-
Notifications
You must be signed in to change notification settings - Fork 253
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Out of memory in Colab (free) #15
Comments
Thats weird havent had that problem on long audio files. Try with another smaller whisper model "--whisper-model" Default is "medium.en", so try with "small.en" it will reduce the accuracy but should use 3x less memory. Edit: adding "torch.cuda.empty_cache()", will free up memory thats not in use. |
Hi, this is strange as nemo automatically segments the file to avoid OOM errors, setting the max split size might help because there's plenty of memory available |
Hello, I would like your help to know where to set the max split size. in the create_config function, i could not find any such parameter. |
@rashi-budati check |
If you're having OOM issues and don't mind a performance hit, always remember that you can run this with --device cpu, that way you aren't worrying about your GPU at all. I'm running whisper-diarization locally though, unsure how this would interact with GColab |
Thanks for the Colab.
I ran into the dreaded out memory while
processing the embedding files saved to nemo_outputs/speaker_outputs/embeddings
Just trying to understand: My audio file is only 1h18 minutes long, but its embeddings amount to: "Dataset loaded with 9209 items, total duration of 2.55 hours." Since segmentations and subsegmentations are done automatically, I wonder what the upper limit filesize for the script to work on Colab would be.
Full error:
RuntimeError: CUDA out of memory. Tried to allocate 7.61 GiB (GPU 0; 14.75 GiB total capacity; 7.81 GiB already allocated; 5.65 GiB free; 7.83 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
The text was updated successfully, but these errors were encountered: