[Refiled] IFU-main-2023-07-31

.github/scripts/validate_binaries.sh

@@ -2,49 +2,39 @@ if [[ ${MATRIX_PACKAGE_TYPE} == "libtorch" ]]; then
     curl ${MATRIX_INSTALLATION} -o libtorch.zip
     unzip libtorch.zip
 else
-    #special case for Python 3.11
-    if [[ ${MATRIX_PYTHON_VERSION} == '3.11' ]]; then
-        conda create -y -n ${ENV_NAME} python=${MATRIX_PYTHON_VERSION}
-        conda activate ${ENV_NAME}
-
-        INSTALLATION=${MATRIX_INSTALLATION/"-c pytorch"/"-c malfet -c pytorch"}
-        INSTALLATION=${INSTALLATION/"pytorch-cuda"/"pytorch-${MATRIX_CHANNEL}::pytorch-cuda"}
-        INSTALLATION=${INSTALLATION/"conda install"/"conda install -y"}
-
-        eval $INSTALLATION
-        python ./test/smoke_test/smoke_test.py
-        conda deactivate
-        conda env remove -n ${ENV_NAME}
-    else
-
-
-
-        # Special case Pypi installation package, only applicable to linux nightly CUDA 11.7 builds, wheel package
-        if [[ ${TARGET_OS} == 'linux' && ${MATRIX_GPU_ARCH_VERSION} == '11.7' && ${MATRIX_PACKAGE_TYPE} == 'manywheel' && ${MATRIX_CHANNEL} != 'nightly' ]]; then
-            conda create -yp ${ENV_NAME}_pypi python=${MATRIX_PYTHON_VERSION} numpy ffmpeg
-            INSTALLATION_PYPI=${MATRIX_INSTALLATION/"cu117"/"cu117_pypi_cudnn"}
-            INSTALLATION_PYPI=${INSTALLATION_PYPI/"torchvision torchaudio"/""}
-            INSTALLATION_PYPI=${INSTALLATION_PYPI/"index-url"/"extra-index-url"}
-            conda run -p ${ENV_NAME}_pypi ${INSTALLATION_PYPI}
-            conda run -p ${ENV_NAME}_pypi python ./test/smoke_test/smoke_test.py --package torchonly
-            conda deactivate
-            conda env remove -p ${ENV_NAME}_pypi
-        fi
+    # Please note ffmpeg is required for torchaudio, see https://github.com/pytorch/pytorch/issues/96159
+    conda create -y -n ${ENV_NAME} python=${MATRIX_PYTHON_VERSION} numpy ffmpeg
+    conda activate ${ENV_NAME}
+    INSTALLATION=${MATRIX_INSTALLATION/"conda install"/"conda install -y"}
+
+    export OLD_PATH=${PATH}
+    # Workaround macos-arm64 runners. Issue: https://github.com/pytorch/test-infra/issues/4342
+    if [[ ${TARGET_OS} == 'macos-arm64' ]]; then
+        export PATH="${CONDA_PREFIX}/bin:${PATH}"
+    fi
 
-        # Please note ffmpeg is required for torchaudio, see https://github.com/pytorch/pytorch/issues/96159
-        conda create -y -n ${ENV_NAME} python=${MATRIX_PYTHON_VERSION} numpy ffmpeg
-        conda activate ${ENV_NAME}
-        INSTALLATION=${MATRIX_INSTALLATION/"conda install"/"conda install -y"}
-        eval $INSTALLATION
+    # Make sure we remove previous installation if it exists, this issue seems to affect only
+    if [[ ${MATRIX_PACKAGE_TYPE} == 'wheel' ]]; then
+        pip3 uninstall -y torch torchaudio torchvision
+    fi
+    eval $INSTALLATION
 
-        if [[ ${TARGET_OS} == 'linux' ]]; then
-            export CONDA_LIBRARY_PATH="$(dirname $(which python))/../lib"
-            export LD_LIBRARY_PATH=$CONDA_LIBRARY_PATH:$LD_LIBRARY_PATH
-            ${PWD}/check_binary.sh
-        fi
+    if [[ ${TARGET_OS} == 'linux' ]]; then
+        export CONDA_LIBRARY_PATH="$(dirname $(which python))/../lib"
+        export LD_LIBRARY_PATH=$CONDA_LIBRARY_PATH:$LD_LIBRARY_PATH
+        ${PWD}/check_binary.sh
+    fi
 
