Regarding memory required for the first tutorial of cell2location

[sudeepthi5]

Please use the template below to post a question to https://discourse.scverse.org/c/ecosytem/cell2location/.

Problem

I am trying to run the example in the following page:
https://cell2location.readthedocs.io/en/latest/notebooks/cell2location_tutorial.html#Loading-packages

I used the following to request my node:
srun --ntasks-per-node=4 --cpus-per-task=4 --mem=32000 --time=12:00:00 --partition=amperenodes --job-name=cell2location --gres=gpu:2

So in the training step I am getting 221/250, instead of training complete, and always I am getting error saying 'use_gpu' command is unrecognized. I would like to request you to provide memory requirement details for tutorial. I will provide any other information if required, can you please help me with this, Thank you.

• I follow the instructions from the cell2location tutorial (using on scvi-tools).
• I have adjusted required hyperparameters to my dataset and tissue N_cells_per_location and detection_alpha.
• I have provided 10X reaction/inlet as batch_key for reference NB regression.
• I have checked scverse Discourse and old Cell2location Community Forum, and did not find a solution.

[vitkl]

Please make sure that you have installed the latest and compatible versions of packages and used the corresponding options (latest version doesn't have 'use_gpu' argument).

The tutorial can run on Google Colab with 12GB RAM given the same data. If you have larger data - you need more RAM.

Cell2location also doesn't use multiple GPUs (--gres=gpu:2) so unless you are starting two processes manually, you can request only one GPU.

There is a bug in the latest GitHub version that results in data being recorded during posterior sampling - resulting in huge RAM use. I am working on fixing this issue - however for now, you can use use_quantiles option of the posterior sampling and exclude_vars to exclude 'data_target' (if it exists).

[sudeepthi5]

I will work on the recommended options. Thank you.