trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 1704,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.88,
      "grad_norm": 12.444563865661621,
      "learning_rate": 3.532863849765258e-05,
      "loss": 3.0578,
      "step": 500
    },
    {
      "epoch": 1.76,
      "grad_norm": 8.964831352233887,
      "learning_rate": 2.0657276995305167e-05,
      "loss": 2.7481,
      "step": 1000
    },
    {
      "epoch": 2.64,
      "grad_norm": 16.21776008605957,
      "learning_rate": 5.9859154929577465e-06,
      "loss": 2.2776,
      "step": 1500
    },
    {
      "epoch": 3.0,
      "step": 1704,
      "total_flos": 7276024173600000.0,
      "train_loss": 2.612193721001137,
      "train_runtime": 276.8012,
      "train_samples_per_second": 196.874,
      "train_steps_per_second": 6.156
    }
  ],
  "logging_steps": 500,
  "max_steps": 1704,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 500,
  "total_flos": 7276024173600000.0,
  "train_batch_size": 32,
  "trial_name": null,
  "trial_params": null
}