[QST] Is it possible to install cuML without Anaconda?

[songwenas12]

What is your question?
Thank you for developing such an amazing tool. I am using cuML for a commercial project. Currently we use docker to deploy our model, but found that the docker imge depends on Anaconda, for which I need to obtain licence for commercial use. Is it possible for me to install and deploy cuML without Anaconda?

Thanks in advance for your kind help!

[quasiben]

Yes! You can install cuML through pip:

pip install
--extra-index-url=https://pypi.nvidia.com
"cudf-cu12==24.12." "cuml-cu12==24.12."

You can see all the options on the install guide page

We'd love to hear more about your usage and experience with cuML if you're able to share.

[betatim]

In addition to Ben's reply, you can continue to install the conda package of cuml as it does not depend on packages from the defaults conda channel. As far as I understand the licensing requirements it isn't the use of conda but the use of packages from the defaults channel that counts. An article that explains more: https://prefix.dev/blog/towards_a_vendor_lock_in_free_conda_experience

The docker image you are currently using, could you share a few more details on it? Is it one maintained by RAPIDS or someone else? if it is maintained by RAPIDS we could potentially improve the documentation to help others with the conda licensing.

[bdice]

Adding a few more links to @betatim's excellent reply above:

RAPIDS is built and tested without the use of the defaults channel. We use the miniforge conda distribution, which uses only conda-forge and does not use Anaconda's defaults channel (unless explicitly added by the user). We recommend the Miniforge distribution as a minimal installation that uses conda-forge by default.

The Docker images deployed by RAPIDS are based on miniforge: https://github.com/rapidsai/docker/blob/b2701d6095dae3bd5b4782863c3c461c57b42512/Dockerfile#L37

You can also add nodefaults to your conda channel list to avoid the use of the defaults channel. See this link (scroll down a bit): https://docs.conda.io/projects/conda/en/stable/user-guide/tasks/manage-environments.html#creating-an-environment-file-manually

[songwenas12]

Thank you all for the valuable answers! @betatim we are indeed using the official docker image provided by the RAPIDS selector. So glad to know that we can continue using the default docker image 😊 👍

[betatim]

Great :D

I'll close this then as it seems like the original question is solved. Feel free to re-open it if needed.