Rate this Page

torch.accelerator.current_accelerator#

torch.accelerator.current_accelerator(check_available=False)[source]#

Return the device of the accelerator available at compilation time. If no accelerator were available at compilation time, returns None. See accelerator for details.

Parameters

check_available (bool, optional) – if True, will also do a runtime check to see if the device torch.accelerator.is_available() on top of the compile-time check. Default: False

Returns

return the current accelerator as torch.device.

Return type

torch.device

Note

The index of the returned torch.device will be None, please use torch.accelerator.current_device_index() to know the current index being used. This API does NOT poison fork. For more details, see Poison fork in multiprocessing.

Example:

>>> # If an accelerator is available, sent the model to it
>>> model = torch.nn.Linear(2, 2)
>>> if (current_device := current_accelerator(check_available=True)) is not None:
>>>     model.to(current_device)