mirror of
https://github.com/dogkeeper886/ollama37.git
synced 2025-12-12 00:37:04 +00:00
Code shuffle to clean up the llm dir
This commit is contained in:
Daniel Hiltgen
53
llm/generate/gen_common.sh
Normal file
53
llm/generate/gen_common.sh
Normal file
@@ -0,0 +1,53 @@
|
||||
# common logic accross linux and darwin
|
||||
|
||||
init_vars() {
|
||||
LLAMACPP_DIR=../llama.cpp
|
||||
PATCHES="0001-Expose-callable-API-for-server.patch"
|
||||
CMAKE_DEFS=""
|
||||
CMAKE_TARGETS="--target ggml --target ggml_static --target llama --target build_info --target common --target ext_server --target llava_static"
|
||||
if echo "${CGO_CFLAGS}" | grep -- '-g' >/dev/null; then
|
||||
CMAKE_DEFS="-DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_VERBOSE_MAKEFILE=on -DLLAMA_GPROF=on -DLLAMA_SERVER_VERBOSE=on"
|
||||
else
|
||||
# TODO - add additional optimization flags...
|
||||
CMAKE_DEFS="-DCMAKE_BUILD_TYPE=Release -DLLAMA_SERVER_VERBOSE=off"
|
||||
fi
|
||||
}
|
||||
|
||||
git_module_setup() {
|
||||
if [ -n "${OLLAMA_SKIP_PATCHING}" ]; then
|
||||
echo "Skipping submodule initialization"
|
||||
return
|
||||
fi
|
||||
git submodule init
|
||||
git submodule update --force ${LLAMACPP_DIR}
|
||||
|
||||
}
|
||||
|
||||
apply_patches() {
|
||||
# Wire up our CMakefile
|
||||
if ! grep ollama ${LLAMACPP_DIR}/examples/server/CMakeLists.txt; then
|
||||
echo 'include (../../../ext_server/CMakeLists.txt) # ollama' >>${LLAMACPP_DIR}/examples/server/CMakeLists.txt
|
||||
fi
|
||||
# Avoid duplicate main symbols when we link into the cgo binary
|
||||
sed -e 's/int main(/int __main(/g' <${LLAMACPP_DIR}/examples/server/server.cpp >${LLAMACPP_DIR}/examples/server/server.cpp.tmp &&
|
||||
mv ${LLAMACPP_DIR}/examples/server/server.cpp.tmp ${LLAMACPP_DIR}/examples/server/server.cpp
|
||||
}
|
||||
|
||||
build() {
|
||||
cmake -S ${LLAMACPP_DIR} -B ${BUILD_DIR} ${CMAKE_DEFS}
|
||||
cmake --build ${BUILD_DIR} ${CMAKE_TARGETS} -j8
|
||||
}
|
||||
|
||||
install() {
|
||||
rm -rf ${BUILD_DIR}/lib
|
||||
mkdir -p ${BUILD_DIR}/lib
|
||||
cp ${BUILD_DIR}/examples/server/libext_server.a ${BUILD_DIR}/lib
|
||||
cp ${BUILD_DIR}/common/libcommon.a ${BUILD_DIR}/lib
|
||||
cp ${BUILD_DIR}/libllama.a ${BUILD_DIR}/lib
|
||||
cp ${BUILD_DIR}/libggml_static.a ${BUILD_DIR}/lib
|
||||
}
|
||||
|
||||
# Keep the local tree clean after we're done with the build
|
||||
cleanup() {
|
||||
(cd ${LLAMACPP_DIR}/examples/server/ && git checkout CMakeLists.txt server.cpp)
|
||||
}
|
||||
32
llm/generate/gen_darwin.sh
Executable file
32
llm/generate/gen_darwin.sh
Executable file
@@ -0,0 +1,32 @@
|
||||
#!/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
|
||||
echo "Starting darwin generate script"
|
||||
source $(dirname $0)/gen_common.sh
|
||||
init_vars
|
||||
CMAKE_DEFS="-DCMAKE_OSX_DEPLOYMENT_TARGET=11.0 -DLLAMA_METAL=on -DLLAMA_ACCELERATE=on ${CMAKE_DEFS}"
|
||||
BUILD_DIR="${LLAMACPP_DIR}/build/darwin/metal"
|
||||
case "${GOARCH}" in
|
||||
"amd64")
|
||||
