koboldcpp/ggml
Nathan Wilson c7bda030e7
vulkan: fix FA dequant path engagement (#28190)
Skip the nb[3] check when ne[3] == 1, the shader never reads it for a
single stream. Cache views carry the full-buffer stride there, so the old
check reduced to n_kv == kv_size and the path only engaged with the
cache full.
2026-09-03 10:40:34 +02:00
..
cmake kleidiai: Rework KleidiAI Build System/Integration (#26077) 2026-08-25 14:07:29 -07:00
include CUDA + ggml: add sparse-fa for DSV4/GLM (#27970) 2026-09-02 17:27:37 +03:00
src vulkan: fix FA dequant path engagement (#28190) 2026-09-03 10:40:34 +02:00
.gitignore
CMakeLists.txt metal : add metallib build support for xcframework (#28163) 2026-09-02 07:45:56 +08:00