Saved in:
Bibliographic Details
Main Authors: Wu, Zhengxuan, Zhang, Yuhao, Qi, Peng, Xu, Yumo, Han, Rujun, Zhang, Yian, Chen, Jifan, Min, Bonan, Huang, Zhiheng
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2407.21417
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909274957938688
author Wu, Zhengxuan
Zhang, Yuhao
Qi, Peng
Xu, Yumo
Han, Rujun
Zhang, Yian
Chen, Jifan
Min, Bonan
Huang, Zhiheng
author_facet Wu, Zhengxuan
Zhang, Yuhao
Qi, Peng
Xu, Yumo
Han, Rujun
Zhang, Yian
Chen, Jifan
Min, Bonan
Huang, Zhiheng
contents 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, instruction-tuned 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 Self-instruction 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.
format Preprint
id arxiv_https___arxiv_org_abs_2407_21417
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Dancing in Chains: Reconciling Instruction Following and Faithfulness in Language Models
Wu, Zhengxuan
Zhang, Yuhao
Qi, Peng
Xu, Yumo
Han, Rujun
Zhang, Yian
Chen, Jifan
Min, Bonan
Huang, Zhiheng
Computation and Language
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, instruction-tuned 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 Self-instruction 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.
title Dancing in Chains: Reconciling Instruction Following and Faithfulness in Language Models
topic Computation and Language
url https://arxiv.org/abs/2407.21417