Mixing CPU/GPU parameters in one Chainer model

cpugpu.py

import chainer
from chainer import cuda
import chainer.functions as F
import chainer.links as L
import chainer.optimizers as O
import cupy
import numpy


class MyModel(chainer.Chain):
    def __init__(self):
        super(MyModel, self).__init__(
            embed=L.EmbedID(10, 3),
            W=L.Linear(3, 3),
        )

    def to_gpu(self, device=None):
        with cuda.get_device(device):
            self.W.to_gpu()  # only send `W` to GPU

    def __call__(self, x, y):
        e = self.embed(x)
        e = F.copy(e, 0)  # send `e` to GPU0
        return F.sum(self.W(e) - y)


model = MyModel()
model.to_gpu()

optimizer = O.SGD()
optimizer.setup(model)

x = numpy.array([1, 2, 3], 'i')
y = cupy.arange(9, dtype='f').reshape((3, 3))
l = model(x, y)

model.zerograds()
l.backward()
optimizer.update()

print(model.embed.W.data)
print(type(model.embed.W.data))
print(model.W.W.data)
print(type(model.W.W.data))