You are viewing a plain text version of this content. The canonical link for it is here.

Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2020/09/24 18:42:33 UTC

[GitHub] [incubator-mxnet] ptrendx commented on pull request #19209: [BUGFIX]Fix cuDNN dropout reproducibility

ptrendx commented on pull request #19209:
URL: https://github.com/apache/incubator-mxnet/pull/19209#issuecomment-698519737


   @szha Over the course of testing this PR I discovered a different possibility of the nondeterminism - if there are more than 1 copies of the dropout resource (or random resource for that matter) then you end up with the race condition of getting a new resource and call to `seed` setting the counter to 0. We could either remove the copies > 1 (and also remove parallelrandom resource, which has the same problem), or adding `waitall` to the `seed` call. Which one do you think is a better solution?


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org