mirror of
https://github.com/dogkeeper886/ollama37.git
synced 2025-12-11 00:07:07 +00:00
Refactor linux packaging
This adjusts linux to follow a similar model to windows with a discrete archive (zip/tgz) to cary the primary executable, and dependent libraries. Runners are still carried as payloads inside the main binary Darwin retain the payload model where the go binary is fully self contained.
This commit is contained in:
Daniel Hiltgen
3
llm/ext_server/CMakeLists.txt
vendored
3
llm/ext_server/CMakeLists.txt
vendored
@@ -1,12 +1,13 @@
|
||||
set(TARGET ollama_llama_server)
|
||||
option(LLAMA_SERVER_VERBOSE "Build verbose logging option for Server" ON)
|
||||
set(LLAMA_SERVER_LDFLAGS $ENV{LLAMA_SERVER_LDFLAGS})
|
||||
include_directories(${CMAKE_CURRENT_SOURCE_DIR})
|
||||
add_executable(${TARGET} server.cpp utils.hpp json.hpp httplib.h)
|
||||
install(TARGETS ${TARGET} RUNTIME)
|
||||
target_compile_definitions(${TARGET} PRIVATE
|
||||
SERVER_VERBOSE=$<BOOL:${LLAMA_SERVER_VERBOSE}>
|
||||
)
|
||||
target_link_libraries(${TARGET} PRIVATE ggml llama common llava ${CMAKE_THREAD_LIBS_INIT})
|
||||
target_link_libraries(${TARGET} PRIVATE ggml llama common llava ${CMAKE_THREAD_LIBS_INIT} ${LLAMA_SERVER_LDFLAGS})
|
||||
if (WIN32)
|
||||
TARGET_LINK_LIBRARIES(${TARGET} PRIVATE ws2_32)
|
||||
endif()
|
||||
|
||||
@@ -9,11 +9,14 @@ init_vars() {
|
||||
ARCH="arm64"
|
||||
;;
|
||||
*)
|
||||
ARCH=$(uname -m | sed -e "s/aarch64/arm64/g")
|
||||
echo "GOARCH must be set"
|
||||
echo "this script is meant to be run from within go generate"
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
|
||||
LLAMACPP_DIR=../llama.cpp
|
||||
CMAKE_DEFS=""
|
||||
CMAKE_DEFS="-DCMAKE_SKIP_RPATH=on"
|
||||
CMAKE_TARGETS="--target ollama_llama_server"
|
||||
if echo "${CGO_CFLAGS}" | grep -- '-g' >/dev/null; then
|
||||
CMAKE_DEFS="-DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_VERBOSE_MAKEFILE=on -DLLAMA_GPROF=on -DLLAMA_SERVER_VERBOSE=on ${CMAKE_DEFS}"
|
||||
@@ -27,6 +30,7 @@ init_vars() {
|
||||
WHOLE_ARCHIVE="-Wl,-force_load"
|
||||
NO_WHOLE_ARCHIVE=""
|
||||
GCC_ARCH="-arch ${ARCH}"
|
||||
DIST_BASE=../../dist/darwin-${GOARCH}/
|
||||
;;
|
||||
"Linux")
|
||||
LIB_EXT="so"
|
||||
@@ -35,6 +39,7 @@ init_vars() {
|
||||
|
||||
# Cross compiling not supported on linux - Use docker
|
||||
GCC_ARCH=""
|
||||
DIST_BASE=../../dist/linux-${GOARCH}/
|
||||
;;
|
||||
*)
|
||||
;;
|
||||
@@ -105,6 +110,14 @@ compress() {
|
||||
echo "Finished compression"
|
||||
}
|
||||
|
||||
install() {
|
||||
echo "Installing libraries to bin dir ${BUILD_DIR}/bin/"
|
||||
for lib in $(find ${BUILD_DIR} -name \*.${LIB_EXT}); do
|
||||
rm -f "${BUILD_DIR}/bin/$(basename ${lib})"
|
||||
cp -af "${lib}" "${BUILD_DIR}/bin/"
|
||||
done
|
||||
}
|
||||
|
||||
# Keep the local tree clean after we're done with the build
|
||||
cleanup() {
|
||||
(cd ${LLAMACPP_DIR}/ && git checkout CMakeLists.txt)
|
||||
|
||||
@@ -51,7 +51,7 @@ if [ -z "${CUDACXX}" ]; then
|
||||
export CUDACXX=$(command -v nvcc)
|
||||
fi
|
||||
fi
|
||||
COMMON_CMAKE_DEFS="-DBUILD_SHARED_LIBS=off -DCMAKE_POSITION_INDEPENDENT_CODE=on -DGGML_NATIVE=off -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_OPENMP=off"
