AsyncGpuResult

class nvmolkit.types.AsyncGpuResult(obj, gpu_id: int | None = None)

Handle to a GPU result.

Populates the __cuda_array_interface__ attribute which can be consumed by other libraries. Note that this result is async, and the data cannot be accessed without a sync, such as torch.cuda.synchronize().

__init__(obj, gpu_id: int | None = None)

Internal construction of the AsyncGpuResult object.

Parameters:

• obj – An object exposing __cuda_array_interface__.

• gpu_id – Optional GPU device id where the underlying buffer lives. If omitted, torch infers the device from the CUDA pointer attributes (typically the current device).

property device

Return the device of the underlying data.

torch()

Return the underlying data as a torch tensor. This is an asynchronous operation.

numpy()

Return the underlying data as a numpy array. This is a blocking operation.