A Common Pitfall of Margin-based Language Model Alignment: Gradient Entanglement

TL;DR

This paper uncovers a fundamental flaw in margin-based language model alignment methods, where gradient entanglement causes unintended safety and performance issues, and proposes theoretical and empirical insights to address this problem.

Contribution

The paper identifies gradient entanglement as a key issue in margin-based LM alignment, providing theoretical conditions and empirical validation, and suggesting algorithmic improvements.

Findings

Margin-based losses couple preferred and dispreferred response probabilities.

Gradient entanglement can cause safety and performance failures.

Theoretical conditions for when gradient entanglement becomes problematic.

Abstract

Reinforcement Learning from Human Feedback (RLHF) has become the predominant approach for language model (LM) alignment. At its core, RLHF uses a margin-based loss for preference optimization, specifying ideal LM behavior only by the difference between preferred and dispreferred responses. In this paper, we identify a common pitfall of margin-based methods -- the under-specification of ideal LM behavior on preferred and dispreferred responses individually, which leads to two unintended consequences as the margin increases: (1) The probability of dispreferred (e.g., unsafe) responses may increase, resulting in potential safety alignment failures. (2) The probability of preferred responses may decrease, even when those responses are ideal. We demystify the reasons behind these problematic behaviors: margin-based losses couple the change in the preferred probability to the gradient of the dispreferred one, and vice versa, often preventing the preferred probability from increasing while the dispreferred one decreases, and thus causing a synchronized increase or decrease in both probabilities. We term this effect, inherent in margin-based objectives, gradient entanglement. Formally, we derive conditions for general margin-based alignment objectives under which gradient entanglement becomes concerning: the inner product of the gradients of preferred and dispreferred log-probabilities is large relative to the individual gradient norms. We theoretically investigate why such inner products can be large when aligning language models and empirically validate our findings. Empirical implications of our framework extend to explaining important differences in the training dynamics of various preference optimization algorithms, and suggesting potential algorithm designs to mitigate the under-specification issue of margin-based methods and thereby improving language model alignment.