Failure at lr_find and fit_one_cycle with an input class inherited from TensorImage

fastai
1

I experienced failure at lr_find and fit_one_cycle with an input class that is inherited from TensorImage.
I had made a class that inherits from TensorImage that add a few class members and factory functions and a show function:

class ImageArray(TensorImage):
    sample_seq_len = 10
    
    @classmethod 
    def create(cls, seq_path, glob_pattern='*rgb.png', rotate=False, size=128): 
    ...

    @classmethod
    def from_images(cls, imgs):
        return cls(torch.stack((ToTensor()(img)/255.0 for img in imgs)))
    
    def show(self, ctx=None, **kwargs): 
        ...

lr_find and fit_one_cycle work well (but slow) when num_workers=0 but if I leave it unset or at any number>0 I see a bunch of error messaging and the call fails. things like:

Traceback (most recent call last):
  File "/usr/lib/python3.6/multiprocessing/queues.py", line 234, in _feed
    obj = _ForkingPickler.dumps(obj)
Traceback (most recent call last):
  File "/usr/lib/python3.6/multiprocessing/queues.py", line 234, in _feed
    obj = _ForkingPickler.dumps(obj)
Traceback (most recent call last):
  File "/usr/lib/python3.6/multiprocessing/reduction.py", line 51, in dumps
    cls(buf, protocol).dump(obj)
  File "/usr/lib/python3.6/multiprocessing/queues.py", line 234, in _feed
    obj = _ForkingPickler.dumps(obj)
  File "/usr/local/lib/python3.6/dist-packages/fastai/torch_core.py", line 313, in __reduce_ex__
    args = (type(self), self.storage(), self.storage_offset(), tuple(self.size()), self.stride())
TypeError: 'int' object is not callable
  File "/usr/lib/python3.6/multiprocessing/reduction.py", line 51, in dumps
    cls(buf, protocol).dump(obj)
TypeError: 'int' object is not callableTraceback (most recent call last):
  File "/usr/lib/python3.6/multiprocessing/queues.py", line 234, in _feed
    obj = _ForkingPickler.dumps(obj)
Traceback (most recent call last):
  File "/usr/lib/python3.6/multiprocessing/queues.py", line 234, in _feed
    obj = _ForkingPickler.dumps(obj)
Traceback (most recent call last):
  File "/usr/lib/python3.6/multiprocessing/reduction.py", line 51, in dumps
    cls(buf, protocol).dump(obj)
  File "/usr/lib/python3.6/multiprocessing/queues.py", line 234, in _feed
    obj = _ForkingPickler.dumps(obj)
  File "/usr/local/lib/python3.6/dist-packages/fastai/torch_core.py", line 313, in __reduce_ex__
    args = (type(self), self.storage(), self.storage_offset(), tuple(self.size()), self.stride())
TypeError: 'int' object is not callable
  File "/usr/lib/python3.6/multiprocessing/reduction.py", line 51, in dumps
    cls(buf, protocol).dump(obj)
TypeError: 'int' object is not callable

If, instead, I change my factory functions to return a regular TensorImage and patch the show function everything works fine (so multiprocessing in itself works fine).
Unfortunately, my code is a bit long and I don’t have a nice clean reproducible example so can’t say for sure it’s not caused by something else I’ve done but I haven’t mentioned here.

Note that dls.one_batch() and model.forward() were not affected