models/bert/force/trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 12.0,
  "global_step": 816,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 7.35,
      "learning_rate": 7.745098039215687e-06,
      "loss": 0.3037,
      "step": 500
    },
    {
      "epoch": 12.0,
      "step": 816,
      "total_flos": 623013398424576.0,
      "train_loss": 0.24204148965723374,
      "train_runtime": 188.5058,
      "train_samples_per_second": 553.129,
      "train_steps_per_second": 4.329
    }
  ],
  "max_steps": 816,
  "num_train_epochs": 12,
  "total_flos": 623013398424576.0,
  "trial_name": null,
  "trial_params": null
}