trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 4.0,
  "eval_steps": 40,
  "global_step": 24,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.67,
      "grad_norm": 1.1704062223434448,
      "learning_rate": 0.00011666666666666668,
      "loss": 0.4643,
      "step": 10
    },
    {
      "epoch": 3.33,
      "grad_norm": 0.49445489048957825,
      "learning_rate": 3.3333333333333335e-05,
      "loss": 0.1939,
      "step": 20
    },
    {
      "epoch": 4.0,
      "step": 24,
      "total_flos": 2.944695605354496e+16,
      "train_loss": 0.2903190416594346,
      "train_runtime": 752.2275,
      "train_samples_per_second": 0.505,
      "train_steps_per_second": 0.032
    }
  ],
  "logging_steps": 10,
  "max_steps": 24,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 40,
  "total_flos": 2.944695605354496e+16,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": null
}