trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 20.0,
  "eval_steps": 500,
  "global_step": 1140,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 8.771929824561404,
      "grad_norm": 5.5376715660095215,
      "learning_rate": 2.8070175438596492e-05,
      "loss": 2.3748,
      "step": 500
    },
    {
      "epoch": 17.54385964912281,
      "grad_norm": 5.5406880378723145,
      "learning_rate": 6.140350877192982e-06,
      "loss": 1.7925,
      "step": 1000
    },
    {
      "epoch": 20.0,
      "step": 1140,
      "total_flos": 1181039984640000.0,
      "train_loss": 2.0282494394402755,
      "train_runtime": 450.0172,
      "train_samples_per_second": 5.022,
      "train_steps_per_second": 2.533
    }
  ],
  "logging_steps": 500,
  "max_steps": 1140,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 20,
  "save_steps": 500,
  "total_flos": 1181039984640000.0,
  "train_batch_size": 2,
  "trial_name": null,
  "trial_params": null
}