Roberto
Roberto
+1 This is indeed a problem I also run into often
I have exactly the same problem, "first clustering "extremely slow (>200 s/it) probably because no GPU usage, first 2 steps 2 it/s as expected. On Task manager I can isolate...
Here a video of the issue https://www.dropbox.com/scl/fi/vaipnzqoxsyrrvbkuy3o8/Untitled.m4v?rlkey=watrlnib7q6a30a3sbzqovpoc&dl=0
I could solve this by using a totally fresh environment, not just uninstall>reinstall. However now I get the CUDA out of memory error 😰
Sure, compressing the files now.
In the zip there is a jupyter notebook that shows the problem and the specific Xd tensor that causes the crash on my machine. I put the standard and modified...
Just noticed that in the notebook I didn´t include the change at line `vtot = (Xg**2).sum(1)`
This last commit seems to really solve the OOM problems.
I have 12 GB. Without the modification I would get OOM often inside the kmeans_plus_plus func. Which line is problematic to you exactly? and what is the error message saying?...
Mmm never had a crash at that line. If you use the normal version, not my fork, does it also crashes in the same line?