Using cache when loading vocab.Vectors on Windows gives OSError Invalid Argument error

I was trying to load fasttext common-crawl vector (.vec.gz) file into torchtext by using torchtext.vocab.Vectors emb = vocab.Vectors(name=fasttext_name, url=URL, cache=fasttext_cache) For the first time without the .pt file, loading the vectors works fine and it creates<filename> .pt file on the cache folder. However, whenever I load the vector again when the .pt file exists in the folder, it gives this kind of error:

 File "load_emb.py", line 94, in get_fasttext_embs
    name=fasttext_name, url=URL, cache=fasttext_cache
  File "C:\Users\user\Anaconda3\lib\site-packages\torchtext\vocab.py", line 323, in __init__
    self.cache(name, cache, url=url, max_vectors=max_vectors)
  File "C:\Users\user\Anaconda3\lib\site-packages\torchtext\vocab.py", line 432, in cache
    self.itos, self.stoi, self.vectors, self.dim = torch.load(path_pt)
  File "C:\Users\user\Anaconda3\lib\site-packages\torch\serialization.py", line 386, in load
    return _load(f, map_location, pickle_module, **pickle_load_args)
  File "C:\Users\user\Anaconda3\lib\site-packages\torch\serialization.py", line 580, in _load
    deserialized_objects[key]._set_from_file(f, offset, f_should_read_directly)
OSError: [Errno 22] Invalid argument

I don’t have this error when I run the same code inside a linux machine. I wonder what can I do to fix this on Windows machine since I need to delete the .pt file and starts loading fasttext vector from the start again every time I need to run the code.

[mattr1]

Okay, looks like a known issue with pytorch 1.2: https://github.com/pytorch/pytorch/issues/26998

[zhangguanheng66]

Feel free to re-open the issue if you still have the question.