vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-09 11:00:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

i hate doing this

Browse source
This commit is contained in:
Concedo 2024-06-26 13:38:59 +08:00
parent c66371fbb0
commit a7a4e238fb
2 changed files with 4 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

7
CMakeLists.txt
View file

@ -126,14 +126,15 @@ if (LLAMA_CUBLAS)
# 60 == f16 CUDA intrinsics
# 61 == integer CUDA intrinsics
# 70 == (assumed) compute capability at which unrolling a loop in mul_mat_q kernels is faster
# 75 == int8 tensor cores
if (LLAMA_CUDA_F16 OR LLAMA_CUDA_DMMV_F16)
set(CMAKE_CUDA_ARCHITECTURES "60;61;70") # needed for f16 CUDA intrinsics
set(CMAKE_CUDA_ARCHITECTURES "60;61;70;75") # needed for f16 CUDA intrinsics
else()
message("CUDA Toolkit Version: ${CUDAToolkit_VERSION}")
if(CUDAToolkit_VERSION VERSION_GREATER 12)
set(CMAKE_CUDA_ARCHITECTURES "52;61;70") # lowest CUDA 12 standard + lowest for integer intrinsics
set(CMAKE_CUDA_ARCHITECTURES "52;61;70;75") # lowest CUDA 12 standard + lowest for integer intrinsics
else()
set(CMAKE_CUDA_ARCHITECTURES "37;52;61;70") # lowest CUDA 12 standard + lowest for integer intrinsics
set(CMAKE_CUDA_ARCHITECTURES "37;52;61;70;75") # lowest CUDA 12 standard + lowest for integer intrinsics
endif()
endif()
endif()

5
ggml-cuda/mmq.cu
View file

@ -108,11 +108,6 @@ bool ggml_cuda_should_use_mmq(enum ggml_type type, int cc, int64_t ne11) {
return false;
}
if(g_mul_mat_q)
{
return true;
}
if (cc < CC_OFFSET_AMD) {
return cc < CC_VOLTA || ne11 < MMQ_DP4A_MAX_BATCH_SIZE;
}