ARG PYTHON=python3 ARG PYTHON_VERSION=3.9.16 ARG PYTHON_SHORT_VERSION=3.9 ARG MAMBA_VERSION=22.11.1-2 # PyTorch Binaries ARG PT_EC2_TRAINING_URL=https://aws-pytorch-unified-cicd-binaries.s3.us-west-2.amazonaws.com/r1.13.1_ec2/20221219-193736/54406b8eed7fbd61be629cb06229dfb7b6b2954e/torch-1.13.1%2Bcu117-cp39-cp39-linux_x86_64.whl ARG PT_SM_TRAINING_URL=https://aws-pytorch-unified-cicd-binaries.s3.us-west-2.amazonaws.com/r1.13.1_sm/20230104-203046/4ad54e91be2ccb431cf26e7d7d1a2de10e8d82ac/torch-1.13.1%2Bcu117-cp39-cp39-linux_x86_64.whl ARG PT_TORCHVISION_URL=https://download.pytorch.org/whl/cu117/torchvision-0.14.1%2Bcu117-cp39-cp39-linux_x86_64.whl ARG PT_TORCHAUDIO_URL=https://download.pytorch.org/whl/cu117/torchaudio-0.13.1%2Bcu117-cp39-cp39-linux_x86_64.whl ARG PT_TORCHDATA_URL=https://download.pytorch.org/whl/test/torchdata-0.5.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl # depended by the SMP binary ARG FLASH_ATTN_VERSION=0.2.8 ARG TRITON_VERSION=2.0.0.dev20221202 # SMD binaries ARG SMD_DATA_PARALLEL_URL=https://smdataparallel.s3.amazonaws.com/binary/pytorch/1.13.1/cu117/2023-01-09/smdistributed_dataparallel-1.7.0-cp39-cp39-linux_x86_64.whl ARG SMD_MODEL_PARALLEL_URL=https://sagemaker-distributed-model-parallel.s3.us-west-2.amazonaws.com/pytorch-1.13.1/build-artifacts/2023-04-17-15-49/smdistributed_modelparallel-1.15.0-cp39-cp39-linux_x86_64.whl # ZeRO-2D Binary ARG ZERO_2D_URL=https://aws-deepspeed-zero-2d-binaries.s3.us-west-2.amazonaws.com/r1.13.1/20230412-070309/4c3ff1af262257f1636b4aafd497e0c332a1bc1d/deepspeed-0.6.1%2B4c3ff1a-py3-none-any.whl FROM nvidia/cuda:11.7.1-base-ubuntu20.04 AS base_image ENV DEBIAN_FRONTEND=noninteractive \ LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/usr/local/lib" RUN apt-get update \ && apt-get upgrade -y \ && apt-get autoremove -y \ && apt-get clean \ && rm -rf /var/lib/apt/lists/* FROM base_image AS common LABEL maintainer="Amazon AI" LABEL dlc_major_version="1" ARG PYTHON ARG PYTHON_VERSION ARG PYTHON_SHORT_VERSION ARG MAMBA_VERSION ARG CUBLAS_VERSION=11.10.3.66 ARG EFA_PATH=/opt/amazon/efa ARG CUDA_HOME=/usr/local/cuda # This arg required to stop docker build waiting for region configuration while installing tz data from ubuntu 20 ARG DEBIAN_FRONTEND=noninteractive # Python won’t try to write .pyc or .pyo files on the import of source modules # Force stdin, stdout and stderr to be totally unbuffered. Good for logging ENV PYTHONDONTWRITEBYTECODE=1 ENV PYTHONUNBUFFERED=1 ENV LD_LIBRARY_PATH="/usr/local/lib:${LD_LIBRARY_PATH}" ENV LD_LIBRARY_PATH="/opt/conda/lib:${LD_LIBRARY_PATH}" ENV PYTHONIOENCODING=UTF-8 ENV LANG=C.UTF-8 ENV LC_ALL=C.UTF-8 ENV PATH /opt/conda/bin:$PATH ENV TORCH_CUDA_ARCH_LIST="3.7 5.0 7.0+PTX 8.0" ENV TORCH_NVCC_FLAGS="-Xfatbin -compress-all" ENV CUDNN_VERSION=8.5.0.96 ENV NCCL_VERSION=2.14.3 ENV HOROVOD_VERSION=0.26.1 ENV EFA_VERSION=1.19.0 ENV OMPI_VERSION=4.1.1 ENV BRANCH_OFI=1.4.0-aws ENV GDRCOPY_VERSION=2.3.1 