Poly Attention
Implementation of Poly-Attention, a general scheme for higher-order self-attention
Install
$ pip install poly-attention
Usage
import torch
from poly_attention import PolyAttention
attn = PolyAttention(
dim = 512,
heads = 8,
dim_head = 64,
causal = False
)
tokens = torch.randn(1, 1024, 512)
out = attn(tokens) # (1, 1024, 512)
A Vision Transformer based on Poly-Attention
import torch
from poly_attention import PolyViT
vit = PolyViT(
image_size = 256,
patch_size = 32,
num_classes = 1000,
dim = 1024,
depth = 6,
heads = 16,
mlp_dim = 2048,
order = 2 # standard poly attention order 2
)
images = torch.randn(1, 3, 256, 256)
preds = vit(images) # (1, 1000)
Quick test
python train_function_composition.py --poly_layers=1 --base_layers=2
Appreciation
-
@dillfrescott for submitting a stability fix
-
@pranoyr for more efficient caching of the GQA heads!
Citations
@inproceedings{chakrabarti2026poly,
title = {Poly-attention: a general scheme for higher-order self-attention},
author = {Chakrabarti, Sayak and Pitassi, Toniann and Alman, Josh},
booktitle = {International Conference on Learning Representations (ICLR)},
year = {2026}
}
@misc{kayyam2026transformersneedprojectionssystematic,
title = {Do Transformers Need Three Projections? Systematic Study of QKV Variants},
author = {Ali Kayyam and Anusha Madan Gopal and M Anthony Lewis},
year = {2026},
eprint = {2606.04032},
archivePrefix = {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2606.04032},
}
1 maintainer
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
poly_attention-0.3.0.tar.gz
(11.8 kB
view details)
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file poly_attention-0.3.0.tar.gz.
File metadata
- Download URL: poly_attention-0.3.0.tar.gz
- Upload date:
- Size: 11.8 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via:
uv/0.8.17
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
5622f7a7d781e1d00944ab9fde50892cde2bedc15c6717fd9e662583cd1d91e1
|
|
| MD5 |
4ffc7c5d61a8880fd079498c0c1491c7
|
|
| BLAKE2b-256 |
bc29f7fb6b8c366d8ec083fc31883780acd07954c518441100e33b03b50cd093
|
File details
Details for the file poly_attention-0.3.0-py3-none-any.whl.
File metadata
- Download URL: poly_attention-0.3.0-py3-none-any.whl
- Upload date:
- Size: 13.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via:
uv/0.8.17
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
0d1dcd396945d919aa5747dc0247866c46c13b9793649cebd798fce510cd4553
|
|
| MD5 |
d077ab99fcf99ca4a7fd4ccd4b83eddb
|
|
| BLAKE2b-256 |
73124d66d5fb041daecad6345ad83f2236d02860af5fc62982c37f8d227ec1c0
|
