run-21/checkpoint-2108/trainer_state.json

{
  "best_metric": 0.8360091743119266,
  "best_model_checkpoint": "tiny-bert-sst2-distilled/run-21/checkpoint-2108",
  "epoch": 4.0,
  "eval_steps": 500,
  "global_step": 2108,
  "is_hyper_param_search": true,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.0,
      "grad_norm": 16.87346649169922,
      "learning_rate": 0.00013104269686993426,
      "loss": 1.2895,
      "step": 527
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.8222477064220184,
      "eval_loss": 1.1431264877319336,
      "eval_runtime": 2.8225,
      "eval_samples_per_second": 308.945,
      "eval_steps_per_second": 2.48,
      "step": 527
    },
    {
      "epoch": 2.0,
      "grad_norm": 7.701562881469727,
      "learning_rate": 0.00010920224739161189,
      "loss": 0.6357,
      "step": 1054
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.8279816513761468,
      "eval_loss": 1.1658384799957275,
      "eval_runtime": 2.8051,
      "eval_samples_per_second": 310.866,
      "eval_steps_per_second": 2.495,
      "step": 1054
    },
    {
      "epoch": 3.0,
      "grad_norm": NaN,
      "learning_rate": 8.740324089142675e-05,
      "loss": 0.4772,
      "step": 1581
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.8279816513761468,
      "eval_loss": 1.2185826301574707,
      "eval_runtime": 2.8164,
      "eval_samples_per_second": 309.615,
      "eval_steps_per_second": 2.485,
      "step": 1581
    },
    {
      "epoch": 4.0,
      "grad_norm": 7.373347759246826,
      "learning_rate": 6.556279141310437e-05,
      "loss": 0.3914,
      "step": 2108
    },
    {
      "epoch": 4.0,
      "eval_accuracy": 0.8360091743119266,
      "eval_loss": 1.2047698497772217,
      "eval_runtime": 2.8054,
      "eval_samples_per_second": 310.831,
      "eval_steps_per_second": 2.495,
      "step": 2108
    }
  ],
  "logging_steps": 500,
  "max_steps": 3689,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 7,
  "save_steps": 500,
  "total_flos": 32337743850420.0,
  "train_batch_size": 128,
  "trial_name": null,
  "trial_params": {
    "alpha": 0.576065854907962,
    "learning_rate": 0.00015288314634825665,
    "num_train_epochs": 7,
    "temperature": 26
  }
}