environments responsibleai vision

responsibleai-vision

Overview

AzureML Responsible AI Vision environment.

Version: 11

Tags

OS : Ubuntu20.04 Training Preview

View in Studio: https://ml.azure.com/registries/azureml/environments/responsibleai-vision/version/11

Docker image: mcr.microsoft.com/azureml/curated/responsibleai-vision:11

Docker build context

Dockerfile

FROM mcr.microsoft.com/azureml/openmpi4.1.0-ubuntu20.04:20241215.v1
# Install OpenCV native dependencies
RUN apt-get update
RUN apt-get install -y python3-opencv
RUN apt-get remove -y libavutil56 libswresample3 libavformat58
RUN apt-get remove -y libavcodec58 libswscale5 libopenexr24 libpmix2

ENV AZUREML_CONDA_ENVIRONMENT_PATH /azureml-envs/responsibleai-vision

# Prepend path to AzureML conda environment
ENV PATH $AZUREML_CONDA_ENVIRONMENT_PATH/bin:$PATH

# Create conda environment
COPY conda_dependencies.yaml .
RUN conda env create -p $AZUREML_CONDA_ENVIRONMENT_PATH -f conda_dependencies.yaml -q && \
    rm conda_dependencies.yaml && \
    conda run -p $AZUREML_CONDA_ENVIRONMENT_PATH pip cache purge && \
    conda clean -a -y

RUN conda list
# Conda install and pip install could happen side by side. Remove crypytography with vulnerability from conda
RUN conda remove cryptography

RUN pip install --pre 'azure-ai-ml' 'azure-storage-blob<=12.13.0'

# Separating RAI install due to very old statsmodels dependency
# which conflicts with econml and numpy
RUN pip install 'responsibleai~=0.36.0' \
                'raiwidgets~=0.36.0' \
                'vision_explanation_methods' \
                'responsibleai-vision==0.3.8'

# Install azureml packages
RUN pip install 'azureml-core==1.59.0' \
                'azureml-mlflow==1.59.0' \
                'azureml-rai-utils==0.0.6'

RUN pip install 'azureml-dataset-runtime==1.59.0' \
                'azureml-automl-dnn-vision==1.59.0'

RUN pip install 'shap==0.41.0' \
                'scikit-learn>=1.5.1' \
                'interpret-community==0.31.0' \
                'Pillow>=10.3.0' \
                'setuptools>=65.5.1' \
                'numpy==1.26.2' \
                'scipy==1.10.0' \
                'statsmodels==0.14.0' \
                'gunicorn>=22.0.0'

# azureml-dataset-runtime[fuse] upper bound for pyarrow is 11.0.0
# so we install pyarrow in extra step to avoid conflict
RUN pip install 'pyarrow>=14.0.1'

# To resolve vulnerability issue regarding crytography
RUN pip install 'cryptography>=43.0.1'

# To resolve vulnerability issue
RUN pip install 'Werkzeug>=3.0.3' \
                'tqdm>=4.66.3' \
                'onnx>=1.17.0'

# To resolve CVE-2024-5480 vulnerability issue for torch < 2.2.2
RUN pip install 'torch>=2.2.2'
# To resolve vulnerability issue
RUN pip install 'opencv-python-headless==4.8.1.78'

RUN pip freeze

# This is needed for mpi to locate libpython
ENV LD_LIBRARY_PATH $AZUREML_CONDA_ENVIRONMENT_PATH/lib:$LD_LIBRARY_PATH