{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.351206434316354,
  "eval_steps": 500,
  "global_step": 2500,
  "is_hyper_param_search": true,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.6702412868632708,
      "grad_norm": 5.694277763366699,
      "learning_rate": 2.542005392773407e-05,
      "loss": 0.557,
      "step": 500
    },
    {
      "epoch": 1.0,
      "eval_accuracy": 0.7254441976547241,
      "eval_loss": 0.5171247720718384,
      "eval_runtime": 8.773,
      "eval_samples_per_second": 340.02,
      "eval_steps_per_second": 21.315,
      "step": 746
    },
    {
      "epoch": 1.3404825737265416,
      "grad_norm": 8.974740028381348,
      "learning_rate": 2.359390062832789e-05,
      "loss": 0.4156,
      "step": 1000
    },
    {
      "epoch": 2.0,
      "eval_accuracy": 0.7596379518508911,
      "eval_loss": 0.6025224924087524,
      "eval_runtime": 8.8883,
      "eval_samples_per_second": 335.609,
      "eval_steps_per_second": 21.039,
      "step": 1492
    },
    {
      "epoch": 2.0107238605898123,
      "grad_norm": 7.7003068923950195,
      "learning_rate": 2.1767747328921705e-05,
      "loss": 0.2948,
      "step": 1500
    },
    {
      "epoch": 2.680965147453083,
      "grad_norm": 20.24570655822754,
      "learning_rate": 1.9941594029515523e-05,
      "loss": 0.1262,
      "step": 2000
    },
    {
      "epoch": 3.0,
      "eval_accuracy": 0.7703654170036316,
      "eval_loss": 0.822274386882782,
      "eval_runtime": 8.8709,
      "eval_samples_per_second": 336.267,
      "eval_steps_per_second": 21.08,
      "step": 2238
    },
    {
      "epoch": 3.351206434316354,
      "grad_norm": 0.9093023538589478,
      "learning_rate": 1.8115440730109338e-05,
      "loss": 0.1012,
      "step": 2500
    }
  ],
  "logging_steps": 500,
  "max_steps": 7460,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 10,
  "save_steps": 500,
  "total_flos": 5940423509036040.0,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": {
    "learning_rate": 2.7246207227140256e-05,
    "per_device_train_batch_size": 16
  }
}