CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-01 09:01:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: support mixed/deepseekR1 FA head sizes (#14509)

Browse Source 
* vulkan: better parameterize FA by head sizes

* vulkan: support mixed/deepseekR1 FA head sizes
This commit is contained in:
Jeff Bolz
2025-07-03 13:21:14 -05:00
committed by GitHub
parent c8c4495b8d
commit 2b72bedec1
5 changed files with 208 additions and 183 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp
View File

@@ -4,10 +4,10 @@ layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
layout (constant_id = 0) const uint32_t WorkGroupSize = 128;
layout (constant_id = 1) const uint32_t Br = 1;
layout (constant_id = 2) const uint32_t Bc = 32;
layout (constant_id = 3) const uint32_t D = 32;
layout (constant_id = 4) const uint32_t Clamp = 0;
layout (constant_id = 5) const uint32_t D_split = 16;
layout (constant_id = 3) const uint32_t HSK = 32;
layout (constant_id = 4) const uint32_t HSV = 32;
layout (constant_id = 5) const uint32_t Clamp = 0;
layout (constant_id = 6) const uint32_t D_split = 16;
layout (push_constant) uniform parameter {
uint32_t N;