azure-pipelines.yml

# Azure pipelines for Marian NMT
#
# The pipeline need to be added manually to the repository, for example:
# 1. Go to Your repository > Pipelines, click "New pipeline"
# 2. Choose "Azure Repos Git" and a repository
# 3. Choose "Existing Azure Pipelines YAML file" and specify path to this file
# 4. "More actions" > "Save"

parameters:
# Allow skipping the entire 'Build' stage
- name: runBuilds
  displayName: Run builds? Uncheck to run regression tests only.
  type: boolean
  default: true

# Warning: the current branch policies disable the automatic triggering to
# minimize VM usage!
# The configuration below specifies that the pipeline CI trigger is set on the
# branch master only and a PR trigger is on a (non-draft) pull request to any
# branch.
trigger:
  # This minimizes the number of parallel pipeline runs. When a pipeline is
  # running, the CI waits until it is completed before starting another one.
  batch: true
  branches:
    include:
    - master
  paths:
    exclude:
    - azure-regression-tests.yml
    - contrib
    - doc
    - examples
    - regression-tests
    - scripts
    - VERSION
    - vs
    - '**/*.md'
    - '**/*.txt'

pool:
  name: Azure Pipelines

variables:
  - group: marian-regression-tests
  - name: BOOST_ROOT_WINDOWS
    value: "C:/hostedtoolcache/windows/Boost/1.72.0/x86_64"
  - name: BOOST_URL
    value: "https://sourceforge.net/projects/boost/files/boost-binaries/1.72.0/boost_1_72_0-msvc-14.2-64.exe"
  - name: CUDA_PATH_WINDOWS
    value: "C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA"
  - name: MKL_DIR
    value: "$(Build.SourcesDirectory)/mkl"
  - name: MKL_URL
    value: "https://data.statmt.org/romang/marian-regression-tests/ci/mkl-2020.1-windows-static.zip"
  - name: VCPKG_COMMIT
    value: 2022.03.10
  - name: VCPKG_DIR
    value: "$(Build.SourcesDirectory)/vcpkg"
  - name: VCPKG_PACKAGES
    value: "protobuf"
  # The Visual Studio installation directory can be found using:
  #   pushd "C:\Program Files (x86)\Microsoft Visual Studio\Installer\"
  #   for /f "delims=" %%x in ('.\vswhere.exe -latest -property InstallationPath') do set VSPATH=%%x
  #   popd
  - name: VS_PATH
    value: "C:/Program Files (x86)/Microsoft Visual Studio/2019/Enterprise"

stages:
- stage: Builds
  jobs:

  ######################################################################
  - job: BuildWindows
    cancelTimeoutInMinutes: 1
    condition: eq(${{ parameters.runBuilds }}, true)
    displayName: Windows

    strategy:
      matrix:
        # Windows CPU-only build
        "CPU":
          cuda: false
          cuda_version: ""
        # Windows CPU+GPU build
        "CUDA 10.2":
          cuda: true
          cuda_version: 10.2

    pool:
      vmImage: windows-2019

    steps:
    - checkout: self
      submodules: true

    - pwsh: .\scripts\ci\install_cuda_windows.ps1 "$(cuda_version)"
      displayName: Install CUDA
      condition: eq(variables.cuda, true)

    - pwsh: |
        C:\msys64\usr\bin\wget.exe -nv $(MKL_URL) -O mkl.zip
        Expand-Archive -Force mkl.zip $(MKL_DIR)
      displayName: Download MKL

    ## Cache for Boost
    #- task: Cache@2
    #  displayName: Cache Boost
    #  inputs:
    #    # Change the first value (v0) to another value to clear the cache
    #    key: '"v0" | "boost" | "$(BOOST_URL)" | "$(BOOST_ROOT_WINDOWS)" | "$(Agent.OS)"'
    #    path: $(BOOST_ROOT_WINDOWS)
    #    cacheHitVar: CACHE_BOOST_RESTORED

