Installation instructions not working for CUDA 11.4
See original GitHub issueThis is with python 3.9.6 and pip 21.2.4.
On the README it says
pip install --upgrade
Easy enough. Then:
pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-release.html
This ran into an error because of the jaxlib dependency:
Looking in links: https://storage.googleapis.com/jax-release.html
Collecting jax[cuda111]
Using cached jax-0.2.19-py3-none-any.whl
Requirement already satisfied: opt-einsum in /home/eben/anaconda3/envs/refgrads/lib/python3.9/site-packages (from jax[cuda111]) (3.3.0)
Requirement already satisfied: numpy>=1.18 in /home/eben/anaconda3/envs/refgrads/lib/python3.9/site-packages (from jax[cuda111]) (1.20.3)
Requirement already satisfied: absl-py in /home/eben/anaconda3/envs/refgrads/lib/python3.9/site-packages (from jax[cuda111]) (0.13.0)
Using cached jax-0.2.18.tar.gz (708 kB)
Using cached jax-0.2.17.tar.gz (693 kB)
Using cached jax-0.2.16.tar.gz (680 kB)
Using cached jax-0.2.15.tar.gz (681 kB)
ERROR: Packages installed from PyPI cannot depend on packages which are not also hosted on PyPI.
jax depends on jaxlib@ https://storage.googleapis.com/jax-releases/cuda111/jaxlib-0.1.67+cuda111-cp39-none-manylinux2010_x86_64.whl
I tried pip install jaxlib
and then tried re-installing with the same command. That did not work, so I uninstalled jaxlib and then ran pip install jaxlib==0.1.67
, but trying to installing jax itself still yields the above error message.
Issue Analytics
- State:
- Created 2 years ago
- Comments:10
Top Results From Across the Web
NVIDIA CUDA Installation Guide for Linux
NVIDIA CUDA Installation Guide for Linux. The installation instructions for the CUDA Toolkit on Linux. 1. Introduction . CUDA® is a parallel...
Read more >How to install Cuda 11.4 on ubuntu 18.04(or 20.04) - Medium
I need to install Cuda on ubuntu and I installed it on Centos 7 before without any issue. There were some issues on...
Read more >Problem while installing cuda toolkit in ubuntu 18.04
I think the issue is the CUDA driver version. It looks like the installer tries to install the newest version 455.23.05 and the...
Read more >CUDA 11.4 and torch version 1.11.0 not working - PyTorch Live
UNINSTALL torch in Terminal: pip3 uninstall torch · DELETE the pip folder in ~/.cache · Rerun installation of torch and cuda from this...
Read more >Not working on RTX 3060 and RTX 3090 · Issue #14 - GitHub
I used the installation instructions with conda on RTX 3060 and RTX 3090. Cuda 11.4 Ubuntu 20 and 21 both. Getting this error....
Read more >
Top Related Medium Post
No results found
Top Related StackOverflow Question
No results found
Troubleshoot Live Code
Lightrun enables developers to add logs, metrics and snapshots to live code - no restarts or redeploys required.
Start Free
Top Related Reddit Thread
No results found
Top Related Hackernoon Post
No results found
Top Related Tweet
No results found
Top Related Dev.to Post
No results found
Top Related Hashnode Post
No results found
Can you try pinning the JAX version? i.e.
Also, double check your releases URL; the URL is
https://storage.googleapis.com/jax-releases/jax_releases.html
, above you typedhttps://storage.googleapis.com/jax-release.html
. Using the wrong URL would make pip unable to find the jaxlib wheels, and it would keep trying older jax versions until it eventually hits this error on 0.2.15.If the URL is correct, then I suspect it has to do with pip’s new(ish) dependency resolver trying to satisfy version requirements for other packages you have installed.
Great, glad it’s working! Sorry I didn’t notice that URL typo earlier.