← Back to events
ActiveAIRelease

trunk/29c701469ca2df4605194613ca70f3bba41e4abe: Enable FA build for SDPA on Windows+CUDA (#186343)

Photo: PyTorch Releases

What happened

Summary Description Enable building the FlashAttention backend for scaled_dot_product_attention on Windows + CUDA (MSVC). Today, the flag is hard-disabled on MSVC, so every Windows CUDA build ships without the flash SDPA backend and silently falls back to slower, memory-heavy backends. With this PR, the FlashAttention kernels now build cleanly under modern MSVC, which restores the fast path. Why this matters On Wind…

Summary assembled by rule from the sources below

Why it's spreading

Sources