Paddle.js 1.0 Released: Browser‑Based Deep Learning Framework

Paddle.js 1.0 has been released as the web‑focused subproject of Baidu's Paddle ecosystem, providing an open‑source deep‑learning framework that runs entirely in browsers supporting WebGL (both 1.0 and 2.0), including desktop browsers like Chrome, Firefox, Safari and mobile browsers such as Baidu App and QQ Browser.

Key compatibility features include support for both NCHW and NHWC model data formats and the ability to execute on GPU via WebGL, eliminating the need for server‑side computation.

The framework allows developers to deploy trained deep‑learning models in the browser, leveraging client‑side GPU acceleration for fast and secure inference. It also processes multimedia inputs (e.g., images) to generate the required tensors for online inference engines.

Paddle.js includes a model conversion tool that transforms Paddle‑Fluid models into a browser‑friendly format, applying optimizations such as operator fusion. Basic operators are already supported, with plans to add more in future releases.

Demo models bundled with the release cover MobileNetV2, TinyYoloV3, portrait segmentation, anti‑terrorism detection, and hand‑gesture recognition. The runtime supports warm‑up, repeated execution, resource reuse, and chaining of multiple models for sequential inference.

Installation commands:

pip install -f https://paddlepaddle.org.cn/pip/oschina/cpu paddlepaddle

pip install -f https://paddlepaddle.org.cn/pip/oschina/gpu paddlepaddle-gpu

For more information about Paddle, refer to the official website.