mirror of
https://github.com/likelovewant/ollama-for-amd.git
synced 2025-12-24 07:28:27 +00:00
Remove submodule and shift to Go server - 0.4.0 (#7157)
* Remove llama.cpp submodule and shift new build to top * CI: install msys and clang gcc on win Needed for deepseek to work properly on windows
This commit is contained in:
Daniel Hiltgen
@@ -1,137 +0,0 @@
|
||||
# common logic across linux and darwin
|
||||
|
||||
init_vars() {
|
||||
case "${GOARCH}" in
|
||||
"amd64")
|
||||
ARCH="x86_64"
|
||||
;;
|
||||
"arm64")
|
||||
ARCH="arm64"
|
||||
;;
|
||||
*)
|
||||
echo "GOARCH must be set"
|
||||
echo "this script is meant to be run from within go generate"
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
|
||||
LLAMACPP_DIR=../llama.cpp
|
||||
CMAKE_DEFS="-DCMAKE_SKIP_RPATH=on"
|
||||
CMAKE_TARGETS="--target ollama_llama_server"
|
||||
if echo "${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}"
|
||||
else
|
||||
# TODO - add additional optimization flags...
|
||||
CMAKE_DEFS="-DCMAKE_BUILD_TYPE=Release -DLLAMA_SERVER_VERBOSE=off ${CMAKE_DEFS}"
|
||||
fi
|
||||
case $(uname -s) in
|
||||
"Darwin")
|
||||
LIB_EXT="dylib"
|
||||
WHOLE_ARCHIVE="-Wl,-force_load"
|
||||
NO_WHOLE_ARCHIVE=""
|
||||
GCC_ARCH="-arch ${ARCH}"
|
||||
DIST_BASE=../../dist/darwin-${GOARCH}/
|
||||
PAYLOAD_BASE=../../build/darwin/${GOARCH}
|
||||
;;
|
||||
"Linux")
|
||||
LIB_EXT="so"
|
||||
WHOLE_ARCHIVE="-Wl,--whole-archive"
|
||||
NO_WHOLE_ARCHIVE="-Wl,--no-whole-archive"
|
||||
|
||||
# Cross compiling not supported on linux - Use docker
|
||||
GCC_ARCH=""
|
||||
DIST_BASE=../../dist/linux-${GOARCH}/
|
||||
PAYLOAD_BASE=../../build/linux/${GOARCH}
|
||||
;;
|
||||
*)
|
||||
;;
|
||||
esac
|
||||
if [ -z "${CMAKE_CUDA_ARCHITECTURES}" ] ; then
|
||||
CMAKE_CUDA_ARCHITECTURES="50;52;61;70;75;80"
|
||||
fi
|
||||
GZIP=$(command -v pigz 2>/dev/null || echo "gzip")
|
||||
RUNNER_BASE="${DIST_BASE}/lib/ollama/runners"
|
||||
}
|
||||
|
||||
git_module_setup() {
|
||||
if [ -n "${OLLAMA_SKIP_PATCHING}" ]; then
|
||||
echo "Skipping submodule initialization"
|
||||
return
|
||||
fi
|
||||
# Make sure the tree is clean after the directory moves
|
||||
if [ -d "${LLAMACPP_DIR}/gguf" ]; then
|
||||
echo "Cleaning up old submodule"
|
||||
rm -rf ${LLAMACPP_DIR}
|
||||
fi
|
||||
git submodule init
|
||||
git submodule update --force ${LLAMACPP_DIR}
|
||||
|
||||
}
|
||||
|
||||
apply_patches() {
|
||||
# apply temporary patches until fix is upstream
|
||||
for patch in ../patches/*.patch; do
|
||||
git -c 'user.name=nobody' -c 'user.email=<>' -C ${LLAMACPP_DIR} am ${patch}
|
||||
done
|
||||
}
|
||||
|
||||
build() {
|
||||
cmake -S ${LLAMACPP_DIR} -B ${BUILD_DIR} ${CMAKE_DEFS}
|
||||
cmake --build ${BUILD_DIR} ${CMAKE_TARGETS} -j8
|
||||
# remove unnecessary build artifacts
|
||||
rm -f ${BUILD_DIR}/bin/ggml-common.h ${BUILD_DIR}/bin/ggml-metal.metal
|
||||
}
|
||||
|
||||
dist() {
|
||||
[ -z "${RUNNER}" ] && exit 1
|
||||
mkdir -p ${RUNNER_BASE}/${RUNNER}/
|
||||
for f in ${BUILD_DIR}/bin/* ; do
|
||||
cp ${f} ${RUNNER_BASE}/${RUNNER}/
|
||||
done
|
||||
# check for lib directory
|
||||
if [ -d ${BUILD_DIR}/lib ]; then
|
||||
for f in ${BUILD_DIR}/lib/* ; do
|
||||
cp ${f} ${RUNNER_BASE}/${RUNNER}/
|
||||
done
|
||||
fi
|
||||
}
|
||||
|
||||
# Compress from the build $BUILD_DIR into the $PAYLOAD_BASE/$RUNNER dir
|
||||
compress() {
|
||||
[ -z "${RUNNER}" ] && exit 1
|
||||
echo "Compressing payloads with ${GZIP} to reduce overall binary size..."
|
||||
rm -rf "${PAYLOAD_BASE}/${RUNNER}/"
|
||||
mkdir -p "${PAYLOAD_BASE}/${RUNNER}/"
|
||||
for f in ${BUILD_DIR}/bin/* ; do
|
||||
${GZIP} -c --best ${f} > "${PAYLOAD_BASE}/${RUNNER}/$(basename ${f}).gz" &
|
||||
compress_pids+=" $!"
|
||||
done
|
||||
# check for lib directory
|
||||
if [ -d ${BUILD_DIR}/lib ]; then
|
||||
for f in ${BUILD_DIR}/lib/* ; do
|
||||
${GZIP} -c --best ${f} > "${PAYLOAD_BASE}/${RUNNER}/$(basename ${f}).gz" &
|
||||
compress_pids+=" $!"
