Document how to run pytorch-notebook docker container

[weiji14]

Follow up of #315 to document how to run the pytorch-notebook docker image.

Output of docker run -it --rm --gpus all pangeo/pytorch-notebook:master nvidia-smi should show something like this:

Thu Apr 28 02:21:34 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 510.60.02    Driver Version: 510.60.02    CUDA Version: 11.6     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA RTX A500...  Off  | 00000000:01:00.0 Off |                  N/A |
| N/A   51C    P3    24W /  N/A |      5MiB / 16384MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      2574      G                                       4MiB |
+-----------------------------------------------------------------------------+

TODO update https://pangeo-data.github.io/pangeo-stacks/images.html too?

[github-actions]

👈 Try on Mybinder.org!
👈 Try on Pangeo GCP Binder!
👈 Try on Pangeo AWS Binder!

[scottyhq]

TODO update https://pangeo-data.github.io/pangeo-stacks/images.html too?

Good point! I opened #319 to track

how to run the pytorch-notebook docker image.

I think adding this in pytorch-notebook/readme.md would be great for now. Some information about required hardware would be good (maybe just links to the NVIDIA docs from #315).

For those that don't have a local GPU one of the easiest ways I've found to run a Docker container with Pytorch on a GPU is via Azure Container Instances https://github.com/Denolle-Lab/azure/tree/main/aci_plus_volume, not sure if it's worth adding some documentation on that? Currently we just have https://github.com/pangeo-data/pangeo-docker-images#how-to-launch-an-image-with-a-cloud-provider-on-your-own-account

[weiji14]

how to run the pytorch-notebook docker image.

I think adding this in pytorch-notebook/readme.md would be great for now. Some information about required hardware would be good (maybe just links to the NVIDIA docs from #315).

Hmm yes, nvidia-docker isn't exactly a trivial install. I thought to keep it in the main README.md because this will be needed for the tensorflow docker image as well, and to be fair, the guide at https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html seems pretty good already and will be better maintained than here.

For those that don't have a local GPU one of the easiest ways I've found to run a Docker container with Pytorch on a GPU is via Azure Container Instances https://github.com/Denolle-Lab/azure/tree/main/aci_plus_volume, not sure if it's worth adding some documentation on that? Currently we just have https://github.com/pangeo-data/pangeo-docker-images#how-to-launch-an-image-with-a-cloud-provider-on-your-own-account

It looks quite involved, but I suppose we could copy some stuff from that. Ideally there would be a one-click GPU-enabled binder link (behind a login of course). I know that microsoft/torchgeo#316 made a 'Open on Planetary Computer' button like , but this won't be pulling the pangeo/pytorch-notebook docker image, just the files from git.

[scottyhq]

Ideally there would be a one-click GPU-enabled binder link (behind a login of course)

Agreed! Not sure if it's in scope for the upcoming revamped pangeo-binder (2i2c-org/infrastructure#919).

Happy to merge this as is if you'd like.

[scottyhq]

thanks!