Remote Tensorflow Execution (RTF)

Bring the power of the Tensorflow Python API to any language, using gRPC.

This repository contains the code of the RTF server and client generator.

WARNING: experimental, unstable, WIP.

Server

The gRPC server implements the server-side protocol. It accepts the requests of function definitions and executes them. The execution is left to the Python interpreter and the output is streamed back to the client.

Usage

The server waits for messages and sends back the responses.

Requirements

Compile the proto definition is required, however the setup.py script takes care of everything. There are 2 options

1. pip install -e . to work in edit mode, with the compiled protobufs
2. pip install . to install the package (it compiles the proto too)
3. (future) pip install rtf | pip install rtf-gpu

1. Start the gRPC server

python -m rtf.server

Client stub generation

To generate the stub of a client in DEST_DIR use the rtf.generate module.

python -m rtf.generate  --dest_dir DEST_DIR --module tensorflow --target Go

Right now the only generator that is going to be developed is Go.