Zyphra ZYFA: FP8 FlashAttention for AMD MI355X, Up to 1.76x Faster
Zyphra has introduced ZYFA, an open-source implementation of FP8 and BF16 prefill attention kernels designed for the AMD Instinct MI355X. Testing showed these kernels achieve speedups of 1.16 to 1.76 times compared to AMD's AITER, while maintaining low accuracy loss.
Key signals
- 01ZYFA supports FP8 and BF16 attention kernels.
- 02Speed improvements range from 1.16x to 1.76x on MI355X.
- 03Accuracy degradation is minimal, between 62-78% lower error.
Zyphra ZYFA: FP8 FlashAttention for AMD MI355X, Up to 1.76x Faster, published by MarkTechPost.
Zyphra released ZYFA, open-source FP8 and BF16 prefill attention kernels for AMD Instinct MI355X. On real model activations, its FP8 kernels run 1.16-1.76x faster than AMD’s AITER, with 62-78% lower error at head dimension 128. The post Zyphra ZYFA: FP8 FlashAttention for AMD…
MarkTechPost
Read the full story at the source: marktechpost.com.
Curated automatically from MarkTechPost. Infrenic News links to the original publisher and does not republish its content.
Daily briefing
Get the next briefing before everyone else.
Five AI stories, five minutes, every morning.
