samskalicky commented on pull request #20043:
URL: https://github.com/apache/incubator-mxnet/pull/20043#issuecomment-833111127


   > Thanks for the contribution! Some of the changes will be useful for 2.0 
too. One general question I have is whether there's opportunity to clean up the 
current c predict APIs in 1.x so that we can offer a clean version for 2.0 too. 
If you are open to it, it would be great to see a proposal on how a clean c 
predict API could look like given the opportunity to break backward 
compatibility.
   
   While helping with this PR, I noticed a lot of similarity in the TVM/DLR 
APIs:
   https://github.com/neo-ai/neo-ai-dlr/blob/main/include/dlr.h
   or
   https://github.com/neo-ai/neo-ai-dlr/blob/main/include/dlr_tvm.h
   
   Maybe we can use those as a reference as well in this exercise. 
   
   Cleaning up the C Predict API is a great idea, at the same time we should 
think about other C/C++ APIs like the MXGetBytesInUse or MXLoadLib that we want 
to include in our C interface moving forward. 


-- 
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:
[email protected]


Reply via email to