trainer_state.json

{
  "best_metric": 0.8170666666666667,
  "best_model_checkpoint": "./resnet_finetuned_models_dataset/CIFAR100/50_from_100/microsoft_resnet-101/model_idx_0340/checkpoints/checkpoint-2664",
  "epoch": 8.0,
  "eval_steps": 500,
  "global_step": 2664,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.0,
      "grad_norm": 77.07249450683594,
      "learning_rate": 2.6666666666666667e-05,
      "loss": 3.8455,
      "step": 333
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.39973333333333333,
      "eval_loss": 3.691957473754883,
      "eval_runtime": 8.5024,
      "eval_samples_per_second": 441.05,
      "eval_steps_per_second": 6.939,
      "step": 333
    },
    {
      "epoch": 2.0,
      "grad_norm": 91.4114761352539,
      "learning_rate": 2.3333333333333336e-05,
      "loss": 3.2379,
      "step": 666
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.5885333333333334,
      "eval_loss": 2.5701327323913574,
      "eval_runtime": 8.3768,
      "eval_samples_per_second": 447.665,
      "eval_steps_per_second": 7.043,
      "step": 666
    },
    {
      "epoch": 3.0,
      "grad_norm": 113.24809265136719,
      "learning_rate": 1.9999999999999998e-05,
      "loss": 1.9498,
      "step": 999
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.7101333333333333,
      "eval_loss": 1.4248946905136108,
      "eval_runtime": 8.3549,
      "eval_samples_per_second": 448.836,
      "eval_steps_per_second": 7.062,
      "step": 999
    },
    {
      "epoch": 4.0,
      "grad_norm": 135.44256591796875,
      "learning_rate": 1.6666666666666667e-05,
      "loss": 1.2241,
      "step": 1332
    },
    {
      "epoch": 4.0,
      "eval_accuracy": 0.7602666666666666,
      "eval_loss": 1.0275640487670898,
      "eval_runtime": 8.0296,
      "eval_samples_per_second": 467.023,
      "eval_steps_per_second": 7.348,
      "step": 1332
    },
    {
      "epoch": 5.0,
      "grad_norm": 165.29348754882812,
      "learning_rate": 1.3333333333333333e-05,
      "loss": 0.9295,
      "step": 1665
    },
    {
      "epoch": 5.0,
      "eval_accuracy": 0.7877333333333333,
      "eval_loss": 0.8442308306694031,
      "eval_runtime": 7.9519,
      "eval_samples_per_second": 471.586,
      "eval_steps_per_second": 7.42,
      "step": 1665
    },
    {
      "epoch": 6.0,
      "grad_norm": 200.5806427001953,
      "learning_rate": 9.999999999999999e-06,
      "loss": 0.7902,
      "step": 1998
    },
    {
      "epoch": 6.0,
      "eval_accuracy": 0.8010666666666667,
      "eval_loss": 0.7582951188087463,
      "eval_runtime": 8.9819,
      "eval_samples_per_second": 417.504,
      "eval_steps_per_second": 6.569,
      "step": 1998
    },
    {
      "epoch": 7.0,
      "grad_norm": 329.13861083984375,
      "learning_rate": 6.666666666666667e-06,
      "loss": 0.6947,
      "step": 2331
    },
    {
      "epoch": 7.0,
      "eval_accuracy": 0.8072,
      "eval_loss": 0.7045201063156128,
      "eval_runtime": 8.8552,
      "eval_samples_per_second": 423.478,
      "eval_steps_per_second": 6.663,
      "step": 2331
    },
    {
      "epoch": 8.0,
      "grad_norm": 102.03043365478516,
      "learning_rate": 3.3333333333333333e-06,
      "loss": 0.6548,
      "step": 2664
    },
    {
      "epoch": 8.0,
      "eval_accuracy": 0.8170666666666667,
      "eval_loss": 0.6785693764686584,
      "eval_runtime": 8.1895,
      "eval_samples_per_second": 457.901,
      "eval_steps_per_second": 7.204,
      "step": 2664
    }
  ],
  "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": 6.5411433916416e+18,
  "train_batch_size": 64,
  "trial_name": null,
  "trial_params": null
}