Scaf-GRPO: Scaffolded Group Relative Policy Optimization for Enhancing LLM Reasoning

TL;DR

Scaf-GRPO is a progressive training framework that enhances LLM reasoning by strategically providing minimal guidance when models plateau, significantly improving performance on challenging benchmarks.

Contribution

Introduces Scaf-GRPO, a novel scaffolded training method that diagnoses learning stagnation and injects tiered hints to improve LLM reasoning on difficult tasks.

Findings

Boosted Qwen2.5-Math-7B's pass@1 score on AIME24 by 44.3%

Effectively overcomes the 'learning cliff' in policy optimization

Enhances autonomous reasoning in large language models

Abstract

Reinforcement learning from verifiable rewards has emerged as a powerful technique for enhancing the complex reasoning abilities of Large Language Models (LLMs). However, these methods are fundamentally constrained by the ''learning cliff'' phenomenon: when faced with problems far beyond their current capabilities, models consistently fail, yielding a persistent zero-reward signal. In policy optimization algorithms like GRPO, this collapses the advantage calculation to zero, rendering these difficult problems invisible to the learning gradient and stalling progress. To overcome this, we introduce Scaf-GRPO (Scaffolded Group Relative Policy Optimization), a progressive training framework that strategically provides minimal guidance only when a model's independent learning has plateaued. The framework first diagnoses learning stagnation and then intervenes by injecting tiered in-prompt hints, ranging from abstract concepts to concrete steps, enabling the model to construct a valid solution by itself. Extensive experiments on challenging mathematics benchmarks demonstrate Scaf-GRPO's effectiveness, boosting the pass@1 score of the Qwen2.5-Math-7B model on the AIME24 benchmark by a relative 44.3% over a vanilla GRPO baseline. This result demonstrates our framework provides a robust and effective methodology for unlocking a model's ability to solve problems previously beyond its reach, a critical step towards extending the frontier of autonomous reasoning in LLM.