Enhancing Downstream Fine-Tuning of Large Language Models through Empirical Sample Design Engineering

Careful design of training samples can significantly improve the downstream performance of large language models, beyond the impact of prompt engineering.