#FP8

Zyphra ZYFA: 1.76x FP8 FlashAttention Speedup on AMD MI355X

Zyphra open-sources ZYFA, providing FP8 and BF16 prefill attention kernels for AMD Instinct MI355X that are 1.16-1.76x faster than AMD AITER, with 62%-78% lower error at head dimension 128.