Very Basic 'How to Use' Accuracy example is not runnable if I only install PyTorch; error may relate to multi-dispatch behaviour

[REIGN12]

Describe the bug
I try to run example in doc of 'How to use', I only install np and torch in my env.

from mmeval import Accuracy
import numpy as np

accuracy = Accuracy()
labels = np.asarray([0, 1, 2, 3])
preds = np.asarray([0, 2, 1, 3])
accuracy(preds, labels)
# {'top1': 0.5}

And I got the error msg:

Patch `plum.type.TypeMeta` with singleton failed, raise error: module 'plum.type' has no attribute 'TypeMeta'. The multiple dispatch speed may be slow.
Patch plum Type with hash value cache failed, raise error: cannot import name 'Dict' from 'plum.parametric' (/anaconda/envs/onemodel/lib/python3.8/site-packages/plum/parametric.py). The multiple dispatch speed may be slow.
Traceback (most recent call last):
  File "test_mmeval.py", line 8, in <module>
    accuracy(preds, labels)
  File "/anaconda/envs/onemodel/lib/python3.8/site-packages/mmeval/core/base_metric.py", line 105, in __call__
    self.add(*args, **kwargs)
  File "/anaconda/envs/onemodel/lib/python3.8/site-packages/mmeval/metrics/accuracy.py", line 191, in add
    corrects = self._compute_corrects(predictions, labels)
  File "/anaconda/envs/onemodel/lib/python3.8/site-packages/plum/function.py", line 419, in __call__
    return self._f(self._instance, *args, **kw_args)
  File "/anaconda/envs/onemodel/lib/python3.8/site-packages/plum/function.py", line 342, in __call__
    self._resolve_pending_registrations()
  File "/anaconda/envs/onemodel/lib/python3.8/site-packages/plum/function.py", line 220, in _resolve_pending_registrations
    signature = extract_signature(f, precedence=precedence)
  File "/anaconda/envs/onemodel/lib/python3.8/site-packages/plum/signature.py", line 187, in extract_signature
    for k, v in typing.get_type_hints(f).items():
  File "/anaconda/envs/onemodel/lib/python3.8/typing.py", line 1264, in get_type_hints
    value = _eval_type(value, globalns, localns)
  File "/anaconda/envs/onemodel/lib/python3.8/typing.py", line 270, in _eval_type
    return t._evaluate(globalns, localns)
  File "/anaconda/envs/onemodel/lib/python3.8/typing.py", line 518, in _evaluate
    eval(self.__forward_code__, globalns, localns),
  File "<string>", line 1, in <module>
NameError: name 'oneflow' is not defined

Indicating oneflow is not defined.
And I use my debugger try to debug, and found the problem is in accuracy.py L248

@overload  # type: ignore
@dispatch
def _compute_corrects(  # type: ignore
    self, predictions: Union['oneflow.Tensor', Sequence['oneflow.Tensor']],
    labels: Union['oneflow.Tensor',
                  Sequence['oneflow.Tensor']]) -> 'oneflow.Tensor':

when the dispatch try to decorate the _compute_corrects related to oneflow.
And I go deeper to dispatch, found that in dispatcher.py L176,

try:
    module = importlib.import_module(module_name)
    resolved_type = getattr(module, module_attr_basename)
except Exception as e: #L176
    if importable_name not in self._unimportable_types:
        logger.debug(
            f"Unimportable: '{importable_name}', raise error: {e}.")
        resolved_type = type(importable_name, (), {})
        self._unimportable_types[importable_name] = resolved_type
    else:
        resolved_type = self._unimportable_types[importable_name]
return resolved_type

this except should have caught the ModuleNotFoundError and solve this, but the error is not caught and I totally have no idea why.

Reproduction

1. What command or script did you run?

from mmeval import Accuracy
import numpy as np

accuracy = Accuracy()
labels = np.asarray([0, 1, 2, 3])
preds = np.asarray([0, 2, 1, 3])
accuracy(preds, labels)
# {'top1': 0.5}

[C1rN09]

plum-dispatch released its 2.0.0 version, which breaks something. We have temporarily fixed this issue in PR #108 and you can reinstall plum-dispatch < 2.0 to solve this for now.
We will try to figure it out and adapt to plum-dispatch >= 2.0.0 in the future.