This repository holds Dockerfiles needed to start basic example from tensorflow-serving tutorial.
First file Dockerfile.base is taken directly from tensorflow-serving repository and installs software needed to tensorflow-model-server development. Some packages are redundant but few of them are crucial to run example.
To build first image run:
docker build --tag=tensorflow-serving-base:latest -f Dockerfile.base .
To build second image run:
docker build --tag=tensorflow-serving-example:latest -f Dockerfile .
To run model server run:
docker run -p 9000:9000 tensorflow-serving-example:latest
Now you can communicate with model server on localhost:9000. Running inference is as simple as this (taken from tutorial page)
virtualenv ve
. ve/bin/activate
pip install tensorflow-serving-api
git clone git@github.com:tensorflow/serving.git
cd serving
python tensorflow_serving/example/mnist_client.py --num_tests=1000 --server=localhost:9000