update docker file

.github/workflows/docker.yml

@@ -25,16 +25,16 @@ jobs:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       -
-        name: Build and push cu113
+        name: Build and push cu116
         uses: docker/build-push-action@v3
         with:
-          context: ./docker/cu113/
+          context: ./docker/cu116/
           push: true
-          tags: dptechnology/unicore:${{ github.ref_name }}-pytorch1.11.0-cuda11.3
+          tags: dptechnology/unicore:${{ github.ref_name }}-pytorch1.12.1-cuda11.6
       -
-        name: Build and push cu116
+        name: Build and push cu116 with rdma
         uses: docker/build-push-action@v3
         with:
-          context: ./docker/cu116/
+          context: ./docker/rdma/
           push: true
-          tags: dptechnology/unicore:${{ github.ref_name }}-pytorch1.12.1-cuda11.6
+          tags: dptechnology/unicore:${{ github.ref_name }}-pytorch1.12.1-cuda11.6-rdma

.github/workflows/docker_latest.yml

@@ -25,16 +25,16 @@ jobs:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
       -
-        name: Build and push cu113
+        name: Build and push cu116
         uses: docker/build-push-action@v3
         with:
-          context: ./docker/cu113/
+          context: ./docker/cu116/
           push: true
-          tags: dptechnology/unicore:latest-pytorch1.11.0-cuda11.3
+          tags: dptechnology/unicore:latest-pytorch1.12.1-cuda11.6
       -
-        name: Build and push cu116
+        name: Build and push cu116 with rdma
         uses: docker/build-push-action@v3
         with:
-          context: ./docker/cu116/
+          context: ./docker/rdma/
           push: true
-          tags: dptechnology/unicore:latest-pytorch1.12.1-cuda11.6
+          tags: dptechnology/unicore:latest-pytorch1.12.1-cuda11.6-rdma

docker/cu113/Dockerfile → docker/rdma/Dockerfile

-FROM nvidia/cuda:11.3.0-cudnn8-devel-ubuntu20.04
-
-ENV LANG C.UTF-8
-ENV OFED_VERSION=5.3-1.0.0.1
+FROM nvcr.io/nvidia/pytorch:22.04-py3
 
 RUN APT_INSTALL="apt-get install -y --no-install-recommends" && \
     rm -rf /var/lib/apt/lists/* \
@@ -52,58 +49,24 @@ RUN APT_INSTALL="apt-get install -y --no-install-recommends" && \
         debhelper \
         nfs-common
 
-# ==================================================================
-# InfiniBand & RDMA
-# ------------------------------------------------------------------
-
-RUN cd /tmp && \
-    wget -q http://content.mellanox.com/ofed/MLNX_OFED-${OFED_VERSION}/MLNX_OFED_LINUX-${OFED_VERSION}-ubuntu20.04-x86_64.tgz && \
-    tar xzf MLNX_OFED_LINUX-${OFED_VERSION}-ubuntu20.04-x86_64.tgz && \
-    MLNX_OFED_LINUX-${OFED_VERSION}-ubuntu20.04-x86_64/mlnxofedinstall --user-space-only --without-fw-update --force --all && \
-    rm -rf /tmp/MLNX_OFED_LINUX-${OFED_VERSION}*
-
-
-RUN cd /tmp && \
-    mkdir -p /usr/local/nccl-rdma-sharp-plugins && \
-    DEBIAN_FRONTEND=noninteractive apt install -y zlib1g-dev && \
-    git clone --depth=1 https://github.com/Mellanox/nccl-rdma-sharp-plugins.git && \
-    cd nccl-rdma-sharp-plugins && \
-    ./autogen.sh && \
-    ./configure --prefix=/usr/local/nccl-rdma-sharp-plugins --with-cuda=/usr/local/cuda && \
-    make && \
-    make install
-
-# ==================================================================
-# python
-# ------------------------------------------------------------------
-
-# Set timezone
-RUN ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime
-ENV PATH /usr/local/nvidia/bin:/usr/local/nvidia/lib64:$PATH
-ENV LD_LIBRARY_PATH /usr/local/nvidia/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
-ENV PYTHON_VERSION=3.8
-RUN wget -O ~/miniconda.sh https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh  && \
-     chmod +x ~/miniconda.sh && \
-     ~/miniconda.sh -b -p /opt/conda && \
-     rm ~/miniconda.sh
-ENV PATH /opt/conda/bin:$PATH
-
-RUN conda install -y python=3.8  && conda clean -ya
+RUN  pip uninstall -y torch && \
+pip uninstall -y torch && \
+conda clean -ya
 
 RUN conda install -y scipy scikit-learn pyyaml tensorboard tensorboardX && \
     conda clean -ya
 
-RUN ldconfig
+# RUN ldconfig
 
-# ==================================================================
-# pytorch
-# ------------------------------------------------------------------
+# # ==================================================================
+# # pytorch
+# # ------------------------------------------------------------------
 ENV TORCH_CUDA_ARCH_LIST "7.0;7.5;8.0"
 
 RUN conda install -y numpy pyyaml scipy ipython mkl mkl-include ninja cython typing pandas && \
     conda clean -ya
 
-RUN conda install pytorch=1.11.0 cudatoolkit=11.3 -c pytorch && \
+RUN conda install pytorch=1.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge && \
     conda clean -ya
 
 RUN cd /tmp && \
@@ -114,13 +77,9 @@ RUN cd /tmp && \
 
 RUN pip install --no-cache-dir tokenizers lmdb biopython ml-collections timeout-decorator urllib3 tree dm-tree
 
-ENV LD_LIBRARY_PATH=/usr/local/nccl-rdma-sharp-plugins/lib:$LD_LIBRARY_PATH
-ENV PATH=/usr/mpi/gcc/openmpi-4.1.0rc5/bin:$PATH
-ENV LD_LIBRARY_PATH=/usr/mpi/gcc/openmpi-4.1.0rc5/lib:$LD_LIBRARY_PATH
 
 RUN ldconfig && \
     apt-get clean && \
     apt-get autoremove && \
     rm -rf /var/lib/apt/lists/* /tmp/* && \
     conda clean -ya
-