Device Utilities

Device resolution utilities for PyTorch training and inference.

Provides a single entry point for resolving user-requested device strings (auto, cpu, cuda, mps) into concrete PyTorch device identifiers, with availability checks for CUDA and Apple Metal Performance Shaders (MPS).

src.utils.device.is_mps_available() → bool

Check whether Apple Metal Performance Shaders (MPS) backend is available.

Returns:

True if torch.backends.mps is available and built, False otherwise.

src.utils.device.resolve_torch_device(requested: str = 'auto') → str

Resolve a user-requested device string to a concrete PyTorch device.

Parameters:

requested – One of "auto", "cpu", "cuda", or "mps". "auto" selects CUDA if available, then MPS, falling back to CPU.

Returns:

The resolved device string ("cuda", "mps", or "cpu").

Raises:

ValueError – If requested is not a supported choice, or if the requested device is not available on the current machine.