Start with the core CLI, then add only the dependencies required by the command or workload you plan to run.
Follow the canonical installation instructions for a published or editable install. Verify that the active environment provides the command:
aorta --help
The core install does not require a GPU, ROCm, Docker, or PyTorch. Those are runtime requirements for particular workloads and features.
Use the canonical command-location guidance to decide whether the CLI and runtime dependencies belong on the host or in a container. Relative recipe, command, and output paths are resolved from your current directory.
Commands that name recipes/..., config/..., or scripts in this repository
assume the repository root as the current directory:
cd /path/to/aorta
aorta sweep run \
--recipe recipes/llm-determinism/example-llm-determinism.yaml \
--dry-run
From another directory, pass a path that is valid from that location.
Read the selected workload’s guide before installing its runtime:
hw-queue only when you use that
feature; see Optional dependencies.torchrun, and Docker only where
the workload instructions require them.For the repository’s Docker Compose development images for in-tree training
workloads, see docker/README.md. That setup is optional,
not the general AORTA installation path.
The environment probe is a useful first command:
aorta env probe --summary
Unavailable optional components are reported as such. For a recipe, validate its syntax and workload name before starting GPU work:
aorta sweep run --recipe /path/to/recipe.yaml --dry-run