trainer_state.json

{
  "best_metric": 0.544,
  "best_model_checkpoint": "./vit_finetuned_models_dataset/CIFAR100/50_from_100/facebook_vit-mae-base/model_idx_0493/checkpoints/checkpoint-1332",
  "epoch": 4.0,
  "eval_steps": 500,
  "global_step": 1332,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.0,
      "grad_norm": Infinity,
      "learning_rate": 0.0002252252252252252,
      "loss": 3.2371,
      "step": 333
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.244,
      "eval_loss": 2.784247875213623,
      "eval_runtime": 12.2541,
      "eval_samples_per_second": 306.019,
      "eval_steps_per_second": 4.815,
      "step": 333
    },
    {
      "epoch": 2.0,
      "grad_norm": Infinity,
      "learning_rate": 0.00015045045045045046,
      "loss": 2.5035,
      "step": 666
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.376,
      "eval_loss": 2.2615201473236084,
      "eval_runtime": 12.1924,
      "eval_samples_per_second": 307.568,
      "eval_steps_per_second": 4.839,
      "step": 666
    },
    {
      "epoch": 3.0,
      "grad_norm": Infinity,
      "learning_rate": 7.567567567567566e-05,
      "loss": 1.9017,
      "step": 999
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.4789333333333333,
      "eval_loss": 1.8625158071517944,
      "eval_runtime": 12.0732,
      "eval_samples_per_second": 310.606,
      "eval_steps_per_second": 4.887,
      "step": 999
    },
    {
      "epoch": 4.0,
      "grad_norm": 18.77025604248047,
      "learning_rate": 6.756756756756756e-07,
      "loss": 1.3171,
      "step": 1332
    },
    {
      "epoch": 4.0,
      "eval_accuracy": 0.544,
      "eval_loss": 1.6450669765472412,
      "eval_runtime": 12.2351,
      "eval_samples_per_second": 306.496,
      "eval_steps_per_second": 4.822,
      "step": 1332
    }
  ],
  "logging_steps": 500,
  "max_steps": 1332,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 500,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": true
      },
      "attributes": {}
    }
  },
  "total_flos": 6.58965282490368e+18,
  "train_batch_size": 64,
  "trial_name": null,
  "trial_params": null
}