search

tensorflow / java

Java bindings for TensorFlow
Apache License 2.0
822 stars 202 forks source link

Do we support tf.nn.bidirectional_dynamic_rnn ? #421

Open mullerhai opened 2 years ago

mullerhai commented 2 years ago

Hi: I only see CudnnRNN ,the project has not testcase for it,I not know if we support tf.nn.bidirectional_dynamic_rnn or normal simple rnn? thank you

karllessard commented 2 years ago

I don't think we do support out-of-the-box RNN ops, most in TensorFlow has been also deprecated in favor to Keras RNN cells and layers, so I guess this support will also as we port the Keras API into our framework.

mullerhai commented 2 years ago

I don't think we do support out-of-the-box RNN ops, most in TensorFlow has been also deprecated in favor to Keras RNN cells and layers, so I guess this support will also as we port the Keras API into our framework.

wow, when this framework will wrapper keras api?

karllessard commented 2 years ago

This is a own project on its own, leaded by @JimClarke5 , a lot of work as already been done by porting the optimizers, metrics, losses, regularizers and activations of Keras to Java. The plumbing for layers and models are coming on soon and from there it would be interesting to tackle the support for RNNs.

If you are interested to know more about it @mullerhai and if you can, I suggest that you join the next SIG meeting where we discuss various topics related to TF Java.