_gpu": 371.91
    },
    {
      "epoch": 1.3719242431774121,
      "grad_norm": 0.6922226548194885,
      "learning_rate": 4.526369928979113e-05,
      "loss": 0.1802,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13800,
      "tokens_per_second_per_gpu": 392.71
    },
    {
      "epoch": 1.372918427200875,
      "grad_norm": 0.5355440378189087,
      "learning_rate": 4.513241919680546e-05,
      "loss": 0.1622,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 13810,
      "tokens_per_second_per_gpu": 361.78
    },
    {
      "epoch": 1.3739126112243376,
      "grad_norm": 0.3850402534008026,
      "learning_rate": 4.500127425426783e-05,
      "loss": 0.1239,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 13820,
      "tokens_per_second_per_gpu": 314.59
    },
    {
      "epoch": 1.3749067952478002,
      "grad_norm": 0.5045945644378662,
      "learning_rate": 4.4870264785215966e-05,
      "loss": 0.1422,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 13830,
      "tokens_per_second_per_gpu": 329.11
    },
    {
      "epoch": 1.375900979271263,
      "grad_norm": 0.552185595035553,
      "learning_rate": 4.4739391112353915e-05,
      "loss": 0.1577,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 13840,
      "tokens_per_second_per_gpu": 368.68
    },
    {
      "epoch": 1.376895163294726,
      "grad_norm": 0.5017216205596924,
      "learning_rate": 4.460865355805109e-05,
      "loss": 0.1855,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 13850,
      "tokens_per_second_per_gpu": 331.39
    },
    {
      "epoch": 1.3778893473181886,
      "grad_norm": 0.33180108666419983,
      "learning_rate": 4.447805244434184e-05,
      "loss": 0.1298,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.41,
      "memory/max_allocated (GiB)": 21.41,
      "step": 13860,
      "tokens_per_second_per_gpu": 410.56
    },
    {
      "epoch": 1.3788835313416512,
      "grad_norm": 0.5970960855484009,
      "learning_rate": 4.4347588092924206e-05,
      "loss": 0.1828,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 13870,
      "tokens_per_second_per_gpu": 367.8
    },
    {
      "epoch": 1.379877715365114,
      "grad_norm": 0.588447630405426,
      "learning_rate": 4.421726082515953e-05,
      "loss": 0.1518,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 13880,
      "tokens_per_second_per_gpu": 333.81
    },
    {
      "epoch": 1.380871899388577,
      "grad_norm": 0.5242084860801697,
      "learning_rate": 4.4087070962071377e-05,
      "loss": 0.2154,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 13890,
      "tokens_per_second_per_gpu": 424.02
    },
    {
      "epoch": 1.3818660834120395,
      "grad_norm": 0.4128514528274536,
      "learning_rate": 4.395701882434493e-05,
      "loss": 0.1594,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 13900,
      "tokens_per_second_per_gpu": 361.98
    },
    {
      "epoch": 1.3828602674355024,
      "grad_norm": 0.3702273964881897,
      "learning_rate": 4.3827104732326055e-05,
      "loss": 0.1143,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 13910,
      "tokens_per_second_per_gpu": 339.1
    },
    {
      "epoch": 1.383854451458965,
      "grad_norm": 0.5153867602348328,
      "learning_rate": 4.3697329006020614e-05,
      "loss": 0.1272,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13920,
      "tokens_per_second_per_gpu": 296.93
    },
    {
      "epoch": 1.3848486354824279,
      "grad_norm": 0.45272836089134216,
      "learning_rate": 4.356769196509373e-05,
      "loss": 0.1551,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 13930,
      "tokens_per_second_per_gpu": 372.63
    },
    {
      "epoch": 1.3858428195058905,
      "grad_norm": 0.44970136880874634,
      "learning_rate": 4.343819392886873e-05,
      "loss": 0.1434,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 13940,
      "tokens_per_second_per_gpu": 340.91
    },
    {
      "epoch": 1.3868370035293534,
      "grad_norm": 0.5090652108192444,
      "learning_rate": 4.3308835216326696e-05,
      "loss": 0.1548,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 13950,
      "tokens_per_second_per_gpu": 350.01
    },
    {
      "epoch": 1.387831187552816,
      "grad_norm": 0.4184592366218567,
      "learning_rate": 4.3179616146105465e-05,
      "loss": 0.121,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 13960,
      "tokens_per_second_per_gpu": 348.76
    },
    {
      "epoch": 1.3888253715762788,
      "grad_norm": 0.6222097277641296,
      "learning_rate": 4.305053703649897e-05,
      "loss": 0.1996,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 13970,
      "tokens_per_second_per_gpu": 380.55
    },
    {
      "epoch": 1.3898195555997415,
      "grad_norm": 0.34182408452033997,
      "learning_rate": 4.292159820545627e-05,
      "loss": 0.1481,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 13980,
      "tokens_per_second_per_gpu": 368.88
    },
    {
      "epoch": 1.3908137396232043,
      "grad_norm": 0.551923930644989,
      "learning_rate": 4.279279997058101e-05,
      "loss": 0.1954,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 13990,
      "tokens_per_second_per_gpu": 327.45
    },
    {
      "epoch": 1.391807923646667,
      "grad_norm": 0.43246060609817505,
      "learning_rate": 4.266414264913041e-05,
      "loss": 0.1153,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 14000,
      "tokens_per_second_per_gpu": 301.74
    },
    {
      "epoch": 1.3928021076701298,
      "grad_norm": 0.3509690761566162,
      "learning_rate": 4.2535626558014705e-05,
      "loss": 0.1414,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 14010,
      "tokens_per_second_per_gpu": 367.19
    },
    {
      "epoch": 1.3937962916935924,
      "grad_norm": 0.5196961164474487,
      "learning_rate": 4.240725201379614e-05,
      "loss": 0.1538,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 14020,
      "tokens_per_second_per_gpu": 339.78
    },
    {
      "epoch": 1.3947904757170553,
      "grad_norm": 0.46282872557640076,
      "learning_rate": 4.22790193326884e-05,
      "loss": 0.1549,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 14030,
      "tokens_per_second_per_gpu": 375.57
    },
    {
      "epoch": 1.395784659740518,
      "grad_norm": 0.3771131932735443,
      "learning_rate": 4.21509288305556e-05,
      "loss": 0.1359,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 14040,
      "tokens_per_second_per_gpu": 377.38
    },
    {
      "epoch": 1.3967788437639808,
      "grad_norm": 0.639408528804779,
      "learning_rate": 4.2022980822911786e-05,
      "loss": 0.1673,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 14050,
      "tokens_per_second_per_gpu": 333.3
    },
    {
      "epoch": 1.3977730277874434,
      "grad_norm": 0.44624921679496765,
      "learning_rate": 4.189517562491996e-05,
      "loss": 0.1855,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14060,
      "tokens_per_second_per_gpu": 378.56
    },
    {
      "epoch": 1.3987672118109062,
      "grad_norm": 0.5301280617713928,
      "learning_rate": 4.176751355139126e-05,
      "loss": 0.1335,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 14070,
      "tokens_per_second_per_gpu": 256.61
    },
    {
      "epoch": 1.3997613958343689,
      "grad_norm": 0.6008658409118652,
      "learning_rate": 4.163999491678444e-05,
      "loss": 0.1674,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 14080,
      "tokens_per_second_per_gpu": 375.31
    },
    {
      "epoch": 1.4007555798578317,
      "grad_norm": 0.569706916809082,
      "learning_rate": 4.1512620035204784e-05,
      "loss": 0.1309,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 14090,
      "tokens_per_second_per_gpu": 311.74
    },
    {
      "epoch": 1.4017497638812944,
      "grad_norm": 0.5985649228096008,
      "learning_rate": 4.138538922040356e-05,
      "loss": 0.2239,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 14100,
      "tokens_per_second_per_gpu": 431.67
    },
    {
      "epoch": 1.4027439479047572,
      "grad_norm": 0.18843001127243042,
      "learning_rate": 4.125830278577717e-05,
      "loss": 0.1663,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 14110,
      "tokens_per_second_per_gpu": 388.4
    },
    {
      "epoch": 1.4037381319282198,
      "grad_norm": 0.4820663332939148,
      "learning_rate": 4.113136104436639e-05,
      "loss": 0.1385,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 14120,
      "tokens_per_second_per_gpu": 375.91
    },
    {
      "epoch": 1.4047323159516827,
      "grad_norm": 0.47296223044395447,
      "learning_rate": 4.10045643088555e-05,
      "loss": 0.188,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14130,
      "tokens_per_second_per_gpu": 364.31
    },
    {
      "epoch": 1.4057264999751453,
      "grad_norm": 0.5604919791221619,
      "learning_rate": 4.0877912891571725e-05,
      "loss": 0.1479,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 14140,
      "tokens_per_second_per_gpu": 354.18
    },
    {
      "epoch": 1.4067206839986082,
      "grad_norm": 0.529105544090271,
      "learning_rate": 4.075140710448419e-05,
      "loss": 0.1566,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 14150,
      "tokens_per_second_per_gpu": 364.28
    },
    {
      "epoch": 1.407714868022071,
      "grad_norm": 0.36099550127983093,
      "learning_rate": 4.062504725920347e-05,
      "loss": 0.1516,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14160,
      "tokens_per_second_per_gpu": 393.89
    },
    {
      "epoch": 1.4087090520455336,
      "grad_norm": 0.6695134043693542,
      "learning_rate": 4.0498833666980505e-05,
      "loss": 0.1371,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 14170,
      "tokens_per_second_per_gpu": 394.12
    },
    {
      "epoch": 1.4097032360689963,
      "grad_norm": 0.4011685252189636,
      "learning_rate": 4.037276663870607e-05,
      "loss": 0.1888,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14180,
      "tokens_per_second_per_gpu": 401.14
    },
    {
      "epoch": 1.4106974200924591,
      "grad_norm": 0.5134690403938293,
      "learning_rate": 4.024684648490995e-05,
      "loss": 0.1467,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 14190,
      "tokens_per_second_per_gpu": 320.81
    },
    {
      "epoch": 1.411691604115922,
      "grad_norm": 0.4153744876384735,
      "learning_rate": 4.012107351576001e-05,
      "loss": 0.1507,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 14200,
      "tokens_per_second_per_gpu": 358.65
    },
    {
      "epoch": 1.4126857881393846,
      "grad_norm": 0.5884429216384888,
      "learning_rate": 3.999544804106174e-05,
      "loss": 0.1785,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 14210,
      "tokens_per_second_per_gpu": 389.13
    },
    {
      "epoch": 1.4136799721628472,
      "grad_norm": 0.3641144931316376,
      "learning_rate": 3.986997037025716e-05,
      "loss": 0.1133,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 14220,
      "tokens_per_second_per_gpu": 354.06
    },
    {
      "epoch": 1.41467415618631,
      "grad_norm": 0.5971251726150513,
      "learning_rate": 3.974464081242437e-05,
      "loss": 0.18,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 14230,
      "tokens_per_second_per_gpu": 388.25
    },
    {
      "epoch": 1.415668340209773,
      "grad_norm": 0.4704718291759491,
      "learning_rate": 3.961945967627648e-05,
      "loss": 0.1546,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.85,
      "memory/max_allocated (GiB)": 18.85,
      "step": 14240,
      "tokens_per_second_per_gpu": 358.95
    },
    {
      "epoch": 1.4166625242332356,
      "grad_norm": 0.44908052682876587,
      "learning_rate": 3.9494427270161124e-05,
      "loss": 0.1928,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 14250,
      "tokens_per_second_per_gpu": 359.69
    },
    {
      "epoch": 1.4176567082566982,
      "grad_norm": 0.4618172347545624,
      "learning_rate": 3.936954390205955e-05,
      "loss": 0.1701,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 14260,
      "tokens_per_second_per_gpu": 363.08
    },
    {
      "epoch": 1.418650892280161,
      "grad_norm": 0.4540565609931946,
      "learning_rate": 3.924480987958592e-05,
      "loss": 0.1195,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14270,
      "tokens_per_second_per_gpu": 382.37
    },
    {
      "epoch": 1.419645076303624,
      "grad_norm": 0.45630887150764465,
      "learning_rate": 3.912022550998642e-05,
      "loss": 0.1577,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 14280,
      "tokens_per_second_per_gpu": 373.77
    },
    {
      "epoch": 1.4206392603270865,
      "grad_norm": 0.5009377002716064,
      "learning_rate": 3.8995791100138755e-05,
      "loss": 0.2043,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14290,
      "tokens_per_second_per_gpu": 436.57
    },
    {
      "epoch": 1.4216334443505492,
      "grad_norm": 0.5434289574623108,
      "learning_rate": 3.887150695655112e-05,
      "loss": 0.1279,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14300,
      "tokens_per_second_per_gpu": 368.81
    },
    {
      "epoch": 1.422627628374012,
      "grad_norm": 0.4998416602611542,
      "learning_rate": 3.874737338536164e-05,
      "loss": 0.1604,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 14310,
      "tokens_per_second_per_gpu": 332.31
    },
    {
      "epoch": 1.4236218123974749,
      "grad_norm": 0.7324095964431763,
      "learning_rate": 3.862339069233759e-05,
      "loss": 0.1572,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 14320,
      "tokens_per_second_per_gpu": 381.52
    },
    {
      "epoch": 1.4246159964209375,
      "grad_norm": 0.5743668675422668,
      "learning_rate": 3.8499559182874475e-05,
      "loss": 0.1947,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 14330,
      "tokens_per_second_per_gpu": 399.2
    },
    {
      "epoch": 1.4256101804444001,
      "grad_norm": 0.3965088427066803,
      "learning_rate": 3.837587916199554e-05,
      "loss": 0.0943,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 14340,
      "tokens_per_second_per_gpu": 308.93
    },
    {
      "epoch": 1.426604364467863,
      "grad_norm": 0.4278639256954193,
      "learning_rate": 3.825235093435076e-05,
      "loss": 0.1645,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14350,
      "tokens_per_second_per_gpu": 375.76
    },
    {
      "epoch": 1.4275985484913258,
      "grad_norm": 0.3658309876918793,
      "learning_rate": 3.812897480421631e-05,
      "loss": 0.1323,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 14360,
      "tokens_per_second_per_gpu": 370.0
    },
    {
      "epoch": 1.4285927325147885,
      "grad_norm": 0.7314541339874268,
      "learning_rate": 3.800575107549362e-05,
      "loss": 0.1764,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14370,
      "tokens_per_second_per_gpu": 318.75
    },
    {
      "epoch": 1.429586916538251,
      "grad_norm": 0.3194984197616577,
      "learning_rate": 3.788268005170883e-05,
      "loss": 0.1383,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 14380,
      "tokens_per_second_per_gpu": 367.0
    },
    {
      "epoch": 1.430581100561714,
      "grad_norm": 0.24363841116428375,
      "learning_rate": 3.7759762036011856e-05,
      "loss": 0.171,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 14390,
      "tokens_per_second_per_gpu": 350.18
    },
    {
      "epoch": 1.4315752845851768,
      "grad_norm": 0.8045446872711182,
      "learning_rate": 3.7636997331175805e-05,
      "loss": 0.1639,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 14400,
      "tokens_per_second_per_gpu": 338.94
    },
    {
      "epoch": 1.4325694686086394,
      "grad_norm": 0.45409709215164185,
      "learning_rate": 3.751438623959601e-05,
      "loss": 0.1665,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 14410,
      "tokens_per_second_per_gpu": 350.33
    },
    {
      "epoch": 1.4335636526321023,
      "grad_norm": 0.5277268886566162,
      "learning_rate": 3.739192906328958e-05,
      "loss": 0.1444,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 14420,
      "tokens_per_second_per_gpu": 332.77
    },
    {
      "epoch": 1.434557836655565,
      "grad_norm": 0.5323857069015503,
      "learning_rate": 3.726962610389435e-05,
      "loss": 0.1362,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 14430,
      "tokens_per_second_per_gpu": 339.66
    },
    {
      "epoch": 1.4355520206790278,
      "grad_norm": 0.7022963166236877,
      "learning_rate": 3.7147477662668386e-05,
      "loss": 0.1252,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.54,
      "memory/max_allocated (GiB)": 21.54,
      "step": 14440,
      "tokens_per_second_per_gpu": 313.51
    },
    {
      "epoch": 1.4365462047024904,
      "grad_norm": 0.4480128586292267,
      "learning_rate": 3.702548404048917e-05,
      "loss": 0.1617,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 14450,
      "tokens_per_second_per_gpu": 400.86
    },
    {
      "epoch": 1.4375403887259532,
      "grad_norm": 0.508468747138977,
      "learning_rate": 3.690364553785268e-05,
      "loss": 0.1598,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 14460,
      "tokens_per_second_per_gpu": 323.65
    },
    {
      "epoch": 1.4385345727494159,
      "grad_norm": 0.3107985854148865,
      "learning_rate": 3.678196245487299e-05,
      "loss": 0.1208,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 14470,
      "tokens_per_second_per_gpu": 328.38
    },
    {
      "epoch": 1.4395287567728787,
      "grad_norm": 0.5842998027801514,
      "learning_rate": 3.666043509128118e-05,
      "loss": 0.1969,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 14480,
      "tokens_per_second_per_gpu": 419.05
    },
    {
      "epoch": 1.4405229407963414,
      "grad_norm": 0.6047186255455017,
      "learning_rate": 3.6539063746424884e-05,
      "loss": 0.1706,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.82,
      "memory/max_allocated (GiB)": 19.82,
      "step": 14490,
      "tokens_per_second_per_gpu": 352.49
    },
    {
      "epoch": 1.4415171248198042,
      "grad_norm": 0.37089505791664124,
      "learning_rate": 3.641784871926733e-05,
      "loss": 0.1393,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 14500,
      "tokens_per_second_per_gpu": 393.16
    },
    {
      "epoch": 1.4425113088432668,
      "grad_norm": 0.472478985786438,
      "learning_rate": 3.629679030838682e-05,
      "loss": 0.1796,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 14510,
      "tokens_per_second_per_gpu": 386.53
    },
    {
      "epoch": 1.4435054928667297,
      "grad_norm": 0.3645547330379486,
      "learning_rate": 3.617588881197571e-05,
      "loss": 0.1152,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 14520,
      "tokens_per_second_per_gpu": 360.27
    },
    {
      "epoch": 1.4444996768901923,
      "grad_norm": 0.16537845134735107,
      "learning_rate": 3.605514452784e-05,
      "loss": 0.1252,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 14530,
      "tokens_per_second_per_gpu": 358.08
    },
    {
      "epoch": 1.4454938609136552,
      "grad_norm": 0.49066445231437683,
      "learning_rate": 3.593455775339837e-05,
      "loss": 0.1967,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 14540,
      "tokens_per_second_per_gpu": 403.93
    },
    {
      "epoch": 1.4464880449371178,
      "grad_norm": 0.424955815076828,
      "learning_rate": 3.5814128785681554e-05,
      "loss": 0.1566,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 14550,
      "tokens_per_second_per_gpu": 364.0
    },
    {
      "epoch": 1.4474822289605807,
      "grad_norm": 0.5189529657363892,
      "learning_rate": 3.569385792133151e-05,
      "loss": 0.1663,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 14560,
      "tokens_per_second_per_gpu": 384.8
    },
    {
      "epoch": 1.4484764129840433,
      "grad_norm": 0.3756466209888458,
      "learning_rate": 3.5573745456600826e-05,
      "loss": 0.1533,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14570,
      "tokens_per_second_per_gpu": 368.89
    },
    {
      "epoch": 1.4494705970075061,
      "grad_norm": 0.5015901923179626,
      "learning_rate": 3.54537916873519e-05,
      "loss": 0.1883,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 14580,
      "tokens_per_second_per_gpu": 330.11
    },
    {
      "epoch": 1.4504647810309688,
      "grad_norm": 0.49634647369384766,
      "learning_rate": 3.5333996909056176e-05,
      "loss": 0.1883,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 14590,
      "tokens_per_second_per_gpu": 392.27
    },
    {
      "epoch": 1.4514589650544316,
      "grad_norm": 0.27927589416503906,
      "learning_rate": 3.521436141679357e-05,
      "loss": 0.1499,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 14600,
