trainer_state.json

{
  "best_metric": 0.7840577960014343,
  "best_model_checkpoint": "/lustre/projects/cardiff/husu_expts/results/expt3/checkpoint-2000",
  "epoch": 10.52553785514863,
  "global_step": 2000,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 5.26,
      "learning_rate": 3.7851351351351354e-05,
      "loss": 1.2598,
      "step": 1000
    },
    {
      "epoch": 5.26,
      "eval_loss": 0.7679372429847717,
      "eval_runtime": 1717.5381,
      "eval_samples_per_second": 911.289,
      "eval_steps_per_second": 14.239,
      "step": 1000
    },
    {
      "epoch": 10.53,
      "learning_rate": 2.433783783783784e-05,
      "loss": 0.5466,
      "step": 2000
    },
    {
      "epoch": 10.53,
      "eval_loss": 0.7840577960014343,
      "eval_runtime": 1693.2394,
      "eval_samples_per_second": 924.366,
      "eval_steps_per_second": 14.443,
      "step": 2000
    }
  ],
  "max_steps": 3800,
  "num_train_epochs": 20,
  "total_flos": 3.12611882139648e+17,
  "trial_name": null,
  "trial_params": null
}