Colab just notified me "Your session crashed after using all available RAM." 😭

I will be using my own machine with 96 GB GPU for demo. Participants may not be able to run all sections with limited GPU. It is fine. This section is mostly for demonstration purpose only.

OK sounds good then! I'd love to test if this works on CryoCloud so we can demo the Notebook in a familiar environment instead of introducing Colab, which may be unfamiliar to some. Have you tested it already? Does it need any new dependencies in the container environment?

Yes, only the geoai package dependency is needed.

mamba install -c conda-forge geoai "pytorch=*=cuda*"

OK I'll add that to this PR and then the docker build tests will kick in. One moment, please ⌛

The conda package specification style "pytorch=*=cuda*" looks a bit unfamiliar to me. I went ahead and put it in the environment.yml! Does it look good to you?

It forces it to install the pytorch GPU version. Otherwise, some env might just the CPU version

The environment fails to solve with these new dependencies. Trying with python>=3.12 instead of python==3.13. I'm not sure if that will fix it, just a guess. These conflict reports are hard to read. Any idea how to address the failure @giswqs ?

Installing pytorch is aways a tricky issue. I would not try to build it with GitHub actions. Just skip executing the notebook. Otherwise, it will take forever to run the notebook and build the website.

We are currently not executing the Notebook. I'm trying to add pytorch to the Docker image we'll use on CryoCloud so you can execute this Notebook in CryoCloud instead of Colab on the day of the workshop. Any idea how we can fix this so the conda environment is solvable and we can build the Docker image?

Try Python 3.12. Some pytorch GPU dependencies might not support Python 3.13

Tried that in 0f9617f, no luck :(

Try the following sequence to track down which package dependencies cause the issue.

1. "pytorch"
2. "pytorch==cuda"
3. "geoai"

Try the following sequence to track down which package dependencies cause the issue.

1. "pytorch"

2. "pytorch=_=cuda_"

3. "geoai"

Thanks!

1. "pytorch" - builds successfully
2. "pytorch=*=cuda* - fails to solve with similar output to GitHub Actions
3. "geoai" - builds successfully

That means pytorch GPU version can't be installed in that docker image. Although geoai installs successfully, it only installs the pytorch CPU version.

The docker image needs to have cuda installed before installing pytorch GPU. Here is the official docker images: https://hub.docker.com/r/pytorch/pytorch/tags

I am not sure if it is worth your time to go through it.

OK. Let's remove 0f9617f and 1b46f1e and merge this PR? Maybe we can figure out how to do this in cryocloud in the future.

FWIW, I seem to have been able to install the GPU version of pytorch on the CryoCloud hub.

gh pr checkout 30
mamba install geoai
mamba install "pytorch=*=cuda*"

import torch

print("CUDA available:", torch.cuda.is_available())
print("Number of GPUs:", torch.cuda.device_count())
if torch.cuda.is_available():
    current = torch.cuda.current_device()
    print("Current GPU:", torch.cuda.get_device_name(current))

python modules/03-integrating-ai/test_pytorch.py 
CUDA available: True
Number of GPUs: 1
Current GPU: Tesla T4