trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 1653,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.91,
      "grad_norm": 4.16880989074707,
      "learning_rate": 3.487598306110103e-05,
      "loss": 0.4004,
      "step": 500
    },
    {
      "epoch": 1.81,
      "grad_norm": 1.7097141742706299,
      "learning_rate": 1.975196612220206e-05,
      "loss": 0.2175,
      "step": 1000
    },
    {
      "epoch": 2.72,
      "grad_norm": 0.8170027136802673,
      "learning_rate": 4.627949183303086e-06,
      "loss": 0.1255,
      "step": 1500
    },
    {
      "epoch": 3.0,
      "step": 1653,
      "total_flos": 1394635774789920.0,
      "train_loss": 0.23321845325209195,
      "train_runtime": 205.592,
      "train_samples_per_second": 128.526,
      "train_steps_per_second": 8.04
    }
  ],
  "logging_steps": 500,
  "max_steps": 1653,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 500,
  "total_flos": 1394635774789920.0,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": null
}