CRScore++: Reinforcement Learning with Verifiable Tool and AI Feedback for Code Review

TL;DR

CRScore++ is a reinforcement learning framework that combines verifiable signals and AI feedback to enhance code review comment generation, improving model performance and generalization across programming languages.

Contribution

It introduces CRScore++, which integrates verifiable tools and AI feedback into RL for code review, bridging the gap between structured and unstructured feedback methods.

Findings

CRScore++ improves code review comment quality.

The framework enables generalization to new programming languages.

Combines supervised fine-tuning with RL critique for better performance.

Abstract

Reinforcement learning (RL) to improve code review comment generation requires handling unstructured outputs, making reinforcement learning (RL) feedback challenging. The two main RL approaches, namely RL with Verifiable Feedback (RLVR) and RL with AI Feedback (RLAIF), offer trade-offs: RLVR provides reliable feedback for structured tasks like code generation, while RLAIF works for unstructured outputs but is subjective. We bridge this gap with CRScore++, an RL framework that leverages both LLM-based subjective feedback and verifiable signals for training. Extending CRScore, a code review evaluation metric integrating LLMs with verifiers like linters and code smell detectors, CRScore++ transforms these signals into training rewards. We show that CRScore++ improves a weaker student model through a combination of supervised fine-tuning and RL critique from a stronger teacher model, thus enabling generalization to novel programming languages.