Mac GPU Accelerated Sparse Attention
6.5
A cross-platform library, optimized for Mac GPUs using PyTorch 2.13 MPS FlexAttention, providing a significant speed boost for AI inference tasks involving sparse attention mechanisms. Leverages the PyTorch 2.13 MPS speedup to drastically reduce latency, particularly in scenarios with high token counts and limited visibility windows.
80h
mvp estimate
6.5
viability grade
6
views
technology stack
Python
Easy
inspired by
Mac GPU speeds up sparse attention by 7.83x.