:orphan:

Examples gallery
================

These examples execute both custom-op directions using explicit library
paths. They do not implement or register an execution provider.

Build the native libraries first and set ``ONNX_LIGHT_ORT_PLUGIN_LIBRARY``,
``ONNX_LIGHT_ORT_HOST_LIBRARY`` and ``ONNX_LIGHT_ORT_INCOMING_CUSTOM_OP_LIBRARY``
to their absolute paths. Models are built and checked with ``onnx_light.onnx``,
not the ``onnx`` package. See :doc:`../development` for the commands and the
build-tree ``PYTHONPATH``.


.. raw:: html

  <div id='sg-tag-list' class='sphx-glr-tag-list'></div>


.. raw:: html

    <div class="sphx-glr-thumbnails">

.. thumbnail-parent-div-open

.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="The real ORT host loads native LeakyRelu and optimized CPU Log custom ops. The model calls the custom domain explicitly; no EP is involved.">

.. only:: html

  .. image:: /auto_examples/images/thumb/sphx_glr_plot_custom_ops_light_called_by_ort_thumb.png
    :alt:

  :doc:`/auto_examples/plot_custom_ops_light_called_by_ort`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">Custom ops: onnx-light called by ORT</div>
    </div>


.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="The standalone onnx-light host loads an independent ORT LeakyRelu custom-op library, compiled using only the public ORT headers and its own numerical code. It then runs the optimized onnx_light_cpu::Log through the same ORT custom-op ABI. This script does not import or load ONNX Runtime.">

.. only:: html

  .. image:: /auto_examples/images/thumb/sphx_glr_plot_custom_ops_ort_called_by_light_thumb.png
    :alt:

  :doc:`/auto_examples/plot_custom_ops_ort_called_by_light`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">Custom ops: ORT called by onnx-light</div>
    </div>


.. raw:: html

    <div class="sphx-glr-thumbcontainer" tooltip="Compare synchronous CPU calls to Log using NumPy, ORT&#x27;s built-in operator, and the onnx_light_cpu custom operator (not an execution provider). Two graph topologies are used: X -&gt; Log -&gt; Y and X -&gt; Log -&gt; ... -&gt; Log -&gt; Y with ten consecutive nodes. Both use float32, dynamic one-dimensional inputs and models built with onnx_light.onnx.">

.. only:: html

  .. image:: /auto_examples/images/thumb/sphx_glr_plot_benchmark_log_custom_ops_thumb.png
    :alt:

  :doc:`/auto_examples/plot_benchmark_log_custom_ops`

.. raw:: html

      <div class="sphx-glr-thumbnail-title">Benchmark Log: built-in and custom operators</div>
    </div>


.. thumbnail-parent-div-close

.. raw:: html

    </div>


.. toctree::
   :hidden:

   /auto_examples/plot_custom_ops_light_called_by_ort
   /auto_examples/plot_custom_ops_ort_called_by_light
   /auto_examples/plot_benchmark_log_custom_ops



.. only:: html

 .. rst-class:: sphx-glr-signature

    `Gallery generated by Sphinx-Gallery <https://sphinx-gallery.github.io>`_