ci: reduce cuda binary size #79

Workflow file for this run

	name: ci

	permissions:
	contents: read
	pull-requests: read
	actions: read

	env:
	VERSION: "${{ github.ref_name }}"

	on:
	workflow_dispatch: { }
	push:
	tags:
	- "v..*"
	branches:
	- main
	paths-ignore:
	- "docs/**"
	- "**.md"
	- "**.mdx"
	- "**.png"
	- "**.jpg"
	pull_request:
	branches:
	- main
	paths-ignore:
	- "docs/**"
	- "**.md"
	- "**.mdx"
	- "**.png"
	- "**.jpg"

	concurrency:
	group: ${{ github.workflow }}-${{ github.head_ref && github.ref \|\| github.run_id }}
	cancel-in-progress: true

	jobs:

	# cache-able building with ccache.
	darwin-metal:
	strategy:
	fail-fast: false
	matrix:
	arch: [ amd64, arm64 ]
	version: [ '3.0' ]
	# see https://github.com/actions/runner-images?tab=readme-ov-file#available-images,
	# https://support.apple.com/en-us/102894.
	runs-on: ${{ matrix.arch == 'amd64' && 'macos-13' \|\| 'macos-14' }}
	steps:
	- name: Checkout
	uses: actions/checkout@v4
	with:
	fetch-depth: 0
	submodules: 'recursive'
	- name: Deps
	run: \|
	brew update && brew install ccache
	- name: Setup XCode
	uses: maxim-lobanov/setup-xcode@v1
	with:
	xcode-version: '15.2'
	- name: Setup Cache
	timeout-minutes: 5
	uses: actions/cache@v3
	with:
	key: cache-darwin-metal-${{ matrix.arch }}-${{ matrix.version }}
	path: \|
	${{ github.workspace }}/.cache
	- name: Build
	# disable OpenMP,
	# see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
	# https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
	env:
	CCACHE_DIR: "${{ github.workspace }}/.cache/ccache"
	run: \|
	echo "===== BUILD ====="
	mkdir -p ${{ github.workspace }}/.cache
	cmake -S ${{ github.workspace }} -B ${{ github.workspace }}/build -DCMAKE_BUILD_TYPE=Release \
	-DGGML_ACCELERATE=on -DGGML_METAL=on -DGGML_METAL_EMBED_LIBRARY=on \
	${{ matrix.arch == 'amd64' && '-DGGML_NATIVE=off' \|\| '-DGGML_NATIVE=on' }} \
	-DGGML_OPENMP=off
	cmake --build ${{ github.workspace }}/build --target llama-box --config Release -- -j $(nproc)

	echo "===== RESULT ====="
	if [ -f ${{ github.workspace }}/build/bin/llama-box ]; then
	otool -L ${{ github.workspace }}/build/bin/llama-box
	else
	exit 1
	fi

	echo "===== PACKAGE ====="
	mkdir -p ${{ github.workspace }}/out
	zip -j ${{ github.workspace }}/out/llama-box-darwin-${{ matrix.arch }}-metal.zip ${{ github.workspace }}/build/bin/llama-box
	- name: Upload Artifact
	uses: actions/upload-artifact@v4
	with:
	path: ${{ github.workspace }}/out/*.zip
	name: llama-box-darwin-${{ matrix.arch }}-metal-${{ matrix.version }}

