## Description
Current implementation of `NDArray.asnumpy()` method does not have the option 
to avoid the physical copy in the memory. Another framework of PyTorch has a 
similar method of `Tensor.numpy()`, which does the reference copy. That means 
PyTorch create a `numpy.array()` with the same memory of the `torch.Tensor`. We 
could similarly allow the reference copy option by setting an optional `copy` 
option in `NDArray.asnumpy(copy=False)` method.

@szha @eric-haibin-lin may have further suggestions


[ Full content available at: 
https://github.com/apache/incubator-mxnet/issues/12699 ]
This message was relayed via gitbox.apache.org for [email protected]

Reply via email to