device

Torch device selection shared by enrichment detectors.

Functions

get_device([force])

Select the best available inference device: cuda, DirectML, or CPU.

Module Contents

openplaces.io.enricher.detectors.device.get_device(force: str | None = None)

Select the best available inference device: cuda, DirectML, or CPU.

The choice is printed once per process so a silent CPU fallback is visible instead of looking like a hang.

Parameters:

force (str, optional) – Device string (e.g. ‘cpu’, ‘cuda:0’) that overrides detection. The OPENPLACES_DEVICE environment variable plays the same role when set.

Returns:

CUDA when an NVIDIA GPU is available (ROCm builds also surface here), otherwise DirectML when the optional torch_directml package is importable (AMD/Intel GPUs on Windows; never install it into the main conda env), otherwise CPU.

Return type:

torch.device