LLaMA-Excitor: General Instruction Tuning via Indirect Feature Interaction

TL;DR

LLaMA-Excitor is a lightweight fine-tuning method that enhances instruction-following in LLMs and multi-modal models by selectively adjusting attention without altering hidden states, preserving core abilities while improving performance.

Contribution

The paper introduces LLaMA-Excitor, a novel attention bypass module that improves instruction-following and multi-modal tuning without compromising pre-trained knowledge.

Findings

Achieves +6% on MMLU benchmark.

Sets new state-of-the-art in image captioning with 157.5 CIDEr on MSCOCO.

Maintains core capabilities while enhancing task performance.

Abstract

Existing methods to fine-tune LLMs, like Adapter, Prefix-tuning, and LoRA, which introduce extra modules or additional input sequences to inject new skills or knowledge, may compromise the innate abilities of LLMs. In this paper, we propose LLaMA-Excitor, a lightweight method that stimulates the LLMs' potential to better follow instructions by gradually paying more attention to worthwhile information. Specifically, the LLaMA-Excitor does not directly change the intermediate hidden state during the self-attention calculation of the transformer structure. We designed the Excitor block as a bypass module for the similarity score computation in LLMs' self-attention to reconstruct keys and change the importance of values by learnable prompts. LLaMA-Excitor ensures a self-adaptive allocation of additional attention to input instructions, thus effectively preserving LLMs' pre-trained knowledge when fine-tuning LLMs on low-quality instruction-following datasets. Furthermore, we unify the modeling of multi-modal tuning and language-only tuning, extending LLaMA-Excitor to a powerful visual instruction follower without the need for complex multi-modal alignment. Our proposed approach is evaluated in language-only and multi-modal tuning experimental scenarios. Notably, LLaMA-Excitor is the only method that maintains basic capabilities while achieving a significant improvement (+6%) on the MMLU benchmark. In the visual instruction tuning, we achieve a new state-of-the-art image captioning performance of 157.5 CIDEr on MSCOCO, and a comparable performance (88.39%) on ScienceQA to cutting-edge models with more parameters and extensive vision-language pertaining.