Skip to content

Commit

Permalink
ci: refactor caching logic
Browse files Browse the repository at this point in the history 
Signed-off-by: thxCode <[email protected]>
  • Loading branch information
@thxCode
thxCode committed Jun 27, 2024
1 parent d697388 commit 31132b7
Showing 1 changed file with 110 additions and 29 deletions.
139 changes: 110 additions & 29 deletions .github/workflows/ci.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,8 @@ concurrency:
cancel-in-progress: true

jobs:

# cache-able building with ccache.
darwin-metal:
strategy:
fail-fast: false
Expand All @@ -59,7 +61,7 @@ jobs:
timeout-minutes: 5
uses: actions/cache@v3
with:
key: cache-darwin-metal-${{ matrix.arch }}-${{ matrix.version }}-${{ hashFiles('**/.gitmodules') }}
key: cache-darwin-metal-${{ matrix.arch }}-${{ matrix.version }}
path: |
${{ github.workspace }}/.cache
- name: Deps
Expand Down Expand Up @@ -96,6 +98,7 @@ jobs:
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
Expand Down Expand Up @@ -124,7 +127,7 @@ jobs:
timeout-minutes: 5
uses: actions/cache@v3
with:
key: cache-linux-hip-${{ matrix.arch }}-${{ matrix.version }}-${{ hashFiles('**/.gitmodules') }}
key: cache-linux-hip-${{ matrix.arch }}-${{ matrix.version }}
path: |
${{ github.workspace }}/.cache
- name: Setup QEMU
Expand Down Expand Up @@ -193,6 +196,7 @@ jobs:
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
Expand Down Expand Up @@ -221,7 +225,7 @@ jobs:
timeout-minutes: 5
uses: actions/cache@v3
with:
key: cache-linux-cuda-${{ matrix.arch }}-${{ matrix.version }}-${{ hashFiles('**/.gitmodules') }}
key: cache-linux-cuda-${{ matrix.arch }}-${{ matrix.version }}
path: |
${{ github.workspace }}/.cache
- name: Setup QEMU
Expand Down Expand Up @@ -284,6 +288,97 @@ jobs:
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 \
-DLLAMA_SYCL=on -DLLAMA_SYCL_F16=on \
${{ matrix.arch == 'amd64' && '-DLLAMA_NATIVE=off' || '-DLLAMA_NATIVE=on' }} \
-DLLAMA_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 ${{ github.workspace }} \
--env CC=icx \
--env CXX=icx \
--env CCACHE_DIR \
--volume /tmp/entrypoint.sh:/entrypoint.sh \
--entrypoint /entrypoint.sh \
intel/oneapi-basekit:${{ 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 }}-openapi-${{ 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
Expand All @@ -304,35 +399,24 @@ jobs:
timeout-minutes: 5
uses: actions/cache@v3
with:
key: cache-windows-hip-${{ matrix.arch }}-${{ matrix.version }}-${{ hashFiles('**/.gitmodules') }}
key: cache-windows-hip-${{ matrix.arch }}-${{ matrix.version }}
path: |
${{ github.workspace }}\.cache
${{ github.workspace }}\tmp
- name: Deps
run: |
$ErrorActionPreference = "Stop"
$WarningPreference = 'SilentlyContinue'
$VerbosePreference = 'SilentlyContinue'
$DebugPreference = 'SilentlyContinue'
$ProgressPreference = 'SilentlyContinue'
choco install ccache -y
- name: Setup HIP
id: sdk
run: |
$ErrorActionPreference = "Stop"
$WarningPreference = 'SilentlyContinue'
$VerbosePreference = 'SilentlyContinue'
$DebugPreference = 'SilentlyContinue'
$ProgressPreference = 'SilentlyContinue'
Write-Host "I install AMD ROCm HIP SDK"
if (-not (Test-Path -Path "${{ github.workspace }}\tmp\rocm-install.exe" -PathType Leaf)) {
New-Item -Force -ItemType Directory -Path "${{ github.workspace }}\tmp" -ErrorAction Ignore | Out-Null
Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-${{ matrix.version == '5.7' && '23.Q4' || '23.Q3' }}-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
Move-Item -Path "${env:RUNNER_TEMP}\rocm-install.exe" -Destination "${{ github.workspace }}\tmp\rocm-install.exe"
}
Start-Process "${{ github.workspace }}\tmp\rocm-install.exe" -ArgumentList '-install' -NoNewWindow -Wait
Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-${{ matrix.version == '5.7' && '23.Q4' || '23.Q3' }}-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
Start-Process "${env:RUNNER_TEMP}\rocm-install.exe" -ArgumentList '-install' -NoNewWindow -Wait
Write-Host "I verify AMD ROCm HIP SDK"
& 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' --version
Expand Down Expand Up @@ -380,6 +464,8 @@ jobs:
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
Expand All @@ -398,33 +484,27 @@ jobs:
with:
fetch-depth: 0
submodules: 'recursive'
- name: Setup Cache
# ccache doesn't work with Visual Studio generator,
# see https://stackoverflow.com/questions/72829476/how-to-use-ccache-4-6-1-on-windows-msvc-with-cmake.
timeout-minutes: 5
uses: actions/cache@v3
with:
key: cache-windows-cuda-${{ matrix.arch }}-${{ matrix.version }}-${{ hashFiles('**/.gitmodules') }}
path: |
${{ github.workspace }}\\build
- name: Setup CUDA
# ensure MSBuildExtensions has been configured,
# see https://github.com/NVlabs/tiny-cuda-nn/issues/164#issuecomment-1280749170.
id: sdk
uses: Jimver/[email protected]
with:
cuda: ${{ matrix.version == '12.5' && '12.5.0' || '11.7.1' }}
method: 'local'
method: 'network'
sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust", "visual_studio_integration"]'
- name: Build
# disable OpenMP,
# see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
# https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
# build fat binary,
# see https://developer.nvidia.com/cuda-gpus.
# ensure MSBuildExtensions has been set,
# see https://github.com/NVlabs/tiny-cuda-nn/issues/164#issuecomment-1280749170.
env:
CUDA_ARCHITECTURES: "${{ startsWith(matrix.version, '12.') && '50;52;53;60;61;62;70;72;75;80;86;87;89;90' || '50;52;53;60;61;62;70;72;75;80;86;87' }}"
run: |
$ErrorActionPreference = "Stop"
$ProgressPreference = 'SilentlyContinue'
Write-Host "===== BUILD ====="
Write-Host "CUDA_PATH=${{ steps.sdk.outputs.CUDA_PATH }}"
$env:CUDA_PATH = "${{ steps.sdk.outputs.CUDA_PATH }}"
Expand Down Expand Up @@ -461,6 +541,7 @@ jobs:
- darwin-metal
- linux-hip
- linux-cuda
- linux-openapi
- windows-hip
- windows-cuda
steps:
Expand Down

0 comments on commit 31132b7

Please sign in to comment.