[XNNPACK] Yolo12 model quantization

🚀 The feature, motivation and pitch

During the development of the Yolo12 detection sample, it was found that the model quantized by the XNNPACKQuantizer could not be lowered to the XNNPACK:

After installation with the XNNPACK backend:

python /examples/models/yolo12/export_and_validate.py --model_name yolo12s  --backend xnnpack --video_path cycling.mp4 --quantize

leads to the following error

Quantized succsessfully!
Traceback (most recent call last):
  File "/home/devuser/dlyakhov/executorch/examples/models/yolo12/export_and_validate.py", line 395, in <module>
    subset_size=args.subset_size,
  File "/home/devuser/dlyakhov/executorch/examples/models/yolo12/export_and_validate.py", line 263, in main
    subset_size=subset_size,
  File "/home/devuser/dlyakhov/executorch/examples/models/yolo12/export_and_validate.py", line 161, in lower_to_xnnpack
    ),
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/exir/program/_program.py", line 113, in wrapper
    return func(self, *args, **kwargs)
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/exir/program/_program.py", line 1261, in to_edge_transform_and_lower
    edge_manager = edge_manager.to_backend(method_to_partitioner)
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/exir/program/_program.py", line 113, in wrapper
    return func(self, *args, **kwargs)
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/exir/program/_program.py", line 1532, in to_backend
    new_edge_programs = to_backend(method_to_programs_and_partitioners)
  File "/usr/lib/python3.10/functools.py", line 889, in wrapper
    return dispatch(args[0].__class__)(*args, **kw)
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/exir/backend/backend_api.py", line 761, in _
    lower_all_submodules_to_backend(
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/exir/backend/backend_api.py", line 595, in lower_all_submodules_to_backend
    backend_name_to_subclass[backend_id].preprocess_multimethod(
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/exir/backend/backend_details.py", line 124, in preprocess_multimethod
    preprocess_result = cls.preprocess(program, compile_spec_for_program)
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/backends/xnnpack/xnnpack_preprocess.py", line 149, in preprocess
    assert_default_dim_order(graph_module)
  File "/home/devuser/dlyakhov/executorch/env/lib/python3.10/site-packages/executorch/backends/xnnpack/xnnpack_preprocess.py", line 94, in assert_default_dim_order
    raise RuntimeError(
RuntimeError: XNNPACK backend only supports contiguous memory format for inputs.Expecting dim_order: (0, 1, 2), but got (2, 0, 1) for a placeholder node b__frozen_param129.

The task is to support quantization of the yolo12 models family in XNNPACK backend

Alternatives

No response

Additional context

No response

RFC (Optional)

No response