trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 588,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 2.55,
      "grad_norm": 1.3594532012939453,
      "learning_rate": 7.482993197278912e-06,
      "loss": 0.4814,
      "step": 500
    },
    {
      "epoch": 3.0,
      "step": 588,
      "total_flos": 2.0027592e+16,
      "train_loss": 0.4145632661929747,
      "train_runtime": 941.5846,
      "train_samples_per_second": 159.306,
      "train_steps_per_second": 0.624
    }
  ],
  "logging_steps": 500,
  "max_steps": 588,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 500,
  "total_flos": 2.0027592e+16,
  "train_batch_size": 256,
  "trial_name": null,
  "trial_params": null
}