Transformation-Augmented GRPO for Enhancing Exploration in Reasoning of Large Language Models

TL;DR

TA-GRPO enhances exploration in large language models by using question rephrasing to generate diverse responses, addressing gradient vanishing and diversity collapse issues in reinforcement learning.

Contribution

It introduces a simple method that automatically generates question rephrasings to improve exploration and diversity in reinforcement learning for large language models.

Findings

TA-GRPO improves pass@$k$ on multiple benchmarks.

It increases average pass@32 by 4.97 and 4.34 points for two models.

It matches exploration quality of larger-data baselines.

Abstract

Group Relative Policy Optimization (GRPO) has become the dominant method for reinforcement learning with verifiable rewards in large language models, but it suffers from two critical limitations: gradient vanishing and diversity collapse. When training questions are too easy or too hard, all sampled responses receive identical rewards, yielding zero gradients. Meanwhile, the model tends to collapse its responses toward a single reasoning pattern rather than exploring diverse strategies. We propose Transformation-Augmented GRPO (TA-GRPO), a simple but effective method that addresses both issues via question rephrasing. For each training question, we automatically generate multiple problem-equivalent rephrasings that alter wording, format, and information order while preserving the underlying meaning. Because these rephrasings shift the model's perceived difficulty, pooling responses across the original and its rephrasings yields mixed rewards and more diverse reasoning paths. TA-GRPO jointly computes advantages over this expanded response set and aligns all importance ratios to the original question, enabling the model to learn from a richer set of solution attempts. Experiments on four LLMs (Qwen3-1.7B, Qwen3-4B, Llama-3.2-1B, Llama-3.2-3B) show that TA-GRPO consistently improves pass@$k$ on competition-level benchmarks (AMC, OlympiadBench, AIME24, AIME25) and out-of-distribution benchmarks (Minerva, GPQA-Diamond). Notably, it improves the average pass@32 of Qwen3-1.7B and Qwen3-4B by \textbf{4.97} and \textbf{4.34} points, respectively, and matches the exploration quality of baselines trained on up to 2.5$\times$ more data.