szha commented on pull request #18987: URL: https://github.com/apache/incubator-mxnet/pull/18987#issuecomment-679455893
> I think CNNs are generally static shape while models in NLP are generally dynamic shape. I don't think we can generalize like this. For example, object detection and segmentation are based on CNN and are usually not static-shaped. > Do we have any plan for improving the memory usage? Of course we do. I think @ArmageddonKnight is currently fixing some missed allocation entries in memory profiler, and plans on developing a memory usage visualization tool later this week to help narrow down the focus for memory optimization. We also intend to add mirror option to cached op to allow training for larger model. ---------------------------------------------------------------- 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]
