Upgrade CI and prebuild to cu118 (#2447)

* update

* fix

* update docs

* fix

* remove ort-fp16 in reg test

* fix build

* debug

* debug

* Revert "debug"

This reverts commit 2505464.

* Revert "debug"

This reverts commit 7779f0b.

* update

* load ort lib for prebuilt mmdeploy

Author: RunningLeon

.github/scripts/prepare_reg_test.py

@@ -112,8 +112,12 @@ def install_torch(torch_version):
     tv = version.parse(torch_version)
     if tv < version.parse('1.10.0'):
         cuda_int = '111'
-    elif tv >= version.parse('1.13.0'):
+    elif tv < version.parse('1.13.0'):
+        cuda_int = '113'
+    elif tv < version.parse('2.0.0'):
         cuda_int = '117'
+    else:
+        cuda_int = '118'
 
     is_torch_v2 = tv >= version.parse('2.0.0')
     if is_torch_v2:

.github/workflows/build.yml

@@ -268,12 +268,14 @@ jobs:
           COLOR: ${{ steps.badge_status.conclusion == 'success' && 'green' || 'red' }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
-  build_cuda113_linux:
+  build_cuda118_linux:
     needs: [build_cpu_model_convert, build_cpu_sdk, build_cuda117]
     runs-on: [self-hosted, linux-3090]
     container:
-      image: openmmlab/mmdeploy:ubuntu20.04-cuda11.3
+      image: openmmlab/mmdeploy:ubuntu20.04-cuda11.8
       options: "--gpus=all --ipc=host"
+      volumes:
+        - /data2/pip-cache:/root/.cache/pip
     steps:
       - name: Checkout repository
         uses: actions/checkout@v3
@@ -313,17 +315,17 @@ jobs:
         if: always()
         uses: RubbaBoy/[email protected]
         with:
-          NAME: build_cuda113_linux
+          NAME: build_cuda118_linux
           LABEL: 'build'
           STATUS: ${{ steps.badge_status.conclusion == 'success' && 'passing' || 'failing' }}
           COLOR: ${{ steps.badge_status.conclusion == 'success' && 'green' || 'red' }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
-  build_cuda113_windows:
+  build_cuda118_windows:
     needs: [build_cpu_model_convert, build_cpu_sdk, build_cuda117]
     runs-on: [self-hosted, win10-3080]
     env:
-      BASE_ENV: cuda11.3-cudnn8.2-py3.8-torch1.10
+      BASE_ENV: mmdeploy-cuda11.8-torch2.0-py38
     defaults:
       run:
         shell: powershell
@@ -356,18 +358,18 @@ jobs:
           python -V
           mkdir build
           cd build
-          cmake ..  -A x64 -T v142 `
+          cmake ..  -A x64 -T v142,cuda=$env:CUDA_PATH `
             -DMMDEPLOY_BUILD_TEST=ON `
             -DMMDEPLOY_BUILD_SDK_CSHARP_API=ON `
             -DMMDEPLOY_BUILD_SDK_PYTHON_API=ON `
             -DMMDEPLOY_BUILD_SDK=ON `
             -DMMDEPLOY_TARGET_DEVICES='cuda' `
             -DMMDEPLOY_TARGET_BACKENDS='ort;trt' `
             -DMMDEPLOY_CODEBASES='all' `
-            -Dpplcv_DIR="$env:PPLCV_DIR\pplcv-build\install\lib\cmake\ppl" `
-            -DOpenCV_DIR="$env:OPENCV_DIR\build\x64\vc15\lib" `
+            -Dpplcv_DIR="$env:pplcv_DIR" `
+            -DOpenCV_DIR="$env:OpenCV_DIR" `
             -DTENSORRT_DIR="$env:TENSORRT_DIR" `
-            -DONNXRUNTIME_DIR="$env:ONNXRUNTIME_DIR" `
+            -DONNXRUNTIME_DIR="$env:ONNXRUNTIME_GPU_DIR" `
             -DMMDEPLOY_BUILD_EXAMPLES=ON `
             -DCUDNN_DIR="$env:CUDNN_DIR"
           cmake --build . --config Release -- /m
@@ -382,21 +384,23 @@ jobs:
         run: |
           conda activate $env:TEMP_ENV
           $env:path = "$pwd\build\bin\Release;" + $env:path
+          $env:path = "$env:ONNXRUNTIME_GPU_DIR\lib;" + $env:path
+          echo $env:path
           .github\scripts\windows\test_full_pipeline.ps1 -Backend trt -Device cuda
       - name: Clear temp env
         if: always()
         run: |
           conda env remove --prefix "$env:TEMP_ENV"
 
