Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.
And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.
Side effect of jit_set_profiling_*(False) at import time?
See original GitHub issueI just spent 2 days wondering why my code got way slower and finally narrowed it down to a side effect of this line:
import speechbrain
How ironic! 😃
It looks like, when imported, it runs the following two lines that seem to have a huge impact on the speed of the rest of my (otherwise speechbrain-free) PyTorch code:
I found this related issue on pytorch repo: https://github.com/pytorch/pytorch/issues/38342
Any up-to-date details of why those two lines are needed?
Additional information about my setup:
import torch
torch.__version__
# 1.8.1+cu102
speechbrain.__version__
# 0.5.10
Issue Analytics
- State:
- Created 2 years ago
- Comments:11
Top Results From Across the Web
Top Related Medium Post
Top Related StackOverflow Question
Troubleshoot Live Code
Top Related Reddit Thread
Top Related Hackernoon Post
Top Related Tweet
Top Related Dev.to Post
Top Related Hashnode Post
oh no, not jit again. Yeah, we did this due to a regression observed in between pytorch version with Jit based code. @mravanelli should we try to see if higher version of Pytorch are now free of this previously observed regression ? Then we could maybe remove these lines …
Anyway, feel free to rename/close this issue as it somehow diverged from what I thought it was about…