Unveiling the Key Factors for Distilling Chain-of-Thought Reasoning

TL;DR

This paper investigates how different factors like granularity, format, and teacher model choice affect the effectiveness of distilling Chain-of-Thought reasoning from large to small language models, providing insights for optimizing this process.

Contribution

It systematically analyzes key factors influencing CoT distillation, revealing how model strength and supervision strategies impact performance in small language models.

Findings

SLMs show non-monotonic performance with granularity, benefiting from simpler CoT supervision.

CoT format impacts LLMs significantly but has minimal effect on SLMs.

Stronger teacher models do not always produce better student models due to diversity and complexity in CoT supervision.

Abstract

Large Language Models (LLMs) excel in reasoning tasks through Chain-of-Thought (CoT) prompting. However, CoT prompting greatly increases computational demands, which has prompted growing interest in distilling CoT capabilities into Small Language Models (SLMs). This study systematically examines the factors influencing CoT distillation, including the choice of granularity, format and teacher model. Through experiments involving four teacher models and seven student models across seven mathematical and commonsense reasoning datasets, we uncover three key findings: (1) Unlike LLMs, SLMs exhibit a non-monotonic relationship with granularity, with stronger models benefiting from finer-grained reasoning and weaker models performing better with simpler CoT supervision; (2) CoT format significantly impacts LLMs but has minimal effect on SLMs, likely due to their reliance on supervised fine-tuning rather than pretraining preferences; (3) Stronger teacher models do NOT always produce better student models, as diversity and complexity in CoT supervision can outweigh accuracy alone. These findings emphasize the need to tailor CoT strategies to specific student model, offering actionable insights for optimizing CoT distillation in SLMs. The code and datasets are available at https://github.com/EIT-NLP/Distilling-CoT-Reasoning.