LLaMA-Excitor: General Instruction Tuning via Indirect Feature Interaction

arXiv:2404.00913v1 Announce Type: new
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 …

abstract adapter arxiv cs.ai cs.cl cs.cv extra feature general knowledge llama llms lora modules paper skills type via