0

First of all, I am fairly new to running models on GPU, so sorry in advance for stupid questions.

I use RAPIDS cuML to GPU-accelerate some algorithms, but I noticed I cannot use the latest version (23.2.0) in a Kaggle notebook. When importing cuML, version 21.12.0 is installed, which is the wrong version. I don't know how to make it work or whether the latest version simply is not supported.

Could anyone help me with this? Is there another reason it does install the latest version? Thanks in advance!

(I need it because the older version does not let me define a metric for its UMAP algorithm from cuml.manifold)

I first tried to run this: !pip install cuml-cu11 --extra-index-url=https://pypi.nvidia.com

which gave me version 21.12.0

Then I tried this: !pip install cuml-cu11==23.2.0 --extra-index-url=https://pypi.nvidia.com

which did not work.

svaladou
  • 1
  • You can't use a more recent version while Kaggle only provides Python 3.7. Kaggle just merged support for Python 3.10 into their Docker-python Github repository, so when this is available you should hopefully be able to pip install a recent version of cuML. – Nick Becker Apr 06 '23 at 21:43
  • If you need the latest version or the night build you could try [colab](https://colab.research.google.com/) – rpanai Apr 16 '23 at 22:05

0 Answers0