ENV CMAKE_PREFIX_PATH="$(dirname $(which conda))/../" ENV OPEN_MPI_PATH=/opt/amazon/openmpi ENV DGLBACKEND=pytorch ENV MANUAL_BUILD=0 ENV RDMAV_FORK_SAFE=1 ENV DLC_CONTAINER_TYPE=training RUN apt-get update \ # TODO: Remove systemd upgrade once it is updated in base image && apt-get -y upgrade --only-upgrade systemd \ && apt-get install -y --allow-change-held-packages --no-install-recommends \ build-essential \ ca-certificates \ cmake \ cuda-command-line-tools-11-7 \ cuda-cudart-11-7 \ cuda-libraries-11-7 \ curl \ emacs \ git \ hwloc \ jq \ libcublas-11-7=${CUBLAS_VERSION}-1 \ libcublas-dev-11-7=${CUBLAS_VERSION}-1 \ libcudnn8=$CUDNN_VERSION-1+cuda11.7 \ libcufft-dev-11-7 \ libcurand-dev-11-7 \ libcurl4-openssl-dev \ libcusolver-dev-11-7 \ libcusparse-dev-11-7 \ libglib2.0-0 \ libgl1-mesa-glx \ libsm6 \ libxext6 \ libxrender-dev \ libgomp1 \ libibverbs-dev \ libhwloc-dev \ libnuma1 \ libnuma-dev \ libssl1.1 \ libssl-dev \ libtool \ openssl \ python3-dev \ unzip \ vim \ wget \ zlib1g-dev \ pkg-config \ check \ libsubunit0 \ libsubunit-dev \ && rm -rf /var/lib/apt/lists/* \ && apt-get clean # Install NCCL RUN cd /tmp \ && git clone https://github.com/NVIDIA/nccl.git -b v${NCCL_VERSION}-1 \ && cd nccl \ && make -j64 src.build BUILDDIR=/usr/local \ && rm -rf /tmp/nccl # Install EFA alone without AWS OPEN_MPI RUN mkdir /tmp/efa \ && cd /tmp/efa \ && curl -O https://s3-us-west-2.amazonaws.com/aws-efa-installer/aws-efa-installer-${EFA_VERSION}.tar.gz \ && tar -xf aws-efa-installer-${EFA_VERSION}.tar.gz \ && cd aws-efa-installer \ && apt-get update \ && ./efa_installer.sh -y --skip-kmod -g \ && rm -rf $OPEN_MPI_PATH \ && rm -rf /tmp/efa \ && rm -rf /tmp/aws-efa-installer-${EFA_VERSION}.tar.gz \ && rm -rf /var/lib/apt/lists/* \ && apt-get clean # Install OpenMPI without libfabric support RUN mkdir /tmp/openmpi \ && cd /tmp/openmpi \ && wget --quiet https://download.open-mpi.org/release/open-mpi/v4.1/openmpi-${OMPI_VERSION}.tar.gz \ && tar zxf openmpi-${OMPI_VERSION}.tar.gz \ && cd openmpi-${OMPI_VERSION} \ && ./configure --enable-orterun-prefix-by-default --prefix=$OPEN_MPI_PATH --with-cuda \ && make -j $(nproc) all \ && make install \ && ldconfig \ && cd / \ && rm -rf /tmp/openmpi ENV PATH="$OPEN_MPI_PATH/bin:$EFA_PATH/bin:$PATH" ENV LD_LIBRARY_PATH=$OPEN_MPI_PATH/lib/:$EFA_PATH/lib/:$LD_LIBRARY_PATH # Install GDRCopy which is a dependency of SM Distributed DataParallel binary # The test binaries requires cuda driver library which could be found in /usr/local/cuda/lib64/stubs # So update the linker path to point to it to avoid -Lcuda not found RUN cd /tmp \ && git clone https://github.com/NVIDIA/gdrcopy.git -b v${GDRCOPY_VERSION} \ && cd gdrcopy \ && sed -ie '12s@$@ -L /usr/local/cuda/lib64/stubs/@' tests/Makefile \ && make install \ && rm -rf /tmp/gdrcopy # for conda ssl verification ENV REQUESTS_CA_BUNDLE=/etc/ssl/certs/ca-certificates.crt RUN ompi_info --parsable --all | grep mpi_built_with_cuda_support:value \ && curl -L -o ~/mambaforge.sh https://github.com/conda-forge/miniforge/releases/download/${MAMBA_VERSION}/Mambaforge-${MAMBA_VERSION}-Linux-x86_64.sh \ && chmod +x ~/mambaforge.sh \ && ~/mambaforge.sh -b -p /opt/conda \ && rm ~/mambaforge.sh \ && /opt/conda/bin/conda install -y -c conda-forge \ python=$PYTHON_VERSION \ cython \ mkl \ mkl-include \ parso \ typing \ h5py \ requests \ pyopenssl \ libgcc \ conda-content-trust \ charset-normalizer \ # needed by fastai.distributed accelerate \ # pin dgl to 0.9.1 as upstream test has not caught up, currently having accuracy error && /opt/conda/bin/conda install -c dglteam -y dgl-cuda11.7=0.9.1 \ && /opt/conda/bin/conda install -c pytorch -y magma-cuda117 \ && /opt/conda/bin/conda install -c fastai fastai \ # fastai conda has a dependency chain of fastai -> spacy -> pathy -> dataclasses # conda pathy installs dataclasses regardless of py version: https://github.com/conda-forge/pathy-feedstock/blob/1e6da43da5c8e11e09c063c79b09236bbacb7780/recipe/meta.yaml#L29 # pypi solves this issue: https://github.com/justindujardin/pathy/blob/master/requirements.txt#L3 # dataclasses package is not needed for py>=37 # install plainly will oversize the image && pip uninstall -y dataclasses \ && /opt/conda/bin/conda clean -ya # remove the default mamba installation RUN rm -rf /root/micromamba/ # Conda installs links for libtinfo.so.6 and libtinfo.so.6.2 both # Which causes "/opt/conda/lib/libtinfo.so.6: no version information available" warning # Removing link for libtinfo.so.6. This change is needed only for ubuntu 20.04-conda, and can be reverted # once conda fixes the issue RUN rm -rf /opt/conda/lib/libtinfo.so.6 RUN /opt/conda/bin/conda config --set ssl_verify False \ && pip install --no-cache-dir --upgrade pip --no-cache-dir --trusted-host pypi.org --trusted-host files.pythonhosted.org \ && ln -s /opt/conda/bin/pip /usr/local/bin/pip3 WORKDIR /root RUN pip install --no-cache-dir -U \ "awscli>1.27,<2" \ boto3 \ "click==8.1.2,<9" \ "cmake>=3.24.3,<3.25" \ "cryptography>39,<40" \ ipython \ "mpi4py>=3.1.4,<3.2" \ "opencv-python>=4.6.0,<4.7" \ packaging \ Pillow \ "psutil>=5.9.4,<5.10" \ "pyyaml>=5.4,<5.5" \ "scipy>=1.9.3,<2" COPY deep_learning_container.py /usr/local/bin/deep_learning_container.py RUN chmod +x /usr/local/bin/deep_learning_container.py RUN curl -o /license.txt https://aws-dlc-licenses.s3.amazonaws.com/pytorch-1.13/license.txt # Removing the cache as it is needed for security verification RUN rm -rf /root/.cache | true ######################################################## # _____ ____ ____ ___ # | ____/ ___|___ \ |_ _|_ __ ___ __ _ __ _ ___ # | _|| | __) | | || '_ ` _ \ / _` |/ _` |/ _ \ # | |__| |___ / __/ | || | | | | | (_| | (_| | __/ # |_____\____|_____| |___|_| |_| |_|\__,_|\__, |\___| # |___/ # ____ _ # | _ \ ___ ___(_)_ __ ___ # | |_) / _ \/ __| | '_ \ / _ \ # | _ < __/ (__| | |_) | __/ # |_| \_\___|\___|_| .