Part I: Tricks or Traps? A Deep Dive into RL for LLM Reasoning

TL;DR

This paper systematically reviews reinforcement learning techniques for large language model reasoning, providing insights, guidelines, and a simple effective combination that improves performance over existing strategies.

Contribution

It offers a comprehensive analysis of RL methods for LLM reasoning, introduces standardized evaluation protocols, and proposes a minimalist technique combination that enhances learning.

Findings

A unified open-source framework for evaluating RL techniques.

Clear guidelines for selecting RL methods based on experimental insights.

A simple two-technique combination surpasses existing strategies like GRPO and DAPO.

Abstract

Reinforcement learning for LLM reasoning has rapidly emerged as a prominent research area, marked by a significant surge in related studies on both algorithmic innovations and practical applications. Despite this progress, several critical challenges remain, including the absence of standardized guidelines for employing RL techniques and a fragmented understanding of their underlying mechanisms. Additionally, inconsistent experimental settings, variations in training data, and differences in model initialization have led to conflicting conclusions, obscuring the key characteristics of these techniques and creating confusion among practitioners when selecting appropriate techniques. This paper systematically reviews widely adopted RL techniques through rigorous reproductions and isolated evaluations within a unified open-source framework. We analyze the internal mechanisms, applicable scenarios, and core principles of each technique through fine-grained experiments, including datasets of varying difficulty, model sizes, and architectures. Based on these insights, we present clear guidelines for selecting RL techniques tailored to specific setups, and provide a reliable roadmap for practitioners navigating the RL for the LLM domain. Finally, we reveal that a minimalist combination of two techniques can unlock the learning capability of critic-free policies using vanilla PPO loss. The results demonstrate that our simple combination consistently improves performance, surpassing strategies like GRPO and DAPO.