    # Boost is no longer pre-installed on Azure/GitHub-hosted Windows runners
    - pwsh: |
        Write-Host "Downloading Boost to $(BOOST_ROOT_WINDOWS)"
        C:\msys64\usr\bin\wget.exe -nv "$(BOOST_URL)" -O "$(Pipeline.Workspace)/boost.exe"
        Start-Process -Wait -FilePath "$(Pipeline.Workspace)/boost.exe" "/SILENT","/SP-","/SUPPRESSMSGBOXES","/DIR=$(BOOST_ROOT_WINDOWS)"
      displayName: Download Boost
      condition: ne(variables.CACHE_BOOST_RESTORED, 'true')

    ## Cache for vcpkg packages. It does not work yet properly due to linker errors after restoring it.
    #- task: Cache@2
    #  displayName: Cache vcpkg
    #  inputs:
    #    # Change the first value (v0) to another value to clear the cache
    #    key: '"v0" | "vcpkg" | "$(VCPKG_COMMIT)" | "$(VCPKG_PACKAGES)" | "$(Agent.OS)"'
    #    path: $(VCPKG_DIR)
    #    cacheHitVar: CACHE_VCPKG_RESTORED

    - pwsh: |
        git clone https://github.com/Microsoft/vcpkg.git $(VCPKG_DIR)
        cd $(VCPKG_DIR)
        git checkout $(VCPKG_COMMIT)
        pushd
        .\bootstrap-vcpkg.bat -disableMetrics
        popd
        # Install packages
        .\vcpkg.exe install --triplet x64-windows-static $(VCPKG_PACKAGES)
        # Clean to make the cache smaller
        Remove-Item $(VCPKG_DIR)\downloads  -Force -Recurse -ErrorAction SilentlyContinue
        Remove-Item $(VCPKG_DIR)\buildtrees -Force -Recurse -ErrorAction SilentlyContinue
      displayName: Prepare vcpkg
      condition: ne(variables.CACHE_VCPKG_RESTORED, 'true')

    - script: |
        :: Load VS environment
        call "$(VS_PATH)/VC/Auxiliary/Build/vcvarsall.bat" x64
        :: Create build directory
        mkdir build
        cd build
        :: Run CMake
        cmake .. -G Ninja ^
          -DCMAKE_BUILD_TYPE="Debug" ^
          -DCMAKE_C_COMPILER="cl.exe" ^
          -DCMAKE_CXX_COMPILER="cl.exe" ^
          -DCMAKE_MAKE_PROGRAM="ninja.exe" ^
          -DCMAKE_TOOLCHAIN_FILE="$(VCPKG_DIR)\scripts\buildsystems\vcpkg.cmake" ^
          -DVCPKG_TARGET_TRIPLET="x64-windows-static" ^
          ^
          -DBOOST_ROOT="$(BOOST_ROOT)" ^
          -DBOOST_INCLUDEDIR="$(BOOST_ROOT)/include" ^
          -DBOOST_LIBRARYDIR="$(BOOST_ROOT)/lib" ^
          -DOPENSSL_USE_STATIC_LIBS="TRUE" ^
          -DOPENSSL_MSVC_STATIC_RT="TRUE" ^
          ^
          -DCOMPILE_CPU="TRUE" ^
          -DCOMPILE_CUDA="$(cuda)" ^
          -DCOMPILE_EXAMPLES="FALSE" ^
          -DCOMPILE_SERVER="TRUE" ^
          -DCOMPILE_TESTS="TRUE" ^
          ^
          -DUSE_CUDNN="FALSE" ^
          -DUSE_FBGEMM="TRUE" ^
          -DUSE_MPI="FALSE" ^
          -DUSE_NCCL="FALSE" ^
          -DUSE_SENTENCEPIECE="TRUE" ^
          -DUSE_STATIC_LIBS="TRUE"
      displayName: Configure CMake
      env:
        # Set envvars so that CMake can find the installed packages
        MKLROOT: $(MKL_DIR)
        CUDA_PATH: $(CUDA_PATH_WINDOWS)/v$(cuda_version)
        BOOST_ROOT: $(BOOST_ROOT_WINDOWS)