	# cache-able building with ccache.
	linux-hip:
	strategy:
	fail-fast: false
	matrix:
	arch: [ amd64 ]
	# see https://hub.docker.com/r/rocm/dev-ubuntu-22.04/tags.
	# 6.1 ==> 6.1.2
	# 5.7 ==> 5.7.1
	version: [ '6.1', '5.7' ]
	runs-on: ubuntu-22.04
	steps:
	- name: Maximize Space
	# see https://github.com/easimon/maximize-build-space/blob/master/action.yml.
	run: \|
	sudo rm -rf /usr/share/dotnet
	sudo rm -rf /usr/local/lib/android
	sudo rm -rf /opt/ghc
	sudo rm -rf /opt/hostedtoolcache/CodeQL
	sudo docker image prune --all --force
	- name: Clone
	uses: actions/checkout@v4
	with:
	fetch-depth: 0
	submodules: 'recursive'
	- name: Setup Cache
	timeout-minutes: 5
	uses: actions/cache@v3
	with:
	key: cache-linux-hip-${{ matrix.arch }}-${{ matrix.version }}
	path: \|
	${{ github.workspace }}/.cache
	- name: Setup QEMU
	if: ${{ matrix.arch == 'arm64' }}
	uses: docker/setup-qemu-action@v3
	with:
	image: tonistiigi/binfmt:qemu-v7.0.0
	platforms: "arm64"
	- name: Build
	# disable OpenMP,
	# see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
	# https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
	# build fat binary,
	# see https://github.com/ggerganov/llama.cpp/pull/1087#issuecomment-1682807878,
	# https://llvm.org/docs/AMDGPUUsage.html.
	# official gpu support list,
	# see https://rocm.docs.amd.com/projects/install-on-linux/en/docs-6.1.2/reference/system-requirements.html,
	# https://rocm.docs.amd.com/en/docs-5.7.1/release/gpu_os_support.html.
	env:
	CCACHE_DIR: "${{ github.workspace }}/.cache/ccache"
	AMDGPU_TARGETS: "gfx803;gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1010;gfx1030;gfx1100;gfx1101;gfx1102"
	run: \|
	echo "===== SCRIPT ====="
	cat <<EOF > /tmp/entrypoint.sh
	#!/bin/bash
	apt-get update && apt-get install -y build-essential git cmake ccache
	git config --system --add safe.directory '*'
	mkdir -p ${{ github.workspace }}/.cache
	cmake -S ${{ github.workspace }} -B ${{ github.workspace }}/build -DCMAKE_BUILD_TYPE=Release \
	-DGGML_HIPBLAS=on -DAMDGPU_TARGETS="${AMDGPU_TARGETS}" \
	${{ matrix.arch == 'amd64' && '-DGGML_NATIVE=off' \|\| '-DGGML_NATIVE=on' }} \
	-DGGML_OPENMP=off
	cmake --build ${{ github.workspace }}/build --target llama-box --config Release -- -j $(nproc)
	EOF
	chmod +x /tmp/entrypoint.sh
	cat /tmp/entrypoint.sh

	echo "===== BUILD ====="
	docker run \
	--rm \
	--privileged \
	--platform linux/${{ matrix.arch }} \
	--volume ${{ github.workspace }}:${{ github.workspace }} \
	--workdir ${{ github.workspace }} \
	--env CC=/opt/rocm/llvm/bin/clang \
	--env CXX=/opt/rocm/llvm/bin/clang++ \
	--env CCACHE_DIR \
	--env AMDGPU_TARGETS \
	--volume /tmp/entrypoint.sh:/entrypoint.sh \
	--entrypoint /entrypoint.sh \
	rocm/dev-ubuntu-22.04:${{ matrix.version == '6.1' && '6.1.2' \|\| '5.7.1' }}-complete

	echo "===== RESULT ====="
	if [ -f ${{ github.workspace }}/build/bin/llama-box ]; then
	ldd ${{ github.workspace }}/build/bin/llama-box
	else
	exit 1
	fi

	echo "===== PACKAGE ====="
	mkdir -p ${{ github.workspace }}/out
	zip -j ${{ github.workspace }}/out/llama-box-linux-${{ matrix.arch }}-hip-${{ matrix.version }}.zip ${{ github.workspace }}/build/bin/*
	- name: Upload Artifact
	uses: actions/upload-artifact@v4
	with:
	path: ${{ github.workspace }}/out/*.zip
	name: llama-box-linux-${{ matrix.arch }}-hip-${{ matrix.version }}

