last-checkpoint/trainer_state.json

{
  "best_metric": 7.173837184906006,
  "best_model_checkpoint": "/home/pcjf/CESGA/works/lmodels/models/large/checkpoint-500",
  "epoch": 0.4516201874223778,
  "global_step": 500,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.45,
      "learning_rate": 9.698885877747666e-05,
      "loss": 7.2539,
      "step": 500
    },
    {
      "epoch": 0.45,
      "eval_loss": 7.173837184906006,
      "eval_runtime": 53.1876,
      "eval_samples_per_second": 216.065,
      "eval_steps_per_second": 27.018,
      "step": 500
    }
  ],
  "max_steps": 16605,
  "num_train_epochs": 15,
  "total_flos": 2.352130715179123e+16,
  "trial_name": null,
  "trial_params": null
}