|
||||
COMMON_CMAKE_DEFS="-DCMAKE_SKIP_RPATH=on -DBUILD_SHARED_LIBS=on -DCMAKE_POSITION_INDEPENDENT_CODE=on -DGGML_NATIVE=off -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_OPENMP=off"
|
||||
source $(dirname $0)/gen_common.sh
|
||||
init_vars
|
||||
git_module_setup
|
||||
@@ -77,10 +77,11 @@ if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
|
||||
if [ -n "${OLLAMA_CUSTOM_CPU_DEFS}" ]; then
|
||||
init_vars
|
||||
echo "OLLAMA_CUSTOM_CPU_DEFS=\"${OLLAMA_CUSTOM_CPU_DEFS}\""
|
||||
CMAKE_DEFS="${OLLAMA_CUSTOM_CPU_DEFS} -DBUILD_SHARED_LIBS=off -DCMAKE_POSITION_INDEPENDENT_CODE=on ${CMAKE_DEFS}"
|
||||
CMAKE_DEFS="${OLLAMA_CUSTOM_CPU_DEFS} -DBUILD_SHARED_LIBS=on -DCMAKE_POSITION_INDEPENDENT_CODE=on ${CMAKE_DEFS}"
|
||||
BUILD_DIR="../build/linux/${ARCH}/cpu"
|
||||
echo "Building custom CPU"
|
||||
build
|
||||
install
|
||||
compress
|
||||
else
|
||||
# Darwin Rosetta x86 emulation does NOT support AVX, AVX2, AVX512
|
||||
@@ -93,7 +94,7 @@ if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
|
||||
# -DGGML_AVX512_VBMI -- 2018 Intel Cannon Lake
|
||||
# -DGGML_AVX512_VNNI -- 2021 Intel Alder Lake
|
||||
|
||||
COMMON_CPU_DEFS="-DBUILD_SHARED_LIBS=off -DCMAKE_POSITION_INDEPENDENT_CODE=on -DGGML_NATIVE=off -DGGML_OPENMP=off"
|
||||
COMMON_CPU_DEFS="-DBUILD_SHARED_LIBS=on -DCMAKE_POSITION_INDEPENDENT_CODE=on -DGGML_NATIVE=off -DGGML_OPENMP=off"
|
||||
if [ -z "${OLLAMA_CPU_TARGET}" -o "${OLLAMA_CPU_TARGET}" = "cpu" ]; then
|
||||
#
|
||||
# CPU first for the default library, set up as lowest common denominator for maximum compatibility (including Rosetta)
|
||||
@@ -103,6 +104,7 @@ if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
|
||||
BUILD_DIR="../build/linux/${ARCH}/cpu"
|
||||
echo "Building LCD CPU"
|
||||
build
|
||||
install
|
||||
compress
|
||||
fi
|
||||
|
||||
@@ -120,6 +122,7 @@ if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
|
||||
BUILD_DIR="../build/linux/${ARCH}/cpu_avx"
|
||||
echo "Building AVX CPU"
|
||||
build
|
||||
install
|
||||
compress
|
||||
fi
|
||||
|
||||
@@ -133,6 +136,7 @@ if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
|
||||
BUILD_DIR="../build/linux/${ARCH}/cpu_avx2"
|
||||
echo "Building AVX2 CPU"
|
||||
build
|
||||
install
|
||||
compress
|
||||
fi
|
||||
fi
|
||||
@@ -178,29 +182,18 @@ if [ -z "${OLLAMA_SKIP_CUDA_GENERATE}" -a -d "${CUDA_LIB_DIR}" ]; then
|
||||
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} ${OLLAMA_CUSTOM_CUDA_DEFS}"
|
||||
echo "Building custom CUDA GPU"
|
||||
else
|
||||
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_FLAGS=-t8 -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}"
|
||||
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}"
|
||||
fi
|
||||
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS} ${CMAKE_CUDA_DEFS}"
|
||||
export CUDAFLAGS="-t8"
|
||||
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS} ${CMAKE_CUDA_DEFS} -DGGML_STATIC=off"
|
||||
BUILD_DIR="../build/linux/${ARCH}/cuda${CUDA_VARIANT}"
|
||||
EXTRA_LIBS="-L${CUDA_LIB_DIR} -lcudart -lcublas -lcublasLt -lcuda"
|
||||
export LLAMA_SERVER_LDFLAGS="-L${CUDA_LIB_DIR} -lcudart -lcublas -lcublasLt -lcuda"
|
||||
CUDA_DIST_DIR="${DIST_BASE}/ollama_libs"
