_new_from_pretrained does not patch from_pretrained correctly with Transformers v4.50

[dragoneye-alex]

Describe the bug

The patch doesn't work now because the function signature for from_pretrained changed in Transformers v4.50 with huggingface/transformers#36033

Steps/Code to reproduce bug

Import modelopt.torch.opt.plugins.huggingface, and then try to load a pretrained model from transformer.

We get the following error:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/conda/envs/pytorch/lib/python3.11/site-packages/modelopt/torch/opt/plugins/huggingface.py", line 84, in _new_from_pretrained
    model = types.MethodType(cls._modelopt_cache["from_pretrained"].__func__, cls)(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/pytorch/lib/python3.11/site-packages/transformers/modeling_utils.py", line 272, in _wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/pytorch/lib/python3.11/site-packages/transformers/modeling_utils.py", line 4455, in from_pretrained
    ) = cls._load_pretrained_model(
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: _new__load_pretrained_model() missing 1 required positional argument: 'pretrained_model_name_or_path'

Expected behavior

It should work.

[ishan-modi]

It should be available in next release - #165

[kevalmorabia97]

nvidia-modelopt==0.27.0 is released now which works with transformers 4.50 as well