Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-18 16:55:14 +02:00
Code Issues Packages Projects Releases Wiki Activity

update for cu13 builds (no ci will be provided)

Browse Source
This commit is contained in:
Concedo
2025-09-26 16:01:43 +08:00
parent 8b5ebfcc8d
commit 1a4f54dd11
3 changed files with 13 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CMakeLists.txt
+4 -1
View File
@@ -130,7 +130,10 @@ if (LLAMA_CUBLAS)
# 75 == int8 tensor cores
# 80 == Ampere, asynchronous data loading, faster tensor core instructions
message("CUDA Toolkit Version: ${CUDAToolkit_VERSION}")
if(CUDAToolkit_VERSION VERSION_GREATER 12)
if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL 13)
add_compile_definitions(GGML_CUDA_USE_GRAPHS) #try enable cuda graphs on cu12 build
set(CMAKE_CUDA_ARCHITECTURES "75-virtual;80-virtual;86-virtual") # lowest CUDA 13 standard
elseif(CUDAToolkit_VERSION VERSION_GREATER 12)
add_compile_definitions(GGML_CUDA_USE_GRAPHS) #try enable cuda graphs on cu12 build
set(CMAKE_CUDA_ARCHITECTURES "50-virtual;61-virtual;70-virtual;75-virtual;80-virtual") # lowest CUDA 12 standard + lowest for integer intrinsics
else()
Makefile
+6
View File
@@ -231,6 +231,12 @@ NVCCFLAGS += -Wno-deprecated-gpu-targets \
-gencode arch=compute_75,code=compute_75 \
-gencode arch=compute_80,code=compute_80
else ifdef LLAMA_ARCHES_CU13
NVCCFLAGS += -Wno-deprecated-gpu-targets \
-gencode arch=compute_75,code=compute_75 \
-gencode arch=compute_80,code=compute_80 \
-gencode arch=compute_86,code=compute_86
else
NVCCFLAGS += -Wno-deprecated-gpu-targets -arch=all
endif
koboldcpp.sh
+3
View File
@@ -39,6 +39,9 @@ fi
if [ -n "$ARCHES_CU12" ]; then
ARCHES_FLAG="LLAMA_ARCHES_CU12=1"
fi
if [ -n "$ARCHES_CU13" ]; then
ARCHES_FLAG="LLAMA_ARCHES_CU13=1"
fi
if [ -n "$NO_WMMA" ]; then
NO_WMMA_FLAG="LLAMA_NO_WMMA=1"
fi