+    if [[ ${TARGET_OS} == 'windows' ]]; then
         python  ./test/smoke_test/smoke_test.py
-        conda deactivate
-        conda env remove -n ${ENV_NAME}
+    else
+        python3  ./test/smoke_test/smoke_test.py
     fi
+
+    if [[ ${TARGET_OS} == 'macos-arm64' ]]; then
+        export PATH=${OLD_PATH}
+    fi
+
+    conda deactivate
+    conda env remove -n ${ENV_NAME}
 fi

.github/scripts/validate_pipy.sh

@@ -0,0 +1,12 @@
+conda create -yp ${ENV_NAME}_pypi python=${MATRIX_PYTHON_VERSION} numpy ffmpeg
+
+if [[ ${MATRIX_CHANNEL} != "release" ]]; then
+    conda run -p ${ENV_NAME}_pypi pip3 install --pre torch --index-url "https://download.pytorch.org/whl/${MATRIX_CHANNEL}/${MATRIX_DESIRED_CUDA}_pypi_cudnn"
+    conda run -p ${ENV_NAME}_pypi pip3 install --pre torchvision torchaudio --index-url "https://download.pytorch.org/whl/${MATRIX_CHANNEL}/${MATRIX_DESIRED_CUDA}"
+else
+    conda run -p ${ENV_NAME}_pypi pip3 install torch torchvision torchaudio
+fi
+
+conda run -p ${ENV_NAME}_pypi python ./test/smoke_test/smoke_test.py
+conda deactivate
+conda env remove -p ${ENV_NAME}_pypi

.github/scripts/validate_poetry.sh

@@ -0,0 +1,27 @@
+
+conda create -y -n ${ENV_NAME}_poetry python=${MATRIX_PYTHON_VERSION} numpy ffmpeg
+conda activate ${ENV_NAME}_poetry
+curl -sSL https://install.python-poetry.org | python3 - --git https://github.com/python-poetry/poetry.git@master
+export PATH="/root/.local/bin:$PATH"
+
+poetry --version
+poetry new test_poetry
+cd test_poetry
+
+if [[ ${MATRIX_CHANNEL} != "release" ]]; then
+    # Installing poetry from our custom repo. We need to configure it before use and disable authentication
+    export PYTHON_KEYRING_BACKEND=keyring.backends.null.Keyring
+    poetry source add --priority=explicit domains "https://download.pytorch.org/whl/${MATRIX_CHANNEL}/${MATRIX_DESIRED_CUDA}"
+    poetry source add --priority=supplemental pytorch-nightly "https://download.pytorch.org/whl/${MATRIX_CHANNEL}"
+    poetry source add --priority=supplemental pytorch "https://download.pytorch.org/whl/${MATRIX_CHANNEL}/${MATRIX_DESIRED_CUDA}_pypi_cudnn"
+    poetry --quiet add --source pytorch torch
+    poetry --quiet add --source domains torchvision torchaudio
+else
+    export PYTHON_KEYRING_BACKEND=keyring.backends.null.Keyring
+    poetry --quiet add torch torchaudio torchvision
+fi
+
+python ../test/smoke_test/smoke_test.py
+conda deactivate
+conda env remove -p ${ENV_NAME}_poetry
+cd ..

.github/workflows/build-conda-images.yml

@@ -3,7 +3,12 @@ name: Build conda docker images
 on:
   push:
     branches:
-      main
+      - main
+      - release/*
+    tags:
+      # NOTE: Binary build pipelines should only get triggered on release candidate or nightly builds
+      # Release candidate tags look like: v1.11.0-rc1
+      - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
     paths:
       - conda/Dockerfile
       - 'common/*'
@@ -19,14 +24,14 @@ env:
   DOCKER_BUILDKIT: 1
   DOCKER_ID: ${{ secrets.DOCKER_ID }}
   DOCKER_TOKEN: ${{ secrets.DOCKER_TOKEN }}
-  WITH_PUSH: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' }}
+  WITH_PUSH: ${{ github.event_name == 'push' && (github.ref == 'refs/heads/main' || startsWith(github.ref, 'refs/heads/release')) }}
 
 jobs:
   build-docker:
     runs-on: ubuntu-22.04
     strategy:
       matrix:
-        cuda_version: ["11.6", "11.7", "11.8", "cpu"]
+        cuda_version: ["11.8", "12.1", "cpu"]
     env:
       CUDA_VERSION: ${{ matrix.cuda_version }}
     steps:

