trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 615,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 2.44,
      "learning_rate": 9.34959349593496e-06,
      "loss": 5.9409,
      "step": 500
    },
    {
      "epoch": 3.0,
      "step": 615,
      "total_flos": 641994522624000.0,
      "train_loss": 5.800324627635924,
      "train_runtime": 171.8298,
      "train_samples_per_second": 14.299,
      "train_steps_per_second": 3.579
    }
  ],
  "logging_steps": 500,
  "max_steps": 615,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 500,
  "total_flos": 641994522624000.0,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}