@safrooze Thank you for your reply. The reason is because in our experiments we see that OpenLSTM is able to beat cuDNN 6 in end-to-end performance evaluation (in terms of training throughput, the benchmarks used are included in this commit). Besides, we also think that it is good to have an open-source LSTM CUDA implementation that other researchers can experiment with or use as a baseline.
The reference is [**here**](https://arxiv.org/pdf/1805.08899.pdf). [ Full content available at: https://github.com/apache/incubator-mxnet/pull/11364 ] This message was relayed via gitbox.apache.org for [email protected]