|
||||
done
|
||||
fi
|
||||
echo
|
||||
}
|
||||
|
||||
wait_for_compress() {
|
||||
for pid in ${compress_pids}; do
|
||||
wait $pid
|
||||
done
|
||||
echo "Finished compression"
|
||||
}
|
||||
|
||||
install() {
|
||||
echo "Installing libraries to bin dir ${BUILD_DIR}/bin/"
|
||||
for lib in $(find ${BUILD_DIR} -name \*.${LIB_EXT} | grep -v "${BUILD_DIR}/bin/" ); do
|
||||
rm -f "${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() {
|
||||
git submodule update --force ${LLAMACPP_DIR}
|
||||
}
|
||||
@@ -1,91 +0,0 @@
|
||||
#!/bin/bash
|
||||
# This script is intended to run inside the go generate
|
||||
# working directory must be ./llm/generate/
|
||||
|
||||
# TODO - add hardening to detect missing tools (cmake, etc.)
|
||||
|
||||
set -ex
|
||||
set -o pipefail
|
||||
compress_pids=""
|
||||
echo "Starting darwin generate script"
|
||||
source $(dirname $0)/gen_common.sh
|
||||
init_vars
|
||||
git_module_setup
|
||||
apply_patches
|
||||
|
||||
sign() {
|
||||
if [ -n "$APPLE_IDENTITY" ]; then
|
||||
codesign -f --timestamp --deep --options=runtime --sign "$APPLE_IDENTITY" --identifier ai.ollama.ollama $1
|
||||
fi
|
||||
}
|
||||
|
||||
COMMON_DARWIN_DEFS="-DBUILD_SHARED_LIBS=off -DCMAKE_OSX_DEPLOYMENT_TARGET=11.3 -DGGML_METAL_MACOSX_VERSION_MIN=11.3 -DCMAKE_SYSTEM_NAME=Darwin -DGGML_METAL_EMBED_LIBRARY=on -DGGML_OPENMP=off"
|
||||
|
||||
case "${GOARCH}" in
|
||||
"amd64")
|
||||
COMMON_CPU_DEFS="${COMMON_DARWIN_DEFS} -DCMAKE_SYSTEM_PROCESSOR=${ARCH} -DCMAKE_OSX_ARCHITECTURES=${ARCH} -DGGML_METAL=off -DGGML_NATIVE=off"
|
||||
|
||||
if [ -z "$OLLAMA_SKIP_CPU_GENERATE" ]; then
|
||||
#
|
||||
# CPU first for the default library, set up as lowest common denominator for maximum compatibility (including Rosetta)
|
||||
#
|
||||
init_vars
|
||||
CMAKE_DEFS="${COMMON_CPU_DEFS} -DGGML_ACCELERATE=off -DGGML_BLAS=off -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off ${CMAKE_DEFS}"
|
||||
RUNNER=cpu
|
||||
BUILD_DIR="../build/darwin/${GOARCH}/${RUNNER}"
|
||||
echo "Building LCD CPU"
|
||||
build
|
||||
sign ${BUILD_DIR}/bin/ollama_llama_server
|
||||
compress
|
||||
|
||||
#
|
||||
# ~2011 CPU Dynamic library with more capabilities turned on to optimize performance
|
||||
# Approximately 400% faster than LCD on same CPU
|
||||
#
|
||||
init_vars
|
||||
CMAKE_DEFS="${COMMON_CPU_DEFS} -DGGML_ACCELERATE=off -DGGML_BLAS=off -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off ${CMAKE_DEFS}"
|
||||
RUNNER=cpu_avx
|
||||
BUILD_DIR="../build/darwin/${GOARCH}/${RUNNER}"
|
||||
echo "Building AVX CPU"
|
||||
build
|
||||
sign ${BUILD_DIR}/bin/ollama_llama_server
|
||||
compress
|
||||
|
||||
#
|
||||
# ~2013 CPU Dynamic library
|
||||
# Approximately 10% faster than AVX on same CPU
|
||||
#
|
||||
init_vars
|
||||
CMAKE_DEFS="${COMMON_CPU_DEFS} -DGGML_ACCELERATE=on -DGGML_BLAS=off -DGGML_AVX=on -DGGML_AVX2=on -DGGML_AVX512=off -DGGML_FMA=on -DGGML_F16C=on ${CMAKE_DEFS}"
|
||||
RUNNER=cpu_avx2
|
||||
BUILD_DIR="../build/darwin/${GOARCH}/${RUNNER}"
|
||||
echo "Building AVX2 CPU"
|
||||
EXTRA_LIBS="${EXTRA_LIBS} -framework Accelerate -framework Foundation"
|
||||
build
|
||||
sign ${BUILD_DIR}/bin/ollama_llama_server
|
||||
compress
|
||||
fi
|
||||
;;
|
||||
"arm64")
|
||||
|
||||
if [ -z "$OLLAMA_SKIP_METAL_GENERATE" ]; then
|
||||
init_vars
|
||||
CMAKE_DEFS="${COMMON_DARWIN_DEFS} -DCMAKE_SYSTEM_PROCESSOR=${ARCH} -DCMAKE_OSX_ARCHITECTURES=${ARCH} ${CMAKE_DEFS}"
|
||||
RUNNER="metal"
|
||||
BUILD_DIR="../build/darwin/${GOARCH}/${RUNNER}"
|
||||
EXTRA_LIBS="${EXTRA_LIBS} -framework Accelerate -framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders"
|
||||
build
|
||||
sign ${BUILD_DIR}/bin/ollama_llama_server
|
||||
compress
|
||||
fi
|
||||
;;
|
||||
*)
