Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/rishit-dagli/compositional-attention
An implementation of Compositional Attention: Disentangling Search and Retrieval by MILA
https://github.com/rishit-dagli/compositional-attention
artificial-intelligence attention attention-mechanism machine-learning
Last synced: 21 days ago
JSON representation
An implementation of Compositional Attention: Disentangling Search and Retrieval by MILA
- Host: GitHub
- URL: https://github.com/rishit-dagli/compositional-attention
- Owner: Rishit-dagli
- License: apache-2.0
- Created: 2022-05-31T04:19:17.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2022-06-01T03:01:46.000Z (over 2 years ago)
- Last Synced: 2024-10-04T15:45:28.955Z (about 1 month ago)
- Topics: artificial-intelligence, attention, attention-mechanism, machine-learning
- Language: Python
- Homepage:
- Size: 162 KB
- Stars: 14
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Codeowners: .github/CODEOWNERS
Awesome Lists containing this project
README
# Compositional Attention
[](https://github.com/Rishit-dagli/compositional-attention/actions/workflows/python-publish.yml)
[](https://github.com/psf/black)
[](https://github.com/Rishit-dagli/compositional-attention/stargazers)
[](https://github.com/Rishit-dagli)
[](https://twitter.com/intent/follow?screen_name=rishit_dagli)
This repository is an implementation of [Compositional Attention: Disentangling Search and Retrieval](https://arxiv.org/abs/2110.09419) by MILA. Revisiting standard Multi-head attention through the lens of multiple parallel and independent search and retrieval mechanisms, this leads to static pairings between searches and retrievals, often leading to redundancy of parameters. They reframe the "heads" of multi-head attention as "searches", and once the multi-headed/searched values are aggregated, there is an extra retrieval step (using attention) off the searched results. The experiments establish this as an easy drop-in replacement for Multi-head attention.
## Installation
Run the following to install:
```sh
pip install compositional-attention
```
## Developing `compositional-attention`
To install `compositional-attention`, along with tools you need to develop and test, run the following in your virtualenv:
```sh
git clone https://github.com/Rishit-dagli/compositional-attention.git
# or clone your own fork
cd compositional-attention
pip install -e .[dev]
```
## Usage
```py
import tensorflow as tf
from compositional_attention import CompositionalAttention
attn = CompositionalAttention(
dim = 1024, # input dimension
dim_head = 64, # dimension per attention 'head' - head is now either search or retrieval
num_searches = 8, # number of searches
num_retrievals = 2, # number of retrievals
dropout = 0.1, # dropout of attention of search and retrieval
)
tokens = tf.random.uniform([1, 512, 1024]) # tokens
mask = tf.ones([1, 512], dtype=tf.dtypes.bool) # mask
out = attn(tokens, mask = mask) # (1, 512, 1024)
```
## Want to Contribute 🙋♂️?
Awesome! If you want to contribute to this project, you're always welcome! See [Contributing Guidelines](CONTRIBUTING.md). You can also take a look at [open issues](https://github.com/Rishit-dagli/Compositional-Attention/issues) for getting more information about current or upcoming tasks.
## Want to discuss? 💬
Have any questions, doubts or want to present your opinions, views? You're always welcome. You can [start discussions](https://github.com/Rishit-dagli/Compositional-Attention/discussions).
## Citations
```bibtex
@misc{https://doi.org/10.48550/arxiv.2110.09419,
doi = {10.48550/ARXIV.2110.09419},
url = {https://arxiv.org/abs/2110.09419},
author = {Mittal, Sarthak and Raparthy, Sharath Chandra and Rish, Irina and Bengio, Yoshua and Lajoie, Guillaume},
keywords = {Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {Compositional Attention: Disentangling Search and Retrieval},
publisher = {arXiv},
year = {2021},
copyright = {arXiv.org perpetual, non-exclusive license}
}
```
Official [PyTorch implmentation](https://github.com/sarthmit/compositional-attention) and Phil Wang's [PyTorch implmenetation](https://github.com/lucidrains/compositional-attention-pytorch).