    - script: |
        call "$(VS_PATH)/VC/Auxiliary/Build/vcvarsall.bat" x64
        ninja
      displayName: Compile
      workingDirectory: build

    - script: |
        call "$(VS_PATH)/VC/Auxiliary/Build/vcvarsall.bat" x64
        ctest
      displayName: Run unit tests
      workingDirectory: build
      condition: eq(variables.cuda, false)

    # Note that versions from Marian executables will not be printed for CUDA builds
    - script: |
        .\marian.exe --version
        .\marian-decoder.exe --version
        .\marian-scorer.exe --version
        .\marian-server.exe --version
        .\spm_encode.exe --version
      displayName: Print versions
      workingDirectory: build

  ######################################################################
  - job: BuildUbuntu
    cancelTimeoutInMinutes: 1
    condition: eq(${{ parameters.runBuilds }}, true)
    displayName: Ubuntu
    timeoutInMinutes: 120

    # Minimal tested configurations for marian-dev v1.11 and C++17:
    # * Ubuntu 16.04, GCC 7.5, CMake 3.10.2, CUDA 9.2 (probably GCC 6 would work too)
    # * Ubuntu 18.04, GCC 7.5, CMake 3.12.2, CUDA 10.0
    strategy:
      matrix:
        ################################################################
        # Ubuntu CPU-only build
        "CPU-only":
          image: ubuntu-20.04
          boost: true
          cpu: true
          gpu: false
          cuda: ""
          gcc: 9
          unit_tests: true
          examples: false
          static: true
        # Ubuntu GPU-only build
        "GPU-only":
          image: ubuntu-20.04
          boost: true
          cpu: false
          gpu: true
          cuda: 11.1
          gcc: 9
          unit_tests: false
          examples: false
          static: false
        ################################################################
        # Ubuntu 22.04 supports CUDA 11+
        #
        # CPU is disabled because FBGEMM + GCC 9+ do not compile on machines with
        # avx512_vnni, see https://github.com/marian-nmt/marian-dev/issues/709
        "22.04 CUDA 11.7 gcc-11":
          image: ubuntu-22.04
          boost: false
          cpu: false
          gpu: true
          cuda: 11.7
          gcc: 11
          unit_tests: false # disable unit tests to minimize compilation time
          examples: false   # disable examples to minimize compilation time
          static: false
        ################################################################
        # Ubuntu 20.04 supports CUDA 11+
        "20.04 CUDA 11.1 gcc-9":
          image: ubuntu-20.04
          boost: true
          cpu: true
          gpu: true
          cuda: 11.1
          gcc: 9
          unit_tests: true
          examples: true
          static: true
        ################################################################
        # Ubuntu 16.04 is no longer available on Azure-hosted machines

    pool:
      vmImage: $(image)

    steps:
    - checkout: self
      submodules: true

    # The following packages are already installed on Azure-hosted runners: build-essential openssl libssl-dev
    # No need to install libprotobuf{17,10,9v5} on Ubuntu {20,18,16}.04 because it is installed together with libprotobuf-dev
    # Note that installation of libunwind-dev is a bug fix for ubuntu-22.04 images on Azure/GitHub-hosted machines
    # and is normally not required
    - bash: sudo apt-get install -y libunwind-dev libgoogle-perftools-dev libprotobuf-dev protobuf-compiler
      displayName: Install packages

    # Boost is no longer pre-installed on Azure/GitHub-hosted runners
    - bash: sudo apt-get install -y libboost-system-dev
      displayName: Install Boost
      condition: eq(variables.boost, true)

    # https://software.intel.com/content/www/us/en/develop/articles/installing-intel-free-libs-and-python-apt-repo.html
    - bash: |
        wget -qO- "https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB" | sudo apt-key add -
        sudo sh -c "echo deb https://apt.repos.intel.com/mkl all main > /etc/apt/sources.list.d/intel-mkl.list"
        sudo apt-get update -o Dir::Etc::sourcelist="/etc/apt/sources.list.d/intel-mkl.list"
        sudo apt-get install -y --no-install-recommends intel-mkl-64bit-2020.0-088
      displayName: Install MKL
      condition: eq(variables.cpu, true)

