Sign up to save your podcasts
Or
Share Teaching Language Models to Critique via Reinforcement Learning
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
Teaching Language Models to Critique via Reinforcement Learning
🤗 Upvotes: 16 | cs.LG, cs.AI, cs.CL
Authors:
Zhihui Xie, Jie chen, Liyu Chen, Weichao Mao, Jingjing Xu, Lingpeng Kong
Title:
Teaching Language Models to Critique via Reinforcement Learning
Arxiv:
http://arxiv.org/abs/2502.03492v1
Abstract:
Teaching large language models (LLMs) to critique and refine their outputs is crucial for building systems that can iteratively improve, yet it is fundamentally limited by the ability to provide accurate judgments and actionable suggestions. In this work, we study LLM critics for code generation and propose $\texttt{CTRL}$, a framework for $\texttt{C}$ritic $\texttt{T}$raining via $\texttt{R}$einforcement $\texttt{L}$earning, which trains a critic model to generate feedback that maximizes correction performance for a fixed generator model without human supervision. Our results demonstrate that critics trained with $\texttt{CTRL}$ significantly enhance pass rates and mitigate compounding errors across both base and stronger generator models. Furthermore, we show that these critic models act as accurate generative reward models and enable test-time scaling through iterative critique-revision, achieving up to 106.1% relative improvements across challenging code generation benchmarks.
View all episodes
By Jingwen Liang, Gengyu Wang🤗 Upvotes: 16 | cs.LG, cs.AI, cs.CL
Authors:
Zhihui Xie, Jie chen, Liyu Chen, Weichao Mao, Jingjing Xu, Lingpeng Kong
Title:
Teaching Language Models to Critique via Reinforcement Learning
Arxiv:
http://arxiv.org/abs/2502.03492v1
Abstract:
Teaching large language models (LLMs) to critique and refine their outputs is crucial for building systems that can iteratively improve, yet it is fundamentally limited by the ability to provide accurate judgments and actionable suggestions. In this work, we study LLM critics for code generation and propose $\texttt{CTRL}$, a framework for $\texttt{C}$ritic $\texttt{T}$raining via $\texttt{R}$einforcement $\texttt{L}$earning, which trains a critic model to generate feedback that maximizes correction performance for a fixed generator model without human supervision. Our results demonstrate that critics trained with $\texttt{CTRL}$ significantly enhance pass rates and mitigate compounding errors across both base and stronger generator models. Furthermore, we show that these critic models act as accurate generative reward models and enable test-time scaling through iterative critique-revision, achieving up to 106.1% relative improvements across challenging code generation benchmarks.