[BUG] Installation with conda doesn't work.

[Amit-Weiss]

Describe the bug When I try to install using the generated command from https://docs.rapids.ai/install#get-rapids, the installation fails with:

Could not solve for environment specs
The following package could not be installed
└─ cuml 24.06**  is not installable because it requires
   └─ cupy >=12.0.0 , which conflicts with any installable versions previously reported.

Trying various types of selections such as different CUDA, different python, different Release results in the same errors. Also, leaving out parts of the command like cuda-version installation or downgrading to different versions still doesn't work.

Steps/Code to reproduce bug Run the command generated from the selector:

conda create -n rapids-24.06 -c rapidsai -c conda-forge -c nvidia  \
    cuml=24.06 python=3.11 cuda-version=11.8

OR By creating a new environment and then running the conda install command OR By using a conda.yml file.

Expected behavior The package should be installed.

Environment details (please complete the following information):

• Environment location: AWS Cloud
• Linux Distro/Architecture: Ubuntu 20.04
• GPU Model/Driver: NVIDIA A10G - g5.4xlarge on AWS.
• CUDA: 11.7 (Also tried with CUDA 12.0 and the appropriate command)
• Method of cuDF & cuML install: conda
  • If method of install is [conda], run conda list and include results here: No conda list, this is the first command.

Additional context I have tried various options and configurations, including installing the package that reported to be conflicting etc. It doesn't work for me and I cannot find a way to install even the slimmest version. Thanks for your help

[dantegd]

Thanks for the issue @Amit-Weiss, seems like the issue is with installing CuPy, to help two things would be really useful:

• Could you post the full error log?
• I would recommend creating a fresh conda environment and trying to install cupy >= 12 on it, and then install cuml on it if it works. This could work or tell us what is the issue better as well.

mamba create -n cupy12cuml2306
mamba activate cupy12cuml2306
mamba install -c conda-forge "cupy>=12" "cuda-version=11.8"
mamba install -c rapidsai -c conda-forge -c nvidia cuml=24.06 cuda-version=11.8

[Amit-Weiss]

Hey @dantegd Thank you for your comment. I followed your steps, installing cupy works but when I run the cuml installation I still get an error. The full error log is:

mamba install -c rapidsai -c conda-forge -c nvidia cuml=24.06 cuda-version=11.8

Looking for: ['cuml=24.06', 'cuda-version=11.8']

conda-forge/linux-64                                        Using cache
conda-forge/noarch                                          Using cache
pytorch/linux-64                                            Using cache
pytorch/noarch                                              Using cache
nvidia/noarch                                                 No change
nvidia/linux-64                                               No change
rapidsai/linux-64                                             No change
rapidsai/noarch                                               No change

Pinned packages:
  - python 3.12.*


warning  libmamba Added empty dependency for problem type SOLVER_RULE_UPDATE
Could not solve for environment specs
The following packages are incompatible
└─ cuml 24.06**  is installable with the potential options
   ├─ cuml [24.06.00|24.06.01] would require
   │  └─ python >=3.10,<3.11.0a0 , which can be installed;
   ├─ cuml [24.06.00|24.06.01] would require
   │  └─ python >=3.11,<3.12.0a0 , which can be installed;
   └─ cuml [24.06.00|24.06.01] would require
      └─ python >=3.9,<3.10.0a0 , which can be installed.

I also tried installing different python versions as suggested in the error logs, but I still get The following packages are incompatible, now with packages like raft-dask or ucx-py.

[dantegd]

Do you happen to have the logs of the errors of raft-dask and ucx-py? I ask, since there are no RAPIDS python 3.12 packages yet, so the error with python 3.12 is not very useful to debug what's going wrong here.