#attention kernels

Zyphra ZYFA: FP8 FlashAttention for AMD MI355X Delivers Up to 1NEW⭐8

Zyphra's open-source ZYFA delivers FP8 FlashAttention kernels for AMD MI355X, running up to 1.76x faster than AITER with 62–78% lower FP8 error at head dim 128.