Incomplete Documentation for doc.tensor

[FloGro3]

When using the pipeline "de_core_news_sm" from Spacy and running it with a GPU the type of doc.tensor isn't numpy.ndarray as stated in the documentation Spacy Doc Attributes.
As you can see the type is "cupy._core.core.ndarray" which means you can't use all the methods you could use with a numpy.ndarray. You'll get a numpy.ndarray when you perform a get() call an the tensor:

Here's the code to reproduce the situation:

import spacy
spacy.require_gpu()
nlp = spacy.load("de_core_news_sm")
doc = nlp("Dies ist ein Test.")
type(doc.tensor)
type(doc.tensor.get())

A small tip or hint regarding this special situation would be nice in the documentation in my opinion :)

[rmitsch]

You're right, the docs could be clearer here. We'll update them. Thank for reporting this!