CMAKE_DEFS="-DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
|
||||
;;
|
||||
"arm64")
|
||||
CMAKE_DEFS="-DCMAKE_SYSTEM_PROCESSOR=arm64 -DCMAKE_OSX_ARCHITECTURES=arm64 ${CMAKE_DEFS}"
|
||||
;;
|
||||
*)
|
||||
echo "GOARCH must be set"
|
||||
echo "this script is meant to be run from within go generate"
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
|
||||
git_module_setup
|
||||
apply_patches
|
||||
build
|
||||
install
|
||||
cleanup
|
||||
116
llm/generate/gen_linux.sh
Executable file
116
llm/generate/gen_linux.sh
Executable file
@@ -0,0 +1,116 @@
|
||||
#!/bin/bash
|
||||
# This script is intended to run inside the go generate
|
||||
# working directory must be llm/generate/
|
||||
|
||||
# First we build our default built-in library which will be linked into the CGO
|
||||
# binary as a normal dependency. This default build is CPU based.
|
||||
#
|
||||
# Then we build a CUDA dynamic library (although statically linked with the CUDA
|
||||
# library dependencies for maximum portability)
|
||||
#
|
||||
# Then if we detect ROCm, we build a dynamically loaded ROCm lib. ROCm is particularly
|
||||
# important to be a dynamic lib even if it's the only GPU library detected because
|
||||
# we can't redistribute the objectfiles but must rely on dynamic libraries at
|
||||
# runtime, which could lead the server not to start if not present.
|
||||
|
||||
set -ex
|
||||
set -o pipefail
|
||||
|
||||
# See https://llvm.org/docs/AMDGPUUsage.html#processors for reference
|
||||
amdGPUs() {
|
||||
GPU_LIST=(
|
||||
"gfx803"
|
||||
"gfx900"
|
||||
"gfx906:xnack-"
|
||||
"gfx908:xnack-"
|
||||
"gfx90a:xnack+"
|
||||
"gfx90a:xnack-"
|
||||
"gfx1010"
|
||||
"gfx1012"
|
||||
"gfx1030"
|
||||
"gfx1100"
|
||||
"gfx1101"
|
||||
"gfx1102"
|
||||
)
|
||||
(
|
||||
IFS=$';'
|
||||
echo "'${GPU_LIST[*]}'"
|
||||
)
|
||||
}
|
||||
|
||||
echo "Starting linux generate script"
|
||||
if [ -z "${CUDACXX}" -a -x /usr/local/cuda/bin/nvcc ]; then
|
||||
export CUDACXX=/usr/local/cuda/bin/nvcc
|
||||
fi
|
||||
COMMON_CMAKE_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off"
|
||||
source $(dirname $0)/gen_common.sh
|
||||
init_vars
|
||||
git_module_setup
|
||||
apply_patches
|
||||
|
||||
#
|
||||
# CPU first for the default library
|
||||
#
|
||||
CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS}"
|
||||
BUILD_DIR="${LLAMACPP_DIR}/build/linux/cpu"
|
||||
|
||||
build
|
||||
install
|
||||
|
||||
# Placeholder to keep go embed happy until we start building dynamic CPU lib variants
|
||||
touch ${BUILD_DIR}/lib/dummy.so
|
||||
|
||||
if [ -d /usr/local/cuda/lib64/ ]; then
|
||||
echo "CUDA libraries detected - building dynamic CUDA library"
|
||||
init_vars
|
||||
CMAKE_DEFS="-DLLAMA_CUBLAS=on ${COMMON_CMAKE_DEFS} ${CMAKE_DEFS}"
|
||||
BUILD_DIR="${LLAMACPP_DIR}/build/linux/cuda"
|
||||
CUDA_LIB_DIR=/usr/local/cuda/lib64
|
||||
build
|
||||
install
|
||||
gcc -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.so \
|
||||
-Wl,--whole-archive \
|
||||
${BUILD_DIR}/lib/libext_server.a \
|
||||
${BUILD_DIR}/lib/libcommon.a \
|
||||
${BUILD_DIR}/lib/libllama.a \
|
||||
-Wl,--no-whole-archive \
|
||||
${CUDA_LIB_DIR}/libcudart_static.a \
|
||||
${CUDA_LIB_DIR}/libcublas_static.a \
|
||||
${CUDA_LIB_DIR}/libcublasLt_static.a \
|
||||
${CUDA_LIB_DIR}/libcudadevrt.a \
|
||||
${CUDA_LIB_DIR}/libculibos.a \
|
||||
-lrt -lpthread -ldl -lstdc++ -lm
|
||||
fi
|
||||
|
||||
if [ -z "${ROCM_PATH}" ]; then
|
||||
# Try the default location in case it exists
|
||||
ROCM_PATH=/opt/rocm
|
||||
fi
|
||||
|
||||
if [ -z "${CLBlast_DIR}" ]; then
|
||||
# Try the default location in case it exists
|
||||
if [ -d /usr/lib/cmake/CLBlast ]; then
|
||||
export CLBlast_DIR=/usr/lib/cmake/CLBlast
|
||||
fi
|
||||
fi
|
||||
|
||||
if [ -d "${ROCM_PATH}" ]; then
|
||||
echo "ROCm libraries detected - building dynamic ROCm library"
|
||||
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=$(amdGPUs)"
|
||||
BUILD_DIR="${LLAMACPP_DIR}/build/linux/rocm"
|
||||
build
|
||||
install
|
||||
gcc -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.so \
|
||||
-Wl,--whole-archive \
|
||||
${BUILD_DIR}/lib/libext_server.a \
|
||||
${BUILD_DIR}/lib/libcommon.a \
|
||||
${BUILD_DIR}/lib/libllama.a \
|
||||
-Wl,--no-whole-archive \
|
||||
-lrt -lpthread -ldl -lstdc++ -lm \
|
||||
-L/opt/rocm/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ \
|
||||
-Wl,-rpath,/opt/rocm/lib,-rpath,/opt/amdgpu/lib/x86_64-linux-gnu/ \
|
||||
-lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu
|
||||
fi
|
||||
|
||||
cleanup
|
||||
88
llm/generate/gen_windows.ps1
Normal file
88
llm/generate/gen_windows.ps1
Normal file
@@ -0,0 +1,88 @@
|
||||
#!powershell
|
||||
|
||||
$ErrorActionPreference = "Stop"
|
||||
|
||||
function init_vars {
|
||||
$script:llamacppDir = "../llama.cpp"
|
||||
$script:patches = @("0001-Expose-callable-API-for-server.patch")
|
||||
$script:cmakeDefs = @("-DBUILD_SHARED_LIBS=on", "-DLLAMA_NATIVE=off", "-DLLAMA_F16C=off", "-DLLAMA_FMA=off", "-DLLAMA_AVX512=off", "-DLLAMA_AVX2=off", "-DLLAMA_AVX=on", "-A","x64")
|
||||
$script:cmakeTargets = @("ggml", "ggml_static", "llama", "build_info", "common", "ext_server_shared", "llava_static")
|
||||
if ($env:CGO_CFLAGS -contains "-g") {
|
||||
$script:cmakeDefs += @("-DCMAKE_VERBOSE_MAKEFILE=on", "-DLLAMA_SERVER_VERBOSE=on")
|
||||
$script:config = "RelWithDebInfo"
|
||||
} else {
|
||||
$script:cmakeDefs += @("-DLLAMA_SERVER_VERBOSE=off")
|
||||
$script:config = "Release"
|
||||
}
|
||||
}
|
||||
|
||||
function git_module_setup {
|
||||
# TODO add flags to skip the init/patch logic to make it easier to mod llama.cpp code in-repo
|
||||
& git submodule init
|
||||
if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
|
||||
& git submodule update --force "${script:llamacppDir}"
|
||||
if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
|
||||
}
|
||||
|
||||
function apply_patches {
|
||||
# Wire up our CMakefile
|
||||
if (!(Select-String -Path "${script:llamacppDir}/examples/server/CMakeLists.txt" -Pattern 'ollama')) {
|
||||
Add-Content -Path "${script:llamacppDir}/examples/server/CMakeLists.txt" -Value 'include (../../../ext_server/CMakeLists.txt) # ollama'
|
||||
}
|
||||
# Avoid duplicate main symbols when we link into the cgo binary
|
||||
$content = Get-Content -Path "${script:llamacppDir}/examples/server/server.cpp"
|
||||
$content = $content -replace 'int main\(', 'int __main('
|
||||
Set-Content -Path "${script:llamacppDir}/examples/server/server.cpp" -Value $content
|
||||
}
|
||||
|
||||
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)}
|
||||
write-host "building with: cmake --build $script:buildDir --config $script:config ($script:cmakeTargets | ForEach-Object { "--target", $_ })"
|
||||
& cmake --build $script:buildDir --config $script:config ($script:cmakeTargets | ForEach-Object { "--target", $_ })
|
||||
if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
|
||||
}
|
||||
|
||||
function install {
|
||||
rm -ea 0 -recurse -force -path "${script:buildDir}/lib"
|
||||
md "${script:buildDir}/lib" -ea 0 > $null
|
||||
cp "${script:buildDir}/bin/${script:config}/ext_server_shared.dll" "${script:buildDir}/lib"
|
||||
cp "${script:buildDir}/bin/${script:config}/llama.dll" "${script:buildDir}/lib"
|
||||
|
||||
# Display the dll dependencies in the build log
|
||||
dumpbin /dependents "${script:buildDir}/bin/${script:config}/ext_server_shared.dll" | select-string ".dll"
|
||||
}
|
||||
|
||||
function cleanup {
|
||||
Set-Location "${script:llamacppDir}/examples/server"
|
||||
git checkout CMakeLists.txt server.cpp
|
||||
}
|
||||
|
||||
init_vars
|
||||
git_module_setup
|
||||
apply_patches
|
||||
|
||||
# first build CPU based
|
||||
$script:buildDir="${script:llamacppDir}/build/windows/cpu"
|
||||
|
||||
build
|
||||
install
|
||||
|
||||
# Then build cuda as a dynamically loaded library
|
||||
init_vars
|
||||
$script:buildDir="${script:llamacppDir}/build/windows/cuda"
|
||||
$script:cmakeDefs += @("-DLLAMA_CUBLAS=ON")
|
||||
build
|
||||
install
|
||||
|
||||
# TODO - actually implement ROCm support on windows
|
||||
$script:buildDir="${script:llamacppDir}/build/windows/rocm"
|
||||
|
||||
rm -ea 0 -recurse -force -path "${script:buildDir}/lib"
|
||||
md "${script:buildDir}/lib" -ea 0 > $null
|
||||
echo $null >> "${script:buildDir}/lib/.generated"
|
||||
|
||||
cleanup
|
||||
write-host "`ngo generate completed"
|
||||
3
llm/generate/generate_darwin.go
Normal file
3
llm/generate/generate_darwin.go
Normal file
@@ -0,0 +1,3 @@
|
||||
package generate
|
||||
|
||||
//go:generate sh ./gen_darwin.sh
|
||||
3
llm/generate/generate_linux.go
Normal file
3
llm/generate/generate_linux.go
Normal file
@@ -0,0 +1,3 @@
|
||||
package generate
|
||||
|
||||
//go:generate bash ./gen_linux.sh
|
||||
3
llm/generate/generate_windows.go
Normal file
3
llm/generate/generate_windows.go
Normal file
@@ -0,0 +1,3 @@
|
||||
package generate
|
||||
|
||||
//go:generate powershell -ExecutionPolicy Bypass -File ./gen_windows.ps1
|
||||
Reference in New Issue
Block a user