|
||||
echo "GOARCH must be set"
|
||||
echo "this script is meant to be run from within go generate"
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
|
||||
cleanup
|
||||
wait_for_compress
|
||||
echo "go generate completed. LLM runners: $(cd ${BUILD_DIR}/..; echo *)"
|
||||
@@ -1,285 +0,0 @@
|
||||
#!/bin/bash
|
||||
# This script is intended to run inside the go generate
|
||||
# working directory must be llm/generate/
|
||||
|
||||
# First we build one or more CPU based LLM libraries
|
||||
#
|
||||
# Then if we detect CUDA, we build a CUDA dynamic library, and carry the required
|
||||
# library dependencies
|
||||
#
|
||||
# Then if we detect ROCm, we build a dynamically loaded ROCm lib. The ROCM
|
||||
# libraries are quite large, and also dynamically load data files at runtime
|
||||
# which in turn are large, so we don't attempt to cary them as payload
|
||||
|
||||
set -ex
|
||||
set -o pipefail
|
||||
compress_pids=""
|
||||
|
||||
# See https://llvm.org/docs/AMDGPUUsage.html#processors for reference
|
||||
amdGPUs() {
|
||||
if [ -n "${AMDGPU_TARGETS}" ]; then
|
||||
echo "${AMDGPU_TARGETS}"
|
||||
return
|
||||
fi
|
||||
GPU_LIST=(
|
||||
"gfx900"
|
||||
"gfx906:xnack-"
|
||||
"gfx908:xnack-"
|
||||
"gfx90a:xnack+"
|
||||
"gfx90a:xnack-"
|
||||
"gfx940"
|
||||
"gfx941"
|
||||
"gfx942"
|
||||
"gfx1010"
|
||||
"gfx1012"
|
||||
"gfx1030"
|
||||
"gfx1100"
|
||||
"gfx1101"
|
||||
"gfx1102"
|
||||
)
|
||||
(
|
||||
IFS=$';'
|
||||
echo "'${GPU_LIST[*]}'"
|
||||
)
|
||||
}
|
||||
|
||||
echo "Starting linux generate script"
|
||||
if [ -z "${CUDACXX}" ]; then
|
||||
if [ -x /usr/local/cuda/bin/nvcc ]; then
|
||||
export CUDACXX=/usr/local/cuda/bin/nvcc
|
||||
else
|
||||
# Try the default location in case it exists
|
||||
export CUDACXX=$(command -v nvcc)
|
||||
fi
|
||||
fi
|
||||
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 $0)/gen_common.sh
|
||||
init_vars
|
||||
git_module_setup
|
||||
apply_patches
|
||||
|
||||
init_vars
|
||||
if [ -z "${OLLAMA_SKIP_CPU_GENERATE}" ]; then
|
||||
# Users building from source can tune the exact flags we pass to cmake for configuring
|
||||
# llama.cpp, and we'll build only 1 CPU variant in that case as the default.
|
||||
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=on -DCMAKE_POSITION_INDEPENDENT_CODE=on ${CMAKE_DEFS}"
|
||||
RUNNER="cpu"
|
||||
BUILD_DIR="../build/linux/${GOARCH}/${RUNNER}"
|
||||
echo "Building custom CPU"
|
||||
build
|
||||
install
|
||||
dist
|
||||
compress
|
||||
else
|
||||
# Darwin Rosetta x86 emulation does NOT support AVX, AVX2, AVX512
|
||||
# -DGGML_AVX -- 2011 Intel Sandy Bridge & AMD Bulldozer
|
||||
# -DGGML_F16C -- 2012 Intel Ivy Bridge & AMD 2011 Bulldozer (No significant improvement over just AVX)
|
||||
# -DGGML_AVX2 -- 2013 Intel Haswell & 2015 AMD Excavator / 2017 AMD Zen
|
||||
# -DGGML_FMA (FMA3) -- 2013 Intel Haswell & 2012 AMD Piledriver
|
||||
# Note: the following seem to yield slower results than AVX2 - ymmv
|
||||
# -DGGML_AVX512 -- 2017 Intel Skylake and High End DeskTop (HEDT)
|
||||
# -DGGML_AVX512_VBMI -- 2018 Intel Cannon Lake
|
||||
# -DGGML_AVX512_VNNI -- 2021 Intel Alder Lake
|
||||
|
||||
COMMON_CPU_DEFS="-DBUILD_SHARED_LIBS=on -DCMAKE_POSITION_INDEPENDENT_CODE=on -DGGML_NATIVE=off -DGGML_OPENMP=off"
|
||||
if [ -z "${OLLAMA_CPU_TARGET}" -o "${OLLAMA_CPU_TARGET}" = "cpu" ]; then
|
||||
#
|
||||
# CPU first for the default library, set up as lowest common denominator for maximum compatibility (including Rosetta)
|
||||
#
|
||||
init_vars
|
||||
CMAKE_DEFS="${COMMON_CPU_DEFS} -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off ${CMAKE_DEFS}"
|
||||
RUNNER=cpu
|
||||
BUILD_DIR="../build/linux/${GOARCH}/${RUNNER}"
|
||||
echo "Building LCD CPU"
