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

Posted to commits@mxnet.apache.org by gi...@git.apache.org on 2017/08/13 19:49:41 UTC

[GitHub] sbodenstein opened a new issue #7445: Using cuDNN for CTC Loss

sbodenstein opened a new issue #7445: Using cuDNN for CTC Loss
URL: https://github.com/apache/incubator-mxnet/issues/7445
 
 
   @piiswrong, @szha: Now that cuDNN 7 supports CTC loss, perhaps we should discard the current GPU implementation in contrib.ctc_loss (adapted from the WarpCTC implementation) and only use cuDNN for GPU? The main reasons:
   1) it requires maintenance effort to ensure the GPU implementation works on new GPU architectures, requiring careful updating of dependencies (like modern gpu). 
   2) Users are still reporting problems with memset issues when using the WarpCTC plugin (#6121)
   
   I don't think the maintenance effort is worthwhile if almost every single user training with CUDA will have cuDNN.
   
   What are your thoughts?
   
   
 
----------------------------------------------------------------
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