method · first seen 1 Nov 2026
FlashAttention-style attention kernels on TPU
Optimized attention kernel implementations tailored to the matrix units and memory layout of Tensor Processing Units.
Mentions per week
PapersCodeDiscussion
Earliest sources
- FlyWheel: Attention Kernels Built for the TPUPapers · 1 Nov 2026 · Zishuo Bao, Zeshen Zhang, Linkun Chen, Yucheng Lu
Latest sources
- FlyWheel: Attention Kernels Built for the TPUPapers · 1 Nov 2026 · Zishuo Bao, Zeshen Zhang, Linkun Chen, Yucheng Lu