|
||||
build
|
||||
install
|
||||
dist
|
||||
compress
|
||||
fi
|
||||
|
||||
if [ "${ARCH}" == "x86_64" ]; then
|
||||
#
|
||||
# ARM chips in M1/M2/M3-based MACs and NVidia Tegra devices do not currently support avx extensions.
|
||||
#
|
||||
if [ -z "${OLLAMA_CPU_TARGET}" -o "${OLLAMA_CPU_TARGET}" = "cpu_avx" ]; then
|
||||
#
|
||||
# ~2011 CPU Dynamic library with more capabilities turned on to optimize performance
|
||||
# Approximately 400% faster than LCD on same CPU
|
||||
#
|
||||
init_vars
|
||||
CMAKE_DEFS="${COMMON_CPU_DEFS} -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off ${CMAKE_DEFS}"
|
||||
RUNNER=cpu_avx
|
||||
BUILD_DIR="../build/linux/${GOARCH}/${RUNNER}"
|
||||
echo "Building AVX CPU"
|
||||
build
|
||||
install
|
||||
dist
|
||||
compress
|
||||
fi
|
||||
|
||||
if [ -z "${OLLAMA_CPU_TARGET}" -o "${OLLAMA_CPU_TARGET}" = "cpu_avx2" ]; then
|
||||
#
|
||||
# ~2013 CPU Dynamic library
|
||||
# Approximately 10% faster than AVX on same CPU
|
||||
#
|
||||
init_vars
|
||||
CMAKE_DEFS="${COMMON_CPU_DEFS} -DGGML_AVX=on -DGGML_AVX2=on -DGGML_AVX512=off -DGGML_FMA=on -DGGML_F16C=on ${CMAKE_DEFS}"
|
||||
RUNNER=cpu_avx2
|
||||
BUILD_DIR="../build/linux/${GOARCH}/${RUNNER}"
|
||||
echo "Building AVX2 CPU"
|
||||
build
|
||||
install
|
||||
dist
|
||||
compress
|
||||
fi
|
||||
fi
|
||||
fi
|
||||
else
|
||||
echo "Skipping CPU generation step as requested"
|
||||
fi
|
||||
|
||||
# If needed, look for the default CUDA toolkit location
|
||||
if [ -z "${CUDA_LIB_DIR}" ] && [ -d /usr/local/cuda/lib64 ]; then
|
||||
CUDA_LIB_DIR=/usr/local/cuda/lib64
|
||||
fi
|
||||
|
||||
# If needed, look for CUDA on Arch Linux
|
||||
if [ -z "${CUDA_LIB_DIR}" ] && [ -d /opt/cuda/targets/x86_64-linux/lib ]; then
|
||||
CUDA_LIB_DIR=/opt/cuda/targets/x86_64-linux/lib
|
||||
fi
|
||||
|
||||
# Allow override in case libcudart is in the wrong place
|
||||
if [ -z "${CUDART_LIB_DIR}" ]; then
|
||||
CUDART_LIB_DIR="${CUDA_LIB_DIR}"
|
||||
fi
|
||||
|
||||
if [ -z "${OLLAMA_SKIP_CUDA_GENERATE}" -a -d "${CUDA_LIB_DIR}" ]; then
|
||||
echo "CUDA libraries detected - building dynamic CUDA library"
|
||||
init_vars
|
||||
CUDA_MAJOR=$(ls "${CUDA_LIB_DIR}"/libcudart.so.* | head -1 | cut -f3 -d. || true)
|
||||
if [ -n "${CUDA_MAJOR}" -a -z "${CUDA_VARIANT}" ]; then
|
||||
CUDA_VARIANT=_v${CUDA_MAJOR}
|
||||
fi
|
||||
if [ "${ARCH}" == "arm64" ]; then
|
||||
echo "ARM CPU detected - disabling unsupported AVX instructions"
|
||||
|
||||
# ARM-based CPUs such as M1 and Tegra do not support AVX extensions.
|
||||
#
|
||||
# CUDA compute < 6.0 lacks proper FP16 support on ARM.
|
||||
# Disabling has minimal performance effect while maintaining compatibility.
|
||||
ARM64_DEFS="-DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_CUDA_F16=off"
|
||||
fi
|
||||
# Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
|
||||
if [ -n "${OLLAMA_CUSTOM_CUDA_DEFS}" ]; then
|
||||
echo "OLLAMA_CUSTOM_CUDA_DEFS=\"${OLLAMA_CUSTOM_CUDA_DEFS}\""
|
||||
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} ${OLLAMA_CUSTOM_CUDA_DEFS}"
|
||||
echo "Building custom CUDA GPU"
|
||||
else
|
||||
CMAKE_CUDA_DEFS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}"
|
||||
fi
|
||||
export CUDAFLAGS="-t8"
|
||||
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS} ${CMAKE_CUDA_DEFS} -DGGML_STATIC=off"
|
||||
RUNNER=cuda${CUDA_VARIANT}
|
||||
BUILD_DIR="../build/linux/${GOARCH}/${RUNNER}"
|
||||
export LLAMA_SERVER_LDFLAGS="-L${CUDA_LIB_DIR} -lcudart -lcublas -lcublasLt -lcuda"
|
||||
CUDA_DIST_DIR="${CUDA_DIST_DIR:-${DIST_BASE}/lib/ollama}"
|
||||
build
|
||||
install
|
||||
dist
|
||||
echo "Installing CUDA dependencies in ${CUDA_DIST_DIR}"
|
||||
mkdir -p "${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}" "${CUDA_DIST_DIR}"
|
||||
done
|
||||
compress
|
||||
|
||||
fi
|
||||
|
||||
if [ -z "${ONEAPI_ROOT}" ]; then
|
||||
# Try the default location in case it exists
|
||||
ONEAPI_ROOT=/opt/intel/oneapi
|
||||
fi
|
||||
|
||||
if [ -z "${OLLAMA_SKIP_ONEAPI_GENERATE}" -a -d "${ONEAPI_ROOT}" ]; then
|
||||
echo "OneAPI libraries detected - building dynamic OneAPI library"
|
||||
init_vars
|
||||
source ${ONEAPI_ROOT}/setvars.sh --force # set up environment variables for oneAPI
|
||||
CC=icx
|
||||
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL=ON -DGGML_SYCL_F16=OFF"
|
||||
RUNNER=oneapi
|
||||
BUILD_DIR="../build/linux/${GOARCH}/${RUNNER}"
|
||||
ONEAPI_DIST_DIR="${DIST_BASE}/lib/ollama"
|
||||
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 -a "${dep}" "${ONEAPI_DIST_DIR}"
|
||||
done
|
||||
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
|
||||
dist
|
||||
compress
|
||||
fi
|
||||
|
||||
if [ -z "${ROCM_PATH}" ]; then
|
||||
# Try the default location in case it exists
|
||||
ROCM_PATH=/opt/rocm
|
||||
fi
|
||||
|
||||
if [ -z "${CLBlast_DIR}" ]; then
|
||||
# Try the default location in case it exists
|
||||
if [ -d /usr/lib/cmake/CLBlast ]; then
|
||||