-  badge_build_cuda113_windows:
-    needs: build_cuda113_windows
+  badge_build_cuda118_windows:
+    needs: build_cuda118_windows
     if: always()
     runs-on: ubuntu-20.04
     steps:
       - name: create badge
         uses: RubbaBoy/[email protected]
         with:
-          NAME: build_cuda113_windows
+          NAME: build_cuda118_windows
           LABEL: 'build'
           STATUS: ${{ needs.build_cuda113_windows.result == 'success' && 'passing' || needs.build_cuda113_windows.result }}
           COLOR: ${{ needs.build_cuda113_windows.result == 'success' && 'green' || 'red' }}

.github/workflows/docker.yml

@@ -20,34 +20,47 @@ on:
 
 jobs:
   publish_docker_image:
-    runs-on: [self-hosted, linux-3090]
+    runs-on: ubuntu-latest
     environment: 'prod'
     env:
-      TAG_PREFIX: openmmlab/mmdeploy:ubuntu20.04-cuda11.3-mmdeploy
+      TAG_PREFIX: openmmlab/mmdeploy:ubuntu20.04-cuda11.8-mmdeploy
     steps:
       - name: Checkout repository
         uses: actions/checkout@v3
-      - name: Get mmdeploy version
-        if: startsWith(github.ref, 'refs/tags/') == false
+      - name: Check disk space
         run: |
-          echo "MMDEPLOY_VERSION=main"  >> $GITHUB_ENV
-          echo "TAG=$TAG_PREFIX"  >> $GITHUB_ENV
-      - name: Get mmdeploy tag
-        if: startsWith(github.ref, 'refs/tags/') == true
+          df -h
+          ls /opt/hostedtoolcache
+          rm -rf ${GITHUB_WORKSPACE}/.git
+          rm -rf  /opt/hostedtoolcache/go
+          rm -rf  /opt/hostedtoolcache/node
+          rm -rf  /opt/hostedtoolcache/Ruby
+          rm -rf  /opt/hostedtoolcache/CodeQL
+          cat /proc/cpuinfo  | grep -ic proc
+          free
+          df -h
+          df . -h
+      - name: Get docker info
         run: |
-          export MMDEPLOY_VERSION=$(python3 -c "import sys; sys.path.append('mmdeploy');from version import __version__;print(__version__)")
-          echo $MMDEPLOY_VERSION
-          echo "TAG=${TAG_PREFIX}${MMDEPLOY_VERSION}"  >> $GITHUB_ENV
-          echo "MMDEPLOY_VERSION=v$MMDEPLOY_VERSION"  >> $GITHUB_ENV
-      - name: Build Docker image
-        continue-on-error: true
+          docker info
+      - name: Login to Docker Hub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Build and push the latest Docker image
         run: |
-          echo $MMDEPLOY_VERSION
+          export TAG=$TAG_PREFIX
+          echo "TAG=${TAG}" >> $GITHUB_ENV
           echo $TAG
-          docker build docker/Release/ -t ${TAG} --no-cache --build-arg MMDEPLOY_VERSION=${MMDEPLOY_VERSION}
-      - name: Push Docker image
-        continue-on-error: true
+          docker ./docker/Release/ -t ${TAG} --no-cache
+          docker push $TAG
+      - name: Push docker image with released tag
+        if: startsWith(github.ref, 'refs/tags/') == true
         run: |
+          export MMDEPLOY_VERSION=$(python3 -c "import sys; sys.path.append('mmdeploy');from version import __version__;print(__version__)")
           echo $MMDEPLOY_VERSION
-          echo $TAG
-          docker push $TAG
+          export RELEASE_TAG=${TAG_PREFIX}${MMDEPLOY_VERSION}
+          echo $RELEASE_TAG
+          docker tag $TAG $RELEASE_TAG
+          docker push $RELEASE_TAG

