Critique-RL: Training Language Models for Critiquing through Two-Stage Reinforcement Learning

TL;DR

Critique-RL introduces a two-stage reinforcement learning method to train critiquing language models without strong supervision, significantly improving their ability to assess and provide feedback on model outputs across various tasks.

Contribution

It proposes a novel two-stage RL approach that enhances critiquing models' discriminability and helpfulness without requiring stronger supervision signals.

Findings

Achieves 9.02% improvement on in-domain tasks

Achieves 5.70% improvement on out-of-domain tasks

Enhances critic helpfulness and discriminability through two-stage optimization

Abstract

Training critiquing language models to assess and provide feedback on model outputs is a promising way to improve LLMs for complex reasoning tasks. However, existing approaches typically rely on stronger supervisors for annotating critique data. To address this, we propose Critique-RL, an online RL approach for developing critiquing language models without stronger supervision. Our approach operates on a two-player paradigm: the actor generates a response, the critic provides feedback, and the actor refines the response accordingly. We first reveal that relying solely on indirect reward signals from the actor's outputs for RL optimization often leads to unsatisfactory critics: while their helpfulness (i.e., providing constructive feedback) improves, the discriminability (i.e., determining whether a response is high-quality or not) remains poor, resulting in marginal performance gains. To overcome this, Critique-RL adopts a two-stage optimization strategy. In stage I, it reinforces the discriminability of the critic with direct rule-based reward signals; in stage II, it introduces indirect rewards based on actor refinement to improve the critic's helpfulness, while maintaining its discriminability via appropriate regularization. Extensive experiments across various tasks and models show that Critique-RL delivers substantial performance improvements. For example, it achieves a 9.02% gain on in-domain tasks and a 5.70% gain on out-of-domain tasks for Qwen2.5-7B, highlighting its potential.