export CLBlast_DIR=/usr/lib/cmake/CLBlast
|
||||
fi
|
||||
fi
|
||||
|
||||
if [ -z "${OLLAMA_SKIP_ROCM_GENERATE}" -a -d "${ROCM_PATH}" ]; then
|
||||
echo "ROCm libraries detected - building dynamic ROCm library"
|
||||
if [ -f ${ROCM_PATH}/lib/librocblas.so.*.*.????? ]; then
|
||||
ROCM_VARIANT=_v$(ls ${ROCM_PATH}/lib/librocblas.so.*.*.????? | cut -f5 -d. || true)
|
||||
fi
|
||||
init_vars
|
||||
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DGGML_HIPBLAS=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)"
|
||||
# Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
|
||||
if [ -n "${OLLAMA_CUSTOM_ROCM_DEFS}" ]; then
|
||||
echo "OLLAMA_CUSTOM_ROCM_DEFS=\"${OLLAMA_CUSTOM_ROCM_DEFS}\""
|
||||
CMAKE_DEFS="${CMAKE_DEFS} ${OLLAMA_CUSTOM_ROCM_DEFS}"
|
||||
echo "Building custom ROCM GPU"
|
||||
fi
|
||||
RUNNER=rocm${ROCM_VARIANT}
|
||||
BUILD_DIR="../build/linux/${GOARCH}/${RUNNER}"
|
||||
# ROCm dependencies are too large to fit into a unified bundle
|
||||
ROCM_DIST_DIR="${DIST_BASE}/../linux-${GOARCH}-rocm/lib/ollama"
|
||||
# TODO figure out how to disable runpath (rpath)
|
||||
# export CMAKE_HIP_FLAGS="-fno-rtlib-add-rpath" # doesn't work
|
||||
export LLAMA_SERVER_LDFLAGS="-L${ROCM_PATH}/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ -lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu"
|
||||
build
|
||||
|
||||
# copy the ROCM dependencies
|
||||
mkdir -p "${ROCM_DIST_DIR}"
|
||||
for dep in $(ldd "${BUILD_DIR}/bin/ollama_llama_server" | grep "=>" | cut -f2 -d= | cut -f2 -d' ' | grep -v "${GOARCH}/rocm${ROCM_VARIANT}" | grep -e rocm -e amdgpu -e libtinfo -e libnuma -e libelf ); do
|
||||
cp -a "${dep}"* "${ROCM_DIST_DIR}"
|
||||
if [ $(readlink -f "${dep}") != "${dep}" ] ; then
|
||||
cp $(readlink -f "${dep}") "${ROCM_DIST_DIR}"
|
||||
fi
|
||||
done
|
||||
install
|
||||
dist
|
||||
compress
|
||||
fi
|
||||
|
||||
cleanup
|
||||
wait_for_compress
|
||||
echo "go generate completed. LLM runners: $(cd ${PAYLOAD_BASE}; echo *)"
|
||||
@@ -1,403 +0,0 @@
|
||||
#!powershell
|
||||
|
||||
$ErrorActionPreference = "Stop"
|
||||
|
||||
function amdGPUs {
|
||||
if ($env:AMDGPU_TARGETS) {
|
||||
return $env:AMDGPU_TARGETS
|
||||
}
|
||||
# Current supported rocblas list from ROCm v6.1.2 on windows
|
||||
# https://rocm.docs.amd.com/projects/install-on-windows/en/latest/reference/system-requirements.html#windows-supported-gpus
|
||||
$GPU_LIST = @(
|
||||
"gfx1030"
|
||||
"gfx1100"
|
||||
"gfx1101"
|
||||
"gfx1102"
|
||||
)
|
||||
$GPU_LIST -join ';'
|
||||
}
|
||||
|
||||
|
||||
function init_vars {
|
||||
write-host "Checking for cmake..."
|
||||
get-command cmake
|
||||
write-host "Checking for ninja..."
|
||||
$d=(get-command -ea 'silentlycontinue' ninja).path
|
||||
if ($null -eq $d) {
|
||||
$MSVC_INSTALL=(Get-CimInstance MSFT_VSInstance -Namespace root/cimv2/vs)[0].InstallLocation
|
||||
$matches=(gci -path $MSVC_INSTALL -r -fi ninja.exe)
|
||||
if ($matches.count -eq 0) {
|
||||
throw "Unable to locate ninja"
|
||||
}
|
||||
$ninjaDir=($matches[0].FullName | split-path -parent)
|
||||
$env:PATH="$env:PATH;$ninjaDir"
|
||||
}
|
||||
if (!$script:SRC_DIR) {
|
||||
$script:SRC_DIR = $(resolve-path "..\..\")
|
||||
}
|
||||
if (!$script:llamacppDir) {
|
||||
$script:llamacppDir = "../llama.cpp"
|
||||
}
|
||||
if (!$script:cmakeTargets) {
|
||||
$script:cmakeTargets = @("ollama_llama_server")
|
||||
}
|
||||
$script:cmakeDefs = @(
|
||||
"-DBUILD_SHARED_LIBS=on",
|
||||
"-DGGML_NATIVE=off",
|
||||
"-DGGML_OPENMP=off"
|
||||
)
|
||||
$script:commonCpuDefs = @("-DCMAKE_POSITION_INDEPENDENT_CODE=on")
|
||||
$script:ARCH = $Env:PROCESSOR_ARCHITECTURE.ToLower()
|
||||
$script:DIST_BASE = "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\runners"
|
||||
md "$script:DIST_BASE" -ea 0 > $null
|
||||
if ($env:CGO_CFLAGS -contains "-g") {
|
||||
$script:cmakeDefs += @("-DCMAKE_VERBOSE_MAKEFILE=on", "-DLLAMA_SERVER_VERBOSE=on", "-DCMAKE_BUILD_TYPE=RelWithDebInfo")
|
||||
$script:config = "RelWithDebInfo"
|
||||
} else {
|
||||
$script:cmakeDefs += @("-DLLAMA_SERVER_VERBOSE=off", "-DCMAKE_BUILD_TYPE=Release")
|
||||
$script:config = "Release"
|
||||
}
|
||||
if ($null -ne $env:CMAKE_SYSTEM_VERSION) {
|
||||
$script:cmakeDefs += @("-DCMAKE_SYSTEM_VERSION=${env:CMAKE_SYSTEM_VERSION}")
|
||||
}
|
||||
# Try to find the CUDA dir
|
||||
if ($env:CUDA_LIB_DIR -eq $null) {
|
||||
$d=(get-command -ea 'silentlycontinue' nvcc).path
|
||||
if ($d -ne $null) {
|
||||
$script:CUDA_LIB_DIR=($d| split-path -parent)
|
||||
$script:CUDA_INCLUDE_DIR=($script:CUDA_LIB_DIR|split-path -parent)+"\include"
|
||||
}
|
||||
} else {
|
||||
$script:CUDA_LIB_DIR=$env:CUDA_LIB_DIR
|
||||
}
|
||||
$script:DUMPBIN=(get-command -ea 'silentlycontinue' dumpbin).path
|
||||
if ($null -eq $env:CMAKE_CUDA_ARCHITECTURES) {
|
||||
$script:CMAKE_CUDA_ARCHITECTURES="50;52;61;70;75;80"
|
||||
} else {
|
||||
$script:CMAKE_CUDA_ARCHITECTURES=$env:CMAKE_CUDA_ARCHITECTURES
|
||||
}
|
||||
