Multi-Layer GRPO: Enhancing Reasoning and Self-Correction in Large Language Models
This addresses training stability issues in reasoning tasks for LLM developers, but it is incremental as it builds on the existing GRPO method.
The paper tackles the problem of inefficient exploration and reward vanishing in the GRPO algorithm for enhancing reasoning in large language models by proposing MGRPO, a two-layer approach that adds a self-correction loop, resulting in significant performance improvements on mathematical reasoning benchmarks.
The Group Relative Policy Optimization (GRPO) algorithm has demonstrated considerable success in enhancing the reasoning capabilities of large language models (LLMs), as evidenced by DeepSeek-R1. However, the absence of intermediate supervision in GRPO frequently leads to inefficient exploration dynamics. A single error in a complex reasoning chain can invalidate the entire solution, resulting in abrupt reward vanishing and compromising training stability.To address these challenges, we propose MGRPO (Multi-layer GRPO). MGRPO operates in two layers: the first layer employs standard GRPO to generate an initial response. This response, along with the original query, is then fed into a second-layer GRPO process. This second layer is specifically trained to identify and correct errors in the initial response, effectively creating a self-correction loop. This mechanism provides implicit process-level supervision by rewarding successful error correction, without requiring an explicit, densely-annotated reward model. Experimental results on several mathematical reasoning benchmarks demonstrate that MGRPO significantly outperforms standard GRPO, achieving superior performance by fostering both reasoning and self-correction abilities.