.github/workflows/prebuild.yml

@@ -22,10 +22,11 @@ jobs:
     if: inputs.run == true || ${{ github.event_name == 'push' }}
     runs-on: [self-hosted, linux-3090]
     container:
-      image: openmmlab/mmdeploy:manylinux2014_x86_64-cuda11.3
+      image: openmmlab/mmdeploy:manylinux2014_x86_64-cuda11.8
       options: "--gpus=all --ipc=host"
       volumes:
         - /data2/actions-runner/prebuild:/__w/mmdeploy/prebuild
+        - /data2/pip-cache:/root/.cache/pip
     steps:
       - name: Checkout repository
         uses: actions/checkout@v3
@@ -80,10 +81,11 @@ jobs:
     if: inputs.run == true || ${{ github.event_name == 'push' }}
     runs-on: [self-hosted, linux-3090]
     container:
-      image: openmmlab/mmdeploy:build-ubuntu16.04-cuda11.3
+      image: openmmlab/mmdeploy:build-ubuntu16.04-cuda11.8
       options: "--gpus=all --ipc=host"
       volumes:
         - /data2/actions-runner/prebuild:/__w/mmdeploy/prebuild
+        - /data2/pip-cache:/root/.cache/pip
     steps:
       - name: Checkout repository
         uses: actions/checkout@v3
@@ -107,7 +109,7 @@ jobs:
           cd pack
           python ../tools/package_tools/generate_build_config.py --backend 'ort;trt' \
             --system linux --output config.yml --device cuda --build-sdk --build-sdk-monolithic \
-            --sdk-dynamic-net --cxx11abi --onnxruntime-dir=$ONNXRUNTIME_GPU_DIR --cudnn-dir /usr
+            --sdk-dynamic-net --cxx11abi --onnxruntime-dir=$ONNXRUNTIME_GPU_DIR --cudnn-dir $CUDNN_DIR
           python ../tools/package_tools/mmdeploy_builder.py --config config.yml
       - name: Zip mmdeploy sdk
         run: |

.github/workflows/regression-test.yml

@@ -27,7 +27,7 @@ on:
         required: true
         description: 'Tested torch versions. Default is ["1.10.0"]'
         type: string
-        default: "['1.10.0']"
+        default: "['2.0.0']"
       models:
         required: true
         description: 'Tested model list, eg: "resnet yolov3". Default is "all".\r\n Example: resnet ssd yolov5 maskrcnn srcnn pointpillars dbnet crnn hrnet fastscnn slowfast RotatedRetinanet'
@@ -65,8 +65,8 @@ jobs:
     timeout-minutes: 4320 # 72hours
     environment: 'prod'
     container:
-      image: openmmlab/mmdeploy:ubuntu20.04-cuda11.3
-      options: "--gpus=all --ipc=host"
+      image: openmmlab/mmdeploy:ubuntu20.04-cuda11.8
+      options: "--gpus=all --ipc=host -e NVIDIA_DRIVER_CAPABILITIES=all -e DISPLAY"
       volumes:
         - /data2/checkpoints:/__w/mmdeploy/mmdeploy_checkpoints
         - /data2/benchmark:/__w/mmdeploy/data
@@ -173,7 +173,7 @@ jobs:
     environment: 'prod'
     runs-on: [self-hosted, win10-3080]
     env:
-      BASE_ENV: cuda11.3-cudnn8.2-py3.8-torch1.10
+      BASE_ENV: mmdeploy-cuda11.8-torch2.0-py38
       DATASET_DIR: D:\reg-test\data
       REGRESSION_DIR: D:\reg-test\regression_log
       CHECKPOINT_DIR: D:\reg-test\checkpoints
@@ -240,7 +240,7 @@ jobs:
           echo "BACKENDS=$env:BACKENDS" >> $env:GITHUB_ENV
           New-Item -Path build -ItemType Directory -Force
           cd build
-          cmake ..  -A x64 -T v142 `
+          cmake ..  -A x64 -T v142,cuda=$env:CUDA_PATH `
             -DMMDEPLOY_BUILD_TEST=OFF `
             -DMMDEPLOY_BUILD_SDK_CSHARP_API=ON `
             -DMMDEPLOY_BUILD_SDK_PYTHON_API=ON `