__/ \___| # |_| ######################################################## FROM common AS ec2 ARG PYTHON ARG PYTHON_VERSION ARG PYTHON_SHORT_VERSION # PyTorch Binaries ARG PT_EC2_TRAINING_URL ARG PT_TORCHVISION_URL ARG PT_TORCHAUDIO_URL ARG PT_TORCHDATA_URL # Install AWS-PyTorch and other torch packages RUN pip uninstall -y torch torchvision torchaudio torchdata \ && pip install --no-cache-dir -U ${PT_EC2_TRAINING_URL} ${PT_TORCHVISION_URL} ${PT_TORCHAUDIO_URL} ${PT_TORCHDATA_URL} torchnet # Install Nvidia Apex (needs pytorch) ## Pin apex commit requested by sm-model-parallel team RUN git clone https://github.com/NVIDIA/apex && \ cd apex && \ git checkout aa756ce && \ pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./ # Configure Open MPI and configure NCCL parameters RUN mv $OPEN_MPI_PATH/bin/mpirun $OPEN_MPI_PATH/bin/mpirun.real \ && echo '#!/bin/bash' > $OPEN_MPI_PATH/bin/mpirun \ && echo "${OPEN_MPI_PATH}/bin/mpirun.real --allow-run-as-root \"\$@\"" >> $OPEN_MPI_PATH/bin/mpirun \ && chmod a+x $OPEN_MPI_PATH/bin/mpirun \ && echo "hwloc_base_binding_policy = none" >> $OPEN_MPI_PATH/etc/openmpi-mca-params.conf \ && echo "rmaps_base_mapping_policy = slot" >> $OPEN_MPI_PATH/etc/openmpi-mca-params.conf \ && echo NCCL_DEBUG=INFO >> /etc/nccl.conf \ && echo NCCL_SOCKET_IFNAME=^docker0 >> /etc/nccl.conf # Install AWS OFI NCCL plug-in RUN apt-get update && apt-get install -y autoconf RUN mkdir /tmp/efa-ofi-nccl \ && cd /tmp/efa-ofi-nccl \ && git clone https://github.com/aws/aws-ofi-nccl.git -b v${BRANCH_OFI} \ && cd aws-ofi-nccl \ && ./autogen.sh \ && ./configure --with-libfabric=/opt/amazon/efa \ --with-mpi=/opt/amazon/openmpi \ --with-cuda=/usr/local/cuda \ --with-nccl=/usr/local --prefix=/usr/local \ && make \ && make install \ && rm -rf /tmp/efa-ofi-nccl \ && rm -rf /var/lib/apt/lists/* \ && apt-get clean # Install OpenSSH for MPI to communicate between containers, allow OpenSSH to talk to containers without asking for confirmation RUN apt-get update \ && apt-get install -y --allow-downgrades --allow-change-held-packages --no-install-recommends \ && apt-get install -y --no-install-recommends openssh-client openssh-server \ && mkdir -p /var/run/sshd \ && cat /etc/ssh/ssh_config | grep -v StrictHostKeyChecking > /etc/ssh/ssh_config.new \ && echo " StrictHostKeyChecking no" >> /etc/ssh/ssh_config.new \ && mv /etc/ssh/ssh_config.new /etc/ssh/ssh_config \ && rm -rf /var/lib/apt/lists/* \ && apt-get clean # Configure OpenSSH so that nodes can communicate with each other RUN mkdir -p /var/run/sshd && \ sed 's@session\s*required\s*pam_loginuid.so@session optional pam_loginuid.so@g' -i /etc/pam.d/sshd RUN rm -rf /root/.ssh/ && \ mkdir -p /root/.ssh/ && \ ssh-keygen -q -t rsa -N '' -f /root/.ssh/id_rsa && \ cp /root/.ssh/id_rsa.pub /root/.ssh/authorized_keys \ && printf "Host *\n StrictHostKeyChecking no\n" >> /root/.ssh/config # Install Horovod (needs pytorch) RUN pip uninstall -y horovod \ && ldconfig /usr/local/cuda-11.7/targets/x86_64-linux/lib/stubs \ && HOROVOD_GPU_ALLREDUCE=NCCL HOROVOD_CUDA_HOME=/usr/local/cuda-11.7 HOROVOD_WITH_PYTORCH=1 pip install --no-cache-dir horovod==${HOROVOD_VERSION} \ && ldconfig RUN mkdir -p /etc/pki/tls/certs && cp /etc/ssl/certs/ca-certificates.crt /etc/pki/tls/certs/ca-bundle.crt WORKDIR / RUN HOME_DIR=/root \ && curl -o ${HOME_DIR}/oss_compliance.zip https://aws-dlinfra-utilities.s3.amazonaws.com/oss_compliance.zip \ && unzip ${HOME_DIR}/oss_compliance.zip -d ${HOME_DIR}/ \ && cp ${HOME_DIR}/oss_compliance/test/testOSSCompliance /usr/local/bin/testOSSCompliance \ && chmod +x /usr/local/bin/testOSSCompliance \ && chmod +x ${HOME_DIR}/oss_compliance/generate_oss_compliance.sh \ && ${HOME_DIR}/oss_compliance/generate_oss_compliance.sh ${HOME_DIR} ${PYTHON} \ && rm -rf ${HOME_DIR}/oss_compliance* \ && rm -rf /tmp/tmp* # Removing the cache as it is needed for security verification RUN rm -rf /root/.cache | true # Starts framework CMD ["/bin/bash"] ################################################################# # ____ __ __ _ # / ___| __ _ __ _ ___| \/ | __ _| | _____ _ __ # \___ \ / _` |/ _` |/ _ \ |\/| |/ _` | |/ / _ \ '__| # ___) | (_| | (_| | __/ | | | (_| | < __/ | # |____/ \__,_|\__, |\___|_| |_|\__,_|_|\_\___|_| # |___/ # ___ ____ _ # |_ _|_ __ ___ __ _ __ _ ___ | _ \ ___ ___(_)_ __ ___ # | || '_ ` _ \ / _` |/ _` |/ _ \ | |_) / _ \/ __| | '_ \ / _ \ # | || | | | | | (_| | (_| | __/ | _ < __/ (__| | |_) | __/ # |___|_| |_| |_|\__,_|\__, |\___| |_| \_\___|\___|_| .__/ \___| # |___/ |_| ################################################################# FROM common AS sagemaker LABEL maintainer="Amazon AI" LABEL dlc_major_version="1" ARG PYTHON ARG PYTHON_VERSION ARG PYTHON_SHORT_VERSION ARG RMM_VERSION=0.15.0 # The smdebug pipeline relies for following format to perform string replace and trigger DLC pipeline for validating # the nightly builds. Therefore, while updating the smdebug version, please ensure that the format is not disturbed. ARG SMDEBUG_VERSION=1.0.34 ENV SAGEMAKER_TRAINING_MODULE=sagemaker_pytorch_container.training:main # SMD model parallel and data parallel binaries ARG SMD_DATA_PARALLEL_URL ARG SMD_MODEL_PARALLEL_URL # ZeRO-2D Binary ARG ZERO_2D_URL # PyTorch Binaries ARG PT_SM_TRAINING_URL ARG PT_TORCHVISION_URL ARG PT_TORCHAUDIO_URL ARG PT_TORCHDATA_URL ARG FLASH_ATTN_VERSION ARG TRITON_VERSION # Install AWS-PyTorch and other torch packages RUN pip uninstall -y torch torchvision torchaudio torchdata \ && pip install --no-cache-dir -U ${PT_SM_TRAINING_URL} ${PT_TORCHVISION_URL} ${PT_TORCHAUDIO_URL} ${PT_TORCHDATA_URL} torchnet # Install Nvidia Apex (needs pytorch) ## Pin apex commit requested by sm-model-parallel team RUN git clone https://github.com/NVIDIA/apex && \ cd apex && \ git checkout aa756ce && \ pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./ # Configure Open MPI and configure NCCL parameters RUN mv $OPEN_MPI_PATH/bin/mpirun $OPEN_MPI_PATH/bin/mpirun.real \ && echo '#!