run-32/checkpoint-144/trainer_state.json

{
  "best_metric": 0.8287671232876712,
  "best_model_checkpoint": "tiny-bert-sst2-distilled/run-32/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": 1.7776581048965454,
      "learning_rate": 0.000302143376211333,
      "loss": 0.5977,
      "step": 48
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.7455968688845401,
      "eval_f1": 0.7906602254428341,
      "eval_loss": 0.5077403783798218,
      "eval_precision": 0.6716826265389877,
      "eval_recall": 0.9608610567514677,
      "eval_runtime": 27.3787,
      "eval_samples_per_second": 37.328,
      "eval_steps_per_second": 0.584,
      "step": 48
    },
    {
      "epoch": 2.0,
      "grad_norm": 9.869495391845703,
      "learning_rate": 0.00020142891747422203,
      "loss": 0.4879,
      "step": 96
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.8052837573385518,
      "eval_f1": 0.8062317429406036,
      "eval_loss": 0.45489782094955444,
      "eval_precision": 0.8023255813953488,
      "eval_recall": 0.8101761252446184,
      "eval_runtime": 27.3555,
      "eval_samples_per_second": 37.36,
      "eval_steps_per_second": 0.585,
      "step": 96
    },
    {
      "epoch": 3.0,
      "grad_norm": 4.101070880889893,
      "learning_rate": 0.00010071445873711101,
      "loss": 0.4431,
      "step": 144
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.8287671232876712,
      "eval_f1": 0.8381128584643849,
      "eval_loss": 0.4164314568042755,
      "eval_precision": 0.7947368421052632,
      "eval_recall": 0.8864970645792564,
      "eval_runtime": 27.378,
      "eval_samples_per_second": 37.329,
      "eval_steps_per_second": 0.584,
      "step": 144
    }
  ],
  "logging_steps": 500,
  "max_steps": 192,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 500,
  "total_flos": 707085591840.0,
  "train_batch_size": 64,
  "trial_name": null,
  "trial_params": {
    "alpha": 0.9963289128276592,
    "learning_rate": 0.00040285783494844405,
    "num_train_epochs": 4,
    "temperature": 26
  }
}