Good question. My colleague opened that issue you referenced (and stefanondisponibile gave a terrific solution). We ended up needing a docker container with tensorflow serving + sentencepiece. Here is the dockerfile.
sentencepiece.Dockerfile
FROM ubuntu:16.04 as base_build
ARG TF_SERVING_BRANCH=r1.14
ARG TF_SERVING_COMMIT=head
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
autoconf \
automake \
build-essential \
ca-certificates \
cmake \
curl \
git \
libcurl3-dev \
libfreetype6-dev \
libgoogle-perftools-dev \
libpng-dev \
libtool \
libzmq3-dev \
make \
mlocate \
openjdk-8-jdk\
openjdk-8-jre-headless \
pkg-config \
python3-dev \
software-properties-common \
swig \
unzip \
wget \
zip \
zlib1g-dev \
gdb \
&& apt-get clean \
&& rm -rf /var/lib/apt/lists/*
ARG PROTOBUF_VERSION=3.7.0
RUN wget https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOBUF_VERSION}/protobuf-cpp-${PROTOBUF_VERSION}.tar.gz \
&& tar -xzvf protobuf-cpp-${PROTOBUF_VERSION}.tar.gz \
&& cd protobuf-${PROTOBUF_VERSION} \
&& ./configure CXXFLAGS="-D_GLIBCXX_USE_CXX11_ABI=0" \
&& make -j "$(nproc)" \
&& make install \
&& ldconfig
RUN curl -fSsL -O https://bootstrap.pypa.io/get-pip.py \
&& python3 get-pip.py \
&& rm get-pip.py
RUN pip3 --no-cache-dir install \
future>=0.17.1 \
grpcio \
h5py \
keras_applications>=1.0.8 \
keras_preprocessing>=1.1.0 \
mock \
numpy \
requests
RUN set -ex \
&& ln -s /usr/bin/python3 usr/bin/python \
&& ln -s /usr/bin/pip3 /usr/bin/pip
ENV BAZEL_VERSION 0.24.1
WORKDIR /
RUN mkdir /bazel \
&& cd /bazel \
&& curl \
-H "User-Agent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/57.0.2987.133 Safari/537.36" \
-fSsL -O https://github.com/bazelbuild/bazel/releases/download/$BAZEL_VERSION/bazel-$BAZEL_VERSION-installer-linux-x86_64.sh \
&& curl \
-H "User-Agent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/57.0.2987.133 Safari/537.36" \
-fSsL -o /bazel/LICENSE.txt https://raw.githubusercontent.com/bazelbuild/bazel/master/LICENSE \
&& chmod +x bazel-*.sh \
&& ./bazel-$BAZEL_VERSION-installer-linux-x86_64.sh \
&& cd / \
&& rm -f /bazel/bazel-$BAZEL_VERSION-installer-linux-x86_64.sh
WORKDIR /tensorflow-serving
RUN git clone --branch=${TF_SERVING_BRANCH} https://github.com/tensorflow/serving . \
&& git remote add upstream https://github.com/tensorflow/serving.git \
&& if [ "${TF_SERVING_COMMIT}" != "head" ]; then git checkout ${TF_SERVING_COMMIT} ; fi
FROM base_build as binary_build
# clone sentencepiece repo and build
RUN set -ex \
&& mkdir -p tensorflow_serving/custom_ops/sentencepiece_processor \
&& working_dir=`pwd` \
&& git clone https://github.com/google/sentencepiece.git tensorflow_serving/custom_ops/sentencepiece_processor/sentencepiece \
&& cd tensorflow_serving/custom_ops/sentencepiece_processor/sentencepiece \
&& mkdir build \
&& cd build \
&& cmake -DSPM_USE_BUILTIN_PROTOBUF=OFF -DSPM_ENABLE_TENSORFLOW_SHARED=ON .. \
&& make -j $(nproc) \
&& make install \
&& ldconfig
COPY BUILD ./BUILD
RUN cp BUILD tensorflow_serving/custom_ops/sentencepiece_processor/BUILD \
&& sed -i.bak '/@org_tensorflow\/\/tensorflow\/contrib:contrib_ops_op_lib/a\ "\/\/tensorflow_serving\/custom_ops\/sentencepiece_processor:sentencepiece_processor_ops",' \
tensorflow_serving/model_servers/BUILD \
&& sed -i '/name = "tensorflow_model_server",/a\ linkopts = ["-Wl,--allow-multiple-definition", "-Wl,-rpath,/usr/lib"],' \
tensorflow_serving/model_servers/BUILD
# Build, and install TensorFlow Serving
ARG BUILD_OPTIONS="--config=nativeopt"
ARG BAZEL_OPTIONS=""
# build bazel
RUN bazel build \
--color=yes \
--curses=yes \
${BAZEL_OPTIONS} \
--verbose_failures \
--output_filter=DONT_MATCH_ANYTHING \
${BUILD_OPTIONS} \
tensorflow_serving/model_servers:tensorflow_model_server \
&& cp bazel-bin/tensorflow_serving/model_servers/tensorflow_model_server /usr/local/bin/
CMD ["/bin/bash"]
To answer your specific question, you need to create a file called BUILD with this in it:
package(
default_visibility = ["//visibility:public"],
features = [
"-layering_check",
"-parse_headers",
],
)
licenses(["notice"]) # Apache 2, BSD, MIT
cc_library(
name = "sentencepiece_processor_ops",
srcs = [
"sentencepiece/tensorflow/sentencepiece_processor_ops.cc"
],
hdrs = [
"sentencepiece/src/sentencepiece_processor.h"
],
strip_include_prefix = "sentencepiece/src/",
alwayslink = 1,
deps =
[
"@org_tensorflow//tensorflow/core:framework",
"@org_tensorflow//tensorflow/core:lib",
"@com_google_protobuf//:protobuf_headers",
"@com_google_protobuf//:protobuf",
":sentencepiece"
]
)
cc_library(
name = "sentencepiece",
srcs = [
"sentencepiece/build/src/libsentencepiece.a",
"sentencepiece/build/src/libsentencepiece_train.a"
],
hdrs = glob(
["sentencepiece/src/*.h"]
),
deps = [
"@com_google_protobuf//:protobuf_headers",
"@com_google_protobuf//:protobuf",
],
strip_include_prefix = "sentencepiece/src/"
)
that is in the same directory as your Dockerfile. i.e.
my-docker-builds
├── BUILD
└── sentencepiece.Dockerfile
This can all be built with:
$ docker build \
-t tensorflow-serving-with-sentencepiece:local \
-f sentencepiece.Dockerfile .
Then wait for like 8 hours :)
Hope this helps.
EDIT:
Also if you're pressed for time (and not money), I spin up an 96 core machine on aws and build this and it completes in ~10 minutes.
