graphcast hub #977
graphcast hub #977
Conversation
julienchastang
force-pushed
the
tm24f
branch
4 times, most recently
from
a95faef to
b35747e
Compare
julienchastang
force-pushed
the
tm24f
branch
from
b35747e to
137c70e
Compare
| @@ -1,7 +1,7 @@ | |||
| # Heavily borrowed from docker-stacks/minimal-notebook/ | |||
| # https://github.com/jupyter/docker-stacks/blob/main/minimal-notebook/Dockerfile | |||
|
|
|||
| ARG BASE_CONTAINER=jupyter/minimal-notebook | |||
| ARG BASE_CONTAINER=jupyter/tensorflow-notebook | |||
There was a problem hiding this comment.
Why the switch from minimal to the tf image?
There was a problem hiding this comment.
It is the only way I can get the GPU to work.
There was a problem hiding this comment.
What are you trying? The image that's live right now uses the minimal-notebook and thomas and I can access the GPU just fine.
There was a problem hiding this comment.
How did you get that to work beyond the usual, i.e., the newly defined environment.yml and jupyterhub_gpu.yaml. Did you have to install anything special CUDA/GPU-wise?
There was a problem hiding this comment.
My understanding is that the current running environment was modified "in-place" for faster experimentation so not everything may have been captured in the Dockerfile, environment.yml, etc. I guess that is what I am asking.
There was a problem hiding this comment.
You can find the most recent Dockerfile/environment combo in the "ana" tmux session on the docker-gpu machine. Your most recent commit has the correct environment file, however using the minimal-notebook as the base still works.
To get the GPU to work there is nothing more needed past the "normal" things--i.e. requesting the additional resource via jupyterhub_gpu.yaml.
I've found that the important parts of getting these things to work are making sure the packages you install (via conda or pip) are compatible with that shown when doing an nvidia-smi on JS2.
For example, in this case I tell pip to look for torch and associated packages from the cuda 12.1 index with this line:
- --extra-index-url https://download.pytorch.org/whl/cu121
We can get together to discuss this in-person after the holidays?
cc @ThomasMGeo @ana-v-espinoza