There is now a CPython + LibTorch ABI stable FlashMLA for if you want to pip install 1 wheel that works across Python 3.10+ and torch 2.10+ at runtime: https://t.co/DrKKG7JOXQ :D
In collaboration with Tri Dao and xformers, PyTorch has uploaded official Flash Attention 3 wheels to https://t.co/KfhpnHoTvl. These wheels support various CUDA versions (12.6+, 13), CPUs (x86, ARM) and OS (Linux, Windows), and are compatible w/ Python 3.10+, torch 2.9+.