Command-V: Pasting LLM Behaviors via Activation Profiles

TL;DR

Command-V is a novel, resource-efficient method for transferring behaviors between large language models by copying activation profiles, avoiding costly retraining and data access, and achieving comparable or better performance.

Contribution

It introduces Command-V, a backpropagation-free technique for behavior transfer in LLMs that requires minimal compute and no training data, outperforming traditional finetuning in key tasks.

Findings

Matches or exceeds finetuning performance in safety and reasoning tasks.

Requires significantly less compute and no training data.

Applicable across different LLM architectures.

Abstract

Retrofitting large language models (LLMs) with new behaviors typically requires full finetuning or distillation-costly steps that must be repeated for every architecture. In this work, we introduce Command-V, a backpropagation-free behavior transfer method that copies an existing residual activation adapter from a donor model and pastes its effect into a recipient model. Command-V profiles layer activations on a small prompt set, derives linear converters between corresponding layers, and applies the donor intervention in the recipient's activation space. This process does not require access to the original training data and needs minimal compute. In three case studies-safety-refusal enhancement, jailbreak facilitation, and automatic chain-of-thought reasoning--Command-V matches or exceeds the performance of direct finetuning while using orders of magnitude less compute. Our code and data are accessible at https://github.com/GithuBarry/Command-V/.