.readthedocs.yml renamed to .readthedocs.yaml

@@ -1,22 +1,24 @@
-ARG CUDA_INT=113
+ARG CUDA_INT=118
 
+FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04 AS cuda-118
 FROM nvidia/cuda:11.3.0-cudnn8-devel-ubuntu20.04 AS cuda-113
-FROM nvidia/cuda:10.2-cudnn8-devel-ubuntu18.04 AS cuda-102
 
 FROM cuda-${CUDA_INT} AS final
 
-ARG TORCH_VERSION=1.10.0
-ARG TORCHVISION_VERSION=0.11.0
+ARG TORCH_VERSION=2.0.0
+ARG TORCHVISION_VERSION=0.15.1
 
 # important dependencies
 ARG OPENCV_VERSION==4.5.4.60
 ARG PPLCV_VERSION=0.7.0
 
 # backends
-ARG ONNXRUNTIME_VERSION=1.8.1
+ARG ONNXRUNTIME_VERSION=1.15.1
 ARG PPLNN_VERSION=0.8.1
-ARG NCNN_VERSION=20221128
-ARG TENSORRT_VERSION=8.2.3.0
+ARG NCNN_VERSION=20230816
+ARG TENSORRT_VERSION=8.6.1.6
+ARG OPENVINO_VERSION=2022.3.0
+
 # tensorrt tar file url
 ARG TENSORRT_URL
 
@@ -47,21 +49,23 @@ ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
 RUN wget https://download.java.net/java/GA/jdk18/43f95e8614114aeaa8e8a5fcf20a682d/36/GPL/openjdk-18_linux-x64_bin.tar.gz &&\
     tar xvf openjdk-18_linux-x64_bin.tar.gz && rm -rf openjdk-18_linux-x64_bin.tar.gz && \
     wget https://github.com/microsoft/onnxruntime/releases/download/v${ONNXRUNTIME_VERSION}/onnxruntime-linux-x64-${ONNXRUNTIME_VERSION}.tgz &&\
-    tar -xzvf onnxruntime-linux-x64-${ONNXRUNTIME_VERSION}.tgz && rm onnxruntime-*.tgz &&\
+    tar -xzvf onnxruntime-linux-x64-${ONNXRUNTIME_VERSION}.tgz &&\
+    wget https://github.com/microsoft/onnxruntime/releases/download/v${ONNXRUNTIME_VERSION}/onnxruntime-linux-x64-gpu-${ONNXRUNTIME_VERSION}.tgz &&\
+    tar -xzvf onnxruntime-linux-x64-gpu-${ONNXRUNTIME_VERSION}.tgz && rm onnxruntime-*.tgz &&\
     wget https://github.com/Kitware/CMake/releases/download/v3.25.2/cmake-3.25.2-linux-x86_64.tar.gz &&\
     tar -xzvf cmake-3.25.2-linux-x86_64.tar.gz && rm cmake-*.tar.gz && mv cmake-* cmake &&\
     export CUDA_INT=$(echo $CUDA_VERSION | awk '{split($0, a, "."); print a[1]a[2]}') &&\
     python3 -m pip install --no-cache-dir --upgrade pip setuptools wheel &&\
     python3 -m pip install --no-cache-dir onnxruntime-gpu==${ONNXRUNTIME_VERSION} &&\
-    python3 -m pip install --no-cache-dir openvino openvino-dev[onnx] &&\
+    python3 -m pip install --no-cache-dir openvino openvino-dev[onnx]==${OPENVINO_VERSION} &&\
     python3 -m pip install --no-cache-dir opencv-python==${OPENCV_VERSION} opencv-python-headless==${OPENCV_VERSION} opencv-contrib-python==${OPENCV_VERSION} &&\
     python3 -m pip install --no-cache-dir torch==${TORCH_VERSION}+cu${CUDA_INT} torchvision==${TORCHVISION_VERSION}+cu${CUDA_INT} -f https://download.pytorch.org/whl/torch_stable.html
 
