Code sample of a full vanilla RNN implementation using dnnl::vanilla_rnn_backward

[dtecta]

I'm looking for and example showing a full implementation of RNN training using oneDNN. The current distribution at https://github.com/uxlfoundation/oneDNN.git shows only the forward pass, dnnl::vanilla_rnn_forward, in vanilla_rnn.cpp. 

In particular, I'm interested in a code sample that show the use of dnnl::vanilla_rnn_backward.

There is a rnn_training_f32.cpp example that shows both a forward and backward pass. However, this example is rather contrived and probably is included to show what to do when vanilla RNN cannot be used. 

Could someone please provide me with a code sample showing the use of dnnl::vanilla_rnn_backward in a typical use case? Thank you!

[Ying_H_Intel]

Hi @dtecta 

thank you for checking the issue here.  You may aslo post your question to oenDNN github : uxlfoundation/oneDNN: oneAPI Deep Neural Network Library (oneDNN) where oneDNN developer will help you directly