RuntimeError: torch.cuda.FloatTensor is not enabled. about unsupervisedmt HOT 5 CLOSED

Done. I follow https://github.com/facebookresearch/UnsupervisedMT/pull/9/files. Moreover, I also change the line 209 to:
if torch.cuda.is_available():
sent1, sent3 = sent1.cuda(), sent3.cuda()
Tnx

from unsupervisedmt.

Hi,
People have tried to make it run on CPU, for instance https://github.com/facebookresearch/UnsupervisedMT/pull/9/files
Maybe you can have a look at this. Overall, removing all occurrences of .cuda() and replacing all occurrences of torch.cuda by torch should be enough.

from unsupervisedmt.

@ehsanzare I'm trying to run the same experiments , Did you benchmark on different no of CPU's . Just curious if it is linearly scalable.

Cheers.

Mohammed Ayub

from unsupervisedmt.

No, I did not try that on no of CPU's.

from unsupervisedmt.

@ehsanzare @glample I'm trying to run in in Windows, multiprocessing_event_loop.py file is giving error saying signal has no attribute SIGUR1 . Seems like Python multiprocessing is slightly different on Windows. For now, I have replaced all occurrences of SIGUR1 with SIGTERM and training seems to initiate fine. Not sure if this is the best solution to avoid this error.

from unsupervisedmt.