Enable CUDA in jupyterlab environment?

HPC Technical
1

Hi Hpc team !
First thank you this great tool that is ondemand :slight_smile:

I am using jupyter lab to perform single cell RNA seq analysis, and there are some tools requiring gpu acceleration (scvi-tools or drvi).
I managed to install it in a venv, but I think it does some mix up with python libraries from GCCore.

More important: even if I start the session on the GPU partition with one GPU. The GPU is detected but in the end I get an error message saying that torch is not compiled with CUDA enabled.

Is there an option I am missing ? Is there a GCCore environment available were CUDA is enabled ?

Otherwise, I see that it is possible to run scvi tools with a docker image (https://hub.docker.com/r/scverse/scvi-tools). Is it possible to have jupyterlab running through this (as for RStudio ?)

Thank you very much, hope it is not to confuse!
Best,

1 Like
2

Dear @Chloe.Mayere

I have a solution for you! I’ve added an extra field in the JupyterLab submission form to specify extra modules you want to load, see in the screenshot below.

firefox_IYgeGoiLne
firefox_IYgeGoiLne710×349 10.8 KB

Best

Yann

3

Thank you so much !
I’ll give it a try straight away :blush:

Best !

4

Another question, where can I see which modules are available?
:slightly_smiling_face:

Do I still have to install these in my env (or venv) ?

(Sorry for these naive questions ^^)