This repository hosts prebuilt Python wheels for flash-attn v3, compiled for NVIDIA H100 GPUs.
- flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl
- Python 3.9+ (abi3)
- Linux x86_64
- Built with CUDA support for H100 (sm_90)
Download the wheel from the Releases page, then install with:
pip install flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl