A Closer Look at the Limitations of Instruction Tuning

TL;DR

This paper critically examines instruction tuning (IT) for large language models, revealing its limitations in knowledge enhancement, response quality, and hallucination, and showing that simpler fine-tuning methods often outperform IT.

Contribution

The paper provides a comprehensive analysis of IT's shortcomings, demonstrating that current methods do not effectively improve LLM knowledge or response quality, and highlighting the superiority of pre-trained knowledge responses.

Findings

IT does not improve LLM knowledge or skills.

Full-parameter fine-tuning can degrade knowledge.

IT responses often decline in quality and increase hallucinations.

Abstract

Instruction Tuning (IT), the process of training large language models (LLMs) using instruction-response pairs, has emerged as the predominant method for transforming base pre-trained LLMs into open-domain conversational agents. While IT has achieved notable success and widespread adoption, its limitations and shortcomings remain underexplored. In this paper, through rigorous experiments and an in-depth analysis of the changes LLMs undergo through IT, we reveal various limitations of IT. In particular, we show that (1) IT fails to enhance knowledge or skills in LLMs. LoRA fine-tuning is limited to learning response initiation and style tokens, and full-parameter fine-tuning leads to knowledge degradation. (2) Copying response patterns from IT datasets derived from knowledgeable sources leads to a decline in response quality. (3) Full-parameter fine-tuning increases hallucination by inaccurately borrowing tokens from conceptually similar instances in the IT dataset for generating responses. (4) Popular methods to improve IT do not lead to performance improvements over a simple LoRA fine-tuned model. Our findings reveal that responses generated solely from pre-trained knowledge consistently outperform responses by models that learn any form of new knowledge from IT on open-source datasets. We hope the insights and challenges revealed in this paper inspire future work in related directions.