#flash-attention

llama.cpp Vulkan backend adds Intel Xe flash attention kernels

New kernels for Xe-LPG Plus, Xe2, and Xe3 target faster long-context token generation on recent Intel GPUs.