      "tokens_per_second_per_gpu": 348.04
    },
    {
      "epoch": 1.4524531490778942,
      "grad_norm": 0.6297810077667236,
      "learning_rate": 3.5094885505251515e-05,
      "loss": 0.1779,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 14610,
      "tokens_per_second_per_gpu": 338.31
    },
    {
      "epoch": 1.453447333101357,
      "grad_norm": 0.6115005612373352,
      "learning_rate": 3.497556946872451e-05,
      "loss": 0.1784,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14620,
      "tokens_per_second_per_gpu": 371.67
    },
    {
      "epoch": 1.4544415171248197,
      "grad_norm": 0.3485526442527771,
      "learning_rate": 3.485641360111309e-05,
      "loss": 0.1118,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 14630,
      "tokens_per_second_per_gpu": 305.03
    },
    {
      "epoch": 1.4554357011482826,
      "grad_norm": 0.5678386688232422,
      "learning_rate": 3.473741819592341e-05,
      "loss": 0.1297,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14640,
      "tokens_per_second_per_gpu": 340.8
    },
    {
      "epoch": 1.4564298851717452,
      "grad_norm": 0.5769549012184143,
      "learning_rate": 3.4618583546266246e-05,
      "loss": 0.1583,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 14650,
      "tokens_per_second_per_gpu": 344.23
    },
    {
      "epoch": 1.457424069195208,
      "grad_norm": 0.48255455493927,
      "learning_rate": 3.449990994485649e-05,
      "loss": 0.1803,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 14660,
      "tokens_per_second_per_gpu": 325.5
    },
    {
      "epoch": 1.458418253218671,
      "grad_norm": 0.5541951060295105,
      "learning_rate": 3.4381397684012296e-05,
      "loss": 0.1496,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 14670,
      "tokens_per_second_per_gpu": 376.55
    },
    {
      "epoch": 1.4594124372421335,
      "grad_norm": 0.4175207316875458,
      "learning_rate": 3.426304705565445e-05,
      "loss": 0.1257,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 14680,
      "tokens_per_second_per_gpu": 338.73
    },
    {
      "epoch": 1.4604066212655962,
      "grad_norm": 0.7478891611099243,
      "learning_rate": 3.4144858351305496e-05,
      "loss": 0.1461,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 14690,
      "tokens_per_second_per_gpu": 313.51
    },
    {
      "epoch": 1.461400805289059,
      "grad_norm": 0.418258935213089,
      "learning_rate": 3.402683186208922e-05,
      "loss": 0.1719,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14700,
      "tokens_per_second_per_gpu": 372.01
    },
    {
      "epoch": 1.4623949893125219,
      "grad_norm": 0.6298655271530151,
      "learning_rate": 3.390896787872985e-05,
      "loss": 0.1876,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 14710,
      "tokens_per_second_per_gpu": 384.18
    },
    {
      "epoch": 1.4633891733359845,
      "grad_norm": 0.2738591134548187,
      "learning_rate": 3.379126669155122e-05,
      "loss": 0.1502,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 14720,
      "tokens_per_second_per_gpu": 316.12
    },
    {
      "epoch": 1.4643833573594471,
      "grad_norm": 0.6010544896125793,
      "learning_rate": 3.3673728590476296e-05,
      "loss": 0.1819,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.44,
      "memory/max_allocated (GiB)": 20.44,
      "step": 14730,
      "tokens_per_second_per_gpu": 376.31
    },
    {
      "epoch": 1.46537754138291,
      "grad_norm": 0.4839765727519989,
      "learning_rate": 3.355635386502619e-05,
      "loss": 0.1465,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14740,
      "tokens_per_second_per_gpu": 339.37
    },
    {
      "epoch": 1.4663717254063728,
      "grad_norm": 0.569148063659668,
      "learning_rate": 3.3439142804319743e-05,
      "loss": 0.1178,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14750,
      "tokens_per_second_per_gpu": 338.15
    },
    {
      "epoch": 1.4673659094298355,
      "grad_norm": 0.7138640284538269,
      "learning_rate": 3.3322095697072496e-05,
      "loss": 0.1648,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14760,
      "tokens_per_second_per_gpu": 334.69
    },
    {
      "epoch": 1.468360093453298,
      "grad_norm": 0.7108574509620667,
      "learning_rate": 3.3205212831596264e-05,
      "loss": 0.1368,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 14770,
      "tokens_per_second_per_gpu": 386.29
    },
    {
      "epoch": 1.469354277476761,
      "grad_norm": 0.347858190536499,
      "learning_rate": 3.30884944957982e-05,
      "loss": 0.164,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 14780,
      "tokens_per_second_per_gpu": 400.72
    },
    {
      "epoch": 1.4703484615002238,
      "grad_norm": 0.4929182827472687,
      "learning_rate": 3.29719409771803e-05,
      "loss": 0.1748,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 14790,
      "tokens_per_second_per_gpu": 369.52
    },
    {
      "epoch": 1.4713426455236864,
      "grad_norm": 0.4838101863861084,
      "learning_rate": 3.2855552562838445e-05,
      "loss": 0.1336,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 14800,
      "tokens_per_second_per_gpu": 353.25
    },
    {
      "epoch": 1.472336829547149,
      "grad_norm": 0.6003445386886597,
      "learning_rate": 3.273932953946193e-05,
      "loss": 0.1646,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14810,
      "tokens_per_second_per_gpu": 359.65
    },
    {
      "epoch": 1.473331013570612,
      "grad_norm": 0.40864112973213196,
      "learning_rate": 3.262327219333262e-05,
      "loss": 0.1307,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 14820,
      "tokens_per_second_per_gpu": 283.81
    },
    {
      "epoch": 1.4743251975940748,
      "grad_norm": 0.5634713768959045,
      "learning_rate": 3.250738081032433e-05,
      "loss": 0.1409,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 14830,
      "tokens_per_second_per_gpu": 370.86
    },
    {
      "epoch": 1.4753193816175374,
      "grad_norm": 0.421989768743515,
      "learning_rate": 3.239165567590197e-05,
      "loss": 0.1392,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 14840,
      "tokens_per_second_per_gpu": 329.01
    },
    {
      "epoch": 1.476313565641,
      "grad_norm": 0.8798142671585083,
      "learning_rate": 3.2276097075121014e-05,
      "loss": 0.2051,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 14850,
      "tokens_per_second_per_gpu": 424.3
    },
    {
      "epoch": 1.4773077496644629,
      "grad_norm": 0.4949728846549988,
      "learning_rate": 3.216070529262678e-05,
      "loss": 0.156,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.65,
      "memory/max_allocated (GiB)": 20.65,
      "step": 14860,
      "tokens_per_second_per_gpu": 344.69
    },
    {
      "epoch": 1.4783019336879257,
      "grad_norm": 0.44336196780204773,
      "learning_rate": 3.204548061265353e-05,
      "loss": 0.1542,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14870,
      "tokens_per_second_per_gpu": 437.39
    },
    {
      "epoch": 1.4792961177113884,
      "grad_norm": 0.3560582399368286,
      "learning_rate": 3.193042331902408e-05,
      "loss": 0.1638,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 14880,
      "tokens_per_second_per_gpu": 316.98
    },
    {
      "epoch": 1.480290301734851,
      "grad_norm": 0.5532661080360413,
      "learning_rate": 3.181553369514881e-05,
      "loss": 0.18,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 14890,
      "tokens_per_second_per_gpu": 356.94
    },
    {
      "epoch": 1.4812844857583138,
      "grad_norm": 0.5040679574012756,
      "learning_rate": 3.170081202402518e-05,
      "loss": 0.1776,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 14900,
      "tokens_per_second_per_gpu": 296.93
    },
    {
      "epoch": 1.4822786697817767,
      "grad_norm": 0.7191395163536072,
      "learning_rate": 3.158625858823688e-05,
      "loss": 0.1919,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 14910,
      "tokens_per_second_per_gpu": 390.26
    },
    {
      "epoch": 1.4832728538052393,
      "grad_norm": 0.6036331653594971,
      "learning_rate": 3.1471873669953275e-05,
      "loss": 0.1501,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 14920,
      "tokens_per_second_per_gpu": 324.83
    },
    {
      "epoch": 1.4842670378287022,
      "grad_norm": 0.4078894555568695,
      "learning_rate": 3.135765755092854e-05,
      "loss": 0.1536,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 14930,
      "tokens_per_second_per_gpu": 434.52
    },
    {
      "epoch": 1.4852612218521648,
      "grad_norm": 0.5916171669960022,
      "learning_rate": 3.1243610512501175e-05,
      "loss": 0.1538,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 14940,
      "tokens_per_second_per_gpu": 439.02
    },
    {
      "epoch": 1.4862554058756277,
      "grad_norm": 0.4087867736816406,
      "learning_rate": 3.1129732835593085e-05,
      "loss": 0.1858,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 14950,
      "tokens_per_second_per_gpu": 284.99
    },
    {
      "epoch": 1.4872495898990903,
      "grad_norm": 0.371852308511734,
      "learning_rate": 3.101602480070909e-05,
      "loss": 0.1251,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 14960,
      "tokens_per_second_per_gpu": 358.19
    },
    {
      "epoch": 1.4882437739225531,
      "grad_norm": 0.529620349407196,
      "learning_rate": 3.0902486687936097e-05,
      "loss": 0.1941,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 14970,
      "tokens_per_second_per_gpu": 369.15
    },
    {
      "epoch": 1.4892379579460158,
      "grad_norm": 0.21512271463871002,
      "learning_rate": 3.0789118776942484e-05,
      "loss": 0.1448,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 14980,
      "tokens_per_second_per_gpu": 350.83
    },
    {
      "epoch": 1.4902321419694786,
      "grad_norm": 0.6459974050521851,
      "learning_rate": 3.067592134697741e-05,
      "loss": 0.1853,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 14990,
      "tokens_per_second_per_gpu": 419.24
    },
    {
      "epoch": 1.4912263259929412,
      "grad_norm": 0.6449222564697266,
      "learning_rate": 3.0562894676870014e-05,
      "loss": 0.1876,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 15000,
      "tokens_per_second_per_gpu": 426.0
    },
    {
      "epoch": 1.492220510016404,
      "grad_norm": 0.5701255202293396,
      "learning_rate": 3.045003904502891e-05,
      "loss": 0.1583,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 15010,
      "tokens_per_second_per_gpu": 355.73
    },
    {
      "epoch": 1.4932146940398667,
      "grad_norm": 0.754385769367218,
      "learning_rate": 3.0337354729441338e-05,
      "loss": 0.2089,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 15020,
      "tokens_per_second_per_gpu": 372.69
    },
    {
      "epoch": 1.4942088780633296,
      "grad_norm": 0.27737346291542053,
      "learning_rate": 3.022484200767264e-05,
      "loss": 0.1358,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 15030,
      "tokens_per_second_per_gpu": 445.86
    },
    {
      "epoch": 1.4952030620867922,
      "grad_norm": 0.35400980710983276,
      "learning_rate": 3.0112501156865348e-05,
      "loss": 0.1361,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 15040,
      "tokens_per_second_per_gpu": 382.91
    },
    {
      "epoch": 1.496197246110255,
      "grad_norm": 0.703973650932312,
      "learning_rate": 3.000033245373881e-05,
      "loss": 0.1716,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 15050,
      "tokens_per_second_per_gpu": 379.21
    },
    {
      "epoch": 1.4971914301337177,
      "grad_norm": 0.47651368379592896,
      "learning_rate": 2.988833617458816e-05,
      "loss": 0.1148,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 15060,
      "tokens_per_second_per_gpu": 311.35
    },
    {
      "epoch": 1.4981856141571805,
      "grad_norm": 0.626930296421051,
      "learning_rate": 2.977651259528399e-05,
      "loss": 0.1478,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 15070,
      "tokens_per_second_per_gpu": 364.12
    },
    {
      "epoch": 1.4991797981806432,
      "grad_norm": 0.5578542351722717,
      "learning_rate": 2.9664861991271343e-05,
      "loss": 0.158,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 15080,
      "tokens_per_second_per_gpu": 428.74
    },
    {
      "epoch": 1.500173982204106,
      "grad_norm": 0.3915964961051941,
      "learning_rate": 2.9553384637569282e-05,
      "loss": 0.1047,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 15090,
      "tokens_per_second_per_gpu": 324.43
    },
    {
      "epoch": 1.5011681662275689,
      "grad_norm": 0.6796404123306274,
      "learning_rate": 2.944208080877008e-05,
      "loss": 0.1921,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 15100,
      "tokens_per_second_per_gpu": 373.42
    },
    {
      "epoch": 1.5021623502510315,
      "grad_norm": 0.3962237238883972,
      "learning_rate": 2.933095077903861e-05,
      "loss": 0.1629,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 15110,
      "tokens_per_second_per_gpu": 322.99
    },
    {
      "epoch": 1.5031565342744941,
      "grad_norm": 1.0471795797348022,
      "learning_rate": 2.921999482211165e-05,
      "loss": 0.1593,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 15120,
      "tokens_per_second_per_gpu": 318.74
    },
    {
      "epoch": 1.504150718297957,
      "grad_norm": 0.3305923640727997,
      "learning_rate": 2.9109213211297103e-05,
      "loss": 0.1913,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 15130,
      "tokens_per_second_per_gpu": 371.82
    },
    {
      "epoch": 1.5051449023214198,
      "grad_norm": 0.5967475771903992,
      "learning_rate": 2.8998606219473555e-05,
      "loss": 0.1741,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 15140,
      "tokens_per_second_per_gpu": 376.88
    },
    {
      "epoch": 1.5061390863448825,
      "grad_norm": 0.5528421998023987,
      "learning_rate": 2.888817411908935e-05,
      "loss": 0.1523,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 15150,
      "tokens_per_second_per_gpu": 323.43
    },
    {
      "epoch": 1.507133270368345,
      "grad_norm": 0.6192378997802734,
      "learning_rate": 2.877791718216214e-05,
      "loss": 0.1434,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.16,
      "memory/max_allocated (GiB)": 18.16,
      "step": 15160,
      "tokens_per_second_per_gpu": 300.91
    },
    {
      "epoch": 1.508127454391808,
      "grad_norm": 0.2290807068347931,
      "learning_rate": 2.866783568027802e-05,
      "loss": 0.1334,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 15170,
      "tokens_per_second_per_gpu": 382.23
    },
    {
      "epoch": 1.5091216384152708,
      "grad_norm": 0.748935341835022,
      "learning_rate": 2.8557929884591038e-05,
      "loss": 0.183,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 15180,
      "tokens_per_second_per_gpu": 358.48
    },
    {
      "epoch": 1.5101158224387334,
      "grad_norm": 0.6596253514289856,
      "learning_rate": 2.844820006582235e-05,
      "loss": 0.1251,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 15190,
      "tokens_per_second_per_gpu": 329.83
    },
    {
      "epoch": 1.511110006462196,
      "grad_norm": 0.5581555962562561,
      "learning_rate": 2.8338646494259746e-05,
      "loss": 0.1731,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 15200,
      "tokens_per_second_per_gpu": 382.43
    },
    {
      "epoch": 1.512104190485659,
      "grad_norm": 0.525514543056488,
      "learning_rate": 2.8229269439756768e-05,
      "loss": 0.1252,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 15210,
      "tokens_per_second_per_gpu": 398.56
    },
    {
      "epoch": 1.5130983745091218,
      "grad_norm": 0.5555036664009094,
      "learning_rate": 2.812006917173229e-05,
      "loss": 0.1561,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 15220,
      "tokens_per_second_per_gpu": 360.58
    },
    {
      "epoch": 1.5140925585325844,
      "grad_norm": 0.4960024058818817,
      "learning_rate": 2.801104595916957e-05,
      "loss": 0.1428,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 15230,
      "tokens_per_second_per_gpu": 378.86
    },
    {
      "epoch": 1.515086742556047,
      "grad_norm": 0.6800406575202942,
      "learning_rate": 2.7902200070615868e-05,
      "loss": 0.2009,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 15240,
      "tokens_per_second_per_gpu": 347.33
    },
    {
      "epoch": 1.5160809265795099,
      "grad_norm": 0.5688208341598511,
      "learning_rate": 2.7793531774181614e-05,
      "loss": 0.1421,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 15250,
      "tokens_per_second_per_gpu": 346.32
    },
    {
      "epoch": 1.5170751106029727,
      "grad_norm": 0.6293416619300842,
      "learning_rate": 2.7685041337539786e-05,
      "loss": 0.1442,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 15260,
      "tokens_per_second_per_gpu": 373.42
    },
    {
      "epoch": 1.5180692946264354,
      "grad_norm": 0.5825393795967102,
      "learning_rate": 2.7576729027925286e-05,
      "loss": 0.1386,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 15270,
      "tokens_per_second_per_gpu": 310.81
    },
    {
      "epoch": 1.519063478649898,
      "grad_norm": 0.37790921330451965,
      "learning_rate": 2.7468595112134165e-05,
      "loss": 0.1603,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 15280,
      "tokens_per_second_per_gpu": 344.02
    },
    {
      "epoch": 1.5200576626733608,
      "grad_norm": 0.4327569603919983,
      "learning_rate": 2.7360639856523172e-05,
      "loss": 0.1681,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 15290,
      "tokens_per_second_per_gpu": 340.19
    },
    {
      "epoch": 1.5210518466968237,
      "grad_norm": 0.272372841835022,
      "learning_rate": 2.7252863527008867e-05,
      "loss": 0.1851,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 15300,
      "tokens_per_second_per_gpu": 380.93
    },
    {
      "epoch": 1.5220460307202863,
      "grad_norm": 0.40148428082466125,
      "learning_rate": 2.7145266389067182e-05,
      "loss": 0.1504,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 15310,
      "tokens_per_second_per_gpu": 331.6
    },
    {
      "epoch": 1.523040214743749,
      "grad_norm": 0.6262397170066833,
      "learning_rate": 2.703784870773255e-05,
      "loss": 0.1614,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 15320,
      "tokens_per_second_per_gpu": 431.35
    },
    {
      "epoch": 1.5240343987672118,
      "grad_norm": 0.3884369730949402,
      "learning_rate": 2.6930610747597483e-05,
      "loss": 0.1573,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.1,
      "memory/max_allocated (GiB)": 19.1,
      "step": 15330,
      "tokens_per_second_per_gpu": 410.18
    },
    {
      "epoch": 1.5250285827906747,
      "grad_norm": 0.285769522190094,
      "learning_rate": 2.682355277281169e-05,
      "loss": 0.1303,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 15340,
      "tokens_per_second_per_gpu": 317.47
    },
    {
      "epoch": 1.5260227668141373,
      "grad_norm": 0.8329140543937683,
      "learning_rate": 2.671667504708163e-05,
      "loss": 0.1356,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 15350,
      "tokens_per_second_per_gpu": 341.69
    },
    {
      "epoch": 1.5270169508376,
      "grad_norm": 0.40556618571281433,
      "learning_rate": 2.6609977833669686e-05,
      "loss": 0.1665,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 15360,
      "tokens_per_second_per_gpu": 382.84
    },
    {
      "epoch": 1.5280111348610628,
      "grad_norm": 0.23340922594070435,
      "learning_rate": 2.650346139539368e-05,
      "loss": 0.1838,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 15370,
      "tokens_per_second_per_gpu": 369.68
    },
    {
      "epoch": 1.5290053188845256,
      "grad_norm": 0.4838123917579651,
      "learning_rate": 2.6397125994626128e-05,
      "loss": 0.1513,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 15380,
      "tokens_per_second_per_gpu": 354.9
    },
    {
      "epoch": 1.5299995029079883,
      "grad_norm": 0.6558499932289124,
      "learning_rate": 2.6290971893293547e-05,
      "loss": 0.1673,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 15390,
      "tokens_per_second_per_gpu": 355.68
    },
    {
      "epoch": 1.5309936869314509,
      "grad_norm": 0.7235273718833923,
      "learning_rate": 2.618499935287595e-05,
      "loss": 0.1269,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 15400,
      "tokens_per_second_per_gpu": 347.78
    },
    {
      "epoch": 1.5319878709549137,
      "grad_norm": 2.2355356216430664,
      "learning_rate": 2.6079208634406106e-05,
