hrx: the fused Q decode kernel compiles at every head size, not only a power of two (engine #1926) - #282

hrx: the fused Q decode kernel compiles at every head size, not only a power of two (engine #1926)#282

bong-water-water-bong merged 1 commit into

1bit/hrx-vulkan-patched1bit-MONSTER/llama.cpp:1bit/hrx-vulkan-patchedfrom

1bit/1926-q-decode-head-size1bit-MONSTER/llama.cpp:1bit/1926-q-decode-head-sizeCopy head branch name to clipboard