checkpoint-500/trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 27.77777777777778,
  "global_step": 500,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 27.78,
      "learning_rate": 3.611111111111111e-05,
      "loss": 0.6295,
      "step": 500
    }
  ],
  "max_steps": 1800,
  "num_train_epochs": 100,
  "total_flos": 259528310784000.0,
  "trial_name": null,
  "trial_params": null
}