# Note: Windows Kits 10 signtool crashes with GCP's plugin
|
||||
if ($null -eq $env:SIGN_TOOL) {
|
||||
${script:SignTool}="C:\Program Files (x86)\Windows Kits\8.1\bin\x64\signtool.exe"
|
||||
} else {
|
||||
${script:SignTool}=${env:SIGN_TOOL}
|
||||
}
|
||||
if ("${env:KEY_CONTAINER}") {
|
||||
${script:OLLAMA_CERT}=$(resolve-path "${script:SRC_DIR}\ollama_inc.crt")
|
||||
}
|
||||
}
|
||||
|
||||
function git_module_setup {
|
||||
# TODO add flags to skip the init/patch logic to make it easier to mod llama.cpp code in-repo
|
||||
& git submodule init
|
||||
if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
|
||||
& git submodule update --force "${script:llamacppDir}"
|
||||
if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
|
||||
}
|
||||
|
||||
function apply_patches {
|
||||
# Apply temporary patches until fix is upstream
|
||||
foreach ($patch in $(Get-ChildItem "../patches/*.patch")) {
|
||||
git -c 'user.name=nobody' -c 'user.email=<>' -C "${script:llamacppDir}" am $patch.FullName
|
||||
}
|
||||
}
|
||||
|
||||
function build {
|
||||
write-host "generating config with: cmake -S ${script:llamacppDir} -B $script:buildDir $script:cmakeDefs"
|
||||
& cmake --version
|
||||
& cmake -S "${script:llamacppDir}" -B $script:buildDir $script:cmakeDefs
|
||||
if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
|
||||
if ($cmakeDefs -contains "-G") {
|
||||
$extra=@("-j8")
|
||||
} else {
|
||||
$extra= @("--", "/maxCpuCount:8")
|
||||
}
|
||||
write-host "building with: cmake --build $script:buildDir --config $script:config $($script:cmakeTargets | ForEach-Object { `"--target`", $_ }) $extra"
|
||||
& cmake --build $script:buildDir --config $script:config ($script:cmakeTargets | ForEach-Object { "--target", $_ }) $extra
|
||||
if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
|
||||
# Rearrange output to be consistent between different generators
|
||||
if ($null -ne ${script:config} -And (test-path -path "${script:buildDir}/bin/${script:config}" ) ) {
|
||||
mv -force "${script:buildDir}/bin/${script:config}/*" "${script:buildDir}/bin/"
|
||||
remove-item "${script:buildDir}/bin/${script:config}"
|
||||
}
|
||||
}
|
||||
|
||||
function sign {
|
||||
if ("${env:KEY_CONTAINER}") {
|
||||
write-host "Signing ${script:buildDir}/bin/*.exe ${script:buildDir}/bin/*.dll"
|
||||
foreach ($file in @(get-childitem "${script:buildDir}/bin/*.exe") + @(get-childitem "${script:buildDir}/bin/*.dll")){
|
||||
& "${script:SignTool}" sign /v /fd sha256 /t http://timestamp.digicert.com /f "${script:OLLAMA_CERT}" `
|
||||
/csp "Google Cloud KMS Provider" /kc "${env:KEY_CONTAINER}" $file
|
||||
if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
function install {
|
||||
write-host "Installing binaries to dist dir ${script:distDir}"
|
||||
mkdir ${script:distDir} -ErrorAction SilentlyContinue
|
||||
$binaries = dir "${script:buildDir}/bin/*.exe"
|
||||
foreach ($file in $binaries) {
|
||||
copy-item -Path $file -Destination ${script:distDir} -Force
|
||||
}
|
||||
|
||||
write-host "Installing dlls to dist dir ${script:distDir}"
|
||||
$dlls = dir "${script:buildDir}/bin/*.dll"
|
||||
foreach ($file in $dlls) {
|
||||
copy-item -Path $file -Destination ${script:distDir} -Force
|
||||
}
|
||||
}
|
||||
|
||||
function cleanup {
|
||||
$patches = Get-ChildItem "../patches/*.diff"
|
||||
foreach ($patch in $patches) {
|
||||
# Extract file paths from the patch file
|
||||
$filePaths = Get-Content $patch.FullName | Where-Object { $_ -match '^\+\+\+ ' } | ForEach-Object {
|
||||
$parts = $_ -split ' '
|
||||
($parts[1] -split '/', 2)[1]
|
||||
}
|
||||
|
||||
# Checkout each file
|
||||
foreach ($file in $filePaths) {
|
||||
git -C "${script:llamacppDir}" checkout $file
|
||||
}
|
||||
git -C "${script:llamacppDir}" checkout CMakeLists.txt
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
# -DGGML_AVX -- 2011 Intel Sandy Bridge & AMD Bulldozer
|
||||
# -DGGML_AVX2 -- 2013 Intel Haswell & 2015 AMD Excavator / 2017 AMD Zen
|
||||
# -DGGML_FMA (FMA3) -- 2013 Intel Haswell & 2012 AMD Piledriver
|
||||
|
||||
|
||||
function build_cpu_x64 {
|
||||
if ((-not "${env:OLLAMA_SKIP_CPU_GENERATE}" ) -and ((-not "${env:OLLAMA_CPU_TARGET}") -or ("${env:OLLAMA_CPU_TARGET}" -eq "cpu"))) {
|
||||
init_vars
|
||||
$script:cmakeDefs = $script:commonCpuDefs + @("-A", "x64", "-DGGML_AVX=off", "-DGGML_AVX2=off", "-DGGML_AVX512=off", "-DGGML_FMA=off", "-DGGML_F16C=off") + $script:cmakeDefs
|
||||
$script:buildDir="../build/windows/${script:ARCH}/cpu"
|
||||
$script:distDir="$script:DIST_BASE\cpu"
|
||||
write-host "Building LCD CPU"
|
||||
build
|
||||
sign
|
||||
install
|
||||
} else {
|
||||
write-host "Skipping CPU generation step as requested"
|
||||
}
|
||||
}
|
||||
|
||||
function build_cpu_arm64 {
|
||||
if ((-not "${env:OLLAMA_SKIP_CPU_GENERATE}" ) -and ((-not "${env:OLLAMA_CPU_TARGET}") -or ("${env:OLLAMA_CPU_TARGET}" -eq "cpu"))) {
|
||||
init_vars
|
||||
write-host "Checking for clang..."
