Dancing in Chains: Reconciling Instruction Following and Faithfulness in Language Models
Zhengxuan Wu, Yuhao Zhang, Peng Qi, Yumo Xu, Rujun Han, Yian Zhang, Jifan Chen, Bonan Min, Zhiheng Huang · 2024
Modern language models (LMs) need to follow human instructions while being faithful; yet, they often fail to achieve both.Here, we provide concrete evidence of a trade-off between instruction following (i.e., follow open-ended instructions) and faithfulness (i.e., ground responses in given context) when training LMs with these objectives.For instance, fine-tuning LLaMA-7B on instruction following datasets renders it less faithful.Conversely, instructiontuned Vicuna-7B shows degraded performance at following instructions when further optimized on tasks that require contextual grounding.One common remedy is multi-task learning (MTL) with data mixing, yet it remains far from achieving a synergic outcome.We propose a simple yet effective method that relies on Rejection Sampling for Continued Selfinstruction Tuning (RESET), which significantly outperforms vanilla MTL.Surprisingly, we find that less is more, as training RESET with high-quality, yet substantially smaller data (three-fold less) yields superior results.Our findings offer a better understanding of objective discrepancies in alignment training of LMs.