trainer_state.json

{
  "best_metric": 0.8690666666666667,
  "best_model_checkpoint": "./vit_finetuned_models_dataset/CIFAR100/50_from_100/facebook_dino-vitb16/model_idx_0312/checkpoints/checkpoint-999",
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 999,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.0,
      "grad_norm": 1.104924201965332,
      "learning_rate": 9e-05,
      "loss": 0.8134,
      "step": 333
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.8386666666666667,
      "eval_loss": 0.5615047812461853,
      "eval_runtime": 10.9263,
      "eval_samples_per_second": 343.208,
      "eval_steps_per_second": 5.4,
      "step": 333
    },
    {
      "epoch": 2.0,
      "grad_norm": 0.29490387439727783,
      "learning_rate": 9e-05,
      "loss": 0.3051,
      "step": 666
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.8338666666666666,
      "eval_loss": 0.5867440104484558,
      "eval_runtime": 10.7194,
      "eval_samples_per_second": 349.832,
      "eval_steps_per_second": 5.504,
      "step": 666
    },
    {
      "epoch": 3.0,
      "grad_norm": Infinity,
      "learning_rate": 9e-05,
      "loss": 0.1956,
      "step": 999
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.8690666666666667,
      "eval_loss": 0.5186358094215393,
      "eval_runtime": 10.4022,
      "eval_samples_per_second": 360.501,
      "eval_steps_per_second": 5.672,
      "step": 999
    }
  ],
  "logging_steps": 500,
  "max_steps": 2997,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 9,
  "save_steps": 500,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": false
      },
      "attributes": {}
    }
  },
  "total_flos": 4.94223961867776e+18,
  "train_batch_size": 64,
  "trial_name": null,
  "trial_params": null
}