FlashAttention-3 (Hopper) Wheel Builder

This Space clones flash-attention, builds FA-3 from hopper/ via python setup.py bdist_wheel, and gives you a downloadable wheel.

Heads-up: FA-3 targets Hopper GPUs (H100/H800) and CUDA ≥ 12.3 (12.8 recommended). If this Space doesn’t have a CUDA toolchain, the build will likely fail.

Commit URL