trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 20.0,
  "eval_steps": 100,
  "global_step": 20,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 5.0,
      "learning_rate": 4.267766952966369e-05,
      "loss": 2.4236,
      "step": 5
    },
    {
      "epoch": 10.0,
      "learning_rate": 2.5e-05,
      "loss": 2.3423,
      "step": 10
    },
    {
      "epoch": 15.0,
      "learning_rate": 7.3223304703363135e-06,
      "loss": 2.302,
      "step": 15
    },
    {
      "epoch": 20.0,
      "learning_rate": 0.0,
      "loss": 2.2859,
      "step": 20
    },
    {
      "epoch": 20.0,
      "step": 20,
      "total_flos": 3.87232947634176e+16,
      "train_loss": 2.338414525985718,
      "train_runtime": 453.0464,
      "train_samples_per_second": 0.662,
      "train_steps_per_second": 0.044
    }
  ],
  "logging_steps": 5,
  "max_steps": 20,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 20,
  "save_steps": 100,
  "total_flos": 3.87232947634176e+16,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}