    # The script simplifies installation of different versions of CUDA
    - bash: ./scripts/ci/install_cuda_ubuntu.sh $(cuda)
      displayName: Install CUDA
      condition: eq(variables.gpu, true)

    - bash: |
        mkdir -p build
        cd build
        CC=/usr/bin/gcc-$(gcc) CXX=/usr/bin/g++-$(gcc) CUDAHOSTCXX=/usr/bin/g++-$(gcc) \
        cmake .. \
          -DCOMPILE_CPU=$(cpu) \
          -DCOMPILE_CUDA=$(gpu) \
          -DCOMPILE_EXAMPLES=$(examples) \
          -DCOMPILE_SERVER=$(boost) \
          -DCOMPILE_TESTS=$(unit_tests) \
          -DUSE_FBGEMM=$(cpu) \
          -DUSE_SENTENCEPIECE=on \
          -DUSE_STATIC_LIBS=$(static) \
          -DBoost_ARCHITECTURE=-x64 \
          -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-$(cuda)
      displayName: Configure CMake

    # Clean build/src/ to safe disk space on Azure-hosted VMs and stay below the 10GB limit
    - bash: |
        make -j3
        rm -rf src/
      displayName: Compile
      workingDirectory: build

    # Unit tests are run only for CPU-only builds because Azure-hosted runners do not have GPUs
    # TODO: add an option to CMake to compile unit tests only for CPU
    - bash: make test
      displayName: Run unit tests
      workingDirectory: build
      condition: and(eq(variables.unit_tests, true), eq(variables.gpu, false))

    - bash: |
        ./marian --version
        ./marian-decoder --version
        ./marian-scorer --version
        ./spm_encode --version
      displayName: Print versions
      workingDirectory: build

  ######################################################################
  - job: BuildMacOS
    cancelTimeoutInMinutes: 1
    condition: eq(${{ parameters.runBuilds }}, true)
    displayName: macOS CPU clang

    pool:
      vmImage: macos-12

    steps:
    - checkout: self
      submodules: true

    - bash: brew install boost openblas openssl protobuf
      displayName: Install packages

    # Openblas location is exported explicitly because openblas is keg-only, which means it was not symlinked into /usr/local/.
    # CMake cannot find BLAS on Azure runners if Marian is being compiled statically, hence USE_STATIC_LIBS=off
    - bash: |
        export LDFLAGS="-L/usr/local/opt/openblas/lib"
        export CPPFLAGS="-I/usr/local/opt/openblas/include"
        mkdir -p build
        cd build
        cmake .. \
          -DCOMPILE_CPU=on \
          -DCOMPILE_CUDA=off \
          -DCOMPILE_EXAMPLES=on \
          -DCOMPILE_SERVER=off \
          -DCOMPILE_TESTS=on \
          -DUSE_FBGEMM=on \
          -DUSE_SENTENCEPIECE=on \
          -DUSE_STATIC_LIBS=off
      displayName: Configure CMake

    - bash: make -j2
      displayName: Compile
      workingDirectory: build

    - bash: make test
      displayName: Run unit tests
      workingDirectory: build

    - bash: |
        ./marian --version
        ./marian-decoder --version
        ./marian-scorer --version
        ./spm_encode --version
      displayName: Print versions
      workingDirectory: build

  ######################################################################
  - job: BuildInstall
    cancelTimeoutInMinutes: 1
    condition: eq(${{ parameters.runBuilds }}, true)
    displayName: Linux CPU library install

    pool:
      vmImage: ubuntu-20.04

    steps:
    - checkout: self
      submodules: true

    # The following packages are already installed on Azure-hosted runners: build-essential openssl libssl-dev
    # No need to install libprotobuf{17,10,9v5} on Ubuntu {20,18,16}.04 because it is installed together with libprotobuf-dev
    - bash: sudo apt-get install -y libgoogle-perftools-dev libprotobuf-dev protobuf-compiler
      displayName: Install packages

