training_args.json

{
    "output_dir": "./results",
    "evaluation_strategy": "steps",
    "eval_steps": 500,
    "save_strategy": "steps",
    "save_steps": 500,
    "learning_rate": 5e-5,
    "per_device_train_batch_size": 4,
    "per_device_eval_batch_size": 4,
    "gradient_accumulation_steps": 8,
    "num_train_epochs": 10,
    "weight_decay": 0.01,
    "warmup_ratio": 0.1,
    "logging_dir": "./logs",
    "logging_steps": 100,
    "save_total_limit": 3,
    "load_best_model_at_end": true,
    "metric_for_best_model": "eval_loss",
    "greater_is_better": false,
    "fp16": true,
    "dataloader_pin_memory": false,
    "report_to": "none",
    "remove_unused_columns": false
}