The exact speedup depends on your model. These layers are available in the dev version on GitHub and will be part of the next PyPI release.
Show this thread
The CuDNNGRU & CuDNNLSTM layers are now available in Keras. Less features than GRU/LSTM, but a 3x-6x speedup on GPU. https://keras.io/layers/recurrent/#cudnngru …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
any updates on your book? My Amazon (de) pre-order tells it will be out on 11/6 – realistic?
-
I finished the draft in early June. It's in the hands of the editor and I cannot tell you for sure when it will be out, sorry...
- Show replies
New conversation -
-
-
Is it possible to import/convert the saved weights to run on the cpu for these layers?
-
`http://model.save _weights(fname)` produces a file that can be loaded into the original model. It's the same weight structure.
- Show replies
New conversation -
-
-
What is the keras version for cuDNNLSTM? I use keras : 2.0.8, tf:1.2.1. But importing error using from keras.layers import CuDNNLSTM. ㅠㅠ
-
Sorry. I miss your twit about dev.
End of conversation
New conversation -
-
-
~15x here
(keras + nilmtk framework + nvidia p4000)
Great job!Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Could anyone please share an example of converting of weights of a model with CuDNNLSTM to a model with LSTM? Just can't get it working no matter what I try
-
Did you get this working? Changing the recurrent activation to sigmoid?
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.