Fine-Tuning or Retrieval? Comparing Knowledge Injection in LLMs

TL;DR

This paper compares fine-tuning and retrieval-augmented generation for knowledge injection in large language models, finding RAG generally outperforms fine-tuning, especially for new information, and highlighting challenges in learning new facts through fine-tuning.

Contribution

It provides a systematic comparison between fine-tuning and retrieval-based methods for updating LLMs with new knowledge, revealing the superiority of RAG in various scenarios.

Findings

RAG outperforms fine-tuning on knowledge-intensive tasks.

LLMs struggle to learn new facts via fine-tuning.

Multiple variations of facts during training can help LLMs learn new information.

Abstract

Large language models (LLMs) encapsulate a vast amount of factual information within their pre-trained weights, as evidenced by their ability to answer diverse questions across different domains. However, this knowledge is inherently limited, relying heavily on the characteristics of the training data. Consequently, using external datasets to incorporate new information or refine the capabilities of LLMs on previously seen information poses a significant challenge. In this study, we compare two common approaches: unsupervised fine-tuning and retrieval-augmented generation (RAG). We evaluate both approaches on a variety of knowledge-intensive tasks across different topics. Our findings reveal that while unsupervised fine-tuning offers some improvement, RAG consistently outperforms it, both for existing knowledge encountered during training and entirely new knowledge. Moreover, we find that LLMs struggle to learn new factual information through unsupervised fine-tuning, and that exposing them to numerous variations of the same fact during training could alleviate this problem.