trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 4.0,
  "eval_steps": 100,
  "global_step": 28,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.43,
      "grad_norm": 1.993449091911316,
      "learning_rate": 0.00012857142857142858,
      "loss": 1.9695,
      "step": 10
    },
    {
      "epoch": 2.86,
      "grad_norm": 1.4520915746688843,
      "learning_rate": 5.714285714285714e-05,
      "loss": 1.524,
      "step": 20
    },
    {
      "epoch": 4.0,
      "step": 28,
      "total_flos": 3.0997907103744e+16,
      "train_loss": 1.6283700466156006,
      "train_runtime": 32.1209,
      "train_samples_per_second": 12.453,
      "train_steps_per_second": 0.872
    }
  ],
  "logging_steps": 10,
  "max_steps": 28,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 100,
  "total_flos": 3.0997907103744e+16,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": null
}