Mentor-KD: Making Small Language Models Better Multi-step Reasoners

TL;DR

Mentor-KD enhances small language models' multi-step reasoning by using a mentor model to improve data quality and soft label provision during knowledge distillation, leading to better reasoning performance.

Contribution

The paper introduces Mentor-KD, a novel approach that employs an intermediate mentor model to improve reasoning distillation for smaller language models.

Findings

Mentor-KD outperforms baseline methods on complex reasoning tasks.

Using a mentor model improves data quality and soft label accuracy.

Mentor-KD achieves significant performance gains across various models.

Abstract

Large Language Models (LLMs) have displayed remarkable performances across various complex tasks by leveraging Chain-of-Thought (CoT) prompting. Recently, studies have proposed a Knowledge Distillation (KD) approach, reasoning distillation, which transfers such reasoning ability of LLMs through fine-tuning language models of multi-step rationales generated by LLM teachers. However, they have inadequately considered two challenges regarding insufficient distillation sets from the LLM teacher model, in terms of 1) data quality and 2) soft label provision. In this paper, we propose Mentor-KD, which effectively distills the multi-step reasoning capability of LLMs to smaller LMs while addressing the aforementioned challenges. Specifically, we exploit a mentor, intermediate-sized task-specific fine-tuned model, to augment additional CoT annotations and provide soft labels for the student model during reasoning distillation. We conduct extensive experiments and confirm Mentor-KD's effectiveness across various models and complex reasoning tasks.