forward_prop fails when the input to the forward pass supposed to be a list or dict

[Erotemic]

Describe the bug
Currently if torchview detects gets a list or dict as the input to forward it assumes the user will have a forward function in their model that accepts that mapping as keyword arguments or the list as a set of positional arguments. This is not always true. For instance, I often will not collate my batches and I will pass a list of dictionaries with a lot of information for how the forward pass should procede. Here is a MWE that demonstrates this:

To Reproduce

import torch

class MyMWE(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.stem1 = torch.nn.Linear(2, 3)
        self.stem2 = torch.nn.Linear(2, 3)
        self.body = torch.nn.Linear(3, 3)
        self.head1 = torch.nn.Linear(3, 5)
        self.head2 = torch.nn.Linear(3, 7)

    def forward(self, batch):
        batch_outputs = []
        for item in batch:
            if 'domain1' in item:
                feat = self.stem1(item['domain1'])
            elif 'domain2' in item:
                feat = self.stem2(item['domain2'])
            else:
                raise ValueError

            hidden = self.body(feat)
            logits1 = self.head1(hidden)
            logits2 = self.head2(hidden)
            output = {
                'logits1': logits1,
                'logits2': logits2,
            }
            batch_outputs.append(output)
        return output

def main():
    from torchview import draw_graph
    model = MyMWE()

    batch = [
        {'domain1': torch.Tensor([1, 2])},
        {'domain2': torch.Tensor([1, 2])},
        {'domain1': torch.Tensor([1, 2])},
    ]

    # Verify a normal forward pass works
    output = model(batch)

    # Check if draw_graph works
    model_graph = draw_graph(
        model,
        input_data=batch,
        expand_nested=True,
        hide_inner_tensors=True,
        device='meta', depth=9001)
    model_graph.visual_graph.view()


if __name__ == '__main__':
    main()

Expected behavior

This errors, but if I remove the * and ** from:

                if isinstance(x, (list, tuple)):
                    _ = model.to(device)(*x, **kwargs)
                elif isinstance(x, Mapping):
                    _ = model.to(device)(**x, **kwargs)

in forward_prop it works fine and I get a nice graph like:

Additional context

I'm not sure what the correct fix for this is, but an easy fix would be to let the user pass a flag that specifies the input data should be passed in as-is without assumptions on trochview's part.