tfjs: a hardware-accelerated JavaScript library for training and deploying machine learning models in the browser and Node.js

What it solves

TensorFlow.js allows developers to train and deploy machine learning models directly in the browser or in Node.js, removing the need for a dedicated backend server for every ML single task. It enables the use of client-side data (like sensor data) for retraining models without sending data to a server.

How it works

It provides a set of APIs for building models from scratch using a low-level linear algebra library or a high-level Layers API (similar to Keras). It supports multiple hardware-accelerated backends to ensure performance, including WebGL, WebGPU, WASM, and a native C++ adapter for Node.js. Additionally, it includes tools to convert and import pre-existing TensorFlow and Keras models for use in JavaScript environments.

Who it’s for

JavaScript developers who want to integrate machine learning into web applications, Node.js server-side applications, or React Native mobile apps.

Highlights

• Multi-platform support: Works in the browser (via WebGL, WebGPU, WASM) and Node.js.
• Flexible APIs: Offers both low-level numerical computation and high-level model building.
• Model Conversion: Ability to import and run pre-trained TensorFlow SavedModels and Keras models.
• Client-side training: Support for retraining models using local sensor or client-side data.
• Visualization: Includes a dedicated library for in-browser model visualization.