docker updates

- Adding user inside docker.
- Updating installation for openfst and  sentencepiece
- Installing custom NCCL. Installing from apt is still failing.

Author: Fhrozen

docker/espnet.devel

@@ -4,9 +4,12 @@ ARG FROM_IMAGE
 FROM ${FROM_IMAGE}
 
 ARG WITH_PROXY
-ENV HTTP_PROXY ${WITH_PROXY}
-ENV HTTPS_PROXY ${WITH_PROXY}
+ARG THIS_USER
+ARG THIS_UID
+ENV http_proxy ${WITH_PROXY}
+ENV https_proxy ${WITH_PROXY}
 ENV CUDA_HOME /usr/local/cuda
+ENV CUDA_PATH ${CUDA_HOME}
 RUN DEBIAN_FRONTEND=noninteractive apt-get update && apt-get -y install --no-install-recommends \ 
         automake \
         autoconf \
@@ -24,13 +27,16 @@ RUN DEBIAN_FRONTEND=noninteractive apt-get update && apt-get -y install --no-ins
         libfreetype6-dev \
         libperl-dev \ 
         libpng12-dev \
+        libprotobuf-dev \
+        libprotobuf9v5 \
         libsndfile1 \
         libsndfile-dev \
         libtool \
         libzmq3-dev \ 
-        perl \
         pciutils \
+        perl \
         pkg-config \
+        protobuf-compiler \
         python-dev \
         python-tk \
         python-numpy-dev \ 
@@ -48,7 +54,7 @@ RUN DEBIAN_FRONTEND=noninteractive apt-get update && apt-get -y install --no-ins
         libgcc-5-dev \
         libstdc++-5-dev \
         && \
-    apt-get -y install --reinstall \
+    apt-get -y remove \
         libnccl-dev \
         libnccl2 \
         && \
@@ -60,6 +66,18 @@ RUN DEBIAN_FRONTEND=noninteractive apt-get update && apt-get -y install --no-ins
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+# Install NCCL
+ENV NCCL_HOME /usr/local/nccl
+RUN git clone https://github.com/NVIDIA/nccl.git && \
+    cd nccl && make PREFIX=${NCCL_HOME} install && \
+    cp ${NCCL_HOME}/include/* ${CUDA_HOME}/include && \
+    cp ${NCCL_HOME}/lib/* ${CUDA_HOME}/lib64
+
+# Add user to container
+RUN if [ ! -z "${THIS_UID}"  ];then \
+    useradd -m -r -u ${THIS_UID} -g root ${THIS_USER}; \
+    fi
+
 # Install pip
 RUN curl -fSsL -O https://bootstrap.pypa.io/get-pip.py && \
     python get-pip.py && \
@@ -72,13 +90,14 @@ COPY ./tools /espnet/tools
 
 WORKDIR /espnet/tools
 RUN make kaldi USE_VENV=OFF && \
-    rm -r ./kaldi/tools/openfst-1.6.5/src && \
+    rm -r ./kaldi/tools/openfst-*/src && \
     find ./kaldi/src -name "*.o" -exec rm -f {} \; && \
     find ./kaldi/src -name "*.o" -exec rm -f {} \; && \
     cd kaldi/tools && ./extras/install_beamformit.sh
 
 RUN make nkf USE_VENV=OFF && make kaldi-io-for-python USE_VENV=OFF
 RUN make venv/lib/python2.7/site-packages/torch USE_VENV=OFF && make warp-ctc USE_VENV=OFF
-RUN make chainer_ctc USE_VENV=OFF && make subword-nmt USE_VENV=OFF
-WORKDIR /
+RUN make chainer_ctc USE_VENV=OFF && make sentencepiece USE_VENV=OFF
 
+USER ${THIS_USER}
+WORKDIR /

docker/run.sh

@@ -5,6 +5,7 @@ docker_egs=
 docker_folders=
 docker_cuda=9.0
 docker_cudnn=7
+docker_user=0
 
 while test $# -gt 0
 do
@@ -80,11 +81,18 @@ if ! [[ -n ${docker_image}  ]]; then
   if [ ! -z "${HTTP_PROXY}" ]; then
     echo "Building with proxy ${HTTP_PROXY}"
     build_args="${build_args} --build-arg WITH_PROXY=${HTTP_PROXY}"
-  fi 
+  fi
+  if [ ${docker_user} -gt 0 ]; then
+    build_args="${build_args} --build-arg THIS_USER=${HOME##*/}"
+    build_args="${build_args} --build-arg THIS_UID=${UID}"
+  else
+    build_args="${build_args} --build-arg THIS_USER=root"
+  fi
+  echo "Now running docker build ${build_args} -f docker/espnet.devel -t ${image_label} ."
   (docker build ${build_args} -f docker/espnet.devel -t ${image_label} .) || exit 1
 fi
 
-vols="-v $PWD/egs:/espnet/egs -v $PWD/src:/espnet/src -v $PWD/test:/espnet/test"
+vols="-v ${PWD}/egs:/espnet/egs -v ${PWD}/src:/espnet/src -v ${PWD}/test:/espnet/test"
 if [ ! -z "${docker_folders}" ]; then
   docker_folders=$(echo ${docker_folders} | tr "," "\n")
   for i in ${docker_folders[@]}
@@ -93,9 +101,17 @@ if [ ! -z "${docker_folders}" ]; then
   done
 fi
 
+# Test if link to kaldi_io.py has been created
+if ! [[ -L ./src/utils/kaldi_io_py.py ]]; then
+  my_dir=${PWD}
+  cd ./src/utils
+  ln -s ../../tools/kaldi-io-for-python/kaldi_io.py kaldi_io_py.py
+  cd ${my_dir}
+fi
+
 cmd1="cd /espnet/egs/${docker_egs}"
 cmd2="./run.sh $@"
-#Required to access to the folder once the training if finished
+# Required to access to the folder once the training if finished
 cmd3="chmod -R 777 /espnet/egs/${docker_egs}"
 
 cmd="${cmd1}; ${cmd2}; ${cmd3}"