tensorflowjs

TensorFlow.js is an open-source JavaScript library for training and deploying machine learning models in the browser and on Node.js. It brings the TensorFlow ecosystem to JavaScript, enabling client-side machine learning in web applications as well as server-side development with JavaScript.

The library provides multiple APIs to support different use cases. The core API offers tensor operations and numerical computations, while the Layers API provides a high-level interface similar to Keras for building and training models. A converter tool allows importing models trained in TensorFlow or Keras into the TensorFlow.js format, facilitating reuse of existing models. In addition, there are specialized packages for Node.js, such as @tensorflow/tfjs-node and @tensorflow/tfjs-node-gpu, which enable server-side acceleration with native bindings.

Backends and performance are a key focus of TensorFlow.js. In web browsers, computations can be accelerated using WebGL, with WebAssembly and CPU backends available in various builds. On Node.js, tfjs-node offers CPU performance, while tfjs-node-gpu leverages CUDA-enabled GPUs for faster execution. This enables a range of workloads from lightweight in-browser inference to more demanding server-side models.

Use cases include in-browser inference and model experimentation, on-device training for small models, and transfer learning by fine-tuning pretrained models. Models are typically loaded from or saved to the TensorFlow.js format, consisting of a JSON topology file and binary weight shards. The ecosystem also includes pre-trained models for tasks such as image classification and pose estimation, and supports importing models from the broader TensorFlow ecosystem.

TensorFlow.js is maintained as part of the TensorFlow project and released under an open-source license. It has an active community and is widely used to enable machine learning in web applications, offering cross-platform deployment with JavaScript.