      "loss": 0.1649,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 15410,
      "tokens_per_second_per_gpu": 418.05
    },
    {
      "epoch": 1.5329820549783766,
      "grad_norm": 0.2927922010421753,
      "learning_rate": 2.5973599998468935e-05,
      "loss": 0.1413,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 15420,
      "tokens_per_second_per_gpu": 348.29
    },
    {
      "epoch": 1.5339762390018392,
      "grad_norm": 0.3300843834877014,
      "learning_rate": 2.586817370520077e-05,
      "loss": 0.1356,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 15430,
      "tokens_per_second_per_gpu": 323.42
    },
    {
      "epoch": 1.5349704230253018,
      "grad_norm": 0.5750816464424133,
      "learning_rate": 2.5762930014288933e-05,
      "loss": 0.1307,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 15440,
      "tokens_per_second_per_gpu": 347.56
    },
    {
      "epoch": 1.5359646070487647,
      "grad_norm": 0.7192912101745605,
      "learning_rate": 2.5657869184970795e-05,
      "loss": 0.1837,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 15450,
      "tokens_per_second_per_gpu": 347.13
    },
    {
      "epoch": 1.5369587910722275,
      "grad_norm": 0.7183093428611755,
      "learning_rate": 2.555299147603345e-05,
      "loss": 0.1978,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 15460,
      "tokens_per_second_per_gpu": 372.76
    },
    {
      "epoch": 1.5379529750956902,
      "grad_norm": 0.27620071172714233,
      "learning_rate": 2.5448297145812805e-05,
      "loss": 0.1732,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 15470,
      "tokens_per_second_per_gpu": 341.71
    },
    {
      "epoch": 1.5389471591191528,
      "grad_norm": 0.3960123062133789,
      "learning_rate": 2.5343786452193185e-05,
      "loss": 0.1229,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 15480,
      "tokens_per_second_per_gpu": 351.22
    },
    {
      "epoch": 1.5399413431426157,
      "grad_norm": 0.4011591970920563,
      "learning_rate": 2.5239459652606457e-05,
      "loss": 0.1504,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 15490,
      "tokens_per_second_per_gpu": 421.47
    },
    {
      "epoch": 1.5409355271660785,
      "grad_norm": 0.40115076303482056,
      "learning_rate": 2.51353170040316e-05,
      "loss": 0.116,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 15500,
      "tokens_per_second_per_gpu": 326.93
    },
    {
      "epoch": 1.5419297111895411,
      "grad_norm": 0.487507164478302,
      "learning_rate": 2.5031358762994005e-05,
      "loss": 0.1467,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.85,
      "memory/max_allocated (GiB)": 18.85,
      "step": 15510,
      "tokens_per_second_per_gpu": 347.74
    },
    {
      "epoch": 1.5429238952130038,
      "grad_norm": 0.4600401222705841,
      "learning_rate": 2.492758518556473e-05,
      "loss": 0.1632,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 15520,
      "tokens_per_second_per_gpu": 389.11
    },
    {
      "epoch": 1.5439180792364666,
      "grad_norm": 0.5790596604347229,
      "learning_rate": 2.482399652736006e-05,
      "loss": 0.1396,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 15530,
      "tokens_per_second_per_gpu": 329.02
    },
    {
      "epoch": 1.5449122632599295,
      "grad_norm": 0.5716336369514465,
      "learning_rate": 2.4720593043540752e-05,
      "loss": 0.201,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 15540,
      "tokens_per_second_per_gpu": 375.4
    },
    {
      "epoch": 1.545906447283392,
      "grad_norm": 0.5810480117797852,
      "learning_rate": 2.461737498881148e-05,
      "loss": 0.1525,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 15550,
      "tokens_per_second_per_gpu": 331.4
    },
    {
      "epoch": 1.5469006313068547,
      "grad_norm": 0.3276031017303467,
      "learning_rate": 2.451434261742005e-05,
      "loss": 0.1237,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 15560,
      "tokens_per_second_per_gpu": 303.69
    },
    {
      "epoch": 1.5478948153303178,
      "grad_norm": 0.40953657031059265,
      "learning_rate": 2.4411496183157045e-05,
      "loss": 0.131,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 15570,
      "tokens_per_second_per_gpu": 341.6
    },
    {
      "epoch": 1.5488889993537804,
      "grad_norm": 0.34955236315727234,
      "learning_rate": 2.4308835939354913e-05,
      "loss": 0.1647,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.09,
      "memory/max_allocated (GiB)": 19.09,
      "step": 15580,
      "tokens_per_second_per_gpu": 373.13
    },
    {
      "epoch": 1.549883183377243,
      "grad_norm": 0.5104487538337708,
      "learning_rate": 2.4206362138887584e-05,
      "loss": 0.1674,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 15590,
      "tokens_per_second_per_gpu": 413.19
    },
    {
      "epoch": 1.550877367400706,
      "grad_norm": 0.3313903510570526,
      "learning_rate": 2.4104075034169628e-05,
      "loss": 0.1848,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 15600,
      "tokens_per_second_per_gpu": 361.41
    },
    {
      "epoch": 1.5518715514241688,
      "grad_norm": 0.35265353322029114,
      "learning_rate": 2.400197487715585e-05,
      "loss": 0.1431,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 15610,
      "tokens_per_second_per_gpu": 352.88
    },
    {
      "epoch": 1.5528657354476314,
      "grad_norm": 0.5764623880386353,
      "learning_rate": 2.390006191934048e-05,
      "loss": 0.145,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 15620,
      "tokens_per_second_per_gpu": 337.91
    },
    {
      "epoch": 1.553859919471094,
      "grad_norm": 0.5069451928138733,
      "learning_rate": 2.3798336411756682e-05,
      "loss": 0.1489,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 15630,
      "tokens_per_second_per_gpu": 411.7
    },
    {
      "epoch": 1.5548541034945569,
      "grad_norm": 0.3448779881000519,
      "learning_rate": 2.3696798604975933e-05,
      "loss": 0.1571,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 15640,
      "tokens_per_second_per_gpu": 383.87
    },
    {
      "epoch": 1.5558482875180197,
      "grad_norm": 0.7202053070068359,
      "learning_rate": 2.359544874910723e-05,
      "loss": 0.1472,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 15650,
      "tokens_per_second_per_gpu": 344.25
    },
    {
      "epoch": 1.5568424715414824,
      "grad_norm": 0.41816800832748413,
      "learning_rate": 2.3494287093796763e-05,
      "loss": 0.1428,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 15660,
      "tokens_per_second_per_gpu": 343.33
    },
    {
      "epoch": 1.557836655564945,
      "grad_norm": 0.47676244378089905,
      "learning_rate": 2.339331388822701e-05,
      "loss": 0.1495,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 15670,
      "tokens_per_second_per_gpu": 386.13
    },
    {
      "epoch": 1.5588308395884078,
      "grad_norm": 0.43887218832969666,
      "learning_rate": 2.3292529381116336e-05,
      "loss": 0.1541,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 15680,
      "tokens_per_second_per_gpu": 345.57
    },
    {
      "epoch": 1.5598250236118707,
      "grad_norm": 0.5106401443481445,
      "learning_rate": 2.319193382071829e-05,
      "loss": 0.1124,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 15690,
      "tokens_per_second_per_gpu": 339.79
    },
    {
      "epoch": 1.5608192076353333,
      "grad_norm": 0.6262336373329163,
      "learning_rate": 2.3091527454821027e-05,
      "loss": 0.1248,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 15700,
      "tokens_per_second_per_gpu": 372.97
    },
    {
      "epoch": 1.561813391658796,
      "grad_norm": 0.48663491010665894,
      "learning_rate": 2.299131053074659e-05,
      "loss": 0.1214,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 15710,
      "tokens_per_second_per_gpu": 292.22
    },
    {
      "epoch": 1.5628075756822588,
      "grad_norm": 0.7886651158332825,
      "learning_rate": 2.2891283295350508e-05,
      "loss": 0.1655,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.41,
      "memory/max_allocated (GiB)": 21.41,
      "step": 15720,
      "tokens_per_second_per_gpu": 390.33
    },
    {
      "epoch": 1.5638017597057217,
      "grad_norm": 0.2919699251651764,
      "learning_rate": 2.2791445995020943e-05,
      "loss": 0.1406,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 15730,
      "tokens_per_second_per_gpu": 323.88
    },
    {
      "epoch": 1.5647959437291843,
      "grad_norm": 0.4909593164920807,
      "learning_rate": 2.2691798875678304e-05,
      "loss": 0.1673,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 15740,
      "tokens_per_second_per_gpu": 320.3
    },
    {
      "epoch": 1.565790127752647,
      "grad_norm": 0.37527504563331604,
      "learning_rate": 2.2592342182774482e-05,
      "loss": 0.1746,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 15750,
      "tokens_per_second_per_gpu": 418.23
    },
    {
      "epoch": 1.5667843117761098,
      "grad_norm": 0.5083606243133545,
      "learning_rate": 2.249307616129237e-05,
      "loss": 0.1401,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 15760,
      "tokens_per_second_per_gpu": 366.1
    },
    {
      "epoch": 1.5677784957995726,
      "grad_norm": 0.5692464113235474,
      "learning_rate": 2.2394001055745107e-05,
      "loss": 0.1795,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 15770,
      "tokens_per_second_per_gpu": 340.36
    },
    {
      "epoch": 1.5687726798230353,
      "grad_norm": 0.6630048751831055,
      "learning_rate": 2.2295117110175645e-05,
      "loss": 0.1981,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 15780,
      "tokens_per_second_per_gpu": 361.48
    },
    {
      "epoch": 1.5697668638464979,
      "grad_norm": 0.6706286072731018,
      "learning_rate": 2.2196424568156073e-05,
      "loss": 0.1255,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 15790,
      "tokens_per_second_per_gpu": 367.95
    },
    {
      "epoch": 1.5707610478699607,
      "grad_norm": 0.6745642423629761,
      "learning_rate": 2.2097923672786913e-05,
      "loss": 0.167,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 15800,
      "tokens_per_second_per_gpu": 420.69
    },
    {
      "epoch": 1.5717552318934236,
      "grad_norm": 0.4416200518608093,
      "learning_rate": 2.1999614666696733e-05,
      "loss": 0.1483,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 15810,
      "tokens_per_second_per_gpu": 343.55
    },
    {
      "epoch": 1.5727494159168862,
      "grad_norm": 0.6229557394981384,
      "learning_rate": 2.1901497792041392e-05,
      "loss": 0.1796,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 15820,
      "tokens_per_second_per_gpu": 333.79
    },
    {
      "epoch": 1.5737435999403488,
      "grad_norm": 0.6999335289001465,
      "learning_rate": 2.1803573290503497e-05,
      "loss": 0.2508,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 15830,
      "tokens_per_second_per_gpu": 421.46
    },
    {
      "epoch": 1.5747377839638117,
      "grad_norm": 0.4594232141971588,
      "learning_rate": 2.170584140329177e-05,
      "loss": 0.1617,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 15840,
      "tokens_per_second_per_gpu": 342.47
    },
    {
      "epoch": 1.5757319679872746,
      "grad_norm": 0.18643365800380707,
      "learning_rate": 2.1608302371140533e-05,
      "loss": 0.142,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 15850,
      "tokens_per_second_per_gpu": 302.45
    },
    {
      "epoch": 1.5767261520107372,
      "grad_norm": 0.42105787992477417,
      "learning_rate": 2.1510956434308992e-05,
      "loss": 0.1691,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 15860,
      "tokens_per_second_per_gpu": 375.79
    },
    {
      "epoch": 1.5777203360341998,
      "grad_norm": 0.5371260046958923,
      "learning_rate": 2.1413803832580813e-05,
      "loss": 0.1398,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 15870,
      "tokens_per_second_per_gpu": 314.27
    },
    {
      "epoch": 1.5787145200576627,
      "grad_norm": 0.5238702297210693,
      "learning_rate": 2.1316844805263346e-05,
      "loss": 0.2099,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 15880,
      "tokens_per_second_per_gpu": 358.19
    },
    {
      "epoch": 1.5797087040811255,
      "grad_norm": 0.2773045599460602,
      "learning_rate": 2.1220079591187214e-05,
      "loss": 0.1528,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 15890,
      "tokens_per_second_per_gpu": 332.88
    },
    {
      "epoch": 1.5807028881045881,
      "grad_norm": 0.6425759196281433,
      "learning_rate": 2.112350842870553e-05,
      "loss": 0.1474,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 15900,
      "tokens_per_second_per_gpu": 303.59
    },
    {
      "epoch": 1.5816970721280508,
      "grad_norm": 0.3497686982154846,
      "learning_rate": 2.1027131555693524e-05,
      "loss": 0.1381,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 15910,
      "tokens_per_second_per_gpu": 300.82
    },
    {
      "epoch": 1.5826912561515136,
      "grad_norm": 0.37605515122413635,
      "learning_rate": 2.0930949209547813e-05,
      "loss": 0.1004,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 15920,
      "tokens_per_second_per_gpu": 326.71
    },
    {
      "epoch": 1.5836854401749765,
      "grad_norm": 0.6107087731361389,
      "learning_rate": 2.08349616271858e-05,
      "loss": 0.1309,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 15930,
      "tokens_per_second_per_gpu": 325.87
    },
    {
      "epoch": 1.584679624198439,
      "grad_norm": 0.4413928687572479,
      "learning_rate": 2.0739169045045237e-05,
      "loss": 0.1575,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 15940,
      "tokens_per_second_per_gpu": 370.56
    },
    {
      "epoch": 1.5856738082219017,
      "grad_norm": 0.3646581470966339,
      "learning_rate": 2.064357169908345e-05,
      "loss": 0.1437,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 15950,
      "tokens_per_second_per_gpu": 362.0
    },
    {
      "epoch": 1.5866679922453646,
      "grad_norm": 0.5044755935668945,
      "learning_rate": 2.054816982477693e-05,
      "loss": 0.1873,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 15960,
      "tokens_per_second_per_gpu": 335.28
    },
    {
      "epoch": 1.5876621762688274,
      "grad_norm": 0.5462661981582642,
      "learning_rate": 2.045296365712066e-05,
      "loss": 0.1646,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 15970,
      "tokens_per_second_per_gpu": 380.25
    },
    {
      "epoch": 1.58865636029229,
      "grad_norm": 0.3154492676258087,
      "learning_rate": 2.0357953430627575e-05,
      "loss": 0.1461,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.82,
      "memory/max_allocated (GiB)": 18.82,
      "step": 15980,
      "tokens_per_second_per_gpu": 362.04
    },
    {
      "epoch": 1.5896505443157527,
      "grad_norm": 0.4407619833946228,
      "learning_rate": 2.02631393793279e-05,
      "loss": 0.1701,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 15990,
      "tokens_per_second_per_gpu": 371.26
    },
    {
      "epoch": 1.5906447283392156,
      "grad_norm": 0.3745664358139038,
      "learning_rate": 2.0168521736768732e-05,
      "loss": 0.1009,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.62,
      "memory/max_allocated (GiB)": 20.62,
      "step": 16000,
      "tokens_per_second_per_gpu": 348.69
    },
    {
      "epoch": 1.5916389123626784,
      "grad_norm": 0.7978280186653137,
      "learning_rate": 2.007410073601326e-05,
      "loss": 0.189,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 16010,
      "tokens_per_second_per_gpu": 396.26
    },
    {
      "epoch": 1.592633096386141,
      "grad_norm": 0.5025820136070251,
      "learning_rate": 1.9979876609640437e-05,
      "loss": 0.1536,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 16020,
      "tokens_per_second_per_gpu": 389.34
    },
    {
      "epoch": 1.5936272804096037,
      "grad_norm": 0.37046942114830017,
      "learning_rate": 1.988584958974412e-05,
      "loss": 0.0939,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 16030,
      "tokens_per_second_per_gpu": 336.01
    },
    {
      "epoch": 1.5946214644330665,
      "grad_norm": 0.40366023778915405,
      "learning_rate": 1.979201990793279e-05,
      "loss": 0.131,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 16040,
      "tokens_per_second_per_gpu": 366.64
    },
    {
      "epoch": 1.5956156484565294,
      "grad_norm": 0.2725692689418793,
      "learning_rate": 1.9698387795328788e-05,
      "loss": 0.1501,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 16050,
      "tokens_per_second_per_gpu": 410.98
    },
    {
      "epoch": 1.596609832479992,
      "grad_norm": 0.5428498983383179,
      "learning_rate": 1.9604953482567756e-05,
      "loss": 0.1891,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 16060,
      "tokens_per_second_per_gpu": 418.32
    },
    {
      "epoch": 1.5976040165034546,
      "grad_norm": 0.5400422215461731,
      "learning_rate": 1.9511717199798208e-05,
      "loss": 0.1726,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 16070,
      "tokens_per_second_per_gpu": 399.7
    },
    {
      "epoch": 1.5985982005269175,
      "grad_norm": 0.5957368612289429,
      "learning_rate": 1.9418679176680743e-05,
      "loss": 0.1893,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 16080,
      "tokens_per_second_per_gpu": 388.42
    },
    {
      "epoch": 1.5995923845503803,
      "grad_norm": 0.8663964867591858,
      "learning_rate": 1.9325839642387755e-05,
      "loss": 0.1314,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 16090,
      "tokens_per_second_per_gpu": 297.89
    },
    {
      "epoch": 1.600586568573843,
      "grad_norm": 0.46175047755241394,
      "learning_rate": 1.9233198825602572e-05,
      "loss": 0.1427,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 16100,
      "tokens_per_second_per_gpu": 322.74
    },
    {
      "epoch": 1.6015807525973058,
      "grad_norm": 0.5290225148200989,
      "learning_rate": 1.9140756954519136e-05,
      "loss": 0.1824,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 16110,
      "tokens_per_second_per_gpu": 423.78
    },
    {
      "epoch": 1.6025749366207687,
      "grad_norm": 0.28334707021713257,
      "learning_rate": 1.904851425684131e-05,
      "loss": 0.1428,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 16120,
      "tokens_per_second_per_gpu": 358.47
    },
    {
      "epoch": 1.6035691206442313,
      "grad_norm": 0.4152858555316925,
      "learning_rate": 1.895647095978238e-05,
      "loss": 0.1508,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 16130,
      "tokens_per_second_per_gpu": 368.35
    },
    {
      "epoch": 1.604563304667694,
      "grad_norm": 0.5704591274261475,
      "learning_rate": 1.8864627290064396e-05,
      "loss": 0.1618,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 16140,
      "tokens_per_second_per_gpu": 371.9
    },
    {
      "epoch": 1.6055574886911568,
      "grad_norm": 0.5318484902381897,
      "learning_rate": 1.877298347391777e-05,
      "loss": 0.1667,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 16150,
      "tokens_per_second_per_gpu": 357.18
    },
    {
      "epoch": 1.6065516727146196,
      "grad_norm": 0.456482470035553,
      "learning_rate": 1.8681539737080543e-05,
      "loss": 0.152,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 16160,
      "tokens_per_second_per_gpu": 340.75
    },
    {
      "epoch": 1.6075458567380823,
      "grad_norm": 0.6000080108642578,
      "learning_rate": 1.8590296304797996e-05,
      "loss": 0.1347,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 16170,
      "tokens_per_second_per_gpu": 329.23
    },
    {
      "epoch": 1.6085400407615449,
      "grad_norm": 0.48490649461746216,
      "learning_rate": 1.8499253401822004e-05,
      "loss": 0.1503,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 16180,
      "tokens_per_second_per_gpu": 304.69
    },
    {
      "epoch": 1.6095342247850077,
      "grad_norm": 0.5004350543022156,
      "learning_rate": 1.840841125241044e-05,
      "loss": 0.18,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.82,
      "memory/max_allocated (GiB)": 19.82,
      "step": 16190,
      "tokens_per_second_per_gpu": 352.28
    },
    {
      "epoch": 1.6105284088084706,
      "grad_norm": 0.5725821256637573,
      "learning_rate": 1.8317770080326757e-05,
      "loss": 0.171,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 16200,
      "tokens_per_second_per_gpu": 427.8
    },
    {
      "epoch": 1.6115225928319332,
      "grad_norm": 0.47481808066368103,
      "learning_rate": 1.822733010883928e-05,
      "loss": 0.1318,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.65,
