Mnist train and pridict

---

Run mnist example from tfserving

Clone the project

git clone https://github.com/tensorflow/serving.git
cd serving

Train in docker

Train (with 100 iterations) and export the first version of model:

tools/run_in_docker.sh -d tensorflow/serving:latest-devel python /tensorflow-serving/tensorflow_serving/example/mnist_saved_model.py --training_iteration=100 --model_version=1 /tmp/mnist

Train (with 1000 iterations) and export the second version of model:

tools/run_in_docker.sh -d tensorflow/serving:latest-devel python /tensorflow-serving/tensorflow_serving/example/mnist_saved_model.py --training_iteration=1000 --model_version=2 /tmp/mnist

You should see model created for each training in your /tmp/mnist directory:

$ ls /tmp/mnist
1  2

Serve the model

Serving with grpc port 8500 and rest API port 8501: The easiest way to serve a model is to provide the --model_name and --model_base_path flags:

docker run -p 8500:8500 -p 8501:8501 -v /tmp/mnist:/models/mnist -t --entrypoint=tensorflow_model_server tensorflow/serving:latest-devel --enable_batching --port=8500 --rest_api_port=8501 --model_name=mnist --model_base_path=/models/mnist

You may provide this configuration file using the --model_config_file flag and instruct Tensorflow Serving to periodically poll for updated versions of this configuration file at the specifed path by setting the --model_config_file_poll_wait_seconds flag.

cp $PWD/models.config /tmp/mnist
docker run -p 8500:8500 -p 8501:8501 -v /tmp/mnist:/models/mnist -t --entrypoint=tensorflow_model_server tensorflow/serving:latest-devel --enable_batching --port=8500 --rest_api_port=8501 --model_config_file=/models/mnist/models.config --model_config_file_poll_wait_seconds=60

Make Prediction with grpc request

serving/tools/run_in_docker.sh -d tensorflow/serving:latest-devel python /tensorflow-serving/tensorflow_serving/example/mnist_client.py --num_tests=1000 --server=127.0.0.1:8500 --concurrency=10

Get model metadata

curl http://localhost:8501/v1/models/mnist

---

Mnist

Train the model

docker run --rm -v $PWD/mnist:/models/mnist tensorflow/serving:latest-devel python /models/mnist/model.py

You should see model created for each training in your ./mnist directory:

ls mnist
assets  keras_metadata.pb  mnist  model.py  predict.py  sample.json  saved_model.pb  variables

Serve the model

docker run -p 8500:8500 -p 8501:8501 -v $PWD/mnist:/models/mnist -t --entrypoint=tensorflow_model_server tensorflow/serving:latest-devel --enable_batching --port=8500 --rest_api_port=8501 -model_name=mnist --model_base_path=/models/mnist

Make Prediction

cd mnist

Predict with predict.py

python predict.py

Predict with curl

curl -X POST -H "Content-Type: application/json" -d @sample.json 

---

Fashion Mnist

Train the model

docker run --rm -v $PWD/fashion_mnist:/models/mnist tensorflow/serving:latest-devel python /models/mnist/fashion_mnist.py

You should see model created for each training in your ./fashion_mnist directory:

ls fashion_mnist
assets  fashion_mnist.json  fashion_mnist_predict.py  fashion_mnist.py  keras_metadata.pb  saved_model.pb  variables

Serve the model

docker run -p 8500:8500 -p 8501:8501 -v $PWD/fashion_mnist:/models/mnist -t --entrypoint=tensorflow_model_server tensorflow/serving:latest-devel --enable_batching --port=8500 --rest_api_port=8501 -model_name=mnist --model_base_path=/models/mnist

Make Prediction

cd fashion_mnist

Predict with fashion_mnist_predict.py

python fashion_mnist_predict.py

Predict with curl

curl -X POST -H "Content-Type: application/json" -d @fashion_mnist.json