mcframework.benchmark.default_backends

mcframework.benchmark.default_backends(*, include_torch: bool = True, include_mps: bool = True, include_cuda: bool = False) → list[BackendSpec]

Build the device-aware default backend matrix.

The CPU backends are always included. Torch backends are added only when their device is actually available, so a run never advertises a backend it cannot execute.

Parameters:

include_torchbool, default True

Include the Torch CPU backend when Torch is installed.

include_mpsbool, default True

Include the Apple-Silicon MPS backend when MPS is available.

include_cudabool, default False

Include the NVIDIA CUDA backend when CUDA is available. Off by default; CUDA support is the next milestone.

Returns:

list[BackendSpec]