Closed PonteIneptique closed 4 years ago
Hey thanks for the PR! I haven't been using pytorch > 1.0 and I am not up to date with these changes. Please have a look at the comment (mostly about readability).
Here you go ! Thanks for the quick answer :)
Thanks for the PR!
Thanks for the quick management ;) When you have time, a bump to "0.2.6" would be neat ;)
Hey ! I spent some time looking at what could be blocking >= 1.2. Most of it was the use of int mask used as boolean, which was touched by a breaking change inserted in 1.2.0 ( https://github.com/pytorch/pytorch/releases/tag/v1.2.0 ) I found two concerned occurrences, trained with my usual models as well as tagged. This does not cover all models but gives a general robustness.
All other masks seems to use the numerical value rather than the boolean one.