	# cache-able building with ccache.
	linux-cuda:
	strategy:
	fail-fast: false
	matrix:
	arch: [ amd64 ]
	# see https://hub.docker.com/r/nvidia/cuda/tags?page=&page_size=&ordering=&name=devel.
	# 12.5 ==> 12.5.0
	# 11.7 ==> 11.7.1
	version: [ '12.5', '11.7' ]
	runs-on: ubuntu-22.04
	steps:
	- name: Maximize Space
	# see https://github.com/easimon/maximize-build-space/blob/master/action.yml.
	run: \|
	sudo rm -rf /usr/share/dotnet
	sudo rm -rf /usr/local/lib/android
	sudo rm -rf /opt/ghc
	sudo rm -rf /opt/hostedtoolcache/CodeQL
	sudo docker image prune --all --force
	- name: Clone
	uses: actions/checkout@v4
	with:
	fetch-depth: 0
	submodules: 'recursive'
	- name: Setup Cache
	timeout-minutes: 5
	uses: actions/cache@v3
	with:
	key: cache-linux-cuda-${{ matrix.arch }}-${{ matrix.version }}
	path: \|
	${{ github.workspace }}/.cache
	- name: Setup QEMU
	if: ${{ matrix.arch == 'arm64' }}
	uses: docker/setup-qemu-action@v3
	with:
	image: tonistiigi/binfmt:qemu-v7.0.0
	platforms: "arm64"
	- name: Build
	# disable OpenMP,
	# see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
	# https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
	# build fat binary,
	# see https://developer.nvidia.com/cuda-gpus.
	env:
	CCACHE_DIR: "${{ github.workspace }}/.cache/ccache"
	CUDA_ARCHITECTURES: "52;61;70;75;80"
	run: \|
	echo "===== SCRIPT ====="
	cat <<EOF > /tmp/entrypoint.sh
	#!/bin/bash
	apt-get update && apt-get install -y build-essential git cmake ccache
	git config --system --add safe.directory '*'
	mkdir -p ${{ github.workspace }}/.cache
	cmake -S ${{ github.workspace }} -B ${{ github.workspace }}/build -DCMAKE_BUILD_TYPE=Release \
	-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES="${CUDA_ARCHITECTURES}" \
	${{ matrix.arch == 'amd64' && '-DGGML_NATIVE=off' \|\| '-DGGML_NATIVE=on' }} \
	-DGGML_OPENMP=off
	cmake --build ${{ github.workspace }}/build --target llama-box --config Release -- -j $(nproc)
	EOF
	chmod +x /tmp/entrypoint.sh
	cat /tmp/entrypoint.sh

	echo "===== BUILD ====="
	docker run \
	--rm \
	--privileged \
	--platform linux/${{ matrix.arch }} \
	--volume ${{ github.workspace }}:${{ github.workspace }} \
	--workdir ${{ github.workspace }} \
	--env CCACHE_DIR \
	--env CUDA_ARCHITECTURES \
	--volume /tmp/entrypoint.sh:/entrypoint.sh \
	--entrypoint /entrypoint.sh \
	nvidia/cuda:${{ matrix.version == '12.5' && '12.5.0' \|\| '11.7.1' }}-devel-ubuntu22.04

	echo "===== RESULT ====="
	if [ -f ${{ github.workspace }}/build/bin/llama-box ]; then
	ldd ${{ github.workspace }}/build/bin/llama-box
	else
	exit 1
	fi

	echo "===== PACKAGE ====="
	mkdir -p ${{ github.workspace }}/out
	zip -j ${{ github.workspace }}/out/llama-box-linux-${{ matrix.arch }}-cuda-${{ matrix.version }}.zip ${{ github.workspace }}/build/bin/*
	- name: Upload Artifact
	uses: actions/upload-artifact@v4
	with:
	path: ${{ github.workspace }}/out/*.zip
	name: llama-box-linux-${{ matrix.arch }}-cuda-${{ matrix.version }}