    # https://software.intel.com/content/www/us/en/develop/articles/installing-intel-free-libs-and-python-apt-repo.html
    - bash: |
        wget -qO- "https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB" | sudo apt-key add -
        sudo sh -c "echo deb https://apt.repos.intel.com/mkl all main > /etc/apt/sources.list.d/intel-mkl.list"
        sudo apt-get update -o Dir::Etc::sourcelist="/etc/apt/sources.list.d/intel-mkl.list"
        sudo apt-get install -y --no-install-recommends intel-mkl-64bit-2020.0-088
      displayName: Install MKL

    - bash: /usr/bin/gcc-9 --version
      displayName: Print GCC version

    - bash: |
        mkdir -p install
        mkdir -p build
        cd build
        CC=/usr/bin/gcc-9 CXX=/usr/bin/g++-9 \
        cmake .. \
          -DCMAKE_INSTALL_PREFIX=../install \
          -DCMAKE_BUILD_TYPE=slim \
          -DCOMPILE_LIBRARY_ONLY=on \
          -DCOMPILE_CUDA=off \
          -DGENERATE_MARIAN_INSTALL_TARGETS=on \
          -DUSE_FBGEMM=on \
          -DUSE_SENTENCEPIECE=on
      displayName: Configure CMake

    - bash: make -j3 install
      displayName: Compile & install
      workingDirectory: build

    - bash: |
        test -e lib/libmarian.a
        test -e lib/libfbgemm.a
        test -e lib/libsentencepiece.a
        ls -lah *
      displayName: Check targets
      workingDirectory: install


# Marian is built in the same job where the regression tests are run to make sure that executables
# are compiled and run on a machine with the same CPU architecture, which is required for
# compilations with FBGEMM.
- stage: Tests
  jobs:

  ######################################################################
  - job: TestWindows
    cancelTimeoutInMinutes: 1
    displayName: Windows CPU+FBGEMM

    pool:
      vmImage: windows-2019

    steps:
    # Due to multiple checkouts this will be commonly cloned into D:\a\1\s\marian-dev
    - checkout: self
      submodules: true

    - pwsh: |
        C:\msys64\usr\bin\wget.exe -nv $(MKL_URL) -O mkl.zip
        Expand-Archive -Force mkl.zip $(MKL_DIR)
      displayName: Download MKL

    # Cache for vcpkg packages
    - task: Cache@2
      displayName: Cache vcpkg
      inputs:
        # Change the first value (v0) to another value to clear the cache
        key: '"v0" | "vcpkg" | "$(VCPKG_COMMIT)" | "$(VCPKG_PACKAGES)" | "$(Agent.OS)"'
        path: $(VCPKG_DIR)
        cacheHitVar: CACHE_VCPKG_RESTORED

    - pwsh: |
        git clone https://github.com/Microsoft/vcpkg.git $(VCPKG_DIR)
        cd $(VCPKG_DIR)
        git checkout $(VCPKG_COMMIT)
        pushd
        .\bootstrap-vcpkg.bat -disableMetrics
        popd
        # Install packages
        .\vcpkg.exe install --triplet x64-windows-static $(VCPKG_PACKAGES)
        # Clean to make the cache smaller
        Remove-Item $(VCPKG_DIR)\downloads  -Force -Recurse -ErrorAction SilentlyContinue
        Remove-Item $(VCPKG_DIR)\buildtrees -Force -Recurse -ErrorAction SilentlyContinue
      displayName: Prepare vcpkg
      condition: ne(variables.CACHE_VCPKG_RESTORED, 'true')

