Issues with torch.compile

We are very happy with the fact that jaxtyping supports Pytorch as well, but we are currently hitting some kind of weird error/edge case and was hoping if you can give some suggestions.
When compiling a module and trying to run it we get this stacktrace:

  File "/build/work/cfc8a89b76634373e85beb2a59a94e9e781a/google3/runfiles/google3/third_party/py/torch/_dynamo/bytecode_transformation.py", [line 646](https://cs.corp.google.com/piper///depot/google3/third_party/py/torch/_dynamo/bytecode_transformation.py?l=646&ws=botev/13260&snapshot=14397), in compute_exception_table
    keys_sorted = sorted(exn_dict.keys(), key=lambda t: (t[0], -t[1]))
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch._dynamo.exc.InternalTorchDynamoError: '<' not supported between instances of 'NoneType' and 'int'

from user code:
   File "/build/work/cfc8a89b76634373e85beb2a59a94e9e781a/google3/runfiles/google3/third_party/py/jaxtyping/_decorator.py", [line 411](https://cs.corp.google.com/piper///depot/google3/third_party/py/jaxtyping/_decorator.py?l=411&ws=botev/13260&snapshot=14397), in wrapped_fn
    bound = param_signature.bind(*args, **kwargs)

Start with jaxtyping/_decorator.py around line 411, where wrapped_fn calls param_signature.bind, and compare that path with the reported torch.compile failure in torch._dynamo/bytecode_transformation.py. Reproduce the issue with a compiled module if possible; done means identifying the compatibility problem and verifying the resulting behavior with a regression test or a documented limitation.