Summary of Llama-excitor: General Instruction Tuning Via Indirect Feature Interaction, by Bo Zou et al.
LLaMA-Excitor: General Instruction Tuning via Indirect Feature Interaction
by Bo Zou, Chao Yang, Yu Qiao, Chengbin Quan, Youjian Zhao
First submitted to arxiv on: 1 Apr 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed LLaMA-Excitor method is a lightweight approach to fine-tune large language models (LLMs) without compromising their innate abilities. The Excitor block acts as a bypass module for the self-attention calculation in transformer structures, allowing for a self-adaptive allocation of attention to input instructions. This approach preserves pre-trained knowledge when fine-tuning LLMs on low-quality instruction-following datasets. The method is evaluated in both language-only and multi-modal tuning scenarios, achieving state-of-the-art performance in image captioning and comparable results in ScienceQA. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper proposes a new way to make large language models better at following instructions without changing their core abilities. It introduces a simple module that helps the model focus on important information, allowing it to learn from low-quality training data. The method is tested on two different types of tasks and achieves impressive results, outperforming other approaches. |
Keywords
» Artificial intelligence » Attention » Fine tuning » Image captioning » Llama » Multi modal » Self attention » Transformer