.github/workflows/build-libtorch-images.yml

@@ -3,7 +3,12 @@ name: Build libtorch docker images
 on:
   push:
     branches:
-      main
+      - main
+      - release/*
+    tags:
+      # NOTE: Binary build pipelines should only get triggered on release candidate or nightly builds
+      # Release candidate tags look like: v1.11.0-rc1
+      - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
     paths:
       - .github/workflows/build-libtorch-images.yml
       - libtorch/Dockerfile
@@ -21,14 +26,14 @@ env:
   DOCKER_BUILDKIT: 1
   DOCKER_ID: ${{ secrets.DOCKER_ID }}
   DOCKER_TOKEN: ${{ secrets.DOCKER_TOKEN }}
-  WITH_PUSH: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' }}
+  WITH_PUSH: ${{ github.event_name == 'push' && (github.ref == 'refs/heads/main' || startsWith(github.ref, 'refs/heads/release')) }}
 
 jobs:
   build-docker-cuda:
     runs-on: ubuntu-22.04
     strategy:
       matrix:
-        cuda_version: ["11.8", "11.7", "11.6"]
+        cuda_version: ["12.1", "11.8"]
     env:
       GPU_ARCH_TYPE: cuda
       GPU_ARCH_VERSION: ${{ matrix.cuda_version }}
@@ -44,10 +49,10 @@ jobs:
         run: |
           libtorch/build_docker.sh
   build-docker-rocm:
-    runs-on: ubuntu-22.04
+    runs-on: linux.12xlarge
     strategy:
       matrix:
-        rocm_version: ["5.3", "5.4.2"]
+        rocm_version: ["5.5", "5.6"]
     env:
       GPU_ARCH_TYPE: rocm
       GPU_ARCH_VERSION: ${{ matrix.rocm_version }}

.github/workflows/build-magma-linux.yml

@@ -30,7 +30,7 @@ jobs:
     runs-on: linux.2xlarge
     strategy:
       matrix:
-        cuda_version: ["118", "117", "116"]
+        cuda_version: ["121", "118"]
     steps:
       - name: Checkout PyTorch builder
         uses: actions/checkout@v3

.github/workflows/build-magma-windows.yml

@@ -17,7 +17,7 @@ jobs:
     runs-on: windows-2019
     strategy:
       matrix:
-        cuda_version: ["118", "117", "116"]
+        cuda_version: ["121", "118"]
         config: ["Release", "Debug"]
     env:
       CUDA_VERSION: ${{ matrix.cuda_version }}

.github/workflows/build-manywheel-images.yml

@@ -3,7 +3,12 @@ name: Build manywheel docker images
 on:
   push:
     branches:
-      main
+      - main
+      - release/*
+    tags:
+      # NOTE: Binary build pipelines should only get triggered on release candidate or nightly builds
+      # Release candidate tags look like: v1.11.0-rc1
+      - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
     paths:
       - .github/workflows/build-manywheel-images.yml
       - manywheel/Dockerfile
@@ -23,14 +28,14 @@ env:
   DOCKER_BUILDKIT: 1
   DOCKER_ID: ${{ secrets.DOCKER_ID }}
   DOCKER_TOKEN: ${{ secrets.DOCKER_TOKEN }}
-  WITH_PUSH: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' }}
+  WITH_PUSH: ${{ github.event_name == 'push' && (github.ref == 'refs/heads/main' || startsWith(github.ref, 'refs/heads/release')) }}
 
 jobs:
   build-docker-cuda:
     runs-on: ubuntu-22.04
     strategy:
       matrix:
-        cuda_version: ["11.8", "11.7", "11.6"]
+        cuda_version: ["12.1", "11.8"]
     env:
       GPU_ARCH_TYPE: cuda
       GPU_ARCH_VERSION: ${{ matrix.cuda_version }}
@@ -46,10 +51,10 @@ jobs:
         run: |
           manywheel/build_docker.sh
   build-docker-rocm:
-    runs-on: ubuntu-22.04
+    runs-on: linux.12xlarge
     strategy:
       matrix:
-        rocm_version: ["5.3", "5.4.2"]
+        rocm_version: ["5.5", "5.6"]
     env:
       GPU_ARCH_TYPE: rocm
       GPU_ARCH_VERSION: ${{ matrix.rocm_version }}

