training_args.json

{
  "output_dir": "./results",
  "overwrite_output_dir": True,
  "max_steps": 200000,
  "per_device_train_batch_size": 32,
  "per_device_eval_batch_size": 32,
  "gradient_accumulation_steps": 1,
  "learning_rate": 0.0001,
  "weight_decay": 0.0,
  "adam_beta1": 0.9,
  "adam_beta2": 0.999,
  "adam_epsilon": 1e-08,
  "max_grad_norm": 1.0,
  "num_train_epochs": 20,
  "warmup_steps": 500,
  "logging_steps": 1000,
  "save_steps": 5000,
  "eval_steps": 5000,
  "load_best_model_at_end": True,
  "metric_for_best_model": "eval_loss",
  "greater_is_better": False,
  "fp16": True,
  "fp16_backend": "auto",
  "seed": 42
}