Curriculum Learning for Efficient Chain-of-Thought Distillation via Structure-Aware Masking and GRPO

TL;DR

This paper introduces a three-stage curriculum learning framework for distilling large language models' reasoning into smaller models, improving accuracy and reducing output length through structure-aware masking and GRPO optimization.

Contribution

It proposes a novel curriculum learning approach with structure-aware masking and GRPO to enhance chain-of-thought distillation for smaller models.

Findings

Qwen2.5-3B-Base achieves 11.29% accuracy improvement on GSM8K.

Output length is reduced by 27.4%.

Outperforms prior distillation methods and instruction-tuned variants.

Abstract

Distilling Chain-of-Thought (CoT) reasoning from large language models into compact student models presents a fundamental challenge: teacher rationales are often too verbose for smaller models to faithfully reproduce. Existing approaches either compress reasoning into single-step, losing the interpretability that makes CoT valuable. We present a three-stage curriculum learning framework that addresses this capacity mismatch through progressive skill acquisition. First, we establish structural understanding via masked shuffled reconstruction. Second, we apply Group Relative Policy Optimization (GRPO) on masked completion tasks, enabling the model to discover its own balance between accuracy and brevity. Third, we identify persistent failure cases and guide the student to internalize teacher knowledge through targeted rewriting, again optimized with GRPO. Experiments on GSM8K demonstrate that our approach enables Qwen2.5-3B-Base to achieve an 11.29 percent accuracy improvement while reducing output length by 27.4 percent, surpassing both instruction-tuned variants and prior distillation methods.