Getting Started¶
This guide gets you from “installed” to a first successful submission.
Install¶
In a virtual environment:
pip install hpc-runner
Quick start: submit a job¶
Submit a batch job:
hpc run python -c 'print("hello")'
Show what would be submitted without actually submitting:
hpc run --dry-run python train.py
Run interactively (SGE: qrsh):
hpc run --interactive bash
Pass raw scheduler arguments using -- as a separator:
hpc run -q gpu.q -l gpu=1 -- python train.py
Or use submit for a config-driven shorthand:
submit -t gpu -n 4 python train.py
Pick a scheduler¶
By default, hpc-runner tries to auto-detect the scheduler. You can force it:
hpc --scheduler sge run "python train.py"
Or use the environment variable:
export HPC_SCHEDULER=sge
First config¶
Create a project-local config (safe to commit):
hpc config init
Create a user config:
hpc config init --global
Then open the file and set your site defaults, e.g. a default SGE queue:
[defaults]
queue = "batch.q"
See Configuration for the full configuration format and examples.