graph\_utils
============

.. List the submodules





.. Autodoc anything defined in the module itself

   TODO: WE DON'T USE THIS OPTION RIGHT NOW BUT WE CAN REACTIVATE IF WANTED
   We use :ignore-module-all: so sphinx does not document the same module twice, even if it is reimported
   For reimports that should be documented somewhere other than where they are defined, the re-imports
   __module__ should be manually overridden -- i.e. in the ``__init__.py`` which contains ``from xxx import YYY``,
   add in ``YYY.__module__ = __name__``.

.. automodule:: modelopt.onnx.quantization.graph_utils
   :members:
   :undoc-members:

   .. Also show members without docstrings. Only members from __all__ are considered as per conf.py
   .. Ideally we should add docstrings for these members.


   .. Overview table of available classes in the module
   
   
   


   .. Overview table of available functions in the module
   
   
   .. rubric:: Functions

   .. autosummary::
      :nosignatures:
   
      add_fp16_fp32_cast
      build_non_residual_input_map
      classify_partition_nodes
      convert_fp16_io
      expand_node_names_from_patterns
      filter_quantizable_kgen_heads
      find_mha_partitions
      find_nodes_from_matmul_to_exclude
      find_nodes_from_mha_to_exclude
      find_nodes_to_exclude
      get_concat_eliminated_tensors
      get_extended_model_outputs
      get_fusible_backbone
      get_resize_scales
      get_tensor_consumer_nodes
      get_tensor_producer_nodes
      has_const_input
      has_path_type
      insert_fp8_mha_casts
      insert_matmul_casts
      is_const_input
      match_fp8_mha_pattern
      print_stat
      remove_partial_input_qdq
      replace_resize_scales