Logic for reading input for the model, augmented model have the same
            input with the original model.
        session_options: Optional OnnxRuntime session options for controlling model run.
            By default graph optimization is turned off
        execution_providers: Collection of execution providers for running the model.
            Only CPU EP is used by default.

    Returns:
        A dictionary where the key is tensor name and values are list of tensors from each batch
    NZ