readme

Export and serve model on tensorflow serving

A lightweight, RESTful remote inference library for decoupling deep learning development and deployment.
Includes serving trained Keras model for pointer-based meter reading

Usage

1. Follow instructions in save_keras_model.py to export model to SavedModel/1 folder. run python save_keras_model.py --output_dir $(path) --model_version $(model_version) in your bash to specify the output directory and model version for tensorflow serving

2. Make sure tensorflow/serving docker image is pulled by running docker image in your bash and check if tensorflow/serving is pulled. If not, run docker pull tensorflow/serving. It will take a couple of minutes.

3. In your terminal, run docker run -d --name serving_base tensorflow/serving to start a docker container on local machine

4. make a diretory for saving exported model.
   mkdir -p /tmp/pointer_model
cp -r $(Path_to_SavedModel) /tmp/pointer_model

5. Run docker cp /tmp/pointer_model serving_base:/models/pointer_model to copy the exported model to docker container

6. (Optional) stop serving_base container by running docker kill serving_base

7. Run docker run -p 8501:8501 --mount type=bind,source=/tmp/pointer_model,target=/models/pointer_model -e MODEL_NAME=pointer_model -t tensorflow/serving. By running this command, this will run in the docker container tensorflow_model_server --port=8500 --rest_api_port=8501 \ --model_name=${MODEL_NAME} --model_base_path=${MODEL_BASE_PATH}/${MODEL_NAME}

8. Now you can serve your image by running in your terminal python pointer_server_client -i $(input_image_path)