/bin/bash' > $OPEN_MPI_PATH/bin/mpirun \ && echo "${OPEN_MPI_PATH}/bin/mpirun.real --allow-run-as-root \"\$@\"" >> $OPEN_MPI_PATH/bin/mpirun \ && chmod a+x $OPEN_MPI_PATH/bin/mpirun \ && echo "hwloc_base_binding_policy = none" >> $OPEN_MPI_PATH/etc/openmpi-mca-params.conf \ && echo "rmaps_base_mapping_policy = slot" >> $OPEN_MPI_PATH/etc/openmpi-mca-params.conf \ && echo NCCL_DEBUG=INFO >> /etc/nccl.conf \ && echo NCCL_SOCKET_IFNAME=^docker0 >> /etc/nccl.conf # Install AWS OFI NCCL plug-in RUN apt-get update && apt-get install -y autoconf RUN mkdir /tmp/efa-ofi-nccl \ && cd /tmp/efa-ofi-nccl \ && git clone https://github.com/aws/aws-ofi-nccl.git -b v${BRANCH_OFI} \ && cd aws-ofi-nccl \ && ./autogen.sh \ && ./configure --with-libfabric=/opt/amazon/efa \ --with-mpi=/opt/amazon/openmpi \ --with-cuda=/usr/local/cuda \ --with-nccl=/usr/local --prefix=/usr/local \ && make \ && make install \ && rm -rf /tmp/efa-ofi-nccl \ && rm -rf /var/lib/apt/lists/* \ && apt-get clean # Install OpenSSH for MPI to communicate between containers, allow OpenSSH to talk to containers without asking for confirmation RUN apt-get update \ && apt-get install -y --allow-downgrades --allow-change-held-packages --no-install-recommends \ && apt-get install -y --no-install-recommends openssh-client openssh-server \ && mkdir -p /var/run/sshd \ && cat /etc/ssh/ssh_config | grep -v StrictHostKeyChecking > /etc/ssh/ssh_config.new \ && echo " StrictHostKeyChecking no" >> /etc/ssh/ssh_config.new \ && mv /etc/ssh/ssh_config.new /etc/ssh/ssh_config \ && rm -rf /var/lib/apt/lists/* \ && apt-get clean # Configure OpenSSH so that nodes can communicate with each other RUN mkdir -p /var/run/sshd && \ sed 's@session\s*required\s*pam_loginuid.so@session optional pam_loginuid.so@g' -i /etc/pam.d/sshd RUN rm -rf /root/.ssh/ && \ mkdir -p /root/.ssh/ && \ ssh-keygen -q -t rsa -N '' -f /root/.ssh/id_rsa && \ cp /root/.ssh/id_rsa.pub /root/.ssh/authorized_keys \ && printf "Host *\n StrictHostKeyChecking no\n" >> /root/.ssh/config # Install Horovod (needs pytorch) RUN pip uninstall -y horovod \ && ldconfig /usr/local/cuda-11.7/targets/x86_64-linux/lib/stubs \ && HOROVOD_GPU_ALLREDUCE=NCCL HOROVOD_CUDA_HOME=/usr/local/cuda-11.7 HOROVOD_WITH_PYTORCH=1 pip install --no-cache-dir horovod==${HOROVOD_VERSION} \ && ldconfig RUN mkdir -p /etc/pki/tls/certs && cp /etc/ssl/certs/ca-certificates.crt /etc/pki/tls/certs/ca-bundle.crt WORKDIR / # Install scikit-learn and pandas RUN conda install -y -c conda-forge \ scikit-learn \ pandas WORKDIR / # Install libboost from source. This package is needed for smdataparallel functionality [for networking asynchronous IO]. RUN wget https://sourceforge.net/projects/boost/files/boost/1.73.0/boost_1_73_0.tar.gz/download -O boost_1_73_0.tar.gz \ && tar -xzf boost_1_73_0.tar.gz \ && cd boost_1_73_0 \ && ./bootstrap.sh \ && ./b2 threading=multi --prefix=/opt/conda -j 64 cxxflags=-fPIC