      "memory/max_allocated (GiB)": 20.65,
      "step": 16210,
      "tokens_per_second_per_gpu": 371.38
    },
    {
      "epoch": 1.6125167768553959,
      "grad_norm": 0.9544722437858582,
      "learning_rate": 1.813709156072081e-05,
      "loss": 0.1429,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 16220,
      "tokens_per_second_per_gpu": 353.59
    },
    {
      "epoch": 1.6135109608788587,
      "grad_norm": 0.5491587519645691,
      "learning_rate": 1.804705465824793e-05,
      "loss": 0.1672,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.41,
      "memory/max_allocated (GiB)": 21.41,
      "step": 16230,
      "tokens_per_second_per_gpu": 359.87
    },
    {
      "epoch": 1.6145051449023216,
      "grad_norm": 0.6248442530632019,
      "learning_rate": 1.795721962320057e-05,
      "loss": 0.1793,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 16240,
      "tokens_per_second_per_gpu": 367.54
    },
    {
      "epoch": 1.6154993289257842,
      "grad_norm": 0.28726670145988464,
      "learning_rate": 1.7867586676861416e-05,
      "loss": 0.1438,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 16250,
      "tokens_per_second_per_gpu": 278.32
    },
    {
      "epoch": 1.6164935129492468,
      "grad_norm": 0.36886531114578247,
      "learning_rate": 1.7778156040015393e-05,
      "loss": 0.182,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 16260,
      "tokens_per_second_per_gpu": 396.55
    },
    {
      "epoch": 1.6174876969727097,
      "grad_norm": 0.3643040955066681,
      "learning_rate": 1.7688927932948983e-05,
      "loss": 0.1251,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 16270,
      "tokens_per_second_per_gpu": 315.86
    },
    {
      "epoch": 1.6184818809961725,
      "grad_norm": 0.4020111560821533,
      "learning_rate": 1.7599902575449955e-05,
      "loss": 0.1725,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 16280,
      "tokens_per_second_per_gpu": 434.3
    },
    {
      "epoch": 1.6194760650196351,
      "grad_norm": 0.5112940073013306,
      "learning_rate": 1.7511080186806518e-05,
      "loss": 0.1305,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 16290,
      "tokens_per_second_per_gpu": 260.92
    },
    {
      "epoch": 1.6204702490430978,
      "grad_norm": 0.7122372984886169,
      "learning_rate": 1.742246098580701e-05,
      "loss": 0.1869,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 16300,
      "tokens_per_second_per_gpu": 320.49
    },
    {
      "epoch": 1.6214644330665606,
      "grad_norm": 0.6411992311477661,
      "learning_rate": 1.7334045190739277e-05,
      "loss": 0.2079,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.2,
      "memory/max_allocated (GiB)": 19.2,
      "step": 16310,
      "tokens_per_second_per_gpu": 406.2
    },
    {
      "epoch": 1.6224586170900235,
      "grad_norm": 0.673321008682251,
      "learning_rate": 1.7245833019390055e-05,
      "loss": 0.1607,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 16320,
      "tokens_per_second_per_gpu": 325.22
    },
    {
      "epoch": 1.623452801113486,
      "grad_norm": 0.6042284369468689,
      "learning_rate": 1.7157824689044632e-05,
      "loss": 0.1823,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 16330,
      "tokens_per_second_per_gpu": 340.57
    },
    {
      "epoch": 1.6244469851369487,
      "grad_norm": 0.3080300986766815,
      "learning_rate": 1.7070020416486065e-05,
      "loss": 0.125,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 16340,
      "tokens_per_second_per_gpu": 342.14
    },
    {
      "epoch": 1.6254411691604116,
      "grad_norm": 0.33366823196411133,
      "learning_rate": 1.6982420417994893e-05,
      "loss": 0.1487,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 16350,
      "tokens_per_second_per_gpu": 348.66
    },
    {
      "epoch": 1.6264353531838744,
      "grad_norm": 0.5744956731796265,
      "learning_rate": 1.6895024909348367e-05,
      "loss": 0.1528,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.19,
      "memory/max_allocated (GiB)": 18.19,
      "step": 16360,
      "tokens_per_second_per_gpu": 371.23
    },
    {
      "epoch": 1.627429537207337,
      "grad_norm": 0.578666090965271,
      "learning_rate": 1.6807834105820163e-05,
      "loss": 0.1521,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 16370,
      "tokens_per_second_per_gpu": 356.8
    },
    {
      "epoch": 1.6284237212307997,
      "grad_norm": 0.5622230768203735,
      "learning_rate": 1.6720848222179587e-05,
      "loss": 0.1405,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 16380,
      "tokens_per_second_per_gpu": 351.53
    },
    {
      "epoch": 1.6294179052542626,
      "grad_norm": 0.562519371509552,
      "learning_rate": 1.6634067472691283e-05,
      "loss": 0.1543,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 16390,
      "tokens_per_second_per_gpu": 360.87
    },
    {
      "epoch": 1.6304120892777254,
      "grad_norm": 0.5723958611488342,
      "learning_rate": 1.65474920711146e-05,
      "loss": 0.1881,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 16400,
      "tokens_per_second_per_gpu": 366.21
    },
    {
      "epoch": 1.631406273301188,
      "grad_norm": 0.5237053632736206,
      "learning_rate": 1.646112223070305e-05,
      "loss": 0.148,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 16410,
      "tokens_per_second_per_gpu": 354.74
    },
    {
      "epoch": 1.6324004573246507,
      "grad_norm": 0.576963484287262,
      "learning_rate": 1.6374958164203768e-05,
      "loss": 0.1729,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 16420,
      "tokens_per_second_per_gpu": 374.69
    },
    {
      "epoch": 1.6333946413481135,
      "grad_norm": 0.5174708366394043,
      "learning_rate": 1.6289000083857088e-05,
      "loss": 0.1487,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 16430,
      "tokens_per_second_per_gpu": 414.81
    },
    {
      "epoch": 1.6343888253715764,
      "grad_norm": 0.44571495056152344,
      "learning_rate": 1.620324820139595e-05,
      "loss": 0.1505,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 16440,
      "tokens_per_second_per_gpu": 359.78
    },
    {
      "epoch": 1.635383009395039,
      "grad_norm": 0.5897732377052307,
      "learning_rate": 1.61177027280453e-05,
      "loss": 0.1283,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 16450,
      "tokens_per_second_per_gpu": 328.51
    },
    {
      "epoch": 1.6363771934185016,
      "grad_norm": 0.5686812400817871,
      "learning_rate": 1.6032363874521804e-05,
      "loss": 0.2051,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 16460,
      "tokens_per_second_per_gpu": 373.45
    },
    {
      "epoch": 1.6373713774419645,
      "grad_norm": 0.6155579686164856,
      "learning_rate": 1.5947231851033016e-05,
      "loss": 0.1437,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 16470,
      "tokens_per_second_per_gpu": 329.49
    },
    {
      "epoch": 1.6383655614654273,
      "grad_norm": 0.4496712386608124,
      "learning_rate": 1.5862306867277155e-05,
      "loss": 0.1803,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 16480,
      "tokens_per_second_per_gpu": 373.25
    },
    {
      "epoch": 1.63935974548889,
      "grad_norm": 0.696696937084198,
      "learning_rate": 1.5777589132442373e-05,
      "loss": 0.1214,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 16490,
      "tokens_per_second_per_gpu": 342.62
    },
    {
      "epoch": 1.6403539295123526,
      "grad_norm": 0.5139226317405701,
      "learning_rate": 1.569307885520639e-05,
      "loss": 0.1722,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 16500,
      "tokens_per_second_per_gpu": 391.42
    },
    {
      "epoch": 1.6413481135358154,
      "grad_norm": 0.5510913133621216,
      "learning_rate": 1.5608776243735834e-05,
      "loss": 0.1549,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 16510,
      "tokens_per_second_per_gpu": 353.41
    },
    {
      "epoch": 1.6423422975592783,
      "grad_norm": 0.4640497863292694,
      "learning_rate": 1.5524681505685888e-05,
      "loss": 0.1431,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 16520,
      "tokens_per_second_per_gpu": 387.36
    },
    {
      "epoch": 1.643336481582741,
      "grad_norm": 0.59361332654953,
      "learning_rate": 1.5440794848199657e-05,
      "loss": 0.1749,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 16530,
      "tokens_per_second_per_gpu": 407.49
    },
    {
      "epoch": 1.6443306656062036,
      "grad_norm": 0.552768349647522,
      "learning_rate": 1.5357116477907728e-05,
      "loss": 0.1278,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 16540,
      "tokens_per_second_per_gpu": 381.79
    },
    {
      "epoch": 1.6453248496296664,
      "grad_norm": 0.7618371248245239,
      "learning_rate": 1.5273646600927583e-05,
      "loss": 0.163,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 16550,
      "tokens_per_second_per_gpu": 364.2
    },
    {
      "epoch": 1.6463190336531293,
      "grad_norm": 0.5775349140167236,
      "learning_rate": 1.5190385422863174e-05,
      "loss": 0.1467,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 16560,
      "tokens_per_second_per_gpu": 367.94
    },
    {
      "epoch": 1.647313217676592,
      "grad_norm": 0.5621904134750366,
      "learning_rate": 1.5107333148804414e-05,
      "loss": 0.1599,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 16570,
      "tokens_per_second_per_gpu": 324.44
    },
    {
      "epoch": 1.6483074017000545,
      "grad_norm": 0.565984845161438,
      "learning_rate": 1.5024489983326562e-05,
      "loss": 0.2052,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.67,
      "memory/max_allocated (GiB)": 19.67,
      "step": 16580,
      "tokens_per_second_per_gpu": 412.88
    },
    {
      "epoch": 1.6493015857235174,
      "grad_norm": 0.456301748752594,
      "learning_rate": 1.4941856130489884e-05,
      "loss": 0.1494,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 16590,
      "tokens_per_second_per_gpu": 374.55
    },
    {
      "epoch": 1.6502957697469802,
      "grad_norm": 0.414070725440979,
      "learning_rate": 1.4859431793838995e-05,
      "loss": 0.131,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 16600,
      "tokens_per_second_per_gpu": 396.68
    },
    {
      "epoch": 1.6512899537704429,
      "grad_norm": 0.6677629947662354,
      "learning_rate": 1.477721717640248e-05,
      "loss": 0.1532,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 16610,
      "tokens_per_second_per_gpu": 419.18
    },
    {
      "epoch": 1.6522841377939057,
      "grad_norm": 0.22176572680473328,
      "learning_rate": 1.4695212480692277e-05,
      "loss": 0.1519,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 16620,
      "tokens_per_second_per_gpu": 401.78
    },
    {
      "epoch": 1.6532783218173686,
      "grad_norm": 0.6775956749916077,
      "learning_rate": 1.4613417908703342e-05,
      "loss": 0.1728,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 16630,
      "tokens_per_second_per_gpu": 349.56
    },
    {
      "epoch": 1.6542725058408312,
      "grad_norm": 0.4293968081474304,
      "learning_rate": 1.4531833661912942e-05,
      "loss": 0.1076,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 16640,
      "tokens_per_second_per_gpu": 314.88
    },
    {
      "epoch": 1.6552666898642938,
      "grad_norm": 0.5610436797142029,
      "learning_rate": 1.445045994128037e-05,
      "loss": 0.1682,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 16650,
      "tokens_per_second_per_gpu": 363.34
    },
    {
      "epoch": 1.6562608738877567,
      "grad_norm": 0.616041362285614,
      "learning_rate": 1.4369296947246236e-05,
      "loss": 0.1587,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 16660,
      "tokens_per_second_per_gpu": 362.03
    },
    {
      "epoch": 1.6572550579112195,
      "grad_norm": 0.3607831299304962,
      "learning_rate": 1.4288344879732185e-05,
      "loss": 0.1889,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.86,
      "memory/max_allocated (GiB)": 18.86,
      "step": 16670,
      "tokens_per_second_per_gpu": 371.37
    },
    {
      "epoch": 1.6582492419346821,
      "grad_norm": 0.2895027697086334,
      "learning_rate": 1.420760393814028e-05,
      "loss": 0.1525,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 16680,
      "tokens_per_second_per_gpu": 375.97
    },
    {
      "epoch": 1.6592434259581448,
      "grad_norm": 0.35571709275245667,
      "learning_rate": 1.4127074321352517e-05,
      "loss": 0.1653,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 16690,
      "tokens_per_second_per_gpu": 359.31
    },
    {
      "epoch": 1.6602376099816076,
      "grad_norm": 0.6197568774223328,
      "learning_rate": 1.404675622773034e-05,
      "loss": 0.1795,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 16700,
      "tokens_per_second_per_gpu": 320.06
    },
    {
      "epoch": 1.6612317940050705,
      "grad_norm": 0.5286340713500977,
      "learning_rate": 1.3966649855114211e-05,
      "loss": 0.1592,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 16710,
      "tokens_per_second_per_gpu": 333.26
    },
    {
      "epoch": 1.6622259780285331,
      "grad_norm": 0.4165874719619751,
      "learning_rate": 1.3886755400823071e-05,
      "loss": 0.1282,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 16720,
      "tokens_per_second_per_gpu": 383.56
    },
    {
      "epoch": 1.6632201620519957,
      "grad_norm": 0.4585205316543579,
      "learning_rate": 1.3807073061653809e-05,
      "loss": 0.2177,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 16730,
      "tokens_per_second_per_gpu": 423.57
    },
    {
      "epoch": 1.6642143460754586,
      "grad_norm": 0.4411196708679199,
      "learning_rate": 1.372760303388091e-05,
      "loss": 0.104,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 16740,
      "tokens_per_second_per_gpu": 305.5
    },
    {
      "epoch": 1.6652085300989214,
      "grad_norm": 0.5241126418113708,
      "learning_rate": 1.36483455132558e-05,
      "loss": 0.1809,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 16750,
      "tokens_per_second_per_gpu": 356.92
    },
    {
      "epoch": 1.666202714122384,
      "grad_norm": 0.6349103450775146,
      "learning_rate": 1.3569300695006548e-05,
      "loss": 0.1311,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 16760,
      "tokens_per_second_per_gpu": 388.79
    },
    {
      "epoch": 1.6671968981458467,
      "grad_norm": 0.46360382437705994,
      "learning_rate": 1.3490468773837217e-05,
      "loss": 0.1204,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 16770,
      "tokens_per_second_per_gpu": 317.52
    },
    {
      "epoch": 1.6681910821693096,
      "grad_norm": 0.49904516339302063,
      "learning_rate": 1.3411849943927513e-05,
      "loss": 0.1616,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 16780,
      "tokens_per_second_per_gpu": 355.09
    },
    {
      "epoch": 1.6691852661927724,
      "grad_norm": 0.19544100761413574,
      "learning_rate": 1.3333444398932205e-05,
      "loss": 0.1091,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 16790,
      "tokens_per_second_per_gpu": 289.19
    },
    {
      "epoch": 1.670179450216235,
      "grad_norm": 0.31488218903541565,
      "learning_rate": 1.325525233198076e-05,
      "loss": 0.1004,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 16800,
      "tokens_per_second_per_gpu": 316.73
    },
    {
      "epoch": 1.6711736342396977,
      "grad_norm": 0.5296607613563538,
      "learning_rate": 1.3177273935676715e-05,
      "loss": 0.1112,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 16810,
      "tokens_per_second_per_gpu": 343.62
    },
    {
      "epoch": 1.6721678182631605,
      "grad_norm": 0.4741019308567047,
      "learning_rate": 1.3099509402097377e-05,
      "loss": 0.1181,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.11,
      "memory/max_allocated (GiB)": 19.11,
      "step": 16820,
      "tokens_per_second_per_gpu": 337.04
    },
    {
      "epoch": 1.6731620022866234,
      "grad_norm": 0.545759916305542,
      "learning_rate": 1.3021958922793209e-05,
      "loss": 0.1247,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 16830,
      "tokens_per_second_per_gpu": 347.83
    },
    {
      "epoch": 1.674156186310086,
      "grad_norm": 0.5977072715759277,
      "learning_rate": 1.2944622688787445e-05,
      "loss": 0.1867,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 16840,
      "tokens_per_second_per_gpu": 371.62
    },
    {
      "epoch": 1.6751503703335486,
      "grad_norm": 0.6563873887062073,
      "learning_rate": 1.2867500890575601e-05,
      "loss": 0.2069,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 16850,
      "tokens_per_second_per_gpu": 339.57
    },
    {
      "epoch": 1.6761445543570115,
      "grad_norm": 0.5459251999855042,
      "learning_rate": 1.279059371812491e-05,
      "loss": 0.1413,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 16860,
      "tokens_per_second_per_gpu": 355.16
    },
    {
      "epoch": 1.6771387383804743,
      "grad_norm": 0.5088791847229004,
      "learning_rate": 1.2713901360874037e-05,
      "loss": 0.1136,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 16870,
      "tokens_per_second_per_gpu": 327.52
    },
    {
      "epoch": 1.678132922403937,
      "grad_norm": 0.6554404497146606,
      "learning_rate": 1.2637424007732434e-05,
      "loss": 0.1708,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 16880,
      "tokens_per_second_per_gpu": 335.39
    },
    {
      "epoch": 1.6791271064273996,
      "grad_norm": 0.49274370074272156,
      "learning_rate": 1.2561161847080028e-05,
      "loss": 0.1766,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 16890,
      "tokens_per_second_per_gpu": 382.39
    },
    {
      "epoch": 1.6801212904508624,
      "grad_norm": 0.6049704551696777,
      "learning_rate": 1.2485115066766584e-05,
      "loss": 0.1887,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 16900,
      "tokens_per_second_per_gpu": 395.86
    },
    {
      "epoch": 1.6811154744743253,
      "grad_norm": 0.45951762795448303,
      "learning_rate": 1.2409283854111442e-05,
      "loss": 0.1627,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 16910,
      "tokens_per_second_per_gpu": 374.46
    },
    {
      "epoch": 1.682109658497788,
      "grad_norm": 0.31577378511428833,
      "learning_rate": 1.2333668395902875e-05,
      "loss": 0.1635,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 16920,
      "tokens_per_second_per_gpu": 302.79
    },
    {
      "epoch": 1.6831038425212506,
      "grad_norm": 0.3830484449863434,
      "learning_rate": 1.225826887839776e-05,
      "loss": 0.0991,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 16930,
      "tokens_per_second_per_gpu": 384.62
    },
    {
      "epoch": 1.6840980265447134,
      "grad_norm": 0.4875403642654419,
      "learning_rate": 1.2183085487321022e-05,
      "loss": 0.1612,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 16940,
      "tokens_per_second_per_gpu": 393.48
    },
    {
      "epoch": 1.6850922105681763,
      "grad_norm": 0.4757268726825714,
      "learning_rate": 1.2108118407865254e-05,
      "loss": 0.1823,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.61,
      "memory/max_allocated (GiB)": 20.61,
      "step": 16950,
      "tokens_per_second_per_gpu": 357.56
    },
    {
      "epoch": 1.686086394591639,
      "grad_norm": 0.3833984136581421,
      "learning_rate": 1.2033367824690223e-05,
      "loss": 0.1261,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 16960,
      "tokens_per_second_per_gpu": 431.85
    },
    {
      "epoch": 1.6870805786151015,
      "grad_norm": 0.35014206171035767,
      "learning_rate": 1.1958833921922418e-05,
      "loss": 0.1545,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 16970,
      "tokens_per_second_per_gpu": 309.84
    },
    {
      "epoch": 1.6880747626385644,
      "grad_norm": 0.5870986580848694,
      "learning_rate": 1.1884516883154606e-05,
      "loss": 0.1116,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 16980,
      "tokens_per_second_per_gpu": 330.54
    },
    {
      "epoch": 1.6890689466620272,
      "grad_norm": 0.3304491639137268,
      "learning_rate": 1.1810416891445319e-05,
      "loss": 0.1723,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 16990,
      "tokens_per_second_per_gpu": 352.93
    },
    {
      "epoch": 1.6900631306854899,
      "grad_norm": 0.295502632856369,
      "learning_rate": 1.1736534129318532e-05,
      "loss": 0.1596,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 17000,
      "tokens_per_second_per_gpu": 365.23
    },
    {
      "epoch": 1.6910573147089525,
      "grad_norm": 0.48762744665145874,
      "learning_rate": 1.1662868778763092e-05,
