tensorflow-serving-client

README

tensorflow-serving-client

A prebuilt tensorflow serving client from the tensorflow serving proto files.

Currently supported build: C++, Java, Python, Node, Go, Mono, Rust.

Check tensorflow serving project for details: https://tensorflow.github.io/serving/

update protos

Set desired version in the VERSION file and invoke ./update.sh.

build libraries

Grpc tools are needed for building variant packages.

• Install protobuf-compiler-grpc and libprotobuf-dev on Ubuntu
• Install grpc and protobuf on macOS

See Dockerfile for details.

NOTE: grpc@1.35 and protobuf@3.14 are required

Target	Command	Outputs	Artifacts
native	gradle cmake	build/cmake
java	gradle java	build/libs
python	gradle wheel	build/dist
node	gradle node	build/node
go	gradle golang	build/go
mono	gradle mono	build/mono
rust	gradle rust	build/cargo

using docker

Invoke ./build.sh instead of gradle to build artifacts using the docker.

The building container runs as user gradle whose UID:GID is 1000:1000. It goes well given that the UID:GID pair of the user matches the one in the host OS.

Otherwise you have to configure it manually. There are two approaches you can take either:

1. Create a new user in the host OS to match the UID:GID pair of gradle.
2. Make the dirs .gradle, build and obj in the project root to be world-writable.

known issues

• protobuf 3.12.3 is buggy (protocolbuffers/protobuf#7683)