run-15/checkpoint-288/trainer_state.json

{
  "best_metric": 0.812133072407045,
  "best_model_checkpoint": "tiny-bert-sst2-distilled/run-15/checkpoint-288",
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 288,
  "is_hyper_param_search": true,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.0,
      "grad_norm": 1.150837779045105,
      "learning_rate": 0.00014120530022151875,
      "loss": 0.5007,
      "step": 96
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.7681017612524462,
      "eval_f1": 0.7889581478183438,
      "eval_loss": 0.4356862008571625,
      "eval_precision": 0.7238562091503268,
      "eval_recall": 0.8669275929549902,
      "eval_runtime": 28.2732,
      "eval_samples_per_second": 36.147,
      "eval_steps_per_second": 1.132,
      "step": 96
    },
    {
      "epoch": 2.0,
      "grad_norm": 2.0905497074127197,
      "learning_rate": 9.413686681434584e-05,
      "loss": 0.4302,
      "step": 192
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.7984344422700587,
      "eval_f1": 0.81993006993007,
      "eval_loss": 0.4125403165817261,
      "eval_precision": 0.740916271721959,
      "eval_recall": 0.9178082191780822,
      "eval_runtime": 29.852,
      "eval_samples_per_second": 34.236,
      "eval_steps_per_second": 1.072,
      "step": 192
    },
    {
      "epoch": 3.0,
      "grad_norm": 2.322948455810547,
      "learning_rate": 4.706843340717292e-05,
      "loss": 0.4123,
      "step": 288
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.812133072407045,
      "eval_f1": 0.8303886925795053,
      "eval_loss": 0.40187525749206543,
      "eval_precision": 0.7568438003220612,
      "eval_recall": 0.9197651663405088,
      "eval_runtime": 28.0204,
      "eval_samples_per_second": 36.473,
      "eval_steps_per_second": 1.142,
      "step": 288
    }
  ],
  "logging_steps": 500,
  "max_steps": 384,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 500,
  "total_flos": 707085591840.0,
  "train_batch_size": 32,
  "trial_name": null,
  "trial_params": {
    "alpha": 0.7235742113106859,
    "learning_rate": 0.00018827373362869168,
    "num_train_epochs": 4,
    "temperature": 25
  }
}