Kh4L opened a new pull request #18350:
URL: https://github.com/apache/incubator-mxnet/pull/18350


   ## Description ##
   Symbol's `optimize_for` was failing when some parameters are missing in 
`args` or `aux` due to:
   - enforcing both in the Python front-end and back-end
   - no way to specify attributes of input 
   - partial shape inference failing (also related to #18332 and  #18348 fixes)
   
   ### Changes
   - Add `skip_infer` option to `symbol.optimize_for`, which skips the 
pre-partition attributes inference. Can be used for backends that does not need 
this (e.g. TensorRT).
   - Add optional `shape_dict`, `type_dict` and `stype_dict` to `optimize_for` 
to specify input data attributes for the pre-partition attributes inference 
passes. These can be used when some input and/or parameters are missing in 
`args`, e.g. input data (similarly to what is done in the `bind` functions).
   - Changes both Python and C++ functions to allow partial inference.
   
   CC @samskalicky 
   


----------------------------------------------------------------
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