ArmageddonKnight commented on issue #10999: Naming of Contrib Implementations URL: https://github.com/apache/incubator-mxnet/issues/10999#issuecomment-390366561 @szha @marcoabreu Thank you so much for your reply. @szha My implementation is based on the open-source LSTM RNN implementation https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/, except that it supports **input dropout** and also **backward pass** (everything `cuDNN` supports for LSTM RNN). I compared my implementation against `cuDNN 6.0` on the Python interface and it is comparable with `cuDNN`. There are cases in which `cuDNN` is slightly better and worse (both at around `20%`). In my opinion, it is always better for other developers to have a complete baseline implementation to work on and experiment. @marcoabreu I added my implementation into the Python **word-level language modeling** example for benchmarking. For the naming, if organization dependent naming is not preferred, is `OpenLSTMRNN` acceptable?
---------------------------------------------------------------- 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: us...@infra.apache.org With regards, Apache Git Services
