marlowe-slurm-operator

To submit a GPU job on the Stanford Marlowe HPC cluster, use Slurm commands like `sbatch` or `srun` while specifying the correct partition, required account suffix, and necessary module loads to ensure compliant job submission.

Monitoring pending job status and diagnosing reasons on a Slurm HPC cluster involves using standard Slurm tracking commands to review job states, diagnose pending reasons, and inspect finished jobs according to site-specific policies.

Slurm CLI tools must be available in the PATH on the target cluster to operate the Marlowe cluster, as this Skill relies on them to verify partitions, track GPU-hours, and execute job submissions directly within the environment.

You can track GPU-hour consumption for projects on the Marlowe cluster by using Slurm commands to monitor resource usage, ensuring adherence to site-specific policies for accurate accounting of your GPU computing workloads.

Slurm job submissions fail on Marlowe when site-specific account suffix and partition requirements are not met, so verifying live cluster facts and loading necessary modules before proposing commands prevents guesswork and submission errors.

Yes, you can verify the state of the preempt partition on the Marlowe cluster by discovering and verifying live cluster facts before proposing commands, preventing guesswork and ensuring safe job submission into the correct environment.