    - script: |
        :: Load VS environment
        call "$(VS_PATH)/VC/Auxiliary/Build/vcvarsall.bat" x64
        :: Create build directory
        mkdir build
        cd build
        :: Run CMake
        cmake .. -G Ninja ^
          -DCMAKE_BUILD_TYPE="Slim" ^
          -DCMAKE_C_COMPILER="cl.exe" ^
          -DCMAKE_CXX_COMPILER="cl.exe" ^
          -DCMAKE_MAKE_PROGRAM="ninja.exe" ^
          -DCMAKE_TOOLCHAIN_FILE="$(VCPKG_DIR)\scripts\buildsystems\vcpkg.cmake" ^
          -DVCPKG_TARGET_TRIPLET="x64-windows-static" ^
          -DDETERMINISTIC="TRUE" ^
          ^
          -DCOMPILE_CPU="TRUE" ^
          -DCOMPILE_CUDA="FALSE" ^
          ^
          -DUSE_FBGEMM="TRUE" ^
          -DUSE_SENTENCEPIECE="TRUE" ^
          -DUSE_STATIC_LIBS="TRUE"
      displayName: Configure CMake
      env:
        # Set envvars so that CMake can find the installed packages
        MKLROOT: $(MKL_DIR)
      workingDirectory: marian-dev

    - script: |
        pwd
        call "$(VS_PATH)/VC/Auxiliary/Build/vcvarsall.bat" x64
        ninja
      displayName: Compile
      workingDirectory: marian-dev/build

    ## Publish an artifact with Marian executables. Disabled because it is not portable due to
    ## fbgemm and not needed at the moment
    #- pwsh: Compress-Archive -Path marian*.exe spm_*.exe -DestinationPath $(Build.SourcesDirectory)/marian-dev-ci_windows-x64_cpu.zip
      #displayName: Prepare binaries
      #workingDirectory: marian-dev/build
    #- publish: marian-dev-ci_windows-x64_cpu.zip
      #artifact: marian-dev-ci_windows-x64_cpu
      #displayName: Publish binaries

    # Running regression tests
    # Due to multiple checkouts this will be commonly cloned into D:\a\1\s\marian-prod-tests
    - checkout: git://Marian/marian-prod-tests

    # Collect details about the CPU architecture, etc.
    # Because the outputs goes to *.log files, they will be also included in the artifact with test outputs.
    - script: bash -c "cat /proc/cpuinfo | tee cpuinfo.log"
      displayName: Machine statistics
      workingDirectory: marian-prod-tests

    # The current SAS token will expire on 12/31/2023 and a new one will need to be set in Marian > Pipelines > Library
    - bash: |
        cd models
        bash download-models.sh
        ls
      displayName: Prepare tests
      env:
        AZURE_STORAGE_SAS_TOKEN: $(marian-prod-tests-blob-sas-token)
      workingDirectory: marian-prod-tests

    # Avoid using $(Build.SourcesDirectory) in bash tasks because on Windows pools it uses '\'
    # instead of '/', which often breaks the job
    - bash: MARIAN=../marian-dev/build TIMEOUT=10m bash ./run_mrt.sh '#cpu' '#basics' '#devops'
      continueOnError: true
      displayName: Run tests
      workingDirectory: marian-prod-tests

    - bash: |
        # cut -c3- removes './' from paths making 7z to retain the directory structure
        find . -type f \( -name "*.log" -o -name "*.out" -o -name "*.diff" \) -print | cut -c3- > listing.txt
        echo "Creating an artifact with the following files:"
        cat listing.txt
        7z a -tzip ../regression-tests-ci_windows-x64_cpu.zip @listing.txt
      displayName: Collect outputs
      workingDirectory: marian-prod-tests

    - publish: regression-tests-ci_windows-x64_cpu.zip
      artifact: regression-tests-ci_windows-x64_cpu
      displayName: Publish outputs

  ######################################################################
  - job: TestLinux
    cancelTimeoutInMinutes: 1
    displayName: Linux CPU+FBGEMM

    pool:
      vmImage: ubuntu-latest

    steps:
    # Due to multiple checkouts this will be commonly cloned into D:\a\1\s\marian-dev
    - checkout: self
      submodules: true

    # The following packages are already installed on Azure-hosted runners: build-essential openssl libssl-dev
    # No need to install libprotobuf{17,10,9v5} on Ubuntu {20,18,16}.04 because it is installed together with libprotobuf-dev
    # Installing libunwind-dev fixes a bug in 2204 (the libunwind-14 and libunwind-dev conflict)
    - bash: |
        sudo apt-get install -y libunwind-dev
        sudo apt-get install -y libgoogle-perftools-dev libprotobuf-dev protobuf-compiler gcc-9 g++-9
      displayName: Install packages

