Demystifying Instruction Mixing for Fine-tuning Large Language Models

Renxi Wang, Haonan Li, Minghao Wu +4

Instruction tuning significantly enhances the performance of large language models (LLMs) across various tasks. However, the procedure to optimizing the mixing of instruction datas…