flare-fast-low-rank-attention

Implement low-rank attention routing using encode-decode factorization to achieve linear-time complexity on long sequences while maintaining compatibility with optimized attention kernels.

Independent GitHub snapshot · Automated classification · Original documentation governs