github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #8708] Warn users that install.sh may break their system #67700

New Issue
Open
opened 2026-05-04 11:22:57 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-05-04 11:22:57 -05:00
Owner
Copy Link

Originally created by @coffeecoding on GitHub (Jan 30, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8708

System: Debian VM, x64, V100

I had installed my cuda and gpu drivers myself previously.

Shortly after installing OLLAMA, my nvidia installation completely broke:

nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

I suspect that the 374 line long install.sh messed with my driver installations, as I realized that it tries to install (nvidia/cuda) drivers again and changes some system configs.

Would it be possible to warn the user that this way of installing may interfere with their system in unwanted ways, potentially even breaking things?

I suggest inserting something like the following at the start of install.sh.

echo "WARNING: This script modifies system files, installs NVIDIA drivers, and may break existing setups."
echo "Press Ctrl+C to cancel or Enter to continue."
read

I appreciate your kind feedback.

Originally created by @coffeecoding on GitHub (Jan 30, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8708 System: Debian VM, x64, V100 I had installed my cuda and gpu drivers myself previously. Shortly after installing OLLAMA, my nvidia installation completely broke: ``` nvidia-smi NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running. ``` I suspect that the 374 line long `install.sh` messed with my driver installations, as I realized that it tries to install (nvidia/cuda) drivers again and changes some system configs. Would it be possible to warn the user that this way of installing may interfere with their system in unwanted ways, potentially even breaking things? I suggest inserting something like the following at the start of `install.sh`. ``` echo "WARNING: This script modifies system files, installs NVIDIA drivers, and may break existing setups." echo "Press Ctrl+C to cancel or Enter to continue." read ``` I appreciate your kind feedback.
GiteaMirror commented 2026-05-04 11:22:59 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Jan 30, 2025):

The intent is to only install cuda if it doesn't already exist so if it's replacing existing drivers, that would suggest a bug.

The install script hasn't been updated in a while so it's due for some improvements. We'll keep this in mind.

<!-- gh-comment-id:2625687760 --> @mxyng commented on GitHub (Jan 30, 2025): The intent is to only install cuda if it doesn't already exist so if it's replacing existing drivers, that would suggest a bug. The install script hasn't been updated in a while so it's due for some improvements. We'll keep this in mind.
GiteaMirror commented 2026-05-04 11:23:00 -05:00
Author
Owner
Copy Link

@fralund commented on GitHub (Jan 30, 2025):

It did the same for me on Ubuntu 24.04. I hade spent a considerable amount of time to get my system to handle hibernation while using the nvidia driver and the Ollama installations destroyed it completely. I have now spent a day to try to get it working again without success.

Please don't just change drivers without an option to decline.

<!-- gh-comment-id:2625799077 --> @fralund commented on GitHub (Jan 30, 2025): It did the same for me on Ubuntu 24.04. I hade spent a considerable amount of time to get my system to handle hibernation while using the nvidia driver and the Ollama installations destroyed it completely. I have now spent a day to try to get it working again without success. Please don't just change drivers without an option to decline.
GiteaMirror commented 2026-05-04 11:23:02 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 31, 2025):

You can use docker to sandbox ollama (or other applications) and prevent changes to the host environment.

<!-- gh-comment-id:2627140031 --> @rick-github commented on GitHub (Jan 31, 2025): You can use docker to sandbox ollama (or other applications) and prevent changes to the host environment.
GiteaMirror commented 2026-05-04 11:23:04 -05:00
Author
Owner
Copy Link

@SquisherDerHeld commented on GitHub (Jan 31, 2025):

You can use docker to sandbox ollama (or other applications) and prevent changes to the host environment.

(Excuse me if the following sounds rude; I'm not a native speaker and I cannot really judge if my tone is off)

While I can get where your reasoning comes from, I don't think thats the way the installer should behave. Having to run cuda in a container means keeping the nvidia-container-toolkit in sync with the host-kernel-driver. Which is a hassle of its own. Making changes to the system besides installing ollama is highly unexpected.

If cuda is available and working, the install.sh should not attempt to put its own ideas on how modules are loaded and configured into the system. If cuda not working but could be working (nvidia gpu detected but something is off) it should just point to some guide on how to enable cuda.

From a user perspective the install.sh should only download and setup ollama and warn about mis-configuration. Things like adding or changing stuff in the nvidia.conf modules load config is not wanted and it also cannot be reverted follwing the docs on uninstall.

<!-- gh-comment-id:2628348383 --> @SquisherDerHeld commented on GitHub (Jan 31, 2025): > You can use docker to sandbox ollama (or other applications) and prevent changes to the host environment. (Excuse me if the following sounds rude; I'm not a native speaker and I cannot really judge if my tone is off) While I can get where your reasoning comes from, I don't think thats the way the installer should behave. Having to run cuda in a container means keeping the nvidia-container-toolkit in sync with the host-kernel-driver. Which is a hassle of its own. Making changes to the system besides installing ollama is highly unexpected. If cuda is available and working, the install.sh should not attempt to put its own ideas on how modules are loaded and configured into the system. If cuda not working but could be working (nvidia gpu detected but something is off) it should just point to some guide on how to enable cuda. From a user perspective the install.sh should only download and setup ollama and warn about mis-configuration. Things like adding or changing stuff in the nvidia.conf modules load config is not wanted and it also cannot be reverted follwing the docs on uninstall.
GiteaMirror commented 2026-05-04 11:23:05 -05:00
Author
Owner
Copy Link

@fralund commented on GitHub (Jan 31, 2025):

You can use docker to sandbox ollama (or other applications) and prevent changes to the host environment.

I do that on other machines and that is of cource a good practice, but that's not the point. The install.sh script obviously do an apt install -y (or something like that) with all the consequences. If you simply remove the -y it's fine. Now it simply uninstalled and installed 1.5 GB of kernel modules, wayland components and other stuff without warning me. If I simply got the question if I want to do that it would be perfectly fine. Besides this glitch in the installer it's an impressive product you have made.

<!-- gh-comment-id:2628537918 --> @fralund commented on GitHub (Jan 31, 2025): > You can use docker to sandbox ollama (or other applications) and prevent changes to the host environment. I do that on other machines and that is of cource a good practice, but that's not the point. The install.sh script obviously do an `apt install -y` (or something like that) with all the consequences. If you simply remove the `-y` it's fine. Now it simply uninstalled and installed 1.5 GB of kernel modules, wayland components and other stuff without warning me. If I simply got the question if I want to do that it would be perfectly fine. Besides this glitch in the installer it's an impressive product you have made.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#67700
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?