checkpoint-2000/trainer_state.json

{
  "best_metric": 0.3950195480446223,
  "best_model_checkpoint": "output/pretraining/vihealthbert-w_unsup-SynPD/lr3e-5_wr0.1_wd0.0/checkpoint-2000",
  "epoch": 0.34464931931759435,
  "eval_steps": 2000,
  "global_step": 2000,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.00017232465965879716,
      "grad_norm": 35.608882904052734,
      "learning_rate": 5.169739789763915e-09,
      "loss": 10.7382,
      "step": 1
    },
    {
      "epoch": 0.34464931931759435,
      "grad_norm": 4.884454250335693,
      "learning_rate": 1.033947957952783e-05,
      "loss": 5.8543,
      "step": 2000
    },
    {
      "epoch": 0.34464931931759435,
      "eval_accuracy": 0.3950195480446223,
      "eval_loss": 3.896709680557251,
      "eval_runtime": 24.6709,
      "eval_samples_per_second": 396.175,
      "eval_steps_per_second": 24.766,
      "step": 2000
    }
  ],
  "logging_steps": 2000,
  "max_steps": 58030,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 10,
  "save_steps": 2000,
  "total_flos": 0.0,
  "train_batch_size": 32,
  "trial_name": null,
  "trial_params": null
}