1

I have been trying to transfer the scipy function cdist over to the gpu using cupy.

from cupyx.scipy.spatial import distance as cudist
import cupy as cu
a = cu.array([1,2])
b = cu.array([3,4])
c = cudist.cdist(a, b, metric="euclidean")

I get a

RuntimeError: pylibraft is not installed

I after looking on the cupy closed issues I see that they made the decision to use the RAFT algorithms and that I need to install those to get pylibraft. So I do so as described in the pip section of the RAPIDSAI's RAFT git repo and I run:

pip install pylibraft-cu11 --extra-index-url=https://pypi.nvidia.com

Just to receive the rather concise error during the getting requirements to build wheel portion of the pip install

RuntimeError: Bad params

I am using python 3.11

nvcc --versionreveals cuda_12.1.r12.1

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 531.61                 Driver Version: 531.61       CUDA Version: 12.1     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                      TCC/WDDM | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf            Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce RTX 3060       WDDM | 00000000:01:00.0  On |                  N/A |
|  0%   36C    P8               16W / 170W|    345MiB / 12288MiB |      1%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
raceee
  • 477
  • 5
  • 14

1 Answers1

0

Let's get you sorted by addressing the 3 main challenges here, which all circulate around the fact that RAFT requires the same installation prerequisites as the rest of the RAPIDS libraries:

  1. There is no official RAPIDS stand alone release for CUDA 12 yet, and you are using CUDA 12.1. You will need to install CUDA 11.x (like 11.8) on your system so that when you run nvcc --version, it says 11.x
  2. If you looked a the RAPIDS install selector, RAPIDS works on Python 3.8 and 3.10 (, and doesn't yet support Python 3.11
  3. Lastly, it looks like you're running directly on windows. RAPIDS only supports Windows when in WSL2

Here is a quick colab to confirm that, at time of posting, it does successfully install when in a proper environment

TaureanDyerNV
  • 1,208
  • 8
  • 9