SeRL: Self-Play Reinforcement Learning for Large Language Models with Limited Data

TL;DR

SeRL introduces a self-play reinforcement learning approach that enables large language models to improve reasoning skills with limited data by generating instructions and rewards internally, reducing reliance on external annotations.

Contribution

SeRL presents a novel self-play RL framework with self-instruction and self-reward modules, effectively training LLMs in data-scarce environments.

Findings

Outperforms existing methods on reasoning benchmarks

Achieves results comparable to high-quality data approaches

Effective in diverse LLM architectures

Abstract

Recent advances have demonstrated the effectiveness of Reinforcement Learning (RL) in improving the reasoning capabilities of Large Language Models (LLMs). However, existing works inevitably rely on high-quality instructions and verifiable rewards for effective training, both of which are often difficult to obtain in specialized domains. In this paper, we propose Self-play Reinforcement Learning (SeRL) to bootstrap LLM training with limited initial data. Specifically, SeRL comprises two complementary modules: self-instruction and self-rewarding. The former module generates additional instructions based on the available data at each training step, employing robust online filtering strategies to ensure instruction quality, diversity, and difficulty. The latter module introduces a simple yet effective majority-voting mechanism to estimate response rewards for additional instructions, eliminating the need for external annotations. Finally, SeRL performs conventional RL based on the generated data, facilitating iterative self-play learning. Extensive experiments on various reasoning benchmarks and across different LLM backbones demonstrate that the proposed SeRL yields results superior to its counterparts and achieves performance on par with those obtained by high-quality data with verifiable rewards. Our code is available at https://github.com/wantbook-book/SeRL.