      "loss": 0.1732,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 17010,
      "tokens_per_second_per_gpu": 390.51
    },
    {
      "epoch": 1.6920514987324153,
      "grad_norm": 0.4020582437515259,
      "learning_rate": 1.1589421021232338e-05,
      "loss": 0.1351,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17020,
      "tokens_per_second_per_gpu": 345.17
    },
    {
      "epoch": 1.6930456827558782,
      "grad_norm": 0.5929551124572754,
      "learning_rate": 1.1516191037643598e-05,
      "loss": 0.1425,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 17030,
      "tokens_per_second_per_gpu": 344.59
    },
    {
      "epoch": 1.6940398667793408,
      "grad_norm": 0.47912150621414185,
      "learning_rate": 1.1443179008377825e-05,
      "loss": 0.1139,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17040,
      "tokens_per_second_per_gpu": 351.04
    },
    {
      "epoch": 1.6950340508028034,
      "grad_norm": 0.3503008782863617,
      "learning_rate": 1.1370385113279047e-05,
      "loss": 0.1163,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 17050,
      "tokens_per_second_per_gpu": 329.13
    },
    {
      "epoch": 1.6960282348262663,
      "grad_norm": 0.3980488181114197,
      "learning_rate": 1.1297809531654046e-05,
      "loss": 0.1619,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.78,
      "memory/max_allocated (GiB)": 20.78,
      "step": 17060,
      "tokens_per_second_per_gpu": 381.97
    },
    {
      "epoch": 1.6970224188497292,
      "grad_norm": 0.5639301538467407,
      "learning_rate": 1.1225452442271789e-05,
      "loss": 0.1308,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17070,
      "tokens_per_second_per_gpu": 352.08
    },
    {
      "epoch": 1.6980166028731918,
      "grad_norm": 0.20409265160560608,
      "learning_rate": 1.1153314023363126e-05,
      "loss": 0.1434,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 17080,
      "tokens_per_second_per_gpu": 318.47
    },
    {
      "epoch": 1.6990107868966544,
      "grad_norm": 0.3582451343536377,
      "learning_rate": 1.1081394452620164e-05,
      "loss": 0.1646,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 17090,
      "tokens_per_second_per_gpu": 296.97
    },
    {
      "epoch": 1.7000049709201173,
      "grad_norm": 0.396576851606369,
      "learning_rate": 1.100969390719605e-05,
      "loss": 0.1628,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 17100,
      "tokens_per_second_per_gpu": 324.04
    },
    {
      "epoch": 1.7009991549435801,
      "grad_norm": 0.53066086769104,
      "learning_rate": 1.0938212563704364e-05,
      "loss": 0.1304,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 17110,
      "tokens_per_second_per_gpu": 414.9
    },
    {
      "epoch": 1.7019933389670427,
      "grad_norm": 0.4172995388507843,
      "learning_rate": 1.0866950598218772e-05,
      "loss": 0.1336,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 17120,
      "tokens_per_second_per_gpu": 319.1
    },
    {
      "epoch": 1.7029875229905056,
      "grad_norm": 0.8753495216369629,
      "learning_rate": 1.0795908186272585e-05,
      "loss": 0.1354,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 17130,
      "tokens_per_second_per_gpu": 303.94
    },
    {
      "epoch": 1.7039817070139684,
      "grad_norm": 0.5303057432174683,
      "learning_rate": 1.0725085502858223e-05,
      "loss": 0.1675,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 17140,
      "tokens_per_second_per_gpu": 418.9
    },
    {
      "epoch": 1.704975891037431,
      "grad_norm": 0.3423836827278137,
      "learning_rate": 1.0654482722426984e-05,
      "loss": 0.1446,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17150,
      "tokens_per_second_per_gpu": 376.63
    },
    {
      "epoch": 1.7059700750608937,
      "grad_norm": 0.46940287947654724,
      "learning_rate": 1.0584100018888376e-05,
      "loss": 0.173,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 17160,
      "tokens_per_second_per_gpu": 398.16
    },
    {
      "epoch": 1.7069642590843566,
      "grad_norm": 0.5457295775413513,
      "learning_rate": 1.0513937565609922e-05,
      "loss": 0.1594,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 17170,
      "tokens_per_second_per_gpu": 334.42
    },
    {
      "epoch": 1.7079584431078194,
      "grad_norm": 0.5194307565689087,
      "learning_rate": 1.044399553541653e-05,
      "loss": 0.0897,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17180,
      "tokens_per_second_per_gpu": 313.63
    },
    {
      "epoch": 1.708952627131282,
      "grad_norm": 0.6305141448974609,
      "learning_rate": 1.0374274100590254e-05,
      "loss": 0.157,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 17190,
      "tokens_per_second_per_gpu": 356.77
    },
    {
      "epoch": 1.7099468111547447,
      "grad_norm": 0.7084689140319824,
      "learning_rate": 1.0304773432869675e-05,
      "loss": 0.2061,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 17200,
      "tokens_per_second_per_gpu": 396.04
    },
    {
      "epoch": 1.7109409951782075,
      "grad_norm": 0.5986453294754028,
      "learning_rate": 1.0235493703449673e-05,
      "loss": 0.1498,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 17.01,
      "memory/max_allocated (GiB)": 17.01,
      "step": 17210,
      "tokens_per_second_per_gpu": 333.56
    },
    {
      "epoch": 1.7119351792016704,
      "grad_norm": 0.4625334143638611,
      "learning_rate": 1.0166435082980818e-05,
      "loss": 0.1473,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 17220,
      "tokens_per_second_per_gpu": 344.72
    },
    {
      "epoch": 1.712929363225133,
      "grad_norm": 0.3249736726284027,
      "learning_rate": 1.0097597741569109e-05,
      "loss": 0.1247,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 17230,
      "tokens_per_second_per_gpu": 387.91
    },
    {
      "epoch": 1.7139235472485956,
      "grad_norm": 0.395663321018219,
      "learning_rate": 1.0028981848775499e-05,
      "loss": 0.1227,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17240,
      "tokens_per_second_per_gpu": 400.44
    },
    {
      "epoch": 1.7149177312720585,
      "grad_norm": 0.3278411030769348,
      "learning_rate": 9.960587573615376e-06,
      "loss": 0.1161,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 17250,
      "tokens_per_second_per_gpu": 419.85
    },
    {
      "epoch": 1.7159119152955213,
      "grad_norm": 0.27274656295776367,
      "learning_rate": 9.892415084558315e-06,
      "loss": 0.1541,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 17260,
      "tokens_per_second_per_gpu": 357.93
    },
    {
      "epoch": 1.716906099318984,
      "grad_norm": 0.6081755757331848,
      "learning_rate": 9.82446454952759e-06,
      "loss": 0.1644,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17270,
      "tokens_per_second_per_gpu": 381.9
    },
    {
      "epoch": 1.7179002833424466,
      "grad_norm": 0.5921940207481384,
      "learning_rate": 9.756736135899724e-06,
      "loss": 0.1686,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 17280,
      "tokens_per_second_per_gpu": 422.69
    },
    {
      "epoch": 1.7188944673659095,
      "grad_norm": 0.43674856424331665,
      "learning_rate": 9.68923001050408e-06,
      "loss": 0.1546,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 17290,
      "tokens_per_second_per_gpu": 370.0
    },
    {
      "epoch": 1.7198886513893723,
      "grad_norm": 0.39371129870414734,
      "learning_rate": 9.621946339622567e-06,
      "loss": 0.1627,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 17300,
      "tokens_per_second_per_gpu": 404.49
    },
    {
      "epoch": 1.720882835412835,
      "grad_norm": 0.6137861609458923,
      "learning_rate": 9.554885288989035e-06,
      "loss": 0.1846,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 17310,
      "tokens_per_second_per_gpu": 400.34
    },
    {
      "epoch": 1.7218770194362976,
      "grad_norm": 0.43737614154815674,
      "learning_rate": 9.488047023789059e-06,
      "loss": 0.1593,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17320,
      "tokens_per_second_per_gpu": 365.64
    },
    {
      "epoch": 1.7228712034597604,
      "grad_norm": 0.20106801390647888,
      "learning_rate": 9.42143170865939e-06,
      "loss": 0.1476,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 17330,
      "tokens_per_second_per_gpu": 366.76
    },
    {
      "epoch": 1.7238653874832233,
      "grad_norm": 0.652117133140564,
      "learning_rate": 9.355039507687657e-06,
      "loss": 0.191,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 17340,
      "tokens_per_second_per_gpu": 433.9
    },
    {
      "epoch": 1.724859571506686,
      "grad_norm": 0.39825040102005005,
      "learning_rate": 9.288870584411835e-06,
      "loss": 0.1297,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 17350,
      "tokens_per_second_per_gpu": 313.48
    },
    {
      "epoch": 1.7258537555301485,
      "grad_norm": 0.20952925086021423,
      "learning_rate": 9.222925101820012e-06,
      "loss": 0.1617,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17360,
      "tokens_per_second_per_gpu": 382.53
    },
    {
      "epoch": 1.7268479395536114,
      "grad_norm": 0.6603105068206787,
      "learning_rate": 9.157203222349853e-06,
      "loss": 0.2117,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 17370,
      "tokens_per_second_per_gpu": 424.27
    },
    {
      "epoch": 1.7278421235770742,
      "grad_norm": 0.5097489356994629,
      "learning_rate": 9.091705107888204e-06,
      "loss": 0.181,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17380,
      "tokens_per_second_per_gpu": 487.99
    },
    {
      "epoch": 1.7288363076005369,
      "grad_norm": 0.674410879611969,
      "learning_rate": 9.026430919770767e-06,
      "loss": 0.1491,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17390,
      "tokens_per_second_per_gpu": 354.95
    },
    {
      "epoch": 1.7298304916239995,
      "grad_norm": 0.370148241519928,
      "learning_rate": 8.961380818781695e-06,
      "loss": 0.1672,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.54,
      "memory/max_allocated (GiB)": 21.54,
      "step": 17400,
      "tokens_per_second_per_gpu": 397.8
    },
    {
      "epoch": 1.7308246756474623,
      "grad_norm": 0.5135669112205505,
      "learning_rate": 8.896554965153126e-06,
      "loss": 0.1515,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 17410,
      "tokens_per_second_per_gpu": 327.7
    },
    {
      "epoch": 1.7318188596709252,
      "grad_norm": 0.5589896440505981,
      "learning_rate": 8.831953518564816e-06,
      "loss": 0.0882,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 17420,
      "tokens_per_second_per_gpu": 296.23
    },
    {
      "epoch": 1.7328130436943878,
      "grad_norm": 0.6967418789863586,
      "learning_rate": 8.767576638143804e-06,
      "loss": 0.1411,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 17430,
      "tokens_per_second_per_gpu": 365.21
    },
    {
      "epoch": 1.7338072277178505,
      "grad_norm": 0.7863910794258118,
      "learning_rate": 8.70342448246394e-06,
      "loss": 0.1521,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 17440,
      "tokens_per_second_per_gpu": 379.01
    },
    {
      "epoch": 1.7348014117413133,
      "grad_norm": 0.49855732917785645,
      "learning_rate": 8.639497209545556e-06,
      "loss": 0.1717,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 17450,
      "tokens_per_second_per_gpu": 351.33
    },
    {
      "epoch": 1.7357955957647762,
      "grad_norm": 0.5448564291000366,
      "learning_rate": 8.57579497685501e-06,
      "loss": 0.1185,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 17460,
      "tokens_per_second_per_gpu": 338.51
    },
    {
      "epoch": 1.7367897797882388,
      "grad_norm": 0.9186548590660095,
      "learning_rate": 8.512317941304404e-06,
      "loss": 0.1824,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 17470,
      "tokens_per_second_per_gpu": 397.66
    },
    {
      "epoch": 1.7377839638117014,
      "grad_norm": 0.6779747605323792,
      "learning_rate": 8.44906625925106e-06,
      "loss": 0.1354,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 17480,
      "tokens_per_second_per_gpu": 348.68
    },
    {
      "epoch": 1.7387781478351643,
      "grad_norm": 0.6516660451889038,
      "learning_rate": 8.386040086497238e-06,
      "loss": 0.1223,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 17490,
      "tokens_per_second_per_gpu": 335.11
    },
    {
      "epoch": 1.7397723318586271,
      "grad_norm": 0.4572422504425049,
      "learning_rate": 8.323239578289754e-06,
      "loss": 0.1306,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17500,
      "tokens_per_second_per_gpu": 387.29
    },
    {
      "epoch": 1.7407665158820897,
      "grad_norm": 0.5288609862327576,
      "learning_rate": 8.260664889319502e-06,
      "loss": 0.1567,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 17510,
      "tokens_per_second_per_gpu": 339.2
    },
    {
      "epoch": 1.7417606999055524,
      "grad_norm": 0.6885465383529663,
      "learning_rate": 8.198316173721199e-06,
      "loss": 0.1687,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 17520,
      "tokens_per_second_per_gpu": 350.42
    },
    {
      "epoch": 1.7427548839290152,
      "grad_norm": 0.4652736485004425,
      "learning_rate": 8.136193585072871e-06,
      "loss": 0.1494,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 17530,
      "tokens_per_second_per_gpu": 329.48
    },
    {
      "epoch": 1.743749067952478,
      "grad_norm": 0.2998766303062439,
      "learning_rate": 8.074297276395592e-06,
      "loss": 0.1341,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17540,
      "tokens_per_second_per_gpu": 343.21
    },
    {
      "epoch": 1.7447432519759407,
      "grad_norm": 0.5339378118515015,
      "learning_rate": 8.012627400153073e-06,
      "loss": 0.1889,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 17550,
      "tokens_per_second_per_gpu": 366.33
    },
    {
      "epoch": 1.7457374359994033,
      "grad_norm": 0.2314610630273819,
      "learning_rate": 7.951184108251242e-06,
      "loss": 0.1306,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 17560,
      "tokens_per_second_per_gpu": 265.9
    },
    {
      "epoch": 1.7467316200228662,
      "grad_norm": 0.33561253547668457,
      "learning_rate": 7.889967552037913e-06,
      "loss": 0.1782,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.69,
      "memory/max_allocated (GiB)": 19.69,
      "step": 17570,
      "tokens_per_second_per_gpu": 361.12
    },
    {
      "epoch": 1.747725804046329,
      "grad_norm": 0.4189736843109131,
      "learning_rate": 7.828977882302413e-06,
      "loss": 0.1693,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 17580,
      "tokens_per_second_per_gpu": 394.43
    },
    {
      "epoch": 1.7487199880697917,
      "grad_norm": 0.3725840747356415,
      "learning_rate": 7.768215249275168e-06,
      "loss": 0.1826,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17590,
      "tokens_per_second_per_gpu": 345.73
    },
    {
      "epoch": 1.7497141720932543,
      "grad_norm": 0.40016722679138184,
      "learning_rate": 7.707679802627399e-06,
      "loss": 0.1428,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 17600,
      "tokens_per_second_per_gpu": 304.16
    },
    {
      "epoch": 1.7507083561167172,
      "grad_norm": 0.274614542722702,
      "learning_rate": 7.647371691470706e-06,
      "loss": 0.0966,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.68,
      "memory/max_allocated (GiB)": 19.68,
      "step": 17610,
      "tokens_per_second_per_gpu": 348.33
    },
    {
      "epoch": 1.75170254014018,
      "grad_norm": 0.5691856145858765,
      "learning_rate": 7.587291064356716e-06,
      "loss": 0.1495,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 17620,
      "tokens_per_second_per_gpu": 285.38
    },
    {
      "epoch": 1.7526967241636426,
      "grad_norm": 0.529352068901062,
      "learning_rate": 7.5274380692766825e-06,
      "loss": 0.1387,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 17630,
      "tokens_per_second_per_gpu": 328.86
    },
    {
      "epoch": 1.7536909081871053,
      "grad_norm": 0.4417911171913147,
      "learning_rate": 7.467812853661216e-06,
      "loss": 0.136,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 17640,
      "tokens_per_second_per_gpu": 350.36
    },
    {
      "epoch": 1.7546850922105683,
      "grad_norm": 0.45288828015327454,
      "learning_rate": 7.4084155643798335e-06,
      "loss": 0.1326,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 17650,
      "tokens_per_second_per_gpu": 329.09
    },
    {
      "epoch": 1.755679276234031,
      "grad_norm": 0.4209536015987396,
      "learning_rate": 7.349246347740568e-06,
      "loss": 0.179,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 17660,
      "tokens_per_second_per_gpu": 426.76
    },
    {
      "epoch": 1.7566734602574936,
      "grad_norm": 0.49393320083618164,
      "learning_rate": 7.290305349489734e-06,
      "loss": 0.2134,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 17670,
      "tokens_per_second_per_gpu": 411.57
    },
    {
      "epoch": 1.7576676442809565,
      "grad_norm": 0.6331411600112915,
      "learning_rate": 7.2315927148114635e-06,
      "loss": 0.1676,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 17680,
      "tokens_per_second_per_gpu": 306.06
    },
    {
      "epoch": 1.7586618283044193,
      "grad_norm": 0.383672297000885,
      "learning_rate": 7.173108588327415e-06,
      "loss": 0.14,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 17690,
      "tokens_per_second_per_gpu": 342.56
    },
    {
      "epoch": 1.759656012327882,
      "grad_norm": 0.4192826449871063,
      "learning_rate": 7.1148531140962986e-06,
      "loss": 0.1358,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.85,
      "memory/max_allocated (GiB)": 18.85,
      "step": 17700,
      "tokens_per_second_per_gpu": 342.72
    },
    {
      "epoch": 1.7606501963513446,
      "grad_norm": 0.3609708845615387,
      "learning_rate": 7.056826435613706e-06,
      "loss": 0.1483,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 17710,
      "tokens_per_second_per_gpu": 378.78
    },
    {
      "epoch": 1.7616443803748074,
      "grad_norm": 0.535681962966919,
      "learning_rate": 6.999028695811572e-06,
      "loss": 0.1457,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 17720,
      "tokens_per_second_per_gpu": 315.36
    },
    {
      "epoch": 1.7626385643982703,
      "grad_norm": 0.7409544587135315,
      "learning_rate": 6.941460037057979e-06,
      "loss": 0.187,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 17730,
      "tokens_per_second_per_gpu": 339.48
    },
    {
      "epoch": 1.763632748421733,
      "grad_norm": 0.4647164046764374,
      "learning_rate": 6.8841206011566625e-06,
      "loss": 0.1178,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 17740,
      "tokens_per_second_per_gpu": 318.67
    },
    {
      "epoch": 1.7646269324451955,
      "grad_norm": 0.37413302063941956,
      "learning_rate": 6.827010529346822e-06,
      "loss": 0.1156,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 17750,
      "tokens_per_second_per_gpu": 361.82
    },
    {
      "epoch": 1.7656211164686584,
      "grad_norm": 0.7163899540901184,
      "learning_rate": 6.7701299623025846e-06,
      "loss": 0.1514,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17760,
      "tokens_per_second_per_gpu": 346.76
    },
    {
      "epoch": 1.7666153004921212,
      "grad_norm": 0.47301217913627625,
      "learning_rate": 6.713479040132841e-06,
      "loss": 0.1582,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17770,
      "tokens_per_second_per_gpu": 346.61
    },
    {
      "epoch": 1.7676094845155839,
      "grad_norm": 0.4980860948562622,
      "learning_rate": 6.657057902380792e-06,
      "loss": 0.1119,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 17780,
      "tokens_per_second_per_gpu": 302.2
    },
    {
      "epoch": 1.7686036685390465,
      "grad_norm": 0.6229726076126099,
      "learning_rate": 6.600866688023588e-06,
      "loss": 0.1737,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17790,
      "tokens_per_second_per_gpu": 408.97
    },
    {
      "epoch": 1.7695978525625093,
      "grad_norm": 0.5228248834609985,
      "learning_rate": 6.5449055354721125e-06,
      "loss": 0.1809,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 17800,
      "tokens_per_second_per_gpu": 379.16
    },
    {
      "epoch": 1.7705920365859722,
      "grad_norm": 0.4032490849494934,
      "learning_rate": 6.489174582570467e-06,
      "loss": 0.147,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 17810,
      "tokens_per_second_per_gpu": 378.57
