emmi_inference.device¶

Attributes¶

`DeviceEnum`	Supported execution devices.
`DeviceContext`	Carrier for device execution context.

`resolve_device`([device])	Resolve the effective device string.
`sync_device`(device)	Synchronize device streams.
`resolve_autocast`(effective_device, requested)	Decide whether autocast should be enabled.

class emmi_inference.device.DeviceEnum¶

Bases: enum.StrEnum

Supported execution devices.

AUTO¶: Let the library pick the best available backend (CUDA -> MPS -> CPU).

MPS¶: Apple Metal Performance Shaders backend on Apple Silicon (if available).

Initialize self. See help(type(self)) for accurate signature.

emmi_inference.device.resolve_device(device=DeviceEnum.AUTO)¶

Resolve the effective device string.

Parameters:: device (DeviceType) – Either a DeviceEnum or a string (“auto”, “cpu”, “cuda”, “mps”). When set to “auto”, prefers CUDA, then MPS, then CPU.
Returns:: One of “cuda”, “mps”, or “cpu” depending on runtime availability.
Return type:: str

emmi_inference.device.sync_device(device)¶

Synchronize device streams.

Parameters:: device (DeviceType) – Either a DeviceEnum or string; will be resolved to the effective device.
Return type:: None

emmi_inference.device.resolve_autocast(effective_device, requested)¶

Decide whether autocast should be enabled.

Autocast is disabled on Apple MPS as PyTorch doesn’t support it.

Parameters:

effective_device (str) – The resolved device string (“cuda”, “mps”, or “cpu”).
requested (bool) – Whether the user requested autocast.

Returns:

True if autocast should be enabled on the effective device, otherwise False.

Return type:

bool

class emmi_inference.device.DeviceContext(device, dtype, autocast)¶

Carrier for device execution context.

Parameters:

amp()¶

Return an AMP/autocast context manager for the current device.

Returns:: torch.cuda.amp.autocast when enabled on CUDA; otherwise a no-op context manager.
Return type:: ContextManager