ugnelis / tensorflow-hello-serving

Simple TensorFlow Estimator 1.x example with Serving API.

TensorFlow Hello Serving

Simple example of TensorFlow 1.x serving.

Training

Model training:

$ ./train.sh

Prediction

Prediction using the trained model:

$ ./predict.sh

Development (with Docker)

Build Docker Image

In order to run this project as a Docker container, a Docker image has to be built:

$ docker build --file Dockerfile.development -t hello-development:1.0 .

Run Docker Container

Run a Docker container together with NVidia GPUs:

$ docker run -it \
    -v ${PWD}/exported_models:/workspace/exported_models \
    --gpus all \
    --name hello-development \
    hello-development:1.0 bash

In case you need to take your trained models out of the container:

$ docker cp docker-development:/workspace/models/ models/

Read more about it in Docker Hub.

Production (with Docker)

Build Docker Image

Docker Image build which contains of TensorFlow serving:

$ docker build --file Dockerfile.production -t hello-production:1.0 .

Run Docker Container

Run a Docker container:

$ docker run -it --rm \
    -v ${PWD}/exported_models:/models \
    -p 8501:8501 \
    -e MODEL_NAME=hello \
    -e MODEL_PATH=/models/hello \
    --name hello-production \
    hello-production:1.0

Send Request

When production Docker container is running, the prediction requests can be done:

[POST] "http://localhost:8501/v1/models/hello:predict"

JSON body example:

{
  "signature_name": "predict",
  "instances": [
    {
      "f1": 6.4,
      "f2": 2.9,
      "f3": 4.3,
      "f4": 1.3
    }
  ]
}