ml_switcheroo.importers.onnx_reader
===================================

.. py:module:: ml_switcheroo.importers.onnx_reader

.. autoapi-nested-parse::

   Importer for ONNX Markdown Specifications.

   This module parses the official ONNX Operators documentation (Markdown files),
   extracting operator names, summaries, inputs, and attributes. It converts
   these definitions into the Semantic Knowledge Base format.

   Key Features:
   - **Markdown Splitting**: Identifies operators via specific HTML anchors.
   - **Input & Attribute Parsing**: Extracts definitions lists (``<dl>``).
   - **Type Extraction**: Parses HTML type signatures (e.g., ``<dt>x : T</dt>``)
     and maps them to ml-switcheroo/Fuzzer compatible type hints (e.g., ``Tensor``, ``int``).
   - **Sanitization**: Cleans HTML tags like ``<tt>``, ``<b>`` from names.



Classes
-------

.. autoapisummary::

   ml_switcheroo.importers.onnx_reader.OnnxSpecImporter


Module Contents
---------------

.. py:class:: OnnxSpecImporter

   Parses ONNX Markdown specification files into semantic JSON structures.

   This class reads Markdown files (like `Operators.md`), identifies operator
   blocks, and parses their Inputs and Attributes sections to build a rich
   function signature including type hints.


   .. py:method:: parse_file(target_file: pathlib.Path) -> Dict[str, Any]

      Parses a specific ONNX Markdown file (e.g. Operators.md).

      :param target_file: Path to the .md file to parse.

      :returns: Dictionary mapping Operator IDs (e.g., "Conv") to their semantic definition.
                The definition includes 'std_args' as a list of (name, type) tuples.