Instruction-tuned Language Models are Better Knowledge Learners
Authors:
- Zhengbao JiangZJZhengbao Jiang
- Pedro RodriguezPRPedro Rodriguez
- Chunting ZhouCZChunting Zhou
- Graham NeubigGNGraham Neubig
- Xi Victoria LinXLXi Victoria Lin
- Wen-Tau YihWYWen-Tau Yih
- Srinivasan IyerSISrinivasan Iyer
Preprint
Abstract
In order for large language model (LLM)-based assistants to effectively adapt to evolving information needs, it must be possible to update their factual knowledge through continued training on new data. The standard recipe for doing so involves continued pre-training on new documents followed by instruction-tuning on question-answer (QA) pairs. However, we find that LLMs trained with this recipe struggle to answer questions, even though the perplexity of documents is minimized. We found that QA pairs are generally straightforward, while documents are more complex, weaving many factual statements together in an intricate manner. Therefore, we hypothesize that it is beneficial to expose LLMs to QA pairs before continued pre-training on documents so that the process of encoding knowledge from complex documents takes into account how this knowledge is accessed through questions. Based on this, we propose pre-instruction-tuning (PIT), a method that instruction-tunes on questions prior to training on documents. This contrasts with standard instruction-tuning, which learns how to extract knowledge after training on documents. Extensive experiments and ablation studies demonstrate that PIT significantly enhances the ability of LLMs to absorb knowledge from new documents, outperforming standard instruction-tuning by 17.8%.Keywords
All Related Versions
This publication has 15 references indexed in Scilit:
- PMC-LLaMA: toward building open-source language models for medicineJournal of the American Medical Informatics Association, 2024
- WebCPM: Interactive Web Search for Chinese Long-form Question Answering2023
- Retrieval as Attention: End-to-end Learning of Retrieval and Reading within a Single Transformer2022
- You Only Need One Model for Open-domain Question Answering2022
- Cross-Task Generalization via Natural Language Crowdsourcing Instructions2022
- Can Generative Pre-trained Language Models Serve As Knowledge Bases for Closed-book QA?2021
- How Can We Know What Language Models Know?Transactions of the Association for Computational Linguistics, 2020
- How Much Knowledge Can You Pack Into the Parameters of a Language Model?2020
- BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension2020
- Natural Questions: A Benchmark for Question Answering ResearchTransactions of the Association for Computational Linguistics, 2019