Chat Vector: A Simple Approach to Equip LLMs with Instruction Following and Model Alignment in New Languages

TL;DR

This paper introduces the chat vector, a simple arithmetic method to enable open-source LLMs to follow instructions and align with human values in new languages without additional training.

Contribution

The authors propose the chat vector technique, allowing models to acquire chat capabilities through weight subtraction and addition, simplifying multilingual instruction following.

Findings

Chat vector improves instruction following across languages.

It reduces toxicity and enhances multi-turn dialogue performance.

The method is effective across various models and languages.

Abstract

Recently, the development of open-source large language models (LLMs) has advanced rapidly. Nevertheless, due to data constraints, the capabilities of most open-source LLMs are primarily focused on English. To address this issue, we introduce the concept of $\textit{chat vector}$ to equip pre-trained language models with instruction following and human value alignment via simple model arithmetic. The chat vector is derived by subtracting the weights of a pre-trained base model (e.g. LLaMA2) from those of its corresponding chat model (e.g. LLaMA2-chat). By simply adding the chat vector to a continual pre-trained model's weights, we can endow the model with chat capabilities in new languages without the need for further training. Our empirical studies demonstrate the superior efficacy of the chat vector from three different aspects: instruction following, toxicity mitigation, and multi-turn dialogue. Moreover, to showcase the adaptability of our approach, we extend our experiments to encompass various languages, base models, and chat vectors. The results underscore the chat vector's simplicity, effectiveness, and wide applicability, making it a compelling solution for efficiently enabling conversational capabilities in pre-trained language models. Our code is available at https://github.com/aqweteddy/ChatVector.