AttributeError: module 'torch.distributed' has no attribute 'is_initialized'

[Mohd-Misran]

I loaded the crf-con-en model to the Parser. Then I tried to use the predict() function on a list of English tokens.
But I get the above error. Need help on how to debug this!

Here's my code:

from supar import Parser
import nltk

parser = Parser.load("crf-con-en")

text = nltk.word_tokenize("John, who is the CEO of a company, played golf.")

output = parser.predict(data=[text], verbose=False)

Complete error log:

AttributeError Traceback (most recent call last)
in
----> 1 output = parser.predict(data=[text], verbose=False)

~\anaconda3\lib\site-packages\supar\parsers\crf_constituency.py in predict(self, data, pred, buckets, batch_size, prob, mbr, verbose, **kwargs)
121 """
122
--> 123 return super().predict(**Config().update(locals()))
124
125 def _train(self, loader):

~\anaconda3\lib\site-packages\supar\parsers\parser.py in predict(self, data, pred, buckets, batch_size, prob, **kwargs)
120 def predict(self, data, pred=None, buckets=8, batch_size=5000, prob=False, **kwargs):
121 args = self.args.update(locals())
--> 122 init_logger(logger, verbose=args.verbose)
123
124 self.transform.eval()

~\anaconda3\lib\site-packages\supar\utils\logging.py in init_logger(logger, path, mode, level, handlers, verbose)
28 level=level,
29 handlers=handlers)
---> 30 logger.setLevel(logging.INFO if is_master() and verbose else logging.WARNING)
31
32

~\anaconda3\lib\site-packages\supar\utils\parallel.py in is_master()
33
34 def is_master():
---> 35 return not dist.is_initialized() or dist.get_rank() == 0

AttributeError: module 'torch.distributed' has no attribute 'is_initialized'

[yzhangcs]

Hi, thanks for reporting this error. Could you provide the pytorch version and device info. Yu Zhang Soochow University

…

________________________________ From: Mohammed Misran <notifications@github.com> Sent: Thursday, July 30, 2020 8:02:02 PM To: yzhangcs/parser <parser@noreply.github.com> Cc: Subscribed <subscribed@noreply.github.com> Subject: [yzhangcs/parser] AttributeError: module 'torch.distributed' has no attribute 'is_initialized' (#37) I loaded the crf-con-en model to the Parser. Then I tried to use the predict() function on a list of English tokens. But I get the above error. Need help on how to debug this! ` from supar import Parser import nltk parser = Parser.load("crf-con-en") text = nltk.word_tokenize("John, who is the CEO of a company, played golf.") output = parser.predict(data=[text], verbose=False) ` — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<#37>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AEMMYK72IAG3GAWY42BNLK3R6FOLVANCNFSM4PNSP65A>.

[Mohd-Misran]

@yzhangcs

Torch version 1.5.1
Windows 10 Pro

[yzhangcs]

@Mohd-Misran You could check if torch.distributed.is_available() is True (facebookresearch/maskrcnn-benchmark#280).

[Mohd-Misran]

@yzhangcs
torch.distributed.is_available() is False

[yzhangcs]

@Mohd-Misran It seems that distributed training is not supported on Windows.

[yzhangcs]

@Mohd-Misran DDP is supported on Windows since torch 1.7.