PyTorch version:
torch.__version__: 1.3.0

CuPy config:

CuPy Version          : 7.0.0rc1
CUDA Root             : /usr/local/cuda-10.0
CUDA Build Version    : 10000
CUDA Driver Version   : 10020
CUDA Runtime Version  : 10000
cuDNN Build Version   : None
cuDNN Version         : None
NCCL Build Version    : None
NCCL Runtime Version  : None

With PyTorch accepting the cuda_array_interface, I'd expect to be able to use a CuPy generated CUDA array in PyTorch without going through DLPack.

import cupy as cp
import torch

a = cp.random.rand(10000)
b = torch.as_tensor(a)

Throws: ValueError: given array strides not a multiple of the element byte size. Make a copy of the array to reallocate the memory.

If this is on PyTorch's side, please let me know, and I'll file an issue there. Thanks!