baalajimaestro/ollama
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

add support for custom gpu build flags for llama.cpp

Browse source
This commit is contained in:
Jeremy 2024-04-17 16:00:48 -04:00
parent 7c000ec3ed
commit 52f5370c48
1 changed files with 15 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
llm/generate/gen_linux.sh
View file

@ -172,8 +172,15 @@ if [ -d "${CUDA_LIB_DIR}" ]; then
# Disabling has minimal performance effect while maintaining compatibility. # Disabling has minimal performance effect while maintaining compatibility.
ARM64_DEFS="-DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_CUDA_F16=off" ARM64_DEFS="-DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_CUDA_F16=off"
fi fi
CMAKE_DEFS="-DLLAMA_CUDA=on -DLLAMA_CUDA_FORCE_MMQ=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} ${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS}" # Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
CMAKE_DEFS="${CMAKE_DEFS} ${OLLAMA_CUSTOM_GPU_DEPS}" if [ -n "${OLLAMA_CUSTOM_GPU_DEFS}" ]; then
echo "OLLAMA_CUSTOM_GPU_DEFS=\"${OLLAMA_CUSTOM_GPU_DEFS}\""
CMAKE_CUDA_DEFS="-DLLAMA_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} ${OLLAMA_CUSTOM_GPU_DEFS}"
echo "Building custom GPU"
else
CMAKE_CUDA_DEFS="-DLLAMA_CUDA=on -DLLAMA_CUDA_FORCE_MMQ=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}"
fi
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS} ${CMAKE_CUDA_DEFS}"
BUILD_DIR="../build/linux/${ARCH}/cuda${CUDA_VARIANT}" BUILD_DIR="../build/linux/${ARCH}/cuda${CUDA_VARIANT}"
EXTRA_LIBS="-L${CUDA_LIB_DIR} -lcudart -lcublas -lcublasLt -lcuda" EXTRA_LIBS="-L${CUDA_LIB_DIR} -lcudart -lcublas -lcublasLt -lcuda"
build build
@ -218,7 +225,12 @@ if [ -d "${ROCM_PATH}" ]; then
fi fi
init_vars init_vars
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DLLAMA_HIPBLAS=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)" CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DLLAMA_HIPBLAS=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)"
CMAKE_DEFS="${CMAKE_DEFS} ${OLLAMA_CUSTOM_GPU_DEPS}" # Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
if [ -n "${OLLAMA_CUSTOM_GPU_DEFS}" ]; then
echo "OLLAMA_CUSTOM_GPU_DEFS=\"${OLLAMA_CUSTOM_GPU_DEFS}\""
CMAKE_DEFS="${CMAKE_DEFS} ${OLLAMA_CUSTOM_GPU_DEFS}"
echo "Building custom GPU"
fi
BUILD_DIR="../build/linux/${ARCH}/rocm${ROCM_VARIANT}" BUILD_DIR="../build/linux/${ARCH}/rocm${ROCM_VARIANT}"
EXTRA_LIBS="-L${ROCM_PATH}/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ -Wl,-rpath,\$ORIGIN/../../rocm/ -lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu" EXTRA_LIBS="-L${ROCM_PATH}/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ -Wl,-rpath,\$ORIGIN/../../rocm/ -lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu"
build build