Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
An awesome repository & A comprehensive survey on interpretability of LLM attention heads.
| Date | Stars |
|---|---|
| 2026-07-24 | 412 |
| 2026-07-25 | 412 |
| 2026-07-28 | 412 |
| 2026-07-30 | 412 |
| 2026-07-31 | 412 |
| 2026-08-06 | 412 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<h2 align='center'> Attention Heads of Large Language Models: A Survey<br>(Awesome Attention Heads) </h2>
<div align='center'>
<p align="center">
<!-- Awesome badges-->
<a href="https://github.com/IAAR-Shanghai/Awesome-Attention-Heads">
<img src="https://img.shields.io/static/v1?label=&message=Awesome+Attention+Heads&color=black&logo=awesomelists">
</a>
<!-- arxiv badges -->
<a href="https://arxiv.org/abs/2409.03752">
<img src="https://img.shields.io/badge/Paper-red?style=flat&logo=arxiv">
</a>
<!-- hf -->
<a href="https://huggingface.co/papers/2409.03752">
<img src="https://img.shields.io/badge/-%F0%9F%A4%97%20Hugging_Face-orange?style=flat"/>
</a>
<!-- Last commit -->
<img src="https://img.shields.io/github/last-commit/IAAR-Shanghai/Awesome-Attention-Heads?color=green">
</p>
</div>
<!-- <div align="center">
<p>
<a href="https://github.com/fan2goa1">Zifan Zheng</a><sup>1*</sup>,
<a href="https://github.com/wyzh0912">Yezhaohui Wang</a><sup>1*</sup>,
<a href="https://github.com/saythe17">Yuxin Huang</a><sup>2*</sup>,
<a href="https://github.com/Ki-Seki">Shichao Song</a><sup>1</sup>,
Bo Tang<sup>1</sup>,
Feiyu Xiong<sup>1</sup>,
Zhiyu Li<sup>1†</sup>
</p>
<p>
<sup>1</sup><a href="https://www.iaar.ac.cn/">Institute for Advanced Algorithms Research (IAAR), Shanghai</a>, <br>
<sup>2</sup><a href="https://air.tsinghua.edu.cn">Institute for AI Industry Research (AIR), Tsinghua University</a>
</p>
</div> -->
<!-- <div align="center">
<p>
<sup>*</sup>Equal contribution.
<br>
<sup>†</sup>Corresponding author: Zhiyu Li (<a href="mailto:[email protected]">[email protected]</a>).
</p>
</div> -->
> \[!IMPORTANT\]
>
> - About this repo. This is a platform to get the **latest research** on different kinds of LLM's Attention Heads. Also, we released a **survey** based on these fantastic works.
>
> - If you want to **cite our work**, here is our bibtex entry: [CITATION.bib](./CITATION.bib).
>
> - If you only want to see the related **paper list**, please jump directly to [here](#-paper-list).
>
> - If you want to contribute to this repo, refer to [here](#hand-make-a-contribution).
## 📢 News
- **[2025/01/13]** Our paper **was accepted by [Patterns](https://www.cell.com/patterns/fulltext/S2666-3899(25)00024-8) (Cell Press)**.
- **[2024/09/07]** Our paper secured the 2nd place on [Hugging Face's Daily Paper List](https://huggingface.co/papers?date=2024-09-06).
- **[2024/09/06]** Our survey paper is available on the arXiv platform: https://arxiv.org/abs/2409.03752.
## 📰 Table of Contents
- [Background](#-background)
- [About Our Survey](#-about-our-survey)
- [Paper List](#-paper-list)
- [Star Trends](#star-star-trends)
## 🎉 Background
With the development of Large Language Model (LLMs), their underlying network structure, the Transformer, is being extensively studied. Researching the Transformer structure helps us enhance our understanding of this "black box" and improve model interpretability. Recently, there has been an increasing body of work suggesting that the model contains two distinct partitions: attention mechanisms used for behavior, inference, and analysis, and Feed-Forward Networks (FFN) for knowledge storage. The former is crucial for revealing the functional capabilities of the model, leading to a series of studies exploring various functions within attention mechanisms, which we have termed **Attention Head Mining**.
## 🔍 About Our Survey
In this survey, we delve into the potential mechanisms of how attention heads in LLMs contribute to the reasoning process.
**Highlights:**
- We propose an innovative **four-stage framework**, inspired by human cognitive neuroscience, to analyze the reasoning process of LLMs (Knowledge Recalling, In-Context Identification, Latent Reasoning, Expression Preparation).
<div align="center">
<img src="assets/four_steps.png" alt="Survey FramewoExcerpt of 97,191 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:15e8a8f4ee4d2ebc, topic:large-language-models, topic:llm, topic:transformer