This needs a release note. It should be versionadded for 2.1 since I don't think it's an emergency to backport this for 2.0. Also needs new tests.

I've added a release note as per suggested. Is there a way to flag this change for 2.1 in the release note? As for the tests, the underlying implementation for np.asanyarray (_array_fromobject_generic) is already covered by tests surrounding np.array, is there still a need to write tests? If so should I also add tests for np.asarray?

Is there a way to flag this change for 2.1 in the release note?

That is automatic in the release notes; the tooling knows how to assemble the release notes for the next version from these snippets.

And you already used .. versionadded:: correctly in the docs (except there adding the correct version number manually is needed, and it should be 2.1.0 rather than 2.0.0).

As for the tests, the underlying implementation for np.asanyarray (_array_fromobject_generic) is already covered by tests surrounding np.array, is there still a need to write tests? If so should I also add tests for np.asarray?

Yes, this will be useful. Any public API construct should have at least a single direct test even when the underlying machinery is tested extensively in some other way. Probably TestDevice in _core/tests/test_multiarray.py is the place to put this. And yes, doing that for np.asarray too would be nice, good catch.

And you already used .. versionadded:: correctly in the docs (except there adding the correct version number manually is needed, and it should be 2.1.0 rather than 2.0.0).

Oh my bad, I completely forgot I did that, I've made the changes.

If the only new tests to add are related to the device keyword, then I'm not sure what tests to add since all methods discard the device keyword, and arr.device calls the following stub:

static PyObject *
array_device(PyArrayObject *self, void *NPY_UNUSED(ignored))
{
    return PyUnicode_FromString("cpu");
}

I think you only test that np.asanyarray(...), np.asanyarray(..., device=None) and np.asanyarray(..., device='cpu') all work and yield the same result - that ensures that usage of the keyword functions as expected. And that np.asanyarray(..., device='nonsense') raises an understandable exception.

Okay, I've added new tests

Yea I think adding one more newline looks nicer too, Ive commited the suggestions

Thanks @JuliaPoo