mjun0812/flash-attention-prebuild-wheels resource

Health v2 · maintenance only

Adoption not part of the score

1706 stars · 78 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A repository that distributes pre-built flash-attention 2 and 3 wheel packages for Linux and Windows, built via GitHub Actions, AWS CodeBuild, and self-hosted runners. It saves users from the very long and resource-intensive process of compiling flash-attention themselves by providing wheels across many Python, CUDA, and PyTorch version combinations.

Use cases

• install flash-attention without compiling it
• find a flash_attn wheel for my CUDA and PyTorch versions
• avoid long build times for flash-attention on Windows
• get prebuilt flash-attention wheels for Linux ARM64
• pip install flash_attn for Python 3.11 with CUDA 12.4
• set up flash-attention for transformer model training quickly

When to choose

• you need flash-attention installed quickly without a lengthy local build
• your CUDA/PyTorch/Python combination is not officially distributed by flash-attention
• you are on Linux x86_64, Linux ARM64, or Windows with an NVIDIA GPU

When to avoid

• you need flash-attention on macOS or CPU-only environments
• you require a custom flash-attention build with non-standard flags or patches
• you need a CUDA/PyTorch combination not covered by the published wheels

Facets

dataset · maturity active

machine-learning gpu-computing developer-tools deep-learning gpu-computing developer-tools windows python flash-attention prebuilt-wheels cuda pytorch wheel-distribution transformers linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="mjun0812/flash-attention-prebuild-wheels")