change base kernel

Author: hephaex

python-tensorflow/Dockerfile.2.3-py36-cuda10.1

@@ -1,4 +1,4 @@
-FROM lablup/common-base:20.07-py36-cuda10.1
+FROM lablup/common-base:py36-cuda10.1
 
 ARG TF_BUILD_VERSION=r2.3
 # Install the most recent bazel release.
@@ -80,13 +80,6 @@ RUN cp .bazelrc /root/.bazelrc
 
 ENV CI_BUILD_PYTHON ${PYTHON}
 ENV WHL_DIR=/tmp/pip3
-# Set bazel build parameters in .bazelrc in parameterized_docker_build.sh
-# Use --copt=-march values to get optimized builds appropriate for the hardware
-#   platform of your choice.
-# For ivy-bridge or sandy-bridge
-# --copt=-march="avx" \
-# For haswell, broadwell, or skylake
-# --copt=-march="avx2" \
 COPY .bazelrc /root/.mkl.bazelrc
 RUN echo "import /root/.mkl.bazelrc" >>/root/.bazelrc
 
@@ -132,33 +125,27 @@ RUN python3 -m pip --no-cache-dir install \
 
 # Clean up Bazel cache when done.
 
+RUN python3 -m pip install --extra-index-url \
+       https://developer.download.nvidia.com/compute/redist nvidia-dali-cuda100 \
+       && \
+    python3 -m pip install --extra-index-url \
+       https://developer.download.nvidia.com/compute/redist nvidia-dali-tf-plugin-cuda100
+
 # Install Horovod, temporarily using CUDA stubs
-RUN HOROVOD_WITH_TENSORFLOW=1 HOROVOD_WITHOUT_PYTORCH=1 HOROVOD_WITHOUT_MXNET=1 \
-    pip install --no-cache-dir horovod==0.19.5 && \
+RUN ldconfig /usr/local/cuda/targets/x86_64-linux/lib/stubs && \
+    HOROVOD_GPU_ALLREDUCE=NCCL HOROVOD_GPU_BROADCAST=NCCL HOROVOD_NCCL_LINK=SHARED \
+    HOROVOD_WITH_TENSORFLOW=1 HOROVOD_WITHOUT_PYTORCH=1 HOROVOD_WITHOUT_MXNET=1\
+    HOROVOD_GPU=CUDA \
+    python3 -m pip install --no-cache-dir horovod==0.21.1 && \
     ldconfig
 
 RUN python3 -m pip install --no-cache-dir \
-    	    mpi4py==3.0.3 \
-	    nni==1.6 \
-	    scikit-nni==0.2.1
-
-RUN python3 -m pip install --no-cache-dir --extra-index-url \
-    	    https://developer.download.nvidia.com/compute/redist \
-	    nvidia-dali-cuda110
-#RUN python3 -m pip install --no-cache-dir \
-#    	    --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali-tf-plugin-cuda110 
-#WORKDIR /tmp
-#RUN git clone  --recursive https://github.com/NVIDIA/DALI && \
-#    cd DALI && \
-#    mkdir build && \
-#    cd build && \
-#    cmake -D CMAKE_BUILD_TYPE=Release .. && \
-#    make -j"$(nproc)" && \
-#    cd .. && \
-#    python3 -m pip install --no-cache-dir dali/python
-    	
-
-	    
+            mpi4py==3.0.3 \
+            nni==1.9 \
+            mlflow==1.12.1 \
+            scikit-nni==0.2.1
+
+COPY ./service-defs /etc/backend.ai/service-defs
 # Install ipython kernelspec
 Run python3 -m ipykernel install --display-name "TensorFlow 2.3 on Python 3.6 & CUDA 10.1" && \
     cat /usr/local/share/jupyter/kernels/python3/kernel.json
@@ -174,7 +161,7 @@ LABEL ai.backend.kernelspec="1" \
       ai.backend.resource.min.cuda.shares=0 \
       ai.backend.runtime-type="python" \
       ai.backend.runtime-path="/usr/bin/python3" \
-      ai.backend.service-ports="ipython:pty:3000,jupyter:http:8080,jupyterlab:http:8090,vscode:http:8180,tensorboard:http:6006"
+      ai.backend.service-ports="ipython:pty:3000,jupyter:http:8081,jupyterlab:http:8090,vscode:http:8180,tensorboard:http:6006,mlflow-ui:preopen:5000,nniboard:preopen:8080"
 
 WORKDIR /home/work
 # vim: ft=dockerfile