 # create env
 ENV JAVA_HOME=/root/workspace/jdk-18
 ENV PATH=$JAVA_HOME/bin:/root/workspace/cmake/bin:$PATH
-ENV ONNXRUNTIME_VERSION=${ONNXRUNTIME_VERSION}
 ENV ONNXRUNTIME_DIR=/root/workspace/onnxruntime-linux-x64-${ONNXRUNTIME_VERSION}
+ENV ONNXRUNTIME_GPU_DIR=/root/workspace/onnxruntime-linux-x64-gpu-${ONNXRUNTIME_VERSION}
 ENV LD_LIBRARY_PATH=${ONNXRUNTIME_DIR}/lib:$LD_LIBRARY_PATH
 
 ### install ppl.nn

docker/Base/Dockerfile

@@ -1,6 +1,6 @@
-FROM openmmlab/mmdeploy:ubuntu20.04-cuda11.3
+FROM openmmlab/mmdeploy:ubuntu20.04-cuda11.8
 
-ARG MMDEPLOY_VERSION
+ARG MMDEPLOY_VERSION=main
 
 ENV BACKUP_LD_LIBRARY_PATH=$LD_LIBRARY_PATH
 ENV LD_LIBRARY_PATH=/usr/local/cuda/compat:$LD_LIBRARY_PATH
@@ -20,7 +20,7 @@ RUN git clone --recursive -b $MMDEPLOY_VERSION --depth 1 https://github.com/open
     python3 -m pip install -U openmim pycuda &&\
     python3 -m mim install "mmcv>=2.0.0" &&\
     python3 -m pip install -r requirements.txt &&\
-    python3 -m pip install -e .
+    python3 -m pip install -e . --user
 
 ENV MMDeploy_DIR="/root/workspace/mmdeploy/build/install/lib/cmake/MMDeploy"
 ENV LD_LIBRARY_PATH="/root/workspace/mmdeploy/build/lib:${BACKUP_LD_LIBRARY_PATH}"

docker/Release/Dockerfile

@@ -5,19 +5,19 @@ ARG CUDA_URL
 ARG CUDNN_URL
 ARG TENSORRT_URL
 
-ARG CUDA_VERSION=11.3
+ARG CUDA_VERSION=11.8
 
 # important dependencies
 ARG OPENCV_VERSION=4.5.5
 ARG PPLCV_VERSION=0.7.0
 
 # backends
-ARG ONNXRUNTIME_VERSION=1.8.1
-ARG TENSORRT_VERSION=8.2.3.0
+ARG ONNXRUNTIME_VERSION=1.15.1
+ARG TENSORRT_VERSION=8.6.1.6
 
 # torch
-ARG TORCH_VERSION=1.10.0
-ARG TORCHVISION_VERSION=0.11.0
+ARG TORCH_VERSION=2.0.0
+ARG TORCHVISION_VERSION=0.15.0
 
 ARG TOOLSET_VERSION=7
 
@@ -105,6 +105,7 @@ RUN curl -fsSL -v -o ~/miniconda.sh -O  https://repo.anaconda.com/miniconda/Mini
     /opt/conda/bin/conda create -n mmdeploy-3.8 python=3.8 -y &&\
     /opt/conda/bin/conda create -n mmdeploy-3.9 python=3.9 -y &&\
     /opt/conda/bin/conda create -n mmdeploy-3.10 python=3.10 -y &&\
+    /opt/conda/bin/conda create -n mmdeploy-3.11 python=3.11 -y &&\
     export CUDA_INT=$(echo $CUDA_VERSION | awk '{split($0, a, "."); print a[1]a[2]}') &&\
     /opt/conda/bin/conda create -n torch${TORCH_VERSION} python=3.8 -y &&\
     /opt/conda/envs/mmdeploy-3.6/bin/pip install --no-cache-dir setuptools wheel pyyaml packaging &&\