trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 4.0,
  "eval_steps": 1000,
  "global_step": 16,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 2.5,
      "grad_norm": 2.7402358055114746,
      "learning_rate": 1.8750000000000002e-05,
      "loss": 2.28,
      "step": 10
    },
    {
      "epoch": 4.0,
      "step": 16,
      "total_flos": 9299872197181440.0,
      "train_loss": 2.1698924899101257,
      "train_runtime": 8.7843,
      "train_samples_per_second": 13.661,
      "train_steps_per_second": 1.821
    }
  ],
  "logging_steps": 10,
  "max_steps": 16,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 1000,
  "total_flos": 9299872197181440.0,
  "train_batch_size": 8,
  "trial_name": null,
  "trial_params": null
}