{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.351206434316354,
  "eval_steps": 500,
  "global_step": 2500,
  "is_hyper_param_search": true,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.6702412868632708,
      "grad_norm": 23.28346824645996,
      "learning_rate": 9.798511296679954e-06,
      "loss": 0.5531,
      "step": 500
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.8025477528572083,
      "eval_loss": 0.4079710841178894,
      "eval_runtime": 9.2612,
      "eval_samples_per_second": 322.097,
      "eval_steps_per_second": 20.192,
      "step": 746
    },
    {
      "epoch": 1.3404825737265416,
      "grad_norm": 38.64032745361328,
      "learning_rate": 9.094595255251797e-06,
      "loss": 0.3957,
      "step": 1000
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.8276902437210083,
      "eval_loss": 0.42224231362342834,
      "eval_runtime": 9.5099,
      "eval_samples_per_second": 313.673,
      "eval_steps_per_second": 19.664,
      "step": 1492
    },
    {
      "epoch": 2.0107238605898123,
      "grad_norm": 34.970176696777344,
      "learning_rate": 8.390679213823639e-06,
      "loss": 0.3185,
      "step": 1500
    },
    {
      "epoch": 2.680965147453083,
      "grad_norm": 13.735482215881348,
      "learning_rate": 7.686763172395482e-06,
      "loss": 0.1887,
      "step": 2000
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.819309413433075,
      "eval_loss": 0.49816903471946716,
      "eval_runtime": 9.3819,
      "eval_samples_per_second": 317.952,
      "eval_steps_per_second": 19.932,
      "step": 2238
    },
    {
      "epoch": 3.351206434316354,
      "grad_norm": 10.144265174865723,
      "learning_rate": 6.982847130967324e-06,
      "loss": 0.1522,
      "step": 2500
    }
  ],
  "logging_steps": 500,
  "max_steps": 7460,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 10,
  "save_steps": 500,
  "total_flos": 6596358998624820.0,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": {
    "learning_rate": 1.0502427338108112e-05,
    "per_device_train_batch_size": 16
  }
}