ONNX Runtime Web is a Javascript library for running ONNX models on browsers and on Node.js.
ONNX Runtime Web has adopted WebAssembly and WebGL technologies for providing an optimized ONNX model inference runtime for both CPUs and GPUs.
The Open Neural Network Exchange (ONNX) is an open standard for representing machine learning models. The biggest advantage of ONNX is that it allows interoperability across different open source AI frameworks, which itself offers more flexibility for AI frameworks adoption.
With ONNX Runtime Web, web developers can score models directly on browsers with various benefits including reducing server-client communication and protecting user privacy, as well as offering install-free and cross-platform in-browser ML experience.
ONNX Runtime Web can run on both CPU and GPU. On CPU side, WebAssembly is adopted to execute the model at near-native speed. ONNX Runtime Web compiles the native ONNX Runtime CPU engine into WebAssembly backend by using Emscripten, so it supports most functionalities native ONNX Runtime offers, including full ONNX operator coverage, multi-threading, ONNX Runtime Quantization as well as ONNX Runtime Mobile. For performance acceleration with GPUs, ONNX Runtime Web leverages WebGL, a popular standard for accessing GPU capabilities. We are keeping improving op coverage and optimizing performance in WebGL backend.
See Compatibility and Operators Supported for a list of platforms and operators ONNX Runtime Web currently supports.
-
See Get started as a landing page for ONNX Runtime Web documentation.
-
Refer to ONNX Runtime JavaScript examples for samples and tutorials.
-
See also ONNX Runtime Web API reference for detailed API documentation.
Refer to the following links for development information:
| EPs/Browsers | Chrome/Edge (Windows) | Chrome/Edge (Android) | Chrome/Edge (MacOS) | Chrome/Edge (iOS) | Safari (MacOS) | Safari (iOS) | Firefox (Windows) | Node.js |
|---|---|---|---|---|---|---|---|---|
| WebAssembly (CPU) | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️[1] |
| WebGPU | ✔️[2] | ✔️[3] | ✔️ | ❌ | ❌ | ❌ | ❌ | ❌ |
| WebGL | ✔️[4] | ✔️[4] | ✔️[4] | ✔️[4] | ✔️[4] | ✔️[4] | ✔️[4] | ❌ |
| WebNN | ✔️[5] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
- [1]: Node.js only support single-threaded
wasmEP. - [2]: WebGPU requires Chromium v113 or later on Windows. Float16 support requires Chrome v121 or later, and Edge v122 or later.
- [3]: WebGPU requires Chromium v121 or later on Windows.
- [4]: WebGL support is in maintenance mode. It is recommended to use WebGPU for better performance.
- [5]: Requires to launch browser with commandline flag
--enable-experimental-web-platform-features.
ONNX Runtime Web currently support all operators in ai.onnx and ai.onnx.ml.
ONNX Runtime Web currently supports a subset of operators in ai.onnx operator set. See webgl-operators.md for a complete, detailed list of which ONNX operators are supported by WebGL backend.
WebGPU backend is still an experimental feature. See webgpu-operators.md for a detailed list of which ONNX operators are supported by WebGPU backend.
License information can be found here.