Instantiating a model with an InferenceSession yields error in inference

[ir2718]

Hi,

As the title suggests, I tried instantiating an SaT model with an already downloaded ONNX model as I'm interested in using it in offline mode. This works, but doing inference with the split method will raise an error:

        def get_default_threshold(model_str: str):
            if "sm" in model_str:
                return 0.25
            if self.use_lora:
                return 0.5
            if "no-limited-lookahead" in model_str and "sm" not in model_str:
                return 0.01
            return 0.025

        default_threshold = get_default_threshold(self.model_name_or_model)

  File "myvenv/lib/python3.11/site-packages/wtpsplit/__init__.py", line 804, in _split
    default_threshold = get_default_threshold(self.model_name_or_model)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "myvenv/lib/python3.11/site-packages/wtpsplit/__init__.py", line 796, in get_default_threshold
    if "sm" in model_str:
       ^^^^^^^^^^^^^^^^^
TypeError: argument of type 'InferenceSession' is not iterable

I presume the get_default_threshold method expects a string, however this is currently not supported.

I've managed to get it working by cloning the repository from the HuggingFace hub. If this is the intended way of using models in offline mode, please briefly describe it in the documentation.

Thanks for the great work!