Closed racng closed 7 months ago
Hi @racng I had the same problem when installing scarches using the yaml. The problem for me was that the cuda driver did not get the specific cuda version
they needed. Check the nvidia-smi
command line to see which version of cuda do you need.
For me it was the 12.1
name: scarches
channels:
- pytorch
- nvidia
- defaults
dependencies:
- python=3.9
- pip
- numpy
- pytorch
- torchaudio
- torchvision
- pytorch-cuda=12.1
- pip:
- scvi-tools
- scarches
variables:
SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL: True
It may help.
The problem was in pytorch-cuda=11.7, i updated it to pytorch-cuda=11.8. Now it should work, feel free to reopen if it doesn't.
Can I have the list of essential package versions installed in your dev environment?
I have tried installing all requirements using the yaml file provided in this repo; however, it ends up installing cpu versions of torchvision and torchaudio, which prevents the usage of our GPU. If I follow pytorch instructions to install pytorch with CUDA 11.7 (
conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.7 -c pytorch -c nvidia
), I get cuda versions of torchvision and torchaudio installed, but I run into jax errors when trying to import scArches.YAML file used
What exact versions of the following packages in combination would work for scarches?
Thanks for your help!