    },
    {
      "epoch": 1.7715862206094348,
      "grad_norm": 0.44193387031555176,
      "learning_rate": 6.433673966595788e-06,
      "loss": 0.1556,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 17820,
      "tokens_per_second_per_gpu": 330.37
    },
    {
      "epoch": 1.7725804046328975,
      "grad_norm": 0.41534683108329773,
      "learning_rate": 6.3784038242578285e-06,
      "loss": 0.1261,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 17830,
      "tokens_per_second_per_gpu": 317.12
    },
    {
      "epoch": 1.7735745886563603,
      "grad_norm": 0.7784578800201416,
      "learning_rate": 6.323364291698642e-06,
      "loss": 0.1922,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 17840,
      "tokens_per_second_per_gpu": 342.84
    },
    {
      "epoch": 1.7745687726798232,
      "grad_norm": 0.7484685778617859,
      "learning_rate": 6.2685555044921905e-06,
      "loss": 0.1375,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17850,
      "tokens_per_second_per_gpu": 351.93
    },
    {
      "epoch": 1.7755629567032858,
      "grad_norm": 0.5791344046592712,
      "learning_rate": 6.213977597644138e-06,
      "loss": 0.1087,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17860,
      "tokens_per_second_per_gpu": 337.95
    },
    {
      "epoch": 1.7765571407267484,
      "grad_norm": 0.22857262194156647,
      "learning_rate": 6.159630705591379e-06,
      "loss": 0.1257,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 17870,
      "tokens_per_second_per_gpu": 365.98
    },
    {
      "epoch": 1.7775513247502113,
      "grad_norm": 0.43469542264938354,
      "learning_rate": 6.10551496220183e-06,
      "loss": 0.1652,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 17880,
      "tokens_per_second_per_gpu": 371.12
    },
    {
      "epoch": 1.7785455087736741,
      "grad_norm": 0.7687386274337769,
      "learning_rate": 6.0516305007739525e-06,
      "loss": 0.1787,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.72,
      "memory/max_allocated (GiB)": 20.72,
      "step": 17890,
      "tokens_per_second_per_gpu": 369.52
    },
    {
      "epoch": 1.7795396927971368,
      "grad_norm": 0.3083856999874115,
      "learning_rate": 5.997977454036608e-06,
      "loss": 0.1817,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 17900,
      "tokens_per_second_per_gpu": 342.94
    },
    {
      "epoch": 1.7805338768205994,
      "grad_norm": 0.3785597085952759,
      "learning_rate": 5.944555954148579e-06,
      "loss": 0.1563,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 17910,
      "tokens_per_second_per_gpu": 396.7
    },
    {
      "epoch": 1.7815280608440622,
      "grad_norm": 0.5601521730422974,
      "learning_rate": 5.891366132698295e-06,
      "loss": 0.1876,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 17920,
      "tokens_per_second_per_gpu": 357.52
    },
    {
      "epoch": 1.782522244867525,
      "grad_norm": 0.6354232430458069,
      "learning_rate": 5.838408120703554e-06,
      "loss": 0.1495,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17930,
      "tokens_per_second_per_gpu": 367.19
    },
    {
      "epoch": 1.7835164288909877,
      "grad_norm": 0.6532102227210999,
      "learning_rate": 5.785682048611097e-06,
      "loss": 0.172,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 17940,
      "tokens_per_second_per_gpu": 391.52
    },
    {
      "epoch": 1.7845106129144503,
      "grad_norm": 0.4626152515411377,
      "learning_rate": 5.733188046296423e-06,
      "loss": 0.1516,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 17950,
      "tokens_per_second_per_gpu": 343.51
    },
    {
      "epoch": 1.7855047969379132,
      "grad_norm": 0.3500106930732727,
      "learning_rate": 5.680926243063322e-06,
      "loss": 0.1487,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 17960,
      "tokens_per_second_per_gpu": 395.4
    },
    {
      "epoch": 1.786498980961376,
      "grad_norm": 0.5653957724571228,
      "learning_rate": 5.628896767643677e-06,
      "loss": 0.1401,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 17970,
      "tokens_per_second_per_gpu": 372.67
    },
    {
      "epoch": 1.7874931649848387,
      "grad_norm": 0.4583251476287842,
      "learning_rate": 5.577099748197079e-06,
      "loss": 0.1497,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 17980,
      "tokens_per_second_per_gpu": 319.79
    },
    {
      "epoch": 1.7884873490083013,
      "grad_norm": 0.7779932022094727,
      "learning_rate": 5.525535312310559e-06,
      "loss": 0.1878,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.23,
      "memory/max_allocated (GiB)": 19.23,
      "step": 17990,
      "tokens_per_second_per_gpu": 330.6
    },
    {
      "epoch": 1.7894815330317642,
      "grad_norm": 0.5615857243537903,
      "learning_rate": 5.4742035869981726e-06,
      "loss": 0.0951,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 18000,
      "tokens_per_second_per_gpu": 310.54
    },
    {
      "epoch": 1.790475717055227,
      "grad_norm": 0.6377988457679749,
      "learning_rate": 5.423104698700853e-06,
      "loss": 0.15,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 18010,
      "tokens_per_second_per_gpu": 340.19
    },
    {
      "epoch": 1.7914699010786896,
      "grad_norm": 0.6940702199935913,
      "learning_rate": 5.372238773285931e-06,
      "loss": 0.1748,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 18020,
      "tokens_per_second_per_gpu": 323.58
    },
    {
      "epoch": 1.7924640851021523,
      "grad_norm": 0.558322548866272,
      "learning_rate": 5.321605936046947e-06,
      "loss": 0.1468,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 18030,
      "tokens_per_second_per_gpu": 305.22
    },
    {
      "epoch": 1.7934582691256151,
      "grad_norm": 0.3536572754383087,
      "learning_rate": 5.271206311703281e-06,
      "loss": 0.1207,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.19,
      "memory/max_allocated (GiB)": 19.19,
      "step": 18040,
      "tokens_per_second_per_gpu": 336.32
    },
    {
      "epoch": 1.794452453149078,
      "grad_norm": 0.5200352668762207,
      "learning_rate": 5.221040024399848e-06,
      "loss": 0.1664,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 18050,
      "tokens_per_second_per_gpu": 281.71
    },
    {
      "epoch": 1.7954466371725406,
      "grad_norm": 0.3343510329723358,
      "learning_rate": 5.171107197706837e-06,
      "loss": 0.1725,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 18060,
      "tokens_per_second_per_gpu": 345.94
    },
    {
      "epoch": 1.7964408211960032,
      "grad_norm": 0.4875069856643677,
      "learning_rate": 5.121407954619339e-06,
      "loss": 0.1521,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.81,
      "memory/max_allocated (GiB)": 19.81,
      "step": 18070,
      "tokens_per_second_per_gpu": 368.58
    },
    {
      "epoch": 1.797435005219466,
      "grad_norm": 0.524474024772644,
      "learning_rate": 5.071942417557096e-06,
      "loss": 0.1604,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 18080,
      "tokens_per_second_per_gpu": 386.02
    },
    {
      "epoch": 1.798429189242929,
      "grad_norm": 0.3819403052330017,
      "learning_rate": 5.0227107083641756e-06,
      "loss": 0.1682,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 18090,
      "tokens_per_second_per_gpu": 397.67
    },
    {
      "epoch": 1.7994233732663916,
      "grad_norm": 0.4870153069496155,
      "learning_rate": 4.9737129483086845e-06,
      "loss": 0.1726,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 18100,
      "tokens_per_second_per_gpu": 409.71
    },
    {
      "epoch": 1.8004175572898542,
      "grad_norm": 0.48295846581459045,
      "learning_rate": 4.924949258082468e-06,
      "loss": 0.1807,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 18110,
      "tokens_per_second_per_gpu": 386.2
    },
    {
      "epoch": 1.801411741313317,
      "grad_norm": 0.43393832445144653,
      "learning_rate": 4.8764197578008095e-06,
      "loss": 0.1368,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 18120,
      "tokens_per_second_per_gpu": 348.61
    },
    {
      "epoch": 1.80240592533678,
      "grad_norm": 0.4767451584339142,
      "learning_rate": 4.828124567002113e-06,
      "loss": 0.141,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 18130,
      "tokens_per_second_per_gpu": 333.81
    },
    {
      "epoch": 1.8034001093602425,
      "grad_norm": 0.4560125768184662,
      "learning_rate": 4.780063804647639e-06,
      "loss": 0.1356,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 18140,
      "tokens_per_second_per_gpu": 331.92
    },
    {
      "epoch": 1.8043942933837052,
      "grad_norm": 0.23658177256584167,
      "learning_rate": 4.732237589121202e-06,
      "loss": 0.1273,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 18150,
      "tokens_per_second_per_gpu": 289.7
    },
    {
      "epoch": 1.805388477407168,
      "grad_norm": 0.310188353061676,
      "learning_rate": 4.684646038228891e-06,
      "loss": 0.1614,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 18160,
      "tokens_per_second_per_gpu": 406.62
    },
    {
      "epoch": 1.8063826614306309,
      "grad_norm": 0.49459826946258545,
      "learning_rate": 4.6372892691987525e-06,
      "loss": 0.1765,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.67,
      "memory/max_allocated (GiB)": 19.67,
      "step": 18170,
      "tokens_per_second_per_gpu": 340.47
    },
    {
      "epoch": 1.8073768454540935,
      "grad_norm": 0.6684255599975586,
      "learning_rate": 4.5901673986804896e-06,
      "loss": 0.1767,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 18180,
      "tokens_per_second_per_gpu": 361.86
    },
    {
      "epoch": 1.8083710294775563,
      "grad_norm": 0.46647799015045166,
      "learning_rate": 4.5432805427452765e-06,
      "loss": 0.131,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 18190,
      "tokens_per_second_per_gpu": 337.2
    },
    {
      "epoch": 1.8093652135010192,
      "grad_norm": 0.5966955423355103,
      "learning_rate": 4.496628816885318e-06,
      "loss": 0.145,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 18200,
      "tokens_per_second_per_gpu": 401.16
    },
    {
      "epoch": 1.8103593975244818,
      "grad_norm": 0.37769100069999695,
      "learning_rate": 4.450212336013681e-06,
      "loss": 0.1557,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 18210,
      "tokens_per_second_per_gpu": 335.53
    },
    {
      "epoch": 1.8113535815479445,
      "grad_norm": 0.4889051616191864,
      "learning_rate": 4.404031214463966e-06,
      "loss": 0.1208,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 18220,
      "tokens_per_second_per_gpu": 339.5
    },
    {
      "epoch": 1.8123477655714073,
      "grad_norm": 0.4089222848415375,
      "learning_rate": 4.358085565990044e-06,
      "loss": 0.1608,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 18230,
      "tokens_per_second_per_gpu": 362.98
    },
    {
      "epoch": 1.8133419495948702,
      "grad_norm": 0.7150986194610596,
      "learning_rate": 4.312375503765742e-06,
      "loss": 0.1807,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 18240,
      "tokens_per_second_per_gpu": 348.29
    },
    {
      "epoch": 1.8143361336183328,
      "grad_norm": 0.5326732397079468,
      "learning_rate": 4.266901140384616e-06,
      "loss": 0.145,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 18250,
      "tokens_per_second_per_gpu": 346.23
    },
    {
      "epoch": 1.8153303176417954,
      "grad_norm": 0.5794299840927124,
      "learning_rate": 4.221662587859631e-06,
      "loss": 0.1534,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 18260,
      "tokens_per_second_per_gpu": 334.07
    },
    {
      "epoch": 1.8163245016652583,
      "grad_norm": 0.4839019477367401,
      "learning_rate": 4.1766599576229195e-06,
      "loss": 0.189,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 18270,
      "tokens_per_second_per_gpu": 345.54
    },
    {
      "epoch": 1.8173186856887211,
      "grad_norm": 0.5688499212265015,
      "learning_rate": 4.131893360525452e-06,
      "loss": 0.1432,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 18280,
      "tokens_per_second_per_gpu": 337.5
    },
    {
      "epoch": 1.8183128697121838,
      "grad_norm": 0.5413910150527954,
      "learning_rate": 4.087362906836812e-06,
      "loss": 0.1351,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 18290,
      "tokens_per_second_per_gpu": 383.78
    },
    {
      "epoch": 1.8193070537356464,
      "grad_norm": 0.4739817678928375,
      "learning_rate": 4.043068706244957e-06,
      "loss": 0.1253,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 18300,
      "tokens_per_second_per_gpu": 356.75
    },
    {
      "epoch": 1.8203012377591092,
      "grad_norm": 0.34818801283836365,
      "learning_rate": 3.999010867855812e-06,
      "loss": 0.1211,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 18310,
      "tokens_per_second_per_gpu": 319.55
    },
    {
      "epoch": 1.821295421782572,
      "grad_norm": 0.692557692527771,
      "learning_rate": 3.955189500193191e-06,
      "loss": 0.1494,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 18320,
      "tokens_per_second_per_gpu": 378.18
    },
    {
      "epoch": 1.8222896058060347,
      "grad_norm": 0.713403582572937,
      "learning_rate": 3.911604711198358e-06,
      "loss": 0.1781,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 18330,
      "tokens_per_second_per_gpu": 376.29
    },
    {
      "epoch": 1.8232837898294973,
      "grad_norm": 0.5273980498313904,
      "learning_rate": 3.8682566082298695e-06,
      "loss": 0.1802,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.84,
      "memory/max_allocated (GiB)": 18.84,
      "step": 18340,
      "tokens_per_second_per_gpu": 353.26
    },
    {
      "epoch": 1.8242779738529602,
      "grad_norm": 0.4756382703781128,
      "learning_rate": 3.825145298063249e-06,
      "loss": 0.1553,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 18350,
      "tokens_per_second_per_gpu": 343.84
    },
    {
      "epoch": 1.825272157876423,
      "grad_norm": 0.27914801239967346,
      "learning_rate": 3.782270886890793e-06,
      "loss": 0.1831,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.37,
      "memory/max_allocated (GiB)": 21.37,
      "step": 18360,
      "tokens_per_second_per_gpu": 341.0
    },
    {
      "epoch": 1.8262663418998857,
      "grad_norm": 0.6465624570846558,
      "learning_rate": 3.739633480321214e-06,
      "loss": 0.2023,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 18370,
      "tokens_per_second_per_gpu": 395.46
    },
    {
      "epoch": 1.8272605259233483,
      "grad_norm": 0.5615578293800354,
      "learning_rate": 3.697233183379467e-06,
      "loss": 0.189,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 18380,
      "tokens_per_second_per_gpu": 345.47
    },
    {
      "epoch": 1.8282547099468112,
      "grad_norm": 0.5241686701774597,
      "learning_rate": 3.6550701005064413e-06,
      "loss": 0.1616,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 18390,
      "tokens_per_second_per_gpu": 381.72
    },
    {
      "epoch": 1.829248893970274,
      "grad_norm": 0.37639015913009644,
      "learning_rate": 3.613144335558738e-06,
      "loss": 0.1789,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.85,
      "memory/max_allocated (GiB)": 18.85,
      "step": 18400,
      "tokens_per_second_per_gpu": 353.78
    },
    {
      "epoch": 1.8302430779937366,
      "grad_norm": 0.3994421064853668,
      "learning_rate": 3.571455991808348e-06,
      "loss": 0.1619,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 18410,
      "tokens_per_second_per_gpu": 387.6
    },
    {
      "epoch": 1.8312372620171993,
      "grad_norm": 0.6425352692604065,
      "learning_rate": 3.5300051719424854e-06,
      "loss": 0.137,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.23,
      "memory/max_allocated (GiB)": 19.23,
      "step": 18420,
      "tokens_per_second_per_gpu": 346.83
    },
    {
      "epoch": 1.8322314460406621,
      "grad_norm": 0.4775727391242981,
      "learning_rate": 3.4887919780632995e-06,
      "loss": 0.1569,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 18430,
      "tokens_per_second_per_gpu": 349.11
    },
    {
      "epoch": 1.833225630064125,
      "grad_norm": 0.37590330839157104,
      "learning_rate": 3.4478165116875626e-06,
      "loss": 0.1745,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 18440,
      "tokens_per_second_per_gpu": 426.6
    },
    {
      "epoch": 1.8342198140875876,
      "grad_norm": 0.47229719161987305,
      "learning_rate": 3.4070788737465497e-06,
      "loss": 0.1438,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 18450,
      "tokens_per_second_per_gpu": 354.96
    },
    {
      "epoch": 1.8352139981110502,
      "grad_norm": 0.5609657764434814,
      "learning_rate": 3.3665791645856258e-06,
      "loss": 0.151,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 18460,
      "tokens_per_second_per_gpu": 349.0
    },
    {
      "epoch": 1.836208182134513,
      "grad_norm": 0.5257097482681274,
      "learning_rate": 3.326317483964181e-06,
      "loss": 0.1469,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 18470,
      "tokens_per_second_per_gpu": 425.66
    },
    {
      "epoch": 1.837202366157976,
      "grad_norm": 0.9030627608299255,
      "learning_rate": 3.2862939310552065e-06,
      "loss": 0.1525,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 18480,
      "tokens_per_second_per_gpu": 370.97
    },
    {
      "epoch": 1.8381965501814386,
      "grad_norm": 0.3405112326145172,
      "learning_rate": 3.2465086044451976e-06,
      "loss": 0.1649,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 18490,
      "tokens_per_second_per_gpu": 368.6
    },
    {
      "epoch": 1.8391907342049012,
      "grad_norm": 0.3417605459690094,
      "learning_rate": 3.206961602133807e-06,
      "loss": 0.1433,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 18500,
      "tokens_per_second_per_gpu": 373.55
    },
    {
      "epoch": 1.840184918228364,
      "grad_norm": 0.6065083742141724,
      "learning_rate": 3.1676530215336675e-06,
      "loss": 0.1569,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 18510,
      "tokens_per_second_per_gpu": 327.3
    },
    {
      "epoch": 1.841179102251827,
      "grad_norm": 0.43219438195228577,
      "learning_rate": 3.1285829594701165e-06,
      "loss": 0.1324,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.86,
      "memory/max_allocated (GiB)": 18.86,
      "step": 18520,
      "tokens_per_second_per_gpu": 301.88
    },
    {
      "epoch": 1.8421732862752895,
      "grad_norm": 0.58506840467453,
      "learning_rate": 3.089751512180972e-06,
      "loss": 0.1842,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 18530,
      "tokens_per_second_per_gpu": 414.46
    },
    {
      "epoch": 1.8431674702987522,
      "grad_norm": 0.4577626883983612,
      "learning_rate": 3.0511587753163094e-06,
      "loss": 0.1406,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 18540,
      "tokens_per_second_per_gpu": 423.64
    },
    {
      "epoch": 1.844161654322215,
      "grad_norm": 0.6649712324142456,
      "learning_rate": 3.0128048439381886e-06,
      "loss": 0.1355,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.23,
      "memory/max_allocated (GiB)": 19.23,
      "step": 18550,
      "tokens_per_second_per_gpu": 401.52
    },
    {
      "epoch": 1.8451558383456779,
      "grad_norm": 0.3550589084625244,
      "learning_rate": 2.974689812520448e-06,
      "loss": 0.1429,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 18560,
      "tokens_per_second_per_gpu": 331.54
    },
    {
      "epoch": 1.8461500223691405,
      "grad_norm": 0.5486629605293274,
      "learning_rate": 2.9368137749484547e-06,
      "loss": 0.185,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 18570,
      "tokens_per_second_per_gpu": 371.95
    },
    {
      "epoch": 1.8471442063926031,
      "grad_norm": 0.45112037658691406,
      "learning_rate": 2.8991768245189232e-06,
      "loss": 0.1472,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 18580,
      "tokens_per_second_per_gpu": 388.42
    },
    {
      "epoch": 1.848138390416066,
      "grad_norm": 0.40423816442489624,
      "learning_rate": 2.861779053939595e-06,
      "loss": 0.1218,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.85,
      "memory/max_allocated (GiB)": 18.85,
      "step": 18590,
      "tokens_per_second_per_gpu": 357.26
    },
    {
      "epoch": 1.8491325744395288,
      "grad_norm": 0.49648118019104004,
      "learning_rate": 2.824620555329094e-06,
      "loss": 0.181,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 18600,
