Sci-CoT: Leveraging Large Language Models for Enhanced Knowledge Distillation in Small Models for Scientific QA

TL;DR

This paper introduces Sci-CoT, a two-stage knowledge distillation framework that transfers reasoning abilities from large language models to smaller models, significantly improving scientific question-answering performance.

Contribution

The paper presents Sci-CoT, a novel two-stage distillation method that enhances small models' reasoning skills by mimicking large models' chain-of-thought reasoning in scientific QA tasks.

Findings

Small model outperforms BLOOM-176B on ARC-Easy with few-shot learning

Sci-CoT effectively transfers reasoning capabilities from large to small models

80-million parameter model surpasses large models in specific scientific QA benchmarks

Abstract

Large Language Models (LLMs) have shown outstanding performance across wide range of downstream tasks. This competency is attributed to their substantial parameter size and pre-training on extensive corpus. Moreover, LLMs have exhibited enhanced reasoning capabilities in tackling complex reasoning tasks, owing to the utilization of a method named ``Chain-of-Thought (CoT) prompting''. This method is designed to generate intermediate reasoning steps that guide the inference of the final answer. However, it is essential to highlight that these advanced reasoning abilities appear to emerge in models with a minimum of 10 billion parameters, thereby limiting its efficacy in situations where computational resources are constrained. In this paper, we investigate the possibility of transferring the reasoning capabilities of LLMs to smaller models via knowledge distillation. Specifically, we propose Sci-CoT, a two-stage framework that separates the processes of generating rationales and inferring answers. This method enables a more efficient use of rationales during the answer inference stage, leading to improved performance on scientific question-answering tasks. Utilizing Sci-CoT, our 80-million parameter model is able to exceed the performance of BLOOM-176B in the ARC-Easy dataset under the few shot setting.