name: flash-attn-4
version: 4.0.0b30
summary: Flash Attention CUTE (CUDA Template Engine) implementation
license: BSD License
author: Tri Dao
requires_python: >=3.10
python_min: 3.10
install_size: 396.7 KB
module_format: py3-none-any
repo_url: https://github.com/Dao-AILab/flash-attention
last_released_at: 2026-09-09T08:20:46Z
releases_last_12mo: 28
avg_days_between_releases: 7.8
platforms: any

## dependencies
nvidia-cutlass-dsl >=4.6.2
torch
einops
typing_extensions
apache-tvm-ffi <0.2,>=0.1.12
torch-c-dlpack-ext
quack-kernels >=0.5.3

## extras
[cu13]
  nvidia-cutlass-dsl[cu13] >=4.6.2
[dev]
  pytest
  pytest-xdist
  ruff

## maintainers
Tri Dao