This website requires JavaScript.

Mitigating the Alignment Tax of RLHF

Yong LinHangyu LinWei Xiong ...+13 Tong Zhang
Feb 2024
LLMs acquire a wide range of abilities during pre-training, but aligning LLMs under Reinforcement Learning with Human Feedback (RLHF) can lead to forgetting, which is also known as the alignment tax. To empirically verify this hypothesis, we conducted experiments with existing RLHF algorithms using OpenLLaMA-3B, which revealed a pronounced alignment tax in NLP tasks. On the other hand, despite various techniques to mitigate forgetting, they are often at odds with the RLHF performance, leading to a trade-off between reward maximization and forgetting mitigation. In light of the above pressing issue in aligning LLMs, in this paper we explore model averaging, which interpolates between pre and post RLHF model weights, to achieve a more efficient reward-tax Pareto front. To understand its effectiveness, We offer theoretical insights into model averaging, revealing that it enhances performance Pareto front by increasing feature diversity on the layers where tasks share overlapped feature spaces. Empirical evidence corroborates our analysis by showing the benefits of averaging low-level transformer layers. Building on the analysis and the observation that averaging different layers of the transformer leads to significantly different reward-tax trade-offs, we propose Adaptive Model Averaging (AMA) to adaptively find various combination ratios of model layers. AMA seeks to maximize the alignment reward while incurring minimal alignment tax. Moreover, we validate AMA's performance across a range of RLHF algorithms over OpenLLaMA-3B and further extend our findings to Mistral-7B.
发布时间 · 被引用数 · 默认排序
发布时间 · 被引用数 · 默认排序