trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 4.0,
  "eval_steps": 500,
  "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": 0.5573725700378418,
      "learning_rate": 1.0000000000000002e-06,
      "loss": 0.3765,
      "step": 10
    },
    {
      "epoch": 2.86,
      "grad_norm": 0.4425622224807739,
      "learning_rate": 2.0000000000000003e-06,
      "loss": 0.4049,
      "step": 20
    }
  ],
  "logging_steps": 10,
  "max_steps": 28,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 500,
  "total_flos": 121791578112000.0,
  "train_batch_size": 8,
  "trial_name": null,
  "trial_params": null
}