Pytorch install broken on Jetson Thor

[javier41]

A couple months back, I followed the instructions on this post to install a torch wheel built for Cuda 13.0. It worked quite well.

As I was trying to package a small python project, I tried to point my uv install to the index at sbsa/cu130 index where I originally got the working wheels.

I realized that the latest torch build (from 2025-10-23) seems to be broken on Thor. Has anybody else seen this, or have a workaround?

I’m installing this into a uv venv using python 3.12.3.
If I do this:

uv pip install -U https://pypi.jetson-ai-lab.io/sbsa/cu130/+f/d03/870b7c360cc90/torch-2.9.0-cp312-cp312-linux_aarch64.whl#sha256=d03870b7c360cc90c5c12f77650ab47cdf81a76338b292873d164b197cb1a23e

or more simply this:

uv pip install torch   --index-url https://pypi.jetson-ai-lab.io/sbsa/cu130

I get this:

$ uv run python
Python 3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/javier/.venv/lib/python3.12/site-packages/torch/__init__.py", line 427, in <module>
    from torch._C import *  # noqa: F403
    ^^^^^^^^^^^^^^^^^^^^^^
ImportError: libcudss.so.0: cannot open shared object file: No such file or directory

Luckily, I kept the torch wheel I downloaded a couple months ago around. If I install that one, I’m able to import torch, no problem, and it shows Cuda available. (this is the exact url I originally got the wheel from, which is now no longer available).

[whitesscott]

edit: now looking at your error you could resolve it with

sudo apt update && sudo apt install libcudss0-cuda-13

or

---

pytorch starting with version 2.9.0 now builds wheels with cuda. since you are using uv:

uv pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130

[AastaLLL]

Hi,

As @whitesscott mentioned, please try the upstream packages directly:

$ pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130

Thanks.

[javier41]

Thank you both for your help! I still run into the same libcudss.so.0 error with the wheels from https://download.pytorch.org/whl/cu130 . Any other ideas?

If I try to install the missing library it as @whitesscott suggested, my package manager is unable to locate it.

Fetched 1,565 kB in 1s (1,055 kB/s)
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
3 packages can be upgraded. Run 'apt list --upgradable' to see them.
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
E: Unable to locate package libcudss0-cuda-13

I’m running on a Jetson Thor with the latest version of Jetpack 7, which ships with drivers and libraries bundled a specific way (which I’m not very familiar with). It’s been a different experience developing for a Jetson vs a datacenter GPU.

[whitesscott]

First confirm you have this file with this contents and if not create it.

cat /etc/apt/sources.list.d/cuda-ubuntu2404-sbsa.list 
deb [signed-by=/usr/share/keyrings/cuda-archive-keyring.gpg] https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/sbsa/ /

---

Second try again to install, I just installed it with

sudo apt update
sudo apt install libcudss0-cuda-13
The following NEW packages will be installed:
  libcudss0-cuda-13

scott@chithor:~$ find /usr -iname "libcudss*"
/usr/lib/aarch64-linux-gnu/libcudss
/usr/lib/aarch64-linux-gnu/libcudss/13/libcudss.so.0
/usr/lib/aarch64-linux-gnu/libcudss/13/libcudss_mtlayer_gomp.so.0
/usr/lib/aarch64-linux-gnu/libcudss/13/libcudss_commlayer_openmpi.so.0
/usr/lib/aarch64-linux-gnu/libcudss/13/libcudss.so.0.7.1
/usr/lib/aarch64-linux-gnu/libcudss/13/libcudss_mtlayer_gomp.so.0.7.1
/usr/lib/aarch64-linux-gnu/libcudss/13/libcudss_commlayer_nccl.so.0
/usr/lib/aarch64-linux-gnu/libcudss/13/libcudss_commlayer_nccl.so.0.7.1
/usr/lib/aarch64-linux-gnu/libcudss/13/libcudss_commlayer_openmpi.so.0.7.1
/usr/share/doc/libcudss0-cuda-13

scott@chithor:~$ head /etc/nv_tegra_release
# R38 (release), REVISION: 2.2, GCID: 42205042, BOARD: generic, EABI: aarch64, DATE: Thu Sep 25 22:47:11 UTC 2025

---

Third if you have /usr/lib/aarch64-linux-gnu/libcudss/13/libcudss.so.0
Try this export. If it works add to your ~/.bashrc

export LD_LIBRARY_PATH=/usr/local/cuda-13.0/targets/sbsa-linux/lib:$LD_LIBRARY_PATH

[johnny_nv]

wget https://developer.download.nvidia.com/compute/cudss/0.7.1/local_installers/cudss-local-repo-ubuntu2404-0.7.1_0.7.1-1_arm64.deb
sudo dpkg -i cudss-local-repo-ubuntu2404-0.7.1_0.7.1-1_arm64.deb
sudo cp /var/cudss-local-repo-ubuntu2404-0.7.1/cudss-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cudss

[javier41]

Thank you @whitesscott and @johnny_nv .

Both of your last solutions worked for me! Really appreciate the help.