🚀 Feature

Let the dtype keyword argument of torch.as_tensor be either a np.dtype or torch.dtype.

Motivation

Suppose I have two numpy arrays with different types and I want to convert one of them to a torch tensor with the type of the other array.

According to https://discuss.pytorch.org/t/converting-a-numpy-dtype-to-torch-dtype/52279/2, there's no convenient way to convert a numpy tensor to a torch tensor.

Pitch

import numpy as np
import torch


# currently raises the following:
# TypeError: as_tensor(): argument 'dtype' must be 
# torch.dtype, not numpy.dtype
mytensor = torch.tensor(48., dtype=np.float32)

Alternatives

• make this API public
• let the dtype keyword argument of torch.Tensor.to be either a np.dtype or torch.dtype

Additional context

Similar issue: #541

cc @mruberry