trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 100.0,
  "eval_steps": 500,
  "global_step": 2000,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 25.0,
      "learning_rate": 3.7500000000000003e-05,
      "loss": 0.4884,
      "step": 500
    },
    {
      "epoch": 50.0,
      "learning_rate": 2.5e-05,
      "loss": 0.0171,
      "step": 1000
    },
    {
      "epoch": 75.0,
      "learning_rate": 1.25e-05,
      "loss": 0.0058,
      "step": 1500
    },
    {
      "epoch": 100.0,
      "learning_rate": 0.0,
      "loss": 0.0033,
      "step": 2000
    },
    {
      "epoch": 100.0,
      "step": 2000,
      "total_flos": 9.7986283776e+16,
      "train_loss": 0.12865829187631608,
      "train_runtime": 2954.7897,
      "train_samples_per_second": 169.217,
      "train_steps_per_second": 0.677
    }
  ],
  "logging_steps": 500,
  "max_steps": 2000,
  "num_train_epochs": 100,
  "save_steps": 500,
  "total_flos": 9.7986283776e+16,
  "trial_name": null,
  "trial_params": null
}