|
||||
get-command clang
|
||||
$env:CFLAGS="-march=armv8.7-a -fvectorize -ffp-model=fast -fno-finite-math-only"
|
||||
$env:CXXFLAGS="$env:CFLAGS"
|
||||
$env:LDFLAGS="-static-libstdc++"
|
||||
$script:cmakeDefs = $script:commonCpuDefs + @(
|
||||
"-DCMAKE_VERBOSE_MAKEFILE=on",
|
||||
"-DCMAKE_C_COMPILER=clang.exe",
|
||||
"-DCMAKE_CXX_COMPILER=clang++.exe",
|
||||
"-DMSVC_RUNTIME_LIBRARY=MultiThreaded"
|
||||
) + $script:cmakeDefs
|
||||
$script:buildDir="../build/windows/${script:ARCH}/cpu"
|
||||
$script:distDir="$script:DIST_BASE\cpu"
|
||||
write-host "Building LCD CPU"
|
||||
build
|
||||
sign
|
||||
install
|
||||
} else {
|
||||
write-host "Skipping CPU generation step as requested"
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
function build_cpu_avx() {
|
||||
if ((-not "${env:OLLAMA_SKIP_CPU_GENERATE}" ) -and ((-not "${env:OLLAMA_CPU_TARGET}") -or ("${env:OLLAMA_CPU_TARGET}" -eq "cpu_avx"))) {
|
||||
init_vars
|
||||
$script:cmakeDefs = $script:commonCpuDefs + @("-A", "x64", "-DGGML_AVX=on", "-DGGML_AVX2=off", "-DGGML_AVX512=off", "-DGGML_FMA=off", "-DGGML_F16C=off") + $script:cmakeDefs
|
||||
$script:buildDir="../build/windows/${script:ARCH}/cpu_avx"
|
||||
$script:distDir="$script:DIST_BASE\cpu_avx"
|
||||
write-host "Building AVX CPU"
|
||||
build
|
||||
sign
|
||||
install
|
||||
} else {
|
||||
write-host "Skipping CPU AVX generation step as requested"
|
||||
}
|
||||
}
|
||||
|
||||
function build_cpu_avx2() {
|
||||
if ((-not "${env:OLLAMA_SKIP_CPU_GENERATE}" ) -and ((-not "${env:OLLAMA_CPU_TARGET}") -or ("${env:OLLAMA_CPU_TARGET}" -eq "cpu_avx2"))) {
|
||||
init_vars
|
||||
$script:cmakeDefs = $script:commonCpuDefs + @("-A", "x64", "-DGGML_AVX=on", "-DGGML_AVX2=on", "-DGGML_AVX512=off", "-DGGML_FMA=on", "-DGGML_F16C=on") + $script:cmakeDefs
|
||||
$script:buildDir="../build/windows/${script:ARCH}/cpu_avx2"
|
||||
$script:distDir="$script:DIST_BASE\cpu_avx2"
|
||||
write-host "Building AVX2 CPU"
|
||||
build
|
||||
sign
|
||||
install
|
||||
} else {
|
||||
write-host "Skipping CPU AVX2 generation step as requested"
|
||||
}
|
||||
}
|
||||
|
||||
function build_cuda() {
|
||||
if ((-not "${env:OLLAMA_SKIP_CUDA_GENERATE}") -and ("${script:CUDA_LIB_DIR}")) {
|
||||
# Then build cuda as a dynamically loaded library
|
||||
$nvcc = "$script:CUDA_LIB_DIR\nvcc.exe"
|
||||
$script:CUDA_VERSION=((get-item ($nvcc | split-path | split-path)).Basename -Split "\.")[0]
|
||||
if ($null -ne $script:CUDA_VERSION) {
|
||||
$script:CUDA_VARIANT="_"+$script:CUDA_VERSION
|
||||
}
|
||||
init_vars
|
||||
$script:buildDir="../build/windows/${script:ARCH}/cuda$script:CUDA_VARIANT"
|
||||
$script:distDir="$script:DIST_BASE\cuda$script:CUDA_VARIANT"
|
||||
$script:cmakeDefs += @(
|
||||
"-A", "x64",
|
||||
"-DGGML_CUDA=ON",
|
||||
"-DGGML_AVX=on",
|
||||
"-DGGML_AVX2=off",
|
||||
"-DCMAKE_CUDA_FLAGS=-t6",
|
||||
"-DCMAKE_CUDA_ARCHITECTURES=${script:CMAKE_CUDA_ARCHITECTURES}",
|
||||
"-DCMAKE_CUDA_COMPILER_TOOLKIT_ROOT=$env:CUDA_PATH"
|
||||
)
|
||||
if ($null -ne $env:OLLAMA_CUSTOM_CUDA_DEFS) {
|
||||
write-host "OLLAMA_CUSTOM_CUDA_DEFS=`"${env:OLLAMA_CUSTOM_CUDA_DEFS}`""
|
||||
$script:cmakeDefs +=@("${env:OLLAMA_CUSTOM_CUDA_DEFS}")
|
||||
write-host "building custom CUDA GPU"
|
||||
}
|
||||
build
|
||||
sign
|
||||
install
|
||||
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\" -ea 0 > $null
|
||||
write-host "copying CUDA dependencies to ${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${script:CUDA_LIB_DIR}\cudart64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${script:CUDA_LIB_DIR}\cublas64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${script:CUDA_LIB_DIR}\cublasLt64_*.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
} else {
|
||||
write-host "Skipping CUDA generation step"
|
||||
}
|
||||
}
|
||||
|
||||
function build_oneapi() {
|
||||
if ((-not "${env:OLLAMA_SKIP_ONEAPI_GENERATE}") -and ("${env:ONEAPI_ROOT}")) {
|
||||
# Get oneAPI version
|
||||
$script:ONEAPI_VERSION = icpx --version
|
||||
$script:ONEAPI_VERSION = [regex]::Match($script:ONEAPI_VERSION, '(?<=oneAPI DPC\+\+/C\+\+ Compiler )(?<version>\d+\.\d+\.\d+)').Value
|
||||
if ($null -ne $script:ONEAPI_VERSION) {
|
||||
$script:ONEAPI_VARIANT = "_v" + $script:ONEAPI_VERSION
|
||||
}
|
||||
init_vars
|
||||
$script:buildDir = "../build/windows/${script:ARCH}/oneapi$script:ONEAPI_VARIANT"
|
||||
$script:distDir ="$script:DIST_BASE\oneapi$script:ONEAPI_VARIANT"
|
||||
$script:cmakeDefs += @(
|
||||
"-G", "MinGW Makefiles",
|
||||
