tests/llm/config/sft.json

{
    "model_type": "qwen-1_8b-chat",
    "dataset": "jd-sentiment-zh",
    "output_dir": "output/pai_test",
    "train_dataset_sample": 100,
    "eval_steps": 5
}