|
||||
build
|
||||
|
||||
# Carry the CUDA libs as payloads to help reduce dependency burden on users
|
||||
#
|
||||
# TODO - in the future we may shift to packaging these separately and conditionally
|
||||
# downloading them in the install script.
|
||||
DEPS="$(ldd ${BUILD_DIR}/bin/ollama_llama_server )"
|
||||
for lib in libcudart.so libcublas.so libcublasLt.so ; do
|
||||
DEP=$(echo "${DEPS}" | grep ${lib} | cut -f1 -d' ' | xargs || true)
|
||||
if [ -n "${DEP}" -a -e "${CUDA_LIB_DIR}/${DEP}" ]; then
|
||||
cp "${CUDA_LIB_DIR}/${DEP}" "${BUILD_DIR}/bin/"
|
||||
elif [ -e "${CUDA_LIB_DIR}/${lib}.${CUDA_MAJOR}" ]; then
|
||||
cp "${CUDA_LIB_DIR}/${lib}.${CUDA_MAJOR}" "${BUILD_DIR}/bin/"
|
||||
elif [ -e "${CUDART_LIB_DIR}/${lib}" ]; then
|
||||
cp -d ${CUDART_LIB_DIR}/${lib}* "${BUILD_DIR}/bin/"
|
||||
else
|
||||
cp -d "${CUDA_LIB_DIR}/${lib}*" "${BUILD_DIR}/bin/"
|
||||
fi
|
||||
install
|
||||
mkdir -p "${CUDA_DIST_DIR}"
|
||||
for lib in ${CUDA_LIB_DIR}/libcudart.so* ${CUDA_LIB_DIR}/libcublas.so* ${CUDA_LIB_DIR}/libcublasLt.so* ; do
|
||||
cp -a "${lib}" "${CUDA_DIST_DIR}"
|
||||
done
|
||||
compress
|
||||
|
||||
@@ -218,21 +211,24 @@ if [ -z "${OLLAMA_SKIP_ONEAPI_GENERATE}" -a -d "${ONEAPI_ROOT}" ]; then
|
||||
CC=icx
|
||||
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL=ON -DGGML_SYCL_F16=OFF"
|
||||
BUILD_DIR="../build/linux/${ARCH}/oneapi"
|
||||
EXTRA_LIBS="-fsycl -Wl,-rpath,${ONEAPI_ROOT}/compiler/latest/lib,-rpath,${ONEAPI_ROOT}/mkl/latest/lib,-rpath,${ONEAPI_ROOT}/tbb/latest/lib,-rpath,${ONEAPI_ROOT}/compiler/latest/opt/oclfpga/linux64/lib -lOpenCL -lmkl_core -lmkl_sycl_blas -lmkl_intel_ilp64 -lmkl_tbb_thread -ltbb"
|
||||
ONEAPI_DIST_DIR="${DIST_BASE}/ollama_libs"
|
||||
export LLAMA_SERVER_LDFLAGS="-fsycl -lOpenCL -lmkl_core -lmkl_sycl_blas -lmkl_intel_ilp64 -lmkl_tbb_thread -ltbb"
|
||||
DEBUG_FLAGS="" # icx compiles with -O0 if we pass -g, so we must remove it
|
||||
build
|
||||
|
||||
# copy oneAPI dependencies
|
||||
mkdir -p "${ONEAPI_DIST_DIR}"
|
||||
for dep in $(ldd "${BUILD_DIR}/bin/ollama_llama_server" | grep "=>" | cut -f2 -d= | cut -f2 -d' ' | grep -e sycl -e mkl -e tbb); do
|
||||
cp "${dep}" "${BUILD_DIR}/bin/"
|
||||
cp -a "${dep}" "${ONEAPI_DIST_DIR}"
|
||||
done
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libOpenCL.so" "${BUILD_DIR}/bin/"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libimf.so" "${BUILD_DIR}/bin/"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libintlc.so.5" "${BUILD_DIR}/bin/"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libirng.so" "${BUILD_DIR}/bin/"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libpi_level_zero.so" "${BUILD_DIR}/bin/"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libsvml.so" "${BUILD_DIR}/bin/"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libur_loader.so.0" "${BUILD_DIR}/bin/"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libOpenCL.so" "${ONEAPI_DIST_DIR}"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libimf.so" "${ONEAPI_DIST_DIR}"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libintlc.so.5" "${ONEAPI_DIST_DIR}"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libirng.so" "${ONEAPI_DIST_DIR}"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libpi_level_zero.so" "${ONEAPI_DIST_DIR}"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libsvml.so" "${ONEAPI_DIST_DIR}"
|
||||
cp "${ONEAPI_ROOT}/compiler/latest/lib/libur_loader.so.0" "${ONEAPI_DIST_DIR}"
|
||||
install
|
||||
compress
|
||||
fi
|
||||
|
||||
