Closed LucySha closed 5 years ago
Interp weights are used in the learned upsampling layer, so only in certain models. Did you make sure you downloaded all the model checkpoints and put them in the correct place as described in the Readme file? Especially checkpoints/full_44KHz/
should contain three model files also with full_44KHz in its name.
Also which command did you use? python Predict.py with cfg.full_44KHz
as in the README or something else?
@f90 I used the cpu and it worked well, maybe the version of cudnn and ccudatoolkit conflicts. CUDA=9.0 cudatoolkit=7.6
I have CUDA 9 on my end too and it worked with GPU use. But given it works on CPU it seems like the models are properly downloaded and "installed" on your end. If you continue having problems it would be informative to print out a list of all the weights in the checkpoint file and in the model since they need to be the same but apparently aren't, then I might be able to help you out!
Closing due to lack of activity
Thanks for your works! When I prun the Predict.py with the pre-trained model, I got the ERROR:Key separator/interp_0 not found in checkpoint, Can you help me to figure it out?