	# cache-able building with ccache.
	linux-oneapi:
	strategy:
	fail-fast: false
	matrix:
	arch: [ amd64 ]
	# see https://hub.docker.com/r/intel/oneapi-basekit/tags?page=&page_size=&ordering=&name=devel.
	# 2024.2 ==> 2024.2.0
	# 2024.1 ==> 2024.1.1
	version: [ '2024.2', '2024.1' ]
	runs-on: ubuntu-22.04
	steps:
	- name: Maximize Space
	# see https://github.com/easimon/maximize-build-space/blob/master/action.yml.
	run: \|
	sudo rm -rf /usr/share/dotnet
	sudo rm -rf /usr/local/lib/android
	sudo rm -rf /opt/ghc
	sudo rm -rf /opt/hostedtoolcache/CodeQL
	sudo docker image prune --all --force
	- name: Clone
	uses: actions/checkout@v4
	with:
	fetch-depth: 0
	submodules: 'recursive'
	- name: Setup Cache
	timeout-minutes: 5
	uses: actions/cache@v3
	with:
	key: cache-linux-oneapi-${{ matrix.arch }}-${{ matrix.version }}
	path: \|
	${{ github.workspace }}/.cache
	- name: Setup QEMU
	if: ${{ matrix.arch == 'arm64' }}
	uses: docker/setup-qemu-action@v3
	with:
	image: tonistiigi/binfmt:qemu-v7.0.0
	platforms: "arm64"
	- name: Build
	# disable OpenMP,
	# see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
	# https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
	env:
	CCACHE_DIR: "${{ github.workspace }}/.cache/ccache"
	run: \|
	echo "===== SCRIPT ====="
	cat <<EOF > /tmp/entrypoint.sh
	#!/bin/bash
	apt-get update && apt-get install -y build-essential git cmake ccache
	git config --system --add safe.directory '*'
	mkdir -p ${{ github.workspace }}/.cache
	cmake -S ${{ github.workspace }} -B ${{ github.workspace }}/build -DCMAKE_BUILD_TYPE=Release \
	-DGGML_SYCL=on -DGGML_SYCL_F16=on \
	${{ matrix.arch == 'amd64' && '-DGGML_NATIVE=off' \|\| '-DGGML_NATIVE=on' }} \
	-DGGML_OPENMP=off
	cmake --build ${{ github.workspace }}/build --target llama-box --config Release -- -j $(nproc)
	EOF
	chmod +x /tmp/entrypoint.sh
	cat /tmp/entrypoint.sh

	echo "===== BUILD ====="
	docker run \
	--rm \
	--privileged \
	--platform linux/${{ matrix.arch }} \
	--volume ${{ github.workspace }}:${{ github.workspace }} \
	--workdir ${{ github.workspace }} \
	--env CC=icx \
	--env CXX=icpx \
	--env CCACHE_DIR \
	--volume /tmp/entrypoint.sh:/entrypoint.sh \
	--entrypoint /entrypoint.sh \
	intel/oneapi-basekit:${{ matrix.version == '2024.2' && '2024.2.0' \|\| '2024.1.1' }}-devel-ubuntu22.04

	echo "===== RESULT ====="
	if [ -f ${{ github.workspace }}/build/bin/llama-box ]; then
	ldd ${{ github.workspace }}/build/bin/llama-box
	else
	exit 1
	fi

	echo "===== PACKAGE ====="
	mkdir -p ${{ github.workspace }}/out
	zip -j ${{ github.workspace }}/out/llama-box-linux-${{ matrix.arch }}-oneapi-${{ matrix.version }}.zip ${{ github.workspace }}/build/bin/*
	- name: Upload Artifact
	uses: actions/upload-artifact@v4
	with:
	path: ${{ github.workspace }}/out/*.zip
	name: llama-box-linux-${{ matrix.arch }}-oneapi-${{ matrix.version }}

	# cache-able building with ccache.
	windows-hip:
	strategy:
	fail-fast: false
	matrix:
	arch: [ amd64 ]
	# see https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html.
	# 5.7 ==> 5.7.1
	# 5.5 ==> 5.5.1
	version: [ '5.7', '5.5' ]
	runs-on: windows-2022
	steps:
	- name: Clone
	uses: actions/checkout@v4
	with:
	fetch-depth: 0
	submodules: 'recursive'
	- name: Deps
	run: \|
	$ErrorActionPreference = "Stop"
	$ProgressPreference = 'SilentlyContinue'

	choco install ccache curl -y
	- name: Setup Cache
	timeout-minutes: 5
	uses: actions/cache@v3
	with:
	key: cache-windows-hip-${{ matrix.arch }}-${{ matrix.version }}
	path: \|
	${{ github.workspace }}\.cache
	- name: Setup HIP
	run: \|
	$ErrorActionPreference = "Stop"
	$ProgressPreference = 'SilentlyContinue'