      "tokens_per_second_per_gpu": 349.07
    },
    {
      "epoch": 1.8501267584629915,
      "grad_norm": 0.4115513265132904,
      "learning_rate": 2.7877014202166372e-06,
      "loss": 0.1188,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 18610,
      "tokens_per_second_per_gpu": 344.58
    },
    {
      "epoch": 1.851120942486454,
      "grad_norm": 0.5676191449165344,
      "learning_rate": 2.7510217395418815e-06,
      "loss": 0.127,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 18620,
      "tokens_per_second_per_gpu": 435.5
    },
    {
      "epoch": 1.852115126509917,
      "grad_norm": 0.5204625725746155,
      "learning_rate": 2.714581603654609e-06,
      "loss": 0.14,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 18630,
      "tokens_per_second_per_gpu": 374.45
    },
    {
      "epoch": 1.8531093105333798,
      "grad_norm": 0.5703979134559631,
      "learning_rate": 2.6783811023145977e-06,
      "loss": 0.1432,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 18640,
      "tokens_per_second_per_gpu": 393.22
    },
    {
      "epoch": 1.8541034945568424,
      "grad_norm": 1.1127846240997314,
      "learning_rate": 2.6424203246913194e-06,
      "loss": 0.1862,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 18650,
      "tokens_per_second_per_gpu": 385.6
    },
    {
      "epoch": 1.855097678580305,
      "grad_norm": 0.37595993280410767,
      "learning_rate": 2.6066993593637844e-06,
      "loss": 0.1694,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 18660,
      "tokens_per_second_per_gpu": 387.27
    },
    {
      "epoch": 1.856091862603768,
      "grad_norm": 0.35933300852775574,
      "learning_rate": 2.571218294320266e-06,
      "loss": 0.1011,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 18670,
      "tokens_per_second_per_gpu": 354.13
    },
    {
      "epoch": 1.8570860466272308,
      "grad_norm": 0.5620598196983337,
      "learning_rate": 2.5359772169581297e-06,
      "loss": 0.1991,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.7,
      "memory/max_allocated (GiB)": 19.7,
      "step": 18680,
      "tokens_per_second_per_gpu": 389.06
    },
    {
      "epoch": 1.8580802306506934,
      "grad_norm": 0.29674777388572693,
      "learning_rate": 2.5009762140835947e-06,
      "loss": 0.1497,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 18690,
      "tokens_per_second_per_gpu": 373.51
    },
    {
      "epoch": 1.8590744146741562,
      "grad_norm": 0.4018515348434448,
      "learning_rate": 2.4662153719115398e-06,
      "loss": 0.1743,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 18700,
      "tokens_per_second_per_gpu": 373.54
    },
    {
      "epoch": 1.860068598697619,
      "grad_norm": 0.5389109253883362,
      "learning_rate": 2.431694776065263e-06,
      "loss": 0.1169,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.53,
      "memory/max_allocated (GiB)": 20.53,
      "step": 18710,
      "tokens_per_second_per_gpu": 340.4
    },
    {
      "epoch": 1.8610627827210817,
      "grad_norm": 0.526465654373169,
      "learning_rate": 2.397414511576268e-06,
      "loss": 0.1841,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 18720,
      "tokens_per_second_per_gpu": 348.59
    },
    {
      "epoch": 1.8620569667445444,
      "grad_norm": 0.3093705177307129,
      "learning_rate": 2.3633746628841325e-06,
      "loss": 0.1741,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 18730,
      "tokens_per_second_per_gpu": 313.63
    },
    {
      "epoch": 1.8630511507680072,
      "grad_norm": 0.9190280437469482,
      "learning_rate": 2.329575313836152e-06,
      "loss": 0.2199,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 18740,
      "tokens_per_second_per_gpu": 400.19
    },
    {
      "epoch": 1.86404533479147,
      "grad_norm": 0.5890634655952454,
      "learning_rate": 2.2960165476873076e-06,
      "loss": 0.133,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 18750,
      "tokens_per_second_per_gpu": 344.18
    },
    {
      "epoch": 1.8650395188149327,
      "grad_norm": 0.7027098536491394,
      "learning_rate": 2.262698447099898e-06,
      "loss": 0.1459,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 18760,
      "tokens_per_second_per_gpu": 288.79
    },
    {
      "epoch": 1.8660337028383953,
      "grad_norm": 0.7816129922866821,
      "learning_rate": 2.2296210941434745e-06,
      "loss": 0.1699,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 18770,
      "tokens_per_second_per_gpu": 369.86
    },
    {
      "epoch": 1.8670278868618582,
      "grad_norm": 0.4923355281352997,
      "learning_rate": 2.19678457029453e-06,
      "loss": 0.1324,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 18780,
      "tokens_per_second_per_gpu": 316.49
    },
    {
      "epoch": 1.868022070885321,
      "grad_norm": 0.27154669165611267,
      "learning_rate": 2.164188956436386e-06,
      "loss": 0.126,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 18790,
      "tokens_per_second_per_gpu": 363.75
    },
    {
      "epoch": 1.8690162549087836,
      "grad_norm": 0.5623666644096375,
      "learning_rate": 2.1318343328588953e-06,
      "loss": 0.161,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 18800,
      "tokens_per_second_per_gpu": 332.31
    },
    {
      "epoch": 1.8700104389322463,
      "grad_norm": 0.5379459261894226,
      "learning_rate": 2.099720779258352e-06,
      "loss": 0.1738,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 18810,
      "tokens_per_second_per_gpu": 410.92
    },
    {
      "epoch": 1.8710046229557091,
      "grad_norm": 0.37512362003326416,
      "learning_rate": 2.0678483747372247e-06,
      "loss": 0.1164,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 18820,
      "tokens_per_second_per_gpu": 351.41
    },
    {
      "epoch": 1.871998806979172,
      "grad_norm": 0.6244227886199951,
      "learning_rate": 2.03621719780398e-06,
      "loss": 0.1444,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 18830,
      "tokens_per_second_per_gpu": 364.33
    },
    {
      "epoch": 1.8729929910026346,
      "grad_norm": 0.3308027386665344,
      "learning_rate": 2.0048273263729046e-06,
      "loss": 0.1737,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 18840,
      "tokens_per_second_per_gpu": 346.54
    },
    {
      "epoch": 1.8739871750260972,
      "grad_norm": 0.2430552840232849,
      "learning_rate": 1.9736788377638705e-06,
      "loss": 0.1432,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.21,
      "memory/max_allocated (GiB)": 19.21,
      "step": 18850,
      "tokens_per_second_per_gpu": 310.33
    },
    {
      "epoch": 1.87498135904956,
      "grad_norm": 0.4946417212486267,
      "learning_rate": 1.942771808702204e-06,
      "loss": 0.1859,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 18860,
      "tokens_per_second_per_gpu": 314.75
    },
    {
      "epoch": 1.875975543073023,
      "grad_norm": 0.5932151079177856,
      "learning_rate": 1.9121063153184293e-06,
      "loss": 0.1639,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 18870,
      "tokens_per_second_per_gpu": 378.07
    },
    {
      "epoch": 1.8769697270964856,
      "grad_norm": 0.6450536847114563,
      "learning_rate": 1.8816824331481575e-06,
      "loss": 0.1325,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 18880,
      "tokens_per_second_per_gpu": 353.93
    },
    {
      "epoch": 1.8779639111199482,
      "grad_norm": 0.498829185962677,
      "learning_rate": 1.8515002371318312e-06,
      "loss": 0.1227,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 18890,
      "tokens_per_second_per_gpu": 357.94
    },
    {
      "epoch": 1.878958095143411,
      "grad_norm": 0.5611813068389893,
      "learning_rate": 1.8215598016145807e-06,
      "loss": 0.202,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.11,
      "memory/max_allocated (GiB)": 19.11,
      "step": 18900,
      "tokens_per_second_per_gpu": 402.84
    },
    {
      "epoch": 1.879952279166874,
      "grad_norm": 0.42488303780555725,
      "learning_rate": 1.7918612003460234e-06,
      "loss": 0.1108,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 18910,
      "tokens_per_second_per_gpu": 336.67
    },
    {
      "epoch": 1.8809464631903365,
      "grad_norm": 0.7830858826637268,
      "learning_rate": 1.7624045064800975e-06,
      "loss": 0.1591,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 18920,
      "tokens_per_second_per_gpu": 323.72
    },
    {
      "epoch": 1.8819406472137992,
      "grad_norm": 0.4499165415763855,
      "learning_rate": 1.7331897925748518e-06,
      "loss": 0.1812,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 18930,
      "tokens_per_second_per_gpu": 380.35
    },
    {
      "epoch": 1.882934831237262,
      "grad_norm": 0.6066138744354248,
      "learning_rate": 1.7042171305923115e-06,
      "loss": 0.0976,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 18940,
      "tokens_per_second_per_gpu": 327.67
    },
    {
      "epoch": 1.8839290152607249,
      "grad_norm": 0.30252575874328613,
      "learning_rate": 1.6754865918982677e-06,
      "loss": 0.1591,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 18950,
      "tokens_per_second_per_gpu": 301.08
    },
    {
      "epoch": 1.8849231992841875,
      "grad_norm": 0.575855016708374,
      "learning_rate": 1.6469982472621103e-06,
      "loss": 0.146,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 18960,
      "tokens_per_second_per_gpu": 341.39
    },
    {
      "epoch": 1.8859173833076501,
      "grad_norm": 0.7902649641036987,
      "learning_rate": 1.6187521668566518e-06,
      "loss": 0.1789,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 18970,
      "tokens_per_second_per_gpu": 370.72
    },
    {
      "epoch": 1.886911567331113,
      "grad_norm": 0.6112450957298279,
      "learning_rate": 1.5907484202579482e-06,
      "loss": 0.1358,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 18980,
      "tokens_per_second_per_gpu": 340.31
    },
    {
      "epoch": 1.8879057513545758,
      "grad_norm": 0.3243188261985779,
      "learning_rate": 1.562987076445177e-06,
      "loss": 0.1604,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.54,
      "memory/max_allocated (GiB)": 21.54,
      "step": 18990,
      "tokens_per_second_per_gpu": 383.91
    },
    {
      "epoch": 1.8888999353780385,
      "grad_norm": 0.3223695158958435,
      "learning_rate": 1.53546820380035e-06,
      "loss": 0.1408,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 19000,
      "tokens_per_second_per_gpu": 346.51
    },
    {
      "epoch": 1.889894119401501,
      "grad_norm": 0.49110448360443115,
      "learning_rate": 1.508191870108311e-06,
      "loss": 0.1379,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 19010,
      "tokens_per_second_per_gpu": 399.45
    },
    {
      "epoch": 1.890888303424964,
      "grad_norm": 0.3531555235385895,
      "learning_rate": 1.4811581425563936e-06,
      "loss": 0.1755,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 19020,
      "tokens_per_second_per_gpu": 360.95
    },
    {
      "epoch": 1.8918824874484268,
      "grad_norm": 0.3664921522140503,
      "learning_rate": 1.4543670877344207e-06,
      "loss": 0.1661,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 19030,
      "tokens_per_second_per_gpu": 397.9
    },
    {
      "epoch": 1.8928766714718894,
      "grad_norm": 0.6378931403160095,
      "learning_rate": 1.4278187716344039e-06,
      "loss": 0.174,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 19040,
      "tokens_per_second_per_gpu": 417.45
    },
    {
      "epoch": 1.893870855495352,
      "grad_norm": 0.1918243020772934,
      "learning_rate": 1.4015132596504554e-06,
      "loss": 0.1896,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 19050,
      "tokens_per_second_per_gpu": 407.04
    },
    {
      "epoch": 1.894865039518815,
      "grad_norm": 0.5390235781669617,
      "learning_rate": 1.3754506165786108e-06,
      "loss": 0.1789,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.77,
      "memory/max_allocated (GiB)": 19.77,
      "step": 19060,
      "tokens_per_second_per_gpu": 441.99
    },
    {
      "epoch": 1.8958592235422778,
      "grad_norm": 0.5331469178199768,
      "learning_rate": 1.3496309066166724e-06,
      "loss": 0.145,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19070,
      "tokens_per_second_per_gpu": 392.93
    },
    {
      "epoch": 1.8968534075657404,
      "grad_norm": 0.48252302408218384,
      "learning_rate": 1.3240541933640439e-06,
      "loss": 0.159,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 19080,
      "tokens_per_second_per_gpu": 330.47
    },
    {
      "epoch": 1.897847591589203,
      "grad_norm": 0.5958520174026489,
      "learning_rate": 1.298720539821563e-06,
      "loss": 0.1252,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 19090,
      "tokens_per_second_per_gpu": 325.74
    },
    {
      "epoch": 1.8988417756126659,
      "grad_norm": 0.675596296787262,
      "learning_rate": 1.273630008391402e-06,
      "loss": 0.1731,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 19100,
      "tokens_per_second_per_gpu": 356.72
    },
    {
      "epoch": 1.8998359596361287,
      "grad_norm": 0.5923382043838501,
      "learning_rate": 1.2487826608768127e-06,
      "loss": 0.149,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 19110,
      "tokens_per_second_per_gpu": 323.65
    },
    {
      "epoch": 1.9008301436595914,
      "grad_norm": 0.3214609920978546,
      "learning_rate": 1.2241785584820808e-06,
      "loss": 0.1449,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 19120,
      "tokens_per_second_per_gpu": 397.11
    },
    {
      "epoch": 1.901824327683054,
      "grad_norm": 0.47794198989868164,
      "learning_rate": 1.199817761812294e-06,
      "loss": 0.0914,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 19130,
      "tokens_per_second_per_gpu": 309.82
    },
    {
      "epoch": 1.9028185117065168,
      "grad_norm": 0.5440109372138977,
      "learning_rate": 1.175700330873275e-06,
      "loss": 0.1356,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 19140,
      "tokens_per_second_per_gpu": 385.7
    },
    {
      "epoch": 1.9038126957299797,
      "grad_norm": 0.5214424729347229,
      "learning_rate": 1.1518263250713147e-06,
      "loss": 0.1127,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 19150,
      "tokens_per_second_per_gpu": 314.47
    },
    {
      "epoch": 1.9048068797534423,
      "grad_norm": 0.42965662479400635,
      "learning_rate": 1.1281958032131611e-06,
      "loss": 0.1387,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 19160,
      "tokens_per_second_per_gpu": 381.94
    },
    {
      "epoch": 1.905801063776905,
      "grad_norm": 0.2793689966201782,
      "learning_rate": 1.1048088235057762e-06,
      "loss": 0.1781,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 19170,
      "tokens_per_second_per_gpu": 312.99
    },
    {
      "epoch": 1.9067952478003678,
      "grad_norm": 0.8887554407119751,
      "learning_rate": 1.0816654435562458e-06,
      "loss": 0.1757,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 19180,
      "tokens_per_second_per_gpu": 381.31
    },
    {
      "epoch": 1.9077894318238307,
      "grad_norm": 0.4591097831726074,
      "learning_rate": 1.0587657203715795e-06,
      "loss": 0.1417,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.17,
      "memory/max_allocated (GiB)": 18.17,
      "step": 19190,
      "tokens_per_second_per_gpu": 326.93
    },
    {
      "epoch": 1.9087836158472933,
      "grad_norm": 0.4055291414260864,
      "learning_rate": 1.036109710358657e-06,
      "loss": 0.1935,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 19200,
      "tokens_per_second_per_gpu": 363.7
    },
    {
      "epoch": 1.9097777998707561,
      "grad_norm": 0.6825937628746033,
      "learning_rate": 1.0136974693240153e-06,
      "loss": 0.1551,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 17.41,
      "memory/max_allocated (GiB)": 17.41,
      "step": 19210,
      "tokens_per_second_per_gpu": 291.2
    },
    {
      "epoch": 1.910771983894219,
      "grad_norm": 0.43678414821624756,
      "learning_rate": 9.915290524737274e-07,
      "loss": 0.1015,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.69,
      "memory/max_allocated (GiB)": 19.69,
      "step": 19220,
      "tokens_per_second_per_gpu": 363.79
    },
    {
      "epoch": 1.9117661679176816,
      "grad_norm": 0.24454043805599213,
      "learning_rate": 9.696045144133136e-07,
      "loss": 0.1326,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.19,
      "memory/max_allocated (GiB)": 18.19,
      "step": 19230,
      "tokens_per_second_per_gpu": 314.11
    },
    {
      "epoch": 1.9127603519411442,
      "grad_norm": 0.40022847056388855,
      "learning_rate": 9.4792390914753e-07,
      "loss": 0.1666,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.65,
      "memory/max_allocated (GiB)": 20.65,
      "step": 19240,
      "tokens_per_second_per_gpu": 377.05
    },
    {
      "epoch": 1.913754535964607,
      "grad_norm": 0.4418630003929138,
      "learning_rate": 9.264872900802912e-07,
      "loss": 0.2076,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 19250,
      "tokens_per_second_per_gpu": 374.03
    },
    {
      "epoch": 1.91474871998807,
      "grad_norm": 0.1951112598180771,
      "learning_rate": 9.052947100145149e-07,
      "loss": 0.1309,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 19260,
      "tokens_per_second_per_gpu": 372.81
    },
    {
      "epoch": 1.9157429040115326,
      "grad_norm": 0.4262109100818634,
      "learning_rate": 8.843462211520215e-07,
      "loss": 0.159,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 19270,
      "tokens_per_second_per_gpu": 352.44
    },
    {
      "epoch": 1.9167370880349952,
      "grad_norm": 0.35612645745277405,
      "learning_rate": 8.636418750933461e-07,
      "loss": 0.1805,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.78,
      "memory/max_allocated (GiB)": 20.78,
      "step": 19280,
      "tokens_per_second_per_gpu": 423.92
    },
    {
      "epoch": 1.917731272058458,
      "grad_norm": 0.4936760365962982,
      "learning_rate": 8.431817228376937e-07,
      "loss": 0.151,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 19290,
      "tokens_per_second_per_gpu": 382.85
    },
    {
      "epoch": 1.918725456081921,
      "grad_norm": 0.6932761669158936,
      "learning_rate": 8.229658147827169e-07,
      "loss": 0.1807,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 19300,
      "tokens_per_second_per_gpu": 369.43
    },
    {
      "epoch": 1.9197196401053835,
      "grad_norm": 0.40733301639556885,
      "learning_rate": 8.02994200724494e-07,
      "loss": 0.0993,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 19310,
      "tokens_per_second_per_gpu": 290.15
    },
    {
      "epoch": 1.9207138241288462,
      "grad_norm": 0.49551132321357727,
      "learning_rate": 7.83266929857307e-07,
      "loss": 0.1942,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 19320,
      "tokens_per_second_per_gpu": 361.74
    },
    {
      "epoch": 1.921708008152309,
      "grad_norm": 0.5282984972000122,
      "learning_rate": 7.637840507736194e-07,
      "loss": 0.2019,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19330,
      "tokens_per_second_per_gpu": 354.99
    },
    {
      "epoch": 1.9227021921757719,
      "grad_norm": 0.6770442724227905,
      "learning_rate": 7.445456114638539e-07,
      "loss": 0.1387,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 19340,
      "tokens_per_second_per_gpu": 297.34
    },
    {
      "epoch": 1.9236963761992345,
      "grad_norm": 0.35707518458366394,
      "learning_rate": 7.255516593163703e-07,
      "loss": 0.1626,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 19350,
      "tokens_per_second_per_gpu": 361.77
    },
    {
      "epoch": 1.9246905602226971,
      "grad_norm": 0.5072949528694153,
      "learning_rate": 7.06802241117288e-07,
      "loss": 0.19,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.45,
      "memory/max_allocated (GiB)": 20.45,
      "step": 19360,
      "tokens_per_second_per_gpu": 382.78
    },
    {
      "epoch": 1.92568474424616,
      "grad_norm": 0.5360376834869385,
      "learning_rate": 6.882974030503863e-07,
      "loss": 0.1228,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19370,
      "tokens_per_second_per_gpu": 318.51
    },
    {
      "epoch": 1.9266789282696228,
      "grad_norm": 0.5876972079277039,
      "learning_rate": 6.700371906969815e-07,
      "loss": 0.1511,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 19380,
      "tokens_per_second_per_gpu": 409.23
    },
    {
      "epoch": 1.9276731122930855,
      "grad_norm": 0.5715491771697998,
      "learning_rate": 6.520216490358388e-07,
