MMU-VisionLab / TF-Serving-MNIST

Serve an MNIST model with TF Serving.

Tensorflow Serving MNIST

This repository contain codes to train an MNIST model using CNN and build the servable model files. TF serving can be used to serve the servable model files. The instructions are as follows :

1. Run main.py with --train argument. Checkpoint files will be saved under a folder named model_ckpt.
2. Run model_builder.py. The servable model files will be saved under a folder named serve.
3. Pull the tensorflow/serving docker image with

docker pull tensorflow/serving

4. Run the docker with

docker run -t --rm -p 8500:8500 \
-v "PATH_TO_SERVEFOLDER/mnistpb":/models/mnistpb \
-e MODEL_NAME=mnistpb \
tensorflow/serving &

5. Run client.py file to get inference from the served model.

License

MIT