@@ -262,21 +258,18 @@ if [ -z "${OLLAMA_SKIP_ROCM_GENERATE}" -a -d "${ROCM_PATH}" ]; then
|
||||
echo "Building custom ROCM GPU"
|
||||
fi
|
||||
BUILD_DIR="../build/linux/${ARCH}/rocm${ROCM_VARIANT}"
|
||||
EXTRA_LIBS="-L${ROCM_PATH}/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ -Wl,-rpath,\$ORIGIN/../../rocm/ -lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu"
|
||||
ROCM_DIST_DIR="${DIST_BASE}/ollama_libs"
|
||||
# TODO figure out how to disable runpath (rpath)
|
||||
# export CMAKE_HIP_FLAGS="-fno-rtlib-add-rpath" # doesn't work
|
||||
export LLAMA_SERVER_LDFLAGS="-L${ROCM_PATH}/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ -lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu"
|
||||
build
|
||||
|
||||
# Record the ROCM dependencies
|
||||
rm -f "${BUILD_DIR}/bin/deps.txt"
|
||||
touch "${BUILD_DIR}/bin/deps.txt"
|
||||
for dep in $(ldd "${BUILD_DIR}/bin/ollama_llama_server" | grep "=>" | cut -f2 -d= | cut -f2 -d' ' | grep -e rocm -e amdgpu -e libtinfo ); do
|
||||
echo "${dep}" >> "${BUILD_DIR}/bin/deps.txt"
|
||||
# copy the ROCM dependencies
|
||||
mkdir -p "${ROCM_DIST_DIR}"
|
||||
for dep in $(ldd "${BUILD_DIR}/bin/ollama_llama_server" | grep "=>" | cut -f2 -d= | cut -f2 -d' ' | grep -v "${ARCH}/rocm${ROCM_VARIANT}" | grep -e rocm -e amdgpu -e libtinfo ); do
|
||||
cp -a "${dep}"* "${ROCM_DIST_DIR}"
|
||||
done
|
||||
# bomb out if for some reason we didn't get a few deps
|
||||
if [ $(cat "${BUILD_DIR}/bin/deps.txt" | wc -l ) -lt 8 ] ; then
|
||||
cat "${BUILD_DIR}/bin/deps.txt"
|
||||
echo "ERROR: deps file short"
|
||||
exit 1
|
||||
fi
|
||||
install
|
||||
compress
|
||||
fi
|
||||
|
||||
|
||||
@@ -286,12 +286,11 @@ function build_cuda() {
|
||||
sign
|
||||
install
|
||||
|
||||
rm -ea 0 -recurse -force -path "${script:SRC_DIR}\dist\windows-${script:ARCH}\cuda\"
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\cuda\" -ea 0 > $null
|
||||
write-host "copying CUDA dependencies to ${script:SRC_DIR}\dist\windows-${script:ARCH}\cuda\"
|
||||
cp "${script:CUDA_LIB_DIR}\cudart64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\cuda\"
|
||||
cp "${script:CUDA_LIB_DIR}\cublas64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\cuda\"
|
||||
cp "${script:CUDA_LIB_DIR}\cublasLt64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\cuda\"
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\" -ea 0 > $null
|
||||
write-host "copying CUDA dependencies to ${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${script:CUDA_LIB_DIR}\cudart64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${script:CUDA_LIB_DIR}\cublas64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${script:CUDA_LIB_DIR}\cublasLt64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
} else {
|
||||
write-host "Skipping CUDA generation step"
|
||||
}
|
||||
@@ -325,18 +324,17 @@ function build_oneapi() {
|
||||
sign
|
||||
install
|
||||
|
||||
rm -ea 0 -recurse -force -path "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\" -ea 0 > $null
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\libirngmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\libmmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_level_zero.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_unified_runtime.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_win_proxy_loader.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\svml_dispmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\sycl7.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_core.2.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_sycl_blas.4.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_tbb_thread.2.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\oneapi\"
