RLPR: Extrapolating RLVR to General Domains without Verifiers
This addresses the problem of scaling RLVR to general domains for AI researchers and practitioners by eliminating the need for domain-specific verifiers, though it is an incremental improvement building on existing RLVR concepts.
The paper tackles the limitation of Reinforcement Learning with Verifiable Rewards (RLVR) being confined to mathematical and code domains due to reliance on domain-specific verifiers, proposing RLPR, a verifier-free framework that uses LLMs' own token probability scores as reward signals to improve reasoning capabilities. Results show RLPR outperforms concurrent methods by up to 7.6 points on benchmarks like TheoremQA and Minerva, and surpasses verifier-dependent approaches by 1.6 average points across seven benchmarks.
Reinforcement Learning with Verifiable Rewards (RLVR) demonstrates promising potential in advancing the reasoning capabilities of LLMs. However, its success remains largely confined to mathematical and code domains. This primary limitation stems from the heavy reliance on domain-specific verifiers, which results in prohibitive complexity and limited scalability. To address the challenge, our key observation is that LLM's intrinsic probability of generating a correct free-form answer directly indicates its own evaluation of the reasoning reward (i.e., how well the reasoning process leads to the correct answer). Building on this insight, we propose RLPR, a simple verifier-free framework that extrapolates RLVR to broader general domains. RLPR uses the LLM's own token probability scores for reference answers as the reward signal and maximizes the expected reward during training. We find that addressing the high variance of this noisy probability reward is crucial to make it work, and propose prob-to-reward and stabilizing methods to ensure a precise and stable reward from LLM intrinsic probabilities. Comprehensive experiments in four general-domain benchmarks and three mathematical benchmarks show that RLPR consistently improves reasoning capabilities in both areas for Gemma, Llama, and Qwen based models. Notably, RLPR outperforms concurrent VeriFree by 7.6 points on TheoremQA and 7.5 points on Minerva, and even surpasses strong verifier-model-dependent approaches General-Reasoner by 1.6 average points across seven benchmarks.