Serkan Sulun

Question about discriminator input

3

Hi Matt, Thanks for the great code. According to the paper, the input to the discriminator is a sequence of frames (history of frames, and the next frame). If I...

Losses are not normalized over the batch. Though it doesn't affect the results, if someone wants to experiment with different parameters, be aware that increasing batch size will increase loss...

Reversible layers increase memory usage

1

I'm checking memory usage using nvidia-smi. When I turn on reversibility (setting reverse_thres to two times the input length) it's using 8.8 GB memory. When I turn it off (setting...

Hi, thanks for the great work. Could you share the weights for the final linear layer (`nn.Linear(4096, 7)`) (classifier) of VGG_Net? Without this layer, it's impossible to do inference.

Hi, thanks for the great code. Since LMShuffledIterator are putting different samples (unrelated sentences) in different batches, how is the memory still useful? Thanks

How to use in CPU mode

1

Is it possible to run this using CPU only? Which modifications should I make? Thanks

Hi Andrew, thanks for the great code. In my own code for MIDI generation, at inference time, I tried your repeat penalty formula and it worked great. Do you have...

When I run mpegflow, I get the following message for all frames starting from 2: Skipping frame X (frame with pts 0 already processed) It produces no output.

I couldn't run the code due to version compatibility issues. After some trial and error, I found out that tensorflow and dm-sonnet should be version 1. I added the requirements.txt...

Hi! Could you please provide a minimal inference script where we can provide the input audio file and get predictions? I wrote the script below, but the output predictions are...