	Write-Host "I install AMD ROCm HIP SDK"
	curl.exe --retry 5 --retry-delay 5 `
	--output "${{ runner.temp }}\installer.exe" `
	--url "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-${{ matrix.version == '5.7' && '23.Q4' \|\| '23.Q3' }}-WinSvr2022-For-HIP.exe"
	Start-Process "${{ runner.temp }}\installer.exe" -NoNewWindow -Wait `
	-ArgumentList '-install'

	Write-Host "I verify AMD ROCm HIP SDK"
	& 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' --version

	$hipPath = "$(Resolve-Path -Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' \| Split-Path \| Split-Path)"
	"HIP_PATH=${hipPath}" \| Out-File -FilePath $env:GITHUB_ENV -Append
	- name: Build
	# disable OpenMP,
	# see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
	# https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
	# build fat binary,
	# see https://github.com/ggerganov/llama.cpp/pull/1087#issuecomment-1682807878,
	# https://llvm.org/docs/AMDGPUUsage.html.
	# official gpu support list,
	# see https://rocm.docs.amd.com/en/docs-5.7.1/release/windows_support.html,
	# https://rocm.docs.amd.com/en/docs-5.5.1/release/windows_support.html.
	env:
	CCACHE_DIR: "${{ github.workspace }}\\.cache\\ccache"
	AMDGPU_TARGETS: "${{ matrix.version == '5.7' && 'gfx803;gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1010;gfx1030;gfx1100;gfx1101;gfx1102' \|\| 'gfx803;gfx900;gfx906;gfx908;gfx90a;gfx940;gfx1010;gfx1030;gfx1100;gfx1101;gfx1102' }}"
	run: \|
	Write-Host "===== BUILD ====="
	Write-Host "HIP_PATH=${env:HIP_PATH}"
	New-Item -Force -ItemType Directory -Path "${{ github.workspace }}\.cache" -ErrorAction Ignore \| Out-Null
	$env:CMAKE_PREFIX_PATH = "${env:HIP_PATH}"
	cmake -G "Unix Makefiles" -S ${{ github.workspace }} -B ${{ github.workspace }}\build -DCMAKE_BUILD_TYPE=Release `
	-DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
	-DGGML_HIPBLAS=on -DAMDGPU_TARGETS="${env:AMDGPU_TARGETS}" `
	${{ matrix.arch == 'amd64' && '-DGGML_NATIVE=off' \|\| '-DGGML_NATIVE=on' }} `
	-DGGML_OPENMP=off
	cmake --build ${{ github.workspace }}\build --target llama-box --config Release -- -j ${env:NUMBER_OF_PROCESSORS}

	Write-Host "===== RESULT ====="
	if (Test-Path -Path "${{ github.workspace }}\build\bin\llama-box.exe") {
	llvm-objdump.exe -p "${{ github.workspace }}\build\bin\llama-box.exe"
	} else {
	exit 1
	}

	Write-Host "===== PACKAGE ====="
	New-Item -Force -ItemType Directory -Path "${{ github.workspace }}\out" -ErrorAction Ignore \| Out-Null
	Compress-Archive -Path "${{ github.workspace }}\build\bin\*" -DestinationPath "${{ github.workspace }}\out\llama-box-windows-${{ matrix.arch }}-hip-${{ matrix.version }}.zip"
	- name: Upload Artifact
	uses: actions/upload-artifact@v4
	with:
	path: ${{ github.workspace }}\\out\\*.zip
	name: llama-box-windows-${{ matrix.arch }}-hip-${{ matrix.version }}

