Training Task Experts through Retrieval Based Distillation

TL;DR

ReBase is a retrieval-based distillation method that enhances domain-specific data quality and reasoning capabilities of models by leveraging online sources, leading to significant performance improvements on multiple benchmarks.

Contribution

The paper introduces ReBase, a novel retrieval-based distillation approach that improves data diversity and reasoning in models for specialized tasks.

Findings

Up to 7.8% performance improvement on SQuAD

Enhanced reasoning with Chain-of-Thought generation

Significant gains on multiple benchmarks

Abstract

One of the most reliable ways to create deployable models for specialized tasks is to obtain an adequate amount of high-quality task-specific data. However, for specialized tasks, often such datasets do not exist. Existing methods address this by creating such data from large language models (LLMs) and then distilling such knowledge into smaller models. However, these methods are limited by the quality of the LLMs output, and tend to generate repetitive or incorrect data. In this work, we present Retrieval Based Distillation (ReBase), a method that first retrieves data from rich online sources and then transforms them into domain-specific data. This method greatly enhances data diversity. Moreover, ReBase generates Chain-of-Thought reasoning and distills the reasoning capacity of LLMs. We test our method on 4 benchmarks and results show that our method significantly improves performance by up to 7.8% on SQuAD, 1.37% on MNLI, and 1.94% on BigBench-Hard.