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\" -ea 0 > $null
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\libirngmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\libmmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_level_zero.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_unified_runtime.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_win_proxy_loader.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\svml_dispmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\sycl7.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_core.2.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_sycl_blas.4.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_tbb_thread.2.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
} else {
|
||||
Write-Host "Skipping oneAPI generation step"
|
||||
}
|
||||
@@ -386,12 +384,11 @@ function build_rocm() {
|
||||
sign
|
||||
install
|
||||
|
||||
rm -ea 0 -recurse -force -path "${script:SRC_DIR}\dist\windows-${script:ARCH}\rocm\"
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\rocm\rocblas\library\" -ea 0 > $null
|
||||
cp "${env:HIP_PATH}\bin\hipblas.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\rocm\"
|
||||
cp "${env:HIP_PATH}\bin\rocblas.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\rocm\"
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\rocblas\library\" -ea 0 > $null
|
||||
cp "${env:HIP_PATH}\bin\hipblas.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
cp "${env:HIP_PATH}\bin\rocblas.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\"
|
||||
# amdhip64.dll dependency comes from the driver and must be installed on the host to use AMD GPUs
|
||||
cp "${env:HIP_PATH}\bin\rocblas\library\*" "${script:SRC_DIR}\dist\windows-${script:ARCH}\rocm\rocblas\library\"
|
||||
cp "${env:HIP_PATH}\bin\rocblas\library\*" "${script:SRC_DIR}\dist\windows-${script:ARCH}\ollama_libs\rocblas\library\"
|
||||
} else {
|
||||
write-host "Skipping ROCm generation step"
|
||||
}
|
||||
|
||||
@@ -306,20 +306,18 @@ func NewLlamaServer(gpus gpu.GpuInfoList, model string, ggml *GGML, adapters, pr
|
||||
if runtime.GOOS == "windows" {
|
||||
pathEnv = "PATH"
|
||||
}
|
||||
// prepend the server directory to LD_LIBRARY_PATH/PATH and the parent dir for common dependencies
|
||||
libraryPaths := []string{dir, filepath.Dir(dir)}
|
||||
// Start with the server directory for the LD_LIBRARY_PATH/PATH
|
||||
libraryPaths := []string{dir}
|
||||
|
||||
if libraryPath, ok := os.LookupEnv(pathEnv); ok {
|
||||
// Append our runner directory to the path
|
||||
// This will favor system libraries over our bundled library dependencies
|
||||
// favor our bundled library dependencies over system libraries
|
||||
libraryPaths = append(libraryPaths, filepath.SplitList(libraryPath)...)
|
||||
}
|
||||
|
||||
// Note: we always put the dependency path first
|
||||
// since this was the exact version we verified for AMD GPUs
|
||||
// and we favor what the user had in their path
|
||||
// since this was the exact version we compiled/linked against
|
||||
if gpus[0].DependencyPath != "" {
|
||||
// TODO refine for multi-gpu support
|
||||
// assume gpus from the same library have the same dependency path
|
||||
libraryPaths = append([]string{gpus[0].DependencyPath}, libraryPaths...)
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user