Kernel Library for Large Headdim Attention (64~1024, BF16/FP8/FP4), 1.5x~15x↑ vs PyTorch SDPA.
↑ +1 today
Discover the best Sage Attention open source repositories on GitHub. Explore trending tools, live star counts, source code, and developer projects updated daily.