    # https://software.intel.com/content/www/us/en/develop/articles/installing-intel-free-libs-and-python-apt-repo.html
    - bash: |
        wget -qO- "https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB" | sudo apt-key add -
        sudo sh -c "echo deb https://apt.repos.intel.com/mkl all main > /etc/apt/sources.list.d/intel-mkl.list"
        sudo apt-get update -o Dir::Etc::sourcelist="/etc/apt/sources.list.d/intel-mkl.list"
        sudo apt-get install -y --no-install-recommends intel-mkl-64bit-2020.0-088
      displayName: Install MKL

    # Note that COMPILE_CPU=on and USE_SENTENCEPUECE=on are set explicitly to make them detectable
    # by the regression tests framework (not sure if it is still required)
    - bash: |
        mkdir -p install
        mkdir -p build
        cd build
        CC=/usr/bin/gcc-9 CXX=/usr/bin/g++-9 \
        cmake .. \
          -DCMAKE_BUILD_TYPE=slim \
          -DCOMPILE_CPU=on \
          -DCOMPILE_CUDA=off \
          -DDETERMINISTIC=on \
          -DUSE_FBGEMM=on \
          -DUSE_SENTENCEPIECE=on \
          -DUSE_STATIC_LIBS=on
      displayName: Configure CMake
      workingDirectory: marian-dev

    - bash: make -j3
      displayName: Compile
      workingDirectory: marian-dev/build

    ## Publish an artifact with Marian executables. Disabled because it is not portable due to
    ## fbgemm and not needed at the moment
    #- bash: zip $(Build.SourcesDirectory)/marian-dev-ci_linux-x64-static_cpu.zip marian* spm_*
      #displayName: Prepare binaries
      #workingDirectory: marian-dev/build
    #- publish: marian-dev-ci_linux-x64-static_cpu.zip
      #artifact: marian-dev-ci_linux-x64-static_cpu
      #displayName: Publish binaries

    # Running regression tests
    # Due to multiple checkouts this will be commonly cloned into D:\a\1\s\marian-prod-tests
    - checkout: git://Marian/marian-prod-tests

    # Collect details about the CPU architecture, etc.
    # Because the outputs goes to *.log files, they will be also included in the artifact with test outputs.
    - bash: |
        echo ">>> lscpu"
        lscpu | tee lscpu.log
        echo ">>> cpuinfo"
        cat /proc/cpuinfo | tee cpuinfo.log
        /usr/bin/gcc-9 --version | tee gcc.log
      displayName: Machine statistics
      workingDirectory: marian-prod-tests

    # The current SAS token will expire on 12/31/2023 and a new one will need to be set in Marian > Pipelines > Library
    - bash: |
        cd models
        bash download-models.sh
        ls
      displayName: Prepare tests
      env:
        AZURE_STORAGE_SAS_TOKEN: $(marian-prod-tests-blob-sas-token)
      workingDirectory: marian-prod-tests

    - bash: MARIAN=../marian-dev/build bash ./run_mrt.sh '#cpu' '#basics' '#devops'
      continueOnError: true
      displayName: Run tests
      workingDirectory: marian-prod-tests

    - bash: |
        # cut -c3- removes './' from paths making 7z to retain the directory structure
        find . -type f \( -name "*.log" -o -name "*.out" -o -name "*.diff" \) -print | cut -c3- > listing.txt
        echo "Creating an artifact with the following files:"
        cat listing.txt
        7z a -tzip ../regression-tests-ci_linux-x64-static_cpu.zip @listing.txt
      displayName: Collect outputs
      workingDirectory: marian-prod-tests

    - publish: regression-tests-ci_linux-x64-static_cpu.zip
      artifact: regression-tests-ci_linux-x64-static_cpu
      displayName: Publish outputs