Running on a SLURM compute node¶
On an HPC cluster the GPUs live on compute nodes, not the login node. If you have an active allocation you can launch the dashboard on that node from the login node with the wrapper script, without installing anything on the compute node yourself.
Note
scripts/kempnerpulse-on-cnode.sh currently targets the FASRC cluster,
where a shared KempnerPulse virtual environment is available on the compute
nodes. It exits gracefully if that environment is not found. Point it at your
own environment with KEMPNERPULSE_VENVPATH (below).
Find your compute node¶
squeue -u $USER
Launch¶
Using holygpu8a11101 as an example node:
./scripts/kempnerpulse-on-cnode.sh holygpu8a11101
The wrapper SSHes to the node, verifies it has GPUs, activates the KempnerPulse
environment, and launches the dashboard. Any extra arguments are passed straight
through to kempnerpulse:
./scripts/kempnerpulse-on-cnode.sh holygpu8a11101 --focus-gpu 0 --hpc-weights
To exit, press Ctrl-C.
Pointing the wrapper at a different environment¶
If your install lives somewhere other than the shared path, set
KEMPNERPULSE_VENVPATH to the virtual-environment root (the script appends
bin/activate):
KEMPNERPULSE_VENVPATH=$HOME/path/to/your/venv ./scripts/kempnerpulse-on-cnode.sh holygpu8a11101
After a successful run with the override, the script offers to save it as the
new default (rewriting the hardcoded path inside the script and leaving a
.bak alongside it for recovery). Answering anything other than yes leaves the
script unchanged.