.github/workflows/validate-binaries.yml

@@ -22,10 +22,6 @@ on:
         default: ""
         required: false
         type: string
-      limit-win-builds:
-        description: "Limit windows builds to single python/cuda config"
-        default: "disable"
-        type: string
   workflow_dispatch:
     inputs:
       os:
@@ -53,11 +49,6 @@ on:
         default: ""
         required: false
         type: string
-      limit-win-builds:
-        description: "Limit windows builds to single python/cuda config"
-        default: "disable"
-        required: false
-        type: string
 
 jobs:
   win:
@@ -66,7 +57,6 @@ jobs:
     with:
       channel: ${{ inputs.channel }}
       ref: ${{ inputs.ref || github.ref }}
-      limit-win-builds: ${{ inputs.limit-win-builds }}
 
   linux:
     if:  inputs.os == 'linux' || inputs.os == 'all'

.github/workflows/validate-linux-binaries.yml

@@ -55,4 +55,14 @@ jobs:
         export ENV_NAME="conda-env-${{ github.run_id }}"
         export TARGET_OS="linux"
         eval "$(conda shell.bash hook)"
+
+        # Special case PyPi installation package. And Install of PyPi package via poetry
+        if [[ ${MATRIX_PACKAGE_TYPE} == "manywheel" ]] && \
+           ([[ ${MATRIX_GPU_ARCH_VERSION} == "12.1" && ${MATRIX_CHANNEL} != "release" ]] || \
+            [[ ${MATRIX_GPU_ARCH_VERSION} == "11.7" && ${MATRIX_CHANNEL} == "release" ]]); then
+          source ./.github/scripts/validate_pipy.sh --runtime-error-check disabled
+          source ./.github/scripts/validate_poetry.sh --runtime-error-check disabled
+        fi
+
+        # Standart case: Validate binaries
         source ./.github/scripts/validate_binaries.sh

.github/workflows/validate-nightly-binaries.yml

@@ -3,8 +3,8 @@ name: cron
 
 on:
   schedule:
-    # At 2:30 pm UTC (7:30 am PDT)
-    - cron: "30 14 * * *"
+    # At 3:30 pm UTC (8:30 am PDT)
+    - cron: "30 15 * * *"
   # Have the ability to trigger this job manually through the API
   workflow_dispatch:
   push:
@@ -17,19 +17,10 @@ on:
       - .github/workflows/validate-macos-binaries.yml
       - .github/workflows/validate-macos-arm64-binaries.yml
       - test/smoke_test/*
-  pull_request:
-    paths:
-      - .github/workflows/validate-nightly-binaries.yml
-      - .github/workflows/validate-linux-binaries.yml
-      - .github/workflows/validate-windows-binaries.yml
-      - .github/workflows/validate-macos-binaries.yml
-      - .github/workflows/validate-macos-arm64-binaries.yml
-      - .github/scripts/validate_binaries.sh
-      - test/smoke_test/*
+
 jobs:
   nightly:
     uses: ./.github/workflows/validate-binaries.yml
     with:
       channel: nightly
       os: all
-      limit-win-builds: enable

.github/workflows/validate-release-binaries.yml

@@ -3,8 +3,8 @@ name: cron
 
 on:
   schedule:
-    # At 3 am and 2 pm UTC (7 am and 8 pm PDT)
-    - cron: "0 3,14 * * *"
+    # At 3 am UTC (7 am PDT)
+    - cron: "0 3 * * *"
   # Have the ability to trigger this job manually through the API
   workflow_dispatch:
   push:
@@ -17,11 +17,19 @@ on:
       - .github/workflows/validate-macos-binaries.yml
       - .github/workflows/validate-macos-arm64-binaries.yml
       - test/smoke_test/*
+  pull_request:
+    paths:
+      - .github/workflows/validate-nightly-binaries.yml
+      - .github/workflows/validate-linux-binaries.yml
+      - .github/workflows/validate-windows-binaries.yml
+      - .github/workflows/validate-macos-binaries.yml
+      - .github/workflows/validate-macos-arm64-binaries.yml
+      - .github/scripts/validate_binaries.sh
+      - test/smoke_test/*
 
 jobs:
   release:
     uses: ./.github/workflows/validate-binaries.yml
     with:
       channel: release
       os: all
-      limit-win-builds: enable