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 2018/08/18 15:23:01 UTC

[GitHub] al-rigazzi commented on issue #12240: ImageIter crashing with 64k samples per batch

al-rigazzi commented on issue #12240: ImageIter crashing with 64k samples per batch
URL: https://github.com/apache/incubator-mxnet/issues/12240#issuecomment-414065538
 
 
   My approach is similar to the one you pointed out (MPI, no PS), I guess we've worked in parallel. Still, I think there are some differences, and it would be worth checking the final results.
   
   Scale efficiency is quite high on Skylake nodes, but I will publish a blog post about it when I finish the study.
   
   OK, I will check your implementation and see if the difference is still there, at the same time, I would appreciate your independent investigation. I think that, if there is not enough memory, we could do a dry run adapting RN50 to use 1x1x3 images, so that we can fit 64k on fewer nodes. Let me know if I should help you with this.
   
   Thanks,
   Al

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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


With regards,
Apache Git Services