WebNN (Web Neural Network API)
==================================

The WebNN API provides a web standard for neural network inference, enabling high-performance
machine learning directly in the browser. Firefox's implementation uses ONNX Runtime as the
backend for executing neural network operations across different hardware accelerators.

The API is exposed via ``navigator.ml`` and allows web applications to:

* Build computational graphs using common neural network operations
* Execute inference using hardware acceleration (CPU, GPU, NPU when available)
* Work with tensors for efficient data transfer
* Run pre-trained models in various formats

To enable the WebNN API, set the ``dom.ml.enabled`` preference to ``true`` in ``about:config``.

.. code-block:: javascript

  // Create a context
  const context = await navigator.ml.createContext();

  // Build a computational graph
  const builder = new MLGraphBuilder(context);
  const input = builder.input('x', { dataType: 'float32', shape: [3] });
  const output = builder.add(input, constant);
  const graph = await builder.build({ 'output': output });

  // Run inference
  await context.compute(graph,
      { 'x': inputData },
      { 'output': outputData }
  );

Learn more about the WebNN implementation:

.. toctree::
   :maxdepth: 2

   architecture

Demo Pages
----------

The ``dom/webnn/docs/`` directory also contains interactive HTML demos:

* **webnn_demo.html** - Basic WebNN operations (linear operations, matrix multiplication)
* **mobilenet_complete.html** - Full MobileNetV2 image classification with pretrained weights
* **README.md** - Demo documentation and usage instructions

See the README.md file in the docs directory for how to run these demos.

Additional Resources
--------------------

* `W3C WebNN Specification <https://www.w3.org/TR/webnn/>`_
* Test files in ``dom/webnn/test/``
* Implementation in ``dom/webnn/`` C++ source files