Whether computation of torch arrays is supported？

[BaseBlank]

I'm calculating the Chamfer Distance for the point cloud, but I'm getting an error when using a torch array instead of a numpy array.
The calculation of torch array is not supported. When Chamfer Distance is used as the loss function of the model, the calculation is forcibly converted to numpy. The calculated loss function does not have gradient information, so the model cannot be updated.
Code error:
TypeError: k_nearest_neighbors(): incompatible function arguments. The following argument types are supported:
1. (query_points: numpy.ndarray, dataset_points: numpy.ndarray, k: int, squared_distances: bool = False, max_points_per_leaf: int = 10, num_threads: int = -1) -> Tuple[object, object]

Invoked with:

Many thanks!