Skip to content

Latest commit

 

History

History
44 lines (30 loc) · 1.51 KB

README.md

File metadata and controls

44 lines (30 loc) · 1.51 KB
Raw

tensorflow-serving-playground

A playground of Tensorflow Serving

Export Tensorflow models

$ pip install -r requirements.txt
$ python export/mnist_tensorflow.py --model_version=1 models/mnist

(WIP)VGG - Keras

$ python export/keras_vgg.py models/vgg

NOTE: Waiting for a new Tensorflow release with this problem fixed.

Run model server

$ docker build --pull -t tensorflow-model-server -f Dockerfile .
$ docker run -d -v $(pwd):/tmp -p 9000:9000 tensorflow-model-server tensorflow_model_server --port=9000 --model_name=mnist --model_base_path=/tmp/models/mnist

Request to prefictions

You need to setup grpc/grpc-go.

$ git clone --recursive https://github.com/tensorflow/serving.git
$ protoc -I=serving -I serving/tensorflow --go_out=plugins=grpc:$GOPATH/src serving/tensorflow_serving/apis/*.proto
$ protoc -I=serving/tensorflow --go_out=plugins=grpc:$GOPATH/src serving/tensorflow/tensorflow/core/framework/*.proto
$ protoc -I=serving/tensorflow --go_out=plugins=grpc:$GOPATH/src serving/tensorflow/tensorflow/core/protobuf/{saver,meta_graph}.proto
$ protoc -I=serving/tensorflow --go_out=plugins=grpc:$GOPATH/src serving/tensorflow/tensorflow/core/example/*.proto

You can use myleott/mnist_png

$ go run serving_client/mnist/client.go --serving-address localhost:9000 mnist_png/1/1039.png