trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 20.0,
  "eval_steps": 500,
  "global_step": 1900,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 5.2631578947368425,
      "grad_norm": 5.739169120788574,
      "learning_rate": 3.6842105263157895e-05,
      "loss": 4.1507,
      "step": 500
    },
    {
      "epoch": 10.526315789473685,
      "grad_norm": 6.05218505859375,
      "learning_rate": 2.368421052631579e-05,
      "loss": 3.5677,
      "step": 1000
    },
    {
      "epoch": 15.789473684210526,
      "grad_norm": 5.807043552398682,
      "learning_rate": 1.0526315789473684e-05,
      "loss": 3.213,
      "step": 1500
    },
    {
      "epoch": 20.0,
      "step": 1900,
      "total_flos": 1975367761920000.0,
      "train_loss": 3.517263697574013,
      "train_runtime": 756.053,
      "train_samples_per_second": 5.0,
      "train_steps_per_second": 2.513
    }
  ],
  "logging_steps": 500,
  "max_steps": 1900,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 20,
  "save_steps": 500,
  "total_flos": 1975367761920000.0,
  "train_batch_size": 2,
  "trial_name": null,
  "trial_params": null
}