NVIDIA/waveglow

RuntimeError: A tensor was not contiguous.

sciai-ai opened this issue · 2 comments

sciai-ai commented

Hi

I ran the train script and got the following

python train.py -c config.json

image

Selimonder commented

This issue might have been caused by nvidia's apex. You can either disable the fp16 run from config or alternatively you can replace apex with torch native mixed-precision training.

bfs18 commented

change L79 in glow.py to

        W = W.contiguous().view(c, c, 1)
  • 👍9
  • 🎉1
  • 🚀1