cflags=-fPIC install || true \ && cd .. \ && rm -rf boost_1_73_0.tar.gz \ && rm -rf boost_1_73_0 \ && cd /opt/conda/include/boost WORKDIR /opt/pytorch # Copy workaround script for incorrect hostname COPY changehostname.c / COPY start_with_right_hostname.sh /usr/local/bin/start_with_right_hostname.sh RUN chmod +x /usr/local/bin/start_with_right_hostname.sh WORKDIR /root RUN pip install --no-cache-dir -U \ # disable smdebug pip install until available stable smdebug releases # smdebug==${SMDEBUG_VERSION} \ smclarify \ "sagemaker>=2,<3" \ sagemaker-experiments==0.* \ sagemaker-pytorch-training \ flash-attn==$FLASH_ATTN_VERSION \ triton==$TRITON_VERSION # Install smdebug from souce RUN cd /tmp \ && git clone https://github.com/awslabs/sagemaker-debugger --branch ${SMDEBUG_VERSION} --depth 1 --single-branch \ && cd sagemaker-debugger \ && pip install . \ && rm -rf /tmp/* # Install extra packages RUN pip install --no-cache-dir -U \ "bokeh>=3.0.1,<4" \ "imageio>=2.22,<3" \ "opencv-python>=4.6,<5" \ "plotly>=5.11,<6" \ "seaborn>=0.12,<1" \ "numba>=0.56.4,<0.57" \ "shap>=0.41,<1" # install metis RUN rm /etc/apt/sources.list.d/* \ && git clone https://github.com/KarypisLab/GKlib \ && cd GKlib \ && make config \ && make \ && make install \ && cd .. \ && git clone https://github.com/KarypisLab/METIS.git \ && cd METIS \ && make config shared=1 cc=gcc prefix=/root/local \ && make install \ && cd .. \ && rm -rf METIS GKlib \ && rm -rf /var/lib/apt/lists/* \ && apt-get clean # Install RAPIDSMemoryManager. # Requires cmake>=3.14. RUN wget -nv https://github.com/rapidsai/rmm/archive/v${RMM_VERSION}.tar.gz \ && tar -xvf v${RMM_VERSION}.tar.gz \ && cd rmm-${RMM_VERSION} \ && INSTALL_PREFIX=/usr/local ./build.sh librmm \ && cd .. \ && rm -rf v${RMM_VERSION}.tar* \ && rm -rf rmm-${RMM_VERSION} # Install ZeRO-2D Binary RUN pip install --no-cache-dir -U ${ZERO_2D_URL} # Install SM Distributed Modelparallel binary RUN pip install --no-cache-dir -U ${SMD_MODEL_PARALLEL_URL} # Install SM Distributed DataParallel binary RUN SMDATAPARALLEL_PT=1 pip install --no-cache-dir ${SMD_DATA_PARALLEL_URL} ENV LD_LIBRARY_PATH="/opt/conda/lib/python${PYTHON_SHORT_VERSION}/site-packages/smdistributed/dataparallel/lib:$LD_LIBRARY_PATH" WORKDIR / RUN HOME_DIR=/root \ && curl -o ${HOME_DIR}/oss_compliance.zip https://aws-dlinfra-utilities.s3.amazonaws.com/oss_compliance.zip \ && unzip ${HOME_DIR}/oss_compliance.zip -d ${HOME_DIR}/ \ && cp ${HOME_DIR}/oss_compliance/test/testOSSCompliance /usr/local/bin/testOSSCompliance \ && chmod +x /usr/local/bin/testOSSCompliance \ && chmod +x ${HOME_DIR}/oss_compliance/generate_oss_compliance.sh \ && ${HOME_DIR}/oss_compliance/generate_oss_compliance.sh ${HOME_DIR} ${PYTHON} \ && rm -rf ${HOME_DIR}/oss_compliance* \ && rm -rf /tmp/tmp* # Removing the cache as it is needed for security verification RUN rm -rf /root/.cache | true ENTRYPOINT ["bash", "-m", "start_with_right_hostname.sh"] CMD ["/bin/bash"]