run-16/checkpoint-144/trainer_state.json

{
  "best_metric": 0.7671232876712328,
  "best_model_checkpoint": "tiny-bert-sst2-distilled/run-16/checkpoint-144",
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 144,
  "is_hyper_param_search": true,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.0,
      "grad_norm": 1.2878928184509277,
      "learning_rate": 0.00013115653902338105,
      "loss": 0.4962,
      "step": 48
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.700587084148728,
      "eval_f1": 0.662251655629139,
      "eval_loss": 0.4426267147064209,
      "eval_precision": 0.759493670886076,
      "eval_recall": 0.5870841487279843,
      "eval_runtime": 27.3,
      "eval_samples_per_second": 37.436,
      "eval_steps_per_second": 0.586,
      "step": 48
    },
    {
      "epoch": 2.0,
      "grad_norm": 3.258302688598633,
      "learning_rate": 0.00011241989059146946,
      "loss": 0.4374,
      "step": 96
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.6898238747553816,
      "eval_f1": 0.6248520710059171,
      "eval_loss": 0.41270217299461365,
      "eval_precision": 0.7904191616766467,
      "eval_recall": 0.5166340508806262,
      "eval_runtime": 27.1797,
      "eval_samples_per_second": 37.602,
      "eval_steps_per_second": 0.589,
      "step": 96
    },
    {
      "epoch": 3.0,
      "grad_norm": 1.6551486253738403,
      "learning_rate": 9.368324215955789e-05,
      "loss": 0.4148,
      "step": 144
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.7671232876712328,
      "eval_f1": 0.7643564356435644,
      "eval_loss": 0.39426228404045105,
      "eval_precision": 0.7735470941883767,
      "eval_recall": 0.7553816046966731,
      "eval_runtime": 28.5963,
      "eval_samples_per_second": 35.739,
      "eval_steps_per_second": 0.56,
      "step": 144
    }
  ],
  "logging_steps": 500,
  "max_steps": 384,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 8,
  "save_steps": 500,
  "total_flos": 707085591840.0,
  "train_batch_size": 64,
  "trial_name": null,
  "trial_params": {
    "alpha": 0.6758158495370239,
    "learning_rate": 0.00014989318745529262,
    "num_train_epochs": 8,
    "temperature": 20
  }
}