Anup Rajput

We have observed that converting using cudnn.convert doesn't work for all modules, for example cudnn.ClippedReLU doesn't get translated into nn despite mention of API compatibility.

We have observed that converting using cudnn.convert doesn't work for all modules, for example cudnn.ClippedReLU doesn't get translated into nn despite mention of API compatibility.

@pfeodrippe please find attached: [wavenet_models.zip](https://github.com/ibab/tensorflow-wavenet/files/477034/wavenet_models.zip) Training information: GPU used - Titan X last year's model (12 GB VRAM) Runtime - Not measured, but less than an hour. BATCH_SIZE = 1...

No, same issue as you. Still trying to trace it.

First sounds with the current default settings! :) [generated_18Sep_1714.wav.zip](https://github.com/ibab/tensorflow-wavenet/files/478742/generated_18Sep_1714.wav.zip)

I have tried to compile the generated audio in a wiki: [Generated audio samples wiki](https://github.com/adroit91/tensorflow-wavenet/wiki/Generated-samples) We can observe how the training is converting gibberish into natural sounding voice over iterations....

We also noted this behaviour when using the bleeding edge of Tensorflow itself as well as tflearn. Trying checking out the last official releases of both, and it should work.