"-DGGML_SYCL=ON",
|
||||
"-DCMAKE_C_COMPILER=icx",
|
||||
"-DCMAKE_CXX_COMPILER=icx",
|
||||
"-DCMAKE_BUILD_TYPE=Release"
|
||||
)
|
||||
|
||||
Write-Host "Building oneAPI"
|
||||
build
|
||||
# Ninja doesn't prefix with config name
|
||||
if ($null -ne $script:DUMPBIN) {
|
||||
& "$script:DUMPBIN" /dependents "${script:buildDir}/bin/ollama_llama_server.exe" | Select-String ".dll"
|
||||
}
|
||||
sign
|
||||
install
|
||||
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\" -ea 0 > $null
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\libirngmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\libmmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_level_zero.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_unified_runtime.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\pi_win_proxy_loader.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\svml_dispmd.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\compiler\latest\bin\sycl7.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_core.2.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_sycl_blas.4.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:ONEAPI_ROOT}\mkl\latest\bin\mkl_tbb_thread.2.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
} else {
|
||||
Write-Host "Skipping oneAPI generation step"
|
||||
}
|
||||
}
|
||||
|
||||
function build_rocm() {
|
||||
if ((-not "${env:OLLAMA_SKIP_ROCM_GENERATE}") -and ("${env:HIP_PATH}")) {
|
||||
$script:ROCM_VERSION=(get-item $env:HIP_PATH).Basename
|
||||
if ($null -ne $script:ROCM_VERSION) {
|
||||
$script:ROCM_VARIANT="_v"+$script:ROCM_VERSION
|
||||
}
|
||||
|
||||
init_vars
|
||||
$script:buildDir="../build/windows/${script:ARCH}/rocm$script:ROCM_VARIANT"
|
||||
$script:distDir="$script:DIST_BASE\rocm$script:ROCM_VARIANT"
|
||||
$script:cmakeDefs += @(
|
||||
"-G", "Ninja",
|
||||
"-DCMAKE_C_COMPILER=clang.exe",
|
||||
"-DCMAKE_CXX_COMPILER=clang++.exe",
|
||||
"-DGGML_HIPBLAS=on",
|
||||
"-DHIP_PLATFORM=amd",
|
||||
"-DGGML_AVX=on",
|
||||
"-DGGML_AVX2=off",
|
||||
"-DCMAKE_POSITION_INDEPENDENT_CODE=on",
|
||||
"-DAMDGPU_TARGETS=$(amdGPUs)",
|
||||
"-DGPU_TARGETS=$(amdGPUs)"
|
||||
)
|
||||
|
||||
# Make sure the ROCm binary dir is first in the path
|
||||
$env:PATH="$env:HIP_PATH\bin;$env:PATH"
|
||||
|
||||
# We have to clobber the LIB var from the developer shell for clang to work properly
|
||||
$env:LIB=""
|
||||
if ($null -ne $env:OLLAMA_CUSTOM_ROCM_DEFS) {
|
||||
write-host "OLLAMA_CUSTOM_ROCM_DEFS=`"${env:OLLAMA_CUSTOM_ROCM_DEFS}`""
|
||||
$script:cmakeDefs += @("${env:OLLAMA_CUSTOM_ROCM_DEFS}")
|
||||
write-host "building custom ROCM GPU"
|
||||
}
|
||||
write-host "Building ROCm"
|
||||
build
|
||||
# Ninja doesn't prefix with config name
|
||||
${script:config}=""
|
||||
if ($null -ne $script:DUMPBIN) {
|
||||
& "$script:DUMPBIN" /dependents "${script:buildDir}/bin/ollama_llama_server.exe" | select-string ".dll"
|
||||
}
|
||||
sign
|
||||
install
|
||||
|
||||
md "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\rocblas\library\" -ea 0 > $null
|
||||
cp "${env:HIP_PATH}\bin\hipblas.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
cp "${env:HIP_PATH}\bin\rocblas.dll" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\"
|
||||
# amdhip64.dll dependency comes from the driver and must be installed on the host to use AMD GPUs
|
||||
cp "${env:HIP_PATH}\bin\rocblas\library\*" "${script:SRC_DIR}\dist\windows-${script:ARCH}\lib\ollama\rocblas\library\"
|
||||
} else {
|
||||
write-host "Skipping ROCm generation step"
|
||||
}
|
||||
}
|
||||
|
||||
init_vars
|
||||
if ($($args.count) -eq 0) {
|
||||
git_module_setup
|
||||
apply_patches
|
||||
if ($script:ARCH -eq "arm64") {
|
||||
build_cpu_arm64
|
||||
} else { # amd64
|
||||
build_cpu_x64
|
||||
build_cpu_avx
|
||||
build_cpu_avx2
|
||||
build_cuda
|
||||
build_oneapi
|
||||
build_rocm
|
||||
}
|
||||
|
||||
cleanup
|
||||
write-host "`ngo generate completed. LLM runners: $(get-childitem -path $script:DIST_BASE)"
|
||||
} else {
|
||||
for ( $i = 0; $i -lt $args.count; $i++ ) {
|
||||
write-host "performing $($args[$i])"
|
||||
& $($args[$i])
|
||||
}
|
||||
}
|
||||
@@ -1,3 +0,0 @@
|
||||
package generate
|
||||
|
||||
//go:generate bash ./gen_darwin.sh
|
||||
@@ -1,3 +0,0 @@
|
||||
package generate
|
||||
|
||||
//go:generate bash ./gen_linux.sh
|
||||
@@ -1,3 +0,0 @@
|
||||
package generate
|
||||
|
||||
//go:generate powershell -ExecutionPolicy Bypass -File ./gen_windows.ps1
|
||||
Reference in New Issue
Block a user