Issue with model.train() in scVI - Training Stuck at 0%

Help scvi-tools
1

Dear：
I am encountering an issue while using scvi-tools for cell integration. Specifically, when I run model.train(), the training progress remains stuck at 0%, and no iterations occur despite the GPU being detected and utilized. Below is the log output

image
image1336×380 24.5 KB

I have tried the following steps to troubleshoot the issue:

  1. Verified that the dataset is loaded correctly and passed to the model.
  2. Ensured that the GPU is available and CUDA is functional.
  3. Adjusted training parameters such as max_epochs and learning rate.

Despite these efforts, the problem persists. I would greatly appreciate your guidance on how to resolve this issue. Please let me know if you need further details about my setup or code.

and,at the same time, after a few minutes, the following error appears:
Trainer.fit stopped: max_epochs=246 reached.

2

Hi @Ares ,
As discussed on Issue with model.train() in scVI - Training Stuck at 0% · Issue #3094 · scverse/scvi-tools · GitHub
Apriori, seems like a problem in that the interactive progress bar of your jupyter nb is not progressing, as the final message is that the model was trained (this is not an error).
So a problem in the jupyter nb widgets I suspect.

Can you check if model.is_trained gives you True (you can run 1 epoch just to debug this).