Created
October 16, 2017 10:51
-
-
Save lirnli/4282fcdfb383bb160cacf41d8c783c70 to your computer and use it in GitHub Desktop.
Pytorch Wavenet
use generate_slow() doesn't have anything
yes. generate() function takes way too much memory. That might be because pytorch tries to expand the storage (not the tensor) as a whole piece, but never deallocates used portions. One quick work around would be to clone the tensor every few cycles, so the old tensor and storage can be freed by GC. I do not have a more Pythonic solution right now.
Whether I use generate_slow() or generate(), I only get a straight line, do you know why ?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
hi, when i use the generate() function to generate a wav like n=48000, the gpu memory will increase all the time up to run out of memory, why is it?could u help me?