      "loss": 0.1781,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.16,
      "memory/max_allocated (GiB)": 18.16,
      "step": 19390,
      "tokens_per_second_per_gpu": 394.12
    },
    {
      "epoch": 1.928667296316548,
      "grad_norm": 0.47656625509262085,
      "learning_rate": 6.342508224430499e-07,
      "loss": 0.1633,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 19400,
      "tokens_per_second_per_gpu": 406.05
    },
    {
      "epoch": 1.929661480340011,
      "grad_norm": 0.531021237373352,
      "learning_rate": 6.167247546919219e-07,
      "loss": 0.1678,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 19410,
      "tokens_per_second_per_gpu": 428.32
    },
    {
      "epoch": 1.9306556643634738,
      "grad_norm": 0.5396085381507874,
      "learning_rate": 5.994434889528556e-07,
      "loss": 0.1532,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.4,
      "memory/max_allocated (GiB)": 21.4,
      "step": 19420,
      "tokens_per_second_per_gpu": 393.0
    },
    {
      "epoch": 1.9316498483869364,
      "grad_norm": 0.33545199036598206,
      "learning_rate": 5.824070677932558e-07,
      "loss": 0.1298,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 19430,
      "tokens_per_second_per_gpu": 361.14
    },
    {
      "epoch": 1.932644032410399,
      "grad_norm": 0.3300628066062927,
      "learning_rate": 5.656155331774437e-07,
      "loss": 0.155,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 19440,
      "tokens_per_second_per_gpu": 350.34
    },
    {
      "epoch": 1.933638216433862,
      "grad_norm": 0.46381351351737976,
      "learning_rate": 5.490689264665117e-07,
      "loss": 0.1298,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.48,
      "memory/max_allocated (GiB)": 21.48,
      "step": 19450,
      "tokens_per_second_per_gpu": 357.6
    },
    {
      "epoch": 1.9346324004573248,
      "grad_norm": 0.5388414263725281,
      "learning_rate": 5.327672884182455e-07,
      "loss": 0.1173,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.79,
      "memory/max_allocated (GiB)": 19.79,
      "step": 19460,
      "tokens_per_second_per_gpu": 354.37
    },
    {
      "epoch": 1.9356265844807874,
      "grad_norm": 0.3915001153945923,
      "learning_rate": 5.167106591870252e-07,
      "loss": 0.1224,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19470,
      "tokens_per_second_per_gpu": 298.76
    },
    {
      "epoch": 1.93662076850425,
      "grad_norm": 0.3780313730239868,
      "learning_rate": 5.008990783237244e-07,
      "loss": 0.1175,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.52,
      "memory/max_allocated (GiB)": 21.52,
      "step": 19480,
      "tokens_per_second_per_gpu": 300.87
    },
    {
      "epoch": 1.9376149525277129,
      "grad_norm": 0.5083069801330566,
      "learning_rate": 4.853325847755997e-07,
      "loss": 0.144,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 19490,
      "tokens_per_second_per_gpu": 391.62
    },
    {
      "epoch": 1.9386091365511757,
      "grad_norm": 0.8569310307502747,
      "learning_rate": 4.700112168862347e-07,
      "loss": 0.1368,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 19500,
      "tokens_per_second_per_gpu": 364.34
    },
    {
      "epoch": 1.9396033205746384,
      "grad_norm": 0.6087594032287598,
      "learning_rate": 4.549350123953855e-07,
      "loss": 0.1282,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.77,
      "memory/max_allocated (GiB)": 20.77,
      "step": 19510,
      "tokens_per_second_per_gpu": 351.22
    },
    {
      "epoch": 1.940597504598101,
      "grad_norm": 0.5610264539718628,
      "learning_rate": 4.4010400843892407e-07,
      "loss": 0.1441,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19520,
      "tokens_per_second_per_gpu": 344.44
    },
    {
      "epoch": 1.9415916886215638,
      "grad_norm": 0.3209472894668579,
      "learning_rate": 4.255182415487613e-07,
      "loss": 0.1601,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 19530,
      "tokens_per_second_per_gpu": 371.35
    },
    {
      "epoch": 1.9425858726450267,
      "grad_norm": 0.34549176692962646,
      "learning_rate": 4.1117774765270235e-07,
      "loss": 0.1846,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 19540,
      "tokens_per_second_per_gpu": 392.28
    },
    {
      "epoch": 1.9435800566684893,
      "grad_norm": 0.610714852809906,
      "learning_rate": 3.970825620744467e-07,
      "loss": 0.1392,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.83,
      "memory/max_allocated (GiB)": 18.83,
      "step": 19550,
      "tokens_per_second_per_gpu": 300.36
    },
    {
      "epoch": 1.944574240691952,
      "grad_norm": 0.42442601919174194,
      "learning_rate": 3.8323271953338844e-07,
      "loss": 0.2258,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19560,
      "tokens_per_second_per_gpu": 434.15
    },
    {
      "epoch": 1.9455684247154148,
      "grad_norm": 0.30905285477638245,
      "learning_rate": 3.696282541446272e-07,
      "loss": 0.1478,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19570,
      "tokens_per_second_per_gpu": 368.05
    },
    {
      "epoch": 1.9465626087388777,
      "grad_norm": 0.29837194085121155,
      "learning_rate": 3.56269199418835e-07,
      "loss": 0.1548,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 19580,
      "tokens_per_second_per_gpu": 314.0
    },
    {
      "epoch": 1.9475567927623403,
      "grad_norm": 0.2777169346809387,
      "learning_rate": 3.431555882621895e-07,
      "loss": 0.1615,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.63,
      "memory/max_allocated (GiB)": 20.63,
      "step": 19590,
      "tokens_per_second_per_gpu": 391.63
    },
    {
      "epoch": 1.948550976785803,
      "grad_norm": 0.7126250267028809,
      "learning_rate": 3.302874529762745e-07,
      "loss": 0.1598,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 19600,
      "tokens_per_second_per_gpu": 376.12
    },
    {
      "epoch": 1.9495451608092658,
      "grad_norm": 0.5951281785964966,
      "learning_rate": 3.176648252580461e-07,
      "loss": 0.1509,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 19610,
      "tokens_per_second_per_gpu": 361.14
    },
    {
      "epoch": 1.9505393448327286,
      "grad_norm": 0.5065922737121582,
      "learning_rate": 3.0528773619969977e-07,
      "loss": 0.1458,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19620,
      "tokens_per_second_per_gpu": 343.47
    },
    {
      "epoch": 1.9515335288561912,
      "grad_norm": 0.463041216135025,
      "learning_rate": 2.9315621628860366e-07,
      "loss": 0.1204,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 19630,
      "tokens_per_second_per_gpu": 303.97
    },
    {
      "epoch": 1.9525277128796539,
      "grad_norm": 0.7519829869270325,
      "learning_rate": 2.812702954072877e-07,
      "loss": 0.1774,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 19640,
      "tokens_per_second_per_gpu": 321.53
    },
    {
      "epoch": 1.9535218969031167,
      "grad_norm": 0.52646404504776,
      "learning_rate": 2.6963000283325434e-07,
      "loss": 0.1509,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 19650,
      "tokens_per_second_per_gpu": 311.88
    },
    {
      "epoch": 1.9545160809265796,
      "grad_norm": 0.4693813920021057,
      "learning_rate": 2.5823536723902364e-07,
      "loss": 0.1782,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 19660,
      "tokens_per_second_per_gpu": 393.51
    },
    {
      "epoch": 1.9555102649500422,
      "grad_norm": 0.5138804912567139,
      "learning_rate": 2.470864166919884e-07,
      "loss": 0.16,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.54,
      "memory/max_allocated (GiB)": 20.54,
      "step": 19670,
      "tokens_per_second_per_gpu": 343.73
    },
    {
      "epoch": 1.9565044489735048,
      "grad_norm": 0.5430607795715332,
      "learning_rate": 2.361831786543589e-07,
      "loss": 0.0975,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19680,
      "tokens_per_second_per_gpu": 321.72
    },
    {
      "epoch": 1.9574986329969677,
      "grad_norm": 0.39533936977386475,
      "learning_rate": 2.2552567998312957e-07,
      "loss": 0.1112,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19690,
      "tokens_per_second_per_gpu": 342.61
    },
    {
      "epoch": 1.9584928170204305,
      "grad_norm": 0.4892284870147705,
      "learning_rate": 2.151139469299679e-07,
      "loss": 0.1444,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19700,
      "tokens_per_second_per_gpu": 337.57
    },
    {
      "epoch": 1.9594870010438932,
      "grad_norm": 0.6313555836677551,
      "learning_rate": 2.0494800514117007e-07,
      "loss": 0.1839,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.43,
      "memory/max_allocated (GiB)": 20.43,
      "step": 19710,
      "tokens_per_second_per_gpu": 379.84
    },
    {
      "epoch": 1.9604811850673558,
      "grad_norm": 0.3936193585395813,
      "learning_rate": 1.950278796576055e-07,
      "loss": 0.1649,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.42,
      "memory/max_allocated (GiB)": 21.42,
      "step": 19720,
      "tokens_per_second_per_gpu": 352.38
    },
    {
      "epoch": 1.9614753690908189,
      "grad_norm": 0.4717642068862915,
      "learning_rate": 1.8535359491462789e-07,
      "loss": 0.1267,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 19730,
      "tokens_per_second_per_gpu": 297.29
    },
    {
      "epoch": 1.9624695531142815,
      "grad_norm": 0.574255108833313,
      "learning_rate": 1.7592517474205317e-07,
      "loss": 0.1551,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 19740,
      "tokens_per_second_per_gpu": 367.19
    },
    {
      "epoch": 1.9634637371377441,
      "grad_norm": 0.6956222057342529,
      "learning_rate": 1.6674264236408165e-07,
      "loss": 0.1482,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.57,
      "memory/max_allocated (GiB)": 20.57,
      "step": 19750,
      "tokens_per_second_per_gpu": 278.73
    },
    {
      "epoch": 1.964457921161207,
      "grad_norm": 0.5839533805847168,
      "learning_rate": 1.5780602039920932e-07,
      "loss": 0.1813,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.54,
      "memory/max_allocated (GiB)": 21.54,
      "step": 19760,
      "tokens_per_second_per_gpu": 344.8
    },
    {
      "epoch": 1.9654521051846698,
      "grad_norm": 0.4014669954776764,
      "learning_rate": 1.4911533086024997e-07,
      "loss": 0.1851,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19770,
      "tokens_per_second_per_gpu": 318.26
    },
    {
      "epoch": 1.9664462892081325,
      "grad_norm": 0.33930712938308716,
      "learning_rate": 1.406705951541909e-07,
      "loss": 0.1547,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.41,
      "memory/max_allocated (GiB)": 21.41,
      "step": 19780,
      "tokens_per_second_per_gpu": 422.7
    },
    {
      "epoch": 1.967440473231595,
      "grad_norm": 0.7784512639045715,
      "learning_rate": 1.32471834082204e-07,
      "loss": 0.1737,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.38,
      "memory/max_allocated (GiB)": 21.38,
      "step": 19790,
      "tokens_per_second_per_gpu": 379.19
    },
    {
      "epoch": 1.968434657255058,
      "grad_norm": 0.5413442850112915,
      "learning_rate": 1.2451906783957912e-07,
      "loss": 0.1701,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 19800,
      "tokens_per_second_per_gpu": 413.24
    },
    {
      "epoch": 1.9694288412785208,
      "grad_norm": 0.49978771805763245,
      "learning_rate": 1.1681231601564647e-07,
      "loss": 0.1503,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19810,
      "tokens_per_second_per_gpu": 325.42
    },
    {
      "epoch": 1.9704230253019834,
      "grad_norm": 0.5488161444664001,
      "learning_rate": 1.0935159759378755e-07,
      "loss": 0.1421,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19820,
      "tokens_per_second_per_gpu": 347.6
    },
    {
      "epoch": 1.971417209325446,
      "grad_norm": 0.36043041944503784,
      "learning_rate": 1.0213693095130206e-07,
      "loss": 0.1309,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.22,
      "memory/max_allocated (GiB)": 19.22,
      "step": 19830,
      "tokens_per_second_per_gpu": 286.15
    },
    {
      "epoch": 1.972411393348909,
      "grad_norm": 0.30269375443458557,
      "learning_rate": 9.516833385945224e-08,
      "loss": 0.1344,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19840,
      "tokens_per_second_per_gpu": 411.14
    },
    {
      "epoch": 1.9734055773723718,
      "grad_norm": 0.6402490735054016,
      "learning_rate": 8.844582348336294e-08,
      "loss": 0.1754,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.58,
      "memory/max_allocated (GiB)": 20.58,
      "step": 19850,
      "tokens_per_second_per_gpu": 333.73
    },
    {
      "epoch": 1.9743997613958344,
      "grad_norm": 0.3337123394012451,
      "learning_rate": 8.196941638199951e-08,
      "loss": 0.1245,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19860,
      "tokens_per_second_per_gpu": 365.58
    },
    {
      "epoch": 1.975393945419297,
      "grad_norm": 0.4003879427909851,
      "learning_rate": 7.573912850812326e-08,
      "loss": 0.1501,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.39,
      "memory/max_allocated (GiB)": 21.39,
      "step": 19870,
      "tokens_per_second_per_gpu": 344.89
    },
    {
      "epoch": 1.9763881294427599,
      "grad_norm": 0.21886324882507324,
      "learning_rate": 6.975497520824715e-08,
      "loss": 0.1815,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 19880,
      "tokens_per_second_per_gpu": 419.3
    },
    {
      "epoch": 1.9773823134662227,
      "grad_norm": 0.7370038628578186,
      "learning_rate": 6.401697122260241e-08,
      "loss": 0.1328,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 18.18,
      "memory/max_allocated (GiB)": 18.18,
      "step": 19890,
      "tokens_per_second_per_gpu": 332.95
    },
    {
      "epoch": 1.9783764974896854,
      "grad_norm": 0.46660006046295166,
      "learning_rate": 5.852513068511645e-08,
      "loss": 0.1096,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 19900,
      "tokens_per_second_per_gpu": 298.67
    },
    {
      "epoch": 1.979370681513148,
      "grad_norm": 0.5484210848808289,
      "learning_rate": 5.3279467123346086e-08,
      "loss": 0.1374,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 19910,
      "tokens_per_second_per_gpu": 340.44
    },
    {
      "epoch": 1.9803648655366108,
      "grad_norm": 0.4510185122489929,
      "learning_rate": 4.827999345846657e-08,
      "loss": 0.1858,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 19920,
      "tokens_per_second_per_gpu": 407.5
    },
    {
      "epoch": 1.9813590495600737,
      "grad_norm": 0.4410419464111328,
      "learning_rate": 4.352672200523822e-08,
      "loss": 0.1675,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.8,
      "memory/max_allocated (GiB)": 19.8,
      "step": 19930,
      "tokens_per_second_per_gpu": 308.63
    },
    {
      "epoch": 1.9823532335835363,
      "grad_norm": 0.5038071274757385,
      "learning_rate": 3.901966447197314e-08,
      "loss": 0.1349,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.78,
      "memory/max_allocated (GiB)": 19.78,
      "step": 19940,
      "tokens_per_second_per_gpu": 351.5
    },
    {
      "epoch": 1.983347417606999,
      "grad_norm": 1.0024924278259277,
      "learning_rate": 3.4758831960524095e-08,
      "loss": 0.1633,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 19950,
      "tokens_per_second_per_gpu": 326.94
    },
    {
      "epoch": 1.9843416016304618,
      "grad_norm": 0.5408278107643127,
      "learning_rate": 3.0744234966195715e-08,
      "loss": 0.1853,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 19960,
      "tokens_per_second_per_gpu": 380.23
    },
    {
      "epoch": 1.9853357856539247,
      "grad_norm": 0.6154715418815613,
      "learning_rate": 2.6975883377799993e-08,
      "loss": 0.159,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.5,
      "memory/max_allocated (GiB)": 21.5,
      "step": 19970,
      "tokens_per_second_per_gpu": 345.31
    },
    {
      "epoch": 1.9863299696773873,
      "grad_norm": 0.52059006690979,
      "learning_rate": 2.3453786477589668e-08,
      "loss": 0.1489,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.64,
      "memory/max_allocated (GiB)": 20.64,
      "step": 19980,
      "tokens_per_second_per_gpu": 365.37
    },
    {
      "epoch": 1.98732415370085,
      "grad_norm": 0.5241625905036926,
      "learning_rate": 2.0177952941224932e-08,
      "loss": 0.1679,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 19990,
      "tokens_per_second_per_gpu": 361.81
    },
    {
      "epoch": 1.9883183377243128,
      "grad_norm": 0.5427475571632385,
      "learning_rate": 1.7148390837784523e-08,
      "loss": 0.1612,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.75,
      "memory/max_allocated (GiB)": 20.75,
      "step": 20000,
      "tokens_per_second_per_gpu": 405.51
    },
    {
      "epoch": 1.9893125217477756,
      "grad_norm": 0.4156535267829895,
      "learning_rate": 1.4365107629710218e-08,
      "loss": 0.1309,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 20010,
      "tokens_per_second_per_gpu": 371.64
    },
    {
      "epoch": 1.9903067057712382,
      "grad_norm": 0.5698441863059998,
      "learning_rate": 1.182811017281793e-08,
      "loss": 0.1373,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 20020,
      "tokens_per_second_per_gpu": 339.51
    },
    {
      "epoch": 1.9913008897947009,
      "grad_norm": 0.7525358200073242,
      "learning_rate": 9.537404716286613e-09,
      "loss": 0.1489,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 19.67,
      "memory/max_allocated (GiB)": 19.67,
      "step": 20030,
      "tokens_per_second_per_gpu": 322.13
    },
    {
      "epoch": 1.9922950738181637,
      "grad_norm": 0.6561192870140076,
      "learning_rate": 7.49299690258054e-09,
      "loss": 0.1278,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.51,
      "memory/max_allocated (GiB)": 21.51,
      "step": 20040,
      "tokens_per_second_per_gpu": 405.65
    },
    {
      "epoch": 1.9932892578416266,
      "grad_norm": 0.28374719619750977,
      "learning_rate": 5.694891767527022e-09,
      "loss": 0.1298,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.76,
      "memory/max_allocated (GiB)": 20.76,
      "step": 20050,
      "tokens_per_second_per_gpu": 347.78
    },
    {
      "epoch": 1.9942834418650892,
      "grad_norm": 0.46046513319015503,
      "learning_rate": 4.1430937402275885e-09,
      "loss": 0.105,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.53,
      "memory/max_allocated (GiB)": 21.53,
      "step": 20060,
      "tokens_per_second_per_gpu": 354.72
    },
    {
      "epoch": 1.9952776258885518,
      "grad_norm": 0.6896146535873413,
      "learning_rate": 2.837606643102397e-09,
      "loss": 0.1675,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.55,
      "memory/max_allocated (GiB)": 20.55,
      "step": 20070,
      "tokens_per_second_per_gpu": 322.76
    },
    {
      "epoch": 1.9962718099120147,
      "grad_norm": 0.29641345143318176,
      "learning_rate": 1.7784336918347244e-09,
      "loss": 0.1681,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 21.49,
      "memory/max_allocated (GiB)": 21.49,
      "step": 20080,
      "tokens_per_second_per_gpu": 359.15
    },
    {
      "epoch": 1.9972659939354775,
      "grad_norm": 0.6377694010734558,
      "learning_rate": 9.655774953931662e-10,
      "loss": 0.164,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.73,
      "memory/max_allocated (GiB)": 20.73,
      "step": 20090,
      "tokens_per_second_per_gpu": 351.54
    },
    {
      "epoch": 1.9982601779589402,
      "grad_norm": 0.5432741641998291,
      "learning_rate": 3.990400559983343e-10,
      "loss": 0.1053,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.74,
      "memory/max_allocated (GiB)": 20.74,
      "step": 20100,
      "tokens_per_second_per_gpu": 346.46
    },
    {
      "epoch": 1.9992543619824028,
      "grad_norm": 0.35591670870780945,
      "learning_rate": 7.882276917836606e-11,
      "loss": 0.2105,
      "memory/device_reserved (GiB)": 22.49,
      "memory/max_active (GiB)": 20.56,
      "memory/max_allocated (GiB)": 20.56,
      "step": 20110,
      "tokens_per_second_per_gpu": 408.89
    }
  ],
  "logging_steps": 10,
  "max_steps": 20117,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 2,
  "save_steps": 500,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": true
      },
      "attributes": {}
    }
  },
  "total_flos": 2.2374951650690335e+18,
  "train_batch_size": 1,
  "trial_name": null,
  "trial_params": null
}