Global configuration¶

This section describes options that impact all modules.

Options¶

Name	Allowed	Default	Description
`hardware`	`"cpu", "gpu"`	`"gpu"`	Forces execution on the specified hardware. The config-level enum is only `cpu` / `gpu`; when set to `gpu`, RAITAP probes for CUDA (NVIDIA), then XPU (Intel), then MPS (Apple, planned) and binds to the first one it finds. If none is available, RAITAP falls back to CPU and emits a CLI warning. The probed backend must match the installed extra — `gpu` on a CUDA host needs `torch-cuda` / `onnx-cuda`, `gpu` on an Intel host needs `torch-intel` / `onnx-intel`. See 2. Pick assessment extras for the extras matrix.
`experiment_name`	`string`	`"Experiment"`	Name of the experiment (assessment run).
`hydra.run.dir`	`string`	`"./outputs/<date>/<time>"`	Directory where Hydra stores the run outputs. If not specified, Hydra creates a timestamped directory under `./outputs` relative to the terminal working directory where RAITAP was launched. If you want to forward outputs to your tracking software, see Configuration.

YAML example¶

YAML

hydra:
  run:
    dir: "./custom-outputs-dir"

hardware: "gpu"
experiment_name: "My Experiment"

Python

from raitap import AppConfig, Hardware

# Python users construct AppConfig directly and call ``run(config)``;
# ``hydra.run.dir`` is a CLI/YAML-only knob (set the output directory in
# your own code instead).
config = AppConfig(
    hardware=Hardware.gpu,
    experiment_name="My Experiment",
)

CLI override example¶

uv run raitap hardware=cpu experiment_name="My_Experiment"

pip

raitap hardware=cpu experiment_name="My_Experiment"

For module-specific options and examples, refer to Module-specific configurations.