@tensorflow/tfjs-node

TensorFlow backend for TensorFlow.js via Node.js

This repo is under active development and is not production-ready. We are actively developing as an open source project.

Installing

TensorFlow.js for Node currently supports the following platforms:

• Mac OS X 10.12.6 (Siera) or higher
• Linux CPU (Ubuntu 16.04 or higher)
• Linux GPU (Ubuntu 16.04 or higher and Cuda 9.0 w/ CUDNN v7) (see installation instructions)

Other Linux variants might also work but this project matches core TensorFlow installation requirements.

Installing CPU TensorFlow.js for Node:

npm install @tensorflow/tfjs-node
(or)
yarn add @tensorflow/tfjs-node

Installing Linux GPU TensorFlow.js for Node:

npm install @tensorflow/tfjs-node-gpu
(or)
yarn add @tensorflow/tfjs-node-gpu

Before executing any TensorFlow.js code, load and set the backend to 'tensorflow'.

import * as tf from '@tensorflow/tfjs';

// Load the binding
import '@tensorflow/tfjs-node';

// Or if running with GPU:
import '@tensorflow/tfjs-node-gpu';

tf.setBackend('tensorflow');

Development

# Download and install JS dependencies, including libtensorflow 1.8.
yarn

# Run TFJS tests against Node.js backend:
yarn test

# Switch to GPU for local development:
yarn enable-gpu

See the demo directory that trains MNIST using TensorFlow.js with the TensorFlow C backend.

cd demo/
yarn

# Run the training script. See demo/package.json for this script.
yarn mnist

The important line to note is at the top of mnist.ts, which sets the backend to TensorFlow.

Optional: Build libtensorflow From TensorFlow source

This requires installing bazel first.

bazel build --config=monolithic //tensorflow/tools/lib_package:libtensorflow