[BUG] RMM Python Docs use nonexistant rmm.to_device() method

[lphan]

Describe the bug
AttributeError: module 'rmm' has no attribute 'to_device'

Steps/Code to reproduce bug
Step 1: Install RAPIDS from RAPIDS.ai
conda create -n rapids-0.17 -c rapidsai -c nvidia -c conda-forge -c defaults rapids-blazing=0.17 python=3.7 cudatoolkit=10.2

Step 2: activate environment rapids-0.17
conda activate rapids-0.17

Step 3:
import rmm
import numpy as np

DeviceBuffer represents an untyped, uninitialized device memory allocation. DeviceBuffers can be created by providing the size of the allocation in bytes

buf = rmm.DeviceBuffer(size=100)
print("Buffer Size: ", buf.size, "Buffer Pointer: ", buf.ptr)

DeviceBuffers can also be created by copying data from host memory

a = np.array([1, 2, 3], dtype='float64')
buf = rmm.to_device(a.tobytes())
print(buf.size)

Source: https://docs.rapids.ai/api/rmm/stable/basics.html

Expected behavior
function to_device should exist or documentation should be updated.

or maybe do I miss something?

Thank you for your work.
Best, Long Phan

[ysq151944]

maybe the api is rmm.DeviceBuffer.to_device?

[lphan]

Thanks, it worked.

[Wuppus]

Documentation still says rmm.to_device
from https://docs.rapids.ai/api/rmm/stable/basics.html#installation stable(21.10)

import rmm
import numpy as np
a = np.array([1, 2, 3], dtype='float64')
buf = rmm.to_device(a.tobytes())
buf.size

[harrism]

Reopening and changing to a doc bug.

[shwina]

The to_device() method was removed in a previous method of RMM. The solution today is to use CuPy directly to allocate device arrays. See here for how to use RMM with CuPy.

[Wuppus]

Ok thanks!