	# uncache-able building,
	# see https://stackoverflow.com/questions/72829476/how-to-use-ccache-4-6-1-on-windows-msvc-with-cmake.
	windows-cuda:
	strategy:
	fail-fast: false
	matrix:
	arch: [ amd64 ]
	# see https://developer.nvidia.com/cuda-downloads?target_os=Windows&target_arch=x86_64&target_version=Server2022&target_type=exe_network.
	# 12.5 ==> 12.5.0
	# 11.7 ==> 11.7.1
	version: [ '12.5', '11.7' ]
	# see https://github.com/actions/runner-images?tab=readme-ov-file#available-images,
	# https://forums.developer.nvidia.com/t/problems-with-latest-vs2022-update/294150.
	runs-on: ${{ matrix.version == '12.5' && 'windows-2022' \|\| 'windows-2019' }}
	steps:
	- name: Clone
	uses: actions/checkout@v4
	with:
	fetch-depth: 0
	submodules: 'recursive'
	- name: Setup CUDA
	# ensure MSBuildExtensions has been configured,
	# see https://github.com/NVlabs/tiny-cuda-nn/issues/164#issuecomment-1280749170.
	uses: Jimver/cuda-toolkit@v0.2.16
	with:
	cuda: ${{ matrix.version == '12.5' && '12.5.0' \|\| '11.7.1' }}
	method: 'network'
	sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust", "visual_studio_integration"]'
	use-github-cache: false
	use-local-cache: false
	- name: Build
	# disable OpenMP,
	# see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
	# https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
	# build fat binary,
	# see https://developer.nvidia.com/cuda-gpus.
	env:
	CUDA_ARCHITECTURES: "52;61;70;75;80"
	run: \|
	$ErrorActionPreference = "Stop"
	$ProgressPreference = 'SilentlyContinue'

	Write-Host "===== BUILD ====="
	Write-Host "CUDA_PATH=${env:CUDA_PATH}"
	cmake -S ${{ github.workspace }} -B ${{ github.workspace }}\build -DCMAKE_BUILD_TYPE=Release `
	-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES="${env:CUDA_ARCHITECTURES}" `
	${{ matrix.arch == 'amd64' && '-DGGML_NATIVE=off' \|\| '-DGGML_NATIVE=on' }} `
	-DGGML_OPENMP=off
	cmake --build ${{ github.workspace }}\build --target llama-box --config Release -- /m:${env:NUMBER_OF_PROCESSORS}

	Write-Host "===== RESULT ====="
	if (Test-Path -Path "${{ github.workspace }}\build\bin\Release\llama-box.exe") {
	llvm-objdump.exe -p "${{ github.workspace }}\build\bin\Release\llama-box.exe"
	} else {
	exit 1
	}

	Write-Host "===== PACKAGE ====="
	New-Item -Force -ItemType Directory -Path "${{ github.workspace }}\out" -ErrorAction Ignore \| Out-Null
	Compress-Archive -Path "${{ github.workspace }}\build\bin\Release\*" -DestinationPath "${{ github.workspace }}\out\llama-box-windows-${{ matrix.arch }}-cuda-${{ matrix.version }}.zip"
	- name: Upload Artifact
	uses: actions/upload-artifact@v4
	with:
	path: ${{ github.workspace }}\\out\\*.zip
	name: llama-box-windows-${{ matrix.arch }}-cuda-${{ matrix.version }}

	# uncache-able building,
	# as the oneAPI need to configure the environment variables via setvars.bat.
	windows-oneapi:
	strategy:
	fail-fast: false
	matrix:
	arch: [ amd64 ]
	# see https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit-download.html?operatingsystem=windows&windows-install-type=offline.
	# 2024.2 ==> 2024.2.0
	# 2024.1 ==> 2024.1.1
	version: [ '2024.2', '2024.1' ]
	runs-on: windows-2022
	steps:
	- name: Clone
	uses: actions/checkout@v4
	with:
	fetch-depth: 0
	submodules: 'recursive'
	- name: Deps
	run: \|
	$ErrorActionPreference = "Stop"
	$ProgressPreference = 'SilentlyContinue'

	choco install curl ninja -y
	- name: Setup oneAPI
	run: \|
	$ErrorActionPreference = "Stop"
	$ProgressPreference = 'SilentlyContinue'

	Write-Host "I install Intel oneAPI SDK"
	curl.exe --retry 5 --retry-delay 5 `
	--output "${{ runner.temp }}\installer.exe" `
	--url "https://registrationcenter-download.intel.com/akdlm/IRC_NAS/${{ matrix.version == '2024.2' && 'e83a8e64-04fc-45df-85c6-c2208d03bdb5/w_BaseKit_p_2024.2.0.635' \|\| '7dff44ba-e3af-4448-841c-0d616c8da6e7/w_BaseKit_p_2024.1.0.595' }}.exe"
	Start-Process "${{ runner.temp }}\installer.exe" -NoNewWindow -Wait `
	-ArgumentList '-s','--action=install','--components=intel.oneapi.win.cpp-dpcpp-common:intel.oneapi.win.mkl.devel','--eula=accept','-p=NEED_VS2017_INTEGRATION=0','-p=NEED_VS2019_INTEGRATION=0','-p=NEED_VS2022_INTEGRATION=0'

	Write-Host "I verify Intel oneAPI SDK"
	& 'C:\Program Files (x86)\Intel\oneAPI\*\bin\icx.exe' --version

	$oneapiPath = "$(Resolve-Path -Path 'C:\Program Files (x86)\Intel\oneAPI\*\bin\icx.exe' \| Split-Path \| Split-Path)"
	"ONEAPI_PATH=${oneapiPath}" \| Out-File -FilePath $env:GITHUB_ENV -Append
	$oneapiRoot = "$(Split-Path -Path $oneapiPath)"
	"ONEAPI_ROOT=${oneapiRoot}" \| Out-File -FilePath $env:GITHUB_ENV -Append
	- name: Build
	# disable OpenMP,
	# see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
	# https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
	run: \|
	Write-Host "===== BUILD ====="
	Write-Host "ONEAPI_PATH=${env:ONEAPI_PATH}"
	Write-Host "ONEAPI_ROOT=${env:ONEAPI_ROOT}"
	& "${{ github.workspace }}\llama-box\scripts\build-windows-oneapi.bat" "${{ github.workspace }}" "${{ matrix.arch }}"

	Write-Host "===== RESULT ====="
	if (Test-Path -Path "${{ github.workspace }}\build\bin\llama-box.exe") {
	llvm-objdump.exe -p "${{ github.workspace }}\build\bin\llama-box.exe"
	} else {
	exit 1
	}

	Write-Host "===== PACKAGE ====="
	New-Item -Force -ItemType Directory -Path "${{ github.workspace }}\out" -ErrorAction Ignore \| Out-Null
	Compress-Archive -Path "${{ github.workspace }}\build\bin\*" -DestinationPath "${{ github.workspace }}\out\llama-box-windows-${{ matrix.arch }}-oneapi-${{ matrix.version }}.zip"
	- name: Upload Artifact
	uses: actions/upload-artifact@v4
	with:
	path: ${{ github.workspace }}\\out\\*.zip
	name: llama-box-windows-${{ matrix.arch }}-oneapi-${{ matrix.version }}

	release:
	if: ${{ startsWith(github.ref, 'refs/tags/') }}
	permissions:
	contents: write
	actions: read
	id-token: write
	runs-on: ubuntu-22.04
	needs:
	- darwin-metal
	- linux-hip
	- linux-cuda
	- linux-oneapi
	- windows-hip
	- windows-cuda
	- windows-oneapi
	steps:
	- name: Download Artifact
	uses: actions/download-artifact@v4
	with:
	path: ${{ github.workspace }}/out
	merge-multiple: true
	- name: Release
	uses: softprops/action-gh-release@v1
	with:
	fail_on_unmatched_files: true
	tag_name: "${{ env.VERSION }}"
	prerelease: ${{ contains(github.ref, 'rc') }}
	files: ${{ github.workspace }}/out/*

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ci: reduce cuda binary size #79

Workflow